0 files changed, 0 insertions, 0 deletions
diff --git a/drivers/gpu/Makefile b/drivers/gpu/Makefile
index e9ed439a5b65..36a54d456630 100644
--- a/drivers/gpu/Makefile
+++ b/drivers/gpu/Makefile
@@ -1,6 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0-only
 # drm/tegra depends on host1x, so if both drivers are built-in care must be
 # taken to initialize them in the correct order. Link order is the only way
 # to ensure this currently.
-obj-$(CONFIG_TEGRA_HOST1X)	+= host1x/
-obj-y			+= drm/ vga/
+obj-y			+= host1x/ drm/ vga/
 obj-$(CONFIG_IMX_IPUV3_CORE)	+= ipu-v3/
+obj-$(CONFIG_TRACE_GPU_MEM)		+= trace/
+obj-$(CONFIG_NOVA_CORE)		+= nova-core/
diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index 78d7fc0ebb57..7e6bc0b3a589 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: GPL-2.0-only
 #
 # Drm device configuration
 #
@@ -7,12 +8,15 @@
 menuconfig DRM
 	tristate "Direct Rendering Manager (XFree86 4.1.0 and higher DRI support)"
 	depends on (AGP || AGP=n) && !EMULATED_CMPXCHG && HAS_DMA
+	select DRM_PANEL_ORIENTATION_QUIRKS
 	select HDMI
-	select FB_CMDLINE
 	select I2C
-	select I2C_ALGOBIT
 	select DMA_SHARED_BUFFER
 	select SYNC_FILE
+# gallium uses SYS_kcmp for os_same_file_description() to de-duplicate
+# device and dmabuf fd. Let's make sure that is available for our userspace.
+	select KCMP
+	select VIDEO
 	help
 	  Kernel-level support for the Direct Rendering Infrastructure (DRI)
 	  introduced in XFree86 4.0. If you say Y here, you need to select
@@ -22,95 +26,154 @@ menuconfig DRM
 	  details.  You should also select and configure AGP
 	  (/dev/agpgart) support if it is available for your platform.
 
+menu "DRM debugging options"
+depends on DRM
+source "drivers/gpu/drm/Kconfig.debug"
+endmenu
+
+if DRM
+
+config DRM_MIPI_DBI
+	tristate
+	depends on DRM
+	select DRM_KMS_HELPER
+
 config DRM_MIPI_DSI
 	bool
 	depends on DRM
 
-config DRM_DP_AUX_CHARDEV
-	bool "DRM DP AUX Interface"
+config DRM_KMS_HELPER
+	tristate
 	depends on DRM
+	select FB_CORE if DRM_FBDEV_EMULATION
 	help
-	  Choose this option to enable a /dev/drm_dp_auxN node that allows to
-	  read and write values to arbitrary DPCD registers on the DP aux
-	  channel.
-
-config DRM_DEBUG_MM
-	bool "Insert extra checks and debug info into the DRM range managers"
-	default n
-	depends on DRM=y
-	depends on STACKTRACE_SUPPORT
-	select STACKDEPOT
-	help
-	  Enable allocation tracking of memory manager and leak detection on
-	  shutdown.
-
-	  Recommended for driver developers only.
+	  CRTC helpers for KMS drivers.
 
-	  If in doubt, say "N".
+config DRM_DRAW
+	bool
+	depends on DRM
 
-config DRM_DEBUG_MM_SELFTEST
-	tristate "kselftests for DRM range manager (struct drm_mm)"
+config DRM_PANIC
+	bool "Display a user-friendly message when a kernel panic occurs"
 	depends on DRM
-	depends on DEBUG_KERNEL
-	select PRIME_NUMBERS
-	select DRM_LIB_RANDOM
-	default n
+	select FONT_SUPPORT
+	select DRM_DRAW
 	help
-	  This option provides a kernel module that can be used to test
-	  the DRM range manager (drm_mm) and its API. This option is not
-	  useful for distributions or general kernels, but only for kernel
-	  developers working on DRM and associated drivers.
-
+	  Enable a drm panic handler, which will display a user-friendly message
+	  when a kernel panic occurs. It's useful when using a user-space
+	  console instead of fbcon.
+	  It will only work if your graphic driver supports this feature.
+	  To support Hi-DPI Display, you can enable bigger fonts like
+	  FONT_TER16x32
+
+config DRM_PANIC_FOREGROUND_COLOR
+	hex "Drm panic screen foreground color, in RGB"
+	depends on DRM_PANIC
+	default 0xffffff
+
+config DRM_PANIC_BACKGROUND_COLOR
+	hex "Drm panic screen background color, in RGB"
+	depends on DRM_PANIC
+	default 0x000000
+
+config DRM_PANIC_DEBUG
+	bool "Add a debug fs entry to trigger drm_panic"
+	depends on DRM_PANIC && DEBUG_FS
+	help
+	  Add dri/[device]/drm_panic_plane_x in the kernel debugfs, to force the
+	  panic handler to write the panic message to this plane scanout buffer.
+	  This is unsafe and should not be enabled on a production build.
 	  If in doubt, say "N".
 
-config DRM_KMS_HELPER
-	tristate
-	depends on DRM
+config DRM_PANIC_SCREEN
+	string "Panic screen formatter"
+	default "user"
+	depends on DRM_PANIC
 	help
-	  CRTC helpers for KMS drivers.
-
-config DRM_KMS_FB_HELPER
-	bool
-	depends on DRM_KMS_HELPER
-	select FB
-	select FRAMEBUFFER_CONSOLE if !EXPERT
-	select FRAMEBUFFER_CONSOLE_DETECT_PRIMARY if FRAMEBUFFER_CONSOLE
-	select FB_SYS_FOPS
-	select FB_SYS_FILLRECT
-	select FB_SYS_COPYAREA
-	select FB_SYS_IMAGEBLIT
-	select FB_CFB_FILLRECT
-	select FB_CFB_COPYAREA
-	select FB_CFB_IMAGEBLIT
-	select FB_DEFERRED_IO
+	  This option enable to choose what will be displayed when a kernel
+	  panic occurs. You can choose between "user", a short message telling
+	  the user to reboot the system, or "kmsg" which will display the last
+	  lines of kmsg.
+	  This can also be overridden by drm.panic_screen=xxxx kernel parameter
+	  or by writing to /sys/module/drm/parameters/panic_screen sysfs entry
+	  Default is "user"
+
+config DRM_PANIC_SCREEN_QR_CODE
+	bool "Add a panic screen with a QR code"
+	depends on DRM_PANIC && RUST
+	select ZLIB_DEFLATE
 	help
-	  FBDEV helpers for KMS drivers.
+	  This option adds a QR code generator, and a panic screen with a QR
+	  code. The QR code will contain the last lines of kmsg and other debug
+	  information. This should be easier for the user to report a kernel
+	  panic, with all debug information available.
+	  To use this panic screen, also set DRM_PANIC_SCREEN to "qr_code"
+
+config DRM_PANIC_SCREEN_QR_CODE_URL
+	string "Base URL of the QR code in the panic screen"
+	depends on DRM_PANIC_SCREEN_QR_CODE
+	help
+	  This option sets the base URL to report the kernel panic. If it's set
+	  the QR code will contain the URL and the kmsg compressed with zlib as
+	  a URL parameter. If it's empty, the QR code will contain the kmsg as
+	  uncompressed text only.
+	  There is a demo code in javascript, to decode and uncompress the kmsg
+	  data from the URL parameter at https://github.com/kdj0c/panic_report
+
+config DRM_PANIC_SCREEN_QR_VERSION
+	int "Maximum version (size) of the QR code."
+	depends on DRM_PANIC_SCREEN_QR_CODE
+	default 40
+	help
+	  This option limits the version (or size) of the QR code. QR code
+	  version ranges from Version 1 (21x21) to Version 40 (177x177).
+	  Smaller QR code are easier to read, but will contain less debugging
+	  data. Default is 40.
 
-config DRM_FBDEV_EMULATION
-	bool "Enable legacy fbdev support for your modesetting driver"
-	depends on DRM
-	select DRM_KMS_HELPER
-	select DRM_KMS_FB_HELPER
-	default y
+config DRM_DEBUG_DP_MST_TOPOLOGY_REFS
+        bool "Enable refcount backtrace history in the DP MST helpers"
+	depends on STACKTRACE_SUPPORT
+        select STACKDEPOT
+        select DRM_KMS_HELPER
+        depends on DEBUG_KERNEL
+        depends on EXPERT
+        help
+          Enables debug tracing for topology refs in DRM's DP MST helpers. A
+          history of each topology reference/dereference will be printed to the
+          kernel log once a port or branch device's topology refcount reaches 0.
+
+          This has the potential to use a lot of memory and print some very
+          large kernel messages. If in doubt, say "N".
+
+config DRM_DEBUG_MODESET_LOCK
+	bool "Enable backtrace history for lock contention"
+	depends on STACKTRACE_SUPPORT
+	depends on DEBUG_KERNEL
+	depends on EXPERT
+	select STACKDEPOT
+	default y if DEBUG_WW_MUTEX_SLOWPATH
 	help
-	  Choose this option if you have a need for the legacy fbdev
-	  support. Note that this support also provides the linux console
-	  support on top of your modesetting driver.
+	  Enable debug tracing of failures to gracefully handle drm modeset lock
+	  contention. A history of each drm modeset lock path hitting -EDEADLK
+	  will be saved until gracefully handled, and the backtrace will be
+	  printed when attempting to lock a contended lock.
 
-	  If in doubt, say "Y".
+	  If in doubt, say "N".
 
-config DRM_FBDEV_OVERALLOC
-	int "Overallocation of the fbdev buffer"
-	depends on DRM_FBDEV_EMULATION
-	default 100
+config DRM_CLIENT
+	bool
+	depends on DRM
 	help
-	  Defines the fbdev buffer overallocation in percent. Default
-	  is 100. Typical values for double buffering will be 200,
-	  triple buffering 300.
+	  Enables support for DRM clients. DRM drivers that need
+	  struct drm_client_dev and its interfaces should select this
+	  option. Drivers that support the default clients should
+	  select DRM_CLIENT_SELECTION instead.
+
+source "drivers/gpu/drm/clients/Kconfig"
 
 config DRM_LOAD_EDID_FIRMWARE
 	bool "Allow to specify an EDID data set instead of probing for it"
-	depends on DRM_KMS_HELPER
+	depends on DRM
 	help
 	  Say Y here, if you want to use EDID data to be loaded from the
 	  /lib/firmware directory or one of the provided built-in
@@ -118,91 +181,118 @@ config DRM_LOAD_EDID_FIRMWARE
 	  monitor are unable to provide appropriate EDID data. Since this
 	  feature is provided as a workaround for broken hardware, the
 	  default case is N. Details and instructions how to build your own
-	  EDID data are given in Documentation/EDID/HOWTO.txt.
+	  EDID data are given in Documentation/admin-guide/edid.rst.
+
+source "drivers/gpu/drm/display/Kconfig"
 
 config DRM_TTM
 	tristate
 	depends on DRM && MMU
+	select SHMEM
 	help
 	  GPU memory management subsystem for devices with multiple
 	  GPU memory types. Will be enabled automatically if a device driver
 	  uses it.
 
-config DRM_GEM_CMA_HELPER
-	bool
+config DRM_EXEC
+	tristate
 	depends on DRM
 	help
-	  Choose this if you need the GEM CMA helper functions
+	  Execution context for command submissions
 
-config DRM_KMS_CMA_HELPER
-	bool
+config DRM_GPUVM
+	tristate
 	depends on DRM
-	select DRM_GEM_CMA_HELPER
-	select DRM_KMS_FB_HELPER
-	select FB_SYS_FILLRECT
-	select FB_SYS_COPYAREA
-	select FB_SYS_IMAGEBLIT
+	select DRM_EXEC
 	help
-	  Choose this if you need the KMS CMA helper functions
-
-config DRM_VM
-	bool
-	depends on DRM && MMU
+	  GPU-VM representation providing helpers to manage a GPUs virtual
+	  address space
 
-source "drivers/gpu/drm/i2c/Kconfig"
+config DRM_GPUSVM
+	tristate
+	depends on DRM && DEVICE_PRIVATE
+	select HMM_MIRROR
+	select MMU_NOTIFIER
+	help
+	  GPU-SVM representation providing helpers to manage a GPUs shared
+	  virtual memory
 
-source "drivers/gpu/drm/arm/Kconfig"
+config DRM_BUDDY
+	tristate
+	depends on DRM
+	help
+	  A page based buddy allocator
 
-config DRM_RADEON
-	tristate "ATI Radeon"
-	depends on DRM && PCI && MMU
-	select FW_LOADER
-        select DRM_KMS_HELPER
-        select DRM_TTM
-	select POWER_SUPPLY
-	select HWMON
-	select BACKLIGHT_CLASS_DEVICE
-	select BACKLIGHT_LCD_SUPPORT
-	select INTERVAL_TREE
+config DRM_VRAM_HELPER
+	tristate
+	depends on DRM
 	help
-	  Choose this option if you have an ATI Radeon graphics card.  There
-	  are both PCI and AGP versions.  You don't need to choose this to
-	  run the Radeon in plain VGA mode.
+	  Helpers for VRAM memory management
 
-	  If M is selected, the module will be called radeon.
+config DRM_TTM_HELPER
+	tristate
+	depends on DRM
+	select DRM_TTM
+	select DRM_KMS_HELPER if DRM_FBDEV_EMULATION
+	select FB_CORE if DRM_FBDEV_EMULATION
+	select FB_SYSMEM_HELPERS_DEFERRED if DRM_FBDEV_EMULATION
+	help
+	  Helpers for ttm-based gem objects
 
-source "drivers/gpu/drm/radeon/Kconfig"
+config DRM_GEM_DMA_HELPER
+	tristate
+	depends on DRM
+	select DRM_KMS_HELPER if DRM_FBDEV_EMULATION
+	select FB_CORE if DRM_FBDEV_EMULATION
+	select FB_DMAMEM_HELPERS_DEFERRED if DRM_FBDEV_EMULATION
+	help
+	  Choose this if you need the GEM DMA helper functions
 
-config DRM_AMDGPU
-	tristate "AMD GPU"
-	depends on DRM && PCI && MMU
-	select FW_LOADER
-        select DRM_KMS_HELPER
-        select DRM_TTM
-	select POWER_SUPPLY
-	select HWMON
-	select BACKLIGHT_CLASS_DEVICE
-	select BACKLIGHT_LCD_SUPPORT
-	select INTERVAL_TREE
+config DRM_GEM_SHMEM_HELPER
+	tristate
+	depends on DRM && MMU
+	select DRM_KMS_HELPER if DRM_FBDEV_EMULATION
+	select FB_CORE if DRM_FBDEV_EMULATION
+	select FB_SYSMEM_HELPERS_DEFERRED if DRM_FBDEV_EMULATION
 	help
-	  Choose this option if you have a recent AMD Radeon graphics card.
+	  Choose this if you need the GEM shmem helper functions
 
-	  If M is selected, the module will be called amdgpu.
+config DRM_SUBALLOC_HELPER
+	tristate
+	depends on DRM
+
+config DRM_SCHED
+	tristate
+	depends on DRM
+
+source "drivers/gpu/drm/sysfb/Kconfig"
+
+source "drivers/gpu/drm/arm/Kconfig"
+
+source "drivers/gpu/drm/radeon/Kconfig"
 
 source "drivers/gpu/drm/amd/amdgpu/Kconfig"
 
 source "drivers/gpu/drm/nouveau/Kconfig"
 
+source "drivers/gpu/drm/nova/Kconfig"
+
 source "drivers/gpu/drm/i915/Kconfig"
 
+source "drivers/gpu/drm/xe/Kconfig"
+
+source "drivers/gpu/drm/kmb/Kconfig"
+
 config DRM_VGEM
 	tristate "Virtual GEM provider"
-	depends on DRM
+	depends on DRM && MMU
+	select DRM_GEM_SHMEM_HELPER
 	help
 	  Choose this option to get a virtual graphics memory manager,
 	  as used by Mesa's software renderer for enhanced performance.
 	  If M is selected the module will be called vgem.
 
+source "drivers/gpu/drm/vkms/Kconfig"
 
 source "drivers/gpu/drm/exynos/Kconfig"
 
@@ -218,15 +308,11 @@ source "drivers/gpu/drm/ast/Kconfig"
 
 source "drivers/gpu/drm/mgag200/Kconfig"
 
-source "drivers/gpu/drm/cirrus/Kconfig"
-
 source "drivers/gpu/drm/armada/Kconfig"
 
 source "drivers/gpu/drm/atmel-hlcdc/Kconfig"
 
-source "drivers/gpu/drm/rcar-du/Kconfig"
-
-source "drivers/gpu/drm/shmobile/Kconfig"
+source "drivers/gpu/drm/renesas/Kconfig"
 
 source "drivers/gpu/drm/sun4i/Kconfig"
 
@@ -236,8 +322,6 @@ source "drivers/gpu/drm/tilcdc/Kconfig"
 
 source "drivers/gpu/drm/qxl/Kconfig"
 
-source "drivers/gpu/drm/bochs/Kconfig"
-
 source "drivers/gpu/drm/virtio/Kconfig"
 
 source "drivers/gpu/drm/msm/Kconfig"
@@ -246,113 +330,102 @@ source "drivers/gpu/drm/fsl-dcu/Kconfig"
 
 source "drivers/gpu/drm/tegra/Kconfig"
 
+source "drivers/gpu/drm/stm/Kconfig"
+
 source "drivers/gpu/drm/panel/Kconfig"
 
 source "drivers/gpu/drm/bridge/Kconfig"
 
 source "drivers/gpu/drm/sti/Kconfig"
 
-source "drivers/gpu/drm/amd/amdkfd/Kconfig"
-
 source "drivers/gpu/drm/imx/Kconfig"
 
+source "drivers/gpu/drm/ingenic/Kconfig"
+
+source "drivers/gpu/drm/v3d/Kconfig"
+
 source "drivers/gpu/drm/vc4/Kconfig"
 
-source "drivers/gpu/drm/etnaviv/Kconfig"
+source "drivers/gpu/drm/loongson/Kconfig"
 
-source "drivers/gpu/drm/arc/Kconfig"
+source "drivers/gpu/drm/etnaviv/Kconfig"
 
 source "drivers/gpu/drm/hisilicon/Kconfig"
 
-source "drivers/gpu/drm/mediatek/Kconfig"
+source "drivers/gpu/drm/logicvc/Kconfig"
 
-source "drivers/gpu/drm/zte/Kconfig"
+source "drivers/gpu/drm/mediatek/Kconfig"
 
 source "drivers/gpu/drm/mxsfb/Kconfig"
 
 source "drivers/gpu/drm/meson/Kconfig"
 
-source "drivers/gpu/drm/tinydrm/Kconfig"
+source "drivers/gpu/drm/tiny/Kconfig"
 
-# Keep legacy drivers last
+source "drivers/gpu/drm/pl111/Kconfig"
 
-menuconfig DRM_LEGACY
-	bool "Enable legacy drivers (DANGEROUS)"
-	depends on DRM && MMU
-	select DRM_VM
-	help
-	  Enable legacy DRI1 drivers. Those drivers expose unsafe and dangerous
-	  APIs to user-space, which can be used to circumvent access
-	  restrictions and other security measures. For backwards compatibility
-	  those drivers are still available, but their use is highly
-	  inadvisable and might harm your system.
+source "drivers/gpu/drm/tve200/Kconfig"
 
-	  You are recommended to use the safe modeset-only drivers instead, and
-	  perform 3D emulation in user-space.
+source "drivers/gpu/drm/xen/Kconfig"
 
-	  Unless you have strong reasons to go rogue, say "N".
+source "drivers/gpu/drm/vboxvideo/Kconfig"
 
-if DRM_LEGACY
+source "drivers/gpu/drm/lima/Kconfig"
 
-config DRM_TDFX
-	tristate "3dfx Banshee/Voodoo3+"
-	depends on DRM && PCI
-	help
-	  Choose this option if you have a 3dfx Banshee or Voodoo3 (or later),
-	  graphics card.  If M is selected, the module will be called tdfx.
+source "drivers/gpu/drm/panfrost/Kconfig"
 
-config DRM_R128
-	tristate "ATI Rage 128"
-	depends on DRM && PCI
-	select FW_LOADER
-	help
-	  Choose this option if you have an ATI Rage 128 graphics card.  If M
-	  is selected, the module will be called r128.  AGP support for
-	  this card is strongly suggested (unless you have a PCI version).
-
-config DRM_I810
-	tristate "Intel I810"
-	# !PREEMPT because of missing ioctl locking
-	depends on DRM && AGP && AGP_INTEL && (!PREEMPT || BROKEN)
-	help
-	  Choose this option if you have an Intel I810 graphics card.  If M is
-	  selected, the module will be called i810.  AGP support is required
-	  for this driver to work.
-
-config DRM_MGA
-	tristate "Matrox g200/g400"
-	depends on DRM && PCI
-	select FW_LOADER
-	help
-	  Choose this option if you have a Matrox G200, G400 or G450 graphics
-	  card.  If M is selected, the module will be called mga.  AGP
-	  support is required for this driver to work.
-
-config DRM_SIS
-	tristate "SiS video cards"
-	depends on DRM && AGP
-	depends on FB_SIS || FB_SIS=n
-	help
-	  Choose this option if you have a SiS 630 or compatible video
-	  chipset. If M is selected the module will be called sis. AGP
-	  support is required for this driver to work.
+source "drivers/gpu/drm/panthor/Kconfig"
 
-config DRM_VIA
-	tristate "Via unichrome video cards"
-	depends on DRM && PCI
-	help
-	  Choose this option if you have a Via unichrome or compatible video
-	  chipset. If M is selected the module will be called via.
+source "drivers/gpu/drm/aspeed/Kconfig"
+
+source "drivers/gpu/drm/mcde/Kconfig"
+
+source "drivers/gpu/drm/tidss/Kconfig"
+
+source "drivers/gpu/drm/adp/Kconfig"
+
+source "drivers/gpu/drm/xlnx/Kconfig"
 
-config DRM_SAVAGE
-	tristate "Savage video cards"
-	depends on DRM && PCI
+source "drivers/gpu/drm/gud/Kconfig"
+
+source "drivers/gpu/drm/sitronix/Kconfig"
+
+source "drivers/gpu/drm/solomon/Kconfig"
+
+source "drivers/gpu/drm/sprd/Kconfig"
+
+source "drivers/gpu/drm/imagination/Kconfig"
+
+source "drivers/gpu/drm/tyr/Kconfig"
+
+config DRM_HYPERV
+	tristate "DRM Support for Hyper-V synthetic video device"
+	depends on DRM && PCI && HYPERV_VMBUS
+	select DRM_CLIENT_SELECTION
+	select DRM_KMS_HELPER
+	select DRM_GEM_SHMEM_HELPER
 	help
-	  Choose this option if you have a Savage3D/4/SuperSavage/Pro/Twister
-	  chipset. If M is selected the module will be called savage.
+	 This is a KMS driver for Hyper-V synthetic video device. Choose this
+	 option if you would like to enable drm driver for Hyper-V virtual
+	 machine. Unselect Hyper-V framebuffer driver (CONFIG_FB_HYPERV) so
+	 that DRM driver is used by default.
 
-endif # DRM_LEGACY
+	 If M is selected the module will be called hyperv_drm.
+
+# Separate option as not all DRM drivers use it
+config DRM_PANEL_BACKLIGHT_QUIRKS
+	tristate
 
 config DRM_LIB_RANDOM
 	bool
 	default n
+
+config DRM_PRIVACY_SCREEN
+	bool
+	default n
+
+endif
+
+# Separate option because drm_panel_orientation_quirks.c is shared with fbdev
+config DRM_PANEL_ORIENTATION_QUIRKS
+	tristate
diff --git a/drivers/gpu/drm/Kconfig.debug b/drivers/gpu/drm/Kconfig.debug
new file mode 100644
index 000000000000..05dc43c0b8c5
--- /dev/null
+++ b/drivers/gpu/drm/Kconfig.debug
@@ -0,0 +1,117 @@
+config DRM_USE_DYNAMIC_DEBUG
+	bool "use dynamic debug to implement drm.debug"
+	default n
+	depends on BROKEN
+	depends on DRM
+	depends on DYNAMIC_DEBUG || DYNAMIC_DEBUG_CORE
+	depends on JUMP_LABEL
+	help
+	 Use dynamic-debug to avoid drm_debug_enabled() runtime overheads.
+	 Due to callsite counts in DRM drivers (~4k in amdgpu) and 56
+	 bytes per callsite, the .data costs can be substantial, and
+	 are therefore configurable.
+
+config DRM_WERROR
+	bool "Compile the drm subsystem with warnings as errors"
+	depends on DRM && EXPERT
+	depends on !WERROR
+	default n
+	help
+	  A kernel build should not cause any compiler warnings, and this
+	  enables the '-Werror' flag to enforce that rule in the drm subsystem.
+
+	  The drm subsystem enables more warnings than the kernel default, so
+	  this config option is disabled by default.
+
+	  If in doubt, say N.
+
+config DRM_HEADER_TEST
+	bool "Ensure DRM headers are self-contained and pass kernel-doc"
+	depends on DRM && EXPERT && BROKEN
+	default n
+	help
+	  Ensure the DRM subsystem headers both under drivers/gpu/drm and
+	  include/drm compile, are self-contained, have header guards, and have
+	  no kernel-doc warnings.
+
+	  If in doubt, say N.
+
+config DRM_DEBUG_MM
+	bool "Insert extra checks and debug info into the DRM range managers"
+	default n
+	depends on DRM
+	depends on STACKTRACE_SUPPORT
+	select STACKDEPOT
+	help
+	  Enable allocation tracking of memory manager and leak detection on
+	  shutdown.
+
+	  Recommended for driver developers only.
+
+	  If in doubt, say "N".
+
+config DRM_KUNIT_TEST_HELPERS
+	tristate
+	depends on DRM && KUNIT
+	select DRM_KMS_HELPER
+	help
+	  KUnit Helpers for KMS drivers.
+
+config DRM_KUNIT_TEST
+	tristate "KUnit tests for DRM" if !KUNIT_ALL_TESTS
+	depends on DRM && KUNIT && MMU
+	select DRM_BRIDGE_CONNECTOR
+	select DRM_BUDDY
+	select DRM_DISPLAY_DP_HELPER
+	select DRM_DISPLAY_HDMI_STATE_HELPER
+	select DRM_DISPLAY_HELPER
+	select DRM_EXEC
+	select DRM_EXPORT_FOR_TESTS if m
+	select DRM_GEM_SHMEM_HELPER
+	select DRM_KUNIT_TEST_HELPERS
+	select DRM_LIB_RANDOM
+	select DRM_SYSFB_HELPER
+	select PRIME_NUMBERS
+	default KUNIT_ALL_TESTS
+	help
+	  This builds unit tests for DRM. This option is not useful for
+	  distributions or general kernels, but only for kernel
+	  developers working on DRM and associated drivers.
+
+	  For more information on KUnit and unit tests in general,
+	  please refer to the KUnit documentation in
+	  Documentation/dev-tools/kunit/.
+
+	  If in doubt, say "N".
+
+config DRM_TTM_KUNIT_TEST
+	tristate "KUnit tests for TTM" if !KUNIT_ALL_TESTS
+	default n
+	depends on DRM && KUNIT && MMU && (UML || COMPILE_TEST)
+	select DRM_TTM
+	select DRM_BUDDY
+	select DRM_EXPORT_FOR_TESTS if m
+	select DRM_KUNIT_TEST_HELPERS
+	default KUNIT_ALL_TESTS
+	help
+	  Enables unit tests for TTM, a GPU memory manager subsystem used
+	  to manage memory buffers. This option is mostly useful for kernel
+	  developers. It depends on (UML || COMPILE_TEST) since no other driver
+	  which uses TTM can be loaded while running the tests.
+
+	  If in doubt, say "N".
+
+config DRM_SCHED_KUNIT_TEST
+	tristate "KUnit tests for the DRM scheduler" if !KUNIT_ALL_TESTS
+	select DRM_SCHED
+	depends on DRM && KUNIT
+	default KUNIT_ALL_TESTS
+	help
+	  Choose this option to build unit tests for the DRM scheduler.
+
+	  Recommended for driver developers only.
+
+	  If in doubt, say "N".
+
+config DRM_EXPORT_FOR_TESTS
+	bool
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 59f0f9b696eb..0e1c668b46d2 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -1,70 +1,190 @@
+# SPDX-License-Identifier: GPL-2.0
 
 # Makefile for the drm device driver.  This driver provides support for the
 # Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher.
 
-drm-y       :=	drm_auth.o drm_bufs.o drm_cache.o \
-		drm_context.o drm_dma.o \
-		drm_file.o drm_gem.o drm_ioctl.o drm_irq.o \
-		drm_lock.o drm_memory.o drm_drv.o \
-		drm_scatter.o drm_pci.o \
-		drm_sysfs.o drm_hashtab.o drm_mm.o \
-		drm_crtc.o drm_fourcc.o drm_modes.o drm_edid.o \
-		drm_info.o drm_encoder_slave.o \
-		drm_trace_points.o drm_global.o drm_prime.o \
-		drm_rect.o drm_vma_manager.o drm_flip_work.o \
-		drm_modeset_lock.o drm_atomic.o drm_bridge.o \
-		drm_framebuffer.o drm_connector.o drm_blend.o \
-		drm_encoder.o drm_mode_object.o drm_property.o \
-		drm_plane.o drm_color_mgmt.o drm_print.o \
-		drm_dumb_buffers.o drm_mode_config.o
+CFLAGS-$(CONFIG_DRM_USE_DYNAMIC_DEBUG)	+= -DDYNAMIC_DEBUG_MODULE
 
+# Unconditionally enable W=1 warnings locally
+# --- begin copy-paste W=1 warnings from scripts/Makefile.warn
+subdir-ccflags-y += -Wextra -Wunused -Wno-unused-parameter
+subdir-ccflags-y += $(call cc-option, -Wrestrict)
+subdir-ccflags-y += -Wmissing-format-attribute
+subdir-ccflags-y += -Wold-style-definition
+subdir-ccflags-y += -Wmissing-include-dirs
+subdir-ccflags-y += $(call cc-option, -Wunused-but-set-variable)
+subdir-ccflags-y += $(call cc-option, -Wunused-const-variable)
+subdir-ccflags-y += $(call cc-option, -Wpacked-not-aligned)
+subdir-ccflags-y += $(call cc-option, -Wformat-overflow)
+# FIXME: fix -Wformat-truncation warnings and uncomment
+#subdir-ccflags-y += $(call cc-option, -Wformat-truncation)
+subdir-ccflags-y += $(call cc-option, -Wstringop-truncation)
+# The following turn off the warnings enabled by -Wextra
+ifeq ($(findstring 2, $(KBUILD_EXTRA_WARN)),)
+subdir-ccflags-y += -Wno-missing-field-initializers
+subdir-ccflags-y += -Wno-type-limits
+subdir-ccflags-y += -Wno-shift-negative-value
+endif
+ifeq ($(findstring 3, $(KBUILD_EXTRA_WARN)),)
+subdir-ccflags-y += -Wno-sign-compare
+endif
+# --- end copy-paste
+
+# Enable -Werror in CI and development
+subdir-ccflags-$(CONFIG_DRM_WERROR) += -Werror
+
+drm-y := \
+	drm_atomic.o \
+	drm_atomic_uapi.o \
+	drm_auth.o \
+	drm_blend.o \
+	drm_bridge.o \
+	drm_cache.o \
+	drm_color_mgmt.o \
+	drm_colorop.o \
+	drm_connector.o \
+	drm_crtc.o \
+	drm_displayid.o \
+	drm_drv.o \
+	drm_dumb_buffers.o \
+	drm_edid.o \
+	drm_eld.o \
+	drm_encoder.o \
+	drm_file.o \
+	drm_fourcc.o \
+	drm_framebuffer.o \
+	drm_gem.o \
+	drm_ioctl.o \
+	drm_lease.o \
+	drm_managed.o \
+	drm_mm.o \
+	drm_mode_config.o \
+	drm_mode_object.o \
+	drm_modes.o \
+	drm_modeset_lock.o \
+	drm_plane.o \
+	drm_prime.o \
+	drm_print.o \
+	drm_property.o \
+	drm_rect.o \
+	drm_syncobj.o \
+	drm_sysfs.o \
+	drm_trace_points.o \
+	drm_vblank.o \
+	drm_vblank_work.o \
+	drm_vma_manager.o \
+	drm_writeback.o
+drm-$(CONFIG_DRM_CLIENT) += \
+	drm_client.o \
+	drm_client_event.o \
+	drm_client_modeset.o \
+	drm_client_sysrq.o
 drm-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o
-drm-$(CONFIG_DRM_VM) += drm_vm.o
 drm-$(CONFIG_COMPAT) += drm_ioc32.o
-drm-$(CONFIG_DRM_GEM_CMA_HELPER) += drm_gem_cma_helper.o
-drm-$(CONFIG_PCI) += ati_pcigart.o
 drm-$(CONFIG_DRM_PANEL) += drm_panel.o
 drm-$(CONFIG_OF) += drm_of.o
-drm-$(CONFIG_AGP) += drm_agpsupport.o
-drm-$(CONFIG_DEBUG_FS) += drm_debugfs.o drm_debugfs_crc.o
+drm-$(CONFIG_PCI) += drm_pci.o
+drm-$(CONFIG_DEBUG_FS) += \
+	drm_debugfs.o \
+	drm_debugfs_crc.o
+drm-$(CONFIG_DRM_LOAD_EDID_FIRMWARE) += drm_edid_load.o
+drm-$(CONFIG_DRM_PRIVACY_SCREEN) += \
+	drm_privacy_screen.o \
+	drm_privacy_screen_x86.o
+drm-$(CONFIG_DRM_ACCEL) += ../../accel/drm_accel.o
+drm-$(CONFIG_DRM_PANIC) += drm_panic.o
+drm-$(CONFIG_DRM_DRAW) += drm_draw.o
+drm-$(CONFIG_DRM_PANIC_SCREEN_QR_CODE) += drm_panic_qr.o
+obj-$(CONFIG_DRM)	+= drm.o
 
-drm_kms_helper-y := drm_crtc_helper.o drm_dp_helper.o drm_probe_helper.o \
-		drm_plane_helper.o drm_dp_mst_topology.o drm_atomic_helper.o \
-		drm_kms_helper_common.o drm_dp_dual_mode_helper.o \
-		drm_simple_kms_helper.o drm_modeset_helper.o \
-		drm_scdc_helper.o
+obj-$(CONFIG_DRM_PANEL_ORIENTATION_QUIRKS) += drm_panel_orientation_quirks.o
+obj-$(CONFIG_DRM_PANEL_BACKLIGHT_QUIRKS) += drm_panel_backlight_quirks.o
 
-drm_kms_helper-$(CONFIG_DRM_LOAD_EDID_FIRMWARE) += drm_edid_load.o
-drm_kms_helper-$(CONFIG_DRM_FBDEV_EMULATION) += drm_fb_helper.o
-drm_kms_helper-$(CONFIG_DRM_KMS_CMA_HELPER) += drm_fb_cma_helper.o
-drm_kms_helper-$(CONFIG_DRM_DP_AUX_CHARDEV) += drm_dp_aux_dev.o
+#
+# Memory-management helpers
+#
+#
+obj-$(CONFIG_DRM_EXEC) += drm_exec.o
+obj-$(CONFIG_DRM_GPUVM) += drm_gpuvm.o
+
+drm_gpusvm_helper-y := \
+	drm_gpusvm.o\
+	drm_pagemap.o
+obj-$(CONFIG_DRM_GPUSVM) += drm_gpusvm_helper.o
+
+obj-$(CONFIG_DRM_BUDDY) += drm_buddy.o
+
+drm_dma_helper-y := drm_gem_dma_helper.o
+drm_dma_helper-$(CONFIG_DRM_FBDEV_EMULATION) += drm_fbdev_dma.o
+drm_dma_helper-$(CONFIG_DRM_KMS_HELPER) += drm_fb_dma_helper.o
+obj-$(CONFIG_DRM_GEM_DMA_HELPER) += drm_dma_helper.o
+
+drm_shmem_helper-y := drm_gem_shmem_helper.o
+drm_shmem_helper-$(CONFIG_DRM_FBDEV_EMULATION) += drm_fbdev_shmem.o
+obj-$(CONFIG_DRM_GEM_SHMEM_HELPER) += drm_shmem_helper.o
+
+drm_suballoc_helper-y := drm_suballoc.o
+obj-$(CONFIG_DRM_SUBALLOC_HELPER) += drm_suballoc_helper.o
+
+drm_vram_helper-y := drm_gem_vram_helper.o
+obj-$(CONFIG_DRM_VRAM_HELPER) += drm_vram_helper.o
 
+drm_ttm_helper-y := drm_gem_ttm_helper.o
+drm_ttm_helper-$(CONFIG_DRM_FBDEV_EMULATION) += drm_fbdev_ttm.o
+obj-$(CONFIG_DRM_TTM_HELPER) += drm_ttm_helper.o
+
+#
+# Modesetting helpers
+#
+
+drm_kms_helper-y := \
+	drm_atomic_helper.o \
+	drm_atomic_state_helper.o \
+	drm_bridge_helper.o \
+	drm_crtc_helper.o \
+	drm_damage_helper.o \
+	drm_flip_work.o \
+	drm_format_helper.o \
+	drm_gem_atomic_helper.o \
+	drm_gem_framebuffer_helper.o \
+	drm_kms_helper_common.o \
+	drm_modeset_helper.o \
+	drm_plane_helper.o \
+	drm_probe_helper.o \
+	drm_self_refresh_helper.o \
+	drm_simple_kms_helper.o \
+	drm_vblank_helper.o
+drm_kms_helper-$(CONFIG_DRM_PANEL_BRIDGE) += bridge/panel.o
+drm_kms_helper-$(CONFIG_DRM_FBDEV_EMULATION) += drm_fb_helper.o
 obj-$(CONFIG_DRM_KMS_HELPER) += drm_kms_helper.o
-obj-$(CONFIG_DRM_DEBUG_MM_SELFTEST) += selftests/
 
-CFLAGS_drm_trace_points.o := -I$(src)
+#
+# Drivers and the rest
+#
 
-obj-$(CONFIG_DRM)	+= drm.o
+obj-y			+= tests/
+
+obj-$(CONFIG_DRM_MIPI_DBI) += drm_mipi_dbi.o
 obj-$(CONFIG_DRM_MIPI_DSI) += drm_mipi_dsi.o
-obj-$(CONFIG_DRM_ARM)	+= arm/
+obj-y			+= arm/
+obj-y			+= clients/
+obj-y			+= display/
 obj-$(CONFIG_DRM_TTM)	+= ttm/
-obj-$(CONFIG_DRM_TDFX)	+= tdfx/
-obj-$(CONFIG_DRM_R128)	+= r128/
-obj-$(CONFIG_HSA_AMD) += amd/amdkfd/
+obj-$(CONFIG_DRM_SCHED)	+= scheduler/
 obj-$(CONFIG_DRM_RADEON)+= radeon/
 obj-$(CONFIG_DRM_AMDGPU)+= amd/amdgpu/
-obj-$(CONFIG_DRM_MGA)	+= mga/
-obj-$(CONFIG_DRM_I810)	+= i810/
+obj-$(CONFIG_DRM_AMDGPU)+= amd/amdxcp/
 obj-$(CONFIG_DRM_I915)	+= i915/
+obj-$(CONFIG_DRM_XE)	+= xe/
+obj-$(CONFIG_DRM_KMB_DISPLAY)  += kmb/
 obj-$(CONFIG_DRM_MGAG200) += mgag200/
+obj-$(CONFIG_DRM_V3D)  += v3d/
 obj-$(CONFIG_DRM_VC4)  += vc4/
-obj-$(CONFIG_DRM_CIRRUS_QEMU) += cirrus/
-obj-$(CONFIG_DRM_SIS)   += sis/
-obj-$(CONFIG_DRM_SAVAGE)+= savage/
 obj-$(CONFIG_DRM_VMWGFX)+= vmwgfx/
-obj-$(CONFIG_DRM_VIA)	+=via/
 obj-$(CONFIG_DRM_VGEM)	+= vgem/
+obj-$(CONFIG_DRM_VKMS)	+= vkms/
 obj-$(CONFIG_DRM_NOUVEAU) +=nouveau/
+obj-$(CONFIG_DRM_NOVA) += nova/
 obj-$(CONFIG_DRM_EXYNOS) +=exynos/
 obj-$(CONFIG_DRM_ROCKCHIP) +=rockchip/
 obj-$(CONFIG_DRM_GMA500) += gma500/
@@ -72,27 +192,64 @@ obj-$(CONFIG_DRM_UDL) += udl/
 obj-$(CONFIG_DRM_AST) += ast/
 obj-$(CONFIG_DRM_ARMADA) += armada/
 obj-$(CONFIG_DRM_ATMEL_HLCDC)	+= atmel-hlcdc/
-obj-$(CONFIG_DRM_RCAR_DU) += rcar-du/
-obj-$(CONFIG_DRM_SHMOBILE) +=shmobile/
+obj-y			+= renesas/
 obj-y			+= omapdrm/
 obj-$(CONFIG_DRM_SUN4I) += sun4i/
 obj-y			+= tilcdc/
 obj-$(CONFIG_DRM_QXL) += qxl/
-obj-$(CONFIG_DRM_BOCHS) += bochs/
 obj-$(CONFIG_DRM_VIRTIO_GPU) += virtio/
 obj-$(CONFIG_DRM_MSM) += msm/
 obj-$(CONFIG_DRM_TEGRA) += tegra/
+obj-$(CONFIG_DRM_STM) += stm/
 obj-$(CONFIG_DRM_STI) += sti/
-obj-$(CONFIG_DRM_IMX) += imx/
+obj-y 			+= imx/
+obj-$(CONFIG_DRM_INGENIC) += ingenic/
+obj-$(CONFIG_DRM_LOGICVC) += logicvc/
 obj-$(CONFIG_DRM_MEDIATEK) += mediatek/
 obj-$(CONFIG_DRM_MESON)	+= meson/
-obj-y			+= i2c/
 obj-y			+= panel/
 obj-y			+= bridge/
 obj-$(CONFIG_DRM_FSL_DCU) += fsl-dcu/
 obj-$(CONFIG_DRM_ETNAVIV) += etnaviv/
-obj-$(CONFIG_DRM_ARCPGU)+= arc/
 obj-y			+= hisilicon/
-obj-$(CONFIG_DRM_ZTE)	+= zte/
-obj-$(CONFIG_DRM_MXSFB)	+= mxsfb/
-obj-$(CONFIG_DRM_TINYDRM) += tinydrm/
+obj-y			+= mxsfb/
+obj-y			+= sysfb/
+obj-y			+= tiny/
+obj-$(CONFIG_DRM_PL111) += pl111/
+obj-$(CONFIG_DRM_TVE200) += tve200/
+obj-$(CONFIG_DRM_ADP) += adp/
+obj-$(CONFIG_DRM_XEN) += xen/
+obj-$(CONFIG_DRM_VBOXVIDEO) += vboxvideo/
+obj-$(CONFIG_DRM_LIMA)  += lima/
+obj-$(CONFIG_DRM_PANFROST) += panfrost/
+obj-$(CONFIG_DRM_PANTHOR) += panthor/
+obj-$(CONFIG_DRM_TYR) += tyr/
+obj-$(CONFIG_DRM_ASPEED_GFX) += aspeed/
+obj-$(CONFIG_DRM_MCDE) += mcde/
+obj-$(CONFIG_DRM_TIDSS) += tidss/
+obj-y			+= xlnx/
+obj-y			+= gud/
+obj-$(CONFIG_DRM_HYPERV) += hyperv/
+obj-y			+= sitronix/
+obj-y			+= solomon/
+obj-$(CONFIG_DRM_SPRD) += sprd/
+obj-$(CONFIG_DRM_LOONGSON) += loongson/
+obj-$(CONFIG_DRM_POWERVR) += imagination/
+
+# Ensure drm headers are self-contained and pass kernel-doc
+hdrtest-files := \
+	$(shell cd $(src) && find . -maxdepth 1 -name 'drm_*.h') \
+	$(shell cd $(src) && find display lib -name '*.h')
+
+always-$(CONFIG_DRM_HEADER_TEST) += \
+	$(patsubst %.h,%.hdrtest, $(hdrtest-files))
+
+# Include the header twice to detect missing include guard.
+quiet_cmd_hdrtest = HDRTEST $(patsubst %.hdrtest,%.h,$@)
+      cmd_hdrtest = \
+		$(CC) $(c_flags) -fsyntax-only -x c /dev/null -include $< -include $<; \
+		 PYTHONDONTWRITEBYTECODE=1 $(PYTHON3) $(KERNELDOC) -none $(if $(CONFIG_WERROR)$(CONFIG_DRM_WERROR),-Werror) $<; \
+		touch $@
+
+$(obj)/%.hdrtest: $(src)/%.h FORCE
+	$(call if_changed_dep,hdrtest)
diff --git a/drivers/gpu/drm/adp/Kconfig b/drivers/gpu/drm/adp/Kconfig
new file mode 100644
index 000000000000..9fcc27eb200d
--- /dev/null
+++ b/drivers/gpu/drm/adp/Kconfig
@@ -0,0 +1,17 @@
+# SPDX-License-Identifier: GPL-2.0-only OR MIT
+config DRM_ADP
+	tristate "DRM Support for pre-DCP Apple display controllers"
+	depends on DRM && OF && ARM64
+	depends on ARCH_APPLE || COMPILE_TEST
+	select DRM_KMS_HELPER
+	select DRM_BRIDGE_CONNECTOR
+	select DRM_DISPLAY_HELPER
+	select DRM_KMS_DMA_HELPER
+	select DRM_GEM_DMA_HELPER
+	select DRM_PANEL_BRIDGE
+	select VIDEOMODE_HELPERS
+	select DRM_MIPI_DSI
+	help
+	  Chose this option if you have an Apple Arm laptop with a touchbar.
+
+	  If M is selected, this module will be called adpdrm.
diff --git a/drivers/gpu/drm/adp/Makefile b/drivers/gpu/drm/adp/Makefile
new file mode 100644
index 000000000000..8e7b618edd35
--- /dev/null
+++ b/drivers/gpu/drm/adp/Makefile
@@ -0,0 +1,5 @@
+# SPDX-License-Identifier: GPL-2.0-only OR MIT
+
+adpdrm-y := adp_drv.o
+adpdrm-mipi-y := adp-mipi.o
+obj-$(CONFIG_DRM_ADP) += adpdrm.o adpdrm-mipi.o
diff --git a/drivers/gpu/drm/adp/adp-mipi.c b/drivers/gpu/drm/adp/adp-mipi.c
new file mode 100644
index 000000000000..cba7d32150a9
--- /dev/null
+++ b/drivers/gpu/drm/adp/adp-mipi.c
@@ -0,0 +1,277 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#include <linux/component.h>
+#include <linux/iopoll.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+
+#include <drm/drm_bridge.h>
+#include <drm/drm_mipi_dsi.h>
+
+#define DSI_GEN_HDR 0x6c
+#define DSI_GEN_PLD_DATA 0x70
+
+#define DSI_CMD_PKT_STATUS 0x74
+
+#define GEN_PLD_R_EMPTY BIT(4)
+#define GEN_PLD_W_FULL BIT(3)
+#define GEN_PLD_W_EMPTY BIT(2)
+#define GEN_CMD_FULL BIT(1)
+#define GEN_CMD_EMPTY BIT(0)
+#define GEN_RD_CMD_BUSY BIT(6)
+#define CMD_PKT_STATUS_TIMEOUT_US 20000
+
+struct adp_mipi_drv_private {
+	struct mipi_dsi_host dsi;
+	struct drm_bridge bridge;
+	struct drm_bridge *next_bridge;
+	void __iomem *mipi;
+};
+
+#define mipi_to_adp(x) container_of(x, struct adp_mipi_drv_private, dsi)
+
+static int adp_dsi_gen_pkt_hdr_write(struct adp_mipi_drv_private *adp, u32 hdr_val)
+{
+	int ret;
+	u32 val, mask;
+
+	ret = readl_poll_timeout(adp->mipi + DSI_CMD_PKT_STATUS,
+				 val, !(val & GEN_CMD_FULL), 1000,
+				 CMD_PKT_STATUS_TIMEOUT_US);
+	if (ret) {
+		dev_err(adp->dsi.dev, "failed to get available command FIFO\n");
+		return ret;
+	}
+
+	writel(hdr_val, adp->mipi + DSI_GEN_HDR);
+
+	mask = GEN_CMD_EMPTY | GEN_PLD_W_EMPTY;
+	ret = readl_poll_timeout(adp->mipi + DSI_CMD_PKT_STATUS,
+				 val, (val & mask) == mask,
+				 1000, CMD_PKT_STATUS_TIMEOUT_US);
+	if (ret) {
+		dev_err(adp->dsi.dev, "failed to write command FIFO\n");
+		return ret;
+	}
+
+	return 0;
+}
+
+static int adp_dsi_write(struct adp_mipi_drv_private *adp,
+			 const struct mipi_dsi_packet *packet)
+{
+	const u8 *tx_buf = packet->payload;
+	int len = packet->payload_length, pld_data_bytes = sizeof(u32), ret;
+	__le32 word;
+	u32 val;
+
+	while (len) {
+		if (len < pld_data_bytes) {
+			word = 0;
+			memcpy(&word, tx_buf, len);
+			writel(le32_to_cpu(word), adp->mipi + DSI_GEN_PLD_DATA);
+			len = 0;
+		} else {
+			memcpy(&word, tx_buf, pld_data_bytes);
+			writel(le32_to_cpu(word), adp->mipi + DSI_GEN_PLD_DATA);
+			tx_buf += pld_data_bytes;
+			len -= pld_data_bytes;
+		}
+
+		ret = readl_poll_timeout(adp->mipi + DSI_CMD_PKT_STATUS,
+					 val, !(val & GEN_PLD_W_FULL), 1000,
+					 CMD_PKT_STATUS_TIMEOUT_US);
+		if (ret) {
+			dev_err(adp->dsi.dev,
+				"failed to get available write payload FIFO\n");
+			return ret;
+		}
+	}
+
+	word = 0;
+	memcpy(&word, packet->header, sizeof(packet->header));
+	return adp_dsi_gen_pkt_hdr_write(adp, le32_to_cpu(word));
+}
+
+static int adp_dsi_read(struct adp_mipi_drv_private *adp,
+			const struct mipi_dsi_msg *msg)
+{
+	int i, j, ret, len = msg->rx_len;
+	u8 *buf = msg->rx_buf;
+	u32 val;
+
+	/* Wait end of the read operation */
+	ret = readl_poll_timeout(adp->mipi + DSI_CMD_PKT_STATUS,
+				 val, !(val & GEN_RD_CMD_BUSY),
+				 1000, CMD_PKT_STATUS_TIMEOUT_US);
+	if (ret) {
+		dev_err(adp->dsi.dev, "Timeout during read operation\n");
+		return ret;
+	}
+
+	for (i = 0; i < len; i += 4) {
+		/* Read fifo must not be empty before all bytes are read */
+		ret = readl_poll_timeout(adp->mipi + DSI_CMD_PKT_STATUS,
+					 val, !(val & GEN_PLD_R_EMPTY),
+					 1000, CMD_PKT_STATUS_TIMEOUT_US);
+		if (ret) {
+			dev_err(adp->dsi.dev, "Read payload FIFO is empty\n");
+			return ret;
+		}
+
+		val = readl(adp->mipi + DSI_GEN_PLD_DATA);
+		for (j = 0; j < 4 && j + i < len; j++)
+			buf[i + j] = val >> (8 * j);
+	}
+
+	return ret;
+}
+
+static ssize_t adp_dsi_host_transfer(struct mipi_dsi_host *host,
+				     const struct mipi_dsi_msg *msg)
+{
+	struct adp_mipi_drv_private *adp = mipi_to_adp(host);
+	struct mipi_dsi_packet packet;
+	int ret, nb_bytes;
+
+	ret = mipi_dsi_create_packet(&packet, msg);
+	if (ret) {
+		dev_err(adp->dsi.dev, "failed to create packet: %d\n", ret);
+		return ret;
+	}
+
+	ret = adp_dsi_write(adp, &packet);
+	if (ret)
+		return ret;
+
+	if (msg->rx_buf && msg->rx_len) {
+		ret = adp_dsi_read(adp, msg);
+		if (ret)
+			return ret;
+		nb_bytes = msg->rx_len;
+	} else {
+		nb_bytes = packet.size;
+	}
+
+	return nb_bytes;
+}
+
+static int adp_dsi_bind(struct device *dev, struct device *master, void *data)
+{
+	return 0;
+}
+
+static void adp_dsi_unbind(struct device *dev, struct device *master, void *data)
+{
+}
+
+static const struct component_ops adp_dsi_component_ops = {
+	.bind	= adp_dsi_bind,
+	.unbind	= adp_dsi_unbind,
+};
+
+static int adp_dsi_host_attach(struct mipi_dsi_host *host,
+			       struct mipi_dsi_device *dev)
+{
+	struct adp_mipi_drv_private *adp = mipi_to_adp(host);
+	struct drm_bridge *next;
+	int ret;
+
+	next = devm_drm_of_get_bridge(adp->dsi.dev, adp->dsi.dev->of_node, 1, 0);
+	if (IS_ERR(next))
+		return PTR_ERR(next);
+
+	adp->next_bridge = next;
+
+	drm_bridge_add(&adp->bridge);
+
+	ret = component_add(host->dev, &adp_dsi_component_ops);
+	if (ret) {
+		pr_err("failed to add dsi_host component: %d\n", ret);
+		drm_bridge_remove(&adp->bridge);
+		return ret;
+	}
+
+	return 0;
+}
+
+static int adp_dsi_host_detach(struct mipi_dsi_host *host,
+			       struct mipi_dsi_device *dev)
+{
+	struct adp_mipi_drv_private *adp = mipi_to_adp(host);
+
+	component_del(host->dev, &adp_dsi_component_ops);
+	drm_bridge_remove(&adp->bridge);
+	return 0;
+}
+
+static const struct mipi_dsi_host_ops adp_dsi_host_ops = {
+	.transfer = adp_dsi_host_transfer,
+	.attach = adp_dsi_host_attach,
+	.detach = adp_dsi_host_detach,
+};
+
+static int adp_dsi_bridge_attach(struct drm_bridge *bridge,
+				 struct drm_encoder *encoder,
+				 enum drm_bridge_attach_flags flags)
+{
+	struct adp_mipi_drv_private *adp =
+		container_of(bridge, struct adp_mipi_drv_private, bridge);
+
+	return drm_bridge_attach(encoder, adp->next_bridge, bridge, flags);
+}
+
+static const struct drm_bridge_funcs adp_dsi_bridge_funcs = {
+	.attach	= adp_dsi_bridge_attach,
+};
+
+static int adp_mipi_probe(struct platform_device *pdev)
+{
+	struct adp_mipi_drv_private *adp;
+
+	adp = devm_drm_bridge_alloc(&pdev->dev, struct adp_mipi_drv_private,
+				    bridge, &adp_dsi_bridge_funcs);
+	if (IS_ERR(adp))
+		return PTR_ERR(adp);
+
+	adp->mipi = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(adp->mipi)) {
+		dev_err(&pdev->dev, "failed to map mipi mmio");
+		return PTR_ERR(adp->mipi);
+	}
+
+	adp->dsi.dev = &pdev->dev;
+	adp->dsi.ops = &adp_dsi_host_ops;
+	adp->bridge.of_node = pdev->dev.of_node;
+	adp->bridge.type = DRM_MODE_CONNECTOR_DSI;
+	dev_set_drvdata(&pdev->dev, adp);
+	return mipi_dsi_host_register(&adp->dsi);
+}
+
+static void adp_mipi_remove(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct adp_mipi_drv_private *adp = dev_get_drvdata(dev);
+
+	mipi_dsi_host_unregister(&adp->dsi);
+}
+
+static const struct of_device_id adp_mipi_of_match[] = {
+	{ .compatible = "apple,h7-display-pipe-mipi", },
+	{ },
+};
+MODULE_DEVICE_TABLE(of, adp_mipi_of_match);
+
+static struct platform_driver adp_mipi_platform_driver = {
+	.driver = {
+		.name = "adp-mipi",
+		.of_match_table = adp_mipi_of_match,
+	},
+	.probe = adp_mipi_probe,
+	.remove = adp_mipi_remove,
+};
+
+module_platform_driver(adp_mipi_platform_driver);
+
+MODULE_DESCRIPTION("Apple Display Pipe MIPI driver");
+MODULE_LICENSE("GPL");
diff --git a/drivers/gpu/drm/adp/adp_drv.c b/drivers/gpu/drm/adp/adp_drv.c
new file mode 100644
index 000000000000..4554cf75565e
--- /dev/null
+++ b/drivers/gpu/drm/adp/adp_drv.c
@@ -0,0 +1,614 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#include <linux/component.h>
+#include <linux/iopoll.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+
+#include <drm/drm_atomic.h>
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_bridge.h>
+#include <drm/drm_bridge_connector.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_fb_dma_helper.h>
+#include <drm/drm_framebuffer.h>
+#include <drm/drm_gem_atomic_helper.h>
+#include <drm/drm_gem_dma_helper.h>
+#include <drm/drm_gem_framebuffer_helper.h>
+#include <drm/drm_of.h>
+#include <drm/drm_print.h>
+#include <drm/drm_probe_helper.h>
+#include <drm/drm_vblank.h>
+
+#define ADP_INT_STATUS 0x34
+#define ADP_INT_STATUS_INT_MASK 0x7
+#define ADP_INT_STATUS_VBLANK 0x1
+#define ADP_CTRL 0x100
+#define ADP_CTRL_VBLANK_ON 0x12
+#define ADP_CTRL_FIFO_ON 0x601
+#define ADP_SCREEN_SIZE 0x0c
+#define ADP_SCREEN_HSIZE GENMASK(15, 0)
+#define ADP_SCREEN_VSIZE GENMASK(31, 16)
+
+#define ADBE_FIFO 0x10c0
+#define ADBE_FIFO_SYNC 0xc0000000
+
+#define ADBE_BLEND_BYPASS 0x2020
+#define ADBE_BLEND_EN1 0x2028
+#define ADBE_BLEND_EN2 0x2074
+#define ADBE_BLEND_EN3 0x202c
+#define ADBE_BLEND_EN4 0x2034
+#define ADBE_MASK_BUF 0x2200
+
+#define ADBE_SRC_START 0x4040
+#define ADBE_SRC_SIZE 0x4048
+#define ADBE_DST_START 0x4050
+#define ADBE_DST_SIZE 0x4054
+#define ADBE_STRIDE 0x4038
+#define ADBE_FB_BASE 0x4030
+
+#define ADBE_LAYER_EN1 0x4020
+#define ADBE_LAYER_EN2 0x4068
+#define ADBE_LAYER_EN3 0x40b4
+#define ADBE_LAYER_EN4 0x40f4
+#define ADBE_SCALE_CTL 0x40ac
+#define ADBE_SCALE_CTL_BYPASS 0x100000
+
+#define ADBE_LAYER_CTL 0x1038
+#define ADBE_LAYER_CTL_ENABLE 0x10000
+
+#define ADBE_PIX_FMT 0x402c
+#define ADBE_PIX_FMT_XRGB32 0x53e4001
+
+static int adp_open(struct inode *inode, struct file *filp)
+{
+	/*
+	 * The modesetting driver does not check the non-desktop connector
+	 * property and keeps the device open and locked. If the touchbar daemon
+	 * opens the device first, modesetting breaks the whole X session.
+	 * Simply refuse to open the device for X11 server processes as
+	 * workaround.
+	 */
+	if (current->comm[0] == 'X')
+		return -EBUSY;
+
+	return drm_open(inode, filp);
+}
+
+static const struct file_operations adp_fops = {
+	.owner          = THIS_MODULE,
+	.open           = adp_open,
+	.release        = drm_release,
+	.unlocked_ioctl = drm_ioctl,
+	.compat_ioctl   = drm_compat_ioctl,
+	.poll           = drm_poll,
+	.read           = drm_read,
+	.llseek         = noop_llseek,
+	.mmap           = drm_gem_mmap,
+	.fop_flags      = FOP_UNSIGNED_OFFSET,
+	DRM_GEM_DMA_UNMAPPED_AREA_FOPS
+};
+
+static int adp_drm_gem_dumb_create(struct drm_file *file_priv,
+					struct drm_device *drm,
+					struct drm_mode_create_dumb *args)
+{
+	args->height = ALIGN(args->height, 64);
+	args->size = args->pitch * args->height;
+
+	return drm_gem_dma_dumb_create_internal(file_priv, drm, args);
+}
+
+static const struct drm_driver adp_driver = {
+	.driver_features = DRIVER_GEM | DRIVER_MODESET | DRIVER_ATOMIC,
+	.fops = &adp_fops,
+	DRM_GEM_DMA_DRIVER_OPS_VMAP_WITH_DUMB_CREATE(adp_drm_gem_dumb_create),
+	.name = "adp",
+	.desc = "Apple Display Pipe DRM Driver",
+	.major = 0,
+	.minor = 1,
+};
+
+struct adp_drv_private {
+	struct drm_device drm;
+	struct drm_crtc crtc;
+	struct drm_encoder *encoder;
+	struct drm_connector *connector;
+	struct drm_bridge *next_bridge;
+	void __iomem *be;
+	void __iomem *fe;
+	u32 *mask_buf;
+	u64 mask_buf_size;
+	dma_addr_t mask_iova;
+	int be_irq;
+	int fe_irq;
+	struct drm_pending_vblank_event *event;
+};
+
+#define to_adp(x) container_of(x, struct adp_drv_private, drm)
+#define crtc_to_adp(x) container_of(x, struct adp_drv_private, crtc)
+
+static int adp_plane_atomic_check(struct drm_plane *plane,
+				    struct drm_atomic_state *state)
+{
+	struct drm_plane_state *new_plane_state;
+	struct drm_crtc_state *crtc_state;
+
+	new_plane_state = drm_atomic_get_new_plane_state(state, plane);
+
+	if (!new_plane_state->crtc)
+		return 0;
+
+	crtc_state = drm_atomic_get_crtc_state(state, new_plane_state->crtc);
+	if (IS_ERR(crtc_state))
+		return PTR_ERR(crtc_state);
+
+	return drm_atomic_helper_check_plane_state(new_plane_state,
+						   crtc_state,
+						   DRM_PLANE_NO_SCALING,
+						   DRM_PLANE_NO_SCALING,
+						   true, true);
+}
+
+static void adp_plane_atomic_update(struct drm_plane *plane,
+				    struct drm_atomic_state *state)
+{
+	struct adp_drv_private *adp;
+	struct drm_rect src_rect;
+	struct drm_gem_dma_object *obj;
+	struct drm_framebuffer *fb;
+	struct drm_plane_state *new_state = drm_atomic_get_new_plane_state(state, plane);
+	u32 src_pos, src_size, dst_pos, dst_size;
+
+	if (!plane || !new_state)
+		return;
+
+	fb = new_state->fb;
+	if (!fb)
+		return;
+	adp = to_adp(plane->dev);
+
+	drm_rect_fp_to_int(&src_rect, &new_state->src);
+	src_pos = src_rect.x1 << 16 | src_rect.y1;
+	dst_pos = new_state->dst.x1 << 16 | new_state->dst.y1;
+	src_size = drm_rect_width(&src_rect) << 16 | drm_rect_height(&src_rect);
+	dst_size = drm_rect_width(&new_state->dst) << 16 |
+		drm_rect_height(&new_state->dst);
+	writel(src_pos, adp->be + ADBE_SRC_START);
+	writel(src_size, adp->be + ADBE_SRC_SIZE);
+	writel(dst_pos, adp->be + ADBE_DST_START);
+	writel(dst_size, adp->be + ADBE_DST_SIZE);
+	writel(fb->pitches[0], adp->be + ADBE_STRIDE);
+	obj = drm_fb_dma_get_gem_obj(fb, 0);
+	if (obj)
+		writel(obj->dma_addr + fb->offsets[0], adp->be + ADBE_FB_BASE);
+
+	writel(BIT(0), adp->be + ADBE_LAYER_EN1);
+	writel(BIT(0), adp->be + ADBE_LAYER_EN2);
+	writel(BIT(0), adp->be + ADBE_LAYER_EN3);
+	writel(BIT(0), adp->be + ADBE_LAYER_EN4);
+	writel(ADBE_SCALE_CTL_BYPASS, adp->be + ADBE_SCALE_CTL);
+	writel(ADBE_LAYER_CTL_ENABLE | BIT(0), adp->be + ADBE_LAYER_CTL);
+	writel(ADBE_PIX_FMT_XRGB32, adp->be + ADBE_PIX_FMT);
+}
+
+static void adp_plane_atomic_disable(struct drm_plane *plane,
+				     struct drm_atomic_state *state)
+{
+	struct adp_drv_private *adp = to_adp(plane->dev);
+
+	writel(0x0, adp->be + ADBE_LAYER_EN1);
+	writel(0x0, adp->be + ADBE_LAYER_EN2);
+	writel(0x0, adp->be + ADBE_LAYER_EN3);
+	writel(0x0, adp->be + ADBE_LAYER_EN4);
+	writel(ADBE_LAYER_CTL_ENABLE, adp->be + ADBE_LAYER_CTL);
+}
+
+static const struct drm_plane_helper_funcs adp_plane_helper_funcs = {
+	.atomic_check = adp_plane_atomic_check,
+	.atomic_update = adp_plane_atomic_update,
+	.atomic_disable = adp_plane_atomic_disable,
+	DRM_GEM_SHADOW_PLANE_HELPER_FUNCS
+};
+
+static const struct drm_plane_funcs adp_plane_funcs = {
+	.update_plane = drm_atomic_helper_update_plane,
+	.disable_plane = drm_atomic_helper_disable_plane,
+	DRM_GEM_SHADOW_PLANE_FUNCS
+};
+
+static const u32 plane_formats[] = {
+	DRM_FORMAT_XRGB8888,
+};
+
+#define ALL_CRTCS 1
+
+static struct drm_plane *adp_plane_new(struct adp_drv_private *adp)
+{
+	struct drm_device *drm = &adp->drm;
+	struct drm_plane *plane;
+
+	plane = __drmm_universal_plane_alloc(drm, sizeof(struct drm_plane), 0,
+					     ALL_CRTCS, &adp_plane_funcs,
+					     plane_formats, ARRAY_SIZE(plane_formats),
+					     NULL, DRM_PLANE_TYPE_PRIMARY, "plane");
+	if (IS_ERR(plane)) {
+		drm_err(drm, "failed to allocate plane");
+		return plane;
+	}
+
+	drm_plane_helper_add(plane, &adp_plane_helper_funcs);
+	return plane;
+}
+
+static void adp_enable_vblank(struct adp_drv_private *adp)
+{
+	u32 cur_ctrl;
+
+	writel(ADP_INT_STATUS_INT_MASK, adp->fe + ADP_INT_STATUS);
+
+	cur_ctrl = readl(adp->fe + ADP_CTRL);
+	writel(cur_ctrl | ADP_CTRL_VBLANK_ON, adp->fe + ADP_CTRL);
+}
+
+static int adp_crtc_enable_vblank(struct drm_crtc *crtc)
+{
+	struct drm_device *dev = crtc->dev;
+	struct adp_drv_private *adp = to_adp(dev);
+
+	adp_enable_vblank(adp);
+
+	return 0;
+}
+
+static void adp_disable_vblank(struct adp_drv_private *adp)
+{
+	u32 cur_ctrl;
+
+	cur_ctrl = readl(adp->fe + ADP_CTRL);
+	writel(cur_ctrl & ~ADP_CTRL_VBLANK_ON, adp->fe + ADP_CTRL);
+	writel(ADP_INT_STATUS_INT_MASK, adp->fe + ADP_INT_STATUS);
+}
+
+static void adp_crtc_disable_vblank(struct drm_crtc *crtc)
+{
+	struct drm_device *dev = crtc->dev;
+	struct adp_drv_private *adp = to_adp(dev);
+
+	adp_disable_vblank(adp);
+}
+
+static void adp_crtc_atomic_enable(struct drm_crtc *crtc,
+				   struct drm_atomic_state *state)
+{
+	struct adp_drv_private *adp = crtc_to_adp(crtc);
+
+	writel(BIT(0), adp->be + ADBE_BLEND_EN2);
+	writel(BIT(4), adp->be + ADBE_BLEND_EN1);
+	writel(BIT(0), adp->be + ADBE_BLEND_EN3);
+	writel(BIT(0), adp->be + ADBE_BLEND_BYPASS);
+	writel(BIT(0), adp->be + ADBE_BLEND_EN4);
+	drm_crtc_vblank_on(crtc);
+}
+
+static void adp_crtc_atomic_disable(struct drm_crtc *crtc,
+				    struct drm_atomic_state *state)
+{
+	struct adp_drv_private *adp = crtc_to_adp(crtc);
+	struct drm_crtc_state *old_state = drm_atomic_get_old_crtc_state(state, crtc);
+
+	drm_atomic_helper_disable_planes_on_crtc(old_state, false);
+
+	writel(0x0, adp->be + ADBE_BLEND_EN2);
+	writel(0x0, adp->be + ADBE_BLEND_EN1);
+	writel(0x0, adp->be + ADBE_BLEND_EN3);
+	writel(0x0, adp->be + ADBE_BLEND_BYPASS);
+	writel(0x0, adp->be + ADBE_BLEND_EN4);
+	drm_crtc_vblank_off(crtc);
+}
+
+static void adp_crtc_atomic_flush(struct drm_crtc *crtc,
+				  struct drm_atomic_state *state)
+{
+	u32 frame_num = 1;
+	unsigned long flags;
+	struct adp_drv_private *adp = crtc_to_adp(crtc);
+	struct drm_crtc_state *new_state = drm_atomic_get_new_crtc_state(state, crtc);
+	u64 new_size = ALIGN(new_state->mode.hdisplay *
+			     new_state->mode.vdisplay * 4, PAGE_SIZE);
+
+	if (new_size != adp->mask_buf_size) {
+		if (adp->mask_buf)
+			dma_free_coherent(crtc->dev->dev, adp->mask_buf_size,
+					  adp->mask_buf, adp->mask_iova);
+		adp->mask_buf = NULL;
+		if (new_size != 0) {
+			adp->mask_buf = dma_alloc_coherent(crtc->dev->dev, new_size,
+							   &adp->mask_iova, GFP_KERNEL);
+			memset(adp->mask_buf, 0xFF, new_size);
+			writel(adp->mask_iova, adp->be + ADBE_MASK_BUF);
+		}
+		adp->mask_buf_size = new_size;
+	}
+	writel(ADBE_FIFO_SYNC | frame_num, adp->be + ADBE_FIFO);
+	//FIXME: use adbe flush interrupt
+	if (crtc->state->event) {
+		struct drm_pending_vblank_event *event = crtc->state->event;
+
+		crtc->state->event = NULL;
+		spin_lock_irqsave(&crtc->dev->event_lock, flags);
+
+		if (drm_crtc_vblank_get(crtc) != 0)
+			drm_crtc_send_vblank_event(crtc, event);
+		else
+			adp->event = event;
+
+		spin_unlock_irqrestore(&crtc->dev->event_lock, flags);
+	}
+}
+
+static const struct drm_crtc_funcs adp_crtc_funcs = {
+	.destroy = drm_crtc_cleanup,
+	.set_config = drm_atomic_helper_set_config,
+	.page_flip = drm_atomic_helper_page_flip,
+	.reset = drm_atomic_helper_crtc_reset,
+	.atomic_duplicate_state = drm_atomic_helper_crtc_duplicate_state,
+	.atomic_destroy_state = drm_atomic_helper_crtc_destroy_state,
+	.enable_vblank = adp_crtc_enable_vblank,
+	.disable_vblank = adp_crtc_disable_vblank,
+};
+
+
+static const struct drm_crtc_helper_funcs adp_crtc_helper_funcs = {
+	.atomic_enable = adp_crtc_atomic_enable,
+	.atomic_disable = adp_crtc_atomic_disable,
+	.atomic_flush = adp_crtc_atomic_flush,
+};
+
+static int adp_setup_crtc(struct adp_drv_private *adp)
+{
+	struct drm_device *drm = &adp->drm;
+	struct drm_plane *primary;
+	int ret;
+
+	primary = adp_plane_new(adp);
+	if (IS_ERR(primary))
+		return PTR_ERR(primary);
+
+	ret = drm_crtc_init_with_planes(drm, &adp->crtc, primary,
+					NULL, &adp_crtc_funcs, NULL);
+	if (ret)
+		return ret;
+
+	drm_crtc_helper_add(&adp->crtc, &adp_crtc_helper_funcs);
+	return 0;
+}
+
+static const struct drm_mode_config_funcs adp_mode_config_funcs = {
+	.fb_create = drm_gem_fb_create_with_dirty,
+	.atomic_check = drm_atomic_helper_check,
+	.atomic_commit = drm_atomic_helper_commit,
+};
+
+static int adp_setup_mode_config(struct adp_drv_private *adp)
+{
+	struct drm_device *drm = &adp->drm;
+	int ret;
+	u32 size;
+
+	ret = drmm_mode_config_init(drm);
+	if (ret)
+		return ret;
+
+	/*
+	 * Query screen size restrict the frame buffer size to the screen size
+	 * aligned to the next multiple of 64. This is not necessary but can be
+	 * used as simple check for non-desktop devices.
+	 * Xorg's modesetting driver does not care about the connector
+	 * "non-desktop" property. The max frame buffer width or height can be
+	 * easily checked and a device can be reject if the max width/height is
+	 * smaller than 120 for example.
+	 * Any touchbar daemon is not limited by this small framebuffer size.
+	 */
+	size = readl(adp->fe + ADP_SCREEN_SIZE);
+
+	drm->mode_config.min_width = 32;
+	drm->mode_config.min_height = 32;
+	drm->mode_config.max_width = ALIGN(FIELD_GET(ADP_SCREEN_HSIZE, size), 64);
+	drm->mode_config.max_height = ALIGN(FIELD_GET(ADP_SCREEN_VSIZE, size), 64);
+	drm->mode_config.preferred_depth = 24;
+	drm->mode_config.prefer_shadow = 0;
+	drm->mode_config.funcs = &adp_mode_config_funcs;
+
+	ret = adp_setup_crtc(adp);
+	if (ret) {
+		drm_err(drm, "failed to create crtc");
+		return ret;
+	}
+
+	adp->encoder = drmm_plain_encoder_alloc(drm, NULL, DRM_MODE_ENCODER_DSI, NULL);
+	if (IS_ERR(adp->encoder)) {
+		drm_err(drm, "failed to init encoder");
+		return PTR_ERR(adp->encoder);
+	}
+	adp->encoder->possible_crtcs = ALL_CRTCS;
+
+	ret = drm_bridge_attach(adp->encoder, adp->next_bridge, NULL,
+				DRM_BRIDGE_ATTACH_NO_CONNECTOR);
+	if (ret) {
+		drm_err(drm, "failed to init bridge chain");
+		return ret;
+	}
+
+	adp->connector = drm_bridge_connector_init(drm, adp->encoder);
+	if (IS_ERR(adp->connector))
+		return PTR_ERR(adp->connector);
+
+	drm_connector_attach_encoder(adp->connector, adp->encoder);
+
+	ret = drm_vblank_init(drm, drm->mode_config.num_crtc);
+	if (ret < 0) {
+		drm_err(drm, "failed to initialize vblank");
+		return ret;
+	}
+
+	drm_mode_config_reset(drm);
+
+	return 0;
+}
+
+static int adp_parse_of(struct platform_device *pdev, struct adp_drv_private *adp)
+{
+	struct device *dev = &pdev->dev;
+
+	adp->be = devm_platform_ioremap_resource_byname(pdev, "be");
+	if (IS_ERR(adp->be)) {
+		dev_err(dev, "failed to map display backend mmio");
+		return PTR_ERR(adp->be);
+	}
+
+	adp->fe = devm_platform_ioremap_resource_byname(pdev, "fe");
+	if (IS_ERR(adp->fe)) {
+		dev_err(dev, "failed to map display pipe mmio");
+		return PTR_ERR(adp->fe);
+	}
+
+	adp->be_irq = platform_get_irq_byname(pdev, "be");
+	if (adp->be_irq < 0)
+		return adp->be_irq;
+
+	adp->fe_irq = platform_get_irq_byname(pdev, "fe");
+	if (adp->fe_irq < 0)
+		return adp->fe_irq;
+
+	return 0;
+}
+
+static irqreturn_t adp_fe_irq(int irq, void *arg)
+{
+	struct adp_drv_private *adp = (struct adp_drv_private *)arg;
+	u32 int_status;
+	u32 int_ctl;
+
+	int_status = readl(adp->fe + ADP_INT_STATUS);
+	if (int_status & ADP_INT_STATUS_VBLANK) {
+		drm_crtc_handle_vblank(&adp->crtc);
+		spin_lock(&adp->crtc.dev->event_lock);
+		if (adp->event) {
+			int_ctl = readl(adp->fe + ADP_CTRL);
+			if ((int_ctl & 0xF00) == 0x600) {
+				drm_crtc_send_vblank_event(&adp->crtc, adp->event);
+				adp->event = NULL;
+				drm_crtc_vblank_put(&adp->crtc);
+			}
+		}
+		spin_unlock(&adp->crtc.dev->event_lock);
+	}
+
+	writel(int_status, adp->fe + ADP_INT_STATUS);
+
+
+	return IRQ_HANDLED;
+}
+
+static int adp_drm_bind(struct device *dev)
+{
+	struct drm_device *drm = dev_get_drvdata(dev);
+	struct adp_drv_private *adp = to_adp(drm);
+	int err;
+
+	writel(ADP_CTRL_FIFO_ON, adp->fe + ADP_CTRL);
+
+	adp->next_bridge = drmm_of_get_bridge(&adp->drm, dev->of_node, 0, 0);
+	if (IS_ERR(adp->next_bridge)) {
+		dev_err(dev, "failed to find next bridge");
+		return PTR_ERR(adp->next_bridge);
+	}
+
+	err = adp_setup_mode_config(adp);
+	if (err < 0)
+		return err;
+
+	err = request_irq(adp->fe_irq, adp_fe_irq, 0, "adp-fe", adp);
+	if (err)
+		return err;
+
+	err = drm_dev_register(&adp->drm, 0);
+	if (err)
+		return err;
+
+	return 0;
+}
+
+static void adp_drm_unbind(struct device *dev)
+{
+	struct drm_device *drm = dev_get_drvdata(dev);
+	struct adp_drv_private *adp = to_adp(drm);
+
+	drm_dev_unregister(drm);
+	drm_atomic_helper_shutdown(drm);
+	free_irq(adp->fe_irq, adp);
+}
+
+static const struct component_master_ops adp_master_ops = {
+	.bind	= adp_drm_bind,
+	.unbind = adp_drm_unbind,
+};
+
+static int compare_dev(struct device *dev, void *data)
+{
+	return dev->of_node == data;
+}
+
+static int adp_probe(struct platform_device *pdev)
+{
+	struct device_node *port;
+	struct component_match *match = NULL;
+	struct adp_drv_private *adp;
+	int err;
+
+	adp = devm_drm_dev_alloc(&pdev->dev, &adp_driver, struct adp_drv_private, drm);
+	if (IS_ERR(adp))
+		return PTR_ERR(adp);
+
+	dev_set_drvdata(&pdev->dev, &adp->drm);
+
+	err = adp_parse_of(pdev, adp);
+	if (err < 0)
+		return err;
+
+	port = of_graph_get_remote_node(pdev->dev.of_node, 0, 0);
+	if (!port)
+		return -ENODEV;
+
+	drm_of_component_match_add(&pdev->dev, &match, compare_dev, port);
+	of_node_put(port);
+
+	return component_master_add_with_match(&pdev->dev, &adp_master_ops, match);
+}
+
+static void adp_remove(struct platform_device *pdev)
+{
+	component_master_del(&pdev->dev, &adp_master_ops);
+	dev_set_drvdata(&pdev->dev, NULL);
+}
+
+static const struct of_device_id adp_of_match[] = {
+	{ .compatible = "apple,h7-display-pipe", },
+	{ },
+};
+MODULE_DEVICE_TABLE(of, adp_of_match);
+
+static struct platform_driver adp_platform_driver = {
+	.driver = {
+		.name = "adp",
+		.of_match_table = adp_of_match,
+	},
+	.probe = adp_probe,
+	.remove = adp_remove,
+};
+
+module_platform_driver(adp_platform_driver);
+
+MODULE_DESCRIPTION("Apple Display Pipe DRM driver");
+MODULE_LICENSE("GPL");
diff --git a/drivers/gpu/drm/amd/acp/Kconfig b/drivers/gpu/drm/amd/acp/Kconfig
index e503e3d6d920..216d932a7831 100644
--- a/drivers/gpu/drm/amd/acp/Kconfig
+++ b/drivers/gpu/drm/amd/acp/Kconfig
@@ -1,11 +1,13 @@
+# SPDX-License-Identifier: MIT
 menu "ACP (Audio CoProcessor) Configuration"
+	depends on DRM_AMDGPU
 
 config DRM_AMD_ACP
-       bool "Enable AMD Audio CoProcessor IP support"
-       depends on DRM_AMDGPU
-       select MFD_CORE
-       select PM_GENERIC_DOMAINS if PM
-       help
+	bool "Enable AMD Audio CoProcessor IP support"
+	depends on DRM_AMDGPU
+	select MFD_CORE
+	select PM_GENERIC_DOMAINS if PM
+	help
 	Choose this option to enable ACP IP support for AMD SOCs.
 	This adds the ACP (Audio CoProcessor) IP driver and wires
 	it up into the amdgpu driver.  The ACP block provides the DMA
diff --git a/drivers/gpu/drm/amd/acp/Makefile b/drivers/gpu/drm/amd/acp/Makefile
index 8a08e81ee90d..d4176a3fb706 100644
--- a/drivers/gpu/drm/amd/acp/Makefile
+++ b/drivers/gpu/drm/amd/acp/Makefile
@@ -1,4 +1,25 @@
 #
+# Copyright 2017 Advanced Micro Devices, Inc.
+#
+# Permission is hereby granted, free of charge, to any person obtaining a
+# copy of this software and associated documentation files (the "Software"),
+# to deal in the Software without restriction, including without limitation
+# the rights to use, copy, modify, merge, publish, distribute, sublicense,
+# and/or sell copies of the Software, and to permit persons to whom the
+# Software is furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+# THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+# OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+# ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+# OTHER DEALINGS IN THE SOFTWARE.
+#
+#
 # Makefile for the ACP, which is a sub-component
 # of AMDSOC/AMDGPU drm driver.
 # It provides the HW control for ACP related functionalities.
diff --git a/drivers/gpu/drm/amd/acp/include/acp_gfx_if.h b/drivers/gpu/drm/amd/acp/include/acp_gfx_if.h
index a72ddb2f69ac..b26710cae801 100644
--- a/drivers/gpu/drm/amd/acp/include/acp_gfx_if.h
+++ b/drivers/gpu/drm/amd/acp/include/acp_gfx_if.h
@@ -19,13 +19,12 @@
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  * OTHER DEALINGS IN THE SOFTWARE.
  *
-*/
+ */
 
 #ifndef _ACP_GFX_IF_H
 #define _ACP_GFX_IF_H
 
 #include <linux/types.h>
-#include "cgs_linux.h"
 #include "cgs_common.h"
 
 int amd_acp_hw_init(struct cgs_device *cgs_device,
diff --git a/drivers/gpu/drm/amd/amdgpu/Kconfig b/drivers/gpu/drm/amd/amdgpu/Kconfig
index 61360e27715f..7f515be5185d 100644
--- a/drivers/gpu/drm/amd/amdgpu/Kconfig
+++ b/drivers/gpu/drm/amd/amdgpu/Kconfig
@@ -1,35 +1,108 @@
+# SPDX-License-Identifier: MIT
+
+config DRM_AMDGPU
+	tristate "AMD GPU"
+	depends on DRM && PCI
+	depends on !UML
+	select FW_LOADER
+	select DRM_CLIENT
+	select DRM_CLIENT_SELECTION
+	select DRM_DISPLAY_DP_HELPER
+	select DRM_DISPLAY_DSC_HELPER
+	select DRM_DISPLAY_HDMI_HELPER
+	select DRM_DISPLAY_HDCP_HELPER
+	select DRM_DISPLAY_HELPER
+	select DRM_KMS_HELPER
+	select DRM_SCHED
+	select DRM_TTM
+	select DRM_TTM_HELPER
+	select POWER_SUPPLY
+	select HWMON
+	select I2C
+	select I2C_ALGOBIT
+	select CRC16
+	select BACKLIGHT_CLASS_DEVICE
+	select INTERVAL_TREE
+	select DRM_BUDDY
+	select DRM_SUBALLOC_HELPER
+	select DRM_EXEC
+	select DRM_PANEL_BACKLIGHT_QUIRKS
+	# amdgpu depends on ACPI_VIDEO when ACPI is enabled, for select to work
+	# ACPI_VIDEO's dependencies must also be selected.
+	select INPUT if ACPI
+	select ACPI_VIDEO if ACPI
+	# On x86 ACPI_VIDEO also needs ACPI_WMI
+	select X86_PLATFORM_DEVICES if ACPI && X86
+	select ACPI_WMI if ACPI && X86
+	help
+	  Choose this option if you have a recent AMD Radeon graphics card.
+
+	  If M is selected, the module will be called amdgpu.
+
 config DRM_AMDGPU_SI
 	bool "Enable amdgpu support for SI parts"
 	depends on DRM_AMDGPU
 	help
-	  Choose this option if you want to enable experimental support
-	  for SI asics.
+	  Choose this option if you want to enable support
+	  for SI (Southern Islands) asics.
+
+	  SI (Southern Islands) are first generation GCN GPUs,
+	  supported by both drivers: radeon (old) and amdgpu (new).
+	  By default, SI dedicated GPUs are supported by amdgpu.
+
+	  Use module options to override this:
+	  To use radeon for SI,
+	  radeon.si_support=1 amdgpu.si_support=0
 
 config DRM_AMDGPU_CIK
 	bool "Enable amdgpu support for CIK parts"
 	depends on DRM_AMDGPU
 	help
-	  Choose this option if you want to enable experimental support
-	  for CIK asics.
+	  Choose this option if you want to enable support for CIK (Sea
+	  Islands) asics.
+
+	  CIK (Sea Islands) are second generation GCN GPUs,
+	  supported by both drivers: radeon (old) and amdgpu (new).
+	  By default,
+	  CIK dedicated GPUs are supported by amdgpu
+	  CIK APUs are supported by radeon
 
-	  CIK is already supported in radeon.  CIK support in amdgpu
-	  is for experimentation and testing.
+	  Use module options to override this:
+	  To use amdgpu for CIK,
+	  radeon.cik_support=0 amdgpu.cik_support=1
+	  To use radeon for CIK,
+	  radeon.cik_support=1 amdgpu.cik_support=0
 
 config DRM_AMDGPU_USERPTR
 	bool "Always enable userptr write support"
 	depends on DRM_AMDGPU
+	select HMM_MIRROR
 	select MMU_NOTIFIER
 	help
-	  This option selects CONFIG_MMU_NOTIFIER if it isn't already
-	  selected to enabled full userptr support.
+	  This option selects CONFIG_HMM and CONFIG_HMM_MIRROR if it
+	  isn't already selected to enabled full userptr support.
+
+config DRM_AMD_ISP
+	bool "Enable AMD Image Signal Processor IP support"
+	depends on DRM_AMDGPU && ACPI
+	select MFD_CORE
+	select PM_GENERIC_DOMAINS if PM
+	help
+	Choose this option to enable ISP IP support for AMD SOCs.
+	This adds the ISP (Image Signal Processor) IP driver and wires
+	it up into the amdgpu driver.  It is required for cameras
+	on APUs which utilize mipi cameras.
 
-config DRM_AMDGPU_GART_DEBUGFS
-	bool "Allow GART access through debugfs"
+config DRM_AMDGPU_WERROR
+	bool "Force the compiler to throw an error instead of a warning when compiling"
 	depends on DRM_AMDGPU
-	depends on DEBUG_FS
+	depends on EXPERT
+	depends on !COMPILE_TEST
 	default n
 	help
-	  Selecting this option creates a debugfs file to inspect the mapped
-	  pages. Uses more memory for housekeeping, enable only for debugging.
+	  Add -Werror to the build flags for amdgpu.ko.
+	  Only enable this if you are warning code for amdgpu.ko.
 
 source "drivers/gpu/drm/amd/acp/Kconfig"
+source "drivers/gpu/drm/amd/display/Kconfig"
+source "drivers/gpu/drm/amd/amdkfd/Kconfig"
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index 660786aba7d2..c88760fb52ea 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -1,46 +1,114 @@
 #
+# Copyright 2017-2024 Advanced Micro Devices, Inc. All rights reserved.
+#
+# Permission is hereby granted, free of charge, to any person obtaining a
+# copy of this software and associated documentation files (the "Software"),
+# to deal in the Software without restriction, including without limitation
+# the rights to use, copy, modify, merge, publish, distribute, sublicense,
+# and/or sell copies of the Software, and to permit persons to whom the
+# Software is furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+# THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+# OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+# ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+# OTHER DEALINGS IN THE SOFTWARE.
+#
+#
 # Makefile for the drm device driver.  This driver provides support for the
 # Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher.
 
 FULL_AMD_PATH=$(src)/..
+DISPLAY_FOLDER_NAME=display
+FULL_AMD_DISPLAY_PATH = $(FULL_AMD_PATH)/$(DISPLAY_FOLDER_NAME)
 
-ccflags-y := -Iinclude/drm -I$(FULL_AMD_PATH)/include/asic_reg \
+ccflags-y := -I$(FULL_AMD_PATH)/include/asic_reg \
 	-I$(FULL_AMD_PATH)/include \
 	-I$(FULL_AMD_PATH)/amdgpu \
-	-I$(FULL_AMD_PATH)/scheduler \
-	-I$(FULL_AMD_PATH)/powerplay/inc \
-	-I$(FULL_AMD_PATH)/acp/include
+	-I$(FULL_AMD_PATH)/pm/inc \
+	-I$(FULL_AMD_PATH)/acp/include \
+	-I$(FULL_AMD_DISPLAY_PATH) \
+	-I$(FULL_AMD_DISPLAY_PATH)/include \
+	-I$(FULL_AMD_DISPLAY_PATH)/modules/inc \
+	-I$(FULL_AMD_DISPLAY_PATH)/dc \
+	-I$(FULL_AMD_DISPLAY_PATH)/amdgpu_dm \
+	-I$(FULL_AMD_PATH)/amdkfd \
+	-I$(FULL_AMD_PATH)/ras/ras_mgr
+
+# Locally disable W=1 warnings enabled in drm subsystem Makefile
+subdir-ccflags-y += -Wno-override-init
+subdir-ccflags-$(CONFIG_DRM_AMDGPU_WERROR) += -Werror
 
 amdgpu-y := amdgpu_drv.o
 
 # add KMS driver
-amdgpu-y += amdgpu_device.o amdgpu_kms.o \
+amdgpu-y += amdgpu_device.o amdgpu_doorbell_mgr.o amdgpu_kms.o \
 	amdgpu_atombios.o atombios_crtc.o amdgpu_connectors.o \
 	atom.o amdgpu_fence.o amdgpu_ttm.o amdgpu_object.o amdgpu_gart.o \
 	amdgpu_encoders.o amdgpu_display.o amdgpu_i2c.o \
-	amdgpu_fb.o amdgpu_gem.o amdgpu_ring.o \
-	amdgpu_cs.o amdgpu_bios.o amdgpu_benchmark.o amdgpu_test.o \
-	amdgpu_pm.o atombios_dp.o amdgpu_afmt.o amdgpu_trace_points.o \
+	amdgpu_gem.o amdgpu_ring.o \
+	amdgpu_cs.o amdgpu_bios.o amdgpu_benchmark.o \
+	atombios_dp.o amdgpu_afmt.o amdgpu_trace_points.o \
 	atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
-	amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \
+	amdgpu_dma_buf.o amdgpu_vm.o amdgpu_vm_pt.o amdgpu_vm_tlb_fence.o \
+	amdgpu_ib.o amdgpu_pll.o \
 	amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
-	amdgpu_gtt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o amdgpu_atomfirmware.o
+	amdgpu_gtt_mgr.o amdgpu_preempt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o \
+	amdgpu_atomfirmware.o amdgpu_vf_error.o amdgpu_sched.o \
+	amdgpu_debugfs.o amdgpu_ids.o amdgpu_gmc.o amdgpu_mmhub.o amdgpu_hdp.o \
+	amdgpu_xgmi.o amdgpu_csa.o amdgpu_ras.o amdgpu_vm_cpu.o \
+	amdgpu_vm_sdma.o amdgpu_discovery.o amdgpu_ras_eeprom.o amdgpu_nbio.o \
+	amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
+	amdgpu_fw_attestation.o amdgpu_securedisplay.o \
+	amdgpu_eeprom.o amdgpu_mca.o amdgpu_psp_ta.o amdgpu_lsdma.o \
+	amdgpu_ring_mux.o amdgpu_xcp.o amdgpu_seq64.o amdgpu_aca.o amdgpu_dev_coredump.o \
+	amdgpu_cper.o amdgpu_userq_fence.o amdgpu_eviction_fence.o amdgpu_ip.o
+
+amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
+
+amdgpu-$(CONFIG_PERF_EVENTS) += amdgpu_pmu.o
 
 # add asic specific block
-amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
-	ci_smc.o ci_dpm.o dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o \
-	amdgpu_amdkfd_gfx_v7.o
+amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o \
+	dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o
+
+amdgpu-$(CONFIG_DRM_AMDGPU_SI)+= si.o gmc_v6_0.o gfx_v6_0.o si_ih.o si_dma.o dce_v6_0.o \
+	uvd_v3_1.o vce_v1_0.o
 
-amdgpu-$(CONFIG_DRM_AMDGPU_SI)+= si.o gmc_v6_0.o gfx_v6_0.o si_ih.o si_dma.o dce_v6_0.o si_dpm.o si_smc.o
+amdgpu-y += \
+	vi.o mxgpu_vi.o nbio_v6_1.o soc15.o emu_soc.o mxgpu_ai.o nbio_v7_0.o vega10_reg_init.o \
+	vega20_reg_init.o nbio_v7_4.o nbio_v2_3.o nv.o arct_reg_init.o mxgpu_nv.o \
+	nbio_v7_2.o hdp_v4_0.o hdp_v5_0.o aldebaran_reg_init.o aldebaran.o soc21.o soc24.o \
+	sienna_cichlid.o smu_v13_0_10.o nbio_v4_3.o hdp_v6_0.o nbio_v7_7.o hdp_v5_2.o lsdma_v6_0.o \
+	nbio_v7_9.o aqua_vanjaram.o nbio_v7_11.o lsdma_v7_0.o hdp_v7_0.o nbif_v6_3_1.o \
+	cyan_skillfish_reg_init.o
 
+# add DF block
 amdgpu-y += \
-	vi.o mxgpu_vi.o nbio_v6_1.o soc15.o mxgpu_ai.o
+	df_v1_7.o \
+	df_v3_6.o \
+	df_v4_3.o \
+	df_v4_6_2.o \
+	df_v4_15.o
 
 # add GMC block
 amdgpu-y += \
 	gmc_v7_0.o \
 	gmc_v8_0.o \
-	gfxhub_v1_0.o mmhub_v1_0.o gmc_v9_0.o
+	gfxhub_v1_0.o mmhub_v1_0.o gmc_v9_0.o gfxhub_v1_1.o mmhub_v9_4.o \
+	gfxhub_v2_0.o mmhub_v2_0.o gmc_v10_0.o gfxhub_v2_1.o mmhub_v2_3.o \
+	mmhub_v1_7.o gfxhub_v3_0.o mmhub_v3_0.o mmhub_v3_0_2.o gmc_v11_0.o \
+	mmhub_v3_0_1.o gfxhub_v3_0_3.o gfxhub_v1_2.o mmhub_v1_8.o mmhub_v3_3.o \
+	gfxhub_v11_5_0.o mmhub_v4_1_0.o gfxhub_v12_0.o gmc_v12_0.o
+
+# add UMC block
+amdgpu-y += \
+	umc_v6_0.o umc_v6_1.o umc_v6_7.o umc_v8_7.o umc_v8_10.o umc_v12_0.o umc_v8_14.o
 
 # add IH block
 amdgpu-y += \
@@ -49,35 +117,68 @@ amdgpu-y += \
 	iceland_ih.o \
 	tonga_ih.o \
 	cz_ih.o \
-	vega10_ih.o
+	vega10_ih.o \
+	vega20_ih.o \
+	navi10_ih.o \
+	ih_v6_0.o \
+	ih_v6_1.o \
+	ih_v7_0.o
 
 # add PSP block
 amdgpu-y += \
 	amdgpu_psp.o \
-	psp_v3_1.o
-
-# add SMC block
-amdgpu-y += \
-	amdgpu_dpm.o \
-	amdgpu_powerplay.o
+	psp_v3_1.o \
+	psp_v10_0.o \
+	psp_v11_0.o \
+	psp_v11_0_8.o \
+	psp_v12_0.o \
+	psp_v13_0.o \
+	psp_v13_0_4.o \
+	psp_v14_0.o
 
 # add DCE block
 amdgpu-y += \
 	dce_v10_0.o \
-	dce_v11_0.o \
-	dce_virtual.o
+	amdgpu_vkms.o
 
 # add GFX block
 amdgpu-y += \
 	amdgpu_gfx.o \
+	amdgpu_rlc.o \
 	gfx_v8_0.o \
-	gfx_v9_0.o
+	gfx_v9_0.o \
+	gfx_v9_4.o \
+	gfx_v9_4_2.o \
+	gfx_v9_4_3.o \
+	gfx_v10_0.o \
+	imu_v11_0.o \
+	gfx_v11_0.o \
+	gfx_v11_0_3.o \
+	imu_v11_0_3.o \
+	gfx_v12_0.o \
+	imu_v12_0.o
 
 # add async DMA block
 amdgpu-y += \
+	amdgpu_sdma.o \
 	sdma_v2_4.o \
 	sdma_v3_0.o \
-	sdma_v4_0.o
+	sdma_v4_0.o \
+	sdma_v4_4.o \
+	sdma_v4_4_2.o \
+	sdma_v5_0.o \
+	sdma_v5_2.o \
+	sdma_v6_0.o \
+	sdma_v7_0.o
+
+# add MES block
+amdgpu-y += \
+	amdgpu_mes.o \
+	mes_v11_0.o \
+	mes_v12_0.o \
+
+# add GFX userqueue support
+amdgpu-y += mes_userqueue.o
 
 # add UVD block
 amdgpu-y += \
@@ -92,19 +193,101 @@ amdgpu-y += \
 	vce_v3_0.o \
 	vce_v4_0.o
 
+# add VCN and JPEG block
+amdgpu-y += \
+	amdgpu_vcn.o \
+	vcn_sw_ring.o \
+	vcn_v1_0.o \
+	vcn_v2_0.o \
+	vcn_v2_5.o \
+	vcn_v3_0.o \
+	vcn_v4_0.o \
+	vcn_v4_0_3.o \
+	vcn_v4_0_5.o \
+	vcn_v5_0_0.o \
+	vcn_v5_0_1.o \
+	amdgpu_jpeg.o \
+	jpeg_v1_0.o \
+	jpeg_v2_0.o \
+	jpeg_v2_5.o \
+	jpeg_v3_0.o \
+	jpeg_v4_0.o \
+	jpeg_v4_0_3.o \
+	jpeg_v4_0_5.o \
+	jpeg_v5_0_0.o \
+	jpeg_v5_0_1.o
+
+# add VPE block
+amdgpu-y += \
+	amdgpu_vpe.o \
+	vpe_v6_1.o
+
+# add UMSCH block
+amdgpu-y += \
+	amdgpu_umsch_mm.o \
+	umsch_mm_v4_0.o
+
+#
+# add ATHUB block
+amdgpu-y += \
+	athub_v1_0.o \
+	athub_v2_0.o \
+	athub_v2_1.o \
+	athub_v3_0.o \
+	athub_v4_1_0.o
+
+# add SMUIO block
+amdgpu-y += \
+	smuio_v9_0.o \
+	smuio_v11_0.o \
+	smuio_v11_0_6.o \
+	smuio_v13_0.o \
+	smuio_v13_0_3.o \
+	smuio_v13_0_6.o \
+	smuio_v14_0_2.o
+
+# add reset block
+amdgpu-y += \
+	amdgpu_reset.o
+
+# add MCA block
+amdgpu-y += \
+	mca_v3_0.o
+
 # add amdkfd interfaces
+amdgpu-y += amdgpu_amdkfd.o
+
+# add gfx usermode queue
+amdgpu-y += amdgpu_userq.o
+
+ifneq ($(CONFIG_HSA_AMD),)
+AMDKFD_PATH := ../amdkfd
+include $(FULL_AMD_PATH)/amdkfd/Makefile
+amdgpu-y += $(AMDKFD_FILES)
 amdgpu-y += \
-	 amdgpu_amdkfd.o \
-	 amdgpu_amdkfd_gfx_v8.o
+	amdgpu_amdkfd_fence.o \
+	amdgpu_amdkfd_gpuvm.o \
+	amdgpu_amdkfd_gfx_v8.o \
+	amdgpu_amdkfd_gfx_v9.o \
+	amdgpu_amdkfd_arcturus.o \
+	amdgpu_amdkfd_aldebaran.o \
+	amdgpu_amdkfd_gc_9_4_3.o \
+	amdgpu_amdkfd_gfx_v10.o \
+	amdgpu_amdkfd_gfx_v10_3.o \
+	amdgpu_amdkfd_gfx_v11.o \
+	amdgpu_amdkfd_gfx_v12.o
+
+ifneq ($(CONFIG_DRM_AMDGPU_CIK),)
+amdgpu-y += amdgpu_amdkfd_gfx_v7.o
+endif
+
+endif
 
 # add cgs
 amdgpu-y += amdgpu_cgs.o
 
 # GPU scheduler
-amdgpu-y += \
-	../scheduler/gpu_scheduler.o \
-	../scheduler/sched_fence.o \
-	amdgpu_job.o
+amdgpu-y += amdgpu_job.o
 
 # ACP componet
 ifneq ($(CONFIG_DRM_AMD_ACP),)
@@ -119,12 +302,32 @@ endif
 amdgpu-$(CONFIG_COMPAT) += amdgpu_ioc32.o
 amdgpu-$(CONFIG_VGA_SWITCHEROO) += amdgpu_atpx_handler.o
 amdgpu-$(CONFIG_ACPI) += amdgpu_acpi.o
-amdgpu-$(CONFIG_MMU_NOTIFIER) += amdgpu_mn.o
+amdgpu-$(CONFIG_HMM_MIRROR) += amdgpu_hmm.o
 
-include $(FULL_AMD_PATH)/powerplay/Makefile
+include $(FULL_AMD_PATH)/pm/Makefile
 
 amdgpu-y += $(AMD_POWERPLAY_FILES)
 
-obj-$(CONFIG_DRM_AMDGPU)+= amdgpu.o
+ifneq ($(CONFIG_DRM_AMD_DC),)
+
+RELATIVE_AMD_DISPLAY_PATH = ../$(DISPLAY_FOLDER_NAME)
+include $(FULL_AMD_DISPLAY_PATH)/Makefile
+
+amdgpu-y += $(AMD_DISPLAY_FILES)
+
+endif
+
+# add isp block
+ifneq ($(CONFIG_DRM_AMD_ISP),)
+amdgpu-y += \
+	amdgpu_isp.o \
+	isp_v4_1_0.o \
+	isp_v4_1_1.o
+endif
+
+AMD_GPU_RAS_PATH := ../ras
+AMD_GPU_RAS_FULL_PATH := $(FULL_AMD_PATH)/ras
+include $(AMD_GPU_RAS_FULL_PATH)/Makefile
+amdgpu-y += $(AMD_GPU_RAS_FILES)
 
-CFLAGS_amdgpu_trace_points.o := -I$(src)
+obj-$(CONFIG_DRM_AMDGPU)+= amdgpu.o
diff --git a/drivers/gpu/drm/amd/amdgpu/ObjectID.h b/drivers/gpu/drm/amd/amdgpu/ObjectID.h
index 06192698bd96..a0f0a17e224f 100644
--- a/drivers/gpu/drm/amd/amdgpu/ObjectID.h
+++ b/drivers/gpu/drm/amd/amdgpu/ObjectID.h
@@ -119,6 +119,7 @@
 #define CONNECTOR_OBJECT_ID_eDP                   0x14
 #define CONNECTOR_OBJECT_ID_MXM                   0x15
 #define CONNECTOR_OBJECT_ID_LVDS_eDP              0x16
+#define CONNECTOR_OBJECT_ID_USBC                  0x17
 
 /* deleted */
 
@@ -136,6 +137,7 @@
 #define GENERIC_OBJECT_ID_PX2_NON_DRIVABLE        0x02
 #define GENERIC_OBJECT_ID_MXM_OPM                 0x03
 #define GENERIC_OBJECT_ID_STEREO_PIN              0x04        //This object could show up from Misc Object table, it follows ATOM_OBJECT format, and contains one ATOM_OBJECT_GPIO_CNTL_RECORD for the stereo pin
+#define GENERIC_OBJECT_ID_BRACKET_LAYOUT          0x05
 
 /****************************************************/
 /* Graphics Object ENUM ID Definition               */
@@ -714,6 +716,13 @@
                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
                                                  GENERIC_OBJECT_ID_STEREO_PIN << OBJECT_ID_SHIFT)
 
+#define GENERICOBJECT_BRACKET_LAYOUT_ENUM_ID1    (GRAPH_OBJECT_TYPE_GENERIC << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                 GENERIC_OBJECT_ID_BRACKET_LAYOUT << OBJECT_ID_SHIFT)
+
+#define GENERICOBJECT_BRACKET_LAYOUT_ENUM_ID2    (GRAPH_OBJECT_TYPE_GENERIC << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
+                                                 GENERIC_OBJECT_ID_BRACKET_LAYOUT << OBJECT_ID_SHIFT)
 /****************************************************/
 /* Object Cap definition - Shared with BIOS         */
 /****************************************************/
diff --git a/drivers/gpu/drm/amd/amdgpu/aldebaran.c b/drivers/gpu/drm/amd/amdgpu/aldebaran.c
new file mode 100644
index 000000000000..daa7b23bc775
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/aldebaran.c
@@ -0,0 +1,473 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "aldebaran.h"
+#include "amdgpu_reset.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_dpm.h"
+#include "amdgpu_job.h"
+#include "amdgpu_ring.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_psp.h"
+#include "amdgpu_xgmi.h"
+
+static bool aldebaran_is_mode2_default(struct amdgpu_reset_control *reset_ctl)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
+
+	if ((amdgpu_ip_version(adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 2) &&
+	     adev->gmc.xgmi.connected_to_cpu))
+		return true;
+
+	return false;
+}
+
+static struct amdgpu_reset_handler *
+aldebaran_get_reset_handler(struct amdgpu_reset_control *reset_ctl,
+			    struct amdgpu_reset_context *reset_context)
+{
+	struct amdgpu_reset_handler *handler;
+	struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
+	int i;
+
+	if (reset_context->method == AMD_RESET_METHOD_NONE) {
+		if (aldebaran_is_mode2_default(reset_ctl))
+			reset_context->method = AMD_RESET_METHOD_MODE2;
+		else
+			reset_context->method = amdgpu_asic_reset_method(adev);
+	}
+
+	if (reset_context->method != AMD_RESET_METHOD_NONE) {
+		dev_dbg(adev->dev, "Getting reset handler for method %d\n",
+			reset_context->method);
+		for_each_handler(i, handler, reset_ctl) {
+			if (handler->reset_method == reset_context->method)
+				return handler;
+		}
+	}
+
+	dev_dbg(adev->dev, "Reset handler not found!\n");
+
+	return NULL;
+}
+
+static inline uint32_t aldebaran_get_ip_block_mask(struct amdgpu_device *adev)
+{
+	uint32_t ip_block_mask = BIT(AMD_IP_BLOCK_TYPE_GFX) |
+				 BIT(AMD_IP_BLOCK_TYPE_SDMA);
+
+	if (adev->aid_mask)
+		ip_block_mask |= BIT(AMD_IP_BLOCK_TYPE_IH);
+
+	return ip_block_mask;
+}
+
+static int aldebaran_mode2_suspend_ip(struct amdgpu_device *adev)
+{
+	uint32_t ip_block_mask = aldebaran_get_ip_block_mask(adev);
+	uint32_t ip_block;
+	int r, i;
+
+	/* Skip suspend of SDMA IP versions >= 4.4.2. They are multi-aid */
+	if (adev->aid_mask)
+		ip_block_mask &= ~BIT(AMD_IP_BLOCK_TYPE_SDMA);
+
+	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
+	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
+
+	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
+		ip_block = BIT(adev->ip_blocks[i].version->type);
+		if (!(ip_block_mask & ip_block))
+			continue;
+
+		r = amdgpu_ip_block_suspend(&adev->ip_blocks[i]);
+		if (r)
+			return r;
+	}
+
+	return 0;
+}
+
+static int
+aldebaran_mode2_prepare_hwcontext(struct amdgpu_reset_control *reset_ctl,
+				  struct amdgpu_reset_context *reset_context)
+{
+	int r = 0;
+	struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
+
+	dev_dbg(adev->dev, "Aldebaran prepare hw context\n");
+	/* Don't suspend on bare metal if we are not going to HW reset the ASIC */
+	if (!amdgpu_sriov_vf(adev))
+		r = aldebaran_mode2_suspend_ip(adev);
+
+	return r;
+}
+
+static void aldebaran_async_reset(struct work_struct *work)
+{
+	struct amdgpu_reset_handler *handler;
+	struct amdgpu_reset_control *reset_ctl =
+		container_of(work, struct amdgpu_reset_control, reset_work);
+	struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
+	int i;
+
+	for_each_handler(i, handler, reset_ctl)	{
+		if (handler->reset_method == reset_ctl->active_reset) {
+			dev_dbg(adev->dev, "Resetting device\n");
+			handler->do_reset(adev);
+			break;
+		}
+	}
+}
+
+static int aldebaran_mode2_reset(struct amdgpu_device *adev)
+{
+	/* disable BM */
+	pci_clear_master(adev->pdev);
+	adev->asic_reset_res = amdgpu_dpm_mode2_reset(adev);
+	return adev->asic_reset_res;
+}
+
+static int
+aldebaran_mode2_perform_reset(struct amdgpu_reset_control *reset_ctl,
+			      struct amdgpu_reset_context *reset_context)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
+	struct list_head *reset_device_list = reset_context->reset_device_list;
+	struct amdgpu_device *tmp_adev = NULL;
+	int r = 0;
+
+	dev_dbg(adev->dev, "aldebaran perform hw reset\n");
+
+	if (reset_device_list == NULL)
+		return -EINVAL;
+
+	if (amdgpu_ip_version(adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 2) &&
+	    reset_context->hive == NULL) {
+		/* Wrong context, return error */
+		return -EINVAL;
+	}
+
+	list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
+		mutex_lock(&tmp_adev->reset_cntl->reset_lock);
+		tmp_adev->reset_cntl->active_reset = AMD_RESET_METHOD_MODE2;
+	}
+	/*
+	 * Mode2 reset doesn't need any sync between nodes in XGMI hive, instead launch
+	 * them together so that they can be completed asynchronously on multiple nodes
+	 */
+	list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
+		/* For XGMI run all resets in parallel to speed up the process */
+		if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
+			if (!queue_work(system_unbound_wq,
+					&tmp_adev->reset_cntl->reset_work))
+				r = -EALREADY;
+		} else
+			r = aldebaran_mode2_reset(tmp_adev);
+		if (r) {
+			dev_err(tmp_adev->dev,
+				"ASIC reset failed with error, %d for drm dev, %s",
+				r, adev_to_drm(tmp_adev)->unique);
+			break;
+		}
+	}
+
+	/* For XGMI wait for all resets to complete before proceed */
+	if (!r) {
+		list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
+			if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
+				flush_work(&tmp_adev->reset_cntl->reset_work);
+				r = tmp_adev->asic_reset_res;
+				if (r)
+					break;
+			}
+		}
+	}
+
+	list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
+		mutex_unlock(&tmp_adev->reset_cntl->reset_lock);
+		tmp_adev->reset_cntl->active_reset = AMD_RESET_METHOD_NONE;
+	}
+
+	return r;
+}
+
+static int aldebaran_mode2_restore_ip(struct amdgpu_device *adev)
+{
+	struct amdgpu_firmware_info *ucode_list[AMDGPU_UCODE_ID_MAXIMUM];
+	uint32_t ip_block_mask = aldebaran_get_ip_block_mask(adev);
+	struct amdgpu_firmware_info *ucode;
+	struct amdgpu_ip_block *cmn_block;
+	struct amdgpu_ip_block *ih_block;
+	int ucode_count = 0;
+	int i, r;
+
+	dev_dbg(adev->dev, "Reloading ucodes after reset\n");
+	for (i = 0; i < adev->firmware.max_ucodes; i++) {
+		ucode = &adev->firmware.ucode[i];
+		if (!ucode->fw)
+			continue;
+		switch (ucode->ucode_id) {
+		case AMDGPU_UCODE_ID_SDMA0:
+		case AMDGPU_UCODE_ID_SDMA1:
+		case AMDGPU_UCODE_ID_SDMA2:
+		case AMDGPU_UCODE_ID_SDMA3:
+		case AMDGPU_UCODE_ID_SDMA4:
+		case AMDGPU_UCODE_ID_SDMA5:
+		case AMDGPU_UCODE_ID_SDMA6:
+		case AMDGPU_UCODE_ID_SDMA7:
+		case AMDGPU_UCODE_ID_CP_MEC1:
+		case AMDGPU_UCODE_ID_CP_MEC1_JT:
+		case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL:
+		case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM:
+		case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM:
+		case AMDGPU_UCODE_ID_RLC_G:
+			ucode_list[ucode_count++] = ucode;
+			break;
+		default:
+			break;
+		}
+	}
+
+	/* Reinit NBIF block */
+	cmn_block =
+		amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_COMMON);
+	if (unlikely(!cmn_block)) {
+		dev_err(adev->dev, "Failed to get BIF handle\n");
+		return -EINVAL;
+	}
+	r = amdgpu_ip_block_resume(cmn_block);
+	if (r)
+		return r;
+
+	if (ip_block_mask & BIT(AMD_IP_BLOCK_TYPE_IH)) {
+		ih_block = amdgpu_device_ip_get_ip_block(adev,
+							 AMD_IP_BLOCK_TYPE_IH);
+		if (unlikely(!ih_block)) {
+			dev_err(adev->dev, "Failed to get IH handle\n");
+			return -EINVAL;
+		}
+		r = amdgpu_ip_block_resume(ih_block);
+		if (r)
+			return r;
+	}
+
+	/* Reinit GFXHUB */
+	adev->gfxhub.funcs->init(adev);
+	r = adev->gfxhub.funcs->gart_enable(adev);
+	if (r) {
+		dev_err(adev->dev, "GFXHUB gart reenable failed after reset\n");
+		return r;
+	}
+
+	/* Reload GFX firmware */
+	r = psp_load_fw_list(&adev->psp, ucode_list, ucode_count);
+	if (r) {
+		dev_err(adev->dev, "GFX ucode load failed after reset\n");
+		return r;
+	}
+
+	/* Resume RLC, FW needs RLC alive to complete reset process */
+	adev->gfx.rlc.funcs->resume(adev);
+
+	/* Wait for FW reset event complete */
+	r = amdgpu_dpm_wait_for_event(adev, SMU_EVENT_RESET_COMPLETE, 0);
+	if (r) {
+		dev_err(adev->dev,
+			"Failed to get response from firmware after reset\n");
+		return r;
+	}
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!(adev->ip_blocks[i].version->type ==
+			      AMD_IP_BLOCK_TYPE_GFX ||
+		      adev->ip_blocks[i].version->type ==
+			      AMD_IP_BLOCK_TYPE_SDMA))
+			continue;
+
+		r = amdgpu_ip_block_resume(&adev->ip_blocks[i]);
+		if (r)
+			return r;
+	}
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!(adev->ip_blocks[i].version->type ==
+			      AMD_IP_BLOCK_TYPE_GFX ||
+		      adev->ip_blocks[i].version->type ==
+			      AMD_IP_BLOCK_TYPE_SDMA ||
+		      adev->ip_blocks[i].version->type ==
+			      AMD_IP_BLOCK_TYPE_COMMON))
+			continue;
+
+		if (adev->ip_blocks[i].version->funcs->late_init) {
+			r = adev->ip_blocks[i].version->funcs->late_init(
+				&adev->ip_blocks[i]);
+			if (r) {
+				dev_err(adev->dev,
+					"late_init of IP block <%s> failed %d after reset\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
+				return r;
+			}
+		}
+		adev->ip_blocks[i].status.late_initialized = true;
+	}
+
+	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE);
+	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE);
+
+	return r;
+}
+
+static int
+aldebaran_mode2_restore_hwcontext(struct amdgpu_reset_control *reset_ctl,
+				  struct amdgpu_reset_context *reset_context)
+{
+	struct list_head *reset_device_list = reset_context->reset_device_list;
+	struct amdgpu_device *tmp_adev = NULL;
+	struct amdgpu_ras *con;
+	int r;
+
+	if (reset_device_list == NULL)
+		return -EINVAL;
+
+	if (amdgpu_ip_version(reset_context->reset_req_dev, MP1_HWIP, 0) ==
+		    IP_VERSION(13, 0, 2) &&
+	    reset_context->hive == NULL) {
+		/* Wrong context, return error */
+		return -EINVAL;
+	}
+
+	list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
+		amdgpu_set_init_level(tmp_adev,
+				AMDGPU_INIT_LEVEL_RESET_RECOVERY);
+		dev_info(tmp_adev->dev,
+			 "GPU reset succeeded, trying to resume\n");
+		/*TBD: Ideally should clear only GFX, SDMA blocks*/
+		amdgpu_ras_clear_err_state(tmp_adev);
+		r = aldebaran_mode2_restore_ip(tmp_adev);
+		if (r)
+			goto end;
+
+		/*
+		 * Add this ASIC as tracked as reset was already
+		 * complete successfully.
+		 */
+		amdgpu_register_gpu_instance(tmp_adev);
+
+		/* Resume RAS, ecc_irq */
+		con = amdgpu_ras_get_context(tmp_adev);
+		if (!amdgpu_sriov_vf(tmp_adev) && con) {
+			if (tmp_adev->sdma.ras &&
+				tmp_adev->sdma.ras->ras_block.ras_late_init) {
+				r = tmp_adev->sdma.ras->ras_block.ras_late_init(tmp_adev,
+						&tmp_adev->sdma.ras->ras_block.ras_comm);
+				if (r) {
+					dev_err(tmp_adev->dev, "SDMA failed to execute ras_late_init! ret:%d\n", r);
+					goto end;
+				}
+			}
+
+			if (tmp_adev->gfx.ras &&
+				tmp_adev->gfx.ras->ras_block.ras_late_init) {
+				r = tmp_adev->gfx.ras->ras_block.ras_late_init(tmp_adev,
+						&tmp_adev->gfx.ras->ras_block.ras_comm);
+				if (r) {
+					dev_err(tmp_adev->dev, "GFX failed to execute ras_late_init! ret:%d\n", r);
+					goto end;
+				}
+			}
+		}
+
+		amdgpu_ras_resume(tmp_adev);
+
+		/* Update PSP FW topology after reset */
+		if (reset_context->hive &&
+		    tmp_adev->gmc.xgmi.num_physical_nodes > 1)
+			r = amdgpu_xgmi_update_topology(reset_context->hive,
+							tmp_adev);
+
+		if (!r) {
+			amdgpu_set_init_level(tmp_adev,
+					      AMDGPU_INIT_LEVEL_DEFAULT);
+			amdgpu_irq_gpu_reset_resume_helper(tmp_adev);
+
+			r = amdgpu_ib_ring_tests(tmp_adev);
+			if (r) {
+				dev_err(tmp_adev->dev,
+					"ib ring test failed (%d).\n", r);
+				r = -EAGAIN;
+				tmp_adev->asic_reset_res = r;
+				goto end;
+			}
+		}
+	}
+
+end:
+	return r;
+}
+
+static struct amdgpu_reset_handler aldebaran_mode2_handler = {
+	.reset_method		= AMD_RESET_METHOD_MODE2,
+	.prepare_env		= NULL,
+	.prepare_hwcontext	= aldebaran_mode2_prepare_hwcontext,
+	.perform_reset		= aldebaran_mode2_perform_reset,
+	.restore_hwcontext	= aldebaran_mode2_restore_hwcontext,
+	.restore_env		= NULL,
+	.do_reset		= aldebaran_mode2_reset,
+};
+
+static struct amdgpu_reset_handler
+	*aldebaran_rst_handlers[AMDGPU_RESET_MAX_HANDLERS] = {
+		&aldebaran_mode2_handler,
+		&xgmi_reset_on_init_handler,
+	};
+
+int aldebaran_reset_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_reset_control *reset_ctl;
+
+	reset_ctl = kzalloc(sizeof(*reset_ctl), GFP_KERNEL);
+	if (!reset_ctl)
+		return -ENOMEM;
+
+	reset_ctl->handle = adev;
+	reset_ctl->async_reset = aldebaran_async_reset;
+	reset_ctl->active_reset = AMD_RESET_METHOD_NONE;
+	reset_ctl->get_reset_handler = aldebaran_get_reset_handler;
+
+	INIT_WORK(&reset_ctl->reset_work, reset_ctl->async_reset);
+	/* Only mode2 is handled through reset control now */
+	reset_ctl->reset_handlers = &aldebaran_rst_handlers;
+
+	adev->reset_cntl = reset_ctl;
+
+	return 0;
+}
+
+int aldebaran_reset_fini(struct amdgpu_device *adev)
+{
+	kfree(adev->reset_cntl);
+	adev->reset_cntl = NULL;
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/aldebaran.h b/drivers/gpu/drm/amd/amdgpu/aldebaran.h
new file mode 100644
index 000000000000..a07db5454d49
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/aldebaran.h
@@ -0,0 +1,32 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __ALDEBARAN_H__
+#define __ALDEBARAN_H__
+
+#include "amdgpu.h"
+
+int aldebaran_reset_init(struct amdgpu_device *adev);
+int aldebaran_reset_fini(struct amdgpu_device *adev);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/aldebaran_reg_init.c b/drivers/gpu/drm/amd/amdgpu/aldebaran_reg_init.c
new file mode 100644
index 000000000000..28e6c9ab8767
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/aldebaran_reg_init.c
@@ -0,0 +1,54 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include "amdgpu.h"
+#include "soc15.h"
+
+#include "soc15_common.h"
+#include "aldebaran_ip_offset.h"
+
+int aldebaran_reg_base_init(struct amdgpu_device *adev)
+{
+	/* HW has more IP blocks,  only initialized the block needed by our driver  */
+	uint32_t i;
+	for (i = 0 ; i < MAX_INSTANCE ; ++i) {
+		adev->reg_offset[GC_HWIP][i] = (uint32_t *)(&(GC_BASE.instance[i]));
+		adev->reg_offset[HDP_HWIP][i] = (uint32_t *)(&(HDP_BASE.instance[i]));
+		adev->reg_offset[MMHUB_HWIP][i] = (uint32_t *)(&(MMHUB_BASE.instance[i]));
+		adev->reg_offset[ATHUB_HWIP][i] = (uint32_t *)(&(ATHUB_BASE.instance[i]));
+		adev->reg_offset[NBIO_HWIP][i] = (uint32_t *)(&(NBIO_BASE.instance[i]));
+		adev->reg_offset[MP0_HWIP][i] = (uint32_t *)(&(MP0_BASE.instance[i]));
+		adev->reg_offset[MP1_HWIP][i] = (uint32_t *)(&(MP1_BASE.instance[i]));
+		adev->reg_offset[DF_HWIP][i] = (uint32_t *)(&(DF_BASE.instance[i]));
+		adev->reg_offset[OSSSYS_HWIP][i] = (uint32_t *)(&(OSSSYS_BASE.instance[i]));
+		adev->reg_offset[SDMA0_HWIP][i] = (uint32_t *)(&(SDMA0_BASE.instance[i]));
+		adev->reg_offset[SDMA1_HWIP][i] = (uint32_t *)(&(SDMA1_BASE.instance[i]));
+		adev->reg_offset[SDMA2_HWIP][i] = (uint32_t *)(&(SDMA2_BASE.instance[i]));
+		adev->reg_offset[SDMA3_HWIP][i] = (uint32_t *)(&(SDMA3_BASE.instance[i]));
+		adev->reg_offset[SDMA4_HWIP][i] = (uint32_t *)(&(SDMA4_BASE.instance[i]));
+		adev->reg_offset[SMUIO_HWIP][i] = (uint32_t *)(&(SMUIO_BASE.instance[i]));
+		adev->reg_offset[THM_HWIP][i] = (uint32_t *)(&(THM_BASE.instance[i]));
+		adev->reg_offset[UMC_HWIP][i] = (uint32_t *)(&(UMC_BASE.instance[i]));
+		adev->reg_offset[VCN_HWIP][i] = (uint32_t *)(&(VCN_BASE.instance[i]));
+	}
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 833c3c16501a..9f9774f58ce1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -28,6 +28,20 @@
 #ifndef __AMDGPU_H__
 #define __AMDGPU_H__
 
+#ifdef pr_fmt
+#undef pr_fmt
+#endif
+
+#define pr_fmt(fmt) "amdgpu: " fmt
+
+#ifdef dev_fmt
+#undef dev_fmt
+#endif
+
+#define dev_fmt(fmt) "amdgpu: " fmt
+
+#include "amdgpu_ctx.h"
+
 #include <linux/atomic.h>
 #include <linux/wait.h>
 #include <linux/list.h>
@@ -35,18 +49,21 @@
 #include <linux/rbtree.h>
 #include <linux/hashtable.h>
 #include <linux/dma-fence.h>
+#include <linux/pci.h>
 
-#include <ttm/ttm_bo_api.h>
-#include <ttm/ttm_bo_driver.h>
-#include <ttm/ttm_placement.h>
-#include <ttm/ttm_module.h>
-#include <ttm/ttm_execbuf_util.h>
+#include <drm/ttm/ttm_bo.h>
+#include <drm/ttm/ttm_placement.h>
 
-#include <drm/drmP.h>
-#include <drm/drm_gem.h>
 #include <drm/amdgpu_drm.h>
+#include <drm/drm_gem.h>
+#include <drm/drm_ioctl.h>
+
+#include <kgd_kfd_interface.h>
+#include "dm_pp_interface.h"
+#include "kgd_pp_interface.h"
 
 #include "amd_shared.h"
+#include "amdgpu_utils.h"
 #include "amdgpu_mode.h"
 #include "amdgpu_ih.h"
 #include "amdgpu_irq.h"
@@ -57,71 +74,216 @@
 #include "amdgpu_sync.h"
 #include "amdgpu_ring.h"
 #include "amdgpu_vm.h"
-#include "amd_powerplay.h"
 #include "amdgpu_dpm.h"
 #include "amdgpu_acp.h"
 #include "amdgpu_uvd.h"
 #include "amdgpu_vce.h"
-
-#include "gpu_scheduler.h"
+#include "amdgpu_vcn.h"
+#include "amdgpu_jpeg.h"
+#include "amdgpu_vpe.h"
+#include "amdgpu_umsch_mm.h"
+#include "amdgpu_gmc.h"
+#include "amdgpu_gfx.h"
+#include "amdgpu_sdma.h"
+#include "amdgpu_lsdma.h"
+#include "amdgpu_nbio.h"
+#include "amdgpu_hdp.h"
+#include "amdgpu_dm.h"
 #include "amdgpu_virt.h"
+#include "amdgpu_csa.h"
+#include "amdgpu_mes_ctx.h"
+#include "amdgpu_gart.h"
+#include "amdgpu_debugfs.h"
+#include "amdgpu_job.h"
+#include "amdgpu_bo_list.h"
+#include "amdgpu_gem.h"
+#include "amdgpu_doorbell.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_discovery.h"
+#include "amdgpu_mes.h"
+#include "amdgpu_umc.h"
+#include "amdgpu_mmhub.h"
+#include "amdgpu_gfxhub.h"
+#include "amdgpu_df.h"
+#include "amdgpu_smuio.h"
+#include "amdgpu_fdinfo.h"
+#include "amdgpu_mca.h"
+#include "amdgpu_aca.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_cper.h"
+#include "amdgpu_xcp.h"
+#include "amdgpu_seq64.h"
+#include "amdgpu_reg_state.h"
+#include "amdgpu_userq.h"
+#include "amdgpu_eviction_fence.h"
+#if defined(CONFIG_DRM_AMD_ISP)
+#include "amdgpu_isp.h"
+#endif
+
+#define MAX_GPU_INSTANCE		64
+
+#define GFX_SLICE_PERIOD_MS		250
+
+struct amdgpu_gpu_instance {
+	struct amdgpu_device		*adev;
+	int				mgpu_fan_enabled;
+};
+
+struct amdgpu_mgpu_info {
+	struct amdgpu_gpu_instance	gpu_ins[MAX_GPU_INSTANCE];
+	struct mutex			mutex;
+	uint32_t			num_gpu;
+	uint32_t			num_dgpu;
+	uint32_t			num_apu;
+};
+
+enum amdgpu_ss {
+	AMDGPU_SS_DRV_LOAD,
+	AMDGPU_SS_DEV_D0,
+	AMDGPU_SS_DEV_D3,
+	AMDGPU_SS_DRV_UNLOAD
+};
+
+struct amdgpu_hwip_reg_entry {
+	u32		hwip;
+	u32		inst;
+	u32		seg;
+	u32		reg_offset;
+	const char	*reg_name;
+};
+
+struct amdgpu_watchdog_timer {
+	bool timeout_fatal_disable;
+	uint32_t period; /* maxCycles = (1 << period), the number of cycles before a timeout */
+};
+
+#define AMDGPU_MAX_TIMEOUT_PARAM_LENGTH	256
 
 /*
  * Modules parameters.
  */
 extern int amdgpu_modeset;
-extern int amdgpu_vram_limit;
+extern unsigned int amdgpu_vram_limit;
+extern int amdgpu_vis_vram_limit;
 extern int amdgpu_gart_size;
+extern int amdgpu_gtt_size;
 extern int amdgpu_moverate;
-extern int amdgpu_benchmarking;
-extern int amdgpu_testing;
 extern int amdgpu_audio;
 extern int amdgpu_disp_priority;
 extern int amdgpu_hw_i2c;
 extern int amdgpu_pcie_gen2;
 extern int amdgpu_msi;
-extern int amdgpu_lockup_timeout;
+extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH];
 extern int amdgpu_dpm;
 extern int amdgpu_fw_load_type;
 extern int amdgpu_aspm;
 extern int amdgpu_runtime_pm;
-extern unsigned amdgpu_ip_block_mask;
+extern uint amdgpu_ip_block_mask;
 extern int amdgpu_bapm;
 extern int amdgpu_deep_color;
 extern int amdgpu_vm_size;
 extern int amdgpu_vm_block_size;
+extern int amdgpu_vm_fragment_size;
 extern int amdgpu_vm_fault_stop;
 extern int amdgpu_vm_debug;
+extern int amdgpu_vm_update_mode;
+extern int amdgpu_exp_hw_support;
+extern int amdgpu_dc;
 extern int amdgpu_sched_jobs;
 extern int amdgpu_sched_hw_submission;
-extern int amdgpu_no_evict;
-extern int amdgpu_direct_gma_size;
-extern unsigned amdgpu_pcie_gen_cap;
-extern unsigned amdgpu_pcie_lane_cap;
-extern unsigned amdgpu_cg_mask;
-extern unsigned amdgpu_pg_mask;
+extern uint amdgpu_pcie_gen_cap;
+extern uint amdgpu_pcie_lane_cap;
+extern u64 amdgpu_cg_mask;
+extern uint amdgpu_pg_mask;
+extern uint amdgpu_sdma_phase_quantum;
 extern char *amdgpu_disable_cu;
 extern char *amdgpu_virtual_display;
-extern unsigned amdgpu_pp_feature_mask;
-extern int amdgpu_vram_page_split;
-extern int amdgpu_ngg;
-extern int amdgpu_prim_buf_per_se;
-extern int amdgpu_pos_buf_per_se;
-extern int amdgpu_cntl_sb_buf_per_se;
-extern int amdgpu_param_buf_per_se;
-
-#define AMDGPU_DEFAULT_GTT_SIZE_MB		3072ULL /* 3GB by default */
+extern uint amdgpu_pp_feature_mask;
+extern uint amdgpu_force_long_training;
+extern int amdgpu_lbpw;
+extern int amdgpu_compute_multipipe;
+extern int amdgpu_gpu_recovery;
+extern int amdgpu_emu_mode;
+extern uint amdgpu_smu_memory_pool_size;
+extern int amdgpu_smu_pptable_id;
+extern uint amdgpu_dc_feature_mask;
+extern uint amdgpu_freesync_vid_mode;
+extern uint amdgpu_dc_debug_mask;
+extern uint amdgpu_dc_visual_confirm;
+extern int amdgpu_dm_abm_level;
+extern int amdgpu_backlight;
+extern int amdgpu_damage_clips;
+extern struct amdgpu_mgpu_info mgpu_info;
+extern int amdgpu_ras_enable;
+extern uint amdgpu_ras_mask;
+extern int amdgpu_bad_page_threshold;
+extern bool amdgpu_ignore_bad_page_threshold;
+extern struct amdgpu_watchdog_timer amdgpu_watchdog_timer;
+extern int amdgpu_async_gfx_ring;
+extern int amdgpu_mcbp;
+extern int amdgpu_discovery;
+extern int amdgpu_mes;
+extern int amdgpu_mes_log_enable;
+extern int amdgpu_mes_kiq;
+extern int amdgpu_uni_mes;
+extern int amdgpu_noretry;
+extern int amdgpu_force_asic_type;
+extern int amdgpu_smartshift_bias;
+extern int amdgpu_use_xgmi_p2p;
+extern int amdgpu_mtype_local;
+extern int amdgpu_enforce_isolation;
+#ifdef CONFIG_HSA_AMD
+extern int sched_policy;
+extern bool debug_evictions;
+extern bool no_system_mem_limit;
+extern int halt_if_hws_hang;
+extern uint amdgpu_svm_default_granularity;
+#else
+static const int __maybe_unused sched_policy = KFD_SCHED_POLICY_HWS;
+static const bool __maybe_unused debug_evictions; /* = false */
+static const bool __maybe_unused no_system_mem_limit;
+static const int __maybe_unused halt_if_hws_hang;
+#endif
+#ifdef CONFIG_HSA_AMD_P2P
+extern bool pcie_p2p;
+#endif
+
+extern int amdgpu_tmz;
+extern int amdgpu_reset_method;
+
+#ifdef CONFIG_DRM_AMDGPU_SI
+extern int amdgpu_si_support;
+#endif
+#ifdef CONFIG_DRM_AMDGPU_CIK
+extern int amdgpu_cik_support;
+#endif
+extern int amdgpu_num_kcq;
+
+#define AMDGPU_VCNFW_LOG_SIZE (32 * 1024)
+#define AMDGPU_UMSCHFW_LOG_SIZE (32 * 1024)
+extern int amdgpu_vcnfw_log;
+extern int amdgpu_sg_display;
+extern int amdgpu_umsch_mm;
+extern int amdgpu_seamless;
+extern int amdgpu_umsch_mm_fwlog;
+
+extern int amdgpu_user_partt_mode;
+extern int amdgpu_agp;
+extern int amdgpu_rebar;
+
+extern int amdgpu_wbrf;
+extern int amdgpu_user_queue;
+
+#define AMDGPU_VM_MAX_NUM_CTX			4096
+#define AMDGPU_SG_THRESHOLD			(256*1024*1024)
 #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS	        3000
 #define AMDGPU_MAX_USEC_TIMEOUT			100000	/* 100 ms */
 #define AMDGPU_FENCE_JIFFIES_TIMEOUT		(HZ / 2)
-/* AMDGPU_IB_POOL_SIZE must be a power of 2 */
-#define AMDGPU_IB_POOL_SIZE			16
 #define AMDGPU_DEBUGFS_MAX_COMPONENTS		32
 #define AMDGPUFB_CONN_LIMIT			4
 #define AMDGPU_BIOS_NUM_SCRATCH			16
 
-/* max number of IP instances */
-#define AMDGPU_MAX_SDMA_INSTANCES		2
+#define AMDGPU_VBIOS_VGA_ALLOCATION		(9 * 1024 * 1024) /* reserve 8MB for vga emulator and 1 MB for FB */
 
 /* hard reset data */
 #define AMDGPU_ASIC_RESET_DATA                  0x39d5e86b
@@ -143,26 +305,36 @@ extern int amdgpu_param_buf_per_se;
 #define AMDGPU_RESET_VCE			(1 << 13)
 #define AMDGPU_RESET_VCE1			(1 << 14)
 
-/* GFX current status */
-#define AMDGPU_GFX_NORMAL_MODE			0x00000000L
-#define AMDGPU_GFX_SAFE_MODE			0x00000001L
-#define AMDGPU_GFX_PG_DISABLED_MODE		0x00000002L
-#define AMDGPU_GFX_CG_DISABLED_MODE		0x00000004L
-#define AMDGPU_GFX_LBPW_DISABLED_MODE		0x00000008L
+/* reset mask */
+#define AMDGPU_RESET_TYPE_FULL (1 << 0) /* full adapter reset, mode1/mode2/BACO/etc. */
+#define AMDGPU_RESET_TYPE_SOFT_RESET (1 << 1) /* IP level soft reset */
+#define AMDGPU_RESET_TYPE_PER_QUEUE (1 << 2) /* per queue */
+#define AMDGPU_RESET_TYPE_PER_PIPE (1 << 3) /* per pipe */
 
 /* max cursor sizes (in pixels) */
 #define CIK_CURSOR_WIDTH 128
 #define CIK_CURSOR_HEIGHT 128
 
+/* smart shift bias level limits */
+#define AMDGPU_SMARTSHIFT_MAX_BIAS (100)
+#define AMDGPU_SMARTSHIFT_MIN_BIAS (-100)
+
+/* Extra time delay(in ms) to eliminate the influence of temperature momentary fluctuation */
+#define AMDGPU_SWCTF_EXTRA_DELAY		50
+
+struct amdgpu_xcp_mgr;
 struct amdgpu_device;
-struct amdgpu_ib;
-struct amdgpu_cs_parser;
-struct amdgpu_job;
 struct amdgpu_irq_src;
 struct amdgpu_fpriv;
+struct amdgpu_bo_va_mapping;
+struct kfd_vm_fault_info;
+struct amdgpu_hive_info;
+struct amdgpu_reset_context;
+struct amdgpu_reset_control;
 
 enum amdgpu_cp_irq {
-	AMDGPU_CP_IRQ_GFX_EOP = 0,
+	AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP = 0,
+	AMDGPU_CP_IRQ_GFX_ME0_PIPE1_EOP,
 	AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP,
 	AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP,
 	AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP,
@@ -175,13 +347,6 @@ enum amdgpu_cp_irq {
 	AMDGPU_CP_IRQ_LAST
 };
 
-enum amdgpu_sdma_irq {
-	AMDGPU_SDMA_IRQ_TRAP0 = 0,
-	AMDGPU_SDMA_IRQ_TRAP1,
-
-	AMDGPU_SDMA_IRQ_LAST
-};
-
 enum amdgpu_thermal_irq {
 	AMDGPU_THERMAL_IRQ_LOW_TO_HIGH = 0,
 	AMDGPU_THERMAL_IRQ_HIGH_TO_LOW,
@@ -193,20 +358,29 @@ enum amdgpu_kiq_irq {
 	AMDGPU_CP_KIQ_IRQ_DRIVER0 = 0,
 	AMDGPU_CP_KIQ_IRQ_LAST
 };
-
-int amdgpu_set_clockgating_state(struct amdgpu_device *adev,
-				  enum amd_ip_block_type block_type,
-				  enum amd_clockgating_state state);
-int amdgpu_set_powergating_state(struct amdgpu_device *adev,
-				  enum amd_ip_block_type block_type,
-				  enum amd_powergating_state state);
-void amdgpu_get_clockgating_state(struct amdgpu_device *adev, u32 *flags);
-int amdgpu_wait_for_idle(struct amdgpu_device *adev,
-			 enum amd_ip_block_type block_type);
-bool amdgpu_is_idle(struct amdgpu_device *adev,
-		    enum amd_ip_block_type block_type);
-
-#define AMDGPU_MAX_IP_NUM 16
+#define MAX_KIQ_REG_WAIT       5000 /* in usecs, 5ms */
+#define MAX_KIQ_REG_BAILOUT_INTERVAL   5 /* in msecs, 5ms */
+#define MAX_KIQ_REG_TRY 1000
+
+int amdgpu_device_ip_set_clockgating_state(void *dev,
+					   enum amd_ip_block_type block_type,
+					   enum amd_clockgating_state state);
+int amdgpu_device_ip_set_powergating_state(void *dev,
+					   enum amd_ip_block_type block_type,
+					   enum amd_powergating_state state);
+void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev,
+					    u64 *flags);
+int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev,
+				   enum amd_ip_block_type block_type);
+bool amdgpu_device_ip_is_hw(struct amdgpu_device *adev,
+			    enum amd_ip_block_type block_type);
+bool amdgpu_device_ip_is_valid(struct amdgpu_device *adev,
+			      enum amd_ip_block_type block_type);
+int amdgpu_ip_block_suspend(struct amdgpu_ip_block *ip_block);
+
+int amdgpu_ip_block_resume(struct amdgpu_ip_block *ip_block);
+
+#define AMDGPU_MAX_IP_NUM AMD_IP_BLOCK_TYPE_NUM
 
 struct amdgpu_ip_block_status {
 	bool valid;
@@ -227,115 +401,28 @@ struct amdgpu_ip_block_version {
 struct amdgpu_ip_block {
 	struct amdgpu_ip_block_status status;
 	const struct amdgpu_ip_block_version *version;
+	struct amdgpu_device *adev;
 };
 
-int amdgpu_ip_block_version_cmp(struct amdgpu_device *adev,
-				enum amd_ip_block_type type,
-				u32 major, u32 minor);
-
-struct amdgpu_ip_block * amdgpu_get_ip_block(struct amdgpu_device *adev,
-					     enum amd_ip_block_type type);
-
-int amdgpu_ip_block_add(struct amdgpu_device *adev,
-			const struct amdgpu_ip_block_version *ip_block_version);
-
-/* provided by hw blocks that can move/clear data.  e.g., gfx or sdma */
-struct amdgpu_buffer_funcs {
-	/* maximum bytes in a single operation */
-	uint32_t	copy_max_bytes;
-
-	/* number of dw to reserve per operation */
-	unsigned	copy_num_dw;
-
-	/* used for buffer migration */
-	void (*emit_copy_buffer)(struct amdgpu_ib *ib,
-				 /* src addr in bytes */
-				 uint64_t src_offset,
-				 /* dst addr in bytes */
-				 uint64_t dst_offset,
-				 /* number of byte to transfer */
-				 uint32_t byte_count);
-
-	/* maximum bytes in a single operation */
-	uint32_t	fill_max_bytes;
-
-	/* number of dw to reserve per operation */
-	unsigned	fill_num_dw;
-
-	/* used for buffer clearing */
-	void (*emit_fill_buffer)(struct amdgpu_ib *ib,
-				 /* value to write to memory */
-				 uint32_t src_data,
-				 /* dst addr in bytes */
-				 uint64_t dst_offset,
-				 /* number of byte to fill */
-				 uint32_t byte_count);
-};
+int amdgpu_device_ip_block_version_cmp(struct amdgpu_device *adev,
+				       enum amd_ip_block_type type,
+				       u32 major, u32 minor);
 
-/* provided by hw blocks that can write ptes, e.g., sdma */
-struct amdgpu_vm_pte_funcs {
-	/* copy pte entries from GART */
-	void (*copy_pte)(struct amdgpu_ib *ib,
-			 uint64_t pe, uint64_t src,
-			 unsigned count);
-	/* write pte one entry at a time with addr mapping */
-	void (*write_pte)(struct amdgpu_ib *ib, uint64_t pe,
-			  uint64_t value, unsigned count,
-			  uint32_t incr);
-	/* for linear pte/pde updates without addr mapping */
-	void (*set_pte_pde)(struct amdgpu_ib *ib,
-			    uint64_t pe,
-			    uint64_t addr, unsigned count,
-			    uint32_t incr, uint64_t flags);
-};
-
-/* provided by the gmc block */
-struct amdgpu_gart_funcs {
-	/* flush the vm tlb via mmio */
-	void (*flush_gpu_tlb)(struct amdgpu_device *adev,
-			      uint32_t vmid);
-	/* write pte/pde updates using the cpu */
-	int (*set_pte_pde)(struct amdgpu_device *adev,
-			   void *cpu_pt_addr, /* cpu addr of page table */
-			   uint32_t gpu_page_idx, /* pte/pde to update */
-			   uint64_t addr, /* addr to write into pte/pde */
-			   uint64_t flags); /* access flags */
-	/* enable/disable PRT support */
-	void (*set_prt)(struct amdgpu_device *adev, bool enable);
-	/* set pte flags based per asic */
-	uint64_t (*get_vm_pte_flags)(struct amdgpu_device *adev,
-				     uint32_t flags);
-	/* adjust mc addr in fb for APU case */
-	u64 (*adjust_mc_addr)(struct amdgpu_device *adev, u64 addr);
-	uint32_t (*get_invalidate_req)(unsigned int vm_id);
-};
+struct amdgpu_ip_block *
+amdgpu_device_ip_get_ip_block(struct amdgpu_device *adev,
+			      enum amd_ip_block_type type);
 
-/* provided by the ih block */
-struct amdgpu_ih_funcs {
-	/* ring read/write ptr handling, called from interrupt context */
-	u32 (*get_wptr)(struct amdgpu_device *adev);
-	void (*decode_iv)(struct amdgpu_device *adev,
-			  struct amdgpu_iv_entry *entry);
-	void (*set_rptr)(struct amdgpu_device *adev);
-};
+int amdgpu_device_ip_block_add(struct amdgpu_device *adev,
+			       const struct amdgpu_ip_block_version *ip_block_version);
 
 /*
  * BIOS.
  */
 bool amdgpu_get_bios(struct amdgpu_device *adev);
 bool amdgpu_read_bios(struct amdgpu_device *adev);
-
-/*
- * Dummy page
- */
-struct amdgpu_dummy_page {
-	struct page	*page;
-	dma_addr_t	addr;
-};
-int amdgpu_dummy_page_init(struct amdgpu_device *adev);
-void amdgpu_dummy_page_fini(struct amdgpu_device *adev);
-
-
+bool amdgpu_soc15_read_bios_from_rom(struct amdgpu_device *adev,
+				     u8 *bios, u32 length_bytes);
+void amdgpu_bios_release(struct amdgpu_device *adev);
 /*
  * Clocks
  */
@@ -350,107 +437,10 @@ struct amdgpu_clock {
 	uint32_t default_mclk;
 	uint32_t default_sclk;
 	uint32_t default_dispclk;
-	uint32_t current_dispclk;
 	uint32_t dp_extclk;
 	uint32_t max_pixel_clock;
 };
 
-/*
- * BO.
- */
-struct amdgpu_bo_list_entry {
-	struct amdgpu_bo		*robj;
-	struct ttm_validate_buffer	tv;
-	struct amdgpu_bo_va		*bo_va;
-	uint32_t			priority;
-	struct page			**user_pages;
-	int				user_invalidated;
-};
-
-struct amdgpu_bo_va_mapping {
-	struct list_head		list;
-	struct rb_node			rb;
-	uint64_t			start;
-	uint64_t			last;
-	uint64_t			__subtree_last;
-	uint64_t			offset;
-	uint64_t			flags;
-};
-
-/* bo virtual addresses in a specific vm */
-struct amdgpu_bo_va {
-	/* protected by bo being reserved */
-	struct list_head		bo_list;
-	struct dma_fence	        *last_pt_update;
-	unsigned			ref_count;
-
-	/* protected by vm mutex and spinlock */
-	struct list_head		vm_status;
-
-	/* mappings for this bo_va */
-	struct list_head		invalids;
-	struct list_head		valids;
-
-	/* constant after initialization */
-	struct amdgpu_vm		*vm;
-	struct amdgpu_bo		*bo;
-};
-
-#define AMDGPU_GEM_DOMAIN_MAX		0x3
-
-struct amdgpu_bo {
-	/* Protected by tbo.reserved */
-	u32				prefered_domains;
-	u32				allowed_domains;
-	struct ttm_place		placements[AMDGPU_GEM_DOMAIN_MAX + 1];
-	struct ttm_placement		placement;
-	struct ttm_buffer_object	tbo;
-	struct ttm_bo_kmap_obj		kmap;
-	u64				flags;
-	unsigned			pin_count;
-	void				*kptr;
-	u64				tiling_flags;
-	u64				metadata_flags;
-	void				*metadata;
-	u32				metadata_size;
-	unsigned			prime_shared_count;
-	/* list of all virtual address to which this bo
-	 * is associated to
-	 */
-	struct list_head		va;
-	/* Constant after initialization */
-	struct drm_gem_object		gem_base;
-	struct amdgpu_bo		*parent;
-	struct amdgpu_bo		*shadow;
-
-	struct ttm_bo_kmap_obj		dma_buf_vmap;
-	struct amdgpu_mn		*mn;
-	struct list_head		mn_list;
-	struct list_head		shadow_list;
-};
-#define gem_to_amdgpu_bo(gobj) container_of((gobj), struct amdgpu_bo, gem_base)
-
-void amdgpu_gem_object_free(struct drm_gem_object *obj);
-int amdgpu_gem_object_open(struct drm_gem_object *obj,
-				struct drm_file *file_priv);
-void amdgpu_gem_object_close(struct drm_gem_object *obj,
-				struct drm_file *file_priv);
-unsigned long amdgpu_gem_timeout(uint64_t timeout_ns);
-struct sg_table *amdgpu_gem_prime_get_sg_table(struct drm_gem_object *obj);
-struct drm_gem_object *
-amdgpu_gem_prime_import_sg_table(struct drm_device *dev,
-				 struct dma_buf_attachment *attach,
-				 struct sg_table *sg);
-struct dma_buf *amdgpu_gem_prime_export(struct drm_device *dev,
-					struct drm_gem_object *gobj,
-					int flags);
-int amdgpu_gem_prime_pin(struct drm_gem_object *obj);
-void amdgpu_gem_prime_unpin(struct drm_gem_object *obj);
-struct reservation_object *amdgpu_gem_prime_res_obj(struct drm_gem_object *);
-void *amdgpu_gem_prime_vmap(struct drm_gem_object *obj);
-void amdgpu_gem_prime_vunmap(struct drm_gem_object *obj, void *vaddr);
-int amdgpu_gem_debugfs_init(struct amdgpu_device *adev);
-
 /* sub-allocation manager, it has to be protected by another lock.
  * By conception this is an helper for other part of the driver
  * like the indirect buffer or semaphore, which both have their
@@ -475,257 +465,14 @@ int amdgpu_gem_debugfs_init(struct amdgpu_device *adev);
  * alignment).
  */
 
-#define AMDGPU_SA_NUM_FENCE_LISTS	32
-
 struct amdgpu_sa_manager {
-	wait_queue_head_t	wq;
-	struct amdgpu_bo	*bo;
-	struct list_head	*hole;
-	struct list_head	flist[AMDGPU_SA_NUM_FENCE_LISTS];
-	struct list_head	olist;
-	unsigned		size;
-	uint64_t		gpu_addr;
-	void			*cpu_ptr;
-	uint32_t		domain;
-	uint32_t		align;
-};
-
-/* sub-allocation buffer */
-struct amdgpu_sa_bo {
-	struct list_head		olist;
-	struct list_head		flist;
-	struct amdgpu_sa_manager	*manager;
-	unsigned			soffset;
-	unsigned			eoffset;
-	struct dma_fence	        *fence;
-};
-
-/*
- * GEM objects.
- */
-void amdgpu_gem_force_release(struct amdgpu_device *adev);
-int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
-				int alignment, u32 initial_domain,
-				u64 flags, bool kernel,
-				struct drm_gem_object **obj);
-
-int amdgpu_mode_dumb_create(struct drm_file *file_priv,
-			    struct drm_device *dev,
-			    struct drm_mode_create_dumb *args);
-int amdgpu_mode_dumb_mmap(struct drm_file *filp,
-			  struct drm_device *dev,
-			  uint32_t handle, uint64_t *offset_p);
-int amdgpu_fence_slab_init(void);
-void amdgpu_fence_slab_fini(void);
-
-/*
- * GART structures, functions & helpers
- */
-struct amdgpu_mc;
-
-#define AMDGPU_GPU_PAGE_SIZE 4096
-#define AMDGPU_GPU_PAGE_MASK (AMDGPU_GPU_PAGE_SIZE - 1)
-#define AMDGPU_GPU_PAGE_SHIFT 12
-#define AMDGPU_GPU_PAGE_ALIGN(a) (((a) + AMDGPU_GPU_PAGE_MASK) & ~AMDGPU_GPU_PAGE_MASK)
-
-struct amdgpu_gart {
-	dma_addr_t			table_addr;
-	struct amdgpu_bo		*robj;
-	void				*ptr;
-	unsigned			num_gpu_pages;
-	unsigned			num_cpu_pages;
-	unsigned			table_size;
-#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
-	struct page			**pages;
-#endif
-	bool				ready;
-
-	/* Asic default pte flags */
-	uint64_t			gart_pte_flags;
-
-	const struct amdgpu_gart_funcs *gart_funcs;
-};
-
-int amdgpu_gart_table_ram_alloc(struct amdgpu_device *adev);
-void amdgpu_gart_table_ram_free(struct amdgpu_device *adev);
-int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev);
-void amdgpu_gart_table_vram_free(struct amdgpu_device *adev);
-int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev);
-void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev);
-int amdgpu_gart_init(struct amdgpu_device *adev);
-void amdgpu_gart_fini(struct amdgpu_device *adev);
-void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset,
-			int pages);
-int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
-		     int pages, struct page **pagelist,
-		     dma_addr_t *dma_addr, uint64_t flags);
-int amdgpu_ttm_recover_gart(struct amdgpu_device *adev);
-
-/*
- * VMHUB structures, functions & helpers
- */
-struct amdgpu_vmhub {
-	uint32_t	ctx0_ptb_addr_lo32;
-	uint32_t	ctx0_ptb_addr_hi32;
-	uint32_t	vm_inv_eng0_req;
-	uint32_t	vm_inv_eng0_ack;
-	uint32_t	vm_context0_cntl;
-	uint32_t	vm_l2_pro_fault_status;
-	uint32_t	vm_l2_pro_fault_cntl;
-};
-
-/*
- * GPU MC structures, functions & helpers
- */
-struct amdgpu_mc {
-	resource_size_t		aper_size;
-	resource_size_t		aper_base;
-	resource_size_t		agp_base;
-	/* for some chips with <= 32MB we need to lie
-	 * about vram size near mc fb location */
-	u64			mc_vram_size;
-	u64			visible_vram_size;
-	u64			gtt_size;
-	u64			gtt_start;
-	u64			gtt_end;
-	u64			vram_start;
-	u64			vram_end;
-	unsigned		vram_width;
-	u64			real_vram_size;
-	int			vram_mtrr;
-	u64                     gtt_base_align;
-	u64                     mc_mask;
-	const struct firmware   *fw;	/* MC firmware */
-	uint32_t                fw_version;
-	struct amdgpu_irq_src	vm_fault;
-	uint32_t		vram_type;
-	uint32_t                srbm_soft_reset;
-	struct amdgpu_mode_mc_save save;
-	bool			prt_warning;
-	/* apertures */
-	u64					shared_aperture_start;
-	u64					shared_aperture_end;
-	u64					private_aperture_start;
-	u64					private_aperture_end;
-	/* protects concurrent invalidation */
-	spinlock_t		invalidate_lock;
-};
-
-/*
- * GPU doorbell structures, functions & helpers
- */
-typedef enum _AMDGPU_DOORBELL_ASSIGNMENT
-{
-	AMDGPU_DOORBELL_KIQ                     = 0x000,
-	AMDGPU_DOORBELL_HIQ                     = 0x001,
-	AMDGPU_DOORBELL_DIQ                     = 0x002,
-	AMDGPU_DOORBELL_MEC_RING0               = 0x010,
-	AMDGPU_DOORBELL_MEC_RING1               = 0x011,
-	AMDGPU_DOORBELL_MEC_RING2               = 0x012,
-	AMDGPU_DOORBELL_MEC_RING3               = 0x013,
-	AMDGPU_DOORBELL_MEC_RING4               = 0x014,
-	AMDGPU_DOORBELL_MEC_RING5               = 0x015,
-	AMDGPU_DOORBELL_MEC_RING6               = 0x016,
-	AMDGPU_DOORBELL_MEC_RING7               = 0x017,
-	AMDGPU_DOORBELL_GFX_RING0               = 0x020,
-	AMDGPU_DOORBELL_sDMA_ENGINE0            = 0x1E0,
-	AMDGPU_DOORBELL_sDMA_ENGINE1            = 0x1E1,
-	AMDGPU_DOORBELL_IH                      = 0x1E8,
-	AMDGPU_DOORBELL_MAX_ASSIGNMENT          = 0x3FF,
-	AMDGPU_DOORBELL_INVALID                 = 0xFFFF
-} AMDGPU_DOORBELL_ASSIGNMENT;
-
-struct amdgpu_doorbell {
-	/* doorbell mmio */
-	resource_size_t		base;
-	resource_size_t		size;
-	u32 __iomem		*ptr;
-	u32			num_doorbells;	/* Number of doorbells actually reserved for amdgpu. */
+	struct drm_suballoc_manager	base;
+	struct amdgpu_bo		*bo;
+	uint64_t			gpu_addr;
+	void				*cpu_ptr;
 };
 
 /*
- * 64bit doorbell, offset are in QWORD, occupy 2KB doorbell space
- */
-typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT
-{
-	/*
-	 * All compute related doorbells: kiq, hiq, diq, traditional compute queue, user queue, should locate in
-	 * a continues range so that programming CP_MEC_DOORBELL_RANGE_LOWER/UPPER can cover this range.
-	 *  Compute related doorbells are allocated from 0x00 to 0x8a
-	 */
-
-
-	/* kernel scheduling */
-	AMDGPU_DOORBELL64_KIQ                     = 0x00,
-
-	/* HSA interface queue and debug queue */
-	AMDGPU_DOORBELL64_HIQ                     = 0x01,
-	AMDGPU_DOORBELL64_DIQ                     = 0x02,
-
-	/* Compute engines */
-	AMDGPU_DOORBELL64_MEC_RING0               = 0x03,
-	AMDGPU_DOORBELL64_MEC_RING1               = 0x04,
-	AMDGPU_DOORBELL64_MEC_RING2               = 0x05,
-	AMDGPU_DOORBELL64_MEC_RING3               = 0x06,
-	AMDGPU_DOORBELL64_MEC_RING4               = 0x07,
-	AMDGPU_DOORBELL64_MEC_RING5               = 0x08,
-	AMDGPU_DOORBELL64_MEC_RING6               = 0x09,
-	AMDGPU_DOORBELL64_MEC_RING7               = 0x0a,
-
-	/* User queue doorbell range (128 doorbells) */
-	AMDGPU_DOORBELL64_USERQUEUE_START         = 0x0b,
-	AMDGPU_DOORBELL64_USERQUEUE_END           = 0x8a,
-
-	/* Graphics engine */
-	AMDGPU_DOORBELL64_GFX_RING0               = 0x8b,
-
-	/*
-	 * Other graphics doorbells can be allocated here: from 0x8c to 0xef
-	 * Graphics voltage island aperture 1
-	 * default non-graphics QWORD index is 0xF0 - 0xFF inclusive
-	 */
-
-	/* sDMA engines */
-	AMDGPU_DOORBELL64_sDMA_ENGINE0            = 0xF0,
-	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE0     = 0xF1,
-	AMDGPU_DOORBELL64_sDMA_ENGINE1            = 0xF2,
-	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE1     = 0xF3,
-
-	/* Interrupt handler */
-	AMDGPU_DOORBELL64_IH                      = 0xF4,  /* For legacy interrupt ring buffer */
-	AMDGPU_DOORBELL64_IH_RING1                = 0xF5,  /* For page migration request log */
-	AMDGPU_DOORBELL64_IH_RING2                = 0xF6,  /* For page migration translation/invalidation log */
-
-	/* VCN engine use 32 bits doorbell  */
-	AMDGPU_DOORBELL64_VCN0_1                  = 0xF8, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */
-	AMDGPU_DOORBELL64_VCN2_3                  = 0xF9,
-	AMDGPU_DOORBELL64_VCN4_5                  = 0xFA,
-	AMDGPU_DOORBELL64_VCN6_7                  = 0xFB,
-
-	/* overlap the doorbell assignment with VCN as they are  mutually exclusive
-	 * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD
-	 */
-	AMDGPU_DOORBELL64_RING0_1                 = 0xF8,
-	AMDGPU_DOORBELL64_RING2_3                 = 0xF9,
-	AMDGPU_DOORBELL64_RING4_5                 = 0xFA,
-	AMDGPU_DOORBELL64_RING6_7                 = 0xFB,
-
-	AMDGPU_DOORBELL64_UVD_RING0_1             = 0xFC,
-	AMDGPU_DOORBELL64_UVD_RING2_3             = 0xFD,
-	AMDGPU_DOORBELL64_UVD_RING4_5             = 0xFE,
-	AMDGPU_DOORBELL64_UVD_RING6_7             = 0xFF,
-
-	AMDGPU_DOORBELL64_MAX_ASSIGNMENT          = 0xFF,
-	AMDGPU_DOORBELL64_INVALID                 = 0xFFFF
-} AMDGPU_DOORBELL64_ASSIGNMENT;
-
-
-void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev,
-				phys_addr_t *aperture_base,
-				size_t *aperture_size,
-				size_t *start_offset);
-
-/*
  * IRQS.
  */
 
@@ -738,80 +485,12 @@ struct amdgpu_flip_work {
 	uint64_t			base;
 	struct drm_pending_vblank_event *event;
 	struct amdgpu_bo		*old_abo;
-	struct dma_fence		*excl;
 	unsigned			shared_count;
 	struct dma_fence		**shared;
 	struct dma_fence_cb		cb;
 	bool				async;
 };
 
-
-/*
- * CP & rings.
- */
-
-struct amdgpu_ib {
-	struct amdgpu_sa_bo		*sa_bo;
-	uint32_t			length_dw;
-	uint64_t			gpu_addr;
-	uint32_t			*ptr;
-	uint32_t			flags;
-};
-
-extern const struct amd_sched_backend_ops amdgpu_sched_ops;
-
-int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
-		     struct amdgpu_job **job, struct amdgpu_vm *vm);
-int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
-			     struct amdgpu_job **job);
-
-void amdgpu_job_free_resources(struct amdgpu_job *job);
-void amdgpu_job_free(struct amdgpu_job *job);
-int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
-		      struct amd_sched_entity *entity, void *owner,
-		      struct dma_fence **f);
-
-/*
- * context related structures
- */
-
-struct amdgpu_ctx_ring {
-	uint64_t		sequence;
-	struct dma_fence	**fences;
-	struct amd_sched_entity	entity;
-};
-
-struct amdgpu_ctx {
-	struct kref		refcount;
-	struct amdgpu_device    *adev;
-	unsigned		reset_counter;
-	spinlock_t		ring_lock;
-	struct dma_fence	**fences;
-	struct amdgpu_ctx_ring	rings[AMDGPU_MAX_RINGS];
-	bool preamble_presented;
-};
-
-struct amdgpu_ctx_mgr {
-	struct amdgpu_device	*adev;
-	struct mutex		lock;
-	/* protected by lock */
-	struct idr		ctx_handles;
-};
-
-struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
-int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
-
-uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
-			      struct dma_fence *fence);
-struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
-				   struct amdgpu_ring *ring, uint64_t seq);
-
-int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
-		     struct drm_file *filp);
-
-void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr);
-void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
-
 /*
  * file private structure
  */
@@ -819,487 +498,160 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
 struct amdgpu_fpriv {
 	struct amdgpu_vm	vm;
 	struct amdgpu_bo_va	*prt_va;
+	struct amdgpu_bo_va	*csa_va;
+	struct amdgpu_bo_va	*seq64_va;
 	struct mutex		bo_list_lock;
 	struct idr		bo_list_handles;
 	struct amdgpu_ctx_mgr	ctx_mgr;
-};
+	struct amdgpu_userq_mgr	userq_mgr;
 
-/*
- * residency list
- */
+	/* Eviction fence infra */
+	struct amdgpu_eviction_fence_mgr evf_mgr;
 
-struct amdgpu_bo_list {
-	struct mutex lock;
-	struct amdgpu_bo *gds_obj;
-	struct amdgpu_bo *gws_obj;
-	struct amdgpu_bo *oa_obj;
-	unsigned first_userptr;
-	unsigned num_entries;
-	struct amdgpu_bo_list_entry *array;
+	/** GPU partition selection */
+	uint32_t		xcp_id;
 };
 
-struct amdgpu_bo_list *
-amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id);
-void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
-			     struct list_head *validated);
-void amdgpu_bo_list_put(struct amdgpu_bo_list *list);
-void amdgpu_bo_list_free(struct amdgpu_bo_list *list);
-
-/*
- * GFX stuff
- */
-#include "clearstate_defs.h"
-
-struct amdgpu_rlc_funcs {
-	void (*enter_safe_mode)(struct amdgpu_device *adev);
-	void (*exit_safe_mode)(struct amdgpu_device *adev);
-};
-
-struct amdgpu_rlc {
-	/* for power gating */
-	struct amdgpu_bo	*save_restore_obj;
-	uint64_t		save_restore_gpu_addr;
-	volatile uint32_t	*sr_ptr;
-	const u32               *reg_list;
-	u32                     reg_list_size;
-	/* for clear state */
-	struct amdgpu_bo	*clear_state_obj;
-	uint64_t		clear_state_gpu_addr;
-	volatile uint32_t	*cs_ptr;
-	const struct cs_section_def   *cs_data;
-	u32                     clear_state_size;
-	/* for cp tables */
-	struct amdgpu_bo	*cp_table_obj;
-	uint64_t		cp_table_gpu_addr;
-	volatile uint32_t	*cp_table_ptr;
-	u32                     cp_table_size;
-
-	/* safe mode for updating CG/PG state */
-	bool in_safe_mode;
-	const struct amdgpu_rlc_funcs *funcs;
-
-	/* for firmware data */
-	u32 save_and_restore_offset;
-	u32 clear_state_descriptor_offset;
-	u32 avail_scratch_ram_locations;
-	u32 reg_restore_list_size;
-	u32 reg_list_format_start;
-	u32 reg_list_format_separate_start;
-	u32 starting_offsets_start;
-	u32 reg_list_format_size_bytes;
-	u32 reg_list_size_bytes;
-
-	u32 *register_list_format;
-	u32 *register_restore;
-};
-
-struct amdgpu_mec {
-	struct amdgpu_bo	*hpd_eop_obj;
-	u64			hpd_eop_gpu_addr;
-	struct amdgpu_bo	*mec_fw_obj;
-	u64			mec_fw_gpu_addr;
-	u32 num_pipe;
-	u32 num_mec;
-	u32 num_queue;
-	void			*mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1];
-};
-
-struct amdgpu_kiq {
-	u64			eop_gpu_addr;
-	struct amdgpu_bo	*eop_obj;
-	struct amdgpu_ring	ring;
-	struct amdgpu_irq_src	irq;
-};
-
-/*
- * GPU scratch registers structures, functions & helpers
- */
-struct amdgpu_scratch {
-	unsigned		num_reg;
-	uint32_t                reg_base;
-	uint32_t		free_mask;
-};
-
-/*
- * GFX configurations
- */
-#define AMDGPU_GFX_MAX_SE 4
-#define AMDGPU_GFX_MAX_SH_PER_SE 2
-
-struct amdgpu_rb_config {
-	uint32_t rb_backend_disable;
-	uint32_t user_rb_backend_disable;
-	uint32_t raster_config;
-	uint32_t raster_config_1;
-};
-
-struct gb_addr_config {
-	uint16_t pipe_interleave_size;
-	uint8_t num_pipes;
-	uint8_t max_compress_frags;
-	uint8_t num_banks;
-	uint8_t num_se;
-	uint8_t num_rb_per_se;
-};
-
-struct amdgpu_gfx_config {
-	unsigned max_shader_engines;
-	unsigned max_tile_pipes;
-	unsigned max_cu_per_sh;
-	unsigned max_sh_per_se;
-	unsigned max_backends_per_se;
-	unsigned max_texture_channel_caches;
-	unsigned max_gprs;
-	unsigned max_gs_threads;
-	unsigned max_hw_contexts;
-	unsigned sc_prim_fifo_size_frontend;
-	unsigned sc_prim_fifo_size_backend;
-	unsigned sc_hiz_tile_fifo_size;
-	unsigned sc_earlyz_tile_fifo_size;
-
-	unsigned num_tile_pipes;
-	unsigned backend_enable_mask;
-	unsigned mem_max_burst_length_bytes;
-	unsigned mem_row_size_in_kb;
-	unsigned shader_engine_tile_size;
-	unsigned num_gpus;
-	unsigned multi_gpu_tile_size;
-	unsigned mc_arb_ramcfg;
-	unsigned gb_addr_config;
-	unsigned num_rbs;
-	unsigned gs_vgt_table_depth;
-	unsigned gs_prim_buffer_depth;
-
-	uint32_t tile_mode_array[32];
-	uint32_t macrotile_mode_array[16];
-
-	struct gb_addr_config gb_addr_config_fields;
-	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
-
-	/* gfx configure feature */
-	uint32_t double_offchip_lds_buf;
-};
-
-struct amdgpu_cu_info {
-	uint32_t number; /* total active CU number */
-	uint32_t ao_cu_mask;
-	uint32_t wave_front_size;
-	uint32_t bitmap[4][4];
-};
-
-struct amdgpu_gfx_funcs {
-	/* get the gpu clock counter */
-	uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev);
-	void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 instance);
-	void (*read_wave_data)(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields);
-	void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t thread, uint32_t start, uint32_t size, uint32_t *dst);
-	void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t start, uint32_t size, uint32_t *dst);
-};
-
-struct amdgpu_ngg_buf {
-	struct amdgpu_bo	*bo;
-	uint64_t		gpu_addr;
-	uint32_t		size;
-	uint32_t		bo_size;
-};
-
-enum {
-	NGG_PRIM = 0,
-	NGG_POS,
-	NGG_CNTL,
-	NGG_PARAM,
-	NGG_BUF_MAX
-};
-
-struct amdgpu_ngg {
-	struct amdgpu_ngg_buf	buf[NGG_BUF_MAX];
-	uint32_t		gds_reserve_addr;
-	uint32_t		gds_reserve_size;
-	bool			init;
-};
-
-struct amdgpu_gfx {
-	struct mutex			gpu_clock_mutex;
-	struct amdgpu_gfx_config	config;
-	struct amdgpu_rlc		rlc;
-	struct amdgpu_mec		mec;
-	struct amdgpu_kiq		kiq;
-	struct amdgpu_scratch		scratch;
-	const struct firmware		*me_fw;	/* ME firmware */
-	uint32_t			me_fw_version;
-	const struct firmware		*pfp_fw; /* PFP firmware */
-	uint32_t			pfp_fw_version;
-	const struct firmware		*ce_fw;	/* CE firmware */
-	uint32_t			ce_fw_version;
-	const struct firmware		*rlc_fw; /* RLC firmware */
-	uint32_t			rlc_fw_version;
-	const struct firmware		*mec_fw; /* MEC firmware */
-	uint32_t			mec_fw_version;
-	const struct firmware		*mec2_fw; /* MEC2 firmware */
-	uint32_t			mec2_fw_version;
-	uint32_t			me_feature_version;
-	uint32_t			ce_feature_version;
-	uint32_t			pfp_feature_version;
-	uint32_t			rlc_feature_version;
-	uint32_t			mec_feature_version;
-	uint32_t			mec2_feature_version;
-	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
-	unsigned			num_gfx_rings;
-	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
-	unsigned			num_compute_rings;
-	struct amdgpu_irq_src		eop_irq;
-	struct amdgpu_irq_src		priv_reg_irq;
-	struct amdgpu_irq_src		priv_inst_irq;
-	/* gfx status */
-	uint32_t			gfx_current_status;
-	/* ce ram size*/
-	unsigned			ce_ram_size;
-	struct amdgpu_cu_info		cu_info;
-	const struct amdgpu_gfx_funcs	*funcs;
-
-	/* reset mask */
-	uint32_t                        grbm_soft_reset;
-	uint32_t                        srbm_soft_reset;
-	bool                            in_reset;
-	/* NGG */
-	struct amdgpu_ngg		ngg;
-};
-
-int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-		  unsigned size, struct amdgpu_ib *ib);
-void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib,
-		    struct dma_fence *f);
-int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
-		       struct amdgpu_ib *ibs, struct amdgpu_job *job,
-		       struct dma_fence **f);
-int amdgpu_ib_pool_init(struct amdgpu_device *adev);
-void amdgpu_ib_pool_fini(struct amdgpu_device *adev);
-int amdgpu_ib_ring_tests(struct amdgpu_device *adev);
-
-/*
- * CS.
- */
-struct amdgpu_cs_chunk {
-	uint32_t		chunk_id;
-	uint32_t		length_dw;
-	void			*kdata;
-};
-
-struct amdgpu_cs_parser {
-	struct amdgpu_device	*adev;
-	struct drm_file		*filp;
-	struct amdgpu_ctx	*ctx;
-
-	/* chunks */
-	unsigned		nchunks;
-	struct amdgpu_cs_chunk	*chunks;
-
-	/* scheduler job object */
-	struct amdgpu_job	*job;
-
-	/* buffer objects */
-	struct ww_acquire_ctx		ticket;
-	struct amdgpu_bo_list		*bo_list;
-	struct amdgpu_bo_list_entry	vm_pd;
-	struct list_head		validated;
-	struct dma_fence		*fence;
-	uint64_t			bytes_moved_threshold;
-	uint64_t			bytes_moved;
-	struct amdgpu_bo_list_entry	*evictable;
-
-	/* user fence */
-	struct amdgpu_bo_list_entry	uf_entry;
-};
-
-#define AMDGPU_PREAMBLE_IB_PRESENT          (1 << 0) /* bit set means command submit involves a preamble IB */
-#define AMDGPU_PREAMBLE_IB_PRESENT_FIRST    (1 << 1) /* bit set means preamble IB is first presented in belonging context */
-#define AMDGPU_HAVE_CTX_SWITCH              (1 << 2) /* bit set means context switch occured */
-#define AMDGPU_VM_DOMAIN                    (1 << 3) /* bit set means in virtual memory context */
-
-struct amdgpu_job {
-	struct amd_sched_job    base;
-	struct amdgpu_device	*adev;
-	struct amdgpu_vm	*vm;
-	struct amdgpu_ring	*ring;
-	struct amdgpu_sync	sync;
-	struct amdgpu_ib	*ibs;
-	struct dma_fence	*fence; /* the hw fence */
-	uint32_t		preamble_status;
-	uint32_t		num_ibs;
-	void			*owner;
-	uint64_t		fence_ctx; /* the fence_context this job uses */
-	bool                    vm_needs_flush;
-	bool			need_pipeline_sync;
-	unsigned		vm_id;
-	uint64_t		vm_pd_addr;
-	uint32_t		gds_base, gds_size;
-	uint32_t		gws_base, gws_size;
-	uint32_t		oa_base, oa_size;
-
-	/* user fence handling */
-	uint64_t		uf_addr;
-	uint64_t		uf_sequence;
-
-};
-#define to_amdgpu_job(sched_job)		\
-		container_of((sched_job), struct amdgpu_job, base)
-
-static inline u32 amdgpu_get_ib_value(struct amdgpu_cs_parser *p,
-				      uint32_t ib_idx, int idx)
-{
-	return p->job->ibs[ib_idx].ptr[idx];
-}
-
-static inline void amdgpu_set_ib_value(struct amdgpu_cs_parser *p,
-				       uint32_t ib_idx, int idx,
-				       uint32_t value)
-{
-	p->job->ibs[ib_idx].ptr[idx] = value;
-}
+int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv);
 
 /*
  * Writeback
  */
 #define AMDGPU_MAX_WB 1024	/* Reserve at most 1024 WB slots for amdgpu-owned rings. */
 
+/**
+ * amdgpu_wb - This struct is used for small GPU memory allocation.
+ *
+ * This struct is used to allocate a small amount of GPU memory that can be
+ * used to shadow certain states into the memory. This is especially useful for
+ * providing easy CPU access to some states without requiring register access
+ * (e.g., if some block is power gated, reading register may be problematic).
+ *
+ * Note: the term writeback was initially used because many of the amdgpu
+ * components had some level of writeback memory, and this struct initially
+ * described those components.
+ */
 struct amdgpu_wb {
+
+	/**
+	 * @wb_obj:
+	 *
+	 * Buffer Object used for the writeback memory.
+	 */
 	struct amdgpu_bo	*wb_obj;
-	volatile uint32_t	*wb;
-	uint64_t		gpu_addr;
-	u32			num_wb;	/* Number of wb slots actually reserved for amdgpu. */
-	unsigned long		used[DIV_ROUND_UP(AMDGPU_MAX_WB, BITS_PER_LONG)];
-};
 
-int amdgpu_wb_get(struct amdgpu_device *adev, u32 *wb);
-void amdgpu_wb_free(struct amdgpu_device *adev, u32 wb);
-int amdgpu_wb_get_64bit(struct amdgpu_device *adev, u32 *wb);
-void amdgpu_wb_free_64bit(struct amdgpu_device *adev, u32 wb);
+	/**
+	 * @wb:
+	 *
+	 * Pointer to the first writeback slot. In terms of CPU address
+	 * this value can be accessed directly by using the offset as an index.
+	 * For the GPU address, it is necessary to use gpu_addr and the offset.
+	 */
+	uint32_t		*wb;
 
-void amdgpu_get_pcie_info(struct amdgpu_device *adev);
+	/**
+	 * @gpu_addr:
+	 *
+	 * Writeback base address in the GPU.
+	 */
+	uint64_t		gpu_addr;
 
-/*
- * SDMA
- */
-struct amdgpu_sdma_instance {
-	/* SDMA firmware */
-	const struct firmware	*fw;
-	uint32_t		fw_version;
-	uint32_t		feature_version;
-
-	struct amdgpu_ring	ring;
-	bool			burst_nop;
-};
+	/**
+	 * @num_wb:
+	 *
+	 * Number of writeback slots reserved for amdgpu.
+	 */
+	u32			num_wb;
 
-struct amdgpu_sdma {
-	struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
-#ifdef CONFIG_DRM_AMDGPU_SI
-	//SI DMA has a difference trap irq number for the second engine
-	struct amdgpu_irq_src	trap_irq_1;
-#endif
-	struct amdgpu_irq_src	trap_irq;
-	struct amdgpu_irq_src	illegal_inst_irq;
-	int			num_instances;
-	uint32_t                    srbm_soft_reset;
-};
+	/**
+	 * @used:
+	 *
+	 * Track the writeback slot already used.
+	 */
+	unsigned long		used[DIV_ROUND_UP(AMDGPU_MAX_WB, BITS_PER_LONG)];
 
-/*
- * Firmware
- */
-enum amdgpu_firmware_load_type {
-	AMDGPU_FW_LOAD_DIRECT = 0,
-	AMDGPU_FW_LOAD_SMU,
-	AMDGPU_FW_LOAD_PSP,
+	/**
+	 * @lock:
+	 *
+	 * Protects read and write of the used field array.
+	 */
+	spinlock_t		lock;
 };
 
-struct amdgpu_firmware {
-	struct amdgpu_firmware_info ucode[AMDGPU_UCODE_ID_MAXIMUM];
-	enum amdgpu_firmware_load_type load_type;
-	struct amdgpu_bo *fw_buf;
-	unsigned int fw_size;
-	unsigned int max_ucodes;
-	/* firmwares are loaded by psp instead of smu from vega10 */
-	const struct amdgpu_psp_funcs *funcs;
-	struct amdgpu_bo *rbuf;
-	struct mutex mutex;
-};
+int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb);
+void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb);
 
 /*
  * Benchmarking
  */
-void amdgpu_benchmark(struct amdgpu_device *adev, int test_number);
-
-
-/*
- * Testing
- */
-void amdgpu_test_moves(struct amdgpu_device *adev);
-
-/*
- * MMU Notifier
- */
-#if defined(CONFIG_MMU_NOTIFIER)
-int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr);
-void amdgpu_mn_unregister(struct amdgpu_bo *bo);
-#else
-static inline int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr)
-{
-	return -ENODEV;
-}
-static inline void amdgpu_mn_unregister(struct amdgpu_bo *bo) {}
-#endif
-
-/*
- * Debugfs
- */
-struct amdgpu_debugfs {
-	const struct drm_info_list	*files;
-	unsigned		num_files;
-};
-
-int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
-			     const struct drm_info_list *files,
-			     unsigned nfiles);
-int amdgpu_debugfs_fence_init(struct amdgpu_device *adev);
-
-#if defined(CONFIG_DEBUG_FS)
-int amdgpu_debugfs_init(struct drm_minor *minor);
-#endif
-
-int amdgpu_debugfs_firmware_init(struct amdgpu_device *adev);
-
-/*
- * amdgpu smumgr functions
- */
-struct amdgpu_smumgr_funcs {
-	int (*check_fw_load_finish)(struct amdgpu_device *adev, uint32_t fwtype);
-	int (*request_smu_load_fw)(struct amdgpu_device *adev);
-	int (*request_smu_specific_fw)(struct amdgpu_device *adev, uint32_t fwtype);
-};
-
-/*
- * amdgpu smumgr
- */
-struct amdgpu_smumgr {
-	struct amdgpu_bo *toc_buf;
-	struct amdgpu_bo *smu_buf;
-	/* asic priv smu data */
-	void *priv;
-	spinlock_t smu_lock;
-	/* smumgr functions */
-	const struct amdgpu_smumgr_funcs *smumgr_funcs;
-	/* ucode loading complete flag */
-	uint32_t fw_flags;
-};
+int amdgpu_benchmark(struct amdgpu_device *adev, int test_number);
 
 /*
  * ASIC specific register table accessible by UMD
  */
 struct amdgpu_allowed_register_entry {
 	uint32_t reg_offset;
-	bool untouched;
 	bool grbm_indexed;
 };
 
+/**
+ * enum amd_reset_method - Methods for resetting AMD GPU devices
+ *
+ * @AMD_RESET_METHOD_NONE: The device will not be reset.
+ * @AMD_RESET_LEGACY: Method reserved for SI, CIK and VI ASICs.
+ * @AMD_RESET_MODE0: Reset the entire ASIC. Not currently available for the
+ *                   any device.
+ * @AMD_RESET_MODE1: Resets all IP blocks on the ASIC (SDMA, GFX, VCN, etc.)
+ *                   individually. Suitable only for some discrete GPU, not
+ *                   available for all ASICs.
+ * @AMD_RESET_MODE2: Resets a lesser level of IPs compared to MODE1. Which IPs
+ *                   are reset depends on the ASIC. Notably doesn't reset IPs
+ *                   shared with the CPU on APUs or the memory controllers (so
+ *                   VRAM is not lost). Not available on all ASICs.
+ * @AMD_RESET_LINK: Triggers SW-UP link reset on other GPUs
+ * @AMD_RESET_BACO: BACO (Bus Alive, Chip Off) method powers off and on the card
+ *                  but without powering off the PCI bus. Suitable only for
+ *                  discrete GPUs.
+ * @AMD_RESET_PCI: Does a full bus reset using core Linux subsystem PCI reset
+ *                 and does a secondary bus reset or FLR, depending on what the
+ *                 underlying hardware supports.
+ *
+ * Methods available for AMD GPU driver for resetting the device. Not all
+ * methods are suitable for every device. User can override the method using
+ * module parameter `reset_method`.
+ */
+enum amd_reset_method {
+	AMD_RESET_METHOD_NONE = -1,
+	AMD_RESET_METHOD_LEGACY = 0,
+	AMD_RESET_METHOD_MODE0,
+	AMD_RESET_METHOD_MODE1,
+	AMD_RESET_METHOD_MODE2,
+	AMD_RESET_METHOD_LINK,
+	AMD_RESET_METHOD_BACO,
+	AMD_RESET_METHOD_PCI,
+	AMD_RESET_METHOD_ON_INIT,
+};
+
+struct amdgpu_video_codec_info {
+	u32 codec_type;
+	u32 max_width;
+	u32 max_height;
+	u32 max_pixels_per_frame;
+	u32 max_level;
+};
+
+#define codec_info_build(type, width, height, level) \
+			 .codec_type = type,\
+			 .max_width = width,\
+			 .max_height = height,\
+			 .max_pixels_per_frame = height * width,\
+			 .max_level = level,
+
+struct amdgpu_video_codecs {
+	const u32 codec_count;
+	const struct amdgpu_video_codec_info *codec_array;
+};
+
 /*
  * ASIC specific functions.
  */
@@ -1311,6 +663,7 @@ struct amdgpu_asic_funcs {
 			     u32 sh_num, u32 reg_offset, u32 *value);
 	void (*set_vga_state)(struct amdgpu_device *adev, bool state);
 	int (*reset)(struct amdgpu_device *adev);
+	enum amd_reset_method (*reset_method)(struct amdgpu_device *adev);
 	/* get the reference clock */
 	u32 (*get_xclk)(struct amdgpu_device *adev);
 	/* MM block clocks */
@@ -1321,95 +674,60 @@ struct amdgpu_asic_funcs {
 	void (*set_pcie_lanes)(struct amdgpu_device *adev, int lanes);
 	/* get config memsize register */
 	u32 (*get_config_memsize)(struct amdgpu_device *adev);
+	/* flush hdp write queue */
+	void (*flush_hdp)(struct amdgpu_device *adev, struct amdgpu_ring *ring);
+	/* invalidate hdp read cache */
+	void (*invalidate_hdp)(struct amdgpu_device *adev,
+			       struct amdgpu_ring *ring);
+	/* check if the asic needs a full reset of if soft reset will work */
+	bool (*need_full_reset)(struct amdgpu_device *adev);
+	/* initialize doorbell layout for specific asic*/
+	void (*init_doorbell_index)(struct amdgpu_device *adev);
+	/* PCIe bandwidth usage */
+	void (*get_pcie_usage)(struct amdgpu_device *adev, uint64_t *count0,
+			       uint64_t *count1);
+	/* do we need to reset the asic at init time (e.g., kexec) */
+	bool (*need_reset_on_init)(struct amdgpu_device *adev);
+	/* PCIe replay counter */
+	uint64_t (*get_pcie_replay_count)(struct amdgpu_device *adev);
+	/* device supports BACO */
+	int (*supports_baco)(struct amdgpu_device *adev);
+	/* pre asic_init quirks */
+	void (*pre_asic_init)(struct amdgpu_device *adev);
+	/* enter/exit umd stable pstate */
+	int (*update_umd_stable_pstate)(struct amdgpu_device *adev, bool enter);
+	/* query video codecs */
+	int (*query_video_codecs)(struct amdgpu_device *adev, bool encode,
+				  const struct amdgpu_video_codecs **codecs);
+	/* encode "> 32bits" smn addressing */
+	u64 (*encode_ext_smn_addressing)(int ext_id);
+
+	ssize_t (*get_reg_state)(struct amdgpu_device *adev,
+				 enum amdgpu_reg_state reg_state, void *buf,
+				 size_t max_size);
 };
 
 /*
  * IOCTL.
  */
-int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
-			    struct drm_file *filp);
 int amdgpu_bo_list_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *filp);
 
-int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
-			  struct drm_file *filp);
-int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
-			struct drm_file *filp);
-int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
-			  struct drm_file *filp);
-int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
-			      struct drm_file *filp);
-int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
-			  struct drm_file *filp);
-int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
-			struct drm_file *filp);
 int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
+int amdgpu_cs_fence_to_handle_ioctl(struct drm_device *dev, void *data,
+				    struct drm_file *filp);
 int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
 int amdgpu_cs_wait_fences_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *filp);
 
-int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
-				struct drm_file *filp);
-
 /* VRAM scratch page for HDP bug, default vram page */
-struct amdgpu_vram_scratch {
+struct amdgpu_mem_scratch {
 	struct amdgpu_bo		*robj;
-	volatile uint32_t		*ptr;
+	uint32_t			*ptr;
 	u64				gpu_addr;
 };
 
 /*
- * ACPI
- */
-struct amdgpu_atif_notification_cfg {
-	bool enabled;
-	int command_code;
-};
-
-struct amdgpu_atif_notifications {
-	bool display_switch;
-	bool expansion_mode_change;
-	bool thermal_state;
-	bool forced_power_state;
-	bool system_power_state;
-	bool display_conf_change;
-	bool px_gfx_switch;
-	bool brightness_change;
-	bool dgpu_display_event;
-};
-
-struct amdgpu_atif_functions {
-	bool system_params;
-	bool sbios_requests;
-	bool select_active_disp;
-	bool lid_state;
-	bool get_tv_standard;
-	bool set_tv_standard;
-	bool get_panel_expansion_mode;
-	bool set_panel_expansion_mode;
-	bool temperature_change;
-	bool graphics_device_types;
-};
-
-struct amdgpu_atif {
-	struct amdgpu_atif_notifications notifications;
-	struct amdgpu_atif_functions functions;
-	struct amdgpu_atif_notification_cfg notification_cfg;
-	struct amdgpu_encoder *encoder_for_bl;
-};
-
-struct amdgpu_atcs_functions {
-	bool get_ext_state;
-	bool pcie_perf_req;
-	bool pcie_dev_rdy;
-	bool pcie_bus_width;
-};
-
-struct amdgpu_atcs {
-	struct amdgpu_atcs_functions functions;
-};
-
-/*
  * CGS
  */
 struct cgs_device *amdgpu_cgs_create_device(struct amdgpu_device *adev);
@@ -1421,50 +739,251 @@ void amdgpu_cgs_destroy_device(struct cgs_device *cgs_device);
 typedef uint32_t (*amdgpu_rreg_t)(struct amdgpu_device*, uint32_t);
 typedef void (*amdgpu_wreg_t)(struct amdgpu_device*, uint32_t, uint32_t);
 
+typedef uint32_t (*amdgpu_rreg_ext_t)(struct amdgpu_device*, uint64_t);
+typedef void (*amdgpu_wreg_ext_t)(struct amdgpu_device*, uint64_t, uint32_t);
+
+typedef uint64_t (*amdgpu_rreg64_t)(struct amdgpu_device*, uint32_t);
+typedef void (*amdgpu_wreg64_t)(struct amdgpu_device*, uint32_t, uint64_t);
+
+typedef uint64_t (*amdgpu_rreg64_ext_t)(struct amdgpu_device*, uint64_t);
+typedef void (*amdgpu_wreg64_ext_t)(struct amdgpu_device*, uint64_t, uint64_t);
+
 typedef uint32_t (*amdgpu_block_rreg_t)(struct amdgpu_device*, uint32_t, uint32_t);
 typedef void (*amdgpu_block_wreg_t)(struct amdgpu_device*, uint32_t, uint32_t, uint32_t);
 
+struct amdgpu_mmio_remap {
+	u32 reg_offset;
+	resource_size_t bus_addr;
+	struct amdgpu_bo *bo;
+};
+
+/* Define the HW IP blocks will be used in driver , add more if necessary */
+enum amd_hw_ip_block_type {
+	GC_HWIP = 1,
+	HDP_HWIP,
+	SDMA0_HWIP,
+	SDMA1_HWIP,
+	SDMA2_HWIP,
+	SDMA3_HWIP,
+	SDMA4_HWIP,
+	SDMA5_HWIP,
+	SDMA6_HWIP,
+	SDMA7_HWIP,
+	LSDMA_HWIP,
+	MMHUB_HWIP,
+	ATHUB_HWIP,
+	NBIO_HWIP,
+	MP0_HWIP,
+	MP1_HWIP,
+	UVD_HWIP,
+	VCN_HWIP = UVD_HWIP,
+	JPEG_HWIP = VCN_HWIP,
+	VCN1_HWIP,
+	VCE_HWIP,
+	VPE_HWIP,
+	DF_HWIP,
+	DCE_HWIP,
+	OSSSYS_HWIP,
+	SMUIO_HWIP,
+	PWR_HWIP,
+	NBIF_HWIP,
+	THM_HWIP,
+	CLK_HWIP,
+	UMC_HWIP,
+	RSMU_HWIP,
+	XGMI_HWIP,
+	DCI_HWIP,
+	PCIE_HWIP,
+	ISP_HWIP,
+	MAX_HWIP
+};
+
+#define HWIP_MAX_INSTANCE	44
+
+#define HW_ID_MAX		300
+#define IP_VERSION_FULL(mj, mn, rv, var, srev) \
+	(((mj) << 24) | ((mn) << 16) | ((rv) << 8) | ((var) << 4) | (srev))
+#define IP_VERSION(mj, mn, rv)		IP_VERSION_FULL(mj, mn, rv, 0, 0)
+#define IP_VERSION_MAJ(ver)		((ver) >> 24)
+#define IP_VERSION_MIN(ver)		(((ver) >> 16) & 0xFF)
+#define IP_VERSION_REV(ver)		(((ver) >> 8) & 0xFF)
+#define IP_VERSION_VARIANT(ver)		(((ver) >> 4) & 0xF)
+#define IP_VERSION_SUBREV(ver)		((ver) & 0xF)
+#define IP_VERSION_MAJ_MIN_REV(ver)	((ver) >> 8)
+
+struct amdgpu_ip_map_info {
+	/* Map of logical to actual dev instances/mask */
+	uint32_t 		dev_inst[MAX_HWIP][HWIP_MAX_INSTANCE];
+	int8_t (*logical_to_dev_inst)(struct amdgpu_device *adev,
+				      enum amd_hw_ip_block_type block,
+				      int8_t inst);
+	uint32_t (*logical_to_dev_mask)(struct amdgpu_device *adev,
+					enum amd_hw_ip_block_type block,
+					uint32_t mask);
+};
+
+enum amdgpu_uid_type {
+	AMDGPU_UID_TYPE_XCD,
+	AMDGPU_UID_TYPE_AID,
+	AMDGPU_UID_TYPE_SOC,
+	AMDGPU_UID_TYPE_MAX
+};
+
+#define AMDGPU_UID_INST_MAX 8 /* max number of instances for each UID type */
+
+struct amdgpu_uid {
+	uint64_t uid[AMDGPU_UID_TYPE_MAX][AMDGPU_UID_INST_MAX];
+	struct amdgpu_device *adev;
+};
+
+struct amd_powerplay {
+	void *pp_handle;
+	const struct amd_pm_funcs *pp_funcs;
+};
+
+/* polaris10 kickers */
+#define ASICID_IS_P20(did, rid)		(((did == 0x67DF) && \
+					 ((rid == 0xE3) || \
+					  (rid == 0xE4) || \
+					  (rid == 0xE5) || \
+					  (rid == 0xE7) || \
+					  (rid == 0xEF))) || \
+					 ((did == 0x6FDF) && \
+					 ((rid == 0xE7) || \
+					  (rid == 0xEF) || \
+					  (rid == 0xFF))))
+
+#define ASICID_IS_P30(did, rid)		((did == 0x67DF) && \
+					((rid == 0xE1) || \
+					 (rid == 0xF7)))
+
+/* polaris11 kickers */
+#define ASICID_IS_P21(did, rid)		(((did == 0x67EF) && \
+					 ((rid == 0xE0) || \
+					  (rid == 0xE5))) || \
+					 ((did == 0x67FF) && \
+					 ((rid == 0xCF) || \
+					  (rid == 0xEF) || \
+					  (rid == 0xFF))))
+
+#define ASICID_IS_P31(did, rid)		((did == 0x67EF) && \
+					((rid == 0xE2)))
+
+/* polaris12 kickers */
+#define ASICID_IS_P23(did, rid)		(((did == 0x6987) && \
+					 ((rid == 0xC0) || \
+					  (rid == 0xC1) || \
+					  (rid == 0xC3) || \
+					  (rid == 0xC7))) || \
+					 ((did == 0x6981) && \
+					 ((rid == 0x00) || \
+					  (rid == 0x01) || \
+					  (rid == 0x10))))
+
+struct amdgpu_mqd_prop {
+	uint64_t mqd_gpu_addr;
+	uint64_t hqd_base_gpu_addr;
+	uint64_t rptr_gpu_addr;
+	uint64_t wptr_gpu_addr;
+	uint32_t queue_size;
+	bool use_doorbell;
+	uint32_t doorbell_index;
+	uint64_t eop_gpu_addr;
+	uint32_t hqd_pipe_priority;
+	uint32_t hqd_queue_priority;
+	bool allow_tunneling;
+	bool hqd_active;
+	uint64_t shadow_addr;
+	uint64_t gds_bkup_addr;
+	uint64_t csa_addr;
+	uint64_t fence_address;
+	bool tmz_queue;
+	bool kernel_queue;
+};
+
+struct amdgpu_mqd {
+	unsigned mqd_size;
+	int (*init_mqd)(struct amdgpu_device *adev, void *mqd,
+			struct amdgpu_mqd_prop *p);
+};
+
+struct amdgpu_pcie_reset_ctx {
+	bool in_link_reset;
+	bool occurs_dpc;
+	bool audio_suspended;
+	struct pci_dev *swus;
+	struct pci_saved_state *swus_pcistate;
+	struct pci_saved_state *swds_pcistate;
+};
+
+/*
+ * Custom Init levels could be defined for different situations where a full
+ * initialization of all hardware blocks are not expected. Sample cases are
+ * custom init sequences after resume after S0i3/S3, reset on initialization,
+ * partial reset of blocks etc. Presently, this defines only two levels. Levels
+ * are described in corresponding struct definitions - amdgpu_init_default,
+ * amdgpu_init_minimal_xgmi.
+ */
+enum amdgpu_init_lvl_id {
+	AMDGPU_INIT_LEVEL_DEFAULT,
+	AMDGPU_INIT_LEVEL_MINIMAL_XGMI,
+	AMDGPU_INIT_LEVEL_RESET_RECOVERY,
+};
+
+struct amdgpu_init_level {
+	enum amdgpu_init_lvl_id level;
+	uint32_t hwini_ip_block_mask;
+};
+
+#define AMDGPU_RESET_MAGIC_NUM 64
+#define AMDGPU_MAX_DF_PERFMONS 4
+struct amdgpu_reset_domain;
+struct amdgpu_fru_info;
+
+enum amdgpu_enforce_isolation_mode {
+	AMDGPU_ENFORCE_ISOLATION_DISABLE = 0,
+	AMDGPU_ENFORCE_ISOLATION_ENABLE = 1,
+	AMDGPU_ENFORCE_ISOLATION_ENABLE_LEGACY = 2,
+	AMDGPU_ENFORCE_ISOLATION_NO_CLEANER_SHADER = 3,
+};
+
 struct amdgpu_device {
 	struct device			*dev;
-	struct drm_device		*ddev;
 	struct pci_dev			*pdev;
+	struct drm_device		ddev;
 
 #ifdef CONFIG_DRM_AMD_ACP
 	struct amdgpu_acp		acp;
 #endif
-
+	struct amdgpu_hive_info *hive;
+	struct amdgpu_xcp_mgr *xcp_mgr;
 	/* ASIC */
 	enum amd_asic_type		asic_type;
 	uint32_t			family;
 	uint32_t			rev_id;
 	uint32_t			external_rev_id;
 	unsigned long			flags;
+	unsigned long			apu_flags;
 	int				usec_timeout;
 	const struct amdgpu_asic_funcs	*asic_funcs;
 	bool				shutdown;
-	bool				need_dma32;
+	bool				need_swiotlb;
 	bool				accel_working;
-	struct work_struct		reset_work;
 	struct notifier_block		acpi_nb;
+	struct notifier_block		pm_nb;
 	struct amdgpu_i2c_chan		*i2c_bus[AMDGPU_MAX_I2C_BUS];
-	struct amdgpu_debugfs		debugfs[AMDGPU_DEBUGFS_MAX_COMPONENTS];
-	unsigned			debugfs_count;
-#if defined(CONFIG_DEBUG_FS)
-	struct dentry			*debugfs_regs[AMDGPU_DEBUGFS_MAX_COMPONENTS];
-#endif
-	struct amdgpu_atif		atif;
-	struct amdgpu_atcs		atcs;
+	struct debugfs_blob_wrapper debugfs_vbios_blob;
 	struct mutex			srbm_mutex;
 	/* GRBM index mutex. Protects concurrent access to GRBM index */
 	struct mutex                    grbm_idx_mutex;
 	struct dev_pm_domain		vga_pm_domain;
 	bool				have_disp_power_ref;
+	bool                            have_atomics_support;
 
 	/* BIOS */
 	bool				is_atom_fw;
 	uint8_t				*bios;
 	uint32_t			bios_size;
-	struct amdgpu_bo		*stollen_vga_memory;
 	uint32_t			bios_scratch_reg_offset;
 	uint32_t			bios_scratch[AMDGPU_BIOS_NUM_SCRATCH];
 
@@ -1474,6 +993,7 @@ struct amdgpu_device {
 	void __iomem			*rmmio;
 	/* protects concurrent MM_INDEX/DATA based register access */
 	spinlock_t mmio_idx_lock;
+	struct amdgpu_mmio_remap        rmmio_remap;
 	/* protects concurrent SMC based register access */
 	spinlock_t smc_idx_lock;
 	amdgpu_rreg_t			smc_rreg;
@@ -1484,6 +1004,12 @@ struct amdgpu_device {
 	amdgpu_wreg_t			pcie_wreg;
 	amdgpu_rreg_t			pciep_rreg;
 	amdgpu_wreg_t			pciep_wreg;
+	amdgpu_rreg_ext_t		pcie_rreg_ext;
+	amdgpu_wreg_ext_t		pcie_wreg_ext;
+	amdgpu_rreg64_t			pcie_rreg64;
+	amdgpu_wreg64_t			pcie_wreg64;
+	amdgpu_rreg64_ext_t			pcie_rreg64_ext;
+	amdgpu_wreg64_ext_t			pcie_wreg64_ext;
 	/* protects concurrent UVD register access */
 	spinlock_t uvd_ctx_idx_lock;
 	amdgpu_rreg_t			uvd_ctx_rreg;
@@ -1496,73 +1022,95 @@ struct amdgpu_device {
 	spinlock_t gc_cac_idx_lock;
 	amdgpu_rreg_t			gc_cac_rreg;
 	amdgpu_wreg_t			gc_cac_wreg;
+	/* protects concurrent se_cac register access */
+	spinlock_t se_cac_idx_lock;
+	amdgpu_rreg_t			se_cac_rreg;
+	amdgpu_wreg_t			se_cac_wreg;
 	/* protects concurrent ENDPOINT (audio) register access */
 	spinlock_t audio_endpt_idx_lock;
 	amdgpu_block_rreg_t		audio_endpt_rreg;
 	amdgpu_block_wreg_t		audio_endpt_wreg;
-	void __iomem                    *rio_mem;
-	resource_size_t			rio_mem_size;
 	struct amdgpu_doorbell		doorbell;
 
 	/* clock/pll info */
 	struct amdgpu_clock            clock;
 
 	/* MC */
-	struct amdgpu_mc		mc;
+	struct amdgpu_gmc		gmc;
 	struct amdgpu_gart		gart;
-	struct amdgpu_dummy_page	dummy_page;
+	dma_addr_t			dummy_page_addr;
 	struct amdgpu_vm_manager	vm_manager;
 	struct amdgpu_vmhub             vmhub[AMDGPU_MAX_VMHUBS];
+	DECLARE_BITMAP(vmhubs_mask, AMDGPU_MAX_VMHUBS);
 
 	/* memory management */
 	struct amdgpu_mman		mman;
-	struct amdgpu_vram_scratch	vram_scratch;
+	struct amdgpu_mem_scratch	mem_scratch;
 	struct amdgpu_wb		wb;
-	atomic64_t			vram_usage;
-	atomic64_t			vram_vis_usage;
-	atomic64_t			gtt_usage;
 	atomic64_t			num_bytes_moved;
 	atomic64_t			num_evictions;
+	atomic64_t			num_vram_cpu_page_faults;
 	atomic_t			gpu_reset_counter;
+	atomic_t			vram_lost_counter;
 
 	/* data for buffer migration throttling */
 	struct {
 		spinlock_t		lock;
 		s64			last_update_us;
 		s64			accum_us; /* accumulated microseconds */
+		s64			accum_us_vis; /* for visible VRAM */
 		u32			log2_max_MBps;
 	} mm_stats;
 
+	/* discovery*/
+	struct amdgpu_discovery_info discovery;
+
 	/* display */
 	bool				enable_virtual_display;
+	struct amdgpu_vkms_output       *amdgpu_vkms_output;
 	struct amdgpu_mode_info		mode_info;
-	struct work_struct		hotplug_work;
+	/* For pre-DCE11. DCE11 and later are in "struct amdgpu_device->dm" */
+	struct delayed_work         hotplug_work;
 	struct amdgpu_irq_src		crtc_irq;
+	struct amdgpu_irq_src		vline0_irq;
+	struct amdgpu_irq_src		vupdate_irq;
 	struct amdgpu_irq_src		pageflip_irq;
 	struct amdgpu_irq_src		hpd_irq;
+	struct amdgpu_irq_src		dmub_trace_irq;
+	struct amdgpu_irq_src		dmub_outbox_irq;
 
 	/* rings */
 	u64				fence_context;
 	unsigned			num_rings;
 	struct amdgpu_ring		*rings[AMDGPU_MAX_RINGS];
+	struct dma_fence __rcu		*gang_submit;
 	bool				ib_pool_ready;
-	struct amdgpu_sa_manager	ring_tmp_bo;
+	struct amdgpu_sa_manager	ib_pools[AMDGPU_IB_POOL_MAX];
+	struct amdgpu_sched		gpu_sched[AMDGPU_HW_IP_NUM][AMDGPU_RING_PRIO_MAX];
 
 	/* interrupts */
 	struct amdgpu_irq		irq;
 
 	/* powerplay */
 	struct amd_powerplay		powerplay;
-	bool				pp_enabled;
-	bool				pp_force_state_enabled;
-
-	/* dpm */
 	struct amdgpu_pm		pm;
-	u32				cg_flags;
+	u64				cg_flags;
 	u32				pg_flags;
 
-	/* amdgpu smumgr */
-	struct amdgpu_smumgr smu;
+	/* nbio */
+	struct amdgpu_nbio		nbio;
+
+	/* hdp */
+	struct amdgpu_hdp		hdp;
+
+	/* smuio */
+	struct amdgpu_smuio		smuio;
+
+	/* mmhub */
+	struct amdgpu_mmhub		mmhub;
+
+	/* gfxhub */
+	struct amdgpu_gfxhub		gfxhub;
 
 	/* gfx */
 	struct amdgpu_gfx		gfx;
@@ -1570,12 +1118,28 @@ struct amdgpu_device {
 	/* sdma */
 	struct amdgpu_sdma		sdma;
 
+	/* lsdma */
+	struct amdgpu_lsdma		lsdma;
+
 	/* uvd */
 	struct amdgpu_uvd		uvd;
 
 	/* vce */
 	struct amdgpu_vce		vce;
 
+	/* vcn */
+	struct amdgpu_vcn		vcn;
+
+	/* jpeg */
+	struct amdgpu_jpeg		jpeg;
+
+	/* vpe */
+	struct amdgpu_vpe		vpe;
+
+	/* umsch */
+	struct amdgpu_umsch_mm		umsch_mm;
+	bool				enable_umsch_mm;
+
 	/* firmwares */
 	struct amdgpu_firmware		firmware;
 
@@ -1585,79 +1149,317 @@ struct amdgpu_device {
 	/* GDS */
 	struct amdgpu_gds		gds;
 
+	/* for userq and VM fences */
+	struct amdgpu_seq64		seq64;
+
+	/* UMC */
+	struct amdgpu_umc		umc;
+
+	/* display related functionality */
+	struct amdgpu_display_manager dm;
+
+#if defined(CONFIG_DRM_AMD_ISP)
+	/* isp */
+	struct amdgpu_isp		isp;
+#endif
+
+	/* mes */
+	bool                            enable_mes;
+	bool                            enable_mes_kiq;
+	bool                            enable_uni_mes;
+	struct amdgpu_mes               mes;
+	struct amdgpu_mqd               mqds[AMDGPU_HW_IP_NUM];
+	const struct amdgpu_userq_funcs *userq_funcs[AMDGPU_HW_IP_NUM];
+
+	/* xarray used to retrieve the user queue fence driver reference
+	 * in the EOP interrupt handler to signal the particular user
+	 * queue fence.
+	 */
+	struct xarray			userq_xa;
+	/**
+	 * @userq_doorbell_xa: Global user queue map (doorbell index → queue)
+	 * Key: doorbell_index (unique global identifier for the queue)
+	 * Value: struct amdgpu_usermode_queue
+	 */
+	struct xarray userq_doorbell_xa;
+
+	/* df */
+	struct amdgpu_df                df;
+
+	/* MCA */
+	struct amdgpu_mca               mca;
+
+	/* ACA */
+	struct amdgpu_aca		aca;
+
+	/* CPER */
+	struct amdgpu_cper		cper;
+
 	struct amdgpu_ip_block          ip_blocks[AMDGPU_MAX_IP_NUM];
+	uint32_t		        harvest_ip_mask;
 	int				num_ip_blocks;
 	struct mutex	mn_lock;
 	DECLARE_HASHTABLE(mn_hash, 7);
 
 	/* tracking pinned memory */
-	u64 vram_pin_size;
-	u64 invisible_pin_size;
-	u64 gart_pin_size;
+	atomic64_t vram_pin_size;
+	atomic64_t visible_pin_size;
+	atomic64_t gart_pin_size;
 
-	/* amdkfd interface */
-	struct kfd_dev          *kfd;
+	/* soc15 register offset based on ip, instance and  segment */
+	uint32_t		*reg_offset[MAX_HWIP][HWIP_MAX_INSTANCE];
+	struct amdgpu_ip_map_info	ip_map;
 
-	struct amdgpu_virt	virt;
+	/* delayed work_func for deferring clockgating during resume */
+	struct delayed_work     delayed_init_work;
 
-	/* link all shadow bo */
-	struct list_head                shadow_list;
-	struct mutex                    shadow_list_lock;
-	/* link all gtt */
-	spinlock_t			gtt_list_lock;
-	struct list_head                gtt_list;
+	struct amdgpu_virt	virt;
 
 	/* record hw reset is performed */
 	bool has_hw_reset;
+	u8				reset_magic[AMDGPU_RESET_MAGIC_NUM];
+
+	/* s3/s4 mask */
+	bool                            in_suspend;
+	bool				in_s3;
+	bool				in_s4;
+	bool				in_s0ix;
+	suspend_state_t			last_suspend_state;
+
+	enum pp_mp1_state               mp1_state;
+	struct amdgpu_doorbell_index doorbell_index;
+
+	struct mutex			notifier_lock;
+
+	int asic_reset_res;
+	struct work_struct		xgmi_reset_work;
+	struct list_head		reset_list;
+
+	long				gfx_timeout;
+	long				sdma_timeout;
+	long				video_timeout;
+	long				compute_timeout;
+	long				psp_timeout;
+
+	uint64_t			unique_id;
+	uint64_t	df_perfmon_config_assign_mask[AMDGPU_MAX_DF_PERFMONS];
+
+	/* enable runtime pm on the device */
+	bool                            in_runpm;
+	bool                            has_pr3;
+
+	bool                            ucode_sysfs_en;
+
+	struct amdgpu_fru_info		*fru_info;
+	atomic_t			throttling_logging_enabled;
+	struct ratelimit_state		throttling_logging_rs;
+	uint32_t                        ras_hw_enabled;
+	uint32_t                        ras_enabled;
+	bool                            ras_default_ecc_enabled;
+
+	bool                            no_hw_access;
+	struct pci_saved_state          *pci_state;
+	pci_channel_state_t		pci_channel_state;
+
+	struct amdgpu_pcie_reset_ctx	pcie_reset_ctx;
+
+	/* Track auto wait count on s_barrier settings */
+	bool				barrier_has_auto_waitcnt;
+
+	struct amdgpu_reset_control     *reset_cntl;
+	uint32_t                        ip_versions[MAX_HWIP][HWIP_MAX_INSTANCE];
+
+	bool				ram_is_direct_mapped;
+
+	struct list_head                ras_list;
+
+	struct amdgpu_reset_domain	*reset_domain;
+
+	struct mutex			benchmark_mutex;
+
+	bool                            scpm_enabled;
+	uint32_t                        scpm_status;
 
+	struct work_struct		reset_work;
+
+	bool                            dc_enabled;
+	/* Mask of active clusters */
+	uint32_t			aid_mask;
+
+	/* Debug */
+	bool                            debug_vm;
+	bool                            debug_largebar;
+	bool                            debug_disable_soft_recovery;
+	bool                            debug_use_vram_fw_buf;
+	bool                            debug_enable_ras_aca;
+	bool                            debug_exp_resets;
+	bool                            debug_disable_gpu_ring_reset;
+	bool                            debug_vm_userptr;
+	bool                            debug_disable_ce_logs;
+	bool                            debug_enable_ce_cs;
+
+	/* Protection for the following isolation structure */
+	struct mutex                    enforce_isolation_mutex;
+	enum amdgpu_enforce_isolation_mode	enforce_isolation[MAX_XCP];
+	struct amdgpu_isolation {
+		void			*owner;
+		struct dma_fence	*spearhead;
+		struct amdgpu_sync	active;
+		struct amdgpu_sync	prev;
+	} isolation[MAX_XCP];
+
+	struct amdgpu_init_level *init_lvl;
+
+	/* This flag is used to determine how VRAM allocations are handled for APUs
+	 * in KFD: VRAM or GTT.
+	 */
+	bool                            apu_prefer_gtt;
+
+	bool                            userq_halt_for_enforce_isolation;
+	struct work_struct              userq_reset_work;
+	struct amdgpu_uid *uid_info;
+
+	/* KFD
+	 * Must be last --ends in a flexible-array member.
+	 */
+	struct amdgpu_kfd_dev		kfd;
 };
 
-static inline struct amdgpu_device *amdgpu_ttm_adev(struct ttm_bo_device *bdev)
+static inline uint32_t amdgpu_ip_version(const struct amdgpu_device *adev,
+					 uint8_t ip, uint8_t inst)
+{
+	/* This considers only major/minor/rev and ignores
+	 * subrevision/variant fields.
+	 */
+	return adev->ip_versions[ip][inst] & ~0xFFU;
+}
+
+static inline uint32_t amdgpu_ip_version_full(const struct amdgpu_device *adev,
+					      uint8_t ip, uint8_t inst)
+{
+	/* This returns full version - major/minor/rev/variant/subrevision */
+	return adev->ip_versions[ip][inst];
+}
+
+static inline struct amdgpu_device *drm_to_adev(struct drm_device *ddev)
+{
+	return container_of(ddev, struct amdgpu_device, ddev);
+}
+
+static inline struct drm_device *adev_to_drm(struct amdgpu_device *adev)
+{
+	return &adev->ddev;
+}
+
+static inline struct amdgpu_device *amdgpu_ttm_adev(struct ttm_device *bdev)
 {
 	return container_of(bdev, struct amdgpu_device, mman.bdev);
 }
 
-bool amdgpu_device_is_px(struct drm_device *dev);
+static inline bool amdgpu_is_multi_aid(struct amdgpu_device *adev)
+{
+	return !!adev->aid_mask;
+}
+
 int amdgpu_device_init(struct amdgpu_device *adev,
-		       struct drm_device *ddev,
-		       struct pci_dev *pdev,
 		       uint32_t flags);
-void amdgpu_device_fini(struct amdgpu_device *adev);
+void amdgpu_device_fini_hw(struct amdgpu_device *adev);
+void amdgpu_device_fini_sw(struct amdgpu_device *adev);
+
 int amdgpu_gpu_wait_for_idle(struct amdgpu_device *adev);
 
-uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg,
+void amdgpu_device_mm_access(struct amdgpu_device *adev, loff_t pos,
+			     void *buf, size_t size, bool write);
+size_t amdgpu_device_aper_access(struct amdgpu_device *adev, loff_t pos,
+				 void *buf, size_t size, bool write);
+
+void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos,
+			       void *buf, size_t size, bool write);
+uint32_t amdgpu_device_wait_on_rreg(struct amdgpu_device *adev,
+			    uint32_t inst, uint32_t reg_addr, char reg_name[],
+			    uint32_t expected_value, uint32_t mask);
+uint32_t amdgpu_device_rreg(struct amdgpu_device *adev,
+			    uint32_t reg, uint32_t acc_flags);
+u32 amdgpu_device_indirect_rreg_ext(struct amdgpu_device *adev,
+				    u64 reg_addr);
+uint32_t amdgpu_device_xcc_rreg(struct amdgpu_device *adev,
+				uint32_t reg, uint32_t acc_flags,
+				uint32_t xcc_id);
+void amdgpu_device_wreg(struct amdgpu_device *adev,
+			uint32_t reg, uint32_t v,
 			uint32_t acc_flags);
-void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v,
-		    uint32_t acc_flags);
-u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg);
-void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v);
-
-u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index);
-void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v);
-u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index);
-void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v);
+void amdgpu_device_indirect_wreg_ext(struct amdgpu_device *adev,
+				     u64 reg_addr, u32 reg_data);
+void amdgpu_device_xcc_wreg(struct amdgpu_device *adev,
+			    uint32_t reg, uint32_t v,
+			    uint32_t acc_flags,
+			    uint32_t xcc_id);
+void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev,
+			     uint32_t reg, uint32_t v, uint32_t xcc_id);
+void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value);
+uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset);
+
+u32 amdgpu_device_indirect_rreg(struct amdgpu_device *adev,
+				u32 reg_addr);
+u64 amdgpu_device_indirect_rreg64(struct amdgpu_device *adev,
+				  u32 reg_addr);
+u64 amdgpu_device_indirect_rreg64_ext(struct amdgpu_device *adev,
+				  u64 reg_addr);
+void amdgpu_device_indirect_wreg(struct amdgpu_device *adev,
+				 u32 reg_addr, u32 reg_data);
+void amdgpu_device_indirect_wreg64(struct amdgpu_device *adev,
+				   u32 reg_addr, u64 reg_data);
+void amdgpu_device_indirect_wreg64_ext(struct amdgpu_device *adev,
+				   u64 reg_addr, u64 reg_data);
+u32 amdgpu_device_get_rev_id(struct amdgpu_device *adev);
+bool amdgpu_device_asic_has_dc_support(struct pci_dev *pdev,
+				       enum amd_asic_type asic_type);
+bool amdgpu_device_has_dc_support(struct amdgpu_device *adev);
+
+void amdgpu_device_set_sriov_virtual_display(struct amdgpu_device *adev);
+
+int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
+				 struct amdgpu_reset_context *reset_context);
+
+int amdgpu_do_asic_reset(struct list_head *device_list_handle,
+			 struct amdgpu_reset_context *reset_context);
+
+int amdgpu_device_reinit_after_reset(struct amdgpu_reset_context *reset_context);
+
+int emu_soc_asic_init(struct amdgpu_device *adev);
 
 /*
  * Registers read & write functions.
  */
-
-#define AMDGPU_REGS_IDX       (1<<0)
 #define AMDGPU_REGS_NO_KIQ    (1<<1)
+#define AMDGPU_REGS_RLC	(1<<2)
+
+#define RREG32_NO_KIQ(reg) amdgpu_device_rreg(adev, (reg), AMDGPU_REGS_NO_KIQ)
+#define WREG32_NO_KIQ(reg, v) amdgpu_device_wreg(adev, (reg), (v), AMDGPU_REGS_NO_KIQ)
 
-#define RREG32_NO_KIQ(reg) amdgpu_mm_rreg(adev, (reg), AMDGPU_REGS_NO_KIQ)
-#define WREG32_NO_KIQ(reg, v) amdgpu_mm_wreg(adev, (reg), (v), AMDGPU_REGS_NO_KIQ)
+#define RREG32_KIQ(reg) amdgpu_kiq_rreg(adev, (reg), 0)
+#define WREG32_KIQ(reg, v) amdgpu_kiq_wreg(adev, (reg), (v), 0)
 
-#define RREG32(reg) amdgpu_mm_rreg(adev, (reg), 0)
-#define RREG32_IDX(reg) amdgpu_mm_rreg(adev, (reg), AMDGPU_REGS_IDX)
-#define DREG32(reg) printk(KERN_INFO "REGISTER: " #reg " : 0x%08X\n", amdgpu_mm_rreg(adev, (reg), 0))
-#define WREG32(reg, v) amdgpu_mm_wreg(adev, (reg), (v), 0)
-#define WREG32_IDX(reg, v) amdgpu_mm_wreg(adev, (reg), (v), AMDGPU_REGS_IDX)
+#define RREG8(reg) amdgpu_mm_rreg8(adev, (reg))
+#define WREG8(reg, v) amdgpu_mm_wreg8(adev, (reg), (v))
+
+#define RREG32(reg) amdgpu_device_rreg(adev, (reg), 0)
+#define DREG32(reg) printk(KERN_INFO "REGISTER: " #reg " : 0x%08X\n", amdgpu_device_rreg(adev, (reg), 0))
+#define WREG32(reg, v) amdgpu_device_wreg(adev, (reg), (v), 0)
 #define REG_SET(FIELD, v) (((v) << FIELD##_SHIFT) & FIELD##_MASK)
 #define REG_GET(FIELD, v) (((v) << FIELD##_SHIFT) & FIELD##_MASK)
+#define RREG32_XCC(reg, inst) amdgpu_device_xcc_rreg(adev, (reg), 0, inst)
+#define WREG32_XCC(reg, v, inst) amdgpu_device_xcc_wreg(adev, (reg), (v), 0, inst)
 #define RREG32_PCIE(reg) adev->pcie_rreg(adev, (reg))
 #define WREG32_PCIE(reg, v) adev->pcie_wreg(adev, (reg), (v))
 #define RREG32_PCIE_PORT(reg) adev->pciep_rreg(adev, (reg))
 #define WREG32_PCIE_PORT(reg, v) adev->pciep_wreg(adev, (reg), (v))
+#define RREG32_PCIE_EXT(reg) adev->pcie_rreg_ext(adev, (reg))
+#define WREG32_PCIE_EXT(reg, v) adev->pcie_wreg_ext(adev, (reg), (v))
+#define RREG64_PCIE(reg) adev->pcie_rreg64(adev, (reg))
+#define WREG64_PCIE(reg, v) adev->pcie_wreg64(adev, (reg), (v))
+#define RREG64_PCIE_EXT(reg) adev->pcie_rreg64_ext(adev, (reg))
+#define WREG64_PCIE_EXT(reg, v) adev->pcie_wreg64_ext(adev, (reg), (v))
 #define RREG32_SMC(reg) adev->smc_rreg(adev, (reg))
 #define WREG32_SMC(reg, v) adev->smc_wreg(adev, (reg), (v))
 #define RREG32_UVD_CTX(reg) adev->uvd_ctx_rreg(adev, (reg))
@@ -1666,6 +1468,8 @@ void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v);
 #define WREG32_DIDT(reg, v) adev->didt_wreg(adev, (reg), (v))
 #define RREG32_GC_CAC(reg) adev->gc_cac_rreg(adev, (reg))
 #define WREG32_GC_CAC(reg, v) adev->gc_cac_wreg(adev, (reg), (v))
+#define RREG32_SE_CAC(reg) adev->se_cac_rreg(adev, (reg))
+#define WREG32_SE_CAC(reg, v) adev->se_cac_wreg(adev, (reg), (v))
 #define RREG32_AUDIO_ENDPT(block, reg) adev->audio_endpt_rreg(adev, (block), (reg))
 #define WREG32_AUDIO_ENDPT(block, reg, v) adev->audio_endpt_wreg(adev, (block), (reg), (v))
 #define WREG32_P(reg, val, mask)				\
@@ -1684,14 +1488,16 @@ void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v);
 		tmp_ |= ((val) & ~(mask));			\
 		WREG32_PLL(reg, tmp_);				\
 	} while (0)
-#define DREG32_SYS(sqf, adev, reg) seq_printf((sqf), #reg " : 0x%08X\n", amdgpu_mm_rreg((adev), (reg), false))
-#define RREG32_IO(reg) amdgpu_io_rreg(adev, (reg))
-#define WREG32_IO(reg, v) amdgpu_io_wreg(adev, (reg), (v))
 
-#define RDOORBELL32(index) amdgpu_mm_rdoorbell(adev, (index))
-#define WDOORBELL32(index, v) amdgpu_mm_wdoorbell(adev, (index), (v))
-#define RDOORBELL64(index) amdgpu_mm_rdoorbell64(adev, (index))
-#define WDOORBELL64(index, v) amdgpu_mm_wdoorbell64(adev, (index), (v))
+#define WREG32_SMC_P(_Reg, _Val, _Mask)                         \
+	do {                                                    \
+		u32 tmp = RREG32_SMC(_Reg);                     \
+		tmp &= (_Mask);                                 \
+		tmp |= ((_Val) & ~(_Mask));                     \
+		WREG32_SMC(_Reg, tmp);                          \
+	} while (0)
+
+#define DREG32_SYS(sqf, adev, reg) seq_printf((sqf), #reg " : 0x%08X\n", amdgpu_device_rreg((adev), (reg), false))
 
 #define REG_FIELD_SHIFT(reg, field) reg##__##field##__SHIFT
 #define REG_FIELD_MASK(reg, field) reg##__##field##_MASK
@@ -1709,6 +1515,7 @@ void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v);
 #define WREG32_FIELD_OFFSET(reg, offset, field, val)	\
 	WREG32(mm##reg + offset, (RREG32(mm##reg + offset) & ~REG_FIELD_MASK(reg, field)) | (val) << REG_FIELD_SHIFT(reg, field))
 
+#define AMDGPU_GET_REG_FIELD(x, h, l) (((x) & GENMASK_ULL(h, l)) >> (l))
 /*
  * BIOS helpers.
  */
@@ -1717,69 +1524,12 @@ void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v);
 #define RBIOS32(i) ((RBIOS16(i)) | (RBIOS16((i)+2) << 16))
 
 /*
- * RING helpers.
- */
-static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
-{
-	if (ring->count_dw <= 0)
-		DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
-	ring->ring[ring->wptr++ & ring->buf_mask] = v;
-	ring->wptr &= ring->ptr_mask;
-	ring->count_dw--;
-}
-
-static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring, void *src, int count_dw)
-{
-	unsigned occupied, chunk1, chunk2;
-	void *dst;
-
-	if (ring->count_dw < count_dw) {
-		DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
-	} else {
-		occupied = ring->wptr & ring->buf_mask;
-		dst = (void *)&ring->ring[occupied];
-		chunk1 = ring->buf_mask + 1 - occupied;
-		chunk1 = (chunk1 >= count_dw) ? count_dw: chunk1;
-		chunk2 = count_dw - chunk1;
-		chunk1 <<= 2;
-		chunk2 <<= 2;
-
-		if (chunk1)
-			memcpy(dst, src, chunk1);
-
-		if (chunk2) {
-			src += chunk1;
-			dst = (void *)ring->ring;
-			memcpy(dst, src, chunk2);
-		}
-
-		ring->wptr += count_dw;
-		ring->wptr &= ring->ptr_mask;
-		ring->count_dw -= count_dw;
-	}
-}
-
-static inline struct amdgpu_sdma_instance *
-amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
-{
-	struct amdgpu_device *adev = ring->adev;
-	int i;
-
-	for (i = 0; i < adev->sdma.num_instances; i++)
-		if (&adev->sdma.instance[i].ring == ring)
-			break;
-
-	if (i < AMDGPU_MAX_SDMA_INSTANCES)
-		return &adev->sdma.instance[i];
-	else
-		return NULL;
-}
-
-/*
  * ASICs macro.
  */
-#define amdgpu_asic_set_vga_state(adev, state) (adev)->asic_funcs->set_vga_state((adev), (state))
+#define amdgpu_asic_set_vga_state(adev, state) \
+    ((adev)->asic_funcs->set_vga_state ? (adev)->asic_funcs->set_vga_state((adev), (state)) : 0)
 #define amdgpu_asic_reset(adev) (adev)->asic_funcs->reset((adev))
+#define amdgpu_asic_reset_method(adev) (adev)->asic_funcs->reset_method((adev))
 #define amdgpu_asic_get_xclk(adev) (adev)->asic_funcs->get_xclk((adev))
 #define amdgpu_asic_set_uvd_clocks(adev, v, d) (adev)->asic_funcs->set_uvd_clocks((adev), (v), (d))
 #define amdgpu_asic_set_vce_clocks(adev, ev, ec) (adev)->asic_funcs->set_vce_clocks((adev), (ev), (ec))
@@ -1790,107 +1540,88 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
 #define amdgpu_asic_read_bios_from_rom(adev, b, l) (adev)->asic_funcs->read_bios_from_rom((adev), (b), (l))
 #define amdgpu_asic_read_register(adev, se, sh, offset, v)((adev)->asic_funcs->read_register((adev), (se), (sh), (offset), (v)))
 #define amdgpu_asic_get_config_memsize(adev) (adev)->asic_funcs->get_config_memsize((adev))
-#define amdgpu_gart_flush_gpu_tlb(adev, vmid) (adev)->gart.gart_funcs->flush_gpu_tlb((adev), (vmid))
-#define amdgpu_gart_set_pte_pde(adev, pt, idx, addr, flags) (adev)->gart.gart_funcs->set_pte_pde((adev), (pt), (idx), (addr), (flags))
-#define amdgpu_vm_copy_pte(adev, ib, pe, src, count) ((adev)->vm_manager.vm_pte_funcs->copy_pte((ib), (pe), (src), (count)))
-#define amdgpu_vm_write_pte(adev, ib, pe, value, count, incr) ((adev)->vm_manager.vm_pte_funcs->write_pte((ib), (pe), (value), (count), (incr)))
-#define amdgpu_vm_set_pte_pde(adev, ib, pe, addr, count, incr, flags) ((adev)->vm_manager.vm_pte_funcs->set_pte_pde((ib), (pe), (addr), (count), (incr), (flags)))
-#define amdgpu_vm_get_pte_flags(adev, flags) (adev)->gart.gart_funcs->get_vm_pte_flags((adev),(flags))
-#define amdgpu_ring_parse_cs(r, p, ib) ((r)->funcs->parse_cs((p), (ib)))
-#define amdgpu_ring_test_ring(r) (r)->funcs->test_ring((r))
-#define amdgpu_ring_test_ib(r, t) (r)->funcs->test_ib((r), (t))
-#define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r))
-#define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r))
-#define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r))
-#define amdgpu_ring_emit_ib(r, ib, vm_id, c) (r)->funcs->emit_ib((r), (ib), (vm_id), (c))
-#define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r))
-#define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr))
-#define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags))
-#define amdgpu_ring_emit_gds_switch(r, v, db, ds, wb, ws, ab, as) (r)->funcs->emit_gds_switch((r), (v), (db), (ds), (wb), (ws), (ab), (as))
-#define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r))
-#define amdgpu_ring_emit_hdp_invalidate(r) (r)->funcs->emit_hdp_invalidate((r))
-#define amdgpu_ring_emit_switch_buffer(r) (r)->funcs->emit_switch_buffer((r))
-#define amdgpu_ring_emit_cntxcntl(r, d) (r)->funcs->emit_cntxcntl((r), (d))
-#define amdgpu_ring_emit_rreg(r, d) (r)->funcs->emit_rreg((r), (d))
-#define amdgpu_ring_emit_wreg(r, d, v) (r)->funcs->emit_wreg((r), (d), (v))
-#define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib)))
-#define amdgpu_ring_init_cond_exec(r) (r)->funcs->init_cond_exec((r))
-#define amdgpu_ring_patch_cond_exec(r,o) (r)->funcs->patch_cond_exec((r),(o))
-#define amdgpu_ih_get_wptr(adev) (adev)->irq.ih_funcs->get_wptr((adev))
-#define amdgpu_ih_decode_iv(adev, iv) (adev)->irq.ih_funcs->decode_iv((adev), (iv))
-#define amdgpu_ih_set_rptr(adev) (adev)->irq.ih_funcs->set_rptr((adev))
-#define amdgpu_display_set_vga_render_state(adev, r) (adev)->mode_info.funcs->set_vga_render_state((adev), (r))
-#define amdgpu_display_vblank_get_counter(adev, crtc) (adev)->mode_info.funcs->vblank_get_counter((adev), (crtc))
-#define amdgpu_display_vblank_wait(adev, crtc) (adev)->mode_info.funcs->vblank_wait((adev), (crtc))
-#define amdgpu_display_backlight_set_level(adev, e, l) (adev)->mode_info.funcs->backlight_set_level((e), (l))
-#define amdgpu_display_backlight_get_level(adev, e) (adev)->mode_info.funcs->backlight_get_level((e))
-#define amdgpu_display_hpd_sense(adev, h) (adev)->mode_info.funcs->hpd_sense((adev), (h))
-#define amdgpu_display_hpd_set_polarity(adev, h) (adev)->mode_info.funcs->hpd_set_polarity((adev), (h))
-#define amdgpu_display_hpd_get_gpio_reg(adev) (adev)->mode_info.funcs->hpd_get_gpio_reg((adev))
-#define amdgpu_display_bandwidth_update(adev) (adev)->mode_info.funcs->bandwidth_update((adev))
-#define amdgpu_display_page_flip(adev, crtc, base, async) (adev)->mode_info.funcs->page_flip((adev), (crtc), (base), (async))
-#define amdgpu_display_page_flip_get_scanoutpos(adev, crtc, vbl, pos) (adev)->mode_info.funcs->page_flip_get_scanoutpos((adev), (crtc), (vbl), (pos))
-#define amdgpu_display_add_encoder(adev, e, s, c) (adev)->mode_info.funcs->add_encoder((adev), (e), (s), (c))
-#define amdgpu_display_add_connector(adev, ci, sd, ct, ib, coi, h, r) (adev)->mode_info.funcs->add_connector((adev), (ci), (sd), (ct), (ib), (coi), (h), (r))
-#define amdgpu_display_stop_mc_access(adev, s) (adev)->mode_info.funcs->stop_mc_access((adev), (s))
-#define amdgpu_display_resume_mc_access(adev, s) (adev)->mode_info.funcs->resume_mc_access((adev), (s))
-#define amdgpu_emit_copy_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_copy_buffer((ib),  (s), (d), (b))
-#define amdgpu_emit_fill_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((ib), (s), (d), (b))
-#define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
-#define amdgpu_gfx_select_se_sh(adev, se, sh, instance) (adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance))
-#define amdgpu_gds_switch(adev, r, v, d, w, a) (adev)->gds.funcs->patch_gds_switch((r), (v), (d), (w), (a))
-#define amdgpu_psp_check_fw_loading_status(adev, i) (adev)->firmware.funcs->check_fw_loading_status((adev), (i))
+#define amdgpu_asic_need_full_reset(adev) (adev)->asic_funcs->need_full_reset((adev))
+#define amdgpu_asic_init_doorbell_index(adev) (adev)->asic_funcs->init_doorbell_index((adev))
+#define amdgpu_asic_get_pcie_usage(adev, cnt0, cnt1) ((adev)->asic_funcs->get_pcie_usage((adev), (cnt0), (cnt1)))
+#define amdgpu_asic_need_reset_on_init(adev) (adev)->asic_funcs->need_reset_on_init((adev))
+#define amdgpu_asic_get_pcie_replay_count(adev) ((adev)->asic_funcs->get_pcie_replay_count((adev)))
+#define amdgpu_asic_supports_baco(adev) (adev)->asic_funcs->supports_baco((adev))
+#define amdgpu_asic_pre_asic_init(adev) (adev)->asic_funcs->pre_asic_init((adev))
+#define amdgpu_asic_update_umd_stable_pstate(adev, enter) \
+	((adev)->asic_funcs->update_umd_stable_pstate ? (adev)->asic_funcs->update_umd_stable_pstate((adev), (enter)) : 0)
+#define amdgpu_asic_query_video_codecs(adev, e, c) (adev)->asic_funcs->query_video_codecs((adev), (e), (c))
+
+#define amdgpu_inc_vram_lost(adev) atomic_inc(&((adev)->vram_lost_counter))
+
+#define BIT_MASK_UPPER(i) ((i) >= BITS_PER_LONG ? 0 : ~0UL << (i))
+#define for_each_inst(i, inst_mask)        \
+	for (i = ffs(inst_mask); i-- != 0; \
+	     i = ffs(inst_mask & BIT_MASK_UPPER(i + 1)))
 
 /* Common functions */
-int amdgpu_gpu_reset(struct amdgpu_device *adev);
-bool amdgpu_need_backup(struct amdgpu_device *adev);
-void amdgpu_pci_config_reset(struct amdgpu_device *adev);
-bool amdgpu_need_post(struct amdgpu_device *adev);
-void amdgpu_update_display_priority(struct amdgpu_device *adev);
-
-int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data);
-int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
-		       u32 ip_instance, u32 ring,
-		       struct amdgpu_ring **out_ring);
-void amdgpu_cs_report_moved_bytes(struct amdgpu_device *adev, u64 num_bytes);
-void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *abo, u32 domain);
-bool amdgpu_ttm_bo_is_amdgpu_bo(struct ttm_buffer_object *bo);
-int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages);
-int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
-				     uint32_t flags);
-bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm);
-struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm);
-bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
-				  unsigned long end);
-bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
-				       int *last_invalidated);
-bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm);
-uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
-				 struct ttm_mem_reg *mem);
-void amdgpu_vram_location(struct amdgpu_device *adev, struct amdgpu_mc *mc, u64 base);
-void amdgpu_gtt_location(struct amdgpu_device *adev, struct amdgpu_mc *mc);
-void amdgpu_ttm_set_active_vram_size(struct amdgpu_device *adev, u64 size);
-int amdgpu_ttm_init(struct amdgpu_device *adev);
-void amdgpu_ttm_fini(struct amdgpu_device *adev);
-void amdgpu_program_register_sequence(struct amdgpu_device *adev,
+bool amdgpu_device_has_job_running(struct amdgpu_device *adev);
+bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev);
+int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
+			      struct amdgpu_job *job,
+			      struct amdgpu_reset_context *reset_context);
+void amdgpu_device_pci_config_reset(struct amdgpu_device *adev);
+int amdgpu_device_pci_reset(struct amdgpu_device *adev);
+bool amdgpu_device_need_post(struct amdgpu_device *adev);
+bool amdgpu_device_seamless_boot_supported(struct amdgpu_device *adev);
+bool amdgpu_device_should_use_aspm(struct amdgpu_device *adev);
+
+void amdgpu_cs_report_moved_bytes(struct amdgpu_device *adev, u64 num_bytes,
+				  u64 num_vis_bytes);
+int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev);
+void amdgpu_device_program_register_sequence(struct amdgpu_device *adev,
 					     const u32 *registers,
 					     const u32 array_size);
 
-bool amdgpu_device_is_px(struct drm_device *dev);
+int amdgpu_device_mode1_reset(struct amdgpu_device *adev);
+int amdgpu_device_link_reset(struct amdgpu_device *adev);
+bool amdgpu_device_supports_atpx(struct amdgpu_device *adev);
+bool amdgpu_device_supports_px(struct amdgpu_device *adev);
+bool amdgpu_device_supports_boco(struct amdgpu_device *adev);
+bool amdgpu_device_supports_smart_shift(struct amdgpu_device *adev);
+int amdgpu_device_supports_baco(struct amdgpu_device *adev);
+void amdgpu_device_detect_runtime_pm_mode(struct amdgpu_device *adev);
+bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev,
+				      struct amdgpu_device *peer_adev);
+int amdgpu_device_baco_enter(struct amdgpu_device *adev);
+int amdgpu_device_baco_exit(struct amdgpu_device *adev);
+
+void amdgpu_device_flush_hdp(struct amdgpu_device *adev,
+		struct amdgpu_ring *ring);
+void amdgpu_device_invalidate_hdp(struct amdgpu_device *adev,
+		struct amdgpu_ring *ring);
+
+void amdgpu_device_halt(struct amdgpu_device *adev);
+u32 amdgpu_device_pcie_port_rreg(struct amdgpu_device *adev,
+				u32 reg);
+void amdgpu_device_pcie_port_wreg(struct amdgpu_device *adev,
+				u32 reg, u32 v);
+struct dma_fence *amdgpu_device_get_gang(struct amdgpu_device *adev);
+struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev,
+					    struct dma_fence *gang);
+struct dma_fence *amdgpu_device_enforce_isolation(struct amdgpu_device *adev,
+						  struct amdgpu_ring *ring,
+						  struct amdgpu_job *job);
+bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev);
+ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring);
+ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset);
+
 /* atpx handler */
 #if defined(CONFIG_VGA_SWITCHEROO)
 void amdgpu_register_atpx_handler(void);
 void amdgpu_unregister_atpx_handler(void);
 bool amdgpu_has_atpx_dgpu_power_cntl(void);
 bool amdgpu_is_atpx_hybrid(void);
-bool amdgpu_atpx_dgpu_req_power_for_displays(void);
 bool amdgpu_has_atpx(void);
 #else
 static inline void amdgpu_register_atpx_handler(void) {}
 static inline void amdgpu_unregister_atpx_handler(void) {}
 static inline bool amdgpu_has_atpx_dgpu_power_cntl(void) { return false; }
 static inline bool amdgpu_is_atpx_hybrid(void) { return false; }
-static inline bool amdgpu_atpx_dgpu_req_power_for_displays(void) { return false; }
 static inline bool amdgpu_has_atpx(void) { return false; }
 #endif
 
@@ -1900,24 +1631,22 @@ static inline bool amdgpu_has_atpx(void) { return false; }
 extern const struct drm_ioctl_desc amdgpu_ioctls_kms[];
 extern const int amdgpu_max_kms_ioctl;
 
-int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags);
+int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags);
 void amdgpu_driver_unload_kms(struct drm_device *dev);
-void amdgpu_driver_lastclose_kms(struct drm_device *dev);
 int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv);
 void amdgpu_driver_postclose_kms(struct drm_device *dev,
 				 struct drm_file *file_priv);
-int amdgpu_suspend(struct amdgpu_device *adev);
-int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon);
-int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon);
-u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe);
-int amdgpu_enable_vblank_kms(struct drm_device *dev, unsigned int pipe);
-void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe);
-int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, unsigned int pipe,
-				    int *max_error,
-				    struct timeval *vblank_time,
-				    unsigned flags);
-long amdgpu_kms_compat_ioctl(struct file *filp, unsigned int cmd,
-			     unsigned long arg);
+void amdgpu_driver_release_kms(struct drm_device *dev);
+
+int amdgpu_device_prepare(struct drm_device *dev);
+void amdgpu_device_complete(struct drm_device *dev);
+int amdgpu_device_suspend(struct drm_device *dev, bool fbcon);
+int amdgpu_device_resume(struct drm_device *dev, bool fbcon);
+u32 amdgpu_get_vblank_counter_kms(struct drm_crtc *crtc);
+int amdgpu_enable_vblank_kms(struct drm_crtc *crtc);
+void amdgpu_disable_vblank_kms(struct drm_crtc *crtc);
+int amdgpu_info_ioctl(struct drm_device *dev, void *data,
+		      struct drm_file *filp);
 
 /*
  * functions used by amdgpu_encoder.c
@@ -1939,22 +1668,141 @@ struct amdgpu_afmt_acr {
 struct amdgpu_afmt_acr amdgpu_afmt_acr(uint32_t clock);
 
 /* amdgpu_acpi.c */
+
+struct amdgpu_numa_info {
+	uint64_t size;
+	int pxm;
+	int nid;
+};
+
+/* ATCS Device/Driver State */
+#define AMDGPU_ATCS_PSC_DEV_STATE_D0		0
+#define AMDGPU_ATCS_PSC_DEV_STATE_D3_HOT	3
+#define AMDGPU_ATCS_PSC_DRV_STATE_OPR		0
+#define AMDGPU_ATCS_PSC_DRV_STATE_NOT_OPR	1
+
 #if defined(CONFIG_ACPI)
 int amdgpu_acpi_init(struct amdgpu_device *adev);
 void amdgpu_acpi_fini(struct amdgpu_device *adev);
 bool amdgpu_acpi_is_pcie_performance_request_supported(struct amdgpu_device *adev);
+bool amdgpu_acpi_is_power_shift_control_supported(void);
 int amdgpu_acpi_pcie_performance_request(struct amdgpu_device *adev,
 						u8 perf_req, bool advertise);
+int amdgpu_acpi_power_shift_control(struct amdgpu_device *adev,
+				    u8 dev_state, bool drv_state);
+int amdgpu_acpi_smart_shift_update(struct amdgpu_device *adev,
+				   enum amdgpu_ss ss_state);
 int amdgpu_acpi_pcie_notify_device_ready(struct amdgpu_device *adev);
+int amdgpu_acpi_get_tmr_info(struct amdgpu_device *adev, u64 *tmr_offset,
+			     u64 *tmr_size);
+int amdgpu_acpi_get_mem_info(struct amdgpu_device *adev, int xcc_id,
+			     struct amdgpu_numa_info *numa_info);
+
+void amdgpu_acpi_get_backlight_caps(struct amdgpu_dm_backlight_caps *caps);
+bool amdgpu_acpi_should_gpu_reset(struct amdgpu_device *adev);
+void amdgpu_acpi_detect(void);
+void amdgpu_acpi_release(void);
 #else
 static inline int amdgpu_acpi_init(struct amdgpu_device *adev) { return 0; }
+static inline int amdgpu_acpi_get_tmr_info(struct amdgpu_device *adev,
+					   u64 *tmr_offset, u64 *tmr_size)
+{
+	return -EINVAL;
+}
+static inline int amdgpu_acpi_get_mem_info(struct amdgpu_device *adev,
+					   int xcc_id,
+					   struct amdgpu_numa_info *numa_info)
+{
+	return -EINVAL;
+}
 static inline void amdgpu_acpi_fini(struct amdgpu_device *adev) { }
+static inline bool amdgpu_acpi_should_gpu_reset(struct amdgpu_device *adev) { return false; }
+static inline void amdgpu_acpi_detect(void) { }
+static inline void amdgpu_acpi_release(void) { }
+static inline bool amdgpu_acpi_is_power_shift_control_supported(void) { return false; }
+static inline int amdgpu_acpi_power_shift_control(struct amdgpu_device *adev,
+						  u8 dev_state, bool drv_state) { return 0; }
+static inline int amdgpu_acpi_smart_shift_update(struct amdgpu_device *adev,
+						 enum amdgpu_ss ss_state)
+{
+	return 0;
+}
+static inline void amdgpu_acpi_get_backlight_caps(struct amdgpu_dm_backlight_caps *caps) { }
+#endif
+
+#if defined(CONFIG_ACPI) && defined(CONFIG_SUSPEND)
+bool amdgpu_acpi_is_s3_active(struct amdgpu_device *adev);
+bool amdgpu_acpi_is_s0ix_active(struct amdgpu_device *adev);
+#else
+static inline bool amdgpu_acpi_is_s0ix_active(struct amdgpu_device *adev) { return false; }
+static inline bool amdgpu_acpi_is_s3_active(struct amdgpu_device *adev) { return false; }
 #endif
 
-struct amdgpu_bo_va_mapping *
-amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
-		       uint64_t addr, struct amdgpu_bo **bo);
-int amdgpu_cs_sysvm_access_required(struct amdgpu_cs_parser *parser);
+#if defined(CONFIG_DRM_AMD_ISP)
+int amdgpu_acpi_get_isp4_dev(struct acpi_device **dev);
+#endif
+
+void amdgpu_register_gpu_instance(struct amdgpu_device *adev);
+void amdgpu_unregister_gpu_instance(struct amdgpu_device *adev);
+
+pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev,
+					   pci_channel_state_t state);
+pci_ers_result_t amdgpu_pci_mmio_enabled(struct pci_dev *pdev);
+pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev);
+void amdgpu_pci_resume(struct pci_dev *pdev);
+
+bool amdgpu_device_cache_pci_state(struct pci_dev *pdev);
+bool amdgpu_device_load_pci_state(struct pci_dev *pdev);
+
+bool amdgpu_device_skip_hw_access(struct amdgpu_device *adev);
+
+int amdgpu_device_set_cg_state(struct amdgpu_device *adev,
+			       enum amd_clockgating_state state);
+int amdgpu_device_set_pg_state(struct amdgpu_device *adev,
+			       enum amd_powergating_state state);
+
+static inline bool amdgpu_device_has_timeouts_enabled(struct amdgpu_device *adev)
+{
+	return amdgpu_gpu_recovery != 0 &&
+		adev->gfx_timeout != MAX_SCHEDULE_TIMEOUT &&
+		adev->compute_timeout != MAX_SCHEDULE_TIMEOUT &&
+		adev->sdma_timeout != MAX_SCHEDULE_TIMEOUT &&
+		adev->video_timeout != MAX_SCHEDULE_TIMEOUT;
+}
 
 #include "amdgpu_object.h"
+
+static inline bool amdgpu_is_tmz(struct amdgpu_device *adev)
+{
+       return adev->gmc.tmz_enabled;
+}
+
+int amdgpu_in_reset(struct amdgpu_device *adev);
+
+extern const struct attribute_group amdgpu_vram_mgr_attr_group;
+extern const struct attribute_group amdgpu_gtt_mgr_attr_group;
+extern const struct attribute_group amdgpu_flash_attr_group;
+
+void amdgpu_set_init_level(struct amdgpu_device *adev,
+			   enum amdgpu_init_lvl_id lvl);
+
+static inline int amdgpu_device_bus_status_check(struct amdgpu_device *adev)
+{
+       u32 status;
+       int r;
+
+       r = pci_read_config_dword(adev->pdev, PCI_COMMAND, &status);
+       if (r || PCI_POSSIBLE_ERROR(status)) {
+		dev_err(adev->dev, "device lost from bus!");
+		return -ENODEV;
+       }
+
+       return 0;
+}
+
+void amdgpu_device_set_uid(struct amdgpu_uid *uid_info,
+			   enum amdgpu_uid_type type, uint8_t inst,
+			   uint64_t uid);
+uint64_t amdgpu_device_get_uid(struct amdgpu_uid *uid_info,
+			       enum amdgpu_uid_type type, uint8_t inst);
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_aca.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_aca.c
new file mode 100644
index 000000000000..9b3180449150
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_aca.c
@@ -0,0 +1,984 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/list.h>
+#include "amdgpu.h"
+#include "amdgpu_aca.h"
+#include "amdgpu_ras.h"
+
+#define ACA_BANK_HWID(type, hwid, mcatype) [ACA_HWIP_TYPE_##type] = {hwid, mcatype}
+
+typedef int bank_handler_t(struct aca_handle *handle, struct aca_bank *bank, enum aca_smu_type type, void *data);
+
+static struct aca_hwip aca_hwid_mcatypes[ACA_HWIP_TYPE_COUNT] = {
+	ACA_BANK_HWID(SMU,	0x01,	0x01),
+	ACA_BANK_HWID(PCS_XGMI, 0x50,	0x00),
+	ACA_BANK_HWID(UMC,	0x96,	0x00),
+};
+
+static void aca_banks_init(struct aca_banks *banks)
+{
+	if (!banks)
+		return;
+
+	memset(banks, 0, sizeof(*banks));
+	INIT_LIST_HEAD(&banks->list);
+}
+
+static int aca_banks_add_bank(struct aca_banks *banks, struct aca_bank *bank)
+{
+	struct aca_bank_node *node;
+
+	if (!bank)
+		return -EINVAL;
+
+	node = kvzalloc(sizeof(*node), GFP_KERNEL);
+	if (!node)
+		return -ENOMEM;
+
+	memcpy(&node->bank, bank, sizeof(*bank));
+
+	INIT_LIST_HEAD(&node->node);
+	list_add_tail(&node->node, &banks->list);
+
+	banks->nr_banks++;
+
+	return 0;
+}
+
+static void aca_banks_release(struct aca_banks *banks)
+{
+	struct aca_bank_node *node, *tmp;
+
+	if (list_empty(&banks->list))
+		return;
+
+	list_for_each_entry_safe(node, tmp, &banks->list, node) {
+		list_del(&node->node);
+		kvfree(node);
+		banks->nr_banks--;
+	}
+}
+
+static int aca_smu_get_valid_aca_count(struct amdgpu_device *adev, enum aca_smu_type type, u32 *count)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	const struct aca_smu_funcs *smu_funcs = aca->smu_funcs;
+
+	if (!count)
+		return -EINVAL;
+
+	if (!smu_funcs || !smu_funcs->get_valid_aca_count)
+		return -EOPNOTSUPP;
+
+	return smu_funcs->get_valid_aca_count(adev, type, count);
+}
+
+static struct aca_regs_dump {
+	const char *name;
+	int reg_idx;
+} aca_regs[] = {
+	{"CONTROL",		ACA_REG_IDX_CTL},
+	{"STATUS",		ACA_REG_IDX_STATUS},
+	{"ADDR",		ACA_REG_IDX_ADDR},
+	{"MISC",		ACA_REG_IDX_MISC0},
+	{"CONFIG",		ACA_REG_IDX_CONFIG},
+	{"IPID",		ACA_REG_IDX_IPID},
+	{"SYND",		ACA_REG_IDX_SYND},
+	{"DESTAT",		ACA_REG_IDX_DESTAT},
+	{"DEADDR",		ACA_REG_IDX_DEADDR},
+	{"CONTROL_MASK",	ACA_REG_IDX_CTL_MASK},
+};
+
+static void aca_smu_bank_dump(struct amdgpu_device *adev, int idx, int total, struct aca_bank *bank,
+			      struct ras_query_context *qctx)
+{
+	u64 event_id = qctx ? qctx->evid.event_id : RAS_EVENT_INVALID_ID;
+	int i;
+
+	if (adev->debug_disable_ce_logs &&
+	    bank->smu_err_type == ACA_SMU_TYPE_CE &&
+	    !ACA_BANK_ERR_IS_DEFFERED(bank))
+		return;
+
+	RAS_EVENT_LOG(adev, event_id, HW_ERR "Accelerator Check Architecture events logged\n");
+	/* plus 1 for output format, e.g: ACA[08/08]: xxxx */
+	for (i = 0; i < ARRAY_SIZE(aca_regs); i++)
+		RAS_EVENT_LOG(adev, event_id, HW_ERR "ACA[%02d/%02d].%s=0x%016llx\n",
+			      idx + 1, total, aca_regs[i].name, bank->regs[aca_regs[i].reg_idx]);
+
+	if (ACA_REG__STATUS__SCRUB(bank->regs[ACA_REG_IDX_STATUS]))
+		RAS_EVENT_LOG(adev, event_id, HW_ERR "hardware error logged by the scrubber\n");
+}
+
+static bool aca_bank_hwip_is_matched(struct aca_bank *bank, enum aca_hwip_type type)
+{
+
+	struct aca_hwip *hwip;
+	int hwid, mcatype;
+	u64 ipid;
+
+	if (!bank || type == ACA_HWIP_TYPE_UNKNOW)
+		return false;
+
+	hwip = &aca_hwid_mcatypes[type];
+	if (!hwip->hwid)
+		return false;
+
+	ipid = bank->regs[ACA_REG_IDX_IPID];
+	hwid = ACA_REG__IPID__HARDWAREID(ipid);
+	mcatype = ACA_REG__IPID__MCATYPE(ipid);
+
+	return hwip->hwid == hwid && hwip->mcatype == mcatype;
+}
+
+static int aca_smu_get_valid_aca_banks(struct amdgpu_device *adev, enum aca_smu_type type,
+				       int start, int count,
+				       struct aca_banks *banks, struct ras_query_context *qctx)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	const struct aca_smu_funcs *smu_funcs = aca->smu_funcs;
+	struct aca_bank bank;
+	int i, max_count, ret;
+
+	if (!count)
+		return 0;
+
+	if (!smu_funcs || !smu_funcs->get_valid_aca_bank)
+		return -EOPNOTSUPP;
+
+	switch (type) {
+	case ACA_SMU_TYPE_UE:
+		max_count = smu_funcs->max_ue_bank_count;
+		break;
+	case ACA_SMU_TYPE_CE:
+		max_count = smu_funcs->max_ce_bank_count;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (start + count > max_count)
+		return -EINVAL;
+
+	count = min_t(int, count, max_count);
+	for (i = 0; i < count; i++) {
+		memset(&bank, 0, sizeof(bank));
+		ret = smu_funcs->get_valid_aca_bank(adev, type, start + i, &bank);
+		if (ret)
+			return ret;
+
+		bank.smu_err_type = type;
+
+		/*
+		 * Poison being consumed when injecting a UE while running background workloads,
+		 * which are unexpected.
+		 */
+		if (type == ACA_SMU_TYPE_UE &&
+		    ACA_REG__STATUS__POISON(bank.regs[ACA_REG_IDX_STATUS]) &&
+		    !aca_bank_hwip_is_matched(&bank, ACA_HWIP_TYPE_UMC))
+			continue;
+
+		aca_smu_bank_dump(adev, i, count, &bank, qctx);
+
+		ret = aca_banks_add_bank(banks, &bank);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static bool aca_bank_is_valid(struct aca_handle *handle, struct aca_bank *bank, enum aca_smu_type type)
+{
+	const struct aca_bank_ops *bank_ops = handle->bank_ops;
+
+	/* Parse all deferred errors with UMC aca handle */
+	if (ACA_BANK_ERR_IS_DEFFERED(bank))
+		return handle->hwip == ACA_HWIP_TYPE_UMC;
+
+	if (!aca_bank_hwip_is_matched(bank, handle->hwip))
+		return false;
+
+	if (!bank_ops->aca_bank_is_valid)
+		return true;
+
+	return bank_ops->aca_bank_is_valid(handle, bank, type, handle->data);
+}
+
+static struct aca_bank_error *new_bank_error(struct aca_error *aerr, struct aca_bank_info *info)
+{
+	struct aca_bank_error *bank_error;
+
+	bank_error = kvzalloc(sizeof(*bank_error), GFP_KERNEL);
+	if (!bank_error)
+		return NULL;
+
+	INIT_LIST_HEAD(&bank_error->node);
+	memcpy(&bank_error->info, info, sizeof(*info));
+
+	mutex_lock(&aerr->lock);
+	list_add_tail(&bank_error->node, &aerr->list);
+	aerr->nr_errors++;
+	mutex_unlock(&aerr->lock);
+
+	return bank_error;
+}
+
+static struct aca_bank_error *find_bank_error(struct aca_error *aerr, struct aca_bank_info *info)
+{
+	struct aca_bank_error *bank_error = NULL;
+	struct aca_bank_info *tmp_info;
+	bool found = false;
+
+	mutex_lock(&aerr->lock);
+	list_for_each_entry(bank_error, &aerr->list, node) {
+		tmp_info = &bank_error->info;
+		if (tmp_info->socket_id == info->socket_id &&
+		    tmp_info->die_id == info->die_id) {
+			found = true;
+			goto out_unlock;
+		}
+	}
+
+out_unlock:
+	mutex_unlock(&aerr->lock);
+
+	return found ? bank_error : NULL;
+}
+
+static void aca_bank_error_remove(struct aca_error *aerr, struct aca_bank_error *bank_error)
+{
+	if (!aerr || !bank_error)
+		return;
+
+	list_del(&bank_error->node);
+	aerr->nr_errors--;
+
+	kvfree(bank_error);
+}
+
+static struct aca_bank_error *get_bank_error(struct aca_error *aerr, struct aca_bank_info *info)
+{
+	struct aca_bank_error *bank_error;
+
+	if (!aerr || !info)
+		return NULL;
+
+	bank_error = find_bank_error(aerr, info);
+	if (bank_error)
+		return bank_error;
+
+	return new_bank_error(aerr, info);
+}
+
+int aca_error_cache_log_bank_error(struct aca_handle *handle, struct aca_bank_info *info,
+				   enum aca_error_type type, u64 count)
+{
+	struct aca_error_cache *error_cache = &handle->error_cache;
+	struct aca_bank_error *bank_error;
+	struct aca_error *aerr;
+
+	if (!handle || !info || type >= ACA_ERROR_TYPE_COUNT)
+		return -EINVAL;
+
+	if (!count)
+		return 0;
+
+	aerr = &error_cache->errors[type];
+	bank_error = get_bank_error(aerr, info);
+	if (!bank_error)
+		return -ENOMEM;
+
+	bank_error->count += count;
+
+	return 0;
+}
+
+static int aca_bank_parser(struct aca_handle *handle, struct aca_bank *bank, enum aca_smu_type type)
+{
+	const struct aca_bank_ops *bank_ops = handle->bank_ops;
+
+	if (!bank)
+		return -EINVAL;
+
+	if (!bank_ops->aca_bank_parser)
+		return -EOPNOTSUPP;
+
+	return bank_ops->aca_bank_parser(handle, bank, type,
+					 handle->data);
+}
+
+static int handler_aca_log_bank_error(struct aca_handle *handle, struct aca_bank *bank,
+				      enum aca_smu_type type, void *data)
+{
+	int ret;
+
+	ret = aca_bank_parser(handle, bank, type);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int aca_dispatch_bank(struct aca_handle_manager *mgr, struct aca_bank *bank,
+			     enum aca_smu_type type, bank_handler_t handler, void *data)
+{
+	struct aca_handle *handle;
+	int ret;
+
+	if (list_empty(&mgr->list))
+		return 0;
+
+	list_for_each_entry(handle, &mgr->list, node) {
+		if (!aca_bank_is_valid(handle, bank, type))
+			continue;
+
+		ret = handler(handle, bank, type, data);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static int aca_dispatch_banks(struct aca_handle_manager *mgr, struct aca_banks *banks,
+			      enum aca_smu_type type, bank_handler_t handler, void *data)
+{
+	struct aca_bank_node *node;
+	struct aca_bank *bank;
+	int ret;
+
+	if (!mgr || !banks)
+		return -EINVAL;
+
+	/* pre check to avoid unnecessary operations */
+	if (list_empty(&mgr->list) || list_empty(&banks->list))
+		return 0;
+
+	list_for_each_entry(node, &banks->list, node) {
+		bank = &node->bank;
+
+		ret = aca_dispatch_bank(mgr, bank, type, handler, data);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static bool aca_bank_should_update(struct amdgpu_device *adev, enum aca_smu_type type)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	bool ret = true;
+
+	/*
+	 * Because the UE Valid MCA count will only be cleared after reset,
+	 * in order to avoid repeated counting of the error count,
+	 * the aca bank is only updated once during the gpu recovery stage.
+	 */
+	if (type == ACA_SMU_TYPE_UE) {
+		if (amdgpu_ras_intr_triggered())
+			ret = atomic_cmpxchg(&aca->ue_update_flag, 0, 1) == 0;
+		else
+			atomic_set(&aca->ue_update_flag, 0);
+	}
+
+	return ret;
+}
+
+static void aca_banks_generate_cper(struct amdgpu_device *adev,
+				    enum aca_smu_type type,
+				    struct aca_banks *banks,
+				    int count)
+{
+	struct aca_bank_node *node;
+	struct aca_bank *bank;
+	int r;
+
+	if (!adev->cper.enabled)
+		return;
+
+	if (!banks || !count) {
+		dev_warn(adev->dev, "fail to generate cper records\n");
+		return;
+	}
+
+	/* UEs must be encoded into separate CPER entries */
+	if (type == ACA_SMU_TYPE_UE) {
+		struct aca_banks de_banks;
+
+		aca_banks_init(&de_banks);
+		list_for_each_entry(node, &banks->list, node) {
+			bank = &node->bank;
+			if (bank->aca_err_type == ACA_ERROR_TYPE_DEFERRED) {
+				r = aca_banks_add_bank(&de_banks, bank);
+				if (r)
+					dev_warn(adev->dev, "fail to add de banks, ret = %d\n", r);
+			} else {
+				if (amdgpu_cper_generate_ue_record(adev, bank))
+					dev_warn(adev->dev, "fail to generate ue cper records\n");
+			}
+		}
+
+		if (!list_empty(&de_banks.list)) {
+			if (amdgpu_cper_generate_ce_records(adev, &de_banks, de_banks.nr_banks))
+				dev_warn(adev->dev, "fail to generate de cper records\n");
+		}
+
+		aca_banks_release(&de_banks);
+	} else {
+		/*
+		 * SMU_TYPE_CE banks are combined into 1 CPER entries,
+		 * they could be CEs or DEs or both
+		 */
+		if (amdgpu_cper_generate_ce_records(adev, banks, count))
+			dev_warn(adev->dev, "fail to generate ce cper records\n");
+	}
+}
+
+static int aca_banks_update(struct amdgpu_device *adev, enum aca_smu_type type,
+			    bank_handler_t handler, struct ras_query_context *qctx, void *data)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	struct aca_banks banks;
+	u32 count = 0;
+	int ret;
+
+	if (list_empty(&aca->mgr.list))
+		return 0;
+
+	if (!aca_bank_should_update(adev, type))
+		return 0;
+
+	ret = aca_smu_get_valid_aca_count(adev, type, &count);
+	if (ret)
+		return ret;
+
+	if (!count)
+		return 0;
+
+	aca_banks_init(&banks);
+
+	ret = aca_smu_get_valid_aca_banks(adev, type, 0, count, &banks, qctx);
+	if (ret)
+		goto err_release_banks;
+
+	if (list_empty(&banks.list)) {
+		ret = 0;
+		goto err_release_banks;
+	}
+
+	ret = aca_dispatch_banks(&aca->mgr, &banks, type,
+				 handler, data);
+	if (ret)
+		goto err_release_banks;
+
+	aca_banks_generate_cper(adev, type, &banks, count);
+
+err_release_banks:
+	aca_banks_release(&banks);
+
+	return ret;
+}
+
+static int aca_log_aca_error_data(struct aca_bank_error *bank_error, enum aca_error_type type, struct ras_err_data *err_data)
+{
+	struct aca_bank_info *info;
+	struct amdgpu_smuio_mcm_config_info mcm_info;
+	u64 count;
+
+	if (type >= ACA_ERROR_TYPE_COUNT)
+		return -EINVAL;
+
+	count = bank_error->count;
+	if (!count)
+		return 0;
+
+	info = &bank_error->info;
+	mcm_info.die_id = info->die_id;
+	mcm_info.socket_id = info->socket_id;
+
+	switch (type) {
+	case ACA_ERROR_TYPE_UE:
+		amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, count);
+		break;
+	case ACA_ERROR_TYPE_CE:
+		amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, count);
+		break;
+	case ACA_ERROR_TYPE_DEFERRED:
+		amdgpu_ras_error_statistic_de_count(err_data, &mcm_info, count);
+		break;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+static int aca_log_aca_error(struct aca_handle *handle, enum aca_error_type type, struct ras_err_data *err_data)
+{
+	struct aca_error_cache *error_cache = &handle->error_cache;
+	struct aca_error *aerr = &error_cache->errors[type];
+	struct aca_bank_error *bank_error, *tmp;
+
+	mutex_lock(&aerr->lock);
+
+	if (list_empty(&aerr->list))
+		goto out_unlock;
+
+	list_for_each_entry_safe(bank_error, tmp, &aerr->list, node) {
+		aca_log_aca_error_data(bank_error, type, err_data);
+		aca_bank_error_remove(aerr, bank_error);
+	}
+
+out_unlock:
+	mutex_unlock(&aerr->lock);
+
+	return 0;
+}
+
+static int __aca_get_error_data(struct amdgpu_device *adev, struct aca_handle *handle, enum aca_error_type type,
+				struct ras_err_data *err_data, struct ras_query_context *qctx)
+{
+	enum aca_smu_type smu_type;
+	int ret;
+
+	switch (type) {
+	case ACA_ERROR_TYPE_UE:
+		smu_type = ACA_SMU_TYPE_UE;
+		break;
+	case ACA_ERROR_TYPE_CE:
+	case ACA_ERROR_TYPE_DEFERRED:
+		smu_type = ACA_SMU_TYPE_CE;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	/* update aca bank to aca source error_cache first */
+	ret = aca_banks_update(adev, smu_type, handler_aca_log_bank_error, qctx, NULL);
+	if (ret)
+		return ret;
+
+	/* DEs may contain in CEs or UEs */
+	if (type != ACA_ERROR_TYPE_DEFERRED)
+		aca_log_aca_error(handle, ACA_ERROR_TYPE_DEFERRED, err_data);
+
+	return aca_log_aca_error(handle, type, err_data);
+}
+
+static bool aca_handle_is_valid(struct aca_handle *handle)
+{
+	if (!handle->mask || !list_empty(&handle->node))
+		return false;
+
+	return true;
+}
+
+int amdgpu_aca_get_error_data(struct amdgpu_device *adev, struct aca_handle *handle,
+			      enum aca_error_type type, struct ras_err_data *err_data,
+			      struct ras_query_context *qctx)
+{
+	if (!handle || !err_data)
+		return -EINVAL;
+
+	if (aca_handle_is_valid(handle))
+		return -EOPNOTSUPP;
+
+	if ((type < 0) || (!(BIT(type) & handle->mask)))
+		return  0;
+
+	return __aca_get_error_data(adev, handle, type, err_data, qctx);
+}
+
+static void aca_error_init(struct aca_error *aerr, enum aca_error_type type)
+{
+	mutex_init(&aerr->lock);
+	INIT_LIST_HEAD(&aerr->list);
+	aerr->type = type;
+	aerr->nr_errors = 0;
+}
+
+static void aca_init_error_cache(struct aca_handle *handle)
+{
+	struct aca_error_cache *error_cache = &handle->error_cache;
+	int type;
+
+	for (type = ACA_ERROR_TYPE_UE; type < ACA_ERROR_TYPE_COUNT; type++)
+		aca_error_init(&error_cache->errors[type], type);
+}
+
+static void aca_error_fini(struct aca_error *aerr)
+{
+	struct aca_bank_error *bank_error, *tmp;
+
+	mutex_lock(&aerr->lock);
+	if (list_empty(&aerr->list))
+		goto out_unlock;
+
+	list_for_each_entry_safe(bank_error, tmp, &aerr->list, node)
+		aca_bank_error_remove(aerr, bank_error);
+
+out_unlock:
+	mutex_destroy(&aerr->lock);
+}
+
+static void aca_fini_error_cache(struct aca_handle *handle)
+{
+	struct aca_error_cache *error_cache = &handle->error_cache;
+	int type;
+
+	for (type = ACA_ERROR_TYPE_UE; type < ACA_ERROR_TYPE_COUNT; type++)
+		aca_error_fini(&error_cache->errors[type]);
+}
+
+static int add_aca_handle(struct amdgpu_device *adev, struct aca_handle_manager *mgr, struct aca_handle *handle,
+			  const char *name, const struct aca_info *ras_info, void *data)
+{
+	memset(handle, 0, sizeof(*handle));
+
+	handle->adev = adev;
+	handle->mgr = mgr;
+	handle->name = name;
+	handle->hwip = ras_info->hwip;
+	handle->mask = ras_info->mask;
+	handle->bank_ops = ras_info->bank_ops;
+	handle->data = data;
+	aca_init_error_cache(handle);
+
+	INIT_LIST_HEAD(&handle->node);
+	list_add_tail(&handle->node, &mgr->list);
+	mgr->nr_handles++;
+
+	return 0;
+}
+
+static ssize_t aca_sysfs_read(struct device *dev,
+			      struct device_attribute *attr, char *buf)
+{
+	struct aca_handle *handle = container_of(attr, struct aca_handle, aca_attr);
+
+	/* NOTE: the aca cache will be auto cleared once read,
+	 * So the driver should unify the query entry point, forward request to ras query interface directly */
+	return amdgpu_ras_aca_sysfs_read(dev, attr, handle, buf, handle->data);
+}
+
+static int add_aca_sysfs(struct amdgpu_device *adev, struct aca_handle *handle)
+{
+	struct device_attribute *aca_attr = &handle->aca_attr;
+
+	snprintf(handle->attr_name, sizeof(handle->attr_name) - 1, "aca_%s", handle->name);
+	aca_attr->show = aca_sysfs_read;
+	aca_attr->attr.name = handle->attr_name;
+	aca_attr->attr.mode = S_IRUGO;
+	sysfs_attr_init(&aca_attr->attr);
+
+	return sysfs_add_file_to_group(&adev->dev->kobj,
+				       &aca_attr->attr,
+				       "ras");
+}
+
+int amdgpu_aca_add_handle(struct amdgpu_device *adev, struct aca_handle *handle,
+			  const char *name, const struct aca_info *ras_info, void *data)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	int ret;
+
+	if (!amdgpu_aca_is_enabled(adev))
+		return 0;
+
+	ret = add_aca_handle(adev, &aca->mgr, handle, name, ras_info, data);
+	if (ret)
+		return ret;
+
+	return add_aca_sysfs(adev, handle);
+}
+
+static void remove_aca_handle(struct aca_handle *handle)
+{
+	struct aca_handle_manager *mgr = handle->mgr;
+
+	aca_fini_error_cache(handle);
+	list_del(&handle->node);
+	mgr->nr_handles--;
+}
+
+static void remove_aca_sysfs(struct aca_handle *handle)
+{
+	struct amdgpu_device *adev = handle->adev;
+	struct device_attribute *aca_attr = &handle->aca_attr;
+
+	if (adev->dev->kobj.sd)
+		sysfs_remove_file_from_group(&adev->dev->kobj,
+					     &aca_attr->attr,
+					     "ras");
+}
+
+void amdgpu_aca_remove_handle(struct aca_handle *handle)
+{
+	if (!handle || list_empty(&handle->node))
+		return;
+
+	remove_aca_sysfs(handle);
+	remove_aca_handle(handle);
+}
+
+static int aca_manager_init(struct aca_handle_manager *mgr)
+{
+	INIT_LIST_HEAD(&mgr->list);
+	mgr->nr_handles = 0;
+
+	return 0;
+}
+
+static void aca_manager_fini(struct aca_handle_manager *mgr)
+{
+	struct aca_handle *handle, *tmp;
+
+	if (list_empty(&mgr->list))
+		return;
+
+	list_for_each_entry_safe(handle, tmp, &mgr->list, node)
+		amdgpu_aca_remove_handle(handle);
+}
+
+bool amdgpu_aca_is_enabled(struct amdgpu_device *adev)
+{
+	return (adev->aca.is_enabled ||
+		adev->debug_enable_ras_aca);
+}
+
+int amdgpu_aca_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	int ret;
+
+	atomic_set(&aca->ue_update_flag, 0);
+
+	ret = aca_manager_init(&aca->mgr);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+void amdgpu_aca_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+
+	aca_manager_fini(&aca->mgr);
+
+	atomic_set(&aca->ue_update_flag, 0);
+}
+
+int amdgpu_aca_reset(struct amdgpu_device *adev)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+
+	atomic_set(&aca->ue_update_flag, 0);
+
+	return 0;
+}
+
+void amdgpu_aca_set_smu_funcs(struct amdgpu_device *adev, const struct aca_smu_funcs *smu_funcs)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+
+	WARN_ON(aca->smu_funcs);
+	aca->smu_funcs = smu_funcs;
+}
+
+int aca_bank_info_decode(struct aca_bank *bank, struct aca_bank_info *info)
+{
+	u64 ipid;
+	u32 instidhi, instidlo;
+
+	if (!bank || !info)
+		return -EINVAL;
+
+	ipid = bank->regs[ACA_REG_IDX_IPID];
+	info->hwid = ACA_REG__IPID__HARDWAREID(ipid);
+	info->mcatype = ACA_REG__IPID__MCATYPE(ipid);
+	/*
+	 * Unfied DieID Format: SAASS. A:AID, S:Socket.
+	 * Unfied DieID[4:4] = InstanceId[0:0]
+	 * Unfied DieID[0:3] = InstanceIdHi[0:3]
+	 */
+	instidhi = ACA_REG__IPID__INSTANCEIDHI(ipid);
+	instidlo = ACA_REG__IPID__INSTANCEIDLO(ipid);
+	info->die_id = ((instidhi >> 2) & 0x03);
+	info->socket_id = ((instidlo & 0x1) << 2) | (instidhi & 0x03);
+
+	return 0;
+}
+
+static int aca_bank_get_error_code(struct amdgpu_device *adev, struct aca_bank *bank)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	const struct aca_smu_funcs *smu_funcs = aca->smu_funcs;
+
+	if (!smu_funcs || !smu_funcs->parse_error_code)
+		return -EOPNOTSUPP;
+
+	return smu_funcs->parse_error_code(adev, bank);
+}
+
+int aca_bank_check_error_codes(struct amdgpu_device *adev, struct aca_bank *bank, int *err_codes, int size)
+{
+	int i, error_code;
+
+	if (!bank || !err_codes)
+		return -EINVAL;
+
+	error_code = aca_bank_get_error_code(adev, bank);
+	if (error_code < 0)
+		return error_code;
+
+	for (i = 0; i < size; i++) {
+		if (err_codes[i] == error_code)
+			return 0;
+	}
+
+	return -EINVAL;
+}
+
+int amdgpu_aca_smu_set_debug_mode(struct amdgpu_device *adev, bool en)
+{
+	struct amdgpu_aca *aca = &adev->aca;
+	const struct aca_smu_funcs *smu_funcs = aca->smu_funcs;
+
+	if (!smu_funcs || !smu_funcs->set_debug_mode)
+		return -EOPNOTSUPP;
+
+	return smu_funcs->set_debug_mode(adev, en);
+}
+
+#if defined(CONFIG_DEBUG_FS)
+static int amdgpu_aca_smu_debug_mode_set(void *data, u64 val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	int ret;
+
+	ret = amdgpu_ras_set_aca_debug_mode(adev, val ? true : false);
+	if (ret)
+		return ret;
+
+	dev_info(adev->dev, "amdgpu set smu aca debug mode %s success\n", val ? "on" : "off");
+
+	return 0;
+}
+
+static void aca_dump_entry(struct seq_file *m, struct aca_bank *bank, enum aca_smu_type type, int idx)
+{
+	struct aca_bank_info info;
+	int i, ret;
+
+	ret = aca_bank_info_decode(bank, &info);
+	if (ret)
+		return;
+
+	seq_printf(m, "aca entry[%d].type: %s\n", idx, type ==  ACA_SMU_TYPE_UE ? "UE" : "CE");
+	seq_printf(m, "aca entry[%d].info: socketid:%d aid:%d hwid:0x%03x mcatype:0x%04x\n",
+		   idx, info.socket_id, info.die_id, info.hwid, info.mcatype);
+
+	for (i = 0; i < ARRAY_SIZE(aca_regs); i++)
+		seq_printf(m, "aca entry[%d].regs[%d]: 0x%016llx\n", idx, aca_regs[i].reg_idx, bank->regs[aca_regs[i].reg_idx]);
+}
+
+struct aca_dump_context {
+	struct seq_file *m;
+	int idx;
+};
+
+static int handler_aca_bank_dump(struct aca_handle *handle, struct aca_bank *bank,
+				 enum aca_smu_type type, void *data)
+{
+	struct aca_dump_context *ctx = (struct aca_dump_context *)data;
+
+	aca_dump_entry(ctx->m, bank, type, ctx->idx++);
+
+	return handler_aca_log_bank_error(handle, bank, type, NULL);
+}
+
+static int aca_dump_show(struct seq_file *m, enum aca_smu_type type)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+	struct aca_dump_context context = {
+		.m = m,
+		.idx = 0,
+	};
+
+	return aca_banks_update(adev, type, handler_aca_bank_dump, NULL, (void *)&context);
+}
+
+static int aca_dump_ce_show(struct seq_file *m, void *unused)
+{
+	return aca_dump_show(m, ACA_SMU_TYPE_CE);
+}
+
+static int aca_dump_ce_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, aca_dump_ce_show, inode->i_private);
+}
+
+static const struct file_operations aca_ce_dump_debug_fops = {
+	.owner = THIS_MODULE,
+	.open = aca_dump_ce_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = single_release,
+};
+
+static int aca_dump_ue_show(struct seq_file *m, void *unused)
+{
+	return aca_dump_show(m, ACA_SMU_TYPE_UE);
+}
+
+static int aca_dump_ue_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, aca_dump_ue_show, inode->i_private);
+}
+
+static const struct file_operations aca_ue_dump_debug_fops = {
+	.owner = THIS_MODULE,
+	.open = aca_dump_ue_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = single_release,
+};
+
+DEFINE_DEBUGFS_ATTRIBUTE(aca_debug_mode_fops, NULL, amdgpu_aca_smu_debug_mode_set, "%llu\n");
+#endif
+
+void amdgpu_aca_smu_debugfs_init(struct amdgpu_device *adev, struct dentry *root)
+{
+#if defined(CONFIG_DEBUG_FS)
+	if (!root)
+		return;
+
+	debugfs_create_file("aca_debug_mode", 0200, root, adev, &aca_debug_mode_fops);
+	debugfs_create_file("aca_ue_dump", 0400, root, adev, &aca_ue_dump_debug_fops);
+	debugfs_create_file("aca_ce_dump", 0400, root, adev, &aca_ce_dump_debug_fops);
+#endif
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_aca.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_aca.h
new file mode 100644
index 000000000000..38c88897e1ec
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_aca.h
@@ -0,0 +1,232 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_ACA_H__
+#define __AMDGPU_ACA_H__
+
+#include <linux/list.h>
+
+struct ras_err_data;
+struct ras_query_context;
+
+#define ACA_MAX_REGS_COUNT	(16)
+
+#define ACA_REG_FIELD(x, h, l)			(((x) & GENMASK_ULL(h, l)) >> l)
+#define ACA_REG__STATUS__VAL(x)			ACA_REG_FIELD(x, 63, 63)
+#define ACA_REG__STATUS__OVERFLOW(x)		ACA_REG_FIELD(x, 62, 62)
+#define ACA_REG__STATUS__UC(x)			ACA_REG_FIELD(x, 61, 61)
+#define ACA_REG__STATUS__EN(x)			ACA_REG_FIELD(x, 60, 60)
+#define ACA_REG__STATUS__MISCV(x)		ACA_REG_FIELD(x, 59, 59)
+#define ACA_REG__STATUS__ADDRV(x)		ACA_REG_FIELD(x, 58, 58)
+#define ACA_REG__STATUS__PCC(x)			ACA_REG_FIELD(x, 57, 57)
+#define ACA_REG__STATUS__ERRCOREIDVAL(x)	ACA_REG_FIELD(x, 56, 56)
+#define ACA_REG__STATUS__TCC(x)			ACA_REG_FIELD(x, 55, 55)
+#define ACA_REG__STATUS__SYNDV(x)		ACA_REG_FIELD(x, 53, 53)
+#define ACA_REG__STATUS__CECC(x)		ACA_REG_FIELD(x, 46, 46)
+#define ACA_REG__STATUS__UECC(x)		ACA_REG_FIELD(x, 45, 45)
+#define ACA_REG__STATUS__DEFERRED(x)		ACA_REG_FIELD(x, 44, 44)
+#define ACA_REG__STATUS__POISON(x)		ACA_REG_FIELD(x, 43, 43)
+#define ACA_REG__STATUS__SCRUB(x)		ACA_REG_FIELD(x, 40, 40)
+#define ACA_REG__STATUS__ERRCOREID(x)		ACA_REG_FIELD(x, 37, 32)
+#define ACA_REG__STATUS__ADDRLSB(x)		ACA_REG_FIELD(x, 29, 24)
+#define ACA_REG__STATUS__ERRORCODEEXT(x)	ACA_REG_FIELD(x, 21, 16)
+#define ACA_REG__STATUS__ERRORCODE(x)		ACA_REG_FIELD(x, 15, 0)
+
+#define ACA_REG__IPID__MCATYPE(x)		ACA_REG_FIELD(x, 63, 48)
+#define ACA_REG__IPID__INSTANCEIDHI(x)		ACA_REG_FIELD(x, 47, 44)
+#define ACA_REG__IPID__HARDWAREID(x)		ACA_REG_FIELD(x, 43, 32)
+#define ACA_REG__IPID__INSTANCEIDLO(x)		ACA_REG_FIELD(x, 31, 0)
+
+#define ACA_REG__MISC0__VALID(x)		ACA_REG_FIELD(x, 63, 63)
+#define ACA_REG__MISC0__OVRFLW(x)		ACA_REG_FIELD(x, 48, 48)
+#define ACA_REG__MISC0__ERRCNT(x)		ACA_REG_FIELD(x, 43, 32)
+
+#define ACA_REG__SYND__ERRORINFORMATION(x)	ACA_REG_FIELD(x, 17, 0)
+
+/* NOTE: The following codes refers to the smu header file */
+#define ACA_EXTERROR_CODE_CE			0x3a
+#define ACA_EXTERROR_CODE_FAULT			0x3b
+
+#define ACA_ERROR_UE_MASK		BIT_MASK(ACA_ERROR_TYPE_UE)
+#define ACA_ERROR_CE_MASK		BIT_MASK(ACA_ERROR_TYPE_CE)
+#define ACA_ERROR_DEFERRED_MASK		BIT_MASK(ACA_ERROR_TYPE_DEFERRED)
+
+#define mmSMNAID_AID0_MCA_SMU		0x03b30400	/* SMN AID AID0 */
+#define mmSMNAID_XCD0_MCA_SMU		0x36430400	/* SMN AID XCD0 */
+#define mmSMNAID_XCD1_MCA_SMU		0x38430400	/* SMN AID XCD1 */
+#define mmSMNXCD_XCD0_MCA_SMU		0x40430400	/* SMN XCD XCD0 */
+
+#define ACA_BANK_ERR_IS_DEFFERED(bank)                                \
+	(ACA_REG__STATUS__POISON((bank)->regs[ACA_REG_IDX_STATUS]) || \
+	 ACA_REG__STATUS__DEFERRED((bank)->regs[ACA_REG_IDX_STATUS]))
+
+enum aca_reg_idx {
+	ACA_REG_IDX_CTL			= 0,
+	ACA_REG_IDX_STATUS		= 1,
+	ACA_REG_IDX_ADDR		= 2,
+	ACA_REG_IDX_MISC0		= 3,
+	ACA_REG_IDX_CONFIG		= 4,
+	ACA_REG_IDX_IPID		= 5,
+	ACA_REG_IDX_SYND		= 6,
+	ACA_REG_IDX_DESTAT		= 8,
+	ACA_REG_IDX_DEADDR		= 9,
+	ACA_REG_IDX_CTL_MASK		= 10,
+	ACA_REG_IDX_COUNT		= 16,
+};
+
+enum aca_hwip_type {
+	ACA_HWIP_TYPE_UNKNOW = -1,
+	ACA_HWIP_TYPE_PSP = 0,
+	ACA_HWIP_TYPE_UMC,
+	ACA_HWIP_TYPE_SMU,
+	ACA_HWIP_TYPE_PCS_XGMI,
+	ACA_HWIP_TYPE_COUNT,
+};
+
+enum aca_error_type {
+	ACA_ERROR_TYPE_INVALID = -1,
+	ACA_ERROR_TYPE_UE = 0,
+	ACA_ERROR_TYPE_CE,
+	ACA_ERROR_TYPE_DEFERRED,
+	ACA_ERROR_TYPE_COUNT
+};
+
+enum aca_smu_type {
+	ACA_SMU_TYPE_INVALID = -1,
+	ACA_SMU_TYPE_UE = 0,
+	ACA_SMU_TYPE_CE,
+	ACA_SMU_TYPE_COUNT,
+};
+
+struct aca_hwip {
+	int hwid;
+	int mcatype;
+};
+
+struct aca_bank {
+	enum aca_error_type aca_err_type;
+	enum aca_smu_type smu_err_type;
+	u64 regs[ACA_MAX_REGS_COUNT];
+};
+
+struct aca_bank_node {
+	struct aca_bank bank;
+	struct list_head node;
+};
+
+struct aca_banks {
+	int nr_banks;
+	struct list_head list;
+};
+
+struct aca_bank_info {
+	int die_id;
+	int socket_id;
+	int hwid;
+	int mcatype;
+};
+
+struct aca_bank_error {
+	struct list_head node;
+	struct aca_bank_info info;
+	u64 count;
+};
+
+struct aca_error {
+	struct list_head list;
+	struct mutex lock;
+	enum aca_error_type type;
+	int nr_errors;
+};
+
+struct aca_handle_manager {
+	struct list_head list;
+	int nr_handles;
+};
+
+struct aca_error_cache {
+	struct aca_error errors[ACA_ERROR_TYPE_COUNT];
+};
+
+struct aca_handle {
+	struct list_head node;
+	enum aca_hwip_type hwip;
+	struct amdgpu_device *adev;
+	struct aca_handle_manager *mgr;
+	struct aca_error_cache error_cache;
+	const struct aca_bank_ops *bank_ops;
+	struct device_attribute aca_attr;
+	char attr_name[64];
+	const char *name;
+	u32 mask;
+	void *data;
+};
+
+struct aca_bank_ops {
+	int (*aca_bank_parser)(struct aca_handle *handle, struct aca_bank *bank, enum aca_smu_type type, void *data);
+	bool (*aca_bank_is_valid)(struct aca_handle *handle, struct aca_bank *bank, enum aca_smu_type type,
+				  void *data);
+};
+
+struct aca_smu_funcs {
+	int max_ue_bank_count;
+	int max_ce_bank_count;
+	int (*set_debug_mode)(struct amdgpu_device *adev, bool enable);
+	int (*get_valid_aca_count)(struct amdgpu_device *adev, enum aca_smu_type type, u32 *count);
+	int (*get_valid_aca_bank)(struct amdgpu_device *adev, enum aca_smu_type type, int idx, struct aca_bank *bank);
+	int (*parse_error_code)(struct amdgpu_device *adev, struct aca_bank *bank);
+};
+
+struct amdgpu_aca {
+	struct aca_handle_manager mgr;
+	const struct aca_smu_funcs *smu_funcs;
+	atomic_t ue_update_flag;
+	bool is_enabled;
+};
+
+struct aca_info {
+	enum aca_hwip_type hwip;
+	const struct aca_bank_ops *bank_ops;
+	u32 mask;
+};
+
+int amdgpu_aca_init(struct amdgpu_device *adev);
+void amdgpu_aca_fini(struct amdgpu_device *adev);
+int amdgpu_aca_reset(struct amdgpu_device *adev);
+void amdgpu_aca_set_smu_funcs(struct amdgpu_device *adev, const struct aca_smu_funcs *smu_funcs);
+bool amdgpu_aca_is_enabled(struct amdgpu_device *adev);
+
+int aca_bank_info_decode(struct aca_bank *bank, struct aca_bank_info *info);
+int aca_bank_check_error_codes(struct amdgpu_device *adev, struct aca_bank *bank, int *err_codes, int size);
+
+int amdgpu_aca_add_handle(struct amdgpu_device *adev, struct aca_handle *handle,
+			  const char *name, const struct aca_info *aca_info, void *data);
+void amdgpu_aca_remove_handle(struct aca_handle *handle);
+int amdgpu_aca_get_error_data(struct amdgpu_device *adev, struct aca_handle *handle,
+			      enum aca_error_type type, struct ras_err_data *err_data,
+			      struct ras_query_context *qctx);
+int amdgpu_aca_smu_set_debug_mode(struct amdgpu_device *adev, bool en);
+void amdgpu_aca_smu_debugfs_init(struct amdgpu_device *adev, struct dentry *root);
+int aca_error_cache_log_bank_error(struct aca_handle *handle, struct aca_bank_info *info,
+				   enum aca_error_type type, u64 count);
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c
index 06879d1dcabd..381ef205b0df 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c
@@ -24,10 +24,13 @@
  */
 
 #include <linux/irqdomain.h>
+#include <linux/pci.h>
 #include <linux/pm_domain.h>
 #include <linux/platform_device.h>
 #include <sound/designware_i2s.h>
 #include <sound/pcm.h>
+#include <linux/acpi.h>
+#include <linux/dmi.h>
 
 #include "amdgpu.h"
 #include "atom.h"
@@ -35,41 +38,57 @@
 
 #include "acp_gfx_if.h"
 
-#define ACP_TILE_ON_MASK                0x03
-#define ACP_TILE_OFF_MASK               0x02
-#define ACP_TILE_ON_RETAIN_REG_MASK     0x1f
-#define ACP_TILE_OFF_RETAIN_REG_MASK    0x20
-
-#define ACP_TILE_P1_MASK                0x3e
-#define ACP_TILE_P2_MASK                0x3d
-#define ACP_TILE_DSP0_MASK              0x3b
-#define ACP_TILE_DSP1_MASK              0x37
-
-#define ACP_TILE_DSP2_MASK              0x2f
-
-#define ACP_DMA_REGS_END		0x146c0
-#define ACP_I2S_PLAY_REGS_START		0x14840
-#define ACP_I2S_PLAY_REGS_END		0x148b4
-#define ACP_I2S_CAP_REGS_START		0x148b8
-#define ACP_I2S_CAP_REGS_END		0x1496c
-
-#define ACP_I2S_COMP1_CAP_REG_OFFSET	0xac
-#define ACP_I2S_COMP2_CAP_REG_OFFSET	0xa8
-#define ACP_I2S_COMP1_PLAY_REG_OFFSET	0x6c
-#define ACP_I2S_COMP2_PLAY_REG_OFFSET	0x68
-
-#define mmACP_PGFSM_RETAIN_REG		0x51c9
-#define mmACP_PGFSM_CONFIG_REG		0x51ca
-#define mmACP_PGFSM_READ_REG_0		0x51cc
-
-#define mmACP_MEM_SHUT_DOWN_REQ_LO	0x51f8
-#define mmACP_MEM_SHUT_DOWN_REQ_HI	0x51f9
-#define mmACP_MEM_SHUT_DOWN_STS_LO	0x51fa
-#define mmACP_MEM_SHUT_DOWN_STS_HI	0x51fb
-
-#define ACP_TIMEOUT_LOOP		0x000000FF
-#define ACP_DEVS			3
-#define ACP_SRC_ID			162
+#define ST_JADEITE 1
+#define ACP_TILE_ON_MASK			0x03
+#define ACP_TILE_OFF_MASK			0x02
+#define ACP_TILE_ON_RETAIN_REG_MASK		0x1f
+#define ACP_TILE_OFF_RETAIN_REG_MASK		0x20
+
+#define ACP_TILE_P1_MASK			0x3e
+#define ACP_TILE_P2_MASK			0x3d
+#define ACP_TILE_DSP0_MASK			0x3b
+#define ACP_TILE_DSP1_MASK			0x37
+
+#define ACP_TILE_DSP2_MASK			0x2f
+
+#define ACP_DMA_REGS_END			0x146c0
+#define ACP_I2S_PLAY_REGS_START			0x14840
+#define ACP_I2S_PLAY_REGS_END			0x148b4
+#define ACP_I2S_CAP_REGS_START			0x148b8
+#define ACP_I2S_CAP_REGS_END			0x1496c
+
+#define ACP_I2S_COMP1_CAP_REG_OFFSET		0xac
+#define ACP_I2S_COMP2_CAP_REG_OFFSET		0xa8
+#define ACP_I2S_COMP1_PLAY_REG_OFFSET		0x6c
+#define ACP_I2S_COMP2_PLAY_REG_OFFSET		0x68
+#define ACP_BT_PLAY_REGS_START			0x14970
+#define ACP_BT_PLAY_REGS_END			0x14a24
+#define ACP_BT_COMP1_REG_OFFSET			0xac
+#define ACP_BT_COMP2_REG_OFFSET			0xa8
+
+#define mmACP_PGFSM_RETAIN_REG			0x51c9
+#define mmACP_PGFSM_CONFIG_REG			0x51ca
+#define mmACP_PGFSM_READ_REG_0			0x51cc
+
+#define mmACP_MEM_SHUT_DOWN_REQ_LO		0x51f8
+#define mmACP_MEM_SHUT_DOWN_REQ_HI		0x51f9
+#define mmACP_MEM_SHUT_DOWN_STS_LO		0x51fa
+#define mmACP_MEM_SHUT_DOWN_STS_HI		0x51fb
+
+#define mmACP_CONTROL				0x5131
+#define mmACP_STATUS				0x5133
+#define mmACP_SOFT_RESET			0x5134
+#define ACP_CONTROL__ClkEn_MASK			0x1
+#define ACP_SOFT_RESET__SoftResetAud_MASK	0x100
+#define ACP_SOFT_RESET__SoftResetAudDone_MASK	0x1000000
+#define ACP_CLOCK_EN_TIME_OUT_VALUE		0x000000FF
+#define ACP_SOFT_RESET_DONE_TIME_OUT_VALUE	0x000000FF
+
+#define ACP_TIMEOUT_LOOP			0x000000FF
+#define ACP_DEVS				4
+#define ACP_SRC_ID				162
+
+static unsigned long acp_machine_id;
 
 enum {
 	ACP_TILE_P1 = 0,
@@ -79,9 +98,9 @@ enum {
 	ACP_TILE_DSP2,
 };
 
-static int acp_sw_init(void *handle)
+static int acp_sw_init(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	adev->acp.parent = adev->dev;
 
@@ -93,9 +112,9 @@ static int acp_sw_init(void *handle)
 	return 0;
 }
 
-static int acp_sw_fini(void *handle)
+static int acp_sw_fini(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	if (adev->acp.cgs_device)
 		amdgpu_cgs_destroy_device(adev->acp.cgs_device);
@@ -103,312 +122,441 @@ static int acp_sw_fini(void *handle)
 	return 0;
 }
 
-/* power off a tile/block within ACP */
-static int acp_suspend_tile(void *cgs_dev, int tile)
-{
-	u32 val = 0;
-	u32 count = 0;
-
-	if ((tile  < ACP_TILE_P1) || (tile > ACP_TILE_DSP2)) {
-		pr_err("Invalid ACP tile : %d to suspend\n", tile);
-		return -1;
-	}
-
-	val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0 + tile);
-	val &= ACP_TILE_ON_MASK;
-
-	if (val == 0x0) {
-		val = cgs_read_register(cgs_dev, mmACP_PGFSM_RETAIN_REG);
-		val = val | (1 << tile);
-		cgs_write_register(cgs_dev, mmACP_PGFSM_RETAIN_REG, val);
-		cgs_write_register(cgs_dev, mmACP_PGFSM_CONFIG_REG,
-					0x500 + tile);
-
-		count = ACP_TIMEOUT_LOOP;
-		while (true) {
-			val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0
-								+ tile);
-			val = val & ACP_TILE_ON_MASK;
-			if (val == ACP_TILE_OFF_MASK)
-				break;
-			if (--count == 0) {
-				pr_err("Timeout reading ACP PGFSM status\n");
-				return -ETIMEDOUT;
-			}
-			udelay(100);
-		}
-
-		val = cgs_read_register(cgs_dev, mmACP_PGFSM_RETAIN_REG);
-
-		val |= ACP_TILE_OFF_RETAIN_REG_MASK;
-		cgs_write_register(cgs_dev, mmACP_PGFSM_RETAIN_REG, val);
-	}
-	return 0;
-}
-
-/* power on a tile/block within ACP */
-static int acp_resume_tile(void *cgs_dev, int tile)
-{
-	u32 val = 0;
-	u32 count = 0;
-
-	if ((tile  < ACP_TILE_P1) || (tile > ACP_TILE_DSP2)) {
-		pr_err("Invalid ACP tile to resume\n");
-		return -1;
-	}
-
-	val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0 + tile);
-	val = val & ACP_TILE_ON_MASK;
-
-	if (val != 0x0) {
-		cgs_write_register(cgs_dev, mmACP_PGFSM_CONFIG_REG,
-					0x600 + tile);
-		count = ACP_TIMEOUT_LOOP;
-		while (true) {
-			val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0
-							+ tile);
-			val = val & ACP_TILE_ON_MASK;
-			if (val == 0x0)
-				break;
-			if (--count == 0) {
-				pr_err("Timeout reading ACP PGFSM status\n");
-				return -ETIMEDOUT;
-			}
-			udelay(100);
-		}
-		val = cgs_read_register(cgs_dev, mmACP_PGFSM_RETAIN_REG);
-		if (tile == ACP_TILE_P1)
-			val = val & (ACP_TILE_P1_MASK);
-		else if (tile == ACP_TILE_P2)
-			val = val & (ACP_TILE_P2_MASK);
-
-		cgs_write_register(cgs_dev, mmACP_PGFSM_RETAIN_REG, val);
-	}
-	return 0;
-}
-
 struct acp_pm_domain {
-	void *cgs_dev;
+	void *adev;
 	struct generic_pm_domain gpd;
 };
 
 static int acp_poweroff(struct generic_pm_domain *genpd)
 {
-	int i, ret;
 	struct acp_pm_domain *apd;
+	struct amdgpu_device *adev;
 
 	apd = container_of(genpd, struct acp_pm_domain, gpd);
-	if (apd != NULL) {
-		/* Donot return abruptly if any of power tile fails to suspend.
-		 * Log it and continue powering off other tile
-		 */
-		for (i = 4; i >= 0 ; i--) {
-			ret = acp_suspend_tile(apd->cgs_dev, ACP_TILE_P1 + i);
-			if (ret)
-				pr_err("ACP tile %d tile suspend failed\n", i);
-		}
-	}
+	adev = apd->adev;
+	/* call smu to POWER GATE ACP block
+	 * smu will
+	 * 1. turn off the acp clock
+	 * 2. power off the acp tiles
+	 * 3. check and enter ulv state
+	 */
+	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true, 0);
 	return 0;
 }
 
 static int acp_poweron(struct generic_pm_domain *genpd)
 {
-	int i, ret;
 	struct acp_pm_domain *apd;
+	struct amdgpu_device *adev;
 
 	apd = container_of(genpd, struct acp_pm_domain, gpd);
-	if (apd != NULL) {
-		for (i = 0; i < 2; i++) {
-			ret = acp_resume_tile(apd->cgs_dev, ACP_TILE_P1 + i);
-			if (ret) {
-				pr_err("ACP tile %d resume failed\n", i);
-				break;
-			}
-		}
-
-		/* Disable DSPs which are not going to be used */
-		for (i = 0; i < 3; i++) {
-			ret = acp_suspend_tile(apd->cgs_dev, ACP_TILE_DSP0 + i);
-			/* Continue suspending other DSP, even if one fails */
-			if (ret)
-				pr_err("ACP DSP %d suspend failed\n", i);
-		}
-	}
+	adev = apd->adev;
+	/* call smu to UNGATE ACP block
+	 * smu will
+	 * 1. exit ulv
+	 * 2. turn on acp clock
+	 * 3. power on acp tiles
+	 */
+	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false, 0);
 	return 0;
 }
 
-static struct device *get_mfd_cell_dev(const char *device_name, int r)
+static int acp_genpd_add_device(struct device *dev, void *data)
 {
-	char auto_dev_name[25];
-	struct device *dev;
+	struct generic_pm_domain *gpd = data;
+	int ret;
 
-	snprintf(auto_dev_name, sizeof(auto_dev_name),
-		 "%s.%d.auto", device_name, r);
-	dev = bus_find_device_by_name(&platform_bus_type, NULL, auto_dev_name);
-	dev_info(dev, "device %s added to pm domain\n", auto_dev_name);
+	ret = pm_genpd_add_device(gpd, dev);
+	if (ret)
+		dev_err(dev, "Failed to add dev to genpd %d\n", ret);
 
-	return dev;
+	return ret;
 }
 
+static int acp_genpd_remove_device(struct device *dev, void *data)
+{
+	int ret;
+
+	ret = pm_genpd_remove_device(dev);
+	if (ret)
+		dev_err(dev, "Failed to remove dev from genpd %d\n", ret);
+
+	/* Continue to remove */
+	return 0;
+}
+
+static int acp_quirk_cb(const struct dmi_system_id *id)
+{
+	acp_machine_id = ST_JADEITE;
+	return 1;
+}
+
+static const struct dmi_system_id acp_quirk_table[] = {
+	{
+		.callback = acp_quirk_cb,
+		.matches = {
+			DMI_EXACT_MATCH(DMI_BOARD_VENDOR, "AMD"),
+			DMI_EXACT_MATCH(DMI_PRODUCT_NAME, "Jadeite"),
+		}
+	},
+	{
+		.callback = acp_quirk_cb,
+		.matches = {
+			DMI_EXACT_MATCH(DMI_BOARD_VENDOR, "IP3 Technology CO.,Ltd."),
+			DMI_EXACT_MATCH(DMI_PRODUCT_NAME, "ASN1D"),
+		},
+	},
+	{
+		.callback = acp_quirk_cb,
+		.matches = {
+			DMI_EXACT_MATCH(DMI_BOARD_VENDOR, "Standard"),
+			DMI_EXACT_MATCH(DMI_PRODUCT_NAME, "ASN10"),
+		},
+	},
+	{}
+};
+
 /**
  * acp_hw_init - start and test ACP block
  *
- * @adev: amdgpu_device pointer
+ * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
  *
  */
-static int acp_hw_init(void *handle)
+static int acp_hw_init(struct amdgpu_ip_block *ip_block)
 {
-	int r, i;
-	uint64_t acp_base;
-	struct device *dev;
-	struct i2s_platform_data *i2s_pdata;
-
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	const struct amdgpu_ip_block *ip_block =
-		amdgpu_get_ip_block(adev, AMD_IP_BLOCK_TYPE_ACP);
+	int r;
+	u64 acp_base;
+	u32 val = 0;
+	u32 count = 0;
+	struct i2s_platform_data *i2s_pdata = NULL;
 
-	if (!ip_block)
-		return -EINVAL;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	r = amd_acp_hw_init(adev->acp.cgs_device,
 			    ip_block->version->major, ip_block->version->minor);
 	/* -ENODEV means board uses AZ rather than ACP */
-	if (r == -ENODEV)
+	if (r == -ENODEV) {
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true, 0);
 		return 0;
-	else if (r)
+	} else if (r) {
 		return r;
+	}
 
-	r = cgs_get_pci_resource(adev->acp.cgs_device, CGS_RESOURCE_TYPE_MMIO,
-			0x5289, 0, &acp_base);
-	if (r == -ENODEV)
-		return 0;
-	else if (r)
-		return r;
+	if (adev->rmmio_size == 0 || adev->rmmio_size < 0x5289)
+		return -EINVAL;
 
+	acp_base = adev->rmmio_base;
 	adev->acp.acp_genpd = kzalloc(sizeof(struct acp_pm_domain), GFP_KERNEL);
-	if (adev->acp.acp_genpd == NULL)
+	if (!adev->acp.acp_genpd)
 		return -ENOMEM;
 
 	adev->acp.acp_genpd->gpd.name = "ACP_AUDIO";
 	adev->acp.acp_genpd->gpd.power_off = acp_poweroff;
 	adev->acp.acp_genpd->gpd.power_on = acp_poweron;
-
-
-	adev->acp.acp_genpd->cgs_dev = adev->acp.cgs_device;
+	adev->acp.acp_genpd->adev = adev;
 
 	pm_genpd_init(&adev->acp.acp_genpd->gpd, NULL, false);
+	dmi_check_system(acp_quirk_table);
+	switch (acp_machine_id) {
+	case ST_JADEITE:
+	{
+		adev->acp.acp_cell = kcalloc(2, sizeof(struct mfd_cell),
+					     GFP_KERNEL);
+		if (!adev->acp.acp_cell) {
+			r = -ENOMEM;
+			goto failure;
+		}
 
-	adev->acp.acp_cell = kzalloc(sizeof(struct mfd_cell) * ACP_DEVS,
-							GFP_KERNEL);
-
-	if (adev->acp.acp_cell == NULL)
-		return -ENOMEM;
+		adev->acp.acp_res = kcalloc(3, sizeof(struct resource), GFP_KERNEL);
+		if (!adev->acp.acp_res) {
+			r = -ENOMEM;
+			goto failure;
+		}
 
-	adev->acp.acp_res = kzalloc(sizeof(struct resource) * 4, GFP_KERNEL);
+		i2s_pdata = kcalloc(1, sizeof(struct i2s_platform_data), GFP_KERNEL);
+		if (!i2s_pdata) {
+			r = -ENOMEM;
+			goto failure;
+		}
 
-	if (adev->acp.acp_res == NULL) {
-		kfree(adev->acp.acp_cell);
-		return -ENOMEM;
+		i2s_pdata[0].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET |
+				      DW_I2S_QUIRK_16BIT_IDX_OVERRIDE;
+		i2s_pdata[0].cap = DWC_I2S_PLAY | DWC_I2S_RECORD;
+		i2s_pdata[0].snd_rates = SNDRV_PCM_RATE_8000_96000;
+		i2s_pdata[0].i2s_reg_comp1 = ACP_I2S_COMP1_CAP_REG_OFFSET;
+		i2s_pdata[0].i2s_reg_comp2 = ACP_I2S_COMP2_CAP_REG_OFFSET;
+
+		adev->acp.acp_res[0].name = "acp2x_dma";
+		adev->acp.acp_res[0].flags = IORESOURCE_MEM;
+		adev->acp.acp_res[0].start = acp_base;
+		adev->acp.acp_res[0].end = acp_base + ACP_DMA_REGS_END;
+
+		adev->acp.acp_res[1].name = "acp2x_dw_i2s_play_cap";
+		adev->acp.acp_res[1].flags = IORESOURCE_MEM;
+		adev->acp.acp_res[1].start = acp_base + ACP_I2S_CAP_REGS_START;
+		adev->acp.acp_res[1].end = acp_base + ACP_I2S_CAP_REGS_END;
+
+		adev->acp.acp_res[2].name = "acp2x_dma_irq";
+		adev->acp.acp_res[2].flags = IORESOURCE_IRQ;
+		adev->acp.acp_res[2].start = amdgpu_irq_create_mapping(adev, 162);
+		adev->acp.acp_res[2].end = adev->acp.acp_res[2].start;
+
+		adev->acp.acp_cell[0].name = "acp_audio_dma";
+		adev->acp.acp_cell[0].id = 0;
+		adev->acp.acp_cell[0].num_resources = 3;
+		adev->acp.acp_cell[0].resources = &adev->acp.acp_res[0];
+		adev->acp.acp_cell[0].platform_data = &adev->asic_type;
+		adev->acp.acp_cell[0].pdata_size = sizeof(adev->asic_type);
+
+		adev->acp.acp_cell[1].name = "designware-i2s";
+		adev->acp.acp_cell[1].id = 1;
+		adev->acp.acp_cell[1].num_resources = 1;
+		adev->acp.acp_cell[1].resources = &adev->acp.acp_res[1];
+		adev->acp.acp_cell[1].platform_data = &i2s_pdata[0];
+		adev->acp.acp_cell[1].pdata_size = sizeof(struct i2s_platform_data);
+		r = mfd_add_devices(adev->acp.parent, 0, adev->acp.acp_cell, 2, NULL, 0, NULL);
+		if (r)
+			goto failure;
+		r = device_for_each_child(adev->acp.parent, &adev->acp.acp_genpd->gpd,
+					  acp_genpd_add_device);
+		if (r)
+			goto failure;
+		break;
 	}
+	default:
+		adev->acp.acp_cell = kcalloc(ACP_DEVS, sizeof(struct mfd_cell),
+					     GFP_KERNEL);
 
-	i2s_pdata = kzalloc(sizeof(struct i2s_platform_data) * 2, GFP_KERNEL);
-	if (i2s_pdata == NULL) {
-		kfree(adev->acp.acp_res);
-		kfree(adev->acp.acp_cell);
-		return -ENOMEM;
-	}
+		if (!adev->acp.acp_cell) {
+			r = -ENOMEM;
+			goto failure;
+		}
 
-	i2s_pdata[0].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET;
-	i2s_pdata[0].cap = DWC_I2S_PLAY;
-	i2s_pdata[0].snd_rates = SNDRV_PCM_RATE_8000_96000;
-	i2s_pdata[0].i2s_reg_comp1 = ACP_I2S_COMP1_PLAY_REG_OFFSET;
-	i2s_pdata[0].i2s_reg_comp2 = ACP_I2S_COMP2_PLAY_REG_OFFSET;
+		adev->acp.acp_res = kcalloc(5, sizeof(struct resource), GFP_KERNEL);
+		if (!adev->acp.acp_res) {
+			r = -ENOMEM;
+			goto failure;
+		}
 
-	i2s_pdata[1].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET |
+		i2s_pdata = kcalloc(3, sizeof(struct i2s_platform_data), GFP_KERNEL);
+		if (!i2s_pdata) {
+			r = -ENOMEM;
+			goto failure;
+		}
+
+		switch (adev->asic_type) {
+		case CHIP_STONEY:
+			i2s_pdata[0].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET |
+				DW_I2S_QUIRK_16BIT_IDX_OVERRIDE;
+			break;
+		default:
+			i2s_pdata[0].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET;
+		}
+		i2s_pdata[0].cap = DWC_I2S_PLAY;
+		i2s_pdata[0].snd_rates = SNDRV_PCM_RATE_8000_96000;
+		i2s_pdata[0].i2s_reg_comp1 = ACP_I2S_COMP1_PLAY_REG_OFFSET;
+		i2s_pdata[0].i2s_reg_comp2 = ACP_I2S_COMP2_PLAY_REG_OFFSET;
+		switch (adev->asic_type) {
+		case CHIP_STONEY:
+			i2s_pdata[1].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET |
+				DW_I2S_QUIRK_COMP_PARAM1 |
+				DW_I2S_QUIRK_16BIT_IDX_OVERRIDE;
+			break;
+		default:
+			i2s_pdata[1].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET |
 				DW_I2S_QUIRK_COMP_PARAM1;
-	i2s_pdata[1].cap = DWC_I2S_RECORD;
-	i2s_pdata[1].snd_rates = SNDRV_PCM_RATE_8000_96000;
-	i2s_pdata[1].i2s_reg_comp1 = ACP_I2S_COMP1_CAP_REG_OFFSET;
-	i2s_pdata[1].i2s_reg_comp2 = ACP_I2S_COMP2_CAP_REG_OFFSET;
-
-	adev->acp.acp_res[0].name = "acp2x_dma";
-	adev->acp.acp_res[0].flags = IORESOURCE_MEM;
-	adev->acp.acp_res[0].start = acp_base;
-	adev->acp.acp_res[0].end = acp_base + ACP_DMA_REGS_END;
-
-	adev->acp.acp_res[1].name = "acp2x_dw_i2s_play";
-	adev->acp.acp_res[1].flags = IORESOURCE_MEM;
-	adev->acp.acp_res[1].start = acp_base + ACP_I2S_PLAY_REGS_START;
-	adev->acp.acp_res[1].end = acp_base + ACP_I2S_PLAY_REGS_END;
-
-	adev->acp.acp_res[2].name = "acp2x_dw_i2s_cap";
-	adev->acp.acp_res[2].flags = IORESOURCE_MEM;
-	adev->acp.acp_res[2].start = acp_base + ACP_I2S_CAP_REGS_START;
-	adev->acp.acp_res[2].end = acp_base + ACP_I2S_CAP_REGS_END;
-
-	adev->acp.acp_res[3].name = "acp2x_dma_irq";
-	adev->acp.acp_res[3].flags = IORESOURCE_IRQ;
-	adev->acp.acp_res[3].start = amdgpu_irq_create_mapping(adev, 162);
-	adev->acp.acp_res[3].end = adev->acp.acp_res[3].start;
-
-	adev->acp.acp_cell[0].name = "acp_audio_dma";
-	adev->acp.acp_cell[0].num_resources = 4;
-	adev->acp.acp_cell[0].resources = &adev->acp.acp_res[0];
-
-	adev->acp.acp_cell[1].name = "designware-i2s";
-	adev->acp.acp_cell[1].num_resources = 1;
-	adev->acp.acp_cell[1].resources = &adev->acp.acp_res[1];
-	adev->acp.acp_cell[1].platform_data = &i2s_pdata[0];
-	adev->acp.acp_cell[1].pdata_size = sizeof(struct i2s_platform_data);
-
-	adev->acp.acp_cell[2].name = "designware-i2s";
-	adev->acp.acp_cell[2].num_resources = 1;
-	adev->acp.acp_cell[2].resources = &adev->acp.acp_res[2];
-	adev->acp.acp_cell[2].platform_data = &i2s_pdata[1];
-	adev->acp.acp_cell[2].pdata_size = sizeof(struct i2s_platform_data);
-
-	r = mfd_add_hotplug_devices(adev->acp.parent, adev->acp.acp_cell,
-								ACP_DEVS);
-	if (r)
-		return r;
+		}
 
-	for (i = 0; i < ACP_DEVS ; i++) {
-		dev = get_mfd_cell_dev(adev->acp.acp_cell[i].name, i);
-		r = pm_genpd_add_device(&adev->acp.acp_genpd->gpd, dev);
-		if (r) {
-			dev_err(dev, "Failed to add dev to genpd\n");
-			return r;
+		i2s_pdata[1].cap = DWC_I2S_RECORD;
+		i2s_pdata[1].snd_rates = SNDRV_PCM_RATE_8000_96000;
+		i2s_pdata[1].i2s_reg_comp1 = ACP_I2S_COMP1_CAP_REG_OFFSET;
+		i2s_pdata[1].i2s_reg_comp2 = ACP_I2S_COMP2_CAP_REG_OFFSET;
+
+		i2s_pdata[2].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET;
+		switch (adev->asic_type) {
+		case CHIP_STONEY:
+			i2s_pdata[2].quirks |= DW_I2S_QUIRK_16BIT_IDX_OVERRIDE;
+			break;
+		default:
+			break;
 		}
+
+		i2s_pdata[2].cap = DWC_I2S_PLAY | DWC_I2S_RECORD;
+		i2s_pdata[2].snd_rates = SNDRV_PCM_RATE_8000_96000;
+		i2s_pdata[2].i2s_reg_comp1 = ACP_BT_COMP1_REG_OFFSET;
+		i2s_pdata[2].i2s_reg_comp2 = ACP_BT_COMP2_REG_OFFSET;
+
+		adev->acp.acp_res[0].name = "acp2x_dma";
+		adev->acp.acp_res[0].flags = IORESOURCE_MEM;
+		adev->acp.acp_res[0].start = acp_base;
+		adev->acp.acp_res[0].end = acp_base + ACP_DMA_REGS_END;
+
+		adev->acp.acp_res[1].name = "acp2x_dw_i2s_play";
+		adev->acp.acp_res[1].flags = IORESOURCE_MEM;
+		adev->acp.acp_res[1].start = acp_base + ACP_I2S_PLAY_REGS_START;
+		adev->acp.acp_res[1].end = acp_base + ACP_I2S_PLAY_REGS_END;
+
+		adev->acp.acp_res[2].name = "acp2x_dw_i2s_cap";
+		adev->acp.acp_res[2].flags = IORESOURCE_MEM;
+		adev->acp.acp_res[2].start = acp_base + ACP_I2S_CAP_REGS_START;
+		adev->acp.acp_res[2].end = acp_base + ACP_I2S_CAP_REGS_END;
+
+		adev->acp.acp_res[3].name = "acp2x_dw_bt_i2s_play_cap";
+		adev->acp.acp_res[3].flags = IORESOURCE_MEM;
+		adev->acp.acp_res[3].start = acp_base + ACP_BT_PLAY_REGS_START;
+		adev->acp.acp_res[3].end = acp_base + ACP_BT_PLAY_REGS_END;
+
+		adev->acp.acp_res[4].name = "acp2x_dma_irq";
+		adev->acp.acp_res[4].flags = IORESOURCE_IRQ;
+		adev->acp.acp_res[4].start = amdgpu_irq_create_mapping(adev, 162);
+		adev->acp.acp_res[4].end = adev->acp.acp_res[4].start;
+
+		adev->acp.acp_cell[0].name = "acp_audio_dma";
+		adev->acp.acp_cell[0].id = 0;
+		adev->acp.acp_cell[0].num_resources = 5;
+		adev->acp.acp_cell[0].resources = &adev->acp.acp_res[0];
+		adev->acp.acp_cell[0].platform_data = &adev->asic_type;
+		adev->acp.acp_cell[0].pdata_size = sizeof(adev->asic_type);
+
+		adev->acp.acp_cell[1].name = "designware-i2s";
+		adev->acp.acp_cell[1].id = 1;
+		adev->acp.acp_cell[1].num_resources = 1;
+		adev->acp.acp_cell[1].resources = &adev->acp.acp_res[1];
+		adev->acp.acp_cell[1].platform_data = &i2s_pdata[0];
+		adev->acp.acp_cell[1].pdata_size = sizeof(struct i2s_platform_data);
+
+		adev->acp.acp_cell[2].name = "designware-i2s";
+		adev->acp.acp_cell[2].id = 2;
+		adev->acp.acp_cell[2].num_resources = 1;
+		adev->acp.acp_cell[2].resources = &adev->acp.acp_res[2];
+		adev->acp.acp_cell[2].platform_data = &i2s_pdata[1];
+		adev->acp.acp_cell[2].pdata_size = sizeof(struct i2s_platform_data);
+
+		adev->acp.acp_cell[3].name = "designware-i2s";
+		adev->acp.acp_cell[3].id = 3;
+		adev->acp.acp_cell[3].num_resources = 1;
+		adev->acp.acp_cell[3].resources = &adev->acp.acp_res[3];
+		adev->acp.acp_cell[3].platform_data = &i2s_pdata[2];
+		adev->acp.acp_cell[3].pdata_size = sizeof(struct i2s_platform_data);
+
+		r = mfd_add_devices(adev->acp.parent, 0, adev->acp.acp_cell, ACP_DEVS, NULL, 0, NULL);
+		if (r)
+			goto failure;
+
+		r = device_for_each_child(adev->acp.parent, &adev->acp.acp_genpd->gpd,
+					  acp_genpd_add_device);
+		if (r)
+			goto failure;
 	}
 
+	/* Assert Soft reset of ACP */
+	val = cgs_read_register(adev->acp.cgs_device, mmACP_SOFT_RESET);
+
+	val |= ACP_SOFT_RESET__SoftResetAud_MASK;
+	cgs_write_register(adev->acp.cgs_device, mmACP_SOFT_RESET, val);
+
+	count = ACP_SOFT_RESET_DONE_TIME_OUT_VALUE;
+	while (true) {
+		val = cgs_read_register(adev->acp.cgs_device, mmACP_SOFT_RESET);
+		if (ACP_SOFT_RESET__SoftResetAudDone_MASK ==
+		    (val & ACP_SOFT_RESET__SoftResetAudDone_MASK))
+			break;
+		if (--count == 0) {
+			dev_err(&adev->pdev->dev, "Failed to reset ACP\n");
+			r = -ETIMEDOUT;
+			goto failure;
+		}
+		udelay(100);
+	}
+	/* Enable clock to ACP and wait until the clock is enabled */
+	val = cgs_read_register(adev->acp.cgs_device, mmACP_CONTROL);
+	val = val | ACP_CONTROL__ClkEn_MASK;
+	cgs_write_register(adev->acp.cgs_device, mmACP_CONTROL, val);
+
+	count = ACP_CLOCK_EN_TIME_OUT_VALUE;
+
+	while (true) {
+		val = cgs_read_register(adev->acp.cgs_device, mmACP_STATUS);
+		if (val & (u32) 0x1)
+			break;
+		if (--count == 0) {
+			dev_err(&adev->pdev->dev, "Failed to reset ACP\n");
+			r = -ETIMEDOUT;
+			goto failure;
+		}
+		udelay(100);
+	}
+	/* Deassert the SOFT RESET flags */
+	val = cgs_read_register(adev->acp.cgs_device, mmACP_SOFT_RESET);
+	val &= ~ACP_SOFT_RESET__SoftResetAud_MASK;
+	cgs_write_register(adev->acp.cgs_device, mmACP_SOFT_RESET, val);
 	return 0;
+
+failure:
+	kfree(i2s_pdata);
+	kfree(adev->acp.acp_res);
+	kfree(adev->acp.acp_cell);
+	kfree(adev->acp.acp_genpd);
+	return r;
 }
 
 /**
  * acp_hw_fini - stop the hardware block
  *
- * @adev: amdgpu_device pointer
+ * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
  *
  */
-static int acp_hw_fini(void *handle)
+static int acp_hw_fini(struct amdgpu_ip_block *ip_block)
 {
-	int i, ret;
-	struct device *dev;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	u32 val = 0;
+	u32 count = 0;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	/* return early if no ACP */
-	if (!adev->acp.acp_genpd)
+	if (!adev->acp.acp_genpd) {
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false, 0);
 		return 0;
+	}
 
-	for (i = 0; i < ACP_DEVS ; i++) {
-		dev = get_mfd_cell_dev(adev->acp.acp_cell[i].name, i);
-		ret = pm_genpd_remove_device(&adev->acp.acp_genpd->gpd, dev);
-		/* If removal fails, dont giveup and try rest */
-		if (ret)
-			dev_err(dev, "remove dev from genpd failed\n");
+	/* Assert Soft reset of ACP */
+	val = cgs_read_register(adev->acp.cgs_device, mmACP_SOFT_RESET);
+
+	val |= ACP_SOFT_RESET__SoftResetAud_MASK;
+	cgs_write_register(adev->acp.cgs_device, mmACP_SOFT_RESET, val);
+
+	count = ACP_SOFT_RESET_DONE_TIME_OUT_VALUE;
+	while (true) {
+		val = cgs_read_register(adev->acp.cgs_device, mmACP_SOFT_RESET);
+		if (ACP_SOFT_RESET__SoftResetAudDone_MASK ==
+		    (val & ACP_SOFT_RESET__SoftResetAudDone_MASK))
+			break;
+		if (--count == 0) {
+			dev_err(&adev->pdev->dev, "Failed to reset ACP\n");
+			return -ETIMEDOUT;
+		}
+		udelay(100);
+	}
+	/* Disable ACP clock */
+	val = cgs_read_register(adev->acp.cgs_device, mmACP_CONTROL);
+	val &= ~ACP_CONTROL__ClkEn_MASK;
+	cgs_write_register(adev->acp.cgs_device, mmACP_CONTROL, val);
+
+	count = ACP_CLOCK_EN_TIME_OUT_VALUE;
+
+	while (true) {
+		val = cgs_read_register(adev->acp.cgs_device, mmACP_STATUS);
+		if (val & (u32) 0x1)
+			break;
+		if (--count == 0) {
+			dev_err(&adev->pdev->dev, "Failed to reset ACP\n");
+			return -ETIMEDOUT;
+		}
+		udelay(100);
 	}
 
+	device_for_each_child(adev->acp.parent, NULL,
+			      acp_genpd_remove_device);
+
 	mfd_remove_devices(adev->acp.parent);
 	kfree(adev->acp.acp_res);
 	kfree(adev->acp.acp_genpd);
@@ -417,52 +565,50 @@ static int acp_hw_fini(void *handle)
 	return 0;
 }
 
-static int acp_suspend(void *handle)
+static int acp_suspend(struct amdgpu_ip_block *ip_block)
 {
-	return 0;
-}
+	struct amdgpu_device *adev = ip_block->adev;
 
-static int acp_resume(void *handle)
-{
+	/* power up on suspend */
+	if (!adev->acp.acp_cell)
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false, 0);
 	return 0;
 }
 
-static int acp_early_init(void *handle)
+static int acp_resume(struct amdgpu_ip_block *ip_block)
 {
-	return 0;
-}
-
-static bool acp_is_idle(void *handle)
-{
-	return true;
-}
+	struct amdgpu_device *adev = ip_block->adev;
 
-static int acp_wait_for_idle(void *handle)
-{
+	/* power down again on resume */
+	if (!adev->acp.acp_cell)
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true, 0);
 	return 0;
 }
 
-static int acp_soft_reset(void *handle)
+static bool acp_is_idle(struct amdgpu_ip_block *ip_block)
 {
-	return 0;
+	return true;
 }
 
-static int acp_set_clockgating_state(void *handle,
+static int acp_set_clockgating_state(struct amdgpu_ip_block *ip_block,
 				     enum amd_clockgating_state state)
 {
 	return 0;
 }
 
-static int acp_set_powergating_state(void *handle,
+static int acp_set_powergating_state(struct amdgpu_ip_block *ip_block,
 				     enum amd_powergating_state state)
 {
+	struct amdgpu_device *adev = ip_block->adev;
+	bool enable = (state == AMD_PG_STATE_GATE);
+
+	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, enable, 0);
+
 	return 0;
 }
 
 static const struct amd_ip_funcs acp_ip_funcs = {
 	.name = "acp_ip",
-	.early_init = acp_early_init,
-	.late_init = NULL,
 	.sw_init = acp_sw_init,
 	.sw_fini = acp_sw_fini,
 	.hw_init = acp_hw_init,
@@ -470,14 +616,11 @@ static const struct amd_ip_funcs acp_ip_funcs = {
 	.suspend = acp_suspend,
 	.resume = acp_resume,
 	.is_idle = acp_is_idle,
-	.wait_for_idle = acp_wait_for_idle,
-	.soft_reset = acp_soft_reset,
 	.set_clockgating_state = acp_set_clockgating_state,
 	.set_powergating_state = acp_set_powergating_state,
 };
 
-const struct amdgpu_ip_block_version acp_ip_block =
-{
+const struct amdgpu_ip_block_version acp_ip_block = {
 	.type = AMD_IP_BLOCK_TYPE_ACP,
 	.major = 2,
 	.minor = 2,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
index ef79551b4cb7..d31460a9e958 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: MIT
 /*
  * Copyright 2012 Advanced Micro Devices, Inc.
  *
@@ -23,33 +24,130 @@
 
 #include <linux/pci.h>
 #include <linux/acpi.h>
+#include <linux/backlight.h>
 #include <linux/slab.h>
+#include <linux/xarray.h>
 #include <linux/power_supply.h>
 #include <linux/pm_runtime.h>
+#include <linux/suspend.h>
 #include <acpi/video.h>
-#include <drm/drmP.h>
-#include <drm/drm_crtc_helper.h>
+#include <acpi/actbl.h>
+
 #include "amdgpu.h"
+#include "amdgpu_pm.h"
+#include "amdgpu_display.h"
 #include "amd_acpi.h"
 #include "atom.h"
 
-extern void amdgpu_pm_acpi_event_handler(struct amdgpu_device *adev);
+/* Declare GUID for AMD _DSM method for XCCs */
+static const guid_t amd_xcc_dsm_guid = GUID_INIT(0x8267f5d5, 0xa556, 0x44f2,
+						 0xb8, 0xb4, 0x45, 0x56, 0x2e,
+						 0x8c, 0x5b, 0xec);
+
+#define AMD_XCC_HID_START 3000
+#define AMD_XCC_DSM_GET_NUM_FUNCS 0
+#define AMD_XCC_DSM_GET_SUPP_MODE 1
+#define AMD_XCC_DSM_GET_XCP_MODE 2
+#define AMD_XCC_DSM_GET_VF_XCC_MAPPING 4
+#define AMD_XCC_DSM_GET_TMR_INFO 5
+#define AMD_XCC_DSM_NUM_FUNCS 5
+
+#define AMD_XCC_MAX_HID 24
+
+struct xarray numa_info_xa;
+
+/* Encapsulates the XCD acpi object information */
+struct amdgpu_acpi_xcc_info {
+	struct list_head list;
+	struct amdgpu_numa_info *numa_info;
+	uint8_t xcp_node;
+	uint8_t phy_id;
+	acpi_handle handle;
+};
+
+struct amdgpu_acpi_dev_info {
+	struct list_head list;
+	struct list_head xcc_list;
+	uint32_t sbdf;
+	uint16_t supp_xcp_mode;
+	uint16_t xcp_mode;
+	uint16_t mem_mode;
+	uint64_t tmr_base;
+	uint64_t tmr_size;
+};
+
+struct list_head amdgpu_acpi_dev_list;
+
+struct amdgpu_atif_notification_cfg {
+	bool enabled;
+	int command_code;
+};
+
+struct amdgpu_atif_notifications {
+	bool thermal_state;
+	bool forced_power_state;
+	bool system_power_state;
+	bool brightness_change;
+	bool dgpu_display_event;
+	bool gpu_package_power_limit;
+};
+
+struct amdgpu_atif_functions {
+	bool system_params;
+	bool sbios_requests;
+	bool temperature_change;
+	bool query_backlight_transfer_characteristics;
+	bool ready_to_undock;
+	bool external_gpu_information;
+};
+
+struct amdgpu_atif {
+	acpi_handle handle;
+
+	struct amdgpu_atif_notifications notifications;
+	struct amdgpu_atif_functions functions;
+	struct amdgpu_atif_notification_cfg notification_cfg;
+	struct backlight_device *bd;
+	struct amdgpu_dm_backlight_caps backlight_caps;
+};
+
+struct amdgpu_atcs_functions {
+	bool get_ext_state;
+	bool pcie_perf_req;
+	bool pcie_dev_rdy;
+	bool pcie_bus_width;
+	bool power_shift_control;
+};
+
+struct amdgpu_atcs {
+	acpi_handle handle;
+
+	struct amdgpu_atcs_functions functions;
+};
+
+static struct amdgpu_acpi_priv {
+	struct amdgpu_atif atif;
+	struct amdgpu_atcs atcs;
+} amdgpu_acpi_priv;
+
 /* Call the ATIF method
  */
 /**
  * amdgpu_atif_call - call an ATIF method
  *
- * @handle: acpi handle
+ * @atif: atif structure
  * @function: the ATIF function to execute
  * @params: ATIF function params
  *
  * Executes the requested ATIF function (all asics).
  * Returns a pointer to the acpi output buffer.
  */
-static union acpi_object *amdgpu_atif_call(acpi_handle handle, int function,
-		struct acpi_buffer *params)
+static union acpi_object *amdgpu_atif_call(struct amdgpu_atif *atif,
+					   int function,
+					   struct acpi_buffer *params)
 {
 	acpi_status status;
+	union acpi_object *obj;
 	union acpi_object atif_arg_elements[2];
 	struct acpi_object_list atif_arg;
 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
@@ -70,17 +168,26 @@ static union acpi_object *amdgpu_atif_call(acpi_handle handle, int function,
 		atif_arg_elements[1].integer.value = 0;
 	}
 
-	status = acpi_evaluate_object(handle, "ATIF", &atif_arg, &buffer);
+	status = acpi_evaluate_object(atif->handle, NULL, &atif_arg,
+				      &buffer);
+	obj = (union acpi_object *)buffer.pointer;
 
-	/* Fail only if calling the method fails and ATIF is supported */
-	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
+	/* Fail if calling the method fails */
+	if (ACPI_FAILURE(status)) {
 		DRM_DEBUG_DRIVER("failed to evaluate ATIF got %s\n",
 				 acpi_format_exception(status));
-		kfree(buffer.pointer);
+		kfree(obj);
 		return NULL;
 	}
 
-	return buffer.pointer;
+	if (obj->type != ACPI_TYPE_BUFFER) {
+		DRM_DEBUG_DRIVER("bad object returned from ATIF: %d\n",
+				 obj->type);
+		kfree(obj);
+		return NULL;
+	}
+
+	return obj;
 }
 
 /**
@@ -95,15 +202,12 @@ static union acpi_object *amdgpu_atif_call(acpi_handle handle, int function,
  */
 static void amdgpu_atif_parse_notification(struct amdgpu_atif_notifications *n, u32 mask)
 {
-	n->display_switch = mask & ATIF_DISPLAY_SWITCH_REQUEST_SUPPORTED;
-	n->expansion_mode_change = mask & ATIF_EXPANSION_MODE_CHANGE_REQUEST_SUPPORTED;
 	n->thermal_state = mask & ATIF_THERMAL_STATE_CHANGE_REQUEST_SUPPORTED;
 	n->forced_power_state = mask & ATIF_FORCED_POWER_STATE_CHANGE_REQUEST_SUPPORTED;
 	n->system_power_state = mask & ATIF_SYSTEM_POWER_SOURCE_CHANGE_REQUEST_SUPPORTED;
-	n->display_conf_change = mask & ATIF_DISPLAY_CONF_CHANGE_REQUEST_SUPPORTED;
-	n->px_gfx_switch = mask & ATIF_PX_GFX_SWITCH_REQUEST_SUPPORTED;
 	n->brightness_change = mask & ATIF_PANEL_BRIGHTNESS_CHANGE_REQUEST_SUPPORTED;
 	n->dgpu_display_event = mask & ATIF_DGPU_DISPLAY_EVENT_SUPPORTED;
+	n->gpu_package_power_limit = mask & ATIF_GPU_PACKAGE_POWER_LIMIT_REQUEST_SUPPORTED;
 }
 
 /**
@@ -120,20 +224,16 @@ static void amdgpu_atif_parse_functions(struct amdgpu_atif_functions *f, u32 mas
 {
 	f->system_params = mask & ATIF_GET_SYSTEM_PARAMETERS_SUPPORTED;
 	f->sbios_requests = mask & ATIF_GET_SYSTEM_BIOS_REQUESTS_SUPPORTED;
-	f->select_active_disp = mask & ATIF_SELECT_ACTIVE_DISPLAYS_SUPPORTED;
-	f->lid_state = mask & ATIF_GET_LID_STATE_SUPPORTED;
-	f->get_tv_standard = mask & ATIF_GET_TV_STANDARD_FROM_CMOS_SUPPORTED;
-	f->set_tv_standard = mask & ATIF_SET_TV_STANDARD_IN_CMOS_SUPPORTED;
-	f->get_panel_expansion_mode = mask & ATIF_GET_PANEL_EXPANSION_MODE_FROM_CMOS_SUPPORTED;
-	f->set_panel_expansion_mode = mask & ATIF_SET_PANEL_EXPANSION_MODE_IN_CMOS_SUPPORTED;
 	f->temperature_change = mask & ATIF_TEMPERATURE_CHANGE_NOTIFICATION_SUPPORTED;
-	f->graphics_device_types = mask & ATIF_GET_GRAPHICS_DEVICE_TYPES_SUPPORTED;
+	f->query_backlight_transfer_characteristics =
+		mask & ATIF_QUERY_BACKLIGHT_TRANSFER_CHARACTERISTICS_SUPPORTED;
+	f->ready_to_undock = mask & ATIF_READY_TO_UNDOCK_NOTIFICATION_SUPPORTED;
+	f->external_gpu_information = mask & ATIF_GET_EXTERNAL_GPU_INFORMATION_SUPPORTED;
 }
 
 /**
  * amdgpu_atif_verify_interface - verify ATIF
  *
- * @handle: acpi handle
  * @atif: amdgpu atif struct
  *
  * Execute the ATIF_FUNCTION_VERIFY_INTERFACE ATIF function
@@ -141,15 +241,14 @@ static void amdgpu_atif_parse_functions(struct amdgpu_atif_functions *f, u32 mas
  * (all asics).
  * returns 0 on success, error on failure.
  */
-static int amdgpu_atif_verify_interface(acpi_handle handle,
-		struct amdgpu_atif *atif)
+static int amdgpu_atif_verify_interface(struct amdgpu_atif *atif)
 {
 	union acpi_object *info;
 	struct atif_verify_interface output;
 	size_t size;
 	int err = 0;
 
-	info = amdgpu_atif_call(handle, ATIF_FUNCTION_VERIFY_INTERFACE, NULL);
+	info = amdgpu_atif_call(atif, ATIF_FUNCTION_VERIFY_INTERFACE, NULL);
 	if (!info)
 		return -EIO;
 
@@ -179,8 +278,7 @@ out:
 /**
  * amdgpu_atif_get_notification_params - determine notify configuration
  *
- * @handle: acpi handle
- * @n: atif notification configuration struct
+ * @atif: acpi handle
  *
  * Execute the ATIF_FUNCTION_GET_SYSTEM_PARAMETERS ATIF function
  * to determine if a notifier is used and if so which one
@@ -188,15 +286,16 @@ out:
  * where n is specified in the result if a notifier is used.
  * Returns 0 on success, error on failure.
  */
-static int amdgpu_atif_get_notification_params(acpi_handle handle,
-		struct amdgpu_atif_notification_cfg *n)
+static int amdgpu_atif_get_notification_params(struct amdgpu_atif *atif)
 {
 	union acpi_object *info;
+	struct amdgpu_atif_notification_cfg *n = &atif->notification_cfg;
 	struct atif_system_params params;
 	size_t size;
 	int err = 0;
 
-	info = amdgpu_atif_call(handle, ATIF_FUNCTION_GET_SYSTEM_PARAMETERS, NULL);
+	info = amdgpu_atif_call(atif, ATIF_FUNCTION_GET_SYSTEM_PARAMETERS,
+				NULL);
 	if (!info) {
 		err = -EIO;
 		goto out;
@@ -240,9 +339,74 @@ out:
 }
 
 /**
+ * amdgpu_atif_query_backlight_caps - get min and max backlight input signal
+ *
+ * @atif: acpi handle
+ *
+ * Execute the QUERY_BRIGHTNESS_TRANSFER_CHARACTERISTICS ATIF function
+ * to determine the acceptable range of backlight values
+ *
+ * Backlight_caps.caps_valid will be set to true if the query is successful
+ *
+ * The input signals are in range 0-255
+ *
+ * This function assumes the display with backlight is the first LCD
+ *
+ * Returns 0 on success, error on failure.
+ */
+static int amdgpu_atif_query_backlight_caps(struct amdgpu_atif *atif)
+{
+	union acpi_object *info;
+	struct atif_qbtc_output characteristics;
+	struct atif_qbtc_arguments arguments;
+	struct acpi_buffer params;
+	size_t size;
+	int err = 0;
+
+	arguments.size = sizeof(arguments);
+	arguments.requested_display = ATIF_QBTC_REQUEST_LCD1;
+
+	params.length = sizeof(arguments);
+	params.pointer = (void *)&arguments;
+
+	info = amdgpu_atif_call(atif,
+		ATIF_FUNCTION_QUERY_BRIGHTNESS_TRANSFER_CHARACTERISTICS,
+		&params);
+	if (!info) {
+		err = -EIO;
+		goto out;
+	}
+
+	size = *(u16 *) info->buffer.pointer;
+	if (size < 10) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	memset(&characteristics, 0, sizeof(characteristics));
+	size = min(sizeof(characteristics), size);
+	memcpy(&characteristics, info->buffer.pointer, size);
+
+	atif->backlight_caps.caps_valid = true;
+	atif->backlight_caps.min_input_signal =
+			characteristics.min_input_signal;
+	atif->backlight_caps.max_input_signal =
+			characteristics.max_input_signal;
+	atif->backlight_caps.ac_level = characteristics.ac_level;
+	atif->backlight_caps.dc_level = characteristics.dc_level;
+	atif->backlight_caps.data_points = characteristics.number_of_points;
+	memcpy(atif->backlight_caps.luminance_data,
+	       characteristics.data_points,
+	       sizeof(atif->backlight_caps.luminance_data));
+out:
+	kfree(info);
+	return err;
+}
+
+/**
  * amdgpu_atif_get_sbios_requests - get requested sbios event
  *
- * @handle: acpi handle
+ * @atif: acpi handle
  * @req: atif sbios request struct
  *
  * Execute the ATIF_FUNCTION_GET_SYSTEM_BIOS_REQUESTS ATIF function
@@ -250,14 +414,15 @@ out:
  * (all asics).
  * Returns 0 on success, error on failure.
  */
-static int amdgpu_atif_get_sbios_requests(acpi_handle handle,
-		struct atif_sbios_requests *req)
+static int amdgpu_atif_get_sbios_requests(struct amdgpu_atif *atif,
+					  struct atif_sbios_requests *req)
 {
 	union acpi_object *info;
 	size_t size;
 	int count = 0;
 
-	info = amdgpu_atif_call(handle, ATIF_FUNCTION_GET_SYSTEM_BIOS_REQUESTS, NULL);
+	info = amdgpu_atif_call(atif, ATIF_FUNCTION_GET_SYSTEM_BIOS_REQUESTS,
+				NULL);
 	if (!info)
 		return -EIO;
 
@@ -287,14 +452,14 @@ out:
  *
  * Checks the acpi event and if it matches an atif event,
  * handles it.
- * Returns NOTIFY code
+ *
+ * Returns:
+ * NOTIFY_BAD or NOTIFY_DONE, depending on the event.
  */
-int amdgpu_atif_handler(struct amdgpu_device *adev,
-			struct acpi_bus_event *event)
+static int amdgpu_atif_handler(struct amdgpu_device *adev,
+			       struct acpi_bus_event *event)
 {
-	struct amdgpu_atif *atif = &adev->atif;
-	struct atif_sbios_requests req;
-	acpi_handle handle;
+	struct amdgpu_atif *atif = &amdgpu_acpi_priv.atif;
 	int count;
 
 	DRM_DEBUG_DRIVER("event, device_class = %s, type = %#x\n",
@@ -303,48 +468,50 @@ int amdgpu_atif_handler(struct amdgpu_device *adev,
 	if (strcmp(event->device_class, ACPI_VIDEO_CLASS) != 0)
 		return NOTIFY_DONE;
 
+	/* Is this actually our event? */
 	if (!atif->notification_cfg.enabled ||
-	    event->type != atif->notification_cfg.command_code)
-		/* Not our event */
-		return NOTIFY_DONE;
-
-	/* Check pending SBIOS requests */
-	handle = ACPI_HANDLE(&adev->pdev->dev);
-	count = amdgpu_atif_get_sbios_requests(handle, &req);
-
-	if (count <= 0)
-		return NOTIFY_DONE;
-
-	DRM_DEBUG_DRIVER("ATIF: %d pending SBIOS requests\n", count);
+	    event->type != atif->notification_cfg.command_code) {
+		/* These events will generate keypresses otherwise */
+		if (event->type == ACPI_VIDEO_NOTIFY_PROBE)
+			return NOTIFY_BAD;
+		else
+			return NOTIFY_DONE;
+	}
 
-	if (req.pending & ATIF_PANEL_BRIGHTNESS_CHANGE_REQUEST) {
-		struct amdgpu_encoder *enc = atif->encoder_for_bl;
+	if (atif->functions.sbios_requests) {
+		struct atif_sbios_requests req;
 
-		if (enc) {
-			struct amdgpu_encoder_atom_dig *dig = enc->enc_priv;
+		/* Check pending SBIOS requests */
+		count = amdgpu_atif_get_sbios_requests(atif, &req);
 
-			DRM_DEBUG_DRIVER("Changing brightness to %d\n",
-					req.backlight_level);
+		if (count <= 0)
+			return NOTIFY_BAD;
 
-			amdgpu_display_backlight_set_level(adev, enc, req.backlight_level);
+		DRM_DEBUG_DRIVER("ATIF: %d pending SBIOS requests\n", count);
 
-#if defined(CONFIG_BACKLIGHT_CLASS_DEVICE) || defined(CONFIG_BACKLIGHT_CLASS_DEVICE_MODULE)
-			backlight_force_update(dig->bl_dev,
-					       BACKLIGHT_UPDATE_HOTKEY);
-#endif
+		if (req.pending & ATIF_PANEL_BRIGHTNESS_CHANGE_REQUEST) {
+			if (atif->bd) {
+				DRM_DEBUG_DRIVER("Changing brightness to %d\n",
+						 req.backlight_level);
+				/*
+				 * XXX backlight_device_set_brightness() is
+				 * hardwired to post BACKLIGHT_UPDATE_SYSFS.
+				 * It probably should accept 'reason' parameter.
+				 */
+				backlight_device_set_brightness(atif->bd, req.backlight_level);
+			}
 		}
-	}
-	if (req.pending & ATIF_DGPU_DISPLAY_EVENT) {
-		if ((adev->flags & AMD_IS_PX) &&
-		    amdgpu_atpx_dgpu_req_power_for_displays()) {
-			pm_runtime_get_sync(adev->ddev->dev);
-			/* Just fire off a uevent and let userspace tell us what to do */
-			drm_helper_hpd_irq_event(adev->ddev);
-			pm_runtime_mark_last_busy(adev->ddev->dev);
-			pm_runtime_put_autosuspend(adev->ddev->dev);
+
+		if (req.pending & ATIF_DGPU_DISPLAY_EVENT) {
+			if (adev->flags & AMD_IS_PX) {
+				pm_runtime_get_sync(adev_to_drm(adev)->dev);
+				/* Just fire off a uevent and let userspace tell us what to do */
+				drm_helper_hpd_irq_event(adev_to_drm(adev));
+				pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+			}
 		}
+		/* TODO: check other events */
 	}
-	/* TODO: check other events */
 
 	/* We've handled the event, stop the notifier chain. The ACPI interface
 	 * overloads ACPI_VIDEO_NOTIFY_PROBE, we don't want to send that to
@@ -359,14 +526,15 @@ int amdgpu_atif_handler(struct amdgpu_device *adev,
 /**
  * amdgpu_atcs_call - call an ATCS method
  *
- * @handle: acpi handle
+ * @atcs: atcs structure
  * @function: the ATCS function to execute
  * @params: ATCS function params
  *
  * Executes the requested ATCS function (all asics).
  * Returns a pointer to the acpi output buffer.
  */
-static union acpi_object *amdgpu_atcs_call(acpi_handle handle, int function,
+static union acpi_object *amdgpu_atcs_call(struct amdgpu_atcs *atcs,
+					   int function,
 					   struct acpi_buffer *params)
 {
 	acpi_status status;
@@ -390,7 +558,7 @@ static union acpi_object *amdgpu_atcs_call(acpi_handle handle, int function,
 		atcs_arg_elements[1].integer.value = 0;
 	}
 
-	status = acpi_evaluate_object(handle, "ATCS", &atcs_arg, &buffer);
+	status = acpi_evaluate_object(atcs->handle, NULL, &atcs_arg, &buffer);
 
 	/* Fail only if calling the method fails and ATIF is supported */
 	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
@@ -419,12 +587,12 @@ static void amdgpu_atcs_parse_functions(struct amdgpu_atcs_functions *f, u32 mas
 	f->pcie_perf_req = mask & ATCS_PCIE_PERFORMANCE_REQUEST_SUPPORTED;
 	f->pcie_dev_rdy = mask & ATCS_PCIE_DEVICE_READY_NOTIFICATION_SUPPORTED;
 	f->pcie_bus_width = mask & ATCS_SET_PCIE_BUS_WIDTH_SUPPORTED;
+	f->power_shift_control = mask & ATCS_SET_POWER_SHIFT_CONTROL_SUPPORTED;
 }
 
 /**
  * amdgpu_atcs_verify_interface - verify ATCS
  *
- * @handle: acpi handle
  * @atcs: amdgpu atcs struct
  *
  * Execute the ATCS_FUNCTION_VERIFY_INTERFACE ATCS function
@@ -432,15 +600,14 @@ static void amdgpu_atcs_parse_functions(struct amdgpu_atcs_functions *f, u32 mas
  * (all asics).
  * returns 0 on success, error on failure.
  */
-static int amdgpu_atcs_verify_interface(acpi_handle handle,
-					struct amdgpu_atcs *atcs)
+static int amdgpu_atcs_verify_interface(struct amdgpu_atcs *atcs)
 {
 	union acpi_object *info;
 	struct atcs_verify_interface output;
 	size_t size;
 	int err = 0;
 
-	info = amdgpu_atcs_call(handle, ATCS_FUNCTION_VERIFY_INTERFACE, NULL);
+	info = amdgpu_atcs_call(atcs, ATCS_FUNCTION_VERIFY_INTERFACE, NULL);
 	if (!info)
 		return -EIO;
 
@@ -477,7 +644,7 @@ out:
  */
 bool amdgpu_acpi_is_pcie_performance_request_supported(struct amdgpu_device *adev)
 {
-	struct amdgpu_atcs *atcs = &adev->atcs;
+	struct amdgpu_atcs *atcs = &amdgpu_acpi_priv.atcs;
 
 	if (atcs->functions.pcie_perf_req && atcs->functions.pcie_dev_rdy)
 		return true;
@@ -486,6 +653,18 @@ bool amdgpu_acpi_is_pcie_performance_request_supported(struct amdgpu_device *ade
 }
 
 /**
+ * amdgpu_acpi_is_power_shift_control_supported
+ *
+ * Check if the ATCS power shift control method
+ * is supported.
+ * returns true if supported, false if not.
+ */
+bool amdgpu_acpi_is_power_shift_control_supported(void)
+{
+	return amdgpu_acpi_priv.atcs.functions.power_shift_control;
+}
+
+/**
  * amdgpu_acpi_pcie_notify_device_ready
  *
  * @adev: amdgpu_device pointer
@@ -496,19 +675,13 @@ bool amdgpu_acpi_is_pcie_performance_request_supported(struct amdgpu_device *ade
  */
 int amdgpu_acpi_pcie_notify_device_ready(struct amdgpu_device *adev)
 {
-	acpi_handle handle;
 	union acpi_object *info;
-	struct amdgpu_atcs *atcs = &adev->atcs;
-
-	/* Get the device handle */
-	handle = ACPI_HANDLE(&adev->pdev->dev);
-	if (!handle)
-		return -EINVAL;
+	struct amdgpu_atcs *atcs = &amdgpu_acpi_priv.atcs;
 
 	if (!atcs->functions.pcie_dev_rdy)
 		return -EINVAL;
 
-	info = amdgpu_atcs_call(handle, ATCS_FUNCTION_PCIE_DEVICE_READY_NOTIFICATION, NULL);
+	info = amdgpu_atcs_call(atcs, ATCS_FUNCTION_PCIE_DEVICE_READY_NOTIFICATION, NULL);
 	if (!info)
 		return -EIO;
 
@@ -531,18 +704,15 @@ int amdgpu_acpi_pcie_notify_device_ready(struct amdgpu_device *adev)
 int amdgpu_acpi_pcie_performance_request(struct amdgpu_device *adev,
 					 u8 perf_req, bool advertise)
 {
-	acpi_handle handle;
 	union acpi_object *info;
-	struct amdgpu_atcs *atcs = &adev->atcs;
+	struct amdgpu_atcs *atcs = &amdgpu_acpi_priv.atcs;
 	struct atcs_pref_req_input atcs_input;
 	struct atcs_pref_req_output atcs_output;
 	struct acpi_buffer params;
 	size_t size;
 	u32 retry = 3;
 
-	/* Get the device handle */
-	handle = ACPI_HANDLE(&adev->pdev->dev);
-	if (!handle)
+	if (amdgpu_acpi_pcie_notify_device_ready(adev))
 		return -EINVAL;
 
 	if (!atcs->functions.pcie_perf_req)
@@ -550,7 +720,7 @@ int amdgpu_acpi_pcie_performance_request(struct amdgpu_device *adev,
 
 	atcs_input.size = sizeof(struct atcs_pref_req_input);
 	/* client id (bit 2-0: func num, 7-3: dev num, 15-8: bus num) */
-	atcs_input.client_id = adev->pdev->devfn | (adev->pdev->bus->number << 8);
+	atcs_input.client_id = pci_dev_id(adev->pdev);
 	atcs_input.valid_flags_mask = ATCS_VALID_FLAGS_MASK;
 	atcs_input.flags = ATCS_WAIT_FOR_COMPLETION;
 	if (advertise)
@@ -562,7 +732,7 @@ int amdgpu_acpi_pcie_performance_request(struct amdgpu_device *adev,
 	params.pointer = &atcs_input;
 
 	while (retry--) {
-		info = amdgpu_atcs_call(handle, ATCS_FUNCTION_PCIE_PERFORMANCE_REQUEST, &params);
+		info = amdgpu_atcs_call(atcs, ATCS_FUNCTION_PCIE_PERFORMANCE_REQUEST, &params);
 		if (!info)
 			return -EIO;
 
@@ -596,6 +766,438 @@ int amdgpu_acpi_pcie_performance_request(struct amdgpu_device *adev,
 }
 
 /**
+ * amdgpu_acpi_power_shift_control
+ *
+ * @adev: amdgpu_device pointer
+ * @dev_state: device acpi state
+ * @drv_state: driver state
+ *
+ * Executes the POWER_SHIFT_CONTROL method to
+ * communicate current dGPU device state and
+ * driver state to APU/SBIOS.
+ * returns 0 on success, error on failure.
+ */
+int amdgpu_acpi_power_shift_control(struct amdgpu_device *adev,
+				    u8 dev_state, bool drv_state)
+{
+	union acpi_object *info;
+	struct amdgpu_atcs *atcs = &amdgpu_acpi_priv.atcs;
+	struct atcs_pwr_shift_input atcs_input;
+	struct acpi_buffer params;
+
+	if (!amdgpu_acpi_is_power_shift_control_supported())
+		return -EINVAL;
+
+	atcs_input.size = sizeof(struct atcs_pwr_shift_input);
+	/* dGPU id (bit 2-0: func num, 7-3: dev num, 15-8: bus num) */
+	atcs_input.dgpu_id = pci_dev_id(adev->pdev);
+	atcs_input.dev_acpi_state = dev_state;
+	atcs_input.drv_state = drv_state;
+
+	params.length = sizeof(struct atcs_pwr_shift_input);
+	params.pointer = &atcs_input;
+
+	info = amdgpu_atcs_call(atcs, ATCS_FUNCTION_POWER_SHIFT_CONTROL, &params);
+	if (!info) {
+		DRM_ERROR("ATCS PSC update failed\n");
+		return -EIO;
+	}
+
+	kfree(info);
+	return 0;
+}
+
+/**
+ * amdgpu_acpi_smart_shift_update - update dGPU device state to SBIOS
+ *
+ * @adev: amdgpu device pointer
+ * @ss_state: current smart shift event
+ *
+ * returns 0 on success,
+ * otherwise return error number.
+ */
+int amdgpu_acpi_smart_shift_update(struct amdgpu_device *adev,
+				   enum amdgpu_ss ss_state)
+{
+	int r;
+
+	if (!amdgpu_device_supports_smart_shift(adev))
+		return 0;
+
+	switch (ss_state) {
+	/* SBIOS trigger “stop”, “enable” and “start” at D0, Driver Operational.
+	 * SBIOS trigger “stop” at D3, Driver Not Operational.
+	 * SBIOS trigger “stop” and “disable” at D0, Driver NOT operational.
+	 */
+	case AMDGPU_SS_DRV_LOAD:
+		r = amdgpu_acpi_power_shift_control(adev,
+						    AMDGPU_ATCS_PSC_DEV_STATE_D0,
+						    AMDGPU_ATCS_PSC_DRV_STATE_OPR);
+		break;
+	case AMDGPU_SS_DEV_D0:
+		r = amdgpu_acpi_power_shift_control(adev,
+						    AMDGPU_ATCS_PSC_DEV_STATE_D0,
+						    AMDGPU_ATCS_PSC_DRV_STATE_OPR);
+		break;
+	case AMDGPU_SS_DEV_D3:
+		r = amdgpu_acpi_power_shift_control(adev,
+						    AMDGPU_ATCS_PSC_DEV_STATE_D3_HOT,
+						    AMDGPU_ATCS_PSC_DRV_STATE_NOT_OPR);
+		break;
+	case AMDGPU_SS_DRV_UNLOAD:
+		r = amdgpu_acpi_power_shift_control(adev,
+						    AMDGPU_ATCS_PSC_DEV_STATE_D0,
+						    AMDGPU_ATCS_PSC_DRV_STATE_NOT_OPR);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return r;
+}
+
+#ifdef CONFIG_ACPI_NUMA
+static inline uint64_t amdgpu_acpi_get_numa_size(int nid)
+{
+	/* This is directly using si_meminfo_node implementation as the
+	 * function is not exported.
+	 */
+	int zone_type;
+	uint64_t managed_pages = 0;
+
+	pg_data_t *pgdat = NODE_DATA(nid);
+
+	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
+		managed_pages +=
+			zone_managed_pages(&pgdat->node_zones[zone_type]);
+	return managed_pages * PAGE_SIZE;
+}
+
+static struct amdgpu_numa_info *amdgpu_acpi_get_numa_info(uint32_t pxm)
+{
+	struct amdgpu_numa_info *numa_info;
+	int nid;
+
+	numa_info = xa_load(&numa_info_xa, pxm);
+
+	if (!numa_info) {
+		struct sysinfo info;
+
+		numa_info = kzalloc(sizeof(*numa_info), GFP_KERNEL);
+		if (!numa_info)
+			return NULL;
+
+		nid = pxm_to_node(pxm);
+		numa_info->pxm = pxm;
+		numa_info->nid = nid;
+
+		if (numa_info->nid == NUMA_NO_NODE) {
+			si_meminfo(&info);
+			numa_info->size = info.totalram * info.mem_unit;
+		} else {
+			numa_info->size = amdgpu_acpi_get_numa_size(nid);
+		}
+		xa_store(&numa_info_xa, numa_info->pxm, numa_info, GFP_KERNEL);
+	}
+
+	return numa_info;
+}
+#endif
+
+/**
+ * amdgpu_acpi_get_node_id - obtain the NUMA node id for corresponding amdgpu
+ * acpi device handle
+ *
+ * @handle: acpi handle
+ * @numa_info: amdgpu_numa_info structure holding numa information
+ *
+ * Queries the ACPI interface to fetch the corresponding NUMA Node ID for a
+ * given amdgpu acpi device.
+ *
+ * Returns ACPI STATUS OK with Node ID on success or the corresponding failure reason
+ */
+static acpi_status amdgpu_acpi_get_node_id(acpi_handle handle,
+				    struct amdgpu_numa_info **numa_info)
+{
+#ifdef CONFIG_ACPI_NUMA
+	u64 pxm;
+	acpi_status status;
+
+	if (!numa_info)
+		return_ACPI_STATUS(AE_ERROR);
+
+	status = acpi_evaluate_integer(handle, "_PXM", NULL, &pxm);
+
+	if (ACPI_FAILURE(status))
+		return status;
+
+	*numa_info = amdgpu_acpi_get_numa_info(pxm);
+
+	if (!*numa_info)
+		return_ACPI_STATUS(AE_ERROR);
+
+	return_ACPI_STATUS(AE_OK);
+#else
+	return_ACPI_STATUS(AE_NOT_EXIST);
+#endif
+}
+
+static struct amdgpu_acpi_dev_info *amdgpu_acpi_get_dev(u32 sbdf)
+{
+	struct amdgpu_acpi_dev_info *acpi_dev;
+
+	if (list_empty(&amdgpu_acpi_dev_list))
+		return NULL;
+
+	list_for_each_entry(acpi_dev, &amdgpu_acpi_dev_list, list)
+		if (acpi_dev->sbdf == sbdf)
+			return acpi_dev;
+
+	return NULL;
+}
+
+static int amdgpu_acpi_dev_init(struct amdgpu_acpi_dev_info **dev_info,
+				struct amdgpu_acpi_xcc_info *xcc_info, u32 sbdf)
+{
+	struct amdgpu_acpi_dev_info *tmp;
+	union acpi_object *obj;
+	int ret = -ENOENT;
+
+	*dev_info = NULL;
+	tmp = kzalloc(sizeof(struct amdgpu_acpi_dev_info), GFP_KERNEL);
+	if (!tmp)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&tmp->xcc_list);
+	INIT_LIST_HEAD(&tmp->list);
+	tmp->sbdf = sbdf;
+
+	obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
+				      AMD_XCC_DSM_GET_SUPP_MODE, NULL,
+				      ACPI_TYPE_INTEGER);
+
+	if (!obj) {
+		acpi_handle_debug(xcc_info->handle,
+				  "_DSM function %d evaluation failed",
+				  AMD_XCC_DSM_GET_SUPP_MODE);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	tmp->supp_xcp_mode = obj->integer.value & 0xFFFF;
+	ACPI_FREE(obj);
+
+	obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
+				      AMD_XCC_DSM_GET_XCP_MODE, NULL,
+				      ACPI_TYPE_INTEGER);
+
+	if (!obj) {
+		acpi_handle_debug(xcc_info->handle,
+				  "_DSM function %d evaluation failed",
+				  AMD_XCC_DSM_GET_XCP_MODE);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	tmp->xcp_mode = obj->integer.value & 0xFFFF;
+	tmp->mem_mode = (obj->integer.value >> 32) & 0xFFFF;
+	ACPI_FREE(obj);
+
+	/* Evaluate DSMs and fill XCC information */
+	obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
+				      AMD_XCC_DSM_GET_TMR_INFO, NULL,
+				      ACPI_TYPE_PACKAGE);
+
+	if (!obj || obj->package.count < 2) {
+		acpi_handle_debug(xcc_info->handle,
+				  "_DSM function %d evaluation failed",
+				  AMD_XCC_DSM_GET_TMR_INFO);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	tmp->tmr_base = obj->package.elements[0].integer.value;
+	tmp->tmr_size = obj->package.elements[1].integer.value;
+	ACPI_FREE(obj);
+
+	DRM_DEBUG_DRIVER(
+		"New dev(%x): Supported xcp mode: %x curr xcp_mode : %x mem mode : %x, tmr base: %llx tmr size: %llx  ",
+		tmp->sbdf, tmp->supp_xcp_mode, tmp->xcp_mode, tmp->mem_mode,
+		tmp->tmr_base, tmp->tmr_size);
+	list_add_tail(&tmp->list, &amdgpu_acpi_dev_list);
+	*dev_info = tmp;
+
+	return 0;
+
+out:
+	if (obj)
+		ACPI_FREE(obj);
+	kfree(tmp);
+
+	return ret;
+}
+
+static int amdgpu_acpi_get_xcc_info(struct amdgpu_acpi_xcc_info *xcc_info,
+				    u32 *sbdf)
+{
+	union acpi_object *obj;
+	acpi_status status;
+	int ret = -ENOENT;
+
+	obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
+				      AMD_XCC_DSM_GET_NUM_FUNCS, NULL,
+				      ACPI_TYPE_INTEGER);
+
+	if (!obj || obj->integer.value != AMD_XCC_DSM_NUM_FUNCS)
+		goto out;
+	ACPI_FREE(obj);
+
+	/* Evaluate DSMs and fill XCC information */
+	obj = acpi_evaluate_dsm_typed(xcc_info->handle, &amd_xcc_dsm_guid, 0,
+				      AMD_XCC_DSM_GET_VF_XCC_MAPPING, NULL,
+				      ACPI_TYPE_INTEGER);
+
+	if (!obj) {
+		acpi_handle_debug(xcc_info->handle,
+				  "_DSM function %d evaluation failed",
+				  AMD_XCC_DSM_GET_VF_XCC_MAPPING);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	/* PF xcc id [39:32] */
+	xcc_info->phy_id = (obj->integer.value >> 32) & 0xFF;
+	/* xcp node of this xcc [47:40] */
+	xcc_info->xcp_node = (obj->integer.value >> 40) & 0xFF;
+	/* PF domain of this xcc [31:16] */
+	*sbdf = (obj->integer.value) & 0xFFFF0000;
+	/* PF bus/dev/fn of this xcc [63:48] */
+	*sbdf |= (obj->integer.value >> 48) & 0xFFFF;
+	ACPI_FREE(obj);
+	obj = NULL;
+
+	status =
+		amdgpu_acpi_get_node_id(xcc_info->handle, &xcc_info->numa_info);
+
+	/* TODO: check if this check is required */
+	if (ACPI_SUCCESS(status))
+		ret = 0;
+out:
+	if (obj)
+		ACPI_FREE(obj);
+
+	return ret;
+}
+
+static int amdgpu_acpi_enumerate_xcc(void)
+{
+	struct amdgpu_acpi_dev_info *dev_info = NULL;
+	struct amdgpu_acpi_xcc_info *xcc_info;
+	struct acpi_device *acpi_dev;
+	char hid[ACPI_ID_LEN];
+	int ret, id;
+	u32 sbdf;
+
+	INIT_LIST_HEAD(&amdgpu_acpi_dev_list);
+	xa_init(&numa_info_xa);
+
+	for (id = 0; id < AMD_XCC_MAX_HID; id++) {
+		sprintf(hid, "%s%d", "AMD", AMD_XCC_HID_START + id);
+		acpi_dev = acpi_dev_get_first_match_dev(hid, NULL, -1);
+		/* These ACPI objects are expected to be in sequential order. If
+		 * one is not found, no need to check the rest.
+		 */
+		if (!acpi_dev) {
+			DRM_DEBUG_DRIVER("No matching acpi device found for %s",
+					 hid);
+			break;
+		}
+
+		xcc_info = kzalloc(sizeof(struct amdgpu_acpi_xcc_info),
+				   GFP_KERNEL);
+		if (!xcc_info) {
+			DRM_ERROR("Failed to allocate memory for xcc info\n");
+			return -ENOMEM;
+		}
+
+		INIT_LIST_HEAD(&xcc_info->list);
+		xcc_info->handle = acpi_device_handle(acpi_dev);
+		acpi_dev_put(acpi_dev);
+
+		ret = amdgpu_acpi_get_xcc_info(xcc_info, &sbdf);
+		if (ret) {
+			kfree(xcc_info);
+			continue;
+		}
+
+		dev_info = amdgpu_acpi_get_dev(sbdf);
+
+		if (!dev_info)
+			ret = amdgpu_acpi_dev_init(&dev_info, xcc_info, sbdf);
+
+		if (ret == -ENOMEM)
+			return ret;
+
+		if (!dev_info) {
+			kfree(xcc_info);
+			continue;
+		}
+
+		list_add_tail(&xcc_info->list, &dev_info->xcc_list);
+	}
+
+	return 0;
+}
+
+int amdgpu_acpi_get_tmr_info(struct amdgpu_device *adev, u64 *tmr_offset,
+			     u64 *tmr_size)
+{
+	struct amdgpu_acpi_dev_info *dev_info;
+	u32 sbdf;
+
+	if (!tmr_offset || !tmr_size)
+		return -EINVAL;
+
+	sbdf = (pci_domain_nr(adev->pdev->bus) << 16);
+	sbdf |= pci_dev_id(adev->pdev);
+	dev_info = amdgpu_acpi_get_dev(sbdf);
+	if (!dev_info)
+		return -ENOENT;
+
+	*tmr_offset = dev_info->tmr_base;
+	*tmr_size = dev_info->tmr_size;
+
+	return 0;
+}
+
+int amdgpu_acpi_get_mem_info(struct amdgpu_device *adev, int xcc_id,
+			     struct amdgpu_numa_info *numa_info)
+{
+	struct amdgpu_acpi_dev_info *dev_info;
+	struct amdgpu_acpi_xcc_info *xcc_info;
+	u32 sbdf;
+
+	if (!numa_info)
+		return -EINVAL;
+
+	sbdf = (pci_domain_nr(adev->pdev->bus) << 16);
+	sbdf |= pci_dev_id(adev->pdev);
+	dev_info = amdgpu_acpi_get_dev(sbdf);
+	if (!dev_info)
+		return -ENOENT;
+
+	list_for_each_entry(xcc_info, &dev_info->xcc_list, list) {
+		if (xcc_info->phy_id == xcc_id) {
+			memcpy(numa_info, xcc_info->numa_info,
+			       sizeof(*numa_info));
+			return 0;
+		}
+	}
+
+	return -ENOENT;
+}
+
+/**
  * amdgpu_acpi_event - handle notify events
  *
  * @nb: notifier block
@@ -638,47 +1240,181 @@ static int amdgpu_acpi_event(struct notifier_block *nb,
  */
 int amdgpu_acpi_init(struct amdgpu_device *adev)
 {
-	acpi_handle handle;
-	struct amdgpu_atif *atif = &adev->atif;
-	struct amdgpu_atcs *atcs = &adev->atcs;
+	struct amdgpu_atif *atif = &amdgpu_acpi_priv.atif;
+
+	if (atif->notifications.brightness_change) {
+		if (adev->dc_enabled) {
+#if defined(CONFIG_DRM_AMD_DC)
+			struct amdgpu_display_manager *dm = &adev->dm;
+
+			if (dm->backlight_dev[0])
+				atif->bd = dm->backlight_dev[0];
+#endif
+		} else {
+			struct drm_encoder *tmp;
+
+			/* Find the encoder controlling the brightness */
+			list_for_each_entry(tmp, &adev_to_drm(adev)->mode_config.encoder_list,
+					    head) {
+				struct amdgpu_encoder *enc = to_amdgpu_encoder(tmp);
+
+				if ((enc->devices & (ATOM_DEVICE_LCD_SUPPORT)) &&
+				    enc->enc_priv) {
+					struct amdgpu_encoder_atom_dig *dig = enc->enc_priv;
+
+					if (dig->bl_dev) {
+						atif->bd = dig->bl_dev;
+						break;
+					}
+				}
+			}
+		}
+	}
+	adev->acpi_nb.notifier_call = amdgpu_acpi_event;
+	register_acpi_notifier(&adev->acpi_nb);
+
+	return 0;
+}
+
+void amdgpu_acpi_get_backlight_caps(struct amdgpu_dm_backlight_caps *caps)
+{
+	struct amdgpu_atif *atif = &amdgpu_acpi_priv.atif;
+
+	memcpy(caps, &atif->backlight_caps, sizeof(*caps));
+}
+
+/**
+ * amdgpu_acpi_fini - tear down driver acpi support
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Unregisters with the acpi notifier chain (all asics).
+ */
+void amdgpu_acpi_fini(struct amdgpu_device *adev)
+{
+	unregister_acpi_notifier(&adev->acpi_nb);
+}
+
+/**
+ * amdgpu_atif_pci_probe_handle - look up the ATIF handle
+ *
+ * @pdev: pci device
+ *
+ * Look up the ATIF handles (all asics).
+ * Returns true if the handle is found, false if not.
+ */
+static bool amdgpu_atif_pci_probe_handle(struct pci_dev *pdev)
+{
+	char acpi_method_name[255] = { 0 };
+	struct acpi_buffer buffer = {sizeof(acpi_method_name), acpi_method_name};
+	acpi_handle dhandle, atif_handle;
+	acpi_status status;
 	int ret;
 
-	/* Get the device handle */
-	handle = ACPI_HANDLE(&adev->pdev->dev);
+	dhandle = ACPI_HANDLE(&pdev->dev);
+	if (!dhandle)
+		return false;
 
-	if (!adev->bios || !handle)
-		return 0;
+	status = acpi_get_handle(dhandle, "ATIF", &atif_handle);
+	if (ACPI_FAILURE(status))
+		return false;
 
-	/* Call the ATCS method */
-	ret = amdgpu_atcs_verify_interface(handle, atcs);
+	amdgpu_acpi_priv.atif.handle = atif_handle;
+	acpi_get_name(amdgpu_acpi_priv.atif.handle, ACPI_FULL_PATHNAME, &buffer);
+	DRM_DEBUG_DRIVER("Found ATIF handle %s\n", acpi_method_name);
+	ret = amdgpu_atif_verify_interface(&amdgpu_acpi_priv.atif);
 	if (ret) {
-		DRM_DEBUG_DRIVER("Call to ATCS verify_interface failed: %d\n", ret);
+		amdgpu_acpi_priv.atif.handle = 0;
+		return false;
 	}
+	return true;
+}
+
+/**
+ * amdgpu_atcs_pci_probe_handle - look up the ATCS handle
+ *
+ * @pdev: pci device
+ *
+ * Look up the ATCS handles (all asics).
+ * Returns true if the handle is found, false if not.
+ */
+static bool amdgpu_atcs_pci_probe_handle(struct pci_dev *pdev)
+{
+	char acpi_method_name[255] = { 0 };
+	struct acpi_buffer buffer = { sizeof(acpi_method_name), acpi_method_name };
+	acpi_handle dhandle, atcs_handle;
+	acpi_status status;
+	int ret;
+
+	dhandle = ACPI_HANDLE(&pdev->dev);
+	if (!dhandle)
+		return false;
 
-	/* Call the ATIF method */
-	ret = amdgpu_atif_verify_interface(handle, atif);
+	status = acpi_get_handle(dhandle, "ATCS", &atcs_handle);
+	if (ACPI_FAILURE(status))
+		return false;
+
+	amdgpu_acpi_priv.atcs.handle = atcs_handle;
+	acpi_get_name(amdgpu_acpi_priv.atcs.handle, ACPI_FULL_PATHNAME, &buffer);
+	DRM_DEBUG_DRIVER("Found ATCS handle %s\n", acpi_method_name);
+	ret = amdgpu_atcs_verify_interface(&amdgpu_acpi_priv.atcs);
 	if (ret) {
-		DRM_DEBUG_DRIVER("Call to ATIF verify_interface failed: %d\n", ret);
-		goto out;
+		amdgpu_acpi_priv.atcs.handle = 0;
+		return false;
 	}
+	return true;
+}
 
-	if (atif->notifications.brightness_change) {
-		struct drm_encoder *tmp;
-
-		/* Find the encoder controlling the brightness */
-		list_for_each_entry(tmp, &adev->ddev->mode_config.encoder_list,
-				head) {
-			struct amdgpu_encoder *enc = to_amdgpu_encoder(tmp);
-
-			if ((enc->devices & (ATOM_DEVICE_LCD_SUPPORT)) &&
-			    enc->enc_priv) {
-				struct amdgpu_encoder_atom_dig *dig = enc->enc_priv;
-				if (dig->bl_dev) {
-					atif->encoder_for_bl = enc;
-					break;
-				}
-			}
-		}
+
+/**
+ * amdgpu_acpi_should_gpu_reset
+ *
+ * @adev: amdgpu_device_pointer
+ *
+ * returns true if should reset GPU, false if not
+ */
+bool amdgpu_acpi_should_gpu_reset(struct amdgpu_device *adev)
+{
+	if ((adev->flags & AMD_IS_APU) &&
+	    adev->gfx.imu.funcs) /* Not need to do mode2 reset for IMU enabled APUs */
+		return false;
+
+	if ((adev->flags & AMD_IS_APU) &&
+	    amdgpu_acpi_is_s3_active(adev))
+		return false;
+
+	if (amdgpu_sriov_vf(adev))
+		return false;
+
+#if IS_ENABLED(CONFIG_SUSPEND)
+	return pm_suspend_target_state != PM_SUSPEND_TO_IDLE;
+#else
+	return true;
+#endif
+}
+
+/*
+ * amdgpu_acpi_detect - detect ACPI ATIF/ATCS methods
+ *
+ * Check if we have the ATIF/ATCS methods and populate
+ * the structures in the driver.
+ */
+void amdgpu_acpi_detect(void)
+{
+	struct amdgpu_atif *atif = &amdgpu_acpi_priv.atif;
+	struct amdgpu_atcs *atcs = &amdgpu_acpi_priv.atcs;
+	struct pci_dev *pdev = NULL;
+	int ret;
+
+	while ((pdev = pci_get_base_class(PCI_BASE_CLASS_DISPLAY, pdev))) {
+		if ((pdev->class != PCI_CLASS_DISPLAY_VGA << 8) &&
+		    (pdev->class != PCI_CLASS_DISPLAY_OTHER << 8))
+			continue;
+
+		if (!atif->handle)
+			amdgpu_atif_pci_probe_handle(pdev);
+		if (!atcs->handle)
+			amdgpu_atcs_pci_probe_handle(pdev);
 	}
 
 	if (atif->functions.sbios_requests && !atif->functions.system_params) {
@@ -690,8 +1426,7 @@ int amdgpu_acpi_init(struct amdgpu_device *adev)
 	}
 
 	if (atif->functions.system_params) {
-		ret = amdgpu_atif_get_notification_params(handle,
-				&atif->notification_cfg);
+		ret = amdgpu_atif_get_notification_params(atif);
 		if (ret) {
 			DRM_DEBUG_DRIVER("Call to GET_SYSTEM_PARAMS failed: %d\n",
 					ret);
@@ -700,21 +1435,131 @@ int amdgpu_acpi_init(struct amdgpu_device *adev)
 		}
 	}
 
-out:
-	adev->acpi_nb.notifier_call = amdgpu_acpi_event;
-	register_acpi_notifier(&adev->acpi_nb);
+	if (atif->functions.query_backlight_transfer_characteristics) {
+		ret = amdgpu_atif_query_backlight_caps(atif);
+		if (ret) {
+			DRM_DEBUG_DRIVER("Call to QUERY_BACKLIGHT_TRANSFER_CHARACTERISTICS failed: %d\n",
+					ret);
+			atif->backlight_caps.caps_valid = false;
+		}
+	} else {
+		atif->backlight_caps.caps_valid = false;
+	}
 
-	return ret;
+	amdgpu_acpi_enumerate_xcc();
+}
+
+void amdgpu_acpi_release(void)
+{
+	struct amdgpu_acpi_dev_info *dev_info, *dev_tmp;
+	struct amdgpu_acpi_xcc_info *xcc_info, *xcc_tmp;
+	struct amdgpu_numa_info *numa_info;
+	unsigned long index;
+
+	xa_for_each(&numa_info_xa, index, numa_info) {
+		kfree(numa_info);
+		xa_erase(&numa_info_xa, index);
+	}
+
+	if (list_empty(&amdgpu_acpi_dev_list))
+		return;
+
+	list_for_each_entry_safe(dev_info, dev_tmp, &amdgpu_acpi_dev_list,
+				 list) {
+		list_for_each_entry_safe(xcc_info, xcc_tmp, &dev_info->xcc_list,
+					 list) {
+			list_del(&xcc_info->list);
+			kfree(xcc_info);
+		}
+
+		list_del(&dev_info->list);
+		kfree(dev_info);
+	}
 }
 
+#if IS_ENABLED(CONFIG_SUSPEND)
 /**
- * amdgpu_acpi_fini - tear down driver acpi support
+ * amdgpu_acpi_is_s3_active
  *
- * @adev: amdgpu_device pointer
+ * @adev: amdgpu_device_pointer
  *
- * Unregisters with the acpi notifier chain (all asics).
+ * returns true if supported, false if not.
  */
-void amdgpu_acpi_fini(struct amdgpu_device *adev)
+bool amdgpu_acpi_is_s3_active(struct amdgpu_device *adev)
 {
-	unregister_acpi_notifier(&adev->acpi_nb);
+	return !(adev->flags & AMD_IS_APU) ||
+		(pm_suspend_target_state == PM_SUSPEND_MEM);
+}
+
+/**
+ * amdgpu_acpi_is_s0ix_active
+ *
+ * @adev: amdgpu_device_pointer
+ *
+ * returns true if supported, false if not.
+ */
+bool amdgpu_acpi_is_s0ix_active(struct amdgpu_device *adev)
+{
+	if (!(adev->flags & AMD_IS_APU) ||
+	    (pm_suspend_target_state != PM_SUSPEND_TO_IDLE))
+		return false;
+
+	if (adev->asic_type < CHIP_RAVEN)
+		return false;
+
+	if (!(adev->pm.pp_feature & PP_GFXOFF_MASK))
+		return false;
+
+	/*
+	 * If ACPI_FADT_LOW_POWER_S0 is not set in the FADT, it is generally
+	 * risky to do any special firmware-related preparations for entering
+	 * S0ix even though the system is suspending to idle, so return false
+	 * in that case.
+	 */
+	if (!(acpi_gbl_FADT.flags & ACPI_FADT_LOW_POWER_S0)) {
+		dev_err_once(adev->dev,
+			      "Power consumption will be higher as BIOS has not been configured for suspend-to-idle.\n"
+			      "To use suspend-to-idle change the sleep mode in BIOS setup.\n");
+		return false;
+	}
+
+#if !IS_ENABLED(CONFIG_AMD_PMC)
+	dev_err_once(adev->dev,
+		      "Power consumption will be higher as the kernel has not been compiled with CONFIG_AMD_PMC.\n");
+	return false;
+#else
+	return true;
+#endif /* CONFIG_AMD_PMC */
+}
+#endif /* CONFIG_SUSPEND */
+
+#if IS_ENABLED(CONFIG_DRM_AMD_ISP)
+static const struct acpi_device_id isp_sensor_ids[] = {
+	{ "OMNI5C10" },
+	{ }
+};
+
+static int isp_match_acpi_device_ids(struct device *dev, const void *data)
+{
+	return acpi_match_device(data, dev) ? 1 : 0;
+}
+
+int amdgpu_acpi_get_isp4_dev(struct acpi_device **dev)
+{
+	struct device *pdev __free(put_device) = NULL;
+	struct acpi_device *acpi_pdev;
+
+	pdev = bus_find_device(&platform_bus_type, NULL, isp_sensor_ids,
+			       isp_match_acpi_device_ids);
+	if (!pdev)
+		return -EINVAL;
+
+	acpi_pdev = ACPI_COMPANION(pdev);
+	if (!acpi_pdev)
+		return -ENODEV;
+
+	*dev = acpi_pdev;
+
+	return 0;
 }
+#endif /* CONFIG_DRM_AMD_ISP */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_afmt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_afmt.c
index 3889486f71fe..80771b1480ff 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_afmt.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_afmt.c
@@ -25,7 +25,7 @@
  */
 #include <linux/hdmi.h>
 #include <linux/gcd.h>
-#include <drm/drmP.h>
+
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 
@@ -100,6 +100,7 @@ struct amdgpu_afmt_acr amdgpu_afmt_acr(uint32_t clock)
 	amdgpu_afmt_calc_cts(clock, &res.cts_32khz, &res.n_32khz, 32000);
 	amdgpu_afmt_calc_cts(clock, &res.cts_44_1khz, &res.n_44_1khz, 44100);
 	amdgpu_afmt_calc_cts(clock, &res.cts_48khz, &res.n_48khz, 48000);
+	res.clock = clock;
 
 	return res;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index dba8a5b25e66..a2879d2b7c8e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: MIT
 /*
  * Copyright 2014 Advanced Micro Devices, Inc.
  *
@@ -21,226 +22,891 @@
  */
 
 #include "amdgpu_amdkfd.h"
+#include "amd_pcie.h"
 #include "amd_shared.h"
-#include <drm/drmP.h>
+
 #include "amdgpu.h"
+#include "amdgpu_gfx.h"
+#include "amdgpu_dma_buf.h"
+#include <drm/ttm/ttm_tt.h>
 #include <linux/module.h>
+#include <linux/dma-buf.h>
+#include "amdgpu_xgmi.h"
+#include <uapi/linux/kfd_ioctl.h>
+#include "amdgpu_ras.h"
+#include "amdgpu_umc.h"
+#include "amdgpu_reset.h"
+
+/* Total memory size in system memory and all GPU VRAM. Used to
+ * estimate worst case amount of memory to reserve for page tables
+ */
+uint64_t amdgpu_amdkfd_total_mem_size;
 
-const struct kfd2kgd_calls *kfd2kgd;
-const struct kgd2kfd_calls *kgd2kfd;
-bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+static bool kfd_initialized;
 
 int amdgpu_amdkfd_init(void)
 {
+	struct sysinfo si;
 	int ret;
 
-#if defined(CONFIG_HSA_AMD_MODULE)
-	int (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**);
+	si_meminfo(&si);
+	amdgpu_amdkfd_total_mem_size = si.freeram - si.freehigh;
+	amdgpu_amdkfd_total_mem_size *= si.mem_unit;
 
-	kgd2kfd_init_p = symbol_request(kgd2kfd_init);
+	ret = kgd2kfd_init();
+	kfd_initialized = !ret;
 
-	if (kgd2kfd_init_p == NULL)
-		return -ENOENT;
+	return ret;
+}
 
-	ret = kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd);
-	if (ret) {
-		symbol_put(kgd2kfd_init);
-		kgd2kfd = NULL;
+void amdgpu_amdkfd_fini(void)
+{
+	if (kfd_initialized) {
+		kgd2kfd_exit();
+		kfd_initialized = false;
 	}
+}
 
-#elif defined(CONFIG_HSA_AMD)
-	ret = kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd);
-	if (ret)
-		kgd2kfd = NULL;
+void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev)
+{
+	bool vf = amdgpu_sriov_vf(adev);
 
-#else
-	ret = -ENOENT;
-#endif
+	if (!kfd_initialized)
+		return;
 
-	return ret;
+	adev->kfd.dev = kgd2kfd_probe(adev, vf);
 }
 
-bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev)
+/**
+ * amdgpu_doorbell_get_kfd_info - Report doorbell configuration required to
+ *                                setup amdkfd
+ *
+ * @adev: amdgpu_device pointer
+ * @aperture_base: output returning doorbell aperture base physical address
+ * @aperture_size: output returning doorbell aperture size in bytes
+ * @start_offset: output returning # of doorbell bytes reserved for amdgpu.
+ *
+ * amdgpu and amdkfd share the doorbell aperture. amdgpu sets it up,
+ * takes doorbells required for its own rings and reports the setup to amdkfd.
+ * amdgpu reserved doorbells are at the start of the doorbell aperture.
+ */
+static void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev,
+					 phys_addr_t *aperture_base,
+					 size_t *aperture_size,
+					 size_t *start_offset)
 {
-	switch (rdev->asic_type) {
-#ifdef CONFIG_DRM_AMDGPU_CIK
-	case CHIP_KAVERI:
-		kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions();
-		break;
-#endif
-	case CHIP_CARRIZO:
-		kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions();
-		break;
-	default:
-		return false;
+	/*
+	 * The first num_kernel_doorbells are used by amdgpu.
+	 * amdkfd takes whatever's left in the aperture.
+	 */
+	if (adev->enable_mes) {
+		/*
+		 * With MES enabled, we only need to initialize
+		 * the base address. The size and offset are
+		 * not initialized as AMDGPU manages the whole
+		 * doorbell space.
+		 */
+		*aperture_base = adev->doorbell.base;
+		*aperture_size = 0;
+		*start_offset = 0;
+	} else if (adev->doorbell.size > adev->doorbell.num_kernel_doorbells *
+						sizeof(u32)) {
+		*aperture_base = adev->doorbell.base;
+		*aperture_size = adev->doorbell.size;
+		*start_offset = adev->doorbell.num_kernel_doorbells * sizeof(u32);
+	} else {
+		*aperture_base = 0;
+		*aperture_size = 0;
+		*start_offset = 0;
 	}
-
-	return true;
 }
 
-void amdgpu_amdkfd_fini(void)
+
+static void amdgpu_amdkfd_reset_work(struct work_struct *work)
 {
-	if (kgd2kfd) {
-		kgd2kfd->exit();
-		symbol_put(kgd2kfd_init);
-	}
+	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
+						  kfd.reset_work);
+
+	struct amdgpu_reset_context reset_context;
+
+	memset(&reset_context, 0, sizeof(reset_context));
+
+	reset_context.method = AMD_RESET_METHOD_NONE;
+	reset_context.reset_req_dev = adev;
+	reset_context.src = adev->enable_mes ?
+			    AMDGPU_RESET_SRC_MES :
+			    AMDGPU_RESET_SRC_HWS;
+	clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
+
+	amdgpu_device_gpu_recover(adev, NULL, &reset_context);
 }
 
-void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev)
+static const struct drm_client_funcs kfd_client_funcs = {
+	.unregister	= drm_client_release,
+};
+
+int amdgpu_amdkfd_drm_client_create(struct amdgpu_device *adev)
 {
-	if (kgd2kfd)
-		rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev,
-					rdev->pdev, kfd2kgd);
+	int ret;
+
+	if (!adev->kfd.init_complete || adev->kfd.client.dev)
+		return 0;
+
+	ret = drm_client_init(&adev->ddev, &adev->kfd.client, "kfd",
+			      &kfd_client_funcs);
+	if (ret) {
+		dev_err(adev->dev, "Failed to init DRM client: %d\n",
+			ret);
+		return ret;
+	}
+
+	drm_client_register(&adev->kfd.client);
+
+	return 0;
 }
 
-void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev)
+void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
 {
-	if (rdev->kfd) {
-		struct kgd2kfd_shared_resources gpu_resources = {
-			.compute_vmid_bitmap = 0xFF00,
+	int i;
+	int last_valid_bit;
+
+	amdgpu_amdkfd_gpuvm_init_mem_limits();
 
-			.first_compute_pipe = 1,
-			.compute_pipe_count = 4 - 1,
+	if (adev->kfd.dev) {
+		struct kgd2kfd_shared_resources gpu_resources = {
+			.compute_vmid_bitmap =
+				((1 << AMDGPU_NUM_VMID) - 1) -
+				((1 << adev->vm_manager.first_kfd_vmid) - 1),
+			.num_pipe_per_mec = adev->gfx.mec.num_pipe_per_mec,
+			.num_queue_per_pipe = adev->gfx.mec.num_queue_per_pipe,
+			.gpuvm_size = min(adev->vm_manager.max_pfn
+					  << AMDGPU_GPU_PAGE_SHIFT,
+					  AMDGPU_GMC_HOLE_START),
+			.drm_render_minor = adev_to_drm(adev)->render->index,
+			.sdma_doorbell_idx = adev->doorbell_index.sdma_engine,
+			.enable_mes = adev->enable_mes,
 		};
 
-		amdgpu_doorbell_get_kfd_info(rdev,
+		/* this is going to have a few of the MSBs set that we need to
+		 * clear
+		 */
+		bitmap_complement(gpu_resources.cp_queue_bitmap,
+				  adev->gfx.mec_bitmap[0].queue_bitmap,
+				  AMDGPU_MAX_QUEUES);
+
+		/* According to linux/bitmap.h we shouldn't use bitmap_clear if
+		 * nbits is not compile time constant
+		 */
+		last_valid_bit = 1 /* only first MEC can have compute queues */
+				* adev->gfx.mec.num_pipe_per_mec
+				* adev->gfx.mec.num_queue_per_pipe;
+		for (i = last_valid_bit; i < AMDGPU_MAX_QUEUES; ++i)
+			clear_bit(i, gpu_resources.cp_queue_bitmap);
+
+		amdgpu_doorbell_get_kfd_info(adev,
 				&gpu_resources.doorbell_physical_address,
 				&gpu_resources.doorbell_aperture_size,
 				&gpu_resources.doorbell_start_offset);
 
-		kgd2kfd->device_init(rdev->kfd, &gpu_resources);
+		/* Since SOC15, BIF starts to statically use the
+		 * lower 12 bits of doorbell addresses for routing
+		 * based on settings in registers like
+		 * SDMA0_DOORBELL_RANGE etc..
+		 * In order to route a doorbell to CP engine, the lower
+		 * 12 bits of its address has to be outside the range
+		 * set for SDMA, VCN, and IH blocks.
+		 */
+		if (adev->asic_type >= CHIP_VEGA10) {
+			gpu_resources.non_cp_doorbells_start =
+					adev->doorbell_index.first_non_cp;
+			gpu_resources.non_cp_doorbells_end =
+					adev->doorbell_index.last_non_cp;
+		}
+
+		adev->kfd.init_complete = kgd2kfd_device_init(adev->kfd.dev,
+							&gpu_resources);
+
+		amdgpu_amdkfd_total_mem_size += adev->gmc.real_vram_size;
+
+		INIT_WORK(&adev->kfd.reset_work, amdgpu_amdkfd_reset_work);
 	}
 }
 
-void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev)
+void amdgpu_amdkfd_device_fini_sw(struct amdgpu_device *adev)
 {
-	if (rdev->kfd) {
-		kgd2kfd->device_exit(rdev->kfd);
-		rdev->kfd = NULL;
+	if (adev->kfd.dev) {
+		kgd2kfd_device_exit(adev->kfd.dev);
+		adev->kfd.dev = NULL;
+		amdgpu_amdkfd_total_mem_size -= adev->gmc.real_vram_size;
 	}
 }
 
-void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev,
+void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,
 		const void *ih_ring_entry)
 {
-	if (rdev->kfd)
-		kgd2kfd->interrupt(rdev->kfd, ih_ring_entry);
+	if (adev->kfd.dev)
+		kgd2kfd_interrupt(adev->kfd.dev, ih_ring_entry);
+}
+
+void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool suspend_proc)
+{
+	if (adev->kfd.dev) {
+		if (adev->in_s0ix)
+			kgd2kfd_stop_sched_all_nodes(adev->kfd.dev);
+		else
+			kgd2kfd_suspend(adev->kfd.dev, suspend_proc);
+	}
+}
+
+int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool resume_proc)
+{
+	int r = 0;
+
+	if (adev->kfd.dev) {
+		if (adev->in_s0ix)
+			r = kgd2kfd_start_sched_all_nodes(adev->kfd.dev);
+		else
+			r = kgd2kfd_resume(adev->kfd.dev, resume_proc);
+	}
+
+	return r;
+}
+
+void amdgpu_amdkfd_suspend_process(struct amdgpu_device *adev)
+{
+	if (adev->kfd.dev)
+		kgd2kfd_suspend_process(adev->kfd.dev);
+}
+
+int amdgpu_amdkfd_resume_process(struct amdgpu_device *adev)
+{
+	int r = 0;
+
+	if (adev->kfd.dev)
+		r = kgd2kfd_resume_process(adev->kfd.dev);
+
+	return r;
 }
 
-void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev)
+int amdgpu_amdkfd_pre_reset(struct amdgpu_device *adev,
+			    struct amdgpu_reset_context *reset_context)
 {
-	if (rdev->kfd)
-		kgd2kfd->suspend(rdev->kfd);
+	int r = 0;
+
+	if (adev->kfd.dev)
+		r = kgd2kfd_pre_reset(adev->kfd.dev, reset_context);
+
+	return r;
 }
 
-int amdgpu_amdkfd_resume(struct amdgpu_device *rdev)
+int amdgpu_amdkfd_post_reset(struct amdgpu_device *adev)
 {
 	int r = 0;
 
-	if (rdev->kfd)
-		r = kgd2kfd->resume(rdev->kfd);
+	if (adev->kfd.dev)
+		r = kgd2kfd_post_reset(adev->kfd.dev);
 
 	return r;
 }
 
-int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
-			void **mem_obj, uint64_t *gpu_addr,
-			void **cpu_ptr)
+void amdgpu_amdkfd_gpu_reset(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
-	struct kgd_mem **mem = (struct kgd_mem **) mem_obj;
+	if (amdgpu_device_should_recover_gpu(adev))
+		amdgpu_reset_domain_schedule(adev->reset_domain,
+					     &adev->kfd.reset_work);
+}
+
+int amdgpu_amdkfd_alloc_gtt_mem(struct amdgpu_device *adev, size_t size,
+				void **mem_obj, uint64_t *gpu_addr,
+				void **cpu_ptr, bool cp_mqd_gfx9)
+{
+	struct amdgpu_bo *bo = NULL;
+	struct amdgpu_bo_param bp;
 	int r;
+	void *cpu_ptr_tmp = NULL;
 
-	BUG_ON(kgd == NULL);
-	BUG_ON(gpu_addr == NULL);
-	BUG_ON(cpu_ptr == NULL);
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_GTT;
+	bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+	bp.bo_ptr_size = sizeof(struct amdgpu_bo);
 
-	*mem = kmalloc(sizeof(struct kgd_mem), GFP_KERNEL);
-	if ((*mem) == NULL)
-		return -ENOMEM;
+	if (cp_mqd_gfx9)
+		bp.flags |= AMDGPU_GEM_CREATE_CP_MQD_GFX9;
 
-	r = amdgpu_bo_create(rdev, size, PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_GTT,
-			     AMDGPU_GEM_CREATE_CPU_GTT_USWC, NULL, NULL, &(*mem)->bo);
+	r = amdgpu_bo_create(adev, &bp, &bo);
 	if (r) {
-		dev_err(rdev->dev,
+		dev_err(adev->dev,
 			"failed to allocate BO for amdkfd (%d)\n", r);
 		return r;
 	}
 
 	/* map the buffer */
-	r = amdgpu_bo_reserve((*mem)->bo, true);
+	r = amdgpu_bo_reserve(bo, true);
 	if (r) {
-		dev_err(rdev->dev, "(%d) failed to reserve bo for amdkfd\n", r);
+		dev_err(adev->dev, "(%d) failed to reserve bo for amdkfd\n", r);
 		goto allocate_mem_reserve_bo_failed;
 	}
 
-	r = amdgpu_bo_pin((*mem)->bo, AMDGPU_GEM_DOMAIN_GTT,
-				&(*mem)->gpu_addr);
+	r = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT);
 	if (r) {
-		dev_err(rdev->dev, "(%d) failed to pin bo for amdkfd\n", r);
+		dev_err(adev->dev, "(%d) failed to pin bo for amdkfd\n", r);
 		goto allocate_mem_pin_bo_failed;
 	}
-	*gpu_addr = (*mem)->gpu_addr;
 
-	r = amdgpu_bo_kmap((*mem)->bo, &(*mem)->cpu_ptr);
+	r = amdgpu_ttm_alloc_gart(&bo->tbo);
 	if (r) {
-		dev_err(rdev->dev,
+		dev_err(adev->dev, "%p bind failed\n", bo);
+		goto allocate_mem_kmap_bo_failed;
+	}
+
+	r = amdgpu_bo_kmap(bo, &cpu_ptr_tmp);
+	if (r) {
+		dev_err(adev->dev,
 			"(%d) failed to map bo to kernel for amdkfd\n", r);
 		goto allocate_mem_kmap_bo_failed;
 	}
-	*cpu_ptr = (*mem)->cpu_ptr;
 
-	amdgpu_bo_unreserve((*mem)->bo);
+	*mem_obj = bo;
+	*gpu_addr = amdgpu_bo_gpu_offset(bo);
+	*cpu_ptr = cpu_ptr_tmp;
+
+	amdgpu_bo_unreserve(bo);
 
 	return 0;
 
 allocate_mem_kmap_bo_failed:
-	amdgpu_bo_unpin((*mem)->bo);
+	amdgpu_bo_unpin(bo);
 allocate_mem_pin_bo_failed:
-	amdgpu_bo_unreserve((*mem)->bo);
+	amdgpu_bo_unreserve(bo);
 allocate_mem_reserve_bo_failed:
-	amdgpu_bo_unref(&(*mem)->bo);
+	amdgpu_bo_unref(&bo);
 
 	return r;
 }
 
-void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj)
+void amdgpu_amdkfd_free_gtt_mem(struct amdgpu_device *adev, void **mem_obj)
 {
-	struct kgd_mem *mem = (struct kgd_mem *) mem_obj;
+	struct amdgpu_bo **bo = (struct amdgpu_bo **) mem_obj;
+
+	if (!bo || !*bo)
+		return;
+
+	(void)amdgpu_bo_reserve(*bo, true);
+	amdgpu_bo_kunmap(*bo);
+	amdgpu_bo_unpin(*bo);
+	amdgpu_bo_unreserve(*bo);
+	amdgpu_bo_unref(bo);
+}
+
+int amdgpu_amdkfd_alloc_gws(struct amdgpu_device *adev, size_t size,
+				void **mem_obj)
+{
+	struct amdgpu_bo *bo = NULL;
+	struct amdgpu_bo_user *ubo;
+	struct amdgpu_bo_param bp;
+	int r;
+
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = 1;
+	bp.domain = AMDGPU_GEM_DOMAIN_GWS;
+	bp.flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
+	bp.type = ttm_bo_type_device;
+	bp.resv = NULL;
+	bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+
+	r = amdgpu_bo_create_user(adev, &bp, &ubo);
+	if (r) {
+		dev_err(adev->dev,
+			"failed to allocate gws BO for amdkfd (%d)\n", r);
+		return r;
+	}
+
+	bo = &ubo->bo;
+	*mem_obj = bo;
+	return 0;
+}
 
-	BUG_ON(mem == NULL);
+void amdgpu_amdkfd_free_gws(struct amdgpu_device *adev, void *mem_obj)
+{
+	struct amdgpu_bo *bo = (struct amdgpu_bo *)mem_obj;
 
-	amdgpu_bo_reserve(mem->bo, true);
-	amdgpu_bo_kunmap(mem->bo);
-	amdgpu_bo_unpin(mem->bo);
-	amdgpu_bo_unreserve(mem->bo);
-	amdgpu_bo_unref(&(mem->bo));
-	kfree(mem);
+	amdgpu_bo_unref(&bo);
 }
 
-uint64_t get_vmem_size(struct kgd_dev *kgd)
+uint32_t amdgpu_amdkfd_get_fw_version(struct amdgpu_device *adev,
+				      enum kgd_engine_type type)
 {
-	struct amdgpu_device *rdev =
-		(struct amdgpu_device *)kgd;
+	switch (type) {
+	case KGD_ENGINE_PFP:
+		return adev->gfx.pfp_fw_version;
+
+	case KGD_ENGINE_ME:
+		return adev->gfx.me_fw_version;
+
+	case KGD_ENGINE_CE:
+		return adev->gfx.ce_fw_version;
+
+	case KGD_ENGINE_MEC1:
+		return adev->gfx.mec_fw_version;
 
-	BUG_ON(kgd == NULL);
+	case KGD_ENGINE_MEC2:
+		return adev->gfx.mec2_fw_version;
 
-	return rdev->mc.real_vram_size;
+	case KGD_ENGINE_RLC:
+		return adev->gfx.rlc_fw_version;
+
+	case KGD_ENGINE_SDMA1:
+		return adev->sdma.instance[0].fw_version;
+
+	case KGD_ENGINE_SDMA2:
+		return adev->sdma.instance[1].fw_version;
+
+	default:
+		return 0;
+	}
+
+	return 0;
 }
 
-uint64_t get_gpu_clock_counter(struct kgd_dev *kgd)
+void amdgpu_amdkfd_get_local_mem_info(struct amdgpu_device *adev,
+				      struct kfd_local_mem_info *mem_info,
+				      struct amdgpu_xcp *xcp)
 {
-	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+	memset(mem_info, 0, sizeof(*mem_info));
+
+	if (xcp) {
+		if (adev->gmc.real_vram_size == adev->gmc.visible_vram_size)
+			mem_info->local_mem_size_public =
+					KFD_XCP_MEMORY_SIZE(adev, xcp->id);
+		else
+			mem_info->local_mem_size_private =
+					KFD_XCP_MEMORY_SIZE(adev, xcp->id);
+	} else if (adev->apu_prefer_gtt) {
+		mem_info->local_mem_size_public = (ttm_tt_pages_limit() << PAGE_SHIFT);
+		mem_info->local_mem_size_private = 0;
+	} else {
+		mem_info->local_mem_size_public = adev->gmc.visible_vram_size;
+		mem_info->local_mem_size_private = adev->gmc.real_vram_size -
+						adev->gmc.visible_vram_size;
+	}
+	mem_info->vram_width = adev->gmc.vram_width;
+
+	pr_debug("Address base: %pap public 0x%llx private 0x%llx\n",
+			&adev->gmc.aper_base,
+			mem_info->local_mem_size_public,
+			mem_info->local_mem_size_private);
+
+	if (adev->pm.dpm_enabled) {
+		if (amdgpu_emu_mode == 1)
+			mem_info->mem_clk_max = 0;
+		else
+			mem_info->mem_clk_max = amdgpu_dpm_get_mclk(adev, false) / 100;
+	} else
+		mem_info->mem_clk_max = 100;
+}
 
-	if (rdev->gfx.funcs->get_gpu_clock_counter)
-		return rdev->gfx.funcs->get_gpu_clock_counter(rdev);
+uint64_t amdgpu_amdkfd_get_gpu_clock_counter(struct amdgpu_device *adev)
+{
+	if (adev->gfx.funcs->get_gpu_clock_counter)
+		return adev->gfx.funcs->get_gpu_clock_counter(adev);
 	return 0;
 }
 
-uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd)
+uint32_t amdgpu_amdkfd_get_max_engine_clock_in_mhz(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *rdev = (struct amdgpu_device *)kgd;
+	/* the sclk is in quantas of 10kHz */
+	if (adev->pm.dpm_enabled)
+		return amdgpu_dpm_get_sclk(adev, false) / 100;
+	else
+		return 100;
+}
 
-	/* The sclk is in quantas of 10kHz */
-	return rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100;
+int amdgpu_amdkfd_get_dmabuf_info(struct amdgpu_device *adev, int dma_buf_fd,
+				  struct amdgpu_device **dmabuf_adev,
+				  uint64_t *bo_size, void *metadata_buffer,
+				  size_t buffer_size, uint32_t *metadata_size,
+				  uint32_t *flags, int8_t *xcp_id)
+{
+	struct dma_buf *dma_buf;
+	struct drm_gem_object *obj;
+	struct amdgpu_bo *bo;
+	uint64_t metadata_flags;
+	int r = -EINVAL;
+
+	dma_buf = dma_buf_get(dma_buf_fd);
+	if (IS_ERR(dma_buf))
+		return PTR_ERR(dma_buf);
+
+	if (dma_buf->ops != &amdgpu_dmabuf_ops)
+		/* Can't handle non-graphics buffers */
+		goto out_put;
+
+	obj = dma_buf->priv;
+	if (obj->dev->driver != adev_to_drm(adev)->driver)
+		/* Can't handle buffers from different drivers */
+		goto out_put;
+
+	adev = drm_to_adev(obj->dev);
+	bo = gem_to_amdgpu_bo(obj);
+	if (!(bo->preferred_domains & (AMDGPU_GEM_DOMAIN_VRAM |
+				    AMDGPU_GEM_DOMAIN_GTT)))
+		/* Only VRAM and GTT BOs are supported */
+		goto out_put;
+
+	r = 0;
+	if (dmabuf_adev)
+		*dmabuf_adev = adev;
+	if (bo_size)
+		*bo_size = amdgpu_bo_size(bo);
+	if (metadata_buffer)
+		r = amdgpu_bo_get_metadata(bo, metadata_buffer, buffer_size,
+					   metadata_size, &metadata_flags);
+	if (flags) {
+		*flags = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) ?
+				KFD_IOC_ALLOC_MEM_FLAGS_VRAM
+				: KFD_IOC_ALLOC_MEM_FLAGS_GTT;
+
+		if (bo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
+			*flags |= KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC;
+	}
+	if (xcp_id)
+		*xcp_id = bo->xcp_id;
+
+out_put:
+	dma_buf_put(dma_buf);
+	return r;
+}
+
+int amdgpu_amdkfd_get_pcie_bandwidth_mbytes(struct amdgpu_device *adev, bool is_min)
+{
+	int num_lanes_shift = (is_min ? ffs(adev->pm.pcie_mlw_mask) :
+							fls(adev->pm.pcie_mlw_mask)) - 1;
+	int gen_speed_shift = (is_min ? ffs(adev->pm.pcie_gen_mask &
+						CAIL_PCIE_LINK_SPEED_SUPPORT_MASK) :
+					fls(adev->pm.pcie_gen_mask &
+						CAIL_PCIE_LINK_SPEED_SUPPORT_MASK)) - 1;
+	uint32_t num_lanes_mask = 1 << num_lanes_shift;
+	uint32_t gen_speed_mask = 1 << gen_speed_shift;
+	int num_lanes_factor = 0, gen_speed_mbits_factor = 0;
+
+	switch (num_lanes_mask) {
+	case CAIL_PCIE_LINK_WIDTH_SUPPORT_X1:
+		num_lanes_factor = 1;
+		break;
+	case CAIL_PCIE_LINK_WIDTH_SUPPORT_X2:
+		num_lanes_factor = 2;
+		break;
+	case CAIL_PCIE_LINK_WIDTH_SUPPORT_X4:
+		num_lanes_factor = 4;
+		break;
+	case CAIL_PCIE_LINK_WIDTH_SUPPORT_X8:
+		num_lanes_factor = 8;
+		break;
+	case CAIL_PCIE_LINK_WIDTH_SUPPORT_X12:
+		num_lanes_factor = 12;
+		break;
+	case CAIL_PCIE_LINK_WIDTH_SUPPORT_X16:
+		num_lanes_factor = 16;
+		break;
+	case CAIL_PCIE_LINK_WIDTH_SUPPORT_X32:
+		num_lanes_factor = 32;
+		break;
+	}
+
+	switch (gen_speed_mask) {
+	case CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1:
+		gen_speed_mbits_factor = 2500;
+		break;
+	case CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2:
+		gen_speed_mbits_factor = 5000;
+		break;
+	case CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3:
+		gen_speed_mbits_factor = 8000;
+		break;
+	case CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4:
+		gen_speed_mbits_factor = 16000;
+		break;
+	case CAIL_PCIE_LINK_SPEED_SUPPORT_GEN5:
+		gen_speed_mbits_factor = 32000;
+		break;
+	}
+
+	return (num_lanes_factor * gen_speed_mbits_factor)/BITS_PER_BYTE;
+}
+
+int amdgpu_amdkfd_submit_ib(struct amdgpu_device *adev,
+				enum kgd_engine_type engine,
+				uint32_t vmid, uint64_t gpu_addr,
+				uint32_t *ib_cmd, uint32_t ib_len)
+{
+	struct amdgpu_job *job;
+	struct amdgpu_ib *ib;
+	struct amdgpu_ring *ring;
+	struct dma_fence *f = NULL;
+	int ret;
+
+	switch (engine) {
+	case KGD_ENGINE_MEC1:
+		ring = &adev->gfx.compute_ring[0];
+		break;
+	case KGD_ENGINE_SDMA1:
+		ring = &adev->sdma.instance[0].ring;
+		break;
+	case KGD_ENGINE_SDMA2:
+		ring = &adev->sdma.instance[1].ring;
+		break;
+	default:
+		pr_err("Invalid engine in IB submission: %d\n", engine);
+		ret = -EINVAL;
+		goto err;
+	}
+
+	ret = amdgpu_job_alloc(adev, NULL, NULL, NULL, 1, &job, 0);
+	if (ret)
+		goto err;
+
+	ib = &job->ibs[0];
+	memset(ib, 0, sizeof(struct amdgpu_ib));
+
+	ib->gpu_addr = gpu_addr;
+	ib->ptr = ib_cmd;
+	ib->length_dw = ib_len;
+	/* This works for NO_HWS. TODO: need to handle without knowing VMID */
+	job->vmid = vmid;
+	job->num_ibs = 1;
+
+	ret = amdgpu_ib_schedule(ring, 1, ib, job, &f);
+
+	if (ret) {
+		DRM_ERROR("amdgpu: failed to schedule IB.\n");
+		goto err_ib_sched;
+	}
+
+	/* Drop the initial kref_init count (see drm_sched_main as example) */
+	dma_fence_put(f);
+	ret = dma_fence_wait(f, false);
+
+err_ib_sched:
+	amdgpu_job_free(job);
+err:
+	return ret;
+}
+
+void amdgpu_amdkfd_set_compute_idle(struct amdgpu_device *adev, bool idle)
+{
+	enum amd_powergating_state state = idle ? AMD_PG_STATE_GATE : AMD_PG_STATE_UNGATE;
+	if ((IP_VERSION_MAJ(amdgpu_ip_version(adev, GC_HWIP, 0)) == 11 &&
+	    ((adev->mes.kiq_version & AMDGPU_MES_VERSION_MASK) <= 64)) ||
+		(IP_VERSION_MAJ(amdgpu_ip_version(adev, GC_HWIP, 0)) == 12)) {
+		pr_debug("GFXOFF is %s\n", idle ? "enabled" : "disabled");
+		amdgpu_gfx_off_ctrl(adev, idle);
+	} else if ((IP_VERSION_MAJ(amdgpu_ip_version(adev, GC_HWIP, 0)) == 9) &&
+		(adev->flags & AMD_IS_APU)) {
+		/* Disable GFXOFF and PG. Temporary workaround
+		 * to fix some compute applications issue on GFX9.
+		 */
+		struct amdgpu_ip_block *gfx_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_GFX);
+		if (gfx_block != NULL)
+			gfx_block->version->funcs->set_powergating_state((void *)gfx_block, state);
+	}
+	amdgpu_dpm_switch_power_profile(adev,
+					PP_SMC_POWER_PROFILE_COMPUTE,
+					!idle);
+}
+
+bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
+{
+	if (adev->kfd.dev)
+		return vmid >= adev->vm_manager.first_kfd_vmid;
+
+	return false;
+}
+
+bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev)
+{
+	return adev->have_atomics_support;
+}
+
+void amdgpu_amdkfd_debug_mem_fence(struct amdgpu_device *adev)
+{
+	amdgpu_device_flush_hdp(adev, NULL);
+}
+
+bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev)
+{
+	return amdgpu_ras_get_fed_status(adev);
+}
+
+void amdgpu_amdkfd_ras_pasid_poison_consumption_handler(struct amdgpu_device *adev,
+				enum amdgpu_ras_block block, uint16_t pasid,
+				pasid_notify pasid_fn, void *data, uint32_t reset)
+{
+	amdgpu_umc_pasid_poison_handler(adev, block, pasid, pasid_fn, data, reset);
+}
+
+void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev,
+	enum amdgpu_ras_block block, uint32_t reset)
+{
+	amdgpu_umc_pasid_poison_handler(adev, block, 0, NULL, NULL, reset);
+}
+
+int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev,
+					uint32_t *payload)
+{
+	int ret;
+
+	/* Device or IH ring is not ready so bail. */
+	ret = amdgpu_ih_wait_on_checkpoint_process_ts(adev, &adev->irq.ih);
+	if (ret)
+		return ret;
+
+	/* Send payload to fence KFD interrupts */
+	amdgpu_amdkfd_interrupt(adev, payload);
+
+	return 0;
+}
+
+int amdgpu_amdkfd_check_and_lock_kfd(struct amdgpu_device *adev)
+{
+	return kgd2kfd_check_and_lock_kfd(adev->kfd.dev);
+}
+
+void amdgpu_amdkfd_unlock_kfd(struct amdgpu_device *adev)
+{
+	kgd2kfd_unlock_kfd(adev->kfd.dev);
+}
+
+
+u64 amdgpu_amdkfd_xcp_memory_size(struct amdgpu_device *adev, int xcp_id)
+{
+	s8 mem_id = KFD_XCP_MEM_ID(adev, xcp_id);
+	u64 tmp;
+
+	if (adev->gmc.num_mem_partitions && xcp_id >= 0 && mem_id >= 0) {
+		if (adev->gmc.is_app_apu && adev->gmc.num_mem_partitions == 1) {
+			/* In NPS1 mode, we should restrict the vram reporting
+			 * tied to the ttm_pages_limit which is 1/2 of the system
+			 * memory. For other partition modes, the HBM is uniformly
+			 * divided already per numa node reported. If user wants to
+			 * go beyond the default ttm limit and maximize the ROCm
+			 * allocations, they can go up to max ttm and sysmem limits.
+			 */
+
+			tmp = (ttm_tt_pages_limit() << PAGE_SHIFT) / num_online_nodes();
+		} else {
+			tmp = adev->gmc.mem_partitions[mem_id].size;
+		}
+		do_div(tmp, adev->xcp_mgr->num_xcp_per_mem_partition);
+		return ALIGN_DOWN(tmp, PAGE_SIZE);
+	} else if (adev->apu_prefer_gtt) {
+		return (ttm_tt_pages_limit() << PAGE_SHIFT);
+	} else {
+		return adev->gmc.real_vram_size;
+	}
+}
+
+int amdgpu_amdkfd_unmap_hiq(struct amdgpu_device *adev, u32 doorbell_off,
+			    u32 inst)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[inst];
+	struct amdgpu_ring *kiq_ring = &kiq->ring;
+	struct amdgpu_ring_funcs *ring_funcs;
+	struct amdgpu_ring *ring;
+	int r = 0;
+
+	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
+		return -EINVAL;
+
+	if (!kiq_ring->sched.ready || amdgpu_in_reset(adev))
+		return 0;
+
+	ring_funcs = kzalloc(sizeof(*ring_funcs), GFP_KERNEL);
+	if (!ring_funcs)
+		return -ENOMEM;
+
+	ring = kzalloc(sizeof(*ring), GFP_KERNEL);
+	if (!ring) {
+		r = -ENOMEM;
+		goto free_ring_funcs;
+	}
+
+	ring_funcs->type = AMDGPU_RING_TYPE_COMPUTE;
+	ring->doorbell_index = doorbell_off;
+	ring->funcs = ring_funcs;
+
+	spin_lock(&kiq->ring_lock);
+
+	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
+		spin_unlock(&kiq->ring_lock);
+		r = -ENOMEM;
+		goto free_ring;
+	}
+
+	kiq->pmf->kiq_unmap_queues(kiq_ring, ring, RESET_QUEUES, 0, 0);
+
+	/* Submit unmap queue packet */
+	amdgpu_ring_commit(kiq_ring);
+	/*
+	 * Ring test will do a basic scratch register change check. Just run
+	 * this to ensure that unmap queues that is submitted before got
+	 * processed successfully before returning.
+	 */
+	r = amdgpu_ring_test_helper(kiq_ring);
+
+	spin_unlock(&kiq->ring_lock);
+
+free_ring:
+	kfree(ring);
+
+free_ring_funcs:
+	kfree(ring_funcs);
+
+	return r;
+}
+
+/* Stop scheduling on KFD */
+int amdgpu_amdkfd_stop_sched(struct amdgpu_device *adev, uint32_t node_id)
+{
+	if (!adev->kfd.init_complete)
+		return 0;
+
+	return kgd2kfd_stop_sched(adev->kfd.dev, node_id);
+}
+
+/* Start scheduling on KFD */
+int amdgpu_amdkfd_start_sched(struct amdgpu_device *adev, uint32_t node_id)
+{
+	if (!adev->kfd.init_complete)
+		return 0;
+
+	return kgd2kfd_start_sched(adev->kfd.dev, node_id);
+}
+
+/* check if there are KFD queues active */
+bool amdgpu_amdkfd_compute_active(struct amdgpu_device *adev, uint32_t node_id)
+{
+	if (!adev->kfd.init_complete)
+		return false;
+
+	return kgd2kfd_compute_active(adev->kfd.dev, node_id);
+}
+
+/* Config CGTT_SQ_CLK_CTRL */
+int amdgpu_amdkfd_config_sq_perfmon(struct amdgpu_device *adev, uint32_t xcp_id,
+	bool core_override_enable, bool reg_override_enable, bool perfmon_override_enable)
+{
+	int r;
+
+	if (!adev->kfd.init_complete)
+		return 0;
+
+	r = psp_config_sq_perfmon(&adev->psp, xcp_id, core_override_enable,
+					reg_override_enable, perfmon_override_enable);
+
+	return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index de530f68d4e3..8bdfcde2029b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -25,41 +25,526 @@
 #ifndef AMDGPU_AMDKFD_H_INCLUDED
 #define AMDGPU_AMDKFD_H_INCLUDED
 
+#include <linux/list.h>
 #include <linux/types.h>
+#include <linux/mm.h>
+#include <linux/kthread.h>
+#include <linux/workqueue.h>
+#include <linux/mmu_notifier.h>
+#include <linux/memremap.h>
 #include <kgd_kfd_interface.h>
+#include <drm/drm_client.h>
+#include "amdgpu_sync.h"
+#include "amdgpu_vm.h"
+#include "amdgpu_xcp.h"
+
+extern uint64_t amdgpu_amdkfd_total_mem_size;
+
+enum TLB_FLUSH_TYPE {
+	TLB_FLUSH_LEGACY = 0,
+	TLB_FLUSH_LIGHTWEIGHT,
+	TLB_FLUSH_HEAVYWEIGHT
+};
 
 struct amdgpu_device;
+struct kfd_process_device;
+struct amdgpu_reset_context;
+
+enum kfd_mem_attachment_type {
+	KFD_MEM_ATT_SHARED,	/* Share kgd_mem->bo or another attachment's */
+	KFD_MEM_ATT_USERPTR,	/* SG bo to DMA map pages from a userptr bo */
+	KFD_MEM_ATT_DMABUF,	/* DMAbuf to DMA map TTM BOs */
+	KFD_MEM_ATT_SG		/* Tag to DMA map SG BOs */
+};
+
+struct kfd_mem_attachment {
+	struct list_head list;
+	enum kfd_mem_attachment_type type;
+	bool is_mapped;
+	struct amdgpu_bo_va *bo_va;
+	struct amdgpu_device *adev;
+	uint64_t va;
+	uint64_t pte_flags;
+};
 
 struct kgd_mem {
+	struct mutex lock;
 	struct amdgpu_bo *bo;
-	uint64_t gpu_addr;
-	void *cpu_ptr;
+	struct dma_buf *dmabuf;
+	struct amdgpu_hmm_range *range;
+	struct list_head attachments;
+	/* protected by amdkfd_process_info.lock */
+	struct list_head validate_list;
+	uint32_t domain;
+	unsigned int mapped_to_gpu_memory;
+	uint64_t va;
+
+	uint32_t alloc_flags;
+
+	uint32_t invalid;
+	struct amdkfd_process_info *process_info;
+
+	struct amdgpu_sync sync;
+
+	uint32_t gem_handle;
+	bool aql_queue;
+	bool is_imported;
+};
+
+/* KFD Memory Eviction */
+struct amdgpu_amdkfd_fence {
+	struct dma_fence base;
+	struct mm_struct *mm;
+	spinlock_t lock;
+	char timeline_name[TASK_COMM_LEN];
+	struct svm_range_bo *svm_bo;
+};
+
+struct amdgpu_kfd_dev {
+	struct kfd_dev *dev;
+	int64_t vram_used[MAX_XCP];
+	uint64_t vram_used_aligned[MAX_XCP];
+	bool init_complete;
+	struct work_struct reset_work;
+
+	/* Client for KFD BO GEM handle allocations */
+	struct drm_client_dev client;
+
+	/* HMM page migration MEMORY_DEVICE_PRIVATE mapping
+	 * Must be last --ends in a flexible-array member.
+	 */
+	struct dev_pagemap pgmap;
+};
+
+enum kgd_engine_type {
+	KGD_ENGINE_PFP = 1,
+	KGD_ENGINE_ME,
+	KGD_ENGINE_CE,
+	KGD_ENGINE_MEC1,
+	KGD_ENGINE_MEC2,
+	KGD_ENGINE_RLC,
+	KGD_ENGINE_SDMA1,
+	KGD_ENGINE_SDMA2,
+	KGD_ENGINE_MAX
+};
+
+
+struct amdkfd_process_info {
+	/* List head of all VMs that belong to a KFD process */
+	struct list_head vm_list_head;
+	/* List head for all KFD BOs that belong to a KFD process. */
+	struct list_head kfd_bo_list;
+	/* List of userptr BOs that are valid or invalid */
+	struct list_head userptr_valid_list;
+	struct list_head userptr_inval_list;
+	/* Lock to protect kfd_bo_list */
+	struct mutex lock;
+
+	/* Number of VMs */
+	unsigned int n_vms;
+	/* Eviction Fence */
+	struct amdgpu_amdkfd_fence *eviction_fence;
+
+	/* MMU-notifier related fields */
+	struct mutex notifier_lock;
+	uint32_t evicted_bos;
+	struct delayed_work restore_userptr_work;
+	struct pid *pid;
+	bool block_mmu_notifications;
 };
 
 int amdgpu_amdkfd_init(void);
 void amdgpu_amdkfd_fini(void);
 
-bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev);
-
-void amdgpu_amdkfd_suspend(struct amdgpu_device *rdev);
-int amdgpu_amdkfd_resume(struct amdgpu_device *rdev);
-void amdgpu_amdkfd_interrupt(struct amdgpu_device *rdev,
+void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool suspend_proc);
+int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool resume_proc);
+void amdgpu_amdkfd_suspend_process(struct amdgpu_device *adev);
+int amdgpu_amdkfd_resume_process(struct amdgpu_device *adev);
+void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,
 			const void *ih_ring_entry);
-void amdgpu_amdkfd_device_probe(struct amdgpu_device *rdev);
-void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev);
-void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev);
+void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev);
+void amdgpu_amdkfd_device_init(struct amdgpu_device *adev);
+void amdgpu_amdkfd_device_fini_sw(struct amdgpu_device *adev);
+int amdgpu_amdkfd_check_and_lock_kfd(struct amdgpu_device *adev);
+void amdgpu_amdkfd_unlock_kfd(struct amdgpu_device *adev);
+int amdgpu_amdkfd_submit_ib(struct amdgpu_device *adev,
+				enum kgd_engine_type engine,
+				uint32_t vmid, uint64_t gpu_addr,
+				uint32_t *ib_cmd, uint32_t ib_len);
+void amdgpu_amdkfd_set_compute_idle(struct amdgpu_device *adev, bool idle);
+bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev);
+
+bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid);
+
+int amdgpu_amdkfd_pre_reset(struct amdgpu_device *adev,
+			    struct amdgpu_reset_context *reset_context);
+
+int amdgpu_amdkfd_post_reset(struct amdgpu_device *adev);
+
+void amdgpu_amdkfd_gpu_reset(struct amdgpu_device *adev);
 
-struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void);
-struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void);
+int amdgpu_queue_mask_bit_to_set_resource_bit(struct amdgpu_device *adev,
+					int queue_bit);
 
+struct amdgpu_amdkfd_fence *amdgpu_amdkfd_fence_create(u64 context,
+				struct mm_struct *mm,
+				struct svm_range_bo *svm_bo);
+
+int amdgpu_amdkfd_drm_client_create(struct amdgpu_device *adev);
+#if defined(CONFIG_DEBUG_FS)
+int kfd_debugfs_kfd_mem_limits(struct seq_file *m, void *data);
+#endif
+#if IS_ENABLED(CONFIG_HSA_AMD)
+bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm);
+struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f);
+void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo);
+int amdgpu_amdkfd_evict_userptr(struct mmu_interval_notifier *mni,
+				unsigned long cur_seq, struct kgd_mem *mem);
+int amdgpu_amdkfd_bo_validate_and_fence(struct amdgpu_bo *bo,
+					uint32_t domain,
+					struct dma_fence *fence);
+#else
+static inline
+bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm)
+{
+	return false;
+}
+
+static inline
+struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f)
+{
+	return NULL;
+}
+
+static inline
+void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo)
+{
+}
+
+static inline
+int amdgpu_amdkfd_evict_userptr(struct mmu_interval_notifier *mni,
+				unsigned long cur_seq, struct kgd_mem *mem)
+{
+	return 0;
+}
+static inline
+int amdgpu_amdkfd_bo_validate_and_fence(struct amdgpu_bo *bo,
+					uint32_t domain,
+					struct dma_fence *fence)
+{
+	return 0;
+}
+#endif
 /* Shared API */
-int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
-			void **mem_obj, uint64_t *gpu_addr,
-			void **cpu_ptr);
-void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj);
-uint64_t get_vmem_size(struct kgd_dev *kgd);
-uint64_t get_gpu_clock_counter(struct kgd_dev *kgd);
+int amdgpu_amdkfd_alloc_gtt_mem(struct amdgpu_device *adev, size_t size,
+				void **mem_obj, uint64_t *gpu_addr,
+				void **cpu_ptr, bool mqd_gfx9);
+void amdgpu_amdkfd_free_gtt_mem(struct amdgpu_device *adev, void **mem_obj);
+int amdgpu_amdkfd_alloc_gws(struct amdgpu_device *adev, size_t size,
+				void **mem_obj);
+void amdgpu_amdkfd_free_gws(struct amdgpu_device *adev, void *mem_obj);
+int amdgpu_amdkfd_add_gws_to_process(void *info, void *gws, struct kgd_mem **mem);
+int amdgpu_amdkfd_remove_gws_from_process(void *info, void *mem);
+uint32_t amdgpu_amdkfd_get_fw_version(struct amdgpu_device *adev,
+				      enum kgd_engine_type type);
+void amdgpu_amdkfd_get_local_mem_info(struct amdgpu_device *adev,
+				      struct kfd_local_mem_info *mem_info,
+				      struct amdgpu_xcp *xcp);
+uint64_t amdgpu_amdkfd_get_gpu_clock_counter(struct amdgpu_device *adev);
+
+uint32_t amdgpu_amdkfd_get_max_engine_clock_in_mhz(struct amdgpu_device *adev);
+int amdgpu_amdkfd_get_dmabuf_info(struct amdgpu_device *adev, int dma_buf_fd,
+				  struct amdgpu_device **dmabuf_adev,
+				  uint64_t *bo_size, void *metadata_buffer,
+				  size_t buffer_size, uint32_t *metadata_size,
+				  uint32_t *flags, int8_t *xcp_id);
+int amdgpu_amdkfd_get_pcie_bandwidth_mbytes(struct amdgpu_device *adev, bool is_min);
+int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev,
+					uint32_t *payload);
+int amdgpu_amdkfd_unmap_hiq(struct amdgpu_device *adev, u32 doorbell_off,
+				u32 inst);
+int amdgpu_amdkfd_start_sched(struct amdgpu_device *adev, uint32_t node_id);
+int amdgpu_amdkfd_stop_sched(struct amdgpu_device *adev, uint32_t node_id);
+int amdgpu_amdkfd_config_sq_perfmon(struct amdgpu_device *adev, uint32_t xcp_id,
+	bool core_override_enable, bool reg_override_enable, bool perfmon_override_enable);
+bool amdgpu_amdkfd_compute_active(struct amdgpu_device *adev, uint32_t node_id);
+
+
+/* Read user wptr from a specified user address space with page fault
+ * disabled. The memory must be pinned and mapped to the hardware when
+ * this is called in hqd_load functions, so it should never fault in
+ * the first place. This resolves a circular lock dependency involving
+ * four locks, including the DQM lock and mmap_lock.
+ */
+#define read_user_wptr(mmptr, wptr, dst)				\
+	({								\
+		bool valid = false;					\
+		if ((mmptr) && (wptr)) {				\
+			pagefault_disable();				\
+			if ((mmptr) == current->mm) {			\
+				valid = !get_user((dst), (wptr));	\
+			} else if (current->flags & PF_KTHREAD) {	\
+				kthread_use_mm(mmptr);			\
+				valid = !get_user((dst), (wptr));	\
+				kthread_unuse_mm(mmptr);		\
+			}						\
+			pagefault_enable();				\
+		}							\
+		valid;							\
+	})
+
+/* GPUVM API */
+#define drm_priv_to_vm(drm_priv)					\
+	(&((struct amdgpu_fpriv *)					\
+		((struct drm_file *)(drm_priv))->driver_priv)->vm)
+
+int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,
+					struct amdgpu_vm *avm,
+					void **process_info,
+					struct dma_fence **ef);
+uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv);
+size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev,
+					uint8_t xcp_id);
+int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
+		struct amdgpu_device *adev, uint64_t va, uint64_t size,
+		void *drm_priv, struct kgd_mem **mem,
+		uint64_t *offset, uint32_t flags, bool criu_resume);
+int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
+		struct amdgpu_device *adev, struct kgd_mem *mem, void *drm_priv,
+		uint64_t *size);
+int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(struct amdgpu_device *adev,
+					  struct kgd_mem *mem, void *drm_priv);
+int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
+		struct amdgpu_device *adev, struct kgd_mem *mem, void *drm_priv);
+int amdgpu_amdkfd_gpuvm_dmaunmap_mem(struct kgd_mem *mem, void *drm_priv);
+int amdgpu_amdkfd_gpuvm_sync_memory(
+		struct amdgpu_device *adev, struct kgd_mem *mem, bool intr);
+int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_mem *mem,
+					     void **kptr, uint64_t *size);
+void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_mem *mem);
+
+int amdgpu_amdkfd_map_gtt_bo_to_gart(struct amdgpu_bo *bo, struct amdgpu_bo **bo_gart);
+
+int amdgpu_amdkfd_gpuvm_restore_process_bos(void *process_info,
+					    struct dma_fence __rcu **ef);
+int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
+					      struct kfd_vm_fault_info *info);
+int amdgpu_amdkfd_gpuvm_import_dmabuf_fd(struct amdgpu_device *adev, int fd,
+					 uint64_t va, void *drm_priv,
+					 struct kgd_mem **mem, uint64_t *size,
+					 uint64_t *mmap_offset);
+int amdgpu_amdkfd_gpuvm_export_dmabuf(struct kgd_mem *mem,
+				      struct dma_buf **dmabuf);
+void amdgpu_amdkfd_debug_mem_fence(struct amdgpu_device *adev);
+int amdgpu_amdkfd_get_tile_config(struct amdgpu_device *adev,
+				struct tile_config *config);
+void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev,
+			enum amdgpu_ras_block block, uint32_t reset);
+
+void amdgpu_amdkfd_ras_pasid_poison_consumption_handler(struct amdgpu_device *adev,
+			enum amdgpu_ras_block block, uint16_t pasid,
+			pasid_notify pasid_fn, void *data, uint32_t reset);
+
+bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev);
+bool amdgpu_amdkfd_bo_mapped_to_dev(void *drm_priv, struct kgd_mem *mem);
+void amdgpu_amdkfd_block_mmu_notifications(void *p);
+int amdgpu_amdkfd_criu_resume(void *p);
+int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
+		uint64_t size, u32 alloc_flag, int8_t xcp_id);
+void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
+		uint64_t size, u32 alloc_flag, int8_t xcp_id);
+
+u64 amdgpu_amdkfd_xcp_memory_size(struct amdgpu_device *adev, int xcp_id);
+
+#define KFD_XCP_MEM_ID(adev, xcp_id) \
+		((adev)->xcp_mgr && (xcp_id) >= 0 ?\
+		(adev)->xcp_mgr->xcp[(xcp_id)].mem_id : -1)
+
+#define KFD_XCP_MEMORY_SIZE(adev, xcp_id) amdgpu_amdkfd_xcp_memory_size((adev), (xcp_id))
+
+
+#if IS_ENABLED(CONFIG_HSA_AMD)
+void amdgpu_amdkfd_gpuvm_init_mem_limits(void);
+void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
+				struct amdgpu_vm *vm);
+
+/**
+ * @amdgpu_amdkfd_release_notify() - Notify KFD when GEM object is released
+ *
+ * Allows KFD to release its resources associated with the GEM object.
+ */
+void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo);
+void amdgpu_amdkfd_reserve_system_mem(uint64_t size);
+#else
+static inline
+void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
+{
+}
+
+static inline
+void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
+					struct amdgpu_vm *vm)
+{
+}
+
+static inline
+void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo)
+{
+}
+#endif
+
+#if IS_ENABLED(CONFIG_HSA_AMD_SVM)
+int kgd2kfd_init_zone_device(struct amdgpu_device *adev);
+#else
+static inline
+int kgd2kfd_init_zone_device(struct amdgpu_device *adev)
+{
+	return 0;
+}
+#endif
+
+/* KGD2KFD callbacks */
+int kgd2kfd_quiesce_mm(struct mm_struct *mm, uint32_t trigger);
+int kgd2kfd_resume_mm(struct mm_struct *mm);
+int kgd2kfd_schedule_evict_and_restore_process(struct mm_struct *mm,
+						struct dma_fence *fence);
+#if IS_ENABLED(CONFIG_HSA_AMD)
+int kgd2kfd_init(void);
+void kgd2kfd_exit(void);
+struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf);
+bool kgd2kfd_device_init(struct kfd_dev *kfd,
+			 const struct kgd2kfd_shared_resources *gpu_resources);
+void kgd2kfd_device_exit(struct kfd_dev *kfd);
+void kgd2kfd_suspend(struct kfd_dev *kfd, bool suspend_proc);
+int kgd2kfd_resume(struct kfd_dev *kfd, bool resume_proc);
+void kgd2kfd_suspend_process(struct kfd_dev *kfd);
+int kgd2kfd_resume_process(struct kfd_dev *kfd);
+int kgd2kfd_pre_reset(struct kfd_dev *kfd,
+		      struct amdgpu_reset_context *reset_context);
+int kgd2kfd_post_reset(struct kfd_dev *kfd);
+void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry);
+void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd);
+void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask);
+int kgd2kfd_check_and_lock_kfd(struct kfd_dev *kfd);
+void kgd2kfd_unlock_kfd(struct kfd_dev *kfd);
+int kgd2kfd_start_sched(struct kfd_dev *kfd, uint32_t node_id);
+int kgd2kfd_start_sched_all_nodes(struct kfd_dev *kfd);
+int kgd2kfd_stop_sched(struct kfd_dev *kfd, uint32_t node_id);
+int kgd2kfd_stop_sched_all_nodes(struct kfd_dev *kfd);
+bool kgd2kfd_compute_active(struct kfd_dev *kfd, uint32_t node_id);
+bool kgd2kfd_vmfault_fast_path(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry,
+			       bool retry_fault);
+
+#else
+static inline int kgd2kfd_init(void)
+{
+	return -ENOENT;
+}
+
+static inline void kgd2kfd_exit(void)
+{
+}
+
+static inline
+struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf)
+{
+	return NULL;
+}
+
+static inline
+bool kgd2kfd_device_init(struct kfd_dev *kfd,
+				const struct kgd2kfd_shared_resources *gpu_resources)
+{
+	return false;
+}
+
+static inline void kgd2kfd_device_exit(struct kfd_dev *kfd)
+{
+}
+
+static inline void kgd2kfd_suspend(struct kfd_dev *kfd, bool suspend_proc)
+{
+}
+
+static inline int kgd2kfd_resume(struct kfd_dev *kfd, bool resume_proc)
+{
+	return 0;
+}
+
+static inline void kgd2kfd_suspend_process(struct kfd_dev *kfd)
+{
+}
+
+static inline int kgd2kfd_resume_process(struct kfd_dev *kfd)
+{
+	return 0;
+}
+
+static inline int kgd2kfd_pre_reset(struct kfd_dev *kfd,
+				    struct amdgpu_reset_context *reset_context)
+{
+	return 0;
+}
+
+static inline int kgd2kfd_post_reset(struct kfd_dev *kfd)
+{
+	return 0;
+}
+
+static inline
+void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
+{
+}
+
+static inline
+void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd)
+{
+}
+
+static inline
+void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask)
+{
+}
+
+static inline int kgd2kfd_check_and_lock_kfd(struct kfd_dev *kfd)
+{
+	return 0;
+}
+
+static inline void kgd2kfd_unlock_kfd(struct kfd_dev *kfd)
+{
+}
+
+static inline int kgd2kfd_start_sched(struct kfd_dev *kfd, uint32_t node_id)
+{
+	return 0;
+}
+
+static inline int kgd2kfd_start_sched_all_nodes(struct kfd_dev *kfd)
+{
+	return 0;
+}
+
+static inline int kgd2kfd_stop_sched(struct kfd_dev *kfd, uint32_t node_id)
+{
+	return 0;
+}
+
+static inline int kgd2kfd_stop_sched_all_nodes(struct kfd_dev *kfd)
+{
+	return 0;
+}
+
+static inline bool kgd2kfd_compute_active(struct kfd_dev *kfd, uint32_t node_id)
+{
+	return false;
+}
 
-uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd);
+static inline bool kgd2kfd_vmfault_fast_path(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry,
+				      bool retry_fault)
+{
+	return false;
+}
 
+#endif
 #endif /* AMDGPU_AMDKFD_H_INCLUDED */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
new file mode 100644
index 000000000000..7e9f7a280c1b
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
@@ -0,0 +1,197 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_amdkfd_arcturus.h"
+#include "amdgpu_amdkfd_gfx_v9.h"
+#include "amdgpu_amdkfd_aldebaran.h"
+#include "gc/gc_9_4_2_offset.h"
+#include "gc/gc_9_4_2_sh_mask.h"
+#include <uapi/linux/kfd_ioctl.h>
+
+/*
+ * Returns TRAP_EN, EXCP_EN and EXCP_REPLACE.
+ *
+ * restore_dbg_registers is ignored here but is a general interface requirement
+ * for devices that support GFXOFF and where the RLC save/restore list
+ * does not support hw registers for debugging i.e. the driver has to manually
+ * initialize the debug mode registers after it has disabled GFX off during the
+ * debug session.
+ */
+uint32_t kgd_aldebaran_enable_debug_trap(struct amdgpu_device *adev,
+					    bool restore_dbg_registers,
+					    uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
+
+	return data;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_aldebaran_disable_debug_trap(struct amdgpu_device *adev,
+						bool keep_trap_enabled,
+						uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, keep_trap_enabled);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
+
+	return data;
+}
+
+static int kgd_aldebaran_validate_trap_override_request(struct amdgpu_device *adev,
+							uint32_t trap_override,
+							uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_FP_INVALID |
+				KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+				KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+				KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+				KFD_DBG_TRAP_MASK_FP_INEXACT |
+				KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+				KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION;
+
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
+			trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
+		return -EPERM;
+
+	return 0;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_RPLACE. */
+static uint32_t kgd_aldebaran_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					uint32_t vmid,
+					uint32_t trap_override,
+					uint32_t trap_mask_bits,
+					uint32_t trap_mask_request,
+					uint32_t *trap_mask_prev,
+					uint32_t kfd_dbg_trap_cntl_prev)
+
+{
+	uint32_t data = 0;
+
+	*trap_mask_prev = REG_GET_FIELD(kfd_dbg_trap_cntl_prev, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
+	trap_mask_bits = (trap_mask_bits & trap_mask_request) |
+		(*trap_mask_prev & ~trap_mask_request);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, trap_mask_bits);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
+
+	return data;
+}
+
+uint32_t kgd_aldebaran_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode);
+
+	return data;
+}
+
+#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
+static uint32_t kgd_gfx_aldebaran_set_address_watch(
+					struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid,
+					uint32_t inst)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 6);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	return watch_address_cntl;
+}
+
+const struct kfd2kgd_calls aldebaran_kfd2kgd = {
+	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_gfx_v9_init_interrupts,
+	.hqd_load = kgd_gfx_v9_hqd_load,
+	.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
+	.hqd_sdma_load = kgd_arcturus_hqd_sdma_load,
+	.hqd_dump = kgd_gfx_v9_hqd_dump,
+	.hqd_sdma_dump = kgd_arcturus_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_arcturus_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_gfx_v9_hqd_destroy,
+	.hqd_sdma_destroy = kgd_arcturus_hqd_sdma_destroy,
+	.wave_control_execute = kgd_gfx_v9_wave_control_execute,
+	.get_atc_vmid_pasid_mapping_info =
+				kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
+	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
+	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
+	.enable_debug_trap = kgd_aldebaran_enable_debug_trap,
+	.disable_debug_trap = kgd_aldebaran_disable_debug_trap,
+	.validate_trap_override_request = kgd_aldebaran_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_aldebaran_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_aldebaran_set_address_watch,
+	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
+	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
+	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
+	.hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr,
+	.hqd_reset = kgd_gfx_v9_hqd_reset,
+	.hqd_sdma_get_doorbell = kgd_gfx_v9_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h
new file mode 100644
index 000000000000..a7bdaf8d82dd
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h
@@ -0,0 +1,27 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+uint32_t kgd_aldebaran_enable_debug_trap(struct amdgpu_device *adev,
+					bool restore_dbg_registers,
+					uint32_t vmid);
+uint32_t kgd_aldebaran_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
new file mode 100644
index 000000000000..1105a09e55dc
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
@@ -0,0 +1,424 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include <linux/module.h>
+#include <linux/uaccess.h>
+#include <linux/firmware.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_amdkfd_arcturus.h"
+#include "amdgpu_reset.h"
+#include "sdma0/sdma0_4_2_2_offset.h"
+#include "sdma0/sdma0_4_2_2_sh_mask.h"
+#include "sdma1/sdma1_4_2_2_offset.h"
+#include "sdma1/sdma1_4_2_2_sh_mask.h"
+#include "sdma2/sdma2_4_2_2_offset.h"
+#include "sdma2/sdma2_4_2_2_sh_mask.h"
+#include "sdma3/sdma3_4_2_2_offset.h"
+#include "sdma3/sdma3_4_2_2_sh_mask.h"
+#include "sdma4/sdma4_4_2_2_offset.h"
+#include "sdma4/sdma4_4_2_2_sh_mask.h"
+#include "sdma5/sdma5_4_2_2_offset.h"
+#include "sdma5/sdma5_4_2_2_sh_mask.h"
+#include "sdma6/sdma6_4_2_2_offset.h"
+#include "sdma6/sdma6_4_2_2_sh_mask.h"
+#include "sdma7/sdma7_4_2_2_offset.h"
+#include "sdma7/sdma7_4_2_2_sh_mask.h"
+#include "v9_structs.h"
+#include "soc15.h"
+#include "soc15d.h"
+#include "amdgpu_amdkfd_gfx_v9.h"
+#include "gfxhub_v1_0.h"
+#include "mmhub_v9_4.h"
+#include "gc/gc_9_0_offset.h"
+#include "gc/gc_9_0_sh_mask.h"
+
+#define HQD_N_REGS 56
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v9_sdma_mqd *)mqd;
+}
+
+static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
+				unsigned int engine_id,
+				unsigned int queue_id)
+{
+	uint32_t sdma_engine_reg_base = 0;
+	uint32_t sdma_rlc_reg_offset;
+
+	switch (engine_id) {
+	default:
+		dev_warn(adev->dev,
+			 "Invalid sdma engine id (%d), using engine id 0\n",
+			 engine_id);
+		fallthrough;
+	case 0:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
+		break;
+	case 1:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA1, 0,
+				mmSDMA1_RLC0_RB_CNTL) - mmSDMA1_RLC0_RB_CNTL;
+		break;
+	case 2:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA2, 0,
+				mmSDMA2_RLC0_RB_CNTL) - mmSDMA2_RLC0_RB_CNTL;
+		break;
+	case 3:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA3, 0,
+				mmSDMA3_RLC0_RB_CNTL) - mmSDMA3_RLC0_RB_CNTL;
+		break;
+	case 4:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA4, 0,
+				mmSDMA4_RLC0_RB_CNTL) - mmSDMA4_RLC0_RB_CNTL;
+		break;
+	case 5:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA5, 0,
+				mmSDMA5_RLC0_RB_CNTL) - mmSDMA5_RLC0_RB_CNTL;
+		break;
+	case 6:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA6, 0,
+				mmSDMA6_RLC0_RB_CNTL) - mmSDMA6_RLC0_RB_CNTL;
+		break;
+	case 7:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA7, 0,
+				mmSDMA7_RLC0_RB_CNTL) - mmSDMA7_RLC0_RB_CNTL;
+		break;
+	}
+
+	sdma_rlc_reg_offset = sdma_engine_reg_base
+		+ queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL);
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
+			queue_id, sdma_rlc_reg_offset);
+
+	return sdma_rlc_reg_offset;
+}
+
+int kgd_arcturus_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	unsigned long end_jiffies;
+	uint32_t data;
+	uint64_t data64;
+	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET,
+	       m->sdmax_rlcx_doorbell_offset);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
+				m->sdmax_rlcx_rb_rptr);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI,
+				m->sdmax_rlcx_rb_rptr_hi);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
+	if (read_user_wptr(mm, wptr64, data64)) {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       lower_32_bits(data64));
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       upper_32_bits(data64));
+	} else {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       m->sdmax_rlcx_rb_rptr_hi);
+	}
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+int kgd_arcturus_hqd_sdma_dump(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
+			engine_id, queue_id);
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+6+7+10)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
+	     reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
+	     reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+bool kgd_arcturus_hqd_sdma_is_occupied(struct amdgpu_device *adev,
+				void *mqd)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+int kgd_arcturus_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
+				unsigned int utimeout)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t temp;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
+	m->sdmax_rlcx_rb_rptr_hi =
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI);
+
+	return 0;
+}
+
+/*
+ * Helper used to suspend/resume gfx pipe for image post process work to set
+ * barrier behaviour.
+ */
+static int suspend_resume_compute_scheduler(struct amdgpu_device *adev, bool suspend)
+{
+	int i, r = 0;
+
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
+
+		if (!amdgpu_ring_sched_ready(ring))
+			continue;
+
+		/* stop secheduler and drain ring. */
+		if (suspend) {
+			drm_sched_stop(&ring->sched, NULL);
+			r = amdgpu_fence_wait_empty(ring);
+			if (r)
+				goto out;
+		} else {
+			drm_sched_start(&ring->sched, 0);
+		}
+	}
+
+out:
+	/* return on resume or failure to drain rings. */
+	if (!suspend || r)
+		return r;
+
+	return amdgpu_device_ip_wait_for_idle(adev, AMD_IP_BLOCK_TYPE_GFX);
+}
+
+static void set_barrier_auto_waitcnt(struct amdgpu_device *adev, bool enable_waitcnt)
+{
+	uint32_t data;
+
+	WRITE_ONCE(adev->barrier_has_auto_waitcnt, enable_waitcnt);
+
+	if (!down_read_trylock(&adev->reset_domain->sem))
+		return;
+
+	amdgpu_amdkfd_suspend(adev, true);
+
+	if (suspend_resume_compute_scheduler(adev, true))
+		goto out;
+
+	data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_CONFIG));
+	data = REG_SET_FIELD(data, SQ_CONFIG, DISABLE_BARRIER_WAITCNT,
+						!enable_waitcnt);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_CONFIG), data);
+
+out:
+	suspend_resume_compute_scheduler(adev, false);
+
+	amdgpu_amdkfd_resume(adev, true);
+
+	up_read(&adev->reset_domain->sem);
+}
+
+/*
+ * restore_dbg_registers is ignored here but is a general interface requirement
+ * for devices that support GFXOFF and where the RLC save/restore list
+ * does not support hw registers for debugging i.e. the driver has to manually
+ * initialize the debug mode registers after it has disabled GFX off during the
+ * debug session.
+ */
+static uint32_t kgd_arcturus_enable_debug_trap(struct amdgpu_device *adev,
+				bool restore_dbg_registers,
+				uint32_t vmid)
+{
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	set_barrier_auto_waitcnt(adev, true);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+/*
+ * keep_trap_enabled is ignored here but is a general interface requirement
+ * for devices that support multi-process debugging where the performance
+ * overhead from trap temporary setup needs to be bypassed when the debug
+ * session has ended.
+ */
+static uint32_t kgd_arcturus_disable_debug_trap(struct amdgpu_device *adev,
+					bool keep_trap_enabled,
+					uint32_t vmid)
+{
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	set_barrier_auto_waitcnt(adev, false);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+const struct kfd2kgd_calls arcturus_kfd2kgd = {
+	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_gfx_v9_init_interrupts,
+	.hqd_load = kgd_gfx_v9_hqd_load,
+	.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
+	.hqd_sdma_load = kgd_arcturus_hqd_sdma_load,
+	.hqd_dump = kgd_gfx_v9_hqd_dump,
+	.hqd_sdma_dump = kgd_arcturus_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_arcturus_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_gfx_v9_hqd_destroy,
+	.hqd_sdma_destroy = kgd_arcturus_hqd_sdma_destroy,
+	.wave_control_execute = kgd_gfx_v9_wave_control_execute,
+	.get_atc_vmid_pasid_mapping_info =
+				kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
+	.set_vm_context_page_table_base =
+				kgd_gfx_v9_set_vm_context_page_table_base,
+	.enable_debug_trap = kgd_arcturus_enable_debug_trap,
+	.disable_debug_trap = kgd_arcturus_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v9_set_address_watch,
+	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
+	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
+	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
+	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
+	.hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr,
+	.hqd_reset = kgd_gfx_v9_hqd_reset,
+	.hqd_sdma_get_doorbell = kgd_gfx_v9_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.h
new file mode 100644
index 000000000000..756c1a5679c0
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.h
@@ -0,0 +1,31 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+int kgd_arcturus_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm);
+int kgd_arcturus_hqd_sdma_dump(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs);
+bool kgd_arcturus_hqd_sdma_is_occupied(struct amdgpu_device *adev,
+				void *mqd);
+int kgd_arcturus_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
+				unsigned int utimeout);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c
new file mode 100644
index 000000000000..1ef758ac5076
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c
@@ -0,0 +1,188 @@
+/*
+ * Copyright 2016-2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <linux/dma-fence.h>
+#include <linux/spinlock.h>
+#include <linux/atomic.h>
+#include <linux/stacktrace.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/sched/mm.h>
+#include "amdgpu_amdkfd.h"
+#include "kfd_svm.h"
+
+static const struct dma_fence_ops amdkfd_fence_ops;
+static atomic_t fence_seq = ATOMIC_INIT(0);
+
+/* Eviction Fence
+ * Fence helper functions to deal with KFD memory eviction.
+ * Big Idea - Since KFD submissions are done by user queues, a BO cannot be
+ *  evicted unless all the user queues for that process are evicted.
+ *
+ * All the BOs in a process share an eviction fence. When process X wants
+ * to map VRAM memory but TTM can't find enough space, TTM will attempt to
+ * evict BOs from its LRU list. TTM checks if the BO is valuable to evict
+ * by calling ttm_device_funcs->eviction_valuable().
+ *
+ * ttm_device_funcs->eviction_valuable() - will return false if the BO belongs
+ *  to process X. Otherwise, it will return true to indicate BO can be
+ *  evicted by TTM.
+ *
+ * If ttm_device_funcs->eviction_valuable returns true, then TTM will continue
+ * the evcition process for that BO by calling ttm_bo_evict --> amdgpu_bo_move
+ * --> amdgpu_copy_buffer(). This sets up job in GPU scheduler.
+ *
+ * GPU Scheduler (amd_sched_main) - sets up a cb (fence_add_callback) to
+ *  nofity when the BO is free to move. fence_add_callback --> enable_signaling
+ *  --> amdgpu_amdkfd_fence.enable_signaling
+ *
+ * amdgpu_amdkfd_fence.enable_signaling - Start a work item that will quiesce
+ * user queues and signal fence. The work item will also start another delayed
+ * work item to restore BOs
+ */
+
+struct amdgpu_amdkfd_fence *amdgpu_amdkfd_fence_create(u64 context,
+				struct mm_struct *mm,
+				struct svm_range_bo *svm_bo)
+{
+	struct amdgpu_amdkfd_fence *fence;
+
+	fence = kzalloc(sizeof(*fence), GFP_KERNEL);
+	if (fence == NULL)
+		return NULL;
+
+	/* This reference gets released in amdkfd_fence_release */
+	mmgrab(mm);
+	fence->mm = mm;
+	get_task_comm(fence->timeline_name, current);
+	spin_lock_init(&fence->lock);
+	fence->svm_bo = svm_bo;
+	dma_fence_init(&fence->base, &amdkfd_fence_ops, &fence->lock,
+		   context, atomic_inc_return(&fence_seq));
+
+	return fence;
+}
+
+struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f)
+{
+	struct amdgpu_amdkfd_fence *fence;
+
+	if (!f)
+		return NULL;
+
+	fence = container_of(f, struct amdgpu_amdkfd_fence, base);
+	if (f->ops == &amdkfd_fence_ops)
+		return fence;
+
+	return NULL;
+}
+
+static const char *amdkfd_fence_get_driver_name(struct dma_fence *f)
+{
+	return "amdgpu_amdkfd_fence";
+}
+
+static const char *amdkfd_fence_get_timeline_name(struct dma_fence *f)
+{
+	struct amdgpu_amdkfd_fence *fence = to_amdgpu_amdkfd_fence(f);
+
+	return fence->timeline_name;
+}
+
+/**
+ * amdkfd_fence_enable_signaling - This gets called when TTM wants to evict
+ *  a KFD BO and schedules a job to move the BO.
+ *  If fence is already signaled return true.
+ *  If fence is not signaled schedule a evict KFD process work item.
+ *
+ *  @f: dma_fence
+ */
+static bool amdkfd_fence_enable_signaling(struct dma_fence *f)
+{
+	struct amdgpu_amdkfd_fence *fence = to_amdgpu_amdkfd_fence(f);
+
+	if (!fence)
+		return false;
+
+	if (dma_fence_is_signaled(f))
+		return true;
+
+	if (!fence->svm_bo) {
+		if (!kgd2kfd_schedule_evict_and_restore_process(fence->mm, f))
+			return true;
+	} else {
+		if (!svm_range_schedule_evict_svm_bo(fence))
+			return true;
+	}
+	return false;
+}
+
+/**
+ * amdkfd_fence_release - callback that fence can be freed
+ *
+ * @f: dma_fence
+ *
+ * This function is called when the reference count becomes zero.
+ * Drops the mm_struct reference and RCU schedules freeing up the fence.
+ */
+static void amdkfd_fence_release(struct dma_fence *f)
+{
+	struct amdgpu_amdkfd_fence *fence = to_amdgpu_amdkfd_fence(f);
+
+	/* Unconditionally signal the fence. The process is getting
+	 * terminated.
+	 */
+	if (WARN_ON(!fence))
+		return; /* Not an amdgpu_amdkfd_fence */
+
+	mmdrop(fence->mm);
+	kfree_rcu(f, rcu);
+}
+
+/**
+ * amdkfd_fence_check_mm - Check whether to prevent eviction of @f by @mm
+ *
+ * @f: [IN] fence
+ * @mm: [IN] mm that needs to be verified
+ *
+ * Check if @mm is same as that of the fence @f, if same return TRUE else
+ * return FALSE.
+ * For svm bo, which support vram overcommitment, always return FALSE.
+ */
+bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm)
+{
+	struct amdgpu_amdkfd_fence *fence = to_amdgpu_amdkfd_fence(f);
+
+	if (!fence)
+		return false;
+	else if (fence->mm == mm  && !fence->svm_bo)
+		return true;
+
+	return false;
+}
+
+static const struct dma_fence_ops amdkfd_fence_ops = {
+	.get_driver_name = amdkfd_fence_get_driver_name,
+	.get_timeline_name = amdkfd_fence_get_timeline_name,
+	.enable_signaling = amdkfd_fence_enable_signaling,
+	.release = amdkfd_fence_release,
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c
new file mode 100644
index 000000000000..89a45a9218f3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c
@@ -0,0 +1,559 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_amdkfd_gfx_v9.h"
+#include "amdgpu_amdkfd_aldebaran.h"
+#include "gc/gc_9_4_3_offset.h"
+#include "gc/gc_9_4_3_sh_mask.h"
+#include "athub/athub_1_8_0_offset.h"
+#include "athub/athub_1_8_0_sh_mask.h"
+#include "oss/osssys_4_4_2_offset.h"
+#include "oss/osssys_4_4_2_sh_mask.h"
+#include "v9_structs.h"
+#include "soc15.h"
+#include "sdma/sdma_4_4_2_offset.h"
+#include "sdma/sdma_4_4_2_sh_mask.h"
+#include <uapi/linux/kfd_ioctl.h>
+
+static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v9_sdma_mqd *)mqd;
+}
+
+static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
+					unsigned int engine_id,
+					unsigned int queue_id)
+{
+	uint32_t sdma_engine_reg_base =
+		SOC15_REG_OFFSET(SDMA0, GET_INST(SDMA0, engine_id),
+				 regSDMA_RLC0_RB_CNTL) -
+		regSDMA_RLC0_RB_CNTL;
+	uint32_t retval = sdma_engine_reg_base +
+		  queue_id * (regSDMA_RLC1_RB_CNTL - regSDMA_RLC0_RB_CNTL);
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
+							queue_id, retval);
+	return retval;
+}
+
+static int kgd_gfx_v9_4_3_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
+				 uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	unsigned long end_jiffies;
+	uint32_t data;
+	uint64_t data64;
+	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+							m->sdma_queue_id);
+
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_CONTEXT_STATUS);
+		if (data & SDMA_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_DOORBELL_OFFSET,
+		m->sdmax_rlcx_doorbell_offset);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA_RLC0_DOORBELL,
+				ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR,
+					m->sdmax_rlcx_rb_rptr);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_HI,
+					m->sdmax_rlcx_rb_rptr_hi);
+
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_MINOR_PTR_UPDATE, 1);
+	if (read_user_wptr(mm, wptr64, data64)) {
+		WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR,
+			lower_32_bits(data64));
+		WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR_HI,
+			upper_32_bits(data64));
+	} else {
+		WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR,
+			m->sdmax_rlcx_rb_rptr);
+		WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_WPTR_HI,
+			m->sdmax_rlcx_rb_rptr_hi);
+	}
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_MINOR_PTR_UPDATE, 0);
+
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA_RLC0_RB_CNTL,
+				RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_gfx_v9_4_3_hqd_sdma_dump(struct amdgpu_device *adev,
+				 uint32_t engine_id, uint32_t queue_id,
+				 uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
+							engine_id, queue_id);
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+6+7+12)
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))      \
+			break;				\
+		(*dump)[i][0] = (addr) << 2;            \
+		(*dump)[i++][1] = RREG32(addr);         \
+	} while (0)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = regSDMA_RLC0_RB_CNTL; reg <= regSDMA_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = regSDMA_RLC0_STATUS; reg <= regSDMA_RLC0_CSA_ADDR_HI; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = regSDMA_RLC0_IB_SUB_REMAIN;
+	     reg <= regSDMA_RLC0_MINOR_PTR_UPDATE; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = regSDMA_RLC0_MIDCMD_DATA0;
+	     reg <= regSDMA_RLC0_MIDCMD_CNTL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static bool kgd_gfx_v9_4_3_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+							m->sdma_queue_id);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+static int kgd_gfx_v9_4_3_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
+				    unsigned int utimeout)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t temp;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+							m->sdma_queue_id);
+
+	temp = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL);
+	temp = temp & ~SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_CNTL) |
+		SDMA_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr =
+			RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR);
+	m->sdmax_rlcx_rb_rptr_hi =
+			RREG32(sdma_rlc_reg_offset + regSDMA_RLC0_RB_RPTR_HI);
+
+	return 0;
+}
+
+static int kgd_gfx_v9_4_3_set_pasid_vmid_mapping(struct amdgpu_device *adev,
+			u32 pasid, unsigned int vmid, uint32_t xcc_inst)
+{
+	unsigned long timeout;
+	unsigned int reg;
+	unsigned int phy_inst = GET_INST(GC, xcc_inst);
+	/* Every two XCCs share one AID */
+	unsigned int aid = phy_inst / 2;
+
+	/*
+	 * We have to assume that there is no outstanding mapping.
+	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
+	 * a mapping is in progress or because a mapping finished
+	 * and the SW cleared it.
+	 * So the protocol is to always wait & clear.
+	 */
+	uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
+			ATC_VMID0_PASID_MAPPING__VALID_MASK;
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0,
+		regATC_VMID0_PASID_MAPPING) + vmid, pasid_mapping);
+
+	timeout = jiffies + msecs_to_jiffies(10);
+	while (!(RREG32(SOC15_REG_OFFSET(ATHUB, 0,
+			regATC_VMID_PASID_MAPPING_UPDATE_STATUS)) &
+			(1U << vmid))) {
+		if (time_after(jiffies, timeout)) {
+			pr_err("Fail to program VMID-PASID mapping\n");
+			return -ETIME;
+		}
+		cpu_relax();
+	}
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0,
+		regATC_VMID_PASID_MAPPING_UPDATE_STATUS),
+		1U << vmid);
+
+	reg = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX));
+	/* Every 4 numbers is a cycle. 1st is AID, 2nd and 3rd are XCDs,
+	 * and the 4th is reserved. Therefore "aid * 4 + (xcc_inst % 2) + 1"
+	 * programs _LUT for XCC and "aid * 4" for AID where the XCC connects
+	 * to.
+	 */
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX),
+		aid * 4 + (phy_inst % 2) + 1);
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_0_LUT) + vmid,
+		pasid_mapping);
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX),
+		aid * 4);
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_0_LUT_MM) + vmid,
+		pasid_mapping);
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_LUT_INDEX), reg);
+
+	return 0;
+}
+
+static inline struct v9_mqd *get_mqd(void *mqd)
+{
+	return (struct v9_mqd *)mqd;
+}
+
+static int kgd_gfx_v9_4_3_hqd_load(struct amdgpu_device *adev, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr, uint32_t wptr_shift,
+			uint32_t wptr_mask, struct mm_struct *mm, uint32_t inst)
+{
+	struct v9_mqd *m;
+	uint32_t *mqd_hqd;
+	uint32_t reg, hqd_base, hqd_end, data;
+
+	m = get_mqd(mqd);
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+
+	/* HQD registers extend to CP_HQD_AQL_DISPATCH_ID_HI */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
+	hqd_base = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_MQD_BASE_ADDR);
+	hqd_end = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), regCP_HQD_AQL_DISPATCH_ID_HI);
+
+	for (reg = hqd_base; reg <= hqd_end; reg++)
+		WREG32_XCC(reg, mqd_hqd[reg - hqd_base], inst);
+
+
+	/* Activate doorbell logic before triggering WPTR poll. */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_HQD_PQ_DOORBELL_CONTROL, data);
+
+	if (wptr) {
+		/* Don't read wptr with get_user because the user
+		 * context may not be accessible (if this function
+		 * runs in a work queue). Instead trigger a one-shot
+		 * polling read from memory in the CP. This assumes
+		 * that wptr is GPU-accessible in the queue's VMID via
+		 * ATC or SVM. WPTR==RPTR before starting the poll so
+		 * the CP starts fetching new commands from the right
+		 * place.
+		 *
+		 * Guessing a 64-bit WPTR from a 32-bit RPTR is a bit
+		 * tricky. Assume that the queue didn't overflow. The
+		 * number of valid bits in the 32-bit RPTR depends on
+		 * the queue size. The remaining bits are taken from
+		 * the saved 64-bit WPTR. If the WPTR wrapped, add the
+		 * queue size.
+		 */
+		uint32_t queue_size =
+			2 << REG_GET_FIELD(m->cp_hqd_pq_control,
+					   CP_HQD_PQ_CONTROL, QUEUE_SIZE);
+		uint64_t guessed_wptr = m->cp_hqd_pq_rptr & (queue_size - 1);
+
+		if ((m->cp_hqd_pq_wptr_lo & (queue_size - 1)) < guessed_wptr)
+			guessed_wptr += queue_size;
+		guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
+		guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
+
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_HQD_PQ_WPTR_LO,
+			lower_32_bits(guessed_wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_HQD_PQ_WPTR_HI,
+			upper_32_bits(guessed_wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_HQD_PQ_WPTR_POLL_ADDR,
+			lower_32_bits((uintptr_t)wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
+			upper_32_bits((uintptr_t)wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_PQ_WPTR_POLL_CNTL1,
+			(uint32_t)kgd_gfx_v9_get_queue_mask(adev, pipe_id, queue_id));
+	}
+
+	/* Start the EOP fetcher */
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_HQD_EOP_RPTR,
+	       REG_SET_FIELD(m->cp_hqd_eop_rptr, CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
+
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), regCP_HQD_ACTIVE, data);
+
+	kgd_gfx_v9_release_queue(adev, inst);
+
+	return 0;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v9_4_3_disable_debug_trap(struct amdgpu_device *adev,
+						bool keep_trap_enabled,
+						uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
+
+	return data;
+}
+
+static int kgd_gfx_v9_4_3_validate_trap_override_request(
+				struct amdgpu_device *adev,
+				uint32_t trap_override,
+				uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_FP_INVALID |
+				KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+				KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+				KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+				KFD_DBG_TRAP_MASK_FP_INEXACT |
+				KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+				KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION |
+				KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START |
+				KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
+			trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
+		return -EPERM;
+
+	return 0;
+}
+
+static uint32_t trap_mask_map_sw_to_hw(uint32_t mask)
+{
+	uint32_t trap_on_start = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START) ? 1 : 0;
+	uint32_t trap_on_end = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END) ? 1 : 0;
+	uint32_t excp_en = mask & (KFD_DBG_TRAP_MASK_FP_INVALID |
+				KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+				KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+				KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+				KFD_DBG_TRAP_MASK_FP_INEXACT |
+				KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+				KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION);
+	uint32_t ret;
+
+	ret = REG_SET_FIELD(0, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, excp_en);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START, trap_on_start);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END, trap_on_end);
+
+	return ret;
+}
+
+static uint32_t trap_mask_map_hw_to_sw(uint32_t mask)
+{
+	uint32_t ret = REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START;
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+	return ret;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v9_4_3_set_wave_launch_trap_override(
+				struct amdgpu_device *adev,
+				uint32_t vmid,
+				uint32_t trap_override,
+				uint32_t trap_mask_bits,
+				uint32_t trap_mask_request,
+				uint32_t *trap_mask_prev,
+				uint32_t kfd_dbg_trap_cntl_prev)
+
+{
+	uint32_t data = 0;
+
+	*trap_mask_prev = trap_mask_map_hw_to_sw(kfd_dbg_trap_cntl_prev);
+
+	data = (trap_mask_bits & trap_mask_request) |
+	       (*trap_mask_prev & ~trap_mask_request);
+	data = trap_mask_map_sw_to_hw(data);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
+
+	return data;
+}
+
+#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
+static uint32_t kgd_gfx_v9_4_3_set_address_watch(
+				struct amdgpu_device *adev,
+				uint64_t watch_address,
+				uint32_t watch_address_mask,
+				uint32_t watch_id,
+				uint32_t watch_mode,
+				uint32_t debug_vmid,
+				uint32_t inst)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 7);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_XCC((SOC15_REG_OFFSET(GC, GET_INST(GC, inst),
+			regTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high, inst);
+
+	WREG32_XCC((SOC15_REG_OFFSET(GC, GET_INST(GC, inst),
+			regTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low, inst);
+
+	return watch_address_cntl;
+}
+
+static uint32_t kgd_gfx_v9_4_3_clear_address_watch(struct amdgpu_device *adev,
+				uint32_t watch_id)
+{
+	return 0;
+}
+
+static uint32_t kgd_gfx_v9_4_3_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+						     int engine, int queue)
+{
+	uint32_t reg_offset = get_sdma_rlc_reg_offset(adev, engine, queue);
+	uint32_t status = RREG32(regSDMA_RLC0_CONTEXT_STATUS + reg_offset);
+	uint32_t doorbell_off = RREG32(regSDMA_RLC0_DOORBELL_OFFSET + reg_offset);
+	bool is_active = !!REG_GET_FIELD(status, SDMA_RLC0_CONTEXT_STATUS, SELECTED);
+
+	return is_active ? doorbell_off >> 2 : 0;
+}
+
+const struct kfd2kgd_calls gc_9_4_3_kfd2kgd = {
+	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_gfx_v9_4_3_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_gfx_v9_init_interrupts,
+	.hqd_load = kgd_gfx_v9_4_3_hqd_load,
+	.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
+	.hqd_sdma_load = kgd_gfx_v9_4_3_hqd_sdma_load,
+	.hqd_dump = kgd_gfx_v9_hqd_dump,
+	.hqd_sdma_dump = kgd_gfx_v9_4_3_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_gfx_v9_4_3_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_gfx_v9_hqd_destroy,
+	.hqd_sdma_destroy = kgd_gfx_v9_4_3_hqd_sdma_destroy,
+	.wave_control_execute = kgd_gfx_v9_wave_control_execute,
+	.get_atc_vmid_pasid_mapping_info =
+				kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
+	.set_vm_context_page_table_base =
+				kgd_gfx_v9_set_vm_context_page_table_base,
+	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
+	.program_trap_handler_settings =
+				kgd_gfx_v9_program_trap_handler_settings,
+	.build_dequeue_wait_counts_packet_info =
+				kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
+	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
+	.enable_debug_trap = kgd_aldebaran_enable_debug_trap,
+	.disable_debug_trap = kgd_gfx_v9_4_3_disable_debug_trap,
+	.validate_trap_override_request =
+			kgd_gfx_v9_4_3_validate_trap_override_request,
+	.set_wave_launch_trap_override =
+			kgd_gfx_v9_4_3_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v9_4_3_set_address_watch,
+	.clear_address_watch = kgd_gfx_v9_4_3_clear_address_watch,
+	.hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr,
+	.hqd_reset = kgd_gfx_v9_hqd_reset,
+	.hqd_sdma_get_doorbell = kgd_gfx_v9_4_3_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
new file mode 100644
index 000000000000..0239114fb6c4
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
@@ -0,0 +1,1123 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_amdkfd_gfx_v10.h"
+#include "gc/gc_10_1_0_offset.h"
+#include "gc/gc_10_1_0_sh_mask.h"
+#include "athub/athub_2_0_0_offset.h"
+#include "athub/athub_2_0_0_sh_mask.h"
+#include "oss/osssys_5_0_0_offset.h"
+#include "oss/osssys_5_0_0_sh_mask.h"
+#include "soc15_common.h"
+#include "v10_structs.h"
+#include "nv.h"
+#include "nvd.h"
+#include <uapi/linux/kfd_ioctl.h>
+
+enum hqd_dequeue_request_type {
+	NO_ACTION = 0,
+	DRAIN_PIPE,
+	RESET_WAVES,
+	SAVE_WAVES
+};
+
+static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	mutex_lock(&adev->srbm_mutex);
+	nv_grbm_select(adev, mec, pipe, queue, vmid);
+}
+
+static void unlock_srbm(struct amdgpu_device *adev)
+{
+	nv_grbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, queue_id, 0);
+}
+
+static uint64_t get_queue_mask(struct amdgpu_device *adev,
+			       uint32_t pipe_id, uint32_t queue_id)
+{
+	unsigned int bit = pipe_id * adev->gfx.mec.num_queue_per_pipe +
+			queue_id;
+
+	return 1ull << bit;
+}
+
+static void release_queue(struct amdgpu_device *adev)
+{
+	unlock_srbm(adev);
+}
+
+static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases, uint32_t inst)
+{
+	lock_srbm(adev, 0, 0, 0, vmid);
+
+	WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
+	WREG32_SOC15(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
+	/* APE1 no longer exists on GFX9 */
+
+	unlock_srbm(adev);
+}
+
+static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
+					unsigned int vmid, uint32_t inst)
+{
+	/*
+	 * We have to assume that there is no outstanding mapping.
+	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
+	 * a mapping is in progress or because a mapping finished
+	 * and the SW cleared it.
+	 * So the protocol is to always wait & clear.
+	 */
+	uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
+			ATC_VMID0_PASID_MAPPING__VALID_MASK;
+
+	pr_debug("pasid 0x%x vmid %d, reg value %x\n", pasid, vmid, pasid_mapping);
+
+	pr_debug("ATHUB, reg %x\n", SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING) + vmid);
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING) + vmid,
+	       pasid_mapping);
+
+#if 0
+	/* TODO: uncomment this code when the hardware support is ready. */
+	while (!(RREG32(SOC15_REG_OFFSET(
+				ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS)) &
+		 (1U << vmid)))
+		cpu_relax();
+
+	pr_debug("ATHUB mapping update finished\n");
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS),
+	       1U << vmid);
+#endif
+
+	/* Mapping vmid to pasid also for IH block */
+	pr_debug("update mapping for IH block and mmhub");
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT) + vmid,
+	       pasid_mapping);
+
+	return 0;
+}
+
+/* TODO - RING0 form of field is obsolete, seems to date back to SI
+ * but still works
+ */
+
+static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst)
+{
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, 0, 0);
+
+	WREG32_SOC15(GC, 0, mmCPC_INT_CNTL,
+		CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+		CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
+
+	unlock_srbm(adev);
+
+	return 0;
+}
+
+static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
+				unsigned int engine_id,
+				unsigned int queue_id)
+{
+	uint32_t sdma_engine_reg_base[2] = {
+		SOC15_REG_OFFSET(SDMA0, 0,
+				 mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL,
+		/* On gfx10, mmSDMA1_xxx registers are defined NOT based
+		 * on SDMA1 base address (dw 0x1860) but based on SDMA0
+		 * base address (dw 0x1260). Therefore use mmSDMA0_RLC0_RB_CNTL
+		 * instead of mmSDMA1_RLC0_RB_CNTL for the base address calc
+		 * below
+		 */
+		SOC15_REG_OFFSET(SDMA1, 0,
+				 mmSDMA1_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL
+	};
+
+	uint32_t retval = sdma_engine_reg_base[engine_id]
+		+ queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL);
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
+			queue_id, retval);
+
+	return retval;
+}
+
+#if 0
+static uint32_t get_watch_base_addr(struct amdgpu_device *adev)
+{
+	uint32_t retval = SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_H) -
+			mmTCP_WATCH0_ADDR_H;
+
+	pr_debug("kfd: reg watch base address: 0x%x\n", retval);
+
+	return retval;
+}
+#endif
+
+static inline struct v10_compute_mqd *get_mqd(void *mqd)
+{
+	return (struct v10_compute_mqd *)mqd;
+}
+
+static inline struct v10_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v10_sdma_mqd *)mqd;
+}
+
+static int kgd_hqd_load(struct amdgpu_device *adev, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr, uint32_t wptr_shift,
+			uint32_t wptr_mask, struct mm_struct *mm, uint32_t inst)
+{
+	struct v10_compute_mqd *m;
+	uint32_t *mqd_hqd;
+	uint32_t reg, hqd_base, data;
+
+	m = get_mqd(mqd);
+
+	pr_debug("Load hqd of pipe %d queue %d\n", pipe_id, queue_id);
+	acquire_queue(adev, pipe_id, queue_id);
+
+	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
+	hqd_base = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
+
+	for (reg = hqd_base;
+	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+		WREG32_SOC15_IP(GC, reg, mqd_hqd[reg - hqd_base]);
+
+
+	/* Activate doorbell logic before triggering WPTR poll. */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, data);
+
+	if (wptr) {
+		/* Don't read wptr with get_user because the user
+		 * context may not be accessible (if this function
+		 * runs in a work queue). Instead trigger a one-shot
+		 * polling read from memory in the CP. This assumes
+		 * that wptr is GPU-accessible in the queue's VMID via
+		 * ATC or SVM. WPTR==RPTR before starting the poll so
+		 * the CP starts fetching new commands from the right
+		 * place.
+		 *
+		 * Guessing a 64-bit WPTR from a 32-bit RPTR is a bit
+		 * tricky. Assume that the queue didn't overflow. The
+		 * number of valid bits in the 32-bit RPTR depends on
+		 * the queue size. The remaining bits are taken from
+		 * the saved 64-bit WPTR. If the WPTR wrapped, add the
+		 * queue size.
+		 */
+		uint32_t queue_size =
+			2 << REG_GET_FIELD(m->cp_hqd_pq_control,
+					   CP_HQD_PQ_CONTROL, QUEUE_SIZE);
+		uint64_t guessed_wptr = m->cp_hqd_pq_rptr & (queue_size - 1);
+
+		if ((m->cp_hqd_pq_wptr_lo & (queue_size - 1)) < guessed_wptr)
+			guessed_wptr += queue_size;
+		guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
+		guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
+
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_LO,
+		       lower_32_bits(guessed_wptr));
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_HI,
+		       upper_32_bits(guessed_wptr));
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
+		       lower_32_bits((uint64_t)wptr));
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
+		       upper_32_bits((uint64_t)wptr));
+		pr_debug("%s setting CP_PQ_WPTR_POLL_CNTL1 to %x\n", __func__,
+			 (uint32_t)get_queue_mask(adev, pipe_id, queue_id));
+		WREG32_SOC15(GC, 0, mmCP_PQ_WPTR_POLL_CNTL1,
+		       (uint32_t)get_queue_mask(adev, pipe_id, queue_id));
+	}
+
+	/* Start the EOP fetcher */
+	WREG32_SOC15(GC, 0, mmCP_HQD_EOP_RPTR,
+	       REG_SET_FIELD(m->cp_hqd_eop_rptr,
+			     CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
+
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE, data);
+
+	release_queue(adev);
+
+	return 0;
+}
+
+static int kgd_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    uint32_t doorbell_off, uint32_t inst)
+{
+	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
+	struct v10_compute_mqd *m;
+	uint32_t mec, pipe;
+	int r;
+
+	m = get_mqd(mqd);
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+		 mec, pipe, queue_id);
+
+	spin_lock(&adev->gfx.kiq[0].ring_lock);
+	r = amdgpu_ring_alloc(kiq_ring, 7);
+	if (r) {
+		pr_err("Failed to alloc KIQ (%d).\n", r);
+		goto out_unlock;
+	}
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
+			  PACKET3_MAP_QUEUES_VMID(m->cp_hqd_vmid) | /* VMID */
+			  PACKET3_MAP_QUEUES_QUEUE(queue_id) |
+			  PACKET3_MAP_QUEUES_PIPE(pipe) |
+			  PACKET3_MAP_QUEUES_ME((mec - 1)) |
+			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
+			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
+			  PACKET3_MAP_QUEUES_ENGINE_SEL(1) | /* engine_sel: hiq */
+			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(doorbell_off));
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_hi);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_hi);
+	amdgpu_ring_commit(kiq_ring);
+
+out_unlock:
+	spin_unlock(&adev->gfx.kiq[0].ring_lock);
+	release_queue(adev);
+
+	return r;
+}
+
+static int kgd_hqd_dump(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
+{
+	uint32_t i = 0, reg;
+#define HQD_N_REGS 56
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32_SOC15_IP(GC, addr);		\
+	} while (0)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	for (reg = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
+	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+		DUMP_REG(reg);
+
+	release_queue(adev);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct v10_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	unsigned long end_jiffies;
+	uint32_t data;
+	uint64_t data64;
+	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET,
+	       m->sdmax_rlcx_doorbell_offset);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
+				m->sdmax_rlcx_rb_rptr);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI,
+				m->sdmax_rlcx_rb_rptr_hi);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
+	if (read_user_wptr(mm, wptr64, data64)) {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       lower_32_bits(data64));
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       upper_32_bits(data64));
+	} else {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       m->sdmax_rlcx_rb_rptr_hi);
+	}
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
+			engine_id, queue_id);
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+6+7+10)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
+	     reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
+	     reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static bool kgd_hqd_is_occupied(struct amdgpu_device *adev,
+				uint64_t queue_address, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	acquire_queue(adev, pipe_id, queue_id);
+	act = RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE) &&
+		   high == RREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE_HI))
+			retval = true;
+	}
+	release_queue(adev);
+	return retval;
+}
+
+static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
+{
+	struct v10_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+static int kgd_hqd_destroy(struct amdgpu_device *adev, void *mqd,
+				enum kfd_preempt_type reset_type,
+				unsigned int utimeout, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	enum hqd_dequeue_request_type type;
+	unsigned long end_jiffies;
+	uint32_t temp;
+	struct v10_compute_mqd *m = get_mqd(mqd);
+
+	if (amdgpu_in_reset(adev))
+		return -EIO;
+
+#if 0
+	unsigned long flags;
+	int retry;
+#endif
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	if (m->cp_hqd_vmid == 0)
+		WREG32_FIELD15(GC, 0, RLC_CP_SCHEDULERS, scheduler1, 0);
+
+	switch (reset_type) {
+	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
+		type = DRAIN_PIPE;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
+		type = RESET_WAVES;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_SAVE:
+		type = SAVE_WAVES;
+		break;
+	default:
+		type = DRAIN_PIPE;
+		break;
+	}
+
+#if 0 /* Is this still needed? */
+	/* Workaround: If IQ timer is active and the wait time is close to or
+	 * equal to 0, dequeueing is not safe. Wait until either the wait time
+	 * is larger or timer is cleared. Also, ensure that IQ_REQ_PEND is
+	 * cleared before continuing. Also, ensure wait times are set to at
+	 * least 0x3.
+	 */
+	local_irq_save(flags);
+	preempt_disable();
+	retry = 5000; /* wait for 500 usecs at maximum */
+	while (true) {
+		temp = RREG32(mmCP_HQD_IQ_TIMER);
+		if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, PROCESSING_IQ)) {
+			pr_debug("HW is processing IQ\n");
+			goto loop;
+		}
+		if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, ACTIVE)) {
+			if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, RETRY_TYPE)
+					== 3) /* SEM-rearm is safe */
+				break;
+			/* Wait time 3 is safe for CP, but our MMIO read/write
+			 * time is close to 1 microsecond, so check for 10 to
+			 * leave more buffer room
+			 */
+			if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, WAIT_TIME)
+					>= 10)
+				break;
+			pr_debug("IQ timer is active\n");
+		} else
+			break;
+loop:
+		if (!retry) {
+			pr_err("CP HQD IQ timer status time out\n");
+			break;
+		}
+		ndelay(100);
+		--retry;
+	}
+	retry = 1000;
+	while (true) {
+		temp = RREG32(mmCP_HQD_DEQUEUE_REQUEST);
+		if (!(temp & CP_HQD_DEQUEUE_REQUEST__IQ_REQ_PEND_MASK))
+			break;
+		pr_debug("Dequeue request is pending\n");
+
+		if (!retry) {
+			pr_err("CP HQD dequeue request time out\n");
+			break;
+		}
+		ndelay(100);
+		--retry;
+	}
+	local_irq_restore(flags);
+	preempt_enable();
+#endif
+
+	WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, type);
+
+	end_jiffies = (utimeout * HZ / 1000) + jiffies;
+	while (true) {
+		temp = RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE);
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("cp queue preemption time out.\n");
+			release_queue(adev);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	release_queue(adev);
+	return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
+				unsigned int utimeout)
+{
+	struct v10_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t temp;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
+	m->sdmax_rlcx_rb_rptr_hi =
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI);
+
+	return 0;
+}
+
+static bool get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
+{
+	uint32_t value;
+
+	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
+
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
+}
+
+static int kgd_wave_control_execute(struct amdgpu_device *adev,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd, uint32_t inst)
+{
+	uint32_t data = 0;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32_SOC15(GC, 0, mmSQ_CMD, sq_cmd);
+
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		INSTANCE_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SA_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SE_BROADCAST_WRITES, 1);
+
+	WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, data);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+static void set_vm_context_page_table_base(struct amdgpu_device *adev,
+		uint32_t vmid, uint64_t page_table_base)
+{
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("trying to set page table base for wrong VMID %u\n",
+		       vmid);
+		return;
+	}
+
+	/* SDMA is on gfxhub as well for Navi1* series */
+	adev->gfxhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base);
+}
+
+/*
+ * GFX10 helper for wave launch stall requirements on debug trap setting.
+ *
+ * vmid:
+ *   Target VMID to stall/unstall.
+ *
+ * stall:
+ *   0-unstall wave launch (enable), 1-stall wave launch (disable).
+ *   After wavefront launch has been stalled, allocated waves must drain from
+ *   SPI in order for debug trap settings to take effect on those waves.
+ *   This is roughly a ~3500 clock cycle wait on SPI where a read on
+ *   SPI_GDBG_WAVE_CNTL translates to ~32 clock cycles.
+ *   KGD_GFX_V10_WAVE_LAUNCH_SPI_DRAIN_LATENCY indicates the number of reads required.
+ *
+ *   NOTE: We can afford to clear the entire STALL_VMID field on unstall
+ *   because current GFX10 chips cannot support multi-process debugging due to
+ *   trap configuration and masking being limited to global scope.  Always
+ *   assume single process conditions.
+ *
+ */
+
+#define KGD_GFX_V10_WAVE_LAUNCH_SPI_DRAIN_LATENCY	110
+static void kgd_gfx_v10_set_wave_launch_stall(struct amdgpu_device *adev, uint32_t vmid, bool stall)
+{
+	uint32_t data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+	int i;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_VMID,
+							stall ? 1 << vmid : 0);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data);
+
+	if (!stall)
+		return;
+
+	for (i = 0; i < KGD_GFX_V10_WAVE_LAUNCH_SPI_DRAIN_LATENCY; i++)
+		RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+}
+
+uint32_t kgd_gfx_v10_enable_debug_trap(struct amdgpu_device *adev,
+				bool restore_dbg_registers,
+				uint32_t vmid)
+{
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
+
+	/* assume gfx off is disabled for the debug session if rlc restore not supported. */
+	if (restore_dbg_registers) {
+		uint32_t data = 0;
+
+		data = REG_SET_FIELD(data, SPI_GDBG_TRAP_CONFIG,
+				VMID_SEL, 1 << vmid);
+		data = REG_SET_FIELD(data, SPI_GDBG_TRAP_CONFIG,
+				TRAP_EN, 1);
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_CONFIG), data);
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA0), 0);
+		WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA1), 0);
+
+		kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
+
+		mutex_unlock(&adev->grbm_idx_mutex);
+
+		return 0;
+	}
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v10_disable_debug_trap(struct amdgpu_device *adev,
+					bool keep_trap_enabled,
+					uint32_t vmid)
+{
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+int kgd_gfx_v10_validate_trap_override_request(struct amdgpu_device *adev,
+					      uint32_t trap_override,
+					      uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH;
+
+	/* The SPI_GDBG_TRAP_MASK register is global and affects all
+	 * processes. Only allow OR-ing the address-watch bit, since
+	 * this only affects processes under the debugger. Other bits
+	 * should stay 0 to avoid the debugger interfering with other
+	 * processes.
+	 */
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR)
+		return -EINVAL;
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					      uint32_t vmid,
+					      uint32_t trap_override,
+					      uint32_t trap_mask_bits,
+					      uint32_t trap_mask_request,
+					      uint32_t *trap_mask_prev,
+					      uint32_t kfd_dbg_trap_cntl_prev)
+{
+	uint32_t data, wave_cntl_prev;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	wave_cntl_prev = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
+
+	data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK));
+	*trap_mask_prev = REG_GET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN);
+
+	trap_mask_bits = (trap_mask_bits & trap_mask_request) |
+		(*trap_mask_prev & ~trap_mask_request);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN, trap_mask_bits);
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, REPLACE, trap_override);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data);
+
+	/* We need to preserve wave launch mode stall settings. */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), wave_cntl_prev);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+	bool is_mode_set = !!wave_launch_mode;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+			VMID_MASK, is_mode_set ? 1 << vmid : 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+			MODE, is_mode_set ? wave_launch_mode : 0);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+#define TCP_WATCH_STRIDE (mmTCP_WATCH1_ADDR_H - mmTCP_WATCH0_ADDR_H)
+#define SQ_WATCH_STRIDE (mmSQ_WATCH1_ADDR_H - mmSQ_WATCH0_ADDR_H)
+uint32_t kgd_gfx_v10_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid,
+					uint32_t inst)
+{
+	/* SQ_WATCH?_ADDR_* and TCP_WATCH?_ADDR_* are programmed with the
+	 * same values.
+	 */
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t tcp_watch_address_cntl;
+	uint32_t sq_watch_address_cntl;
+
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	tcp_watch_address_cntl = 0;
+	tcp_watch_address_cntl = REG_SET_FIELD(tcp_watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VMID,
+			debug_vmid);
+	tcp_watch_address_cntl = REG_SET_FIELD(tcp_watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+	tcp_watch_address_cntl = REG_SET_FIELD(tcp_watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 7);
+
+	sq_watch_address_cntl = 0;
+	sq_watch_address_cntl = REG_SET_FIELD(sq_watch_address_cntl,
+			SQ_WATCH0_CNTL,
+			VMID,
+			debug_vmid);
+	sq_watch_address_cntl = REG_SET_FIELD(sq_watch_address_cntl,
+			SQ_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+	sq_watch_address_cntl = REG_SET_FIELD(sq_watch_address_cntl,
+			SQ_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 6);
+
+	/* Turning off this watch point until we set all the registers */
+	tcp_watch_address_cntl = REG_SET_FIELD(tcp_watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			0);
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			tcp_watch_address_cntl);
+
+	sq_watch_address_cntl = REG_SET_FIELD(sq_watch_address_cntl,
+			SQ_WATCH0_CNTL,
+			VALID,
+			0);
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmSQ_WATCH0_CNTL) +
+			(watch_id * SQ_WATCH_STRIDE)),
+			sq_watch_address_cntl);
+
+	/* Program {TCP,SQ}_WATCH?_ADDR* */
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmSQ_WATCH0_ADDR_H) +
+			(watch_id * SQ_WATCH_STRIDE)),
+			watch_address_high);
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmSQ_WATCH0_ADDR_L) +
+			(watch_id * SQ_WATCH_STRIDE)),
+			watch_address_low);
+
+	/* Enable the watch point */
+	tcp_watch_address_cntl = REG_SET_FIELD(tcp_watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			tcp_watch_address_cntl);
+
+	sq_watch_address_cntl = REG_SET_FIELD(sq_watch_address_cntl,
+			SQ_WATCH0_CNTL,
+			VALID,
+			1);
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmSQ_WATCH0_CNTL) +
+			(watch_id * SQ_WATCH_STRIDE)),
+			sq_watch_address_cntl);
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v10_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id)
+{
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmSQ_WATCH0_CNTL) +
+			(watch_id * SQ_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	return 0;
+}
+#undef TCP_WATCH_STRIDE
+#undef SQ_WATCH_STRIDE
+
+
+/* kgd_gfx_v10_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
+ * The values read are:
+ *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
+ *     atomic_offload_wait_time -- Wait Count for L2 and GDS Atomics Offloads.
+ *     wrm_offload_wait_time    -- Wait Count for WAIT_REG_MEM Offloads.
+ *     gws_wait_time            -- Wait Count for Global Wave Syncs.
+ *     que_sleep_wait_time      -- Wait Count for Dequeue Retry.
+ *     sch_wave_wait_time       -- Wait Count for Scheduling Wave Message.
+ *     sem_rearm_wait_time      -- Wait Count for Semaphore re-arm.
+ *     deq_retry_wait_time      -- Wait Count for Global Wave Syncs.
+ */
+void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev,
+					uint32_t *wait_times,
+					uint32_t inst)
+
+{
+	*wait_times = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2));
+}
+
+void kgd_gfx_v10_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
+						uint32_t wait_times,
+						uint32_t sch_wave,
+						uint32_t que_sleep,
+						uint32_t *reg_offset,
+						uint32_t *reg_data)
+{
+	*reg_data = wait_times;
+
+	if (sch_wave)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				SCH_WAVE,
+				sch_wave);
+	if (que_sleep)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				QUE_SLEEP,
+				que_sleep);
+
+	*reg_offset = SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2);
+}
+
+static void program_trap_handler_settings(struct amdgpu_device *adev,
+		uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr,
+		uint32_t inst)
+{
+	lock_srbm(adev, 0, 0, 0, vmid);
+
+	/*
+	 * Program TBA registers
+	 */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TBA_LO),
+			lower_32_bits(tba_addr >> 8));
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TBA_HI),
+			upper_32_bits(tba_addr >> 8) |
+			(1 << SQ_SHADER_TBA_HI__TRAP_EN__SHIFT));
+
+	/*
+	 * Program TMA registers
+	 */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TMA_LO),
+			lower_32_bits(tma_addr >> 8));
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TMA_HI),
+			upper_32_bits(tma_addr >> 8));
+
+	unlock_srbm(adev);
+}
+
+uint64_t kgd_gfx_v10_hqd_get_pq_addr(struct amdgpu_device *adev,
+				     uint32_t pipe_id, uint32_t queue_id,
+				     uint32_t inst)
+{
+	return 0;
+}
+
+uint64_t kgd_gfx_v10_hqd_reset(struct amdgpu_device *adev,
+			       uint32_t pipe_id, uint32_t queue_id,
+			       uint32_t inst, unsigned int utimeout)
+{
+	return 0;
+}
+
+uint32_t kgd_gfx_v10_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+					   int engine, int queue)
+{
+	return 0;
+}
+
+const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
+	.program_sh_mem_settings = kgd_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_init_interrupts,
+	.hqd_load = kgd_hqd_load,
+	.hiq_mqd_load = kgd_hiq_mqd_load,
+	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_dump = kgd_hqd_dump,
+	.hqd_sdma_dump = kgd_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_hqd_destroy,
+	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+	.wave_control_execute = kgd_wave_control_execute,
+	.get_atc_vmid_pasid_mapping_info =
+			get_atc_vmid_pasid_mapping_info,
+	.set_vm_context_page_table_base = set_vm_context_page_table_base,
+	.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
+	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v10_set_address_watch,
+	.clear_address_watch = kgd_gfx_v10_clear_address_watch,
+	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v10_build_dequeue_wait_counts_packet_info,
+	.program_trap_handler_settings = program_trap_handler_settings,
+	.hqd_get_pq_addr = kgd_gfx_v10_hqd_get_pq_addr,
+	.hqd_reset = kgd_gfx_v10_hqd_reset,
+	.hqd_sdma_get_doorbell = kgd_gfx_v10_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
new file mode 100644
index 000000000000..a4c607c88178
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
@@ -0,0 +1,70 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+uint32_t kgd_gfx_v10_enable_debug_trap(struct amdgpu_device *adev,
+				      bool restore_dbg_registers,
+				      uint32_t vmid);
+uint32_t kgd_gfx_v10_disable_debug_trap(struct amdgpu_device *adev,
+					bool keep_trap_enabled,
+					uint32_t vmid);
+int kgd_gfx_v10_validate_trap_override_request(struct amdgpu_device *adev,
+					     uint32_t trap_override,
+					     uint32_t *trap_mask_supported);
+uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					     uint32_t vmid,
+					     uint32_t trap_override,
+					     uint32_t trap_mask_bits,
+					     uint32_t trap_mask_request,
+					     uint32_t *trap_mask_prev,
+					     uint32_t kfd_dbg_trap_cntl_prev);
+uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
+					 uint8_t wave_launch_mode,
+					 uint32_t vmid);
+uint32_t kgd_gfx_v10_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid,
+					uint32_t inst);
+uint32_t kgd_gfx_v10_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id);
+void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev,
+				uint32_t *wait_times,
+				uint32_t inst);
+void kgd_gfx_v10_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
+					       uint32_t wait_times,
+					       uint32_t sch_wave,
+					       uint32_t que_sleep,
+					       uint32_t *reg_offset,
+					       uint32_t *reg_data);
+uint64_t kgd_gfx_v10_hqd_get_pq_addr(struct amdgpu_device *adev,
+				    uint32_t pipe_id,
+				    uint32_t queue_id,
+				    uint32_t inst);
+uint64_t kgd_gfx_v10_hqd_reset(struct amdgpu_device *adev,
+			      uint32_t pipe_id,
+			      uint32_t queue_id,
+			      uint32_t inst,
+			      unsigned int utimeout);
+uint32_t kgd_gfx_v10_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+					   int engine, int queue);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
new file mode 100644
index 000000000000..f2278a0937ff
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
@@ -0,0 +1,687 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include <linux/mmu_context.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_amdkfd_gfx_v10.h"
+#include "gc/gc_10_3_0_offset.h"
+#include "gc/gc_10_3_0_sh_mask.h"
+#include "oss/osssys_5_0_0_offset.h"
+#include "oss/osssys_5_0_0_sh_mask.h"
+#include "athub/athub_2_1_0_offset.h"
+#include "athub/athub_2_1_0_sh_mask.h"
+#include "soc15_common.h"
+#include "v10_structs.h"
+#include "nv.h"
+#include "nvd.h"
+
+enum hqd_dequeue_request_type {
+	NO_ACTION = 0,
+	DRAIN_PIPE,
+	RESET_WAVES,
+	SAVE_WAVES
+};
+
+static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	mutex_lock(&adev->srbm_mutex);
+	nv_grbm_select(adev, mec, pipe, queue, vmid);
+}
+
+static void unlock_srbm(struct amdgpu_device *adev)
+{
+	nv_grbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, queue_id, 0);
+}
+
+static uint64_t get_queue_mask(struct amdgpu_device *adev,
+			       uint32_t pipe_id, uint32_t queue_id)
+{
+	unsigned int bit = pipe_id * adev->gfx.mec.num_queue_per_pipe +
+			queue_id;
+
+	return 1ull << bit;
+}
+
+static void release_queue(struct amdgpu_device *adev)
+{
+	unlock_srbm(adev);
+}
+
+static void program_sh_mem_settings_v10_3(struct amdgpu_device *adev, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases, uint32_t inst)
+{
+	lock_srbm(adev, 0, 0, 0, vmid);
+
+	WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
+	WREG32_SOC15(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
+	/* APE1 no longer exists on GFX9 */
+
+	unlock_srbm(adev);
+}
+
+/* ATC is defeatured on Sienna_Cichlid */
+static int set_pasid_vmid_mapping_v10_3(struct amdgpu_device *adev, unsigned int pasid,
+					unsigned int vmid, uint32_t inst)
+{
+	uint32_t value = pasid << IH_VMID_0_LUT__PASID__SHIFT;
+
+	/* Mapping vmid to pasid also for IH block */
+	pr_debug("mapping vmid %d -> pasid %d in IH block for GFX client\n",
+			vmid, pasid);
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT) + vmid, value);
+
+	return 0;
+}
+
+static int init_interrupts_v10_3(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst)
+{
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, 0, 0);
+
+	WREG32_SOC15(GC, 0, mmCPC_INT_CNTL,
+		CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+		CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
+
+	unlock_srbm(adev);
+
+	return 0;
+}
+
+static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
+				unsigned int engine_id,
+				unsigned int queue_id)
+{
+	uint32_t sdma_engine_reg_base = 0;
+	uint32_t sdma_rlc_reg_offset;
+
+	switch (engine_id) {
+	default:
+		dev_warn(adev->dev,
+			 "Invalid sdma engine id (%d), using engine id 0\n",
+			 engine_id);
+		fallthrough;
+	case 0:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
+		break;
+	case 1:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				mmSDMA1_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
+		break;
+	case 2:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				mmSDMA2_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
+		break;
+	case 3:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				mmSDMA3_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
+		break;
+	}
+
+	sdma_rlc_reg_offset = sdma_engine_reg_base
+		+ queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL);
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
+			queue_id, sdma_rlc_reg_offset);
+
+	return sdma_rlc_reg_offset;
+}
+
+static inline struct v10_compute_mqd *get_mqd(void *mqd)
+{
+	return (struct v10_compute_mqd *)mqd;
+}
+
+static inline struct v10_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v10_sdma_mqd *)mqd;
+}
+
+static int hqd_load_v10_3(struct amdgpu_device *adev, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr, uint32_t wptr_shift,
+			uint32_t wptr_mask, struct mm_struct *mm, uint32_t inst)
+{
+	struct v10_compute_mqd *m;
+	uint32_t *mqd_hqd;
+	uint32_t reg, hqd_base, data;
+
+	m = get_mqd(mqd);
+
+	pr_debug("Load hqd of pipe %d queue %d\n", pipe_id, queue_id);
+	acquire_queue(adev, pipe_id, queue_id);
+
+	/* HIQ is set during driver init period with vmid set to 0*/
+	if (m->cp_hqd_vmid == 0) {
+		uint32_t value, mec, pipe;
+
+		mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+		pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+		pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+			mec, pipe, queue_id);
+		value = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
+		value = REG_SET_FIELD(value, RLC_CP_SCHEDULERS, scheduler1,
+			((mec << 5) | (pipe << 3) | queue_id | 0x80));
+		WREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS, value);
+	}
+
+	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
+	hqd_base = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
+
+	for (reg = hqd_base;
+	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+		WREG32_SOC15_IP(GC, reg, mqd_hqd[reg - hqd_base]);
+
+
+	/* Activate doorbell logic before triggering WPTR poll. */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, data);
+
+	if (wptr) {
+		/* Don't read wptr with get_user because the user
+		 * context may not be accessible (if this function
+		 * runs in a work queue). Instead trigger a one-shot
+		 * polling read from memory in the CP. This assumes
+		 * that wptr is GPU-accessible in the queue's VMID via
+		 * ATC or SVM. WPTR==RPTR before starting the poll so
+		 * the CP starts fetching new commands from the right
+		 * place.
+		 *
+		 * Guessing a 64-bit WPTR from a 32-bit RPTR is a bit
+		 * tricky. Assume that the queue didn't overflow. The
+		 * number of valid bits in the 32-bit RPTR depends on
+		 * the queue size. The remaining bits are taken from
+		 * the saved 64-bit WPTR. If the WPTR wrapped, add the
+		 * queue size.
+		 */
+		uint32_t queue_size =
+			2 << REG_GET_FIELD(m->cp_hqd_pq_control,
+					   CP_HQD_PQ_CONTROL, QUEUE_SIZE);
+		uint64_t guessed_wptr = m->cp_hqd_pq_rptr & (queue_size - 1);
+
+		if ((m->cp_hqd_pq_wptr_lo & (queue_size - 1)) < guessed_wptr)
+			guessed_wptr += queue_size;
+		guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
+		guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
+
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_LO,
+		       lower_32_bits(guessed_wptr));
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_HI,
+		       upper_32_bits(guessed_wptr));
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
+		       lower_32_bits((uint64_t)wptr));
+		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
+		       upper_32_bits((uint64_t)wptr));
+		pr_debug("%s setting CP_PQ_WPTR_POLL_CNTL1 to %x\n", __func__,
+			 (uint32_t)get_queue_mask(adev, pipe_id, queue_id));
+		WREG32_SOC15(GC, 0, mmCP_PQ_WPTR_POLL_CNTL1,
+		       (uint32_t)get_queue_mask(adev, pipe_id, queue_id));
+	}
+
+	/* Start the EOP fetcher */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_EOP_RPTR),
+	       REG_SET_FIELD(m->cp_hqd_eop_rptr,
+			     CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
+
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE, data);
+
+	release_queue(adev);
+
+	return 0;
+}
+
+static int hiq_mqd_load_v10_3(struct amdgpu_device *adev, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    uint32_t doorbell_off, uint32_t inst)
+{
+	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
+	struct v10_compute_mqd *m;
+	uint32_t mec, pipe;
+	int r;
+
+	m = get_mqd(mqd);
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+		 mec, pipe, queue_id);
+
+	spin_lock(&adev->gfx.kiq[0].ring_lock);
+	r = amdgpu_ring_alloc(kiq_ring, 7);
+	if (r) {
+		pr_err("Failed to alloc KIQ (%d).\n", r);
+		goto out_unlock;
+	}
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
+			  PACKET3_MAP_QUEUES_VMID(m->cp_hqd_vmid) | /* VMID */
+			  PACKET3_MAP_QUEUES_QUEUE(queue_id) |
+			  PACKET3_MAP_QUEUES_PIPE(pipe) |
+			  PACKET3_MAP_QUEUES_ME((mec - 1)) |
+			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
+			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
+			  PACKET3_MAP_QUEUES_ENGINE_SEL(1) | /* engine_sel: hiq */
+			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(doorbell_off));
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_hi);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_hi);
+	amdgpu_ring_commit(kiq_ring);
+
+out_unlock:
+	spin_unlock(&adev->gfx.kiq[0].ring_lock);
+	release_queue(adev);
+
+	return r;
+}
+
+static int hqd_dump_v10_3(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
+{
+	uint32_t i = 0, reg;
+#define HQD_N_REGS 56
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32_SOC15_IP(GC, addr);		\
+	} while (0)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	for (reg = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR);
+	     reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++)
+		DUMP_REG(reg);
+
+	release_queue(adev);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int hqd_sdma_load_v10_3(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct v10_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	unsigned long end_jiffies;
+	uint32_t data;
+	uint64_t data64;
+	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET,
+	       m->sdmax_rlcx_doorbell_offset);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
+				m->sdmax_rlcx_rb_rptr);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI,
+				m->sdmax_rlcx_rb_rptr_hi);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
+	if (read_user_wptr(mm, wptr64, data64)) {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       lower_32_bits(data64));
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       upper_32_bits(data64));
+	} else {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       m->sdmax_rlcx_rb_rptr_hi);
+	}
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int hqd_sdma_dump_v10_3(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
+			engine_id, queue_id);
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+6+7+12)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
+	     reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
+	     reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static bool hqd_is_occupied_v10_3(struct amdgpu_device *adev,
+				uint64_t queue_address, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	acquire_queue(adev, pipe_id, queue_id);
+	act = RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE) &&
+		   high == RREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE_HI))
+			retval = true;
+	}
+	release_queue(adev);
+	return retval;
+}
+
+static bool hqd_sdma_is_occupied_v10_3(struct amdgpu_device *adev,
+				void *mqd)
+{
+	struct v10_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+static int hqd_destroy_v10_3(struct amdgpu_device *adev, void *mqd,
+				enum kfd_preempt_type reset_type,
+				unsigned int utimeout, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	enum hqd_dequeue_request_type type;
+	unsigned long end_jiffies;
+	uint32_t temp;
+	struct v10_compute_mqd *m = get_mqd(mqd);
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	if (m->cp_hqd_vmid == 0)
+		WREG32_FIELD15(GC, 0, RLC_CP_SCHEDULERS, scheduler1, 0);
+
+	switch (reset_type) {
+	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
+		type = DRAIN_PIPE;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
+		type = RESET_WAVES;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_SAVE:
+		type = SAVE_WAVES;
+		break;
+	default:
+		type = DRAIN_PIPE;
+		break;
+	}
+
+	WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, type);
+
+	end_jiffies = (utimeout * HZ / 1000) + jiffies;
+	while (true) {
+		temp = RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE);
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("cp queue pipe %d queue %d preemption failed\n",
+					pipe_id, queue_id);
+			release_queue(adev);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	release_queue(adev);
+	return 0;
+}
+
+static int hqd_sdma_destroy_v10_3(struct amdgpu_device *adev, void *mqd,
+				unsigned int utimeout)
+{
+	struct v10_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t temp;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
+	m->sdmax_rlcx_rb_rptr_hi =
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI);
+
+	return 0;
+}
+
+static int wave_control_execute_v10_3(struct amdgpu_device *adev,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd, uint32_t inst)
+{
+	uint32_t data = 0;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_CMD), sq_cmd);
+
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		INSTANCE_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SA_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SE_BROADCAST_WRITES, 1);
+
+	WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, data);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+static bool get_atc_vmid_pasid_mapping_info_v10_3(struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
+{
+	uint32_t value;
+
+	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
+
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
+}
+
+static void set_vm_context_page_table_base_v10_3(struct amdgpu_device *adev,
+		uint32_t vmid, uint64_t page_table_base)
+{
+	/* SDMA is on gfxhub as well for Navi1* series */
+	adev->gfxhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base);
+}
+
+static void program_trap_handler_settings_v10_3(struct amdgpu_device *adev,
+			uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr,
+			uint32_t inst)
+{
+	lock_srbm(adev, 0, 0, 0, vmid);
+
+	/*
+	 * Program TBA registers
+	 */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TBA_LO),
+			lower_32_bits(tba_addr >> 8));
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TBA_HI),
+			upper_32_bits(tba_addr >> 8) |
+			(1 << SQ_SHADER_TBA_HI__TRAP_EN__SHIFT));
+
+	/*
+	 * Program TMA registers
+	 */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TMA_LO),
+			lower_32_bits(tma_addr >> 8));
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_SHADER_TMA_HI),
+			 upper_32_bits(tma_addr >> 8));
+
+	unlock_srbm(adev);
+}
+
+const struct kfd2kgd_calls gfx_v10_3_kfd2kgd = {
+	.program_sh_mem_settings = program_sh_mem_settings_v10_3,
+	.set_pasid_vmid_mapping = set_pasid_vmid_mapping_v10_3,
+	.init_interrupts = init_interrupts_v10_3,
+	.hqd_load = hqd_load_v10_3,
+	.hiq_mqd_load = hiq_mqd_load_v10_3,
+	.hqd_sdma_load = hqd_sdma_load_v10_3,
+	.hqd_dump = hqd_dump_v10_3,
+	.hqd_sdma_dump = hqd_sdma_dump_v10_3,
+	.hqd_is_occupied = hqd_is_occupied_v10_3,
+	.hqd_sdma_is_occupied = hqd_sdma_is_occupied_v10_3,
+	.hqd_destroy = hqd_destroy_v10_3,
+	.hqd_sdma_destroy = hqd_sdma_destroy_v10_3,
+	.wave_control_execute = wave_control_execute_v10_3,
+	.get_atc_vmid_pasid_mapping_info = get_atc_vmid_pasid_mapping_info_v10_3,
+	.set_vm_context_page_table_base = set_vm_context_page_table_base_v10_3,
+	.program_trap_handler_settings = program_trap_handler_settings_v10_3,
+	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v10_build_dequeue_wait_counts_packet_info,
+	.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
+	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v10_set_address_watch,
+	.clear_address_watch = kgd_gfx_v10_clear_address_watch,
+	.hqd_get_pq_addr = kgd_gfx_v10_hqd_get_pq_addr,
+	.hqd_reset = kgd_gfx_v10_hqd_reset,
+	.hqd_sdma_get_doorbell = kgd_gfx_v10_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
new file mode 100644
index 000000000000..aaccf0b9947d
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
@@ -0,0 +1,835 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include <linux/mmu_context.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "gc/gc_11_0_0_offset.h"
+#include "gc/gc_11_0_0_sh_mask.h"
+#include "oss/osssys_6_0_0_offset.h"
+#include "oss/osssys_6_0_0_sh_mask.h"
+#include "soc15_common.h"
+#include "soc15d.h"
+#include "v11_structs.h"
+#include "soc21.h"
+#include <uapi/linux/kfd_ioctl.h>
+
+enum hqd_dequeue_request_type {
+	NO_ACTION = 0,
+	DRAIN_PIPE,
+	RESET_WAVES,
+	SAVE_WAVES
+};
+
+static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	mutex_lock(&adev->srbm_mutex);
+	soc21_grbm_select(adev, mec, pipe, queue, vmid);
+}
+
+static void unlock_srbm(struct amdgpu_device *adev)
+{
+	soc21_grbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, queue_id, 0);
+}
+
+static uint64_t get_queue_mask(struct amdgpu_device *adev,
+			       uint32_t pipe_id, uint32_t queue_id)
+{
+	unsigned int bit = pipe_id * adev->gfx.mec.num_queue_per_pipe +
+			queue_id;
+
+	return 1ull << bit;
+}
+
+static void release_queue(struct amdgpu_device *adev)
+{
+	unlock_srbm(adev);
+}
+
+static void program_sh_mem_settings_v11(struct amdgpu_device *adev, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases, uint32_t inst)
+{
+	lock_srbm(adev, 0, 0, 0, vmid);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, regSH_MEM_CONFIG), sh_mem_config);
+	WREG32(SOC15_REG_OFFSET(GC, 0, regSH_MEM_BASES), sh_mem_bases);
+
+	unlock_srbm(adev);
+}
+
+static int set_pasid_vmid_mapping_v11(struct amdgpu_device *adev, unsigned int pasid,
+					unsigned int vmid, uint32_t inst)
+{
+	uint32_t value = pasid << IH_VMID_0_LUT__PASID__SHIFT;
+
+	/* Mapping vmid to pasid also for IH block */
+	pr_debug("mapping vmid %d -> pasid %d in IH block for GFX client\n",
+			vmid, pasid);
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, regIH_VMID_0_LUT) + vmid, value);
+
+	return 0;
+}
+
+static int init_interrupts_v11(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst)
+{
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, 0, 0);
+
+	WREG32_SOC15(GC, 0, regCPC_INT_CNTL,
+		CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+		CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
+
+	unlock_srbm(adev);
+
+	return 0;
+}
+
+static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
+				unsigned int engine_id,
+				unsigned int queue_id)
+{
+	uint32_t sdma_engine_reg_base = 0;
+	uint32_t sdma_rlc_reg_offset;
+
+	switch (engine_id) {
+	case 0:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				regSDMA0_QUEUE0_RB_CNTL) - regSDMA0_QUEUE0_RB_CNTL;
+		break;
+	case 1:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA1, 0,
+				regSDMA1_QUEUE0_RB_CNTL) - regSDMA0_QUEUE0_RB_CNTL;
+		break;
+	default:
+		BUG();
+	}
+
+	sdma_rlc_reg_offset = sdma_engine_reg_base
+		+ queue_id * (regSDMA0_QUEUE1_RB_CNTL - regSDMA0_QUEUE0_RB_CNTL);
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
+			queue_id, sdma_rlc_reg_offset);
+
+	return sdma_rlc_reg_offset;
+}
+
+static inline struct v11_compute_mqd *get_mqd(void *mqd)
+{
+	return (struct v11_compute_mqd *)mqd;
+}
+
+static inline struct v11_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v11_sdma_mqd *)mqd;
+}
+
+static int hqd_load_v11(struct amdgpu_device *adev, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr,
+			uint32_t wptr_shift, uint32_t wptr_mask,
+			struct mm_struct *mm, uint32_t inst)
+{
+	struct v11_compute_mqd *m;
+	uint32_t *mqd_hqd;
+	uint32_t reg, hqd_base, data;
+
+	m = get_mqd(mqd);
+
+	pr_debug("Load hqd of pipe %d queue %d\n", pipe_id, queue_id);
+	acquire_queue(adev, pipe_id, queue_id);
+
+	/* HIQ is set during driver init period with vmid set to 0*/
+	if (m->cp_hqd_vmid == 0) {
+		uint32_t value, mec, pipe;
+
+		mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+		pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+		pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+			mec, pipe, queue_id);
+		value = RREG32(SOC15_REG_OFFSET(GC, 0, regRLC_CP_SCHEDULERS));
+		value = REG_SET_FIELD(value, RLC_CP_SCHEDULERS, scheduler1,
+			((mec << 5) | (pipe << 3) | queue_id | 0x80));
+		WREG32(SOC15_REG_OFFSET(GC, 0, regRLC_CP_SCHEDULERS), value);
+	}
+
+	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
+	hqd_base = SOC15_REG_OFFSET(GC, 0, regCP_MQD_BASE_ADDR);
+
+	for (reg = hqd_base;
+	     reg <= SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_WPTR_HI); reg++)
+		WREG32(reg, mqd_hqd[reg - hqd_base]);
+
+
+	/* Activate doorbell logic before triggering WPTR poll. */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL), data);
+
+	if (wptr) {
+		/* Don't read wptr with get_user because the user
+		 * context may not be accessible (if this function
+		 * runs in a work queue). Instead trigger a one-shot
+		 * polling read from memory in the CP. This assumes
+		 * that wptr is GPU-accessible in the queue's VMID via
+		 * ATC or SVM. WPTR==RPTR before starting the poll so
+		 * the CP starts fetching new commands from the right
+		 * place.
+		 *
+		 * Guessing a 64-bit WPTR from a 32-bit RPTR is a bit
+		 * tricky. Assume that the queue didn't overflow. The
+		 * number of valid bits in the 32-bit RPTR depends on
+		 * the queue size. The remaining bits are taken from
+		 * the saved 64-bit WPTR. If the WPTR wrapped, add the
+		 * queue size.
+		 */
+		uint32_t queue_size =
+			2 << REG_GET_FIELD(m->cp_hqd_pq_control,
+					   CP_HQD_PQ_CONTROL, QUEUE_SIZE);
+		uint64_t guessed_wptr = m->cp_hqd_pq_rptr & (queue_size - 1);
+
+		if ((m->cp_hqd_pq_wptr_lo & (queue_size - 1)) < guessed_wptr)
+			guessed_wptr += queue_size;
+		guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
+		guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
+
+		WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_WPTR_LO),
+		       lower_32_bits(guessed_wptr));
+		WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_WPTR_HI),
+		       upper_32_bits(guessed_wptr));
+		WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR),
+		       lower_32_bits((uint64_t)wptr));
+		WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI),
+		       upper_32_bits((uint64_t)wptr));
+		pr_debug("%s setting CP_PQ_WPTR_POLL_CNTL1 to %x\n", __func__,
+			 (uint32_t)get_queue_mask(adev, pipe_id, queue_id));
+		WREG32(SOC15_REG_OFFSET(GC, 0, regCP_PQ_WPTR_POLL_CNTL1),
+		       (uint32_t)get_queue_mask(adev, pipe_id, queue_id));
+	}
+
+	/* Start the EOP fetcher */
+	WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_EOP_RPTR),
+	       REG_SET_FIELD(m->cp_hqd_eop_rptr,
+			     CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
+
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_ACTIVE), data);
+
+	release_queue(adev);
+
+	return 0;
+}
+
+static int hiq_mqd_load_v11(struct amdgpu_device *adev, void *mqd,
+			      uint32_t pipe_id, uint32_t queue_id,
+			      uint32_t doorbell_off, uint32_t inst)
+{
+	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
+	struct v11_compute_mqd *m;
+	uint32_t mec, pipe;
+	int r;
+
+	m = get_mqd(mqd);
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+		 mec, pipe, queue_id);
+
+	spin_lock(&adev->gfx.kiq[0].ring_lock);
+	r = amdgpu_ring_alloc(kiq_ring, 7);
+	if (r) {
+		pr_err("Failed to alloc KIQ (%d).\n", r);
+		goto out_unlock;
+	}
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
+			  PACKET3_MAP_QUEUES_VMID(m->cp_hqd_vmid) | /* VMID */
+			  PACKET3_MAP_QUEUES_QUEUE(queue_id) |
+			  PACKET3_MAP_QUEUES_PIPE(pipe) |
+			  PACKET3_MAP_QUEUES_ME((mec - 1)) |
+			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
+			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
+			  PACKET3_MAP_QUEUES_ENGINE_SEL(1) | /* engine_sel: hiq */
+			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_MAP_QUEUES_DOORBELL_OFFSET(doorbell_off));
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_hi);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_hi);
+	amdgpu_ring_commit(kiq_ring);
+
+out_unlock:
+	spin_unlock(&adev->gfx.kiq[0].ring_lock);
+	release_queue(adev);
+
+	return r;
+}
+
+static int hqd_dump_v11(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
+{
+	uint32_t i = 0, reg;
+#define HQD_N_REGS 56
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	for (reg = SOC15_REG_OFFSET(GC, 0, regCP_MQD_BASE_ADDR);
+	     reg <= SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_WPTR_HI); reg++)
+		DUMP_REG(reg);
+
+	release_queue(adev);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int hqd_sdma_load_v11(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct v11_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	unsigned long end_jiffies;
+	uint32_t data;
+	uint64_t data64;
+	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_QUEUE0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_CONTEXT_STATUS);
+		if (data & SDMA0_QUEUE0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_DOORBELL_OFFSET,
+	       m->sdmax_rlcx_doorbell_offset);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_QUEUE0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_RPTR,
+				m->sdmax_rlcx_rb_rptr);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_RPTR_HI,
+				m->sdmax_rlcx_rb_rptr_hi);
+
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_MINOR_PTR_UPDATE, 1);
+	if (read_user_wptr(mm, wptr64, data64)) {
+		WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_WPTR,
+		       lower_32_bits(data64));
+		WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_WPTR_HI,
+		       upper_32_bits(data64));
+	} else {
+		WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+		WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_WPTR_HI,
+		       m->sdmax_rlcx_rb_rptr_hi);
+	}
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_MINOR_PTR_UPDATE, 0);
+
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_QUEUE0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int hqd_sdma_dump_v11(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
+			engine_id, queue_id);
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (7+11+1+12+12)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = regSDMA0_QUEUE0_RB_CNTL;
+	     reg <= regSDMA0_QUEUE0_RB_WPTR_HI; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = regSDMA0_QUEUE0_RB_RPTR_ADDR_HI;
+	     reg <= regSDMA0_QUEUE0_DOORBELL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = regSDMA0_QUEUE0_DOORBELL_LOG;
+	     reg <= regSDMA0_QUEUE0_DOORBELL_LOG; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = regSDMA0_QUEUE0_DOORBELL_OFFSET;
+	     reg <= regSDMA0_QUEUE0_RB_PREEMPT; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = regSDMA0_QUEUE0_MIDCMD_DATA0;
+	     reg <= regSDMA0_QUEUE0_MIDCMD_CNTL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static bool hqd_is_occupied_v11(struct amdgpu_device *adev, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id, uint32_t inst)
+{
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	acquire_queue(adev, pipe_id, queue_id);
+	act = RREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_ACTIVE));
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_BASE)) &&
+		   high == RREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_BASE_HI)))
+			retval = true;
+	}
+	release_queue(adev);
+	return retval;
+}
+
+static bool hqd_sdma_is_occupied_v11(struct amdgpu_device *adev, void *mqd)
+{
+	struct v11_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_QUEUE0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+static int hqd_destroy_v11(struct amdgpu_device *adev, void *mqd,
+				enum kfd_preempt_type reset_type,
+				unsigned int utimeout, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	enum hqd_dequeue_request_type type;
+	unsigned long end_jiffies;
+	uint32_t temp;
+	struct v11_compute_mqd *m = get_mqd(mqd);
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	if (m->cp_hqd_vmid == 0)
+		WREG32_FIELD15_PREREG(GC, 0, RLC_CP_SCHEDULERS, scheduler1, 0);
+
+	switch (reset_type) {
+	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
+		type = DRAIN_PIPE;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
+		type = RESET_WAVES;
+		break;
+	default:
+		type = DRAIN_PIPE;
+		break;
+	}
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_DEQUEUE_REQUEST), type);
+
+	end_jiffies = (utimeout * HZ / 1000) + jiffies;
+	while (true) {
+		temp = RREG32(SOC15_REG_OFFSET(GC, 0, regCP_HQD_ACTIVE));
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("cp queue pipe %d queue %d preemption failed\n",
+					pipe_id, queue_id);
+			release_queue(adev);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	release_queue(adev);
+	return 0;
+}
+
+static int hqd_sdma_destroy_v11(struct amdgpu_device *adev, void *mqd,
+				unsigned int utimeout)
+{
+	struct v11_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t temp;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	temp = RREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_CNTL);
+	temp = temp & ~SDMA0_QUEUE0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_CONTEXT_STATUS);
+		if (temp & SDMA0_QUEUE0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_CNTL) |
+		SDMA0_QUEUE0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_RPTR);
+	m->sdmax_rlcx_rb_rptr_hi =
+		RREG32(sdma_rlc_reg_offset + regSDMA0_QUEUE0_RB_RPTR_HI);
+
+	return 0;
+}
+
+static int wave_control_execute_v11(struct amdgpu_device *adev,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd, uint32_t inst)
+{
+	uint32_t data = 0;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_INDEX), gfx_index_val);
+	WREG32(SOC15_REG_OFFSET(GC, 0, regSQ_CMD), sq_cmd);
+
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		INSTANCE_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SA_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SE_BROADCAST_WRITES, 1);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_INDEX), data);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+static void set_vm_context_page_table_base_v11(struct amdgpu_device *adev,
+		uint32_t vmid, uint64_t page_table_base)
+{
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("trying to set page table base for wrong VMID %u\n",
+		       vmid);
+		return;
+	}
+
+	/* SDMA is on gfxhub as well for gfx11 adapters */
+	adev->gfxhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base);
+}
+
+/*
+ * Returns TRAP_EN, EXCP_EN and EXCP_REPLACE.
+ *
+ * restore_dbg_registers is ignored here but is a general interface requirement
+ * for devices that support GFXOFF and where the RLC save/restore list
+ * does not support hw registers for debugging i.e. the driver has to manually
+ * initialize the debug mode registers after it has disabled GFX off during the
+ * debug session.
+ */
+static uint32_t kgd_gfx_v11_enable_debug_trap(struct amdgpu_device *adev,
+					    bool restore_dbg_registers,
+					    uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
+
+	return data;
+}
+
+/* Returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v11_disable_debug_trap(struct amdgpu_device *adev,
+						bool keep_trap_enabled,
+						uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
+
+	return data;
+}
+
+static int kgd_gfx_v11_validate_trap_override_request(struct amdgpu_device *adev,
+							uint32_t trap_override,
+							uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_FP_INVALID |
+				KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+				KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+				KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+				KFD_DBG_TRAP_MASK_FP_INEXACT |
+				KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+				KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION;
+
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 4))
+		*trap_mask_supported |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START |
+					KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
+			trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
+		return -EPERM;
+
+	return 0;
+}
+
+static uint32_t trap_mask_map_sw_to_hw(uint32_t mask)
+{
+	uint32_t trap_on_start = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START) ? 1 : 0;
+	uint32_t trap_on_end = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END) ? 1 : 0;
+	uint32_t excp_en = mask & (KFD_DBG_TRAP_MASK_FP_INVALID |
+			KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+			KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+			KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+			KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+			KFD_DBG_TRAP_MASK_FP_INEXACT |
+			KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+			KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+			KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION);
+	uint32_t ret;
+
+	ret = REG_SET_FIELD(0, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, excp_en);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START, trap_on_start);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END, trap_on_end);
+
+	return ret;
+}
+
+static uint32_t trap_mask_map_hw_to_sw(uint32_t mask)
+{
+	uint32_t ret = REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START;
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+	return ret;
+}
+
+/* Returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v11_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					uint32_t vmid,
+					uint32_t trap_override,
+					uint32_t trap_mask_bits,
+					uint32_t trap_mask_request,
+					uint32_t *trap_mask_prev,
+					uint32_t kfd_dbg_trap_cntl_prev)
+{
+	uint32_t data = 0;
+
+	*trap_mask_prev = trap_mask_map_hw_to_sw(kfd_dbg_trap_cntl_prev);
+
+	data = (trap_mask_bits & trap_mask_request) | (*trap_mask_prev & ~trap_mask_request);
+	data = trap_mask_map_sw_to_hw(data);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
+
+	return data;
+}
+
+static uint32_t kgd_gfx_v11_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode);
+
+	return data;
+}
+
+#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
+static uint32_t kgd_gfx_v11_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid,
+					uint32_t inst)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 7);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	return watch_address_cntl;
+}
+
+static uint32_t kgd_gfx_v11_clear_address_watch(struct amdgpu_device *adev,
+						uint32_t watch_id)
+{
+	return 0;
+}
+
+static uint64_t kgd_gfx_v11_hqd_get_pq_addr(struct amdgpu_device *adev,
+					    uint32_t pipe_id, uint32_t queue_id,
+					    uint32_t inst)
+{
+	return 0;
+}
+
+static uint64_t kgd_gfx_v11_hqd_reset(struct amdgpu_device *adev,
+				      uint32_t pipe_id, uint32_t queue_id,
+				      uint32_t inst, unsigned int utimeout)
+{
+	return 0;
+}
+
+static uint32_t kgd_gfx_v11_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+						  int engine, int queue)
+{
+	return 0;
+}
+
+const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
+	.program_sh_mem_settings = program_sh_mem_settings_v11,
+	.set_pasid_vmid_mapping = set_pasid_vmid_mapping_v11,
+	.init_interrupts = init_interrupts_v11,
+	.hqd_load = hqd_load_v11,
+	.hiq_mqd_load = hiq_mqd_load_v11,
+	.hqd_sdma_load = hqd_sdma_load_v11,
+	.hqd_dump = hqd_dump_v11,
+	.hqd_sdma_dump = hqd_sdma_dump_v11,
+	.hqd_is_occupied = hqd_is_occupied_v11,
+	.hqd_sdma_is_occupied = hqd_sdma_is_occupied_v11,
+	.hqd_destroy = hqd_destroy_v11,
+	.hqd_sdma_destroy = hqd_sdma_destroy_v11,
+	.wave_control_execute = wave_control_execute_v11,
+	.get_atc_vmid_pasid_mapping_info = NULL,
+	.set_vm_context_page_table_base = set_vm_context_page_table_base_v11,
+	.enable_debug_trap = kgd_gfx_v11_enable_debug_trap,
+	.disable_debug_trap = kgd_gfx_v11_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v11_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v11_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v11_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v11_set_address_watch,
+	.clear_address_watch = kgd_gfx_v11_clear_address_watch,
+	.hqd_get_pq_addr = kgd_gfx_v11_hqd_get_pq_addr,
+	.hqd_reset = kgd_gfx_v11_hqd_reset,
+	.hqd_sdma_get_doorbell = kgd_gfx_v11_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v12.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v12.c
new file mode 100644
index 000000000000..e0ceab400b2d
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v12.c
@@ -0,0 +1,384 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "gc/gc_12_0_0_offset.h"
+#include "gc/gc_12_0_0_sh_mask.h"
+#include "soc24.h"
+#include <uapi/linux/kfd_ioctl.h>
+
+static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	mutex_lock(&adev->srbm_mutex);
+	soc24_grbm_select(adev, mec, pipe, queue, vmid);
+}
+
+static void unlock_srbm(struct amdgpu_device *adev)
+{
+	soc24_grbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, queue_id, 0);
+}
+
+static void release_queue(struct amdgpu_device *adev)
+{
+	unlock_srbm(adev);
+}
+
+static int init_interrupts_v12(struct amdgpu_device *adev, uint32_t pipe_id, uint32_t inst)
+{
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	lock_srbm(adev, mec, pipe, 0, 0);
+
+	WREG32_SOC15(GC, 0, regCPC_INT_CNTL,
+		CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+		CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
+
+	unlock_srbm(adev);
+
+	return 0;
+}
+
+static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
+				unsigned int engine_id,
+				unsigned int queue_id)
+{
+	uint32_t sdma_engine_reg_base = 0;
+	uint32_t sdma_rlc_reg_offset;
+
+	switch (engine_id) {
+	case 0:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				regSDMA0_QUEUE0_RB_CNTL) - regSDMA0_QUEUE0_RB_CNTL;
+		break;
+	case 1:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA1, 0,
+				regSDMA1_QUEUE0_RB_CNTL) - regSDMA0_QUEUE0_RB_CNTL;
+		break;
+	default:
+		BUG();
+	}
+
+	sdma_rlc_reg_offset = sdma_engine_reg_base
+		+ queue_id * (regSDMA0_QUEUE1_RB_CNTL - regSDMA0_QUEUE0_RB_CNTL);
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
+			queue_id, sdma_rlc_reg_offset);
+
+	return sdma_rlc_reg_offset;
+}
+
+static int hqd_dump_v12(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
+{
+	uint32_t i = 0, reg;
+#define HQD_N_REGS 56
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	for (reg = SOC15_REG_OFFSET(GC, 0, regCP_MQD_BASE_ADDR);
+	     reg <= SOC15_REG_OFFSET(GC, 0, regCP_HQD_PQ_WPTR_HI); reg++)
+		DUMP_REG(reg);
+
+	release_queue(adev);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int hqd_sdma_dump_v12(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
+			engine_id, queue_id);
+	uint32_t i = 0, reg;
+
+	const uint32_t first_reg = regSDMA0_QUEUE0_RB_CNTL;
+	const uint32_t last_reg = regSDMA0_QUEUE0_CONTEXT_STATUS;
+#undef HQD_N_REGS
+#define HQD_N_REGS (last_reg - first_reg + 1)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = first_reg;
+	     reg <= last_reg; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int wave_control_execute_v12(struct amdgpu_device *adev,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd, uint32_t inst)
+{
+	uint32_t data = 0;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_INDEX), gfx_index_val);
+	WREG32(SOC15_REG_OFFSET(GC, 0, regSQ_CMD), sq_cmd);
+
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		INSTANCE_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SA_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SE_BROADCAST_WRITES, 1);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_INDEX), data);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v12_enable_debug_trap(struct amdgpu_device *adev,
+					    bool restore_dbg_registers,
+					    uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
+
+	return data;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v12_disable_debug_trap(struct amdgpu_device *adev,
+						bool keep_trap_enabled,
+						uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, 0);
+
+	return data;
+}
+
+static int kgd_gfx_v12_validate_trap_override_request(struct amdgpu_device *adev,
+							uint32_t trap_override,
+							uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_FP_INVALID |
+				KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+				KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+				KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+				KFD_DBG_TRAP_MASK_FP_INEXACT |
+				KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+				KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION |
+				KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START |
+				KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
+			trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
+		return -EPERM;
+
+	return 0;
+}
+
+static uint32_t trap_mask_map_sw_to_hw(uint32_t mask)
+{
+	uint32_t trap_on_start = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START) ? 1 : 0;
+	uint32_t trap_on_end = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END) ? 1 : 0;
+	uint32_t excp_en = mask & (KFD_DBG_TRAP_MASK_FP_INVALID |
+			KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+			KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+			KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+			KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+			KFD_DBG_TRAP_MASK_FP_INEXACT |
+			KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+			KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+			KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION);
+	uint32_t ret;
+
+	ret = REG_SET_FIELD(0, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, excp_en);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START, trap_on_start);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END, trap_on_end);
+
+	return ret;
+}
+
+static uint32_t trap_mask_map_hw_to_sw(uint32_t mask)
+{
+	uint32_t ret = REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START;
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+	return ret;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v12_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					uint32_t vmid,
+					uint32_t trap_override,
+					uint32_t trap_mask_bits,
+					uint32_t trap_mask_request,
+					uint32_t *trap_mask_prev,
+					uint32_t kfd_dbg_trap_cntl_prev)
+
+{
+	uint32_t data = 0;
+
+	*trap_mask_prev = trap_mask_map_hw_to_sw(kfd_dbg_trap_cntl_prev);
+
+	data = (trap_mask_bits & trap_mask_request) | (*trap_mask_prev & ~trap_mask_request);
+	data = trap_mask_map_sw_to_hw(data);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
+
+	return data;
+}
+
+/* returns STALL_VMID or LAUNCH_MODE. */
+static uint32_t kgd_gfx_v12_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+	bool is_stall_mode = wave_launch_mode == 4;
+
+	if (is_stall_mode)
+		data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, STALL_VMID,
+									1);
+	else
+		data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE,
+							wave_launch_mode);
+
+	return data;
+}
+
+#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
+static uint32_t kgd_gfx_v12_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid,
+					uint32_t inst)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 7);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	return watch_address_cntl;
+}
+
+static uint32_t kgd_gfx_v12_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id)
+{
+	return 0;
+}
+
+static uint32_t kgd_gfx_v12_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+						 int engine, int queue)
+{
+	return 0;
+}
+
+const struct kfd2kgd_calls gfx_v12_kfd2kgd = {
+	.init_interrupts = init_interrupts_v12,
+	.hqd_dump = hqd_dump_v12,
+	.hqd_sdma_dump = hqd_sdma_dump_v12,
+	.wave_control_execute = wave_control_execute_v12,
+	.get_atc_vmid_pasid_mapping_info = NULL,
+	.enable_debug_trap = kgd_gfx_v12_enable_debug_trap,
+	.disable_debug_trap = kgd_gfx_v12_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v12_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v12_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v12_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v12_set_address_watch,
+	.clear_address_watch = kgd_gfx_v12_clear_address_watch,
+	.hqd_sdma_get_doorbell = kgd_gfx_v12_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
index 1a0a5f7cccbc..df77558e03ef 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
@@ -20,15 +20,11 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  */
 
-#include <linux/fdtable.h>
-#include <linux/uaccess.h>
-#include <linux/firmware.h>
-#include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_amdkfd.h"
 #include "cikd.h"
 #include "cik_sdma.h"
-#include "amdgpu_ucode.h"
+#include "gfx_v7_0.h"
 #include "gca/gfx_7_2_d.h"
 #include "gca/gfx_7_2_enum.h"
 #include "gca/gfx_7_2_sh_mask.h"
@@ -38,187 +34,65 @@
 #include "gmc/gmc_7_1_sh_mask.h"
 #include "cik_structs.h"
 
-#define CIK_PIPE_PER_MEC	(4)
+enum hqd_dequeue_request_type {
+	NO_ACTION = 0,
+	DRAIN_PIPE,
+	RESET_WAVES
+};
 
 enum {
 	MAX_TRAPID = 8,		/* 3 bits in the bitfield. */
 	MAX_WATCH_ADDRESSES = 4
 };
 
-enum {
-	ADDRESS_WATCH_REG_ADDR_HI = 0,
-	ADDRESS_WATCH_REG_ADDR_LO,
-	ADDRESS_WATCH_REG_CNTL,
-	ADDRESS_WATCH_REG_MAX
-};
-
-/*  not defined in the CI/KV reg file  */
-enum {
-	ADDRESS_WATCH_REG_CNTL_ATC_BIT = 0x10000000UL,
-	ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK = 0x00FFFFFF,
-	ADDRESS_WATCH_REG_ADDLOW_MASK_EXTENSION = 0x03000000,
-	/* extend the mask to 26 bits to match the low address field */
-	ADDRESS_WATCH_REG_ADDLOW_SHIFT = 6,
-	ADDRESS_WATCH_REG_ADDHIGH_MASK = 0xFFFF
-};
-
-static const uint32_t watchRegs[MAX_WATCH_ADDRESSES * ADDRESS_WATCH_REG_MAX] = {
-	mmTCP_WATCH0_ADDR_H, mmTCP_WATCH0_ADDR_L, mmTCP_WATCH0_CNTL,
-	mmTCP_WATCH1_ADDR_H, mmTCP_WATCH1_ADDR_L, mmTCP_WATCH1_CNTL,
-	mmTCP_WATCH2_ADDR_H, mmTCP_WATCH2_ADDR_L, mmTCP_WATCH2_CNTL,
-	mmTCP_WATCH3_ADDR_H, mmTCP_WATCH3_ADDR_L, mmTCP_WATCH3_CNTL
-};
-
-union TCP_WATCH_CNTL_BITS {
-	struct {
-		uint32_t mask:24;
-		uint32_t vmid:4;
-		uint32_t atc:1;
-		uint32_t mode:2;
-		uint32_t valid:1;
-	} bitfields, bits;
-	uint32_t u32All;
-	signed int i32All;
-	float f32All;
-};
-
-/*
- * Register access functions
- */
-
-static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
-		uint32_t sh_mem_config,	uint32_t sh_mem_ape1_base,
-		uint32_t sh_mem_ape1_limit, uint32_t sh_mem_bases);
-
-static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
-					unsigned int vmid);
-
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-				uint32_t hpd_size, uint64_t hpd_gpu_addr);
-static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
-static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
-			uint32_t queue_id, uint32_t __user *wptr);
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
-static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
-				uint32_t pipe_id, uint32_t queue_id);
-
-static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
-				unsigned int utimeout, uint32_t pipe_id,
-				uint32_t queue_id);
-static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
-static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
-				unsigned int utimeout);
-static int kgd_address_watch_disable(struct kgd_dev *kgd);
-static int kgd_address_watch_execute(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					uint32_t cntl_val,
-					uint32_t addr_hi,
-					uint32_t addr_lo);
-static int kgd_wave_control_execute(struct kgd_dev *kgd,
-					uint32_t gfx_index_val,
-					uint32_t sq_cmd);
-static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					unsigned int reg_offset);
-
-static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, uint8_t vmid);
-static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
-							uint8_t vmid);
-static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
-
-static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
-
-static const struct kfd2kgd_calls kfd2kgd = {
-	.init_gtt_mem_allocation = alloc_gtt_mem,
-	.free_gtt_mem = free_gtt_mem,
-	.get_vmem_size = get_vmem_size,
-	.get_gpu_clock_counter = get_gpu_clock_counter,
-	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
-	.program_sh_mem_settings = kgd_program_sh_mem_settings,
-	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
-	.init_pipeline = kgd_init_pipeline,
-	.init_interrupts = kgd_init_interrupts,
-	.hqd_load = kgd_hqd_load,
-	.hqd_sdma_load = kgd_hqd_sdma_load,
-	.hqd_is_occupied = kgd_hqd_is_occupied,
-	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
-	.hqd_destroy = kgd_hqd_destroy,
-	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
-	.address_watch_disable = kgd_address_watch_disable,
-	.address_watch_execute = kgd_address_watch_execute,
-	.wave_control_execute = kgd_wave_control_execute,
-	.address_watch_get_offset = kgd_address_watch_get_offset,
-	.get_atc_vmid_pasid_mapping_pasid = get_atc_vmid_pasid_mapping_pasid,
-	.get_atc_vmid_pasid_mapping_valid = get_atc_vmid_pasid_mapping_valid,
-	.write_vmid_invalidate_request = write_vmid_invalidate_request,
-	.get_fw_version = get_fw_version
-};
-
-struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void)
-{
-	return (struct kfd2kgd_calls *)&kfd2kgd;
-}
-
-static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
-{
-	return (struct amdgpu_device *)kgd;
-}
-
-static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
 			uint32_t queue, uint32_t vmid)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
 
 	mutex_lock(&adev->srbm_mutex);
 	WREG32(mmSRBM_GFX_CNTL, value);
 }
 
-static void unlock_srbm(struct kgd_dev *kgd)
+static void unlock_srbm(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
 	WREG32(mmSRBM_GFX_CNTL, 0);
 	mutex_unlock(&adev->srbm_mutex);
 }
 
-static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+static void acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
 				uint32_t queue_id)
 {
-	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
-	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
 
-	lock_srbm(kgd, mec, pipe, queue_id, 0);
+	lock_srbm(adev, mec, pipe, queue_id, 0);
 }
 
-static void release_queue(struct kgd_dev *kgd)
+static void release_queue(struct amdgpu_device *adev)
 {
-	unlock_srbm(kgd);
+	unlock_srbm(adev);
 }
 
-static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
 					uint32_t sh_mem_config,
 					uint32_t sh_mem_ape1_base,
 					uint32_t sh_mem_ape1_limit,
-					uint32_t sh_mem_bases)
+					uint32_t sh_mem_bases, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
-	lock_srbm(kgd, 0, 0, 0, vmid);
+	lock_srbm(adev, 0, 0, 0, vmid);
 
 	WREG32(mmSH_MEM_CONFIG, sh_mem_config);
 	WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
 	WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
 	WREG32(mmSH_MEM_BASES, sh_mem_bases);
 
-	unlock_srbm(kgd);
+	unlock_srbm(adev);
 }
 
-static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
-					unsigned int vmid)
+static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
+					unsigned int vmid, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
 	/*
 	 * We have to assume that there is no outstanding mapping.
 	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
@@ -240,51 +114,34 @@ static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
 	return 0;
 }
 
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-				uint32_t hpd_size, uint64_t hpd_gpu_addr)
-{
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
-	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
-	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
-
-	lock_srbm(kgd, mec, pipe, 0, 0);
-	WREG32(mmCP_HPD_EOP_BASE_ADDR, lower_32_bits(hpd_gpu_addr >> 8));
-	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(hpd_gpu_addr >> 8));
-	WREG32(mmCP_HPD_EOP_VMID, 0);
-	WREG32(mmCP_HPD_EOP_CONTROL, hpd_size);
-	unlock_srbm(kgd);
-
-	return 0;
-}
-
-static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
+static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t mec;
 	uint32_t pipe;
 
-	mec = (pipe_id / CIK_PIPE_PER_MEC) + 1;
-	pipe = (pipe_id % CIK_PIPE_PER_MEC);
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
 
-	lock_srbm(kgd, mec, pipe, 0, 0);
+	lock_srbm(adev, mec, pipe, 0, 0);
 
 	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
 			CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
 
-	unlock_srbm(kgd);
+	unlock_srbm(adev);
 
 	return 0;
 }
 
-static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+static inline uint32_t get_sdma_rlc_reg_offset(struct cik_sdma_rlc_registers *m)
 {
 	uint32_t retval;
 
 	retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET +
 			m->sdma_queue_id * KFD_CIK_SDMA_QUEUE_OFFSET;
 
-	pr_debug("kfd: sdma base address: 0x%x\n", retval);
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n",
+			m->sdma_engine_id, m->sdma_queue_id, retval);
 
 	return retval;
 }
@@ -299,112 +156,176 @@ static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
 	return (struct cik_sdma_rlc_registers *)mqd;
 }
 
-static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
-			uint32_t queue_id, uint32_t __user *wptr)
+static int kgd_hqd_load(struct amdgpu_device *adev, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr, uint32_t wptr_shift,
+			uint32_t wptr_mask, struct mm_struct *mm, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	uint32_t wptr_shadow, is_wptr_shadow_valid;
 	struct cik_mqd *m;
+	uint32_t *mqd_hqd;
+	uint32_t reg, wptr_val, data;
+	bool valid_wptr = false;
 
 	m = get_mqd(mqd);
 
-	is_wptr_shadow_valid = !get_user(wptr_shadow, wptr);
+	acquire_queue(adev, pipe_id, queue_id);
 
-	acquire_queue(kgd, pipe_id, queue_id);
-	WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
-	WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
-	WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
+	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_MQD_CONTROL. */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
 
-	WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
-	WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
-	WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
+	for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_MQD_CONTROL; reg++)
+		WREG32(reg, mqd_hqd[reg - mmCP_MQD_BASE_ADDR]);
 
-	WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
-	WREG32(mmCP_HQD_IB_BASE_ADDR, m->cp_hqd_ib_base_addr_lo);
-	WREG32(mmCP_HQD_IB_BASE_ADDR_HI, m->cp_hqd_ib_base_addr_hi);
-
-	WREG32(mmCP_HQD_IB_RPTR, m->cp_hqd_ib_rptr);
-
-	WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
-	WREG32(mmCP_HQD_SEMA_CMD, m->cp_hqd_sema_cmd);
-	WREG32(mmCP_HQD_MSG_TYPE, m->cp_hqd_msg_type);
+	/* Copy userspace write pointer value to register.
+	 * Activate doorbell logic to monitor subsequent changes.
+	 */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, data);
 
-	WREG32(mmCP_HQD_ATOMIC0_PREOP_LO, m->cp_hqd_atomic0_preop_lo);
-	WREG32(mmCP_HQD_ATOMIC0_PREOP_HI, m->cp_hqd_atomic0_preop_hi);
-	WREG32(mmCP_HQD_ATOMIC1_PREOP_LO, m->cp_hqd_atomic1_preop_lo);
-	WREG32(mmCP_HQD_ATOMIC1_PREOP_HI, m->cp_hqd_atomic1_preop_hi);
+	/* read_user_ptr may take the mm->mmap_lock.
+	 * release srbm_mutex to avoid circular dependency between
+	 * srbm_mutex->mmap_lock->reservation_ww_class_mutex->srbm_mutex.
+	 */
+	release_queue(adev);
+	valid_wptr = read_user_wptr(mm, wptr, wptr_val);
+	acquire_queue(adev, pipe_id, queue_id);
+	if (valid_wptr)
+		WREG32(mmCP_HQD_PQ_WPTR, (wptr_val << wptr_shift) & wptr_mask);
 
-	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
-	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
-			m->cp_hqd_pq_rptr_report_addr_hi);
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32(mmCP_HQD_ACTIVE, data);
 
-	WREG32(mmCP_HQD_PQ_RPTR, m->cp_hqd_pq_rptr);
+	release_queue(adev);
 
-	WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR, m->cp_hqd_pq_wptr_poll_addr_lo);
-	WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR_HI, m->cp_hqd_pq_wptr_poll_addr_hi);
+	return 0;
+}
 
-	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
+static int kgd_hqd_dump(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
+{
+	uint32_t i = 0, reg;
+#define HQD_N_REGS (35+4)
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
 
-	WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
 
-	WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
+	acquire_queue(adev, pipe_id, queue_id);
 
-	WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
-	WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE0);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE1);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE2);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE3);
 
-	WREG32(mmCP_HQD_IQ_RPTR, m->cp_hqd_iq_rptr);
+	for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_MQD_CONTROL; reg++)
+		DUMP_REG(reg);
 
-	if (is_wptr_shadow_valid)
-		WREG32(mmCP_HQD_PQ_WPTR, wptr_shadow);
+	release_queue(adev);
 
-	WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
-	release_queue(kgd);
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
 
 	return 0;
 }
 
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+static int kgd_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct cik_sdma_rlc_registers *m;
-	uint32_t sdma_base_addr;
+	unsigned long end_jiffies;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t data;
 
 	m = get_sdma_mqd(mqd);
-	sdma_base_addr = get_sdma_base_addr(m);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m);
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR,
-			m->sdma_rlc_virtual_addr);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		m->sdma_rlc_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE,
-			m->sdma_rlc_rb_base);
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI,
+	data = REG_SET_FIELD(m->sdma_rlc_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
+				m->sdma_rlc_rb_rptr);
+
+	if (read_user_wptr(mm, wptr, data))
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, data);
+	else
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdma_rlc_rb_rptr);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_VIRTUAL_ADDR,
+				m->sdma_rlc_virtual_addr);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdma_rlc_rb_base);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
 			m->sdma_rlc_rb_base_hi);
-
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
 			m->sdma_rlc_rb_rptr_addr_lo);
-
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
 			m->sdma_rlc_rb_rptr_addr_hi);
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL,
-			m->sdma_rlc_doorbell);
+	data = REG_SET_FIELD(m->sdma_rlc_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_offset = engine_id * SDMA1_REGISTER_OFFSET +
+		queue_id * KFD_CIK_SDMA_QUEUE_OFFSET;
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+4)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
-			m->sdma_rlc_rb_cntl);
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_VIRTUAL_ADDR; reg <= mmSDMA0_RLC0_WATERMARK;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
 
 	return 0;
 }
 
-static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
-				uint32_t pipe_id, uint32_t queue_id)
+static bool kgd_hqd_is_occupied(struct amdgpu_device *adev,
+				uint64_t queue_address, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t act;
 	bool retval = false;
 	uint32_t low, high;
 
-	acquire_queue(kgd, pipe_id, queue_id);
+	acquire_queue(adev, pipe_id, queue_id);
 	act = RREG32(mmCP_HQD_ACTIVE);
 	if (act) {
 		low = lower_32_bits(queue_address >> 8);
@@ -414,21 +335,20 @@ static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
 				high == RREG32(mmCP_HQD_PQ_BASE_HI))
 			retval = true;
 	}
-	release_queue(kgd);
+	release_queue(adev);
 	return retval;
 }
 
-static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
+static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct cik_sdma_rlc_registers *m;
-	uint32_t sdma_base_addr;
+	uint32_t sdma_rlc_reg_offset;
 	uint32_t sdma_rlc_rb_cntl;
 
 	m = get_sdma_mqd(mqd);
-	sdma_base_addr = get_sdma_base_addr(m);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m);
 
-	sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
 
 	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
 		return true;
@@ -436,126 +356,147 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
 	return false;
 }
 
-static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+static int kgd_hqd_destroy(struct amdgpu_device *adev, void *mqd,
+				enum kfd_preempt_type reset_type,
 				unsigned int utimeout, uint32_t pipe_id,
-				uint32_t queue_id)
+				uint32_t queue_id, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t temp;
-	int timeout = utimeout;
+	enum hqd_dequeue_request_type type;
+	unsigned long flags, end_jiffies;
+	int retry;
 
-	acquire_queue(kgd, pipe_id, queue_id);
+	if (amdgpu_in_reset(adev))
+		return -EIO;
+
+	acquire_queue(adev, pipe_id, queue_id);
 	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
 
-	WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type);
+	switch (reset_type) {
+	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
+		type = DRAIN_PIPE;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
+		type = RESET_WAVES;
+		break;
+	default:
+		type = DRAIN_PIPE;
+		break;
+	}
+
+	/* Workaround: If IQ timer is active and the wait time is close to or
+	 * equal to 0, dequeueing is not safe. Wait until either the wait time
+	 * is larger or timer is cleared. Also, ensure that IQ_REQ_PEND is
+	 * cleared before continuing. Also, ensure wait times are set to at
+	 * least 0x3.
+	 */
+	local_irq_save(flags);
+	preempt_disable();
+	retry = 5000; /* wait for 500 usecs at maximum */
+	while (true) {
+		temp = RREG32(mmCP_HQD_IQ_TIMER);
+		if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, PROCESSING_IQ)) {
+			pr_debug("HW is processing IQ\n");
+			goto loop;
+		}
+		if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, ACTIVE)) {
+			if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, RETRY_TYPE)
+					== 3) /* SEM-rearm is safe */
+				break;
+			/* Wait time 3 is safe for CP, but our MMIO read/write
+			 * time is close to 1 microsecond, so check for 10 to
+			 * leave more buffer room
+			 */
+			if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, WAIT_TIME)
+					>= 10)
+				break;
+			pr_debug("IQ timer is active\n");
+		} else
+			break;
+loop:
+		if (!retry) {
+			pr_err("CP HQD IQ timer status time out\n");
+			break;
+		}
+		ndelay(100);
+		--retry;
+	}
+	retry = 1000;
+	while (true) {
+		temp = RREG32(mmCP_HQD_DEQUEUE_REQUEST);
+		if (!(temp & CP_HQD_DEQUEUE_REQUEST__IQ_REQ_PEND_MASK))
+			break;
+		pr_debug("Dequeue request is pending\n");
 
+		if (!retry) {
+			pr_err("CP HQD dequeue request time out\n");
+			break;
+		}
+		ndelay(100);
+		--retry;
+	}
+	local_irq_restore(flags);
+	preempt_enable();
+
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, type);
+
+	end_jiffies = (utimeout * HZ / 1000) + jiffies;
 	while (true) {
 		temp = RREG32(mmCP_HQD_ACTIVE);
-		if (temp & CP_HQD_ACTIVE__ACTIVE_MASK)
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
 			break;
-		if (timeout <= 0) {
-			pr_err("kfd: cp queue preemption time out.\n");
-			release_queue(kgd);
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("cp queue preemption time out\n");
+			release_queue(adev);
 			return -ETIME;
 		}
-		msleep(20);
-		timeout -= 20;
+		usleep_range(500, 1000);
 	}
 
-	release_queue(kgd);
+	release_queue(adev);
 	return 0;
 }
 
-static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+static int kgd_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
 				unsigned int utimeout)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct cik_sdma_rlc_registers *m;
-	uint32_t sdma_base_addr;
+	uint32_t sdma_rlc_reg_offset;
 	uint32_t temp;
-	int timeout = utimeout;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
 
 	m = get_sdma_mqd(mqd);
-	sdma_base_addr = get_sdma_base_addr(m);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m);
 
-	temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+	temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
 	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
 
 	while (true) {
-		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
-		if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT)
+		temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
 			break;
-		if (timeout <= 0)
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
 			return -ETIME;
-		msleep(20);
-		timeout -= 20;
+		}
+		usleep_range(500, 1000);
 	}
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
 
-	return 0;
-}
-
-static int kgd_address_watch_disable(struct kgd_dev *kgd)
-{
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	union TCP_WATCH_CNTL_BITS cntl;
-	unsigned int i;
-
-	cntl.u32All = 0;
-
-	cntl.bitfields.valid = 0;
-	cntl.bitfields.mask = ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK;
-	cntl.bitfields.atc = 1;
-
-	/* Turning off this address until we set all the registers */
-	for (i = 0; i < MAX_WATCH_ADDRESSES; i++)
-		WREG32(watchRegs[i * ADDRESS_WATCH_REG_MAX +
-			ADDRESS_WATCH_REG_CNTL], cntl.u32All);
-
-	return 0;
-}
-
-static int kgd_address_watch_execute(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					uint32_t cntl_val,
-					uint32_t addr_hi,
-					uint32_t addr_lo)
-{
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	union TCP_WATCH_CNTL_BITS cntl;
-
-	cntl.u32All = cntl_val;
-
-	/* Turning off this watch point until we set all the registers */
-	cntl.bitfields.valid = 0;
-	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
-		ADDRESS_WATCH_REG_CNTL], cntl.u32All);
-
-	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
-		ADDRESS_WATCH_REG_ADDR_HI], addr_hi);
-
-	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
-		ADDRESS_WATCH_REG_ADDR_LO], addr_lo);
-
-	/* Enable the watch point */
-	cntl.bitfields.valid = 1;
-
-	WREG32(watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX +
-		ADDRESS_WATCH_REG_CNTL], cntl.u32All);
+	m->sdma_rlc_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
 
 	return 0;
 }
 
-static int kgd_wave_control_execute(struct kgd_dev *kgd,
+static int kgd_wave_control_execute(struct amdgpu_device *adev,
 					uint32_t gfx_index_val,
-					uint32_t sq_cmd)
+					uint32_t sq_cmd, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t data;
 
 	mutex_lock(&adev->grbm_idx_mutex);
@@ -576,96 +517,73 @@ static int kgd_wave_control_execute(struct kgd_dev *kgd,
 	return 0;
 }
 
-static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					unsigned int reg_offset)
+static bool get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
 {
-	return watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + reg_offset];
-}
+	uint32_t value;
 
-static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
-							uint8_t vmid)
-{
-	uint32_t reg;
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	value = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
 
-	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
-	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
 }
 
-static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
-								uint8_t vmid)
+static void set_scratch_backing_va(struct amdgpu_device *adev,
+					uint64_t va, uint32_t vmid)
 {
-	uint32_t reg;
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-
-	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
-	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+	lock_srbm(adev, 0, 0, 0, vmid);
+	WREG32(mmSH_HIDDEN_PRIVATE_BASE_VMID, va);
+	unlock_srbm(adev);
 }
 
-static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+static void set_vm_context_page_table_base(struct amdgpu_device *adev,
+			uint32_t vmid, uint64_t page_table_base)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-
-	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("trying to set page table base for wrong VMID\n");
+		return;
+	}
+	WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8,
+		lower_32_bits(page_table_base));
 }
 
-static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
+ /**
+  * read_vmid_from_vmfault_reg - read vmid from register
+  *
+  * adev: amdgpu_device pointer
+  * @vmid: vmid pointer
+  * read vmid from register (CIK).
+  */
+static uint32_t read_vmid_from_vmfault_reg(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	const union amdgpu_firmware_header *hdr;
-
-	BUG_ON(kgd == NULL);
-
-	switch (type) {
-	case KGD_ENGINE_PFP:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.pfp_fw->data;
-		break;
-
-	case KGD_ENGINE_ME:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.me_fw->data;
-		break;
-
-	case KGD_ENGINE_CE:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.ce_fw->data;
-		break;
-
-	case KGD_ENGINE_MEC1:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.mec_fw->data;
-		break;
-
-	case KGD_ENGINE_MEC2:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.mec2_fw->data;
-		break;
-
-	case KGD_ENGINE_RLC:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.rlc_fw->data;
-		break;
-
-	case KGD_ENGINE_SDMA1:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->sdma.instance[0].fw->data;
-		break;
+	uint32_t status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
 
-	case KGD_ENGINE_SDMA2:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->sdma.instance[1].fw->data;
-		break;
-
-	default:
-		return 0;
-	}
+	return REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
+}
 
-	if (hdr == NULL)
-		return 0;
+static uint32_t kgd_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+					  int engine, int queue)
 
-	/* Only 12 bit in use*/
-	return hdr->common.ucode_version;
+{
+	return 0;
 }
 
+const struct kfd2kgd_calls gfx_v7_kfd2kgd = {
+	.program_sh_mem_settings = kgd_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_init_interrupts,
+	.hqd_load = kgd_hqd_load,
+	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_dump = kgd_hqd_dump,
+	.hqd_sdma_dump = kgd_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_hqd_destroy,
+	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+	.wave_control_execute = kgd_wave_control_execute,
+	.get_atc_vmid_pasid_mapping_info = get_atc_vmid_pasid_mapping_info,
+	.set_scratch_backing_va = set_scratch_backing_va,
+	.set_vm_context_page_table_base = set_vm_context_page_table_base,
+	.read_vmid_from_vmfault_reg = read_vmid_from_vmfault_reg,
+	.hqd_sdma_get_doorbell = kgd_hqd_sdma_get_doorbell,
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
index 6697612239c2..e68c0fa8d751 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
@@ -20,14 +20,9 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  */
 
-#include <linux/module.h>
-#include <linux/fdtable.h>
-#include <linux/uaccess.h>
-#include <linux/firmware.h>
-#include <drm/drmP.h>
 #include "amdgpu.h"
 #include "amdgpu_amdkfd.h"
-#include "amdgpu_ucode.h"
+#include "gfx_v8_0.h"
 #include "gca/gfx_8_0_sh_mask.h"
 #include "gca/gfx_8_0_d.h"
 #include "gca/gfx_8_0_enum.h"
@@ -38,148 +33,60 @@
 #include "vi_structs.h"
 #include "vid.h"
 
-#define VI_PIPE_PER_MEC	(4)
-
-struct cik_sdma_rlc_registers;
-
-/*
- * Register access functions
- */
-
-static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
-		uint32_t sh_mem_config,
-		uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit,
-		uint32_t sh_mem_bases);
-static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
-		unsigned int vmid);
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-		uint32_t hpd_size, uint64_t hpd_gpu_addr);
-static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
-static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
-		uint32_t queue_id, uint32_t __user *wptr);
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
-static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
-		uint32_t pipe_id, uint32_t queue_id);
-static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
-static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
-				unsigned int utimeout, uint32_t pipe_id,
-				uint32_t queue_id);
-static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
-				unsigned int utimeout);
-static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
-static int kgd_address_watch_disable(struct kgd_dev *kgd);
-static int kgd_address_watch_execute(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					uint32_t cntl_val,
-					uint32_t addr_hi,
-					uint32_t addr_lo);
-static int kgd_wave_control_execute(struct kgd_dev *kgd,
-					uint32_t gfx_index_val,
-					uint32_t sq_cmd);
-static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					unsigned int reg_offset);
-
-static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
-		uint8_t vmid);
-static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
-		uint8_t vmid);
-static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
-static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
-
-static const struct kfd2kgd_calls kfd2kgd = {
-	.init_gtt_mem_allocation = alloc_gtt_mem,
-	.free_gtt_mem = free_gtt_mem,
-	.get_vmem_size = get_vmem_size,
-	.get_gpu_clock_counter = get_gpu_clock_counter,
-	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
-	.program_sh_mem_settings = kgd_program_sh_mem_settings,
-	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
-	.init_pipeline = kgd_init_pipeline,
-	.init_interrupts = kgd_init_interrupts,
-	.hqd_load = kgd_hqd_load,
-	.hqd_sdma_load = kgd_hqd_sdma_load,
-	.hqd_is_occupied = kgd_hqd_is_occupied,
-	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
-	.hqd_destroy = kgd_hqd_destroy,
-	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
-	.address_watch_disable = kgd_address_watch_disable,
-	.address_watch_execute = kgd_address_watch_execute,
-	.wave_control_execute = kgd_wave_control_execute,
-	.address_watch_get_offset = kgd_address_watch_get_offset,
-	.get_atc_vmid_pasid_mapping_pasid =
-			get_atc_vmid_pasid_mapping_pasid,
-	.get_atc_vmid_pasid_mapping_valid =
-			get_atc_vmid_pasid_mapping_valid,
-	.write_vmid_invalidate_request = write_vmid_invalidate_request,
-	.get_fw_version = get_fw_version
+enum hqd_dequeue_request_type {
+	NO_ACTION = 0,
+	DRAIN_PIPE,
+	RESET_WAVES
 };
 
-struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void)
-{
-	return (struct kfd2kgd_calls *)&kfd2kgd;
-}
-
-static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
-{
-	return (struct amdgpu_device *)kgd;
-}
-
-static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
 			uint32_t queue, uint32_t vmid)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
 
 	mutex_lock(&adev->srbm_mutex);
 	WREG32(mmSRBM_GFX_CNTL, value);
 }
 
-static void unlock_srbm(struct kgd_dev *kgd)
+static void unlock_srbm(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
 	WREG32(mmSRBM_GFX_CNTL, 0);
 	mutex_unlock(&adev->srbm_mutex);
 }
 
-static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+static void acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
 				uint32_t queue_id)
 {
-	uint32_t mec = (++pipe_id / VI_PIPE_PER_MEC) + 1;
-	uint32_t pipe = (pipe_id % VI_PIPE_PER_MEC);
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
 
-	lock_srbm(kgd, mec, pipe, queue_id, 0);
+	lock_srbm(adev, mec, pipe, queue_id, 0);
 }
 
-static void release_queue(struct kgd_dev *kgd)
+static void release_queue(struct amdgpu_device *adev)
 {
-	unlock_srbm(kgd);
+	unlock_srbm(adev);
 }
 
-static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
 					uint32_t sh_mem_config,
 					uint32_t sh_mem_ape1_base,
 					uint32_t sh_mem_ape1_limit,
-					uint32_t sh_mem_bases)
+					uint32_t sh_mem_bases, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
-	lock_srbm(kgd, 0, 0, 0, vmid);
+	lock_srbm(adev, 0, 0, 0, vmid);
 
 	WREG32(mmSH_MEM_CONFIG, sh_mem_config);
 	WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
 	WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
 	WREG32(mmSH_MEM_BASES, sh_mem_bases);
 
-	unlock_srbm(kgd);
+	unlock_srbm(adev);
 }
 
-static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
-					unsigned int vmid)
+static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
+					unsigned int vmid, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
 	/*
 	 * We have to assume that there is no outstanding mapping.
 	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
@@ -202,33 +109,36 @@ static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
 	return 0;
 }
 
-static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
-				uint32_t hpd_size, uint64_t hpd_gpu_addr)
-{
-	return 0;
-}
-
-static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
+static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t mec;
 	uint32_t pipe;
 
-	mec = (++pipe_id / VI_PIPE_PER_MEC) + 1;
-	pipe = (pipe_id % VI_PIPE_PER_MEC);
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
 
-	lock_srbm(kgd, mec, pipe, 0, 0);
+	lock_srbm(adev, mec, pipe, 0, 0);
 
-	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK);
+	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+			CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
 
-	unlock_srbm(kgd);
+	unlock_srbm(adev);
 
 	return 0;
 }
 
-static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+static inline uint32_t get_sdma_rlc_reg_offset(struct vi_sdma_mqd *m)
 {
-	return 0;
+	uint32_t retval;
+
+	retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET +
+		m->sdma_queue_id * KFD_VI_SDMA_QUEUE_OFFSET;
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n",
+			m->sdma_engine_id, m->sdma_queue_id, retval);
+
+	return retval;
 }
 
 static inline struct vi_mqd *get_mqd(void *mqd)
@@ -236,87 +146,218 @@ static inline struct vi_mqd *get_mqd(void *mqd)
 	return (struct vi_mqd *)mqd;
 }
 
-static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+static inline struct vi_sdma_mqd *get_sdma_mqd(void *mqd)
 {
-	return (struct cik_sdma_rlc_registers *)mqd;
+	return (struct vi_sdma_mqd *)mqd;
 }
 
-static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
-			uint32_t queue_id, uint32_t __user *wptr)
+static int kgd_hqd_load(struct amdgpu_device *adev, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr, uint32_t wptr_shift,
+			uint32_t wptr_mask, struct mm_struct *mm, uint32_t inst)
 {
 	struct vi_mqd *m;
-	uint32_t shadow_wptr, valid_wptr;
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t *mqd_hqd;
+	uint32_t reg, wptr_val, data;
+	bool valid_wptr = false;
 
 	m = get_mqd(mqd);
 
-	valid_wptr = copy_from_user(&shadow_wptr, wptr, sizeof(shadow_wptr));
-	acquire_queue(kgd, pipe_id, queue_id);
+	acquire_queue(adev, pipe_id, queue_id);
 
-	WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
-	WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
-	WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
+	/* HIQ is set during driver init period with vmid set to 0*/
+	if (m->cp_hqd_vmid == 0) {
+		uint32_t value, mec, pipe;
 
-	WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
-	WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
-	WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
-	WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
-	WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
-	WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
-	WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
-	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
-	WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
-			m->cp_hqd_pq_rptr_report_addr_hi);
+		mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+		pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
 
-	if (valid_wptr > 0)
-		WREG32(mmCP_HQD_PQ_WPTR, shadow_wptr);
+		pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+			mec, pipe, queue_id);
+		value = RREG32(mmRLC_CP_SCHEDULERS);
+		value = REG_SET_FIELD(value, RLC_CP_SCHEDULERS, scheduler1,
+			((mec << 5) | (pipe << 3) | queue_id | 0x80));
+		WREG32(mmRLC_CP_SCHEDULERS, value);
+	}
 
-	WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
-	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
+	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
 
-	WREG32(mmCP_HQD_EOP_BASE_ADDR, m->cp_hqd_eop_base_addr_lo);
-	WREG32(mmCP_HQD_EOP_BASE_ADDR_HI, m->cp_hqd_eop_base_addr_hi);
-	WREG32(mmCP_HQD_EOP_CONTROL, m->cp_hqd_eop_control);
-	WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr);
-	WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr);
-	WREG32(mmCP_HQD_EOP_EVENTS, m->cp_hqd_eop_done_events);
+	for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_HQD_EOP_CONTROL; reg++)
+		WREG32(reg, mqd_hqd[reg - mmCP_MQD_BASE_ADDR]);
 
-	WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO, m->cp_hqd_ctx_save_base_addr_lo);
-	WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI, m->cp_hqd_ctx_save_base_addr_hi);
-	WREG32(mmCP_HQD_CTX_SAVE_CONTROL, m->cp_hqd_ctx_save_control);
-	WREG32(mmCP_HQD_CNTL_STACK_OFFSET, m->cp_hqd_cntl_stack_offset);
-	WREG32(mmCP_HQD_CNTL_STACK_SIZE, m->cp_hqd_cntl_stack_size);
-	WREG32(mmCP_HQD_WG_STATE_OFFSET, m->cp_hqd_wg_state_offset);
-	WREG32(mmCP_HQD_CTX_SAVE_SIZE, m->cp_hqd_ctx_save_size);
+	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
+	 * This is safe since EOP RPTR==WPTR for any inactive HQD
+	 * on ASICs that do not support context-save.
+	 * EOP writes/reads can start anywhere in the ring.
+	 */
+	if (adev->asic_type != CHIP_TONGA) {
+		WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr);
+		WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr);
+		WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem);
+	}
 
-	WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
+	for (reg = mmCP_HQD_EOP_EVENTS; reg <= mmCP_HQD_ERROR; reg++)
+		WREG32(reg, mqd_hqd[reg - mmCP_MQD_BASE_ADDR]);
 
-	WREG32(mmCP_HQD_DEQUEUE_REQUEST, m->cp_hqd_dequeue_request);
-	WREG32(mmCP_HQD_ERROR, m->cp_hqd_error);
-	WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem);
-	WREG32(mmCP_HQD_EOP_DONES, m->cp_hqd_eop_dones);
+	/* Copy userspace write pointer value to register.
+	 * Activate doorbell logic to monitor subsequent changes.
+	 */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, data);
+
+	/* read_user_ptr may take the mm->mmap_lock.
+	 * release srbm_mutex to avoid circular dependency between
+	 * srbm_mutex->mmap_lock->reservation_ww_class_mutex->srbm_mutex.
+	 */
+	release_queue(adev);
+	valid_wptr = read_user_wptr(mm, wptr, wptr_val);
+	acquire_queue(adev, pipe_id, queue_id);
+	if (valid_wptr)
+		WREG32(mmCP_HQD_PQ_WPTR, (wptr_val << wptr_shift) & wptr_mask);
 
-	WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32(mmCP_HQD_ACTIVE, data);
 
-	release_queue(kgd);
+	release_queue(adev);
 
 	return 0;
 }
 
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+static int kgd_hqd_dump(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
 {
+	uint32_t i = 0, reg;
+#define HQD_N_REGS (54+4)
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE0);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE1);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE2);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE3);
+
+	for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_HQD_EOP_DONES; reg++)
+		DUMP_REG(reg);
+
+	release_queue(adev);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
 	return 0;
 }
 
-static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
-				uint32_t pipe_id, uint32_t queue_id)
+static int kgd_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct vi_sdma_mqd *m;
+	unsigned long end_jiffies;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t data;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
+				m->sdmax_rlcx_rb_rptr);
+
+	if (read_user_wptr(mm, wptr, data))
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, data);
+	else
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_VIRTUAL_ADDR,
+				m->sdmax_rlcx_virtual_addr);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_offset = engine_id * SDMA1_REGISTER_OFFSET +
+		queue_id * KFD_VI_SDMA_QUEUE_OFFSET;
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+4+2+3+7)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_VIRTUAL_ADDR; reg <= mmSDMA0_RLC0_WATERMARK;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_CSA_ADDR_LO; reg <= mmSDMA0_RLC0_CSA_ADDR_HI;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN; reg <= mmSDMA0_RLC0_DUMMY_REG;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0; reg <= mmSDMA0_RLC0_MIDCMD_CNTL;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static bool kgd_hqd_is_occupied(struct amdgpu_device *adev,
+				uint64_t queue_address, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t act;
 	bool retval = false;
 	uint32_t low, high;
 
-	acquire_queue(kgd, pipe_id, queue_id);
+	acquire_queue(adev, pipe_id, queue_id);
 	act = RREG32(mmCP_HQD_ACTIVE);
 	if (act) {
 		low = lower_32_bits(queue_address >> 8);
@@ -326,21 +367,20 @@ static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
 				high == RREG32(mmCP_HQD_PQ_BASE_HI))
 			retval = true;
 	}
-	release_queue(kgd);
+	release_queue(adev);
 	return retval;
 }
 
-static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
+static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct cik_sdma_rlc_registers *m;
-	uint32_t sdma_base_addr;
+	struct vi_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
 	uint32_t sdma_rlc_rb_cntl;
 
 	m = get_sdma_mqd(mqd);
-	sdma_base_addr = get_sdma_base_addr(m);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m);
 
-	sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
 
 	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
 		return true;
@@ -348,115 +388,161 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
 	return false;
 }
 
-static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+static int kgd_hqd_destroy(struct amdgpu_device *adev, void *mqd,
+				enum kfd_preempt_type reset_type,
 				unsigned int utimeout, uint32_t pipe_id,
-				uint32_t queue_id)
+				uint32_t queue_id, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t temp;
-	int timeout = utimeout;
+	enum hqd_dequeue_request_type type;
+	unsigned long flags, end_jiffies;
+	int retry;
+	struct vi_mqd *m = get_mqd(mqd);
+
+	if (amdgpu_in_reset(adev))
+		return -EIO;
+
+	acquire_queue(adev, pipe_id, queue_id);
+
+	if (m->cp_hqd_vmid == 0)
+		WREG32_FIELD(RLC_CP_SCHEDULERS, scheduler1, 0);
 
-	acquire_queue(kgd, pipe_id, queue_id);
+	switch (reset_type) {
+	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
+		type = DRAIN_PIPE;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
+		type = RESET_WAVES;
+		break;
+	default:
+		type = DRAIN_PIPE;
+		break;
+	}
+
+	/* Workaround: If IQ timer is active and the wait time is close to or
+	 * equal to 0, dequeueing is not safe. Wait until either the wait time
+	 * is larger or timer is cleared. Also, ensure that IQ_REQ_PEND is
+	 * cleared before continuing. Also, ensure wait times are set to at
+	 * least 0x3.
+	 */
+	local_irq_save(flags);
+	preempt_disable();
+	retry = 5000; /* wait for 500 usecs at maximum */
+	while (true) {
+		temp = RREG32(mmCP_HQD_IQ_TIMER);
+		if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, PROCESSING_IQ)) {
+			pr_debug("HW is processing IQ\n");
+			goto loop;
+		}
+		if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, ACTIVE)) {
+			if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, RETRY_TYPE)
+					== 3) /* SEM-rearm is safe */
+				break;
+			/* Wait time 3 is safe for CP, but our MMIO read/write
+			 * time is close to 1 microsecond, so check for 10 to
+			 * leave more buffer room
+			 */
+			if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, WAIT_TIME)
+					>= 10)
+				break;
+			pr_debug("IQ timer is active\n");
+		} else
+			break;
+loop:
+		if (!retry) {
+			pr_err("CP HQD IQ timer status time out\n");
+			break;
+		}
+		ndelay(100);
+		--retry;
+	}
+	retry = 1000;
+	while (true) {
+		temp = RREG32(mmCP_HQD_DEQUEUE_REQUEST);
+		if (!(temp & CP_HQD_DEQUEUE_REQUEST__IQ_REQ_PEND_MASK))
+			break;
+		pr_debug("Dequeue request is pending\n");
+
+		if (!retry) {
+			pr_err("CP HQD dequeue request time out\n");
+			break;
+		}
+		ndelay(100);
+		--retry;
+	}
+	local_irq_restore(flags);
+	preempt_enable();
 
-	WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type);
+	WREG32(mmCP_HQD_DEQUEUE_REQUEST, type);
 
+	end_jiffies = (utimeout * HZ / 1000) + jiffies;
 	while (true) {
 		temp = RREG32(mmCP_HQD_ACTIVE);
-		if (temp & CP_HQD_ACTIVE__ACTIVE_MASK)
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
 			break;
-		if (timeout <= 0) {
-			pr_err("kfd: cp queue preemption time out.\n");
-			release_queue(kgd);
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("cp queue preemption time out.\n");
+			release_queue(adev);
 			return -ETIME;
 		}
-		msleep(20);
-		timeout -= 20;
+		usleep_range(500, 1000);
 	}
 
-	release_queue(kgd);
+	release_queue(adev);
 	return 0;
 }
 
-static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+static int kgd_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
 				unsigned int utimeout)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct cik_sdma_rlc_registers *m;
-	uint32_t sdma_base_addr;
+	struct vi_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
 	uint32_t temp;
-	int timeout = utimeout;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
 
 	m = get_sdma_mqd(mqd);
-	sdma_base_addr = get_sdma_base_addr(m);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m);
 
-	temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+	temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
 	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
 
 	while (true) {
-		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
-		if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT)
+		temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
 			break;
-		if (timeout <= 0)
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
 			return -ETIME;
-		msleep(20);
-		timeout -= 20;
+		}
+		usleep_range(500, 1000);
 	}
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
-
-	return 0;
-}
-
-static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
-							uint8_t vmid)
-{
-	uint32_t reg;
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-
-	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
-	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
-}
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
 
-static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
-								uint8_t vmid)
-{
-	uint32_t reg;
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
 
-	reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
-	return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+	return 0;
 }
 
-static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+static bool get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+	uint32_t value;
 
-	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
-}
+	value = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
 
-static int kgd_address_watch_disable(struct kgd_dev *kgd)
-{
-	return 0;
-}
-
-static int kgd_address_watch_execute(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					uint32_t cntl_val,
-					uint32_t addr_hi,
-					uint32_t addr_lo)
-{
-	return 0;
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
 }
 
-static int kgd_wave_control_execute(struct kgd_dev *kgd,
+static int kgd_wave_control_execute(struct amdgpu_device *adev,
 					uint32_t gfx_index_val,
-					uint32_t sq_cmd)
+					uint32_t sq_cmd, uint32_t inst)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	uint32_t data = 0;
 
 	mutex_lock(&adev->grbm_idx_mutex);
@@ -477,68 +563,48 @@ static int kgd_wave_control_execute(struct kgd_dev *kgd,
 	return 0;
 }
 
-static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
-					unsigned int watch_point_id,
-					unsigned int reg_offset)
+static void set_scratch_backing_va(struct amdgpu_device *adev,
+					uint64_t va, uint32_t vmid)
 {
-	return 0;
+	lock_srbm(adev, 0, 0, 0, vmid);
+	WREG32(mmSH_HIDDEN_PRIVATE_BASE_VMID, va);
+	unlock_srbm(adev);
 }
 
-static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
+static void set_vm_context_page_table_base(struct amdgpu_device *adev,
+		uint32_t vmid, uint64_t page_table_base)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	const union amdgpu_firmware_header *hdr;
-
-	BUG_ON(kgd == NULL);
-
-	switch (type) {
-	case KGD_ENGINE_PFP:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.pfp_fw->data;
-		break;
-
-	case KGD_ENGINE_ME:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.me_fw->data;
-		break;
-
-	case KGD_ENGINE_CE:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.ce_fw->data;
-		break;
-
-	case KGD_ENGINE_MEC1:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.mec_fw->data;
-		break;
-
-	case KGD_ENGINE_MEC2:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.mec2_fw->data;
-		break;
-
-	case KGD_ENGINE_RLC:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->gfx.rlc_fw->data;
-		break;
-
-	case KGD_ENGINE_SDMA1:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->sdma.instance[0].fw->data;
-		break;
-
-	case KGD_ENGINE_SDMA2:
-		hdr = (const union amdgpu_firmware_header *)
-							adev->sdma.instance[1].fw->data;
-		break;
-
-	default:
-		return 0;
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("trying to set page table base for wrong VMID\n");
+		return;
 	}
+	WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8,
+			lower_32_bits(page_table_base));
+}
 
-	if (hdr == NULL)
-		return 0;
+static uint32_t kgd_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+					  int engine, int queue)
 
-	/* Only 12 bit in use*/
-	return hdr->common.ucode_version;
+{
+	return 0;
 }
+
+const struct kfd2kgd_calls gfx_v8_kfd2kgd = {
+	.program_sh_mem_settings = kgd_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_init_interrupts,
+	.hqd_load = kgd_hqd_load,
+	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_dump = kgd_hqd_dump,
+	.hqd_sdma_dump = kgd_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_hqd_destroy,
+	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+	.wave_control_execute = kgd_wave_control_execute,
+	.get_atc_vmid_pasid_mapping_info =
+			get_atc_vmid_pasid_mapping_info,
+	.set_scratch_backing_va = set_scratch_backing_va,
+	.set_vm_context_page_table_base = set_vm_context_page_table_base,
+	.hqd_sdma_get_doorbell = kgd_hqd_sdma_get_doorbell,
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
new file mode 100644
index 000000000000..088d09cc7a72
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -0,0 +1,1264 @@
+/*
+ * Copyright 2014-2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "gc/gc_9_0_offset.h"
+#include "gc/gc_9_0_sh_mask.h"
+#include "vega10_enum.h"
+#include "sdma0/sdma0_4_0_offset.h"
+#include "sdma0/sdma0_4_0_sh_mask.h"
+#include "sdma1/sdma1_4_0_offset.h"
+#include "sdma1/sdma1_4_0_sh_mask.h"
+#include "athub/athub_1_0_offset.h"
+#include "athub/athub_1_0_sh_mask.h"
+#include "oss/osssys_4_0_offset.h"
+#include "oss/osssys_4_0_sh_mask.h"
+#include "soc15_common.h"
+#include "v9_structs.h"
+#include "soc15.h"
+#include "soc15d.h"
+#include "gfx_v9_0.h"
+#include "amdgpu_amdkfd_gfx_v9.h"
+#include <uapi/linux/kfd_ioctl.h>
+
+enum hqd_dequeue_request_type {
+	NO_ACTION = 0,
+	DRAIN_PIPE,
+	RESET_WAVES,
+	SAVE_WAVES
+};
+
+static void kgd_gfx_v9_lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid, uint32_t inst)
+{
+	mutex_lock(&adev->srbm_mutex);
+	soc15_grbm_select(adev, mec, pipe, queue, vmid, GET_INST(GC, inst));
+}
+
+static void kgd_gfx_v9_unlock_srbm(struct amdgpu_device *adev, uint32_t inst)
+{
+	soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, inst));
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+void kgd_gfx_v9_acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	kgd_gfx_v9_lock_srbm(adev, mec, pipe, queue_id, 0, inst);
+}
+
+uint64_t kgd_gfx_v9_get_queue_mask(struct amdgpu_device *adev,
+			       uint32_t pipe_id, uint32_t queue_id)
+{
+	unsigned int bit = pipe_id * adev->gfx.mec.num_queue_per_pipe +
+			queue_id;
+
+	return 1ull << bit;
+}
+
+void kgd_gfx_v9_release_queue(struct amdgpu_device *adev, uint32_t inst)
+{
+	kgd_gfx_v9_unlock_srbm(adev, inst);
+}
+
+void kgd_gfx_v9_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases, uint32_t inst)
+{
+	kgd_gfx_v9_lock_srbm(adev, 0, 0, 0, vmid, inst);
+
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmSH_MEM_CONFIG, sh_mem_config);
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmSH_MEM_BASES, sh_mem_bases);
+	/* APE1 no longer exists on GFX9 */
+
+	kgd_gfx_v9_unlock_srbm(adev, inst);
+}
+
+int kgd_gfx_v9_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
+					unsigned int vmid, uint32_t inst)
+{
+	/*
+	 * We have to assume that there is no outstanding mapping.
+	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
+	 * a mapping is in progress or because a mapping finished
+	 * and the SW cleared it.
+	 * So the protocol is to always wait & clear.
+	 */
+	uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
+			ATC_VMID0_PASID_MAPPING__VALID_MASK;
+
+	/*
+	 * need to do this twice, once for gfx and once for mmhub
+	 * for ATC add 16 to VMID for mmhub, for IH different registers.
+	 * ATC_VMID0..15 registers are separate from ATC_VMID16..31.
+	 */
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING) + vmid,
+	       pasid_mapping);
+
+	while (!(RREG32(SOC15_REG_OFFSET(
+				ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS)) &
+		 (1U << vmid)))
+		cpu_relax();
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS),
+	       1U << vmid);
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT) + vmid,
+	       pasid_mapping);
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID16_PASID_MAPPING) + vmid,
+	       pasid_mapping);
+
+	while (!(RREG32(SOC15_REG_OFFSET(
+				ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS)) &
+		 (1U << (vmid + 16))))
+		cpu_relax();
+
+	WREG32(SOC15_REG_OFFSET(ATHUB, 0,
+				mmATC_VMID_PASID_MAPPING_UPDATE_STATUS),
+	       1U << (vmid + 16));
+
+	/* Mapping vmid to pasid also for IH block */
+	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT_MM) + vmid,
+	       pasid_mapping);
+	return 0;
+}
+
+/* TODO - RING0 form of field is obsolete, seems to date back to SI
+ * but still works
+ */
+
+int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst)
+{
+	uint32_t mec;
+	uint32_t pipe;
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	kgd_gfx_v9_lock_srbm(adev, mec, pipe, 0, 0, inst);
+
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmCPC_INT_CNTL,
+		CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+		CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
+
+	kgd_gfx_v9_unlock_srbm(adev, inst);
+
+	return 0;
+}
+
+static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev,
+				unsigned int engine_id,
+				unsigned int queue_id)
+{
+	uint32_t sdma_engine_reg_base = 0;
+	uint32_t sdma_rlc_reg_offset;
+
+	switch (engine_id) {
+	default:
+		dev_warn(adev->dev,
+			 "Invalid sdma engine id (%d), using engine id 0\n",
+			 engine_id);
+		fallthrough;
+	case 0:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0,
+				mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
+		break;
+	case 1:
+		sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA1, 0,
+				mmSDMA1_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL;
+		break;
+	}
+
+	sdma_rlc_reg_offset = sdma_engine_reg_base
+		+ queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL);
+
+	pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id,
+		 queue_id, sdma_rlc_reg_offset);
+
+	return sdma_rlc_reg_offset;
+}
+
+static inline struct v9_mqd *get_mqd(void *mqd)
+{
+	return (struct v9_mqd *)mqd;
+}
+
+static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct v9_sdma_mqd *)mqd;
+}
+
+int kgd_gfx_v9_hqd_load(struct amdgpu_device *adev, void *mqd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t __user *wptr, uint32_t wptr_shift,
+			uint32_t wptr_mask, struct mm_struct *mm,
+			uint32_t inst)
+{
+	struct v9_mqd *m;
+	uint32_t *mqd_hqd;
+	uint32_t reg, hqd_base, data;
+
+	m = get_mqd(mqd);
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+
+	/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
+	mqd_hqd = &m->cp_mqd_base_addr_lo;
+	hqd_base = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_MQD_BASE_ADDR);
+
+	for (reg = hqd_base;
+	     reg <= SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_HI); reg++)
+		WREG32_XCC(reg, mqd_hqd[reg - hqd_base], inst);
+
+
+	/* Activate doorbell logic before triggering WPTR poll. */
+	data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
+			     CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_DOORBELL_CONTROL, data);
+
+	if (wptr) {
+		/* Don't read wptr with get_user because the user
+		 * context may not be accessible (if this function
+		 * runs in a work queue). Instead trigger a one-shot
+		 * polling read from memory in the CP. This assumes
+		 * that wptr is GPU-accessible in the queue's VMID via
+		 * ATC or SVM. WPTR==RPTR before starting the poll so
+		 * the CP starts fetching new commands from the right
+		 * place.
+		 *
+		 * Guessing a 64-bit WPTR from a 32-bit RPTR is a bit
+		 * tricky. Assume that the queue didn't overflow. The
+		 * number of valid bits in the 32-bit RPTR depends on
+		 * the queue size. The remaining bits are taken from
+		 * the saved 64-bit WPTR. If the WPTR wrapped, add the
+		 * queue size.
+		 */
+		uint32_t queue_size =
+			2 << REG_GET_FIELD(m->cp_hqd_pq_control,
+					   CP_HQD_PQ_CONTROL, QUEUE_SIZE);
+		uint64_t guessed_wptr = m->cp_hqd_pq_rptr & (queue_size - 1);
+
+		if ((m->cp_hqd_pq_wptr_lo & (queue_size - 1)) < guessed_wptr)
+			guessed_wptr += queue_size;
+		guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1);
+		guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32;
+
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_LO,
+			lower_32_bits(guessed_wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_HI,
+			upper_32_bits(guessed_wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_POLL_ADDR,
+			lower_32_bits((uintptr_t)wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
+			upper_32_bits((uintptr_t)wptr));
+		WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_PQ_WPTR_POLL_CNTL1,
+			(uint32_t)kgd_gfx_v9_get_queue_mask(adev, pipe_id, queue_id));
+	}
+
+	/* Start the EOP fetcher */
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_EOP_RPTR,
+	       REG_SET_FIELD(m->cp_hqd_eop_rptr, CP_HQD_EOP_RPTR, INIT_FETCHER, 1));
+
+	data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE, data);
+
+	kgd_gfx_v9_release_queue(adev, inst);
+
+	return 0;
+}
+
+int kgd_gfx_v9_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    uint32_t doorbell_off, uint32_t inst)
+{
+	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[inst].ring;
+	struct v9_mqd *m;
+	uint32_t mec, pipe;
+	int r;
+
+	m = get_mqd(mqd);
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+
+	mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
+	pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
+
+	pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
+		 mec, pipe, queue_id);
+
+	spin_lock(&adev->gfx.kiq[inst].ring_lock);
+	r = amdgpu_ring_alloc(kiq_ring, 7);
+	if (r) {
+		pr_err("Failed to alloc KIQ (%d).\n", r);
+		goto out_unlock;
+	}
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
+			  PACKET3_MAP_QUEUES_VMID(m->cp_hqd_vmid) | /* VMID */
+			  PACKET3_MAP_QUEUES_QUEUE(queue_id) |
+			  PACKET3_MAP_QUEUES_PIPE(pipe) |
+			  PACKET3_MAP_QUEUES_ME((mec - 1)) |
+			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
+			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
+			  PACKET3_MAP_QUEUES_ENGINE_SEL(1) | /* engine_sel: hiq */
+			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(doorbell_off));
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_hi);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_lo);
+	amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_hi);
+	amdgpu_ring_commit(kiq_ring);
+
+out_unlock:
+	spin_unlock(&adev->gfx.kiq[inst].ring_lock);
+	kgd_gfx_v9_release_queue(adev, inst);
+
+	return r;
+}
+
+int kgd_gfx_v9_hqd_dump(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst)
+{
+	uint32_t i = 0, reg;
+#define HQD_N_REGS 56
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+
+	for (reg = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_MQD_BASE_ADDR);
+	     reg <= SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_HI); reg++)
+		DUMP_REG(reg);
+
+	kgd_gfx_v9_release_queue(adev, inst);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	unsigned long end_jiffies;
+	uint32_t data;
+	uint64_t data64;
+	uint64_t __user *wptr64 = (uint64_t __user *)wptr;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET,
+	       m->sdmax_rlcx_doorbell_offset);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR,
+				m->sdmax_rlcx_rb_rptr);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI,
+				m->sdmax_rlcx_rb_rptr_hi);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1);
+	if (read_user_wptr(mm, wptr64, data64)) {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       lower_32_bits(data64));
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       upper_32_bits(data64));
+	} else {
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+		WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI,
+		       m->sdmax_rlcx_rb_rptr_hi);
+	}
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0);
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev,
+			engine_id, queue_id);
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+6+7+10)
+
+	*dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN;
+	     reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0;
+	     reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++)
+		DUMP_REG(sdma_rlc_reg_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+bool kgd_gfx_v9_hqd_is_occupied(struct amdgpu_device *adev,
+				uint64_t queue_address, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+	act = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE) &&
+		   high == RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE_HI))
+			retval = true;
+	}
+	kgd_gfx_v9_release_queue(adev, inst);
+	return retval;
+}
+
+static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t sdma_rlc_rb_cntl;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+
+	if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+		return true;
+
+	return false;
+}
+
+int kgd_gfx_v9_hqd_destroy(struct amdgpu_device *adev, void *mqd,
+				enum kfd_preempt_type reset_type,
+				unsigned int utimeout, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst)
+{
+	enum hqd_dequeue_request_type type;
+	unsigned long end_jiffies;
+	uint32_t temp;
+	struct v9_mqd *m = get_mqd(mqd);
+
+	if (amdgpu_in_reset(adev))
+		return -EIO;
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+
+	if (m->cp_hqd_vmid == 0)
+		WREG32_FIELD15_RLC(GC, GET_INST(GC, inst), RLC_CP_SCHEDULERS, scheduler1, 0);
+
+	switch (reset_type) {
+	case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
+		type = DRAIN_PIPE;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
+		type = RESET_WAVES;
+		break;
+	case KFD_PREEMPT_TYPE_WAVEFRONT_SAVE:
+		type = SAVE_WAVES;
+		break;
+	default:
+		type = DRAIN_PIPE;
+		break;
+	}
+
+	WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_DEQUEUE_REQUEST, type);
+
+	end_jiffies = (utimeout * HZ / 1000) + jiffies;
+	while (true) {
+		temp = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE);
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("cp queue preemption time out.\n");
+			kgd_gfx_v9_release_queue(adev, inst);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	kgd_gfx_v9_release_queue(adev, inst);
+	return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
+				unsigned int utimeout)
+{
+	struct v9_sdma_mqd *m;
+	uint32_t sdma_rlc_reg_offset;
+	uint32_t temp;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	m = get_sdma_mqd(mqd);
+	sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id,
+					    m->sdma_queue_id);
+
+	temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL);
+	temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp);
+
+	while (true) {
+		temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies)) {
+			pr_err("SDMA RLC not idle in %s\n", __func__);
+			return -ETIME;
+		}
+		usleep_range(500, 1000);
+	}
+
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0);
+	WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR);
+	m->sdmax_rlcx_rb_rptr_hi =
+		RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI);
+
+	return 0;
+}
+
+bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
+{
+	uint32_t value;
+
+	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
+
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
+}
+
+int kgd_gfx_v9_wave_control_execute(struct amdgpu_device *adev,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd, uint32_t inst)
+{
+	uint32_t data = 0;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	WREG32_SOC15_RLC_SHADOW(GC, GET_INST(GC, inst), mmGRBM_GFX_INDEX, gfx_index_val);
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_CMD, sq_cmd);
+
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		INSTANCE_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SH_BROADCAST_WRITES, 1);
+	data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+		SE_BROADCAST_WRITES, 1);
+
+	WREG32_SOC15_RLC_SHADOW(GC, GET_INST(GC, inst), mmGRBM_GFX_INDEX, data);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+/*
+ * GFX9 helper for wave launch stall requirements on debug trap setting.
+ *
+ * vmid:
+ *   Target VMID to stall/unstall.
+ *
+ * stall:
+ *   0-unstall wave launch (enable), 1-stall wave launch (disable).
+ *   After wavefront launch has been stalled, allocated waves must drain from
+ *   SPI in order for debug trap settings to take effect on those waves.
+ *   This is roughly a ~96 clock cycle wait on SPI where a read on
+ *   SPI_GDBG_WAVE_CNTL translates to ~32 clock cycles.
+ *   KGD_GFX_V9_WAVE_LAUNCH_SPI_DRAIN_LATENCY indicates the number of reads required.
+ *
+ *   NOTE: We can afford to clear the entire STALL_VMID field on unstall
+ *   because GFX9.4.1 cannot support multi-process debugging due to trap
+ *   configuration and masking being limited to global scope.  Always assume
+ *   single process conditions.
+ */
+#define KGD_GFX_V9_WAVE_LAUNCH_SPI_DRAIN_LATENCY	3
+void kgd_gfx_v9_set_wave_launch_stall(struct amdgpu_device *adev,
+					uint32_t vmid,
+					bool stall)
+{
+	int i;
+	uint32_t data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1))
+		data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_VMID,
+							stall ? 1 << vmid : 0);
+	else
+		data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_RA,
+							stall ? 1 : 0);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data);
+
+	if (!stall)
+		return;
+
+	for (i = 0; i < KGD_GFX_V9_WAVE_LAUNCH_SPI_DRAIN_LATENCY; i++)
+		RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+}
+
+/*
+ * restore_dbg_registers is ignored here but is a general interface requirement
+ * for devices that support GFXOFF and where the RLC save/restore list
+ * does not support hw registers for debugging i.e. the driver has to manually
+ * initialize the debug mode registers after it has disabled GFX off during the
+ * debug session.
+ */
+uint32_t kgd_gfx_v9_enable_debug_trap(struct amdgpu_device *adev,
+				bool restore_dbg_registers,
+				uint32_t vmid)
+{
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+/*
+ * keep_trap_enabled is ignored here but is a general interface requirement
+ * for devices that support multi-process debugging where the performance
+ * overhead from trap temporary setup needs to be bypassed when the debug
+ * session has ended.
+ */
+uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev,
+					bool keep_trap_enabled,
+					uint32_t vmid)
+{
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+int kgd_gfx_v9_validate_trap_override_request(struct amdgpu_device *adev,
+					uint32_t trap_override,
+					uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH;
+
+	/* The SPI_GDBG_TRAP_MASK register is global and affects all
+	 * processes. Only allow OR-ing the address-watch bit, since
+	 * this only affects processes under the debugger. Other bits
+	 * should stay 0 to avoid the debugger interfering with other
+	 * processes.
+	 */
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR)
+		return -EINVAL;
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					     uint32_t vmid,
+					     uint32_t trap_override,
+					     uint32_t trap_mask_bits,
+					     uint32_t trap_mask_request,
+					     uint32_t *trap_mask_prev,
+					     uint32_t kfd_dbg_cntl_prev)
+{
+	uint32_t data, wave_cntl_prev;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	wave_cntl_prev = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK));
+	*trap_mask_prev = REG_GET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN);
+
+	trap_mask_bits = (trap_mask_bits & trap_mask_request) |
+		(*trap_mask_prev & ~trap_mask_request);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN, trap_mask_bits);
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, REPLACE, trap_override);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data);
+
+	/* We need to preserve wave launch mode stall settings. */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), wave_cntl_prev);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v9_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+	bool is_mode_set = !!wave_launch_mode;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+		VMID_MASK, is_mode_set ? 1 << vmid : 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+		MODE, is_mode_set ? wave_launch_mode : 0);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
+#define TCP_WATCH_STRIDE (mmTCP_WATCH1_ADDR_H - mmTCP_WATCH0_ADDR_H)
+uint32_t kgd_gfx_v9_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid,
+					uint32_t inst)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VMID,
+			debug_vmid);
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 6);
+
+	/* Turning off this watch point until we set all the registers */
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			0);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	/* Enable the watch point */
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v9_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id)
+{
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	return 0;
+}
+
+/* kgd_gfx_v9_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
+ * The values read are:
+ *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
+ *     atomic_offload_wait_time -- Wait Count for L2 and GDS Atomics Offloads.
+ *     wrm_offload_wait_time    -- Wait Count for WAIT_REG_MEM Offloads.
+ *     gws_wait_time            -- Wait Count for Global Wave Syncs.
+ *     que_sleep_wait_time      -- Wait Count for Dequeue Retry.
+ *     sch_wave_wait_time       -- Wait Count for Scheduling Wave Message.
+ *     sem_rearm_wait_time      -- Wait Count for Semaphore re-arm.
+ *     deq_retry_wait_time      -- Wait Count for Global Wave Syncs.
+ */
+void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev,
+					uint32_t *wait_times,
+					uint32_t inst)
+
+{
+	*wait_times = RREG32_SOC15_RLC(GC, GET_INST(GC, inst),
+			mmCP_IQ_WAIT_TIME2);
+}
+
+void kgd_gfx_v9_set_vm_context_page_table_base(struct amdgpu_device *adev,
+			uint32_t vmid, uint64_t page_table_base)
+{
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("trying to set page table base for wrong VMID %u\n",
+		       vmid);
+		return;
+	}
+
+	adev->mmhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base);
+
+	adev->gfxhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base);
+}
+
+static void lock_spi_csq_mutexes(struct amdgpu_device *adev)
+{
+	mutex_lock(&adev->srbm_mutex);
+	mutex_lock(&adev->grbm_idx_mutex);
+
+}
+
+static void unlock_spi_csq_mutexes(struct amdgpu_device *adev)
+{
+	mutex_unlock(&adev->grbm_idx_mutex);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
+/**
+ * get_wave_count: Read device registers to get number of waves in flight for
+ * a particular queue. The method also returns the VMID associated with the
+ * queue.
+ *
+ * @adev: Handle of device whose registers are to be read
+ * @queue_idx: Index of queue in the queue-map bit-field
+ * @queue_cnt: Stores the wave count and doorbell offset for an active queue
+ * @inst: xcc's instance number on a multi-XCC setup
+ */
+static void get_wave_count(struct amdgpu_device *adev, int queue_idx,
+		struct kfd_cu_occupancy *queue_cnt, uint32_t inst)
+{
+	int pipe_idx;
+	int queue_slot;
+	unsigned int reg_val;
+	unsigned int wave_cnt;
+	/*
+	 * Program GRBM with appropriate MEID, PIPEID, QUEUEID and VMID
+	 * parameters to read out waves in flight. Get VMID if there are
+	 * non-zero waves in flight.
+	 */
+	pipe_idx = queue_idx / adev->gfx.mec.num_queue_per_pipe;
+	queue_slot = queue_idx % adev->gfx.mec.num_queue_per_pipe;
+	soc15_grbm_select(adev, 1, pipe_idx, queue_slot, 0, GET_INST(GC, inst));
+	reg_val = RREG32_SOC15_IP(GC, SOC15_REG_OFFSET(GC, GET_INST(GC, inst),
+				  mmSPI_CSQ_WF_ACTIVE_COUNT_0) + queue_slot);
+	wave_cnt = reg_val & SPI_CSQ_WF_ACTIVE_COUNT_0__COUNT_MASK;
+	if (wave_cnt != 0) {
+		queue_cnt->wave_cnt += wave_cnt;
+		queue_cnt->doorbell_off =
+			(RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_DOORBELL_CONTROL) &
+			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK) >>
+			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT;
+	}
+}
+
+/**
+ * kgd_gfx_v9_get_cu_occupancy: Reads relevant registers associated with each
+ * shader engine and aggregates the number of waves that are in flight for the
+ * process whose pasid is provided as a parameter. The process could have ZERO
+ * or more queues running and submitting waves to compute units.
+ *
+ * @adev: Handle of device from which to get number of waves in flight
+ * @cu_occupancy: Array that gets filled with wave_cnt and doorbell offset
+ *		  for comparison later.
+ * @max_waves_per_cu: Output parameter updated with maximum number of waves
+ *                    possible per Compute Unit
+ * @inst: xcc's instance number on a multi-XCC setup
+ *
+ * Note: It's possible that the device has too many queues (oversubscription)
+ * in which case a VMID could be remapped to a different PASID. This could lead
+ * to an inaccurate wave count. Following is a high-level sequence:
+ *    Time T1: vmid = getVmid(); vmid is associated with Pasid P1
+ *    Time T2: passId = getPasId(vmid); vmid is associated with Pasid P2
+ * In the sequence above wave count obtained from time T1 will be incorrectly
+ * lost or added to total wave count.
+ *
+ * The registers that provide the waves in flight are:
+ *
+ *  SPI_CSQ_WF_ACTIVE_STATUS - bit-map of queues per pipe. The bit is ON if a
+ *  queue is slotted, OFF if there is no queue. A process could have ZERO or
+ *  more queues slotted and submitting waves to be run on compute units. Even
+ *  when there is a queue it is possible there could be zero wave fronts, this
+ *  can happen when queue is waiting on top-of-pipe events - e.g. waitRegMem
+ *  command
+ *
+ *  For each bit that is ON from above:
+ *
+ *    Read (SPI_CSQ_WF_ACTIVE_COUNT_0 + queue_idx) register. It provides the
+ *    number of waves that are in flight for the queue at specified index. The
+ *    index ranges from 0 to 7.
+ *
+ *    If non-zero waves are in flight, store the corresponding doorbell offset
+ *    of the queue, along with the wave count.
+ *
+ *    Determine if the queue belongs to the process by comparing the doorbell
+ *    offset against the process's queues. If it matches, aggregate the wave
+ *    count for the process.
+ *
+ *  Reading registers referenced above involves programming GRBM appropriately
+ */
+void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev,
+				 struct kfd_cu_occupancy *cu_occupancy,
+				 int *max_waves_per_cu, uint32_t inst)
+{
+	int qidx;
+	int se_idx;
+	int se_cnt;
+	int queue_map;
+	int max_queue_cnt;
+	DECLARE_BITMAP(cp_queue_bitmap, AMDGPU_MAX_QUEUES);
+
+	lock_spi_csq_mutexes(adev);
+	soc15_grbm_select(adev, 1, 0, 0, 0, GET_INST(GC, inst));
+
+	/*
+	 * Iterate through the shader engines and arrays of the device
+	 * to get number of waves in flight
+	 */
+	bitmap_complement(cp_queue_bitmap, adev->gfx.mec_bitmap[0].queue_bitmap,
+			  AMDGPU_MAX_QUEUES);
+	max_queue_cnt = adev->gfx.mec.num_pipe_per_mec *
+			adev->gfx.mec.num_queue_per_pipe;
+	se_cnt = adev->gfx.config.max_shader_engines;
+	for (se_idx = 0; se_idx < se_cnt; se_idx++) {
+		amdgpu_gfx_select_se_sh(adev, se_idx, 0, 0xffffffff, inst);
+		queue_map = RREG32_SOC15(GC, GET_INST(GC, inst), mmSPI_CSQ_WF_ACTIVE_STATUS);
+
+		/*
+		 * Assumption: queue map encodes following schema: four
+		 * pipes per each micro-engine, with each pipe mapping
+		 * eight queues. This schema is true for GFX9 devices
+		 * and must be verified for newer device families
+		 */
+		for (qidx = 0; qidx < max_queue_cnt; qidx++) {
+			/* Skip qeueus that are not associated with
+			 * compute functions
+			 */
+			if (!test_bit(qidx, cp_queue_bitmap))
+				continue;
+
+			if (!(queue_map & (1 << qidx)))
+				continue;
+
+			/* Get number of waves in flight and aggregate them */
+			get_wave_count(adev, qidx, &cu_occupancy[qidx],
+					inst);
+		}
+	}
+
+	amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, inst);
+	soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, inst));
+	unlock_spi_csq_mutexes(adev);
+
+	/* Update the output parameters and return */
+	*max_waves_per_cu = adev->gfx.cu_info.simd_per_cu *
+				adev->gfx.cu_info.max_waves_per_simd;
+}
+
+void kgd_gfx_v9_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
+		uint32_t wait_times,
+		uint32_t sch_wave,
+		uint32_t que_sleep,
+		uint32_t *reg_offset,
+		uint32_t *reg_data)
+{
+	*reg_data = wait_times;
+
+	if (sch_wave)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				SCH_WAVE,
+				sch_wave);
+	if (que_sleep)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				QUE_SLEEP,
+				que_sleep);
+
+	*reg_offset = SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2);
+}
+
+void kgd_gfx_v9_program_trap_handler_settings(struct amdgpu_device *adev,
+		uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr, uint32_t inst)
+{
+	kgd_gfx_v9_lock_srbm(adev, 0, 0, 0, vmid, inst);
+
+	/*
+	 * Program TBA registers
+	 */
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TBA_LO,
+			lower_32_bits(tba_addr >> 8));
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TBA_HI,
+			upper_32_bits(tba_addr >> 8));
+
+	/*
+	 * Program TMA registers
+	 */
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TMA_LO,
+			lower_32_bits(tma_addr >> 8));
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TMA_HI,
+			upper_32_bits(tma_addr >> 8));
+
+	kgd_gfx_v9_unlock_srbm(adev, inst);
+}
+
+uint64_t kgd_gfx_v9_hqd_get_pq_addr(struct amdgpu_device *adev,
+				    uint32_t pipe_id, uint32_t queue_id,
+				    uint32_t inst)
+{
+	uint32_t low, high;
+	uint64_t queue_addr = 0;
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+	amdgpu_gfx_rlc_enter_safe_mode(adev, inst);
+
+	if (!RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE))
+		goto unlock_out;
+
+	low = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE);
+	high = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE_HI);
+
+	/* only concerned with user queues. */
+	if (!high)
+		goto unlock_out;
+
+	queue_addr = (((queue_addr | high) << 32) | low) << 8;
+
+unlock_out:
+	amdgpu_gfx_rlc_exit_safe_mode(adev, inst);
+	kgd_gfx_v9_release_queue(adev, inst);
+
+	return queue_addr;
+}
+
+/* assume queue acquired  */
+static int kgd_gfx_v9_hqd_dequeue_wait(struct amdgpu_device *adev, uint32_t inst,
+				       unsigned int utimeout)
+{
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
+
+	while (true) {
+		uint32_t temp = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE);
+
+		if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
+			return 0;
+
+		if (time_after(jiffies, end_jiffies))
+			return -ETIME;
+
+		usleep_range(500, 1000);
+	}
+}
+
+uint64_t kgd_gfx_v9_hqd_reset(struct amdgpu_device *adev,
+			      uint32_t pipe_id, uint32_t queue_id,
+			      uint32_t inst, unsigned int utimeout)
+{
+	uint32_t low, high, pipe_reset_data = 0;
+	uint64_t queue_addr = 0;
+
+	kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst);
+	amdgpu_gfx_rlc_enter_safe_mode(adev, inst);
+
+	if (!RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE))
+		goto unlock_out;
+
+	low = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE);
+	high = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE_HI);
+
+	/* only concerned with user queues. */
+	if (!high)
+		goto unlock_out;
+
+	queue_addr = (((queue_addr | high) << 32) | low) << 8;
+
+	pr_debug("Attempting queue reset on XCC %i pipe id %i queue id %i\n",
+		 inst, pipe_id, queue_id);
+
+	/* assume previous dequeue request issued will take affect after reset */
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmSPI_COMPUTE_QUEUE_RESET, 0x1);
+
+	if (!kgd_gfx_v9_hqd_dequeue_wait(adev, inst, utimeout))
+		goto unlock_out;
+
+	pr_debug("Attempting pipe reset on XCC %i pipe id %i\n", inst, pipe_id);
+
+	pipe_reset_data = REG_SET_FIELD(pipe_reset_data, CP_MEC_CNTL, MEC_ME1_PIPE0_RESET, 1);
+	pipe_reset_data = pipe_reset_data << pipe_id;
+
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmCP_MEC_CNTL, pipe_reset_data);
+	WREG32_SOC15(GC, GET_INST(GC, inst), mmCP_MEC_CNTL, 0);
+
+	if (kgd_gfx_v9_hqd_dequeue_wait(adev, inst, utimeout))
+		queue_addr = 0;
+
+unlock_out:
+	pr_debug("queue reset on XCC %i pipe id %i queue id %i %s\n",
+		 inst, pipe_id, queue_id, !!queue_addr ? "succeeded!" : "failed!");
+	amdgpu_gfx_rlc_exit_safe_mode(adev, inst);
+	kgd_gfx_v9_release_queue(adev, inst);
+
+	return queue_addr;
+}
+
+uint32_t kgd_gfx_v9_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+					  int engine, int queue)
+
+{
+	return 0;
+}
+
+const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
+	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
+	.init_interrupts = kgd_gfx_v9_init_interrupts,
+	.hqd_load = kgd_gfx_v9_hqd_load,
+	.hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load,
+	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_dump = kgd_gfx_v9_hqd_dump,
+	.hqd_sdma_dump = kgd_hqd_sdma_dump,
+	.hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied,
+	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+	.hqd_destroy = kgd_gfx_v9_hqd_destroy,
+	.hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+	.wave_control_execute = kgd_gfx_v9_wave_control_execute,
+	.get_atc_vmid_pasid_mapping_info =
+			kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
+	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
+	.enable_debug_trap = kgd_gfx_v9_enable_debug_trap,
+	.disable_debug_trap = kgd_gfx_v9_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v9_set_address_watch,
+	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
+	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
+	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
+	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
+	.hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr,
+	.hqd_reset = kgd_gfx_v9_hqd_reset,
+	.hqd_sdma_get_doorbell = kgd_gfx_v9_hqd_sdma_get_doorbell
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
new file mode 100644
index 000000000000..704452ca62f8
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
@@ -0,0 +1,116 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+void kgd_gfx_v9_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
+		uint32_t sh_mem_config,
+		uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit,
+		uint32_t sh_mem_bases, uint32_t inst);
+int kgd_gfx_v9_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
+		unsigned int vmid, uint32_t inst);
+int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t inst);
+int kgd_gfx_v9_hqd_load(struct amdgpu_device *adev, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr,
+			uint32_t wptr_shift, uint32_t wptr_mask,
+			struct mm_struct *mm, uint32_t inst);
+int kgd_gfx_v9_hiq_mqd_load(struct amdgpu_device *adev, void *mqd,
+			    uint32_t pipe_id, uint32_t queue_id,
+			    uint32_t doorbell_off, uint32_t inst);
+int kgd_gfx_v9_hqd_dump(struct amdgpu_device *adev,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst);
+bool kgd_gfx_v9_hqd_is_occupied(struct amdgpu_device *adev,
+			uint64_t queue_address, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t inst);
+int kgd_gfx_v9_hqd_destroy(struct amdgpu_device *adev, void *mqd,
+				enum kfd_preempt_type reset_type,
+				unsigned int utimeout, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst);
+int kgd_gfx_v9_wave_control_execute(struct amdgpu_device *adev,
+					uint32_t gfx_index_val,
+					uint32_t sq_cmd, uint32_t inst);
+bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid);
+void kgd_gfx_v9_set_vm_context_page_table_base(struct amdgpu_device *adev,
+			uint32_t vmid, uint64_t page_table_base);
+void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev,
+				 struct kfd_cu_occupancy *cu_occupancy,
+				 int *max_waves_per_cu, uint32_t inst);
+void kgd_gfx_v9_program_trap_handler_settings(struct amdgpu_device *adev,
+		uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr,
+		uint32_t inst);
+void kgd_gfx_v9_acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
+				uint32_t queue_id, uint32_t inst);
+uint64_t kgd_gfx_v9_get_queue_mask(struct amdgpu_device *adev,
+				uint32_t pipe_id, uint32_t queue_id);
+void kgd_gfx_v9_release_queue(struct amdgpu_device *adev, uint32_t inst);
+void kgd_gfx_v9_set_wave_launch_stall(struct amdgpu_device *adev,
+					uint32_t vmid,
+					bool stall);
+uint32_t kgd_gfx_v9_enable_debug_trap(struct amdgpu_device *adev,
+				      bool restore_dbg_registers,
+				      uint32_t vmid);
+uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev,
+					bool keep_trap_enabled,
+					uint32_t vmid);
+int kgd_gfx_v9_validate_trap_override_request(struct amdgpu_device *adev,
+					     uint32_t trap_override,
+					     uint32_t *trap_mask_supported);
+uint32_t kgd_gfx_v9_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid);
+uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					     uint32_t vmid,
+					     uint32_t trap_override,
+					     uint32_t trap_mask_bits,
+					     uint32_t trap_mask_request,
+					     uint32_t *trap_mask_prev,
+					     uint32_t kfd_dbg_trap_cntl_prev);
+uint32_t kgd_gfx_v9_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid,
+					uint32_t inst);
+uint32_t kgd_gfx_v9_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id);
+void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev,
+				uint32_t *wait_times,
+				uint32_t inst);
+void kgd_gfx_v9_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
+					       uint32_t wait_times,
+					       uint32_t sch_wave,
+					       uint32_t que_sleep,
+					       uint32_t *reg_offset,
+					       uint32_t *reg_data);
+uint64_t kgd_gfx_v9_hqd_get_pq_addr(struct amdgpu_device *adev,
+				    uint32_t pipe_id,
+				    uint32_t queue_id,
+				    uint32_t inst);
+uint64_t kgd_gfx_v9_hqd_reset(struct amdgpu_device *adev,
+			      uint32_t pipe_id,
+			      uint32_t queue_id,
+			      uint32_t inst,
+			      unsigned int utimeout);
+uint32_t kgd_gfx_v9_hqd_sdma_get_doorbell(struct amdgpu_device *adev,
+					  int engine, int queue);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
new file mode 100644
index 000000000000..b1c24c8fa686
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -0,0 +1,3252 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2014-2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include <linux/dma-buf.h>
+#include <linux/list.h>
+#include <linux/pagemap.h>
+#include <linux/sched/mm.h>
+#include <linux/sched/task.h>
+#include <drm/ttm/ttm_tt.h>
+
+#include <drm/drm_exec.h>
+
+#include "amdgpu_object.h"
+#include "amdgpu_gem.h"
+#include "amdgpu_vm.h"
+#include "amdgpu_hmm.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_dma_buf.h"
+#include <uapi/linux/kfd_ioctl.h>
+#include "amdgpu_xgmi.h"
+#include "kfd_priv.h"
+#include "kfd_smi_events.h"
+
+/* Userptr restore delay, just long enough to allow consecutive VM
+ * changes to accumulate
+ */
+#define AMDGPU_USERPTR_RESTORE_DELAY_MS 1
+#define AMDGPU_RESERVE_MEM_LIMIT			(3UL << 29)
+
+/*
+ * Align VRAM availability to 2MB to avoid fragmentation caused by 4K allocations in the tail 2MB
+ * BO chunk
+ */
+#define VRAM_AVAILABLITY_ALIGN (1 << 21)
+
+/* Impose limit on how much memory KFD can use */
+static struct {
+	uint64_t max_system_mem_limit;
+	uint64_t max_ttm_mem_limit;
+	int64_t system_mem_used;
+	int64_t ttm_mem_used;
+	spinlock_t mem_limit_lock;
+} kfd_mem_limit;
+
+static const char * const domain_bit_to_string[] = {
+		"CPU",
+		"GTT",
+		"VRAM",
+		"GDS",
+		"GWS",
+		"OA"
+};
+
+#define domain_string(domain) domain_bit_to_string[ffs(domain)-1]
+
+static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work);
+
+static bool kfd_mem_is_attached(struct amdgpu_vm *avm,
+		struct kgd_mem *mem)
+{
+	struct kfd_mem_attachment *entry;
+
+	list_for_each_entry(entry, &mem->attachments, list)
+		if (entry->bo_va->base.vm == avm)
+			return true;
+
+	return false;
+}
+
+/**
+ * reuse_dmamap() - Check whether adev can share the original
+ * userptr BO
+ *
+ * If both adev and bo_adev are in direct mapping or
+ * in the same iommu group, they can share the original BO.
+ *
+ * @adev: Device to which can or cannot share the original BO
+ * @bo_adev: Device to which allocated BO belongs to
+ *
+ * Return: returns true if adev can share original userptr BO,
+ * false otherwise.
+ */
+static bool reuse_dmamap(struct amdgpu_device *adev, struct amdgpu_device *bo_adev)
+{
+	return (adev->ram_is_direct_mapped && bo_adev->ram_is_direct_mapped) ||
+			(adev->dev->iommu_group == bo_adev->dev->iommu_group);
+}
+
+/* Set memory usage limits. Current, limits are
+ *  System (TTM + userptr) memory - 15/16th System RAM
+ *  TTM memory - 3/8th System RAM
+ */
+void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
+{
+	struct sysinfo si;
+	uint64_t mem;
+
+	if (kfd_mem_limit.max_system_mem_limit)
+		return;
+
+	si_meminfo(&si);
+	mem = si.totalram - si.totalhigh;
+	mem *= si.mem_unit;
+
+	spin_lock_init(&kfd_mem_limit.mem_limit_lock);
+	kfd_mem_limit.max_system_mem_limit = mem - (mem >> 6);
+	if (kfd_mem_limit.max_system_mem_limit < 2 * AMDGPU_RESERVE_MEM_LIMIT)
+		kfd_mem_limit.max_system_mem_limit >>= 1;
+	else
+		kfd_mem_limit.max_system_mem_limit -= AMDGPU_RESERVE_MEM_LIMIT;
+
+	kfd_mem_limit.max_ttm_mem_limit = ttm_tt_pages_limit() << PAGE_SHIFT;
+	pr_debug("Kernel memory limit %lluM, TTM limit %lluM\n",
+		(kfd_mem_limit.max_system_mem_limit >> 20),
+		(kfd_mem_limit.max_ttm_mem_limit >> 20));
+}
+
+void amdgpu_amdkfd_reserve_system_mem(uint64_t size)
+{
+	kfd_mem_limit.system_mem_used += size;
+}
+
+/* Estimate page table size needed to represent a given memory size
+ *
+ * With 4KB pages, we need one 8 byte PTE for each 4KB of memory
+ * (factor 512, >> 9). With 2MB pages, we need one 8 byte PTE for 2MB
+ * of memory (factor 256K, >> 18). ROCm user mode tries to optimize
+ * for 2MB pages for TLB efficiency. However, small allocations and
+ * fragmented system memory still need some 4KB pages. We choose a
+ * compromise that should work in most cases without reserving too
+ * much memory for page tables unnecessarily (factor 16K, >> 14).
+ */
+
+#define ESTIMATE_PT_SIZE(mem_size) max(((mem_size) >> 14), AMDGPU_VM_RESERVED_VRAM)
+
+/**
+ * amdgpu_amdkfd_reserve_mem_limit() - Decrease available memory by size
+ * of buffer.
+ *
+ * @adev: Device to which allocated BO belongs to
+ * @size: Size of buffer, in bytes, encapsulated by B0. This should be
+ * equivalent to amdgpu_bo_size(BO)
+ * @alloc_flag: Flag used in allocating a BO as noted above
+ * @xcp_id: xcp_id is used to get xcp from xcp manager, one xcp is
+ * managed as one compute node in driver for app
+ *
+ * Return:
+ *	returns -ENOMEM in case of error, ZERO otherwise
+ */
+int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
+		uint64_t size, u32 alloc_flag, int8_t xcp_id)
+{
+	uint64_t reserved_for_pt =
+		ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	uint64_t reserved_for_ras = (con ? con->reserved_pages_in_bytes : 0);
+	size_t system_mem_needed, ttm_mem_needed, vram_needed;
+	int ret = 0;
+	uint64_t vram_size = 0;
+
+	system_mem_needed = 0;
+	ttm_mem_needed = 0;
+	vram_needed = 0;
+	if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
+		system_mem_needed = size;
+		ttm_mem_needed = size;
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
+		/*
+		 * Conservatively round up the allocation requirement to 2 MB
+		 * to avoid fragmentation caused by 4K allocations in the tail
+		 * 2M BO chunk.
+		 */
+		vram_needed = size;
+		/*
+		 * For GFX 9.4.3, get the VRAM size from XCP structs
+		 */
+		if (WARN_ONCE(xcp_id < 0, "invalid XCP ID %d", xcp_id))
+			return -EINVAL;
+
+		vram_size = KFD_XCP_MEMORY_SIZE(adev, xcp_id);
+		if (adev->apu_prefer_gtt) {
+			system_mem_needed = size;
+			ttm_mem_needed = size;
+		}
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
+		system_mem_needed = size;
+	} else if (!(alloc_flag &
+				(KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+				 KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
+		pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag);
+		return -ENOMEM;
+	}
+
+	spin_lock(&kfd_mem_limit.mem_limit_lock);
+
+	if (kfd_mem_limit.system_mem_used + system_mem_needed >
+	    kfd_mem_limit.max_system_mem_limit) {
+		pr_debug("Set no_system_mem_limit=1 if using shared memory\n");
+		if (!no_system_mem_limit) {
+			ret = -ENOMEM;
+			goto release;
+		}
+	}
+
+	if (kfd_mem_limit.ttm_mem_used + ttm_mem_needed >
+		kfd_mem_limit.max_ttm_mem_limit) {
+		ret = -ENOMEM;
+		goto release;
+	}
+
+	/*if is_app_apu is false and apu_prefer_gtt is true, it is an APU with
+	 * carve out < gtt. In that case, VRAM allocation will go to gtt domain, skip
+	 * VRAM check since ttm_mem_limit check already cover this allocation
+	 */
+
+	if (adev && xcp_id >= 0 && (!adev->apu_prefer_gtt || adev->gmc.is_app_apu)) {
+		uint64_t vram_available =
+			vram_size - reserved_for_pt - reserved_for_ras -
+			atomic64_read(&adev->vram_pin_size);
+		if (adev->kfd.vram_used[xcp_id] + vram_needed > vram_available) {
+			ret = -ENOMEM;
+			goto release;
+		}
+	}
+
+	/* Update memory accounting by decreasing available system
+	 * memory, TTM memory and GPU memory as computed above
+	 */
+	WARN_ONCE(vram_needed && !adev,
+		  "adev reference can't be null when vram is used");
+	if (adev && xcp_id >= 0) {
+		adev->kfd.vram_used[xcp_id] += vram_needed;
+		adev->kfd.vram_used_aligned[xcp_id] +=
+				adev->apu_prefer_gtt ?
+				vram_needed :
+				ALIGN(vram_needed, VRAM_AVAILABLITY_ALIGN);
+	}
+	kfd_mem_limit.system_mem_used += system_mem_needed;
+	kfd_mem_limit.ttm_mem_used += ttm_mem_needed;
+
+release:
+	spin_unlock(&kfd_mem_limit.mem_limit_lock);
+	return ret;
+}
+
+void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
+		uint64_t size, u32 alloc_flag, int8_t xcp_id)
+{
+	spin_lock(&kfd_mem_limit.mem_limit_lock);
+
+	if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
+		kfd_mem_limit.system_mem_used -= size;
+		kfd_mem_limit.ttm_mem_used -= size;
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
+		WARN_ONCE(!adev,
+			  "adev reference can't be null when alloc mem flags vram is set");
+		if (WARN_ONCE(xcp_id < 0, "invalid XCP ID %d", xcp_id))
+			goto release;
+
+		if (adev) {
+			adev->kfd.vram_used[xcp_id] -= size;
+			if (adev->apu_prefer_gtt) {
+				adev->kfd.vram_used_aligned[xcp_id] -= size;
+				kfd_mem_limit.system_mem_used -= size;
+				kfd_mem_limit.ttm_mem_used -= size;
+			} else {
+				adev->kfd.vram_used_aligned[xcp_id] -=
+					ALIGN(size, VRAM_AVAILABLITY_ALIGN);
+			}
+		}
+	} else if (alloc_flag & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
+		kfd_mem_limit.system_mem_used -= size;
+	} else if (!(alloc_flag &
+				(KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+				 KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
+		pr_err("%s: Invalid BO type %#x\n", __func__, alloc_flag);
+		goto release;
+	}
+	WARN_ONCE(adev && xcp_id >= 0 && adev->kfd.vram_used[xcp_id] < 0,
+		  "KFD VRAM memory accounting unbalanced for xcp: %d", xcp_id);
+	WARN_ONCE(kfd_mem_limit.ttm_mem_used < 0,
+		  "KFD TTM memory accounting unbalanced");
+	WARN_ONCE(kfd_mem_limit.system_mem_used < 0,
+		  "KFD system memory accounting unbalanced");
+
+release:
+	spin_unlock(&kfd_mem_limit.mem_limit_lock);
+}
+
+void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	u32 alloc_flags = bo->kfd_bo->alloc_flags;
+	u64 size = amdgpu_bo_size(bo);
+
+	amdgpu_amdkfd_unreserve_mem_limit(adev, size, alloc_flags,
+					  bo->xcp_id);
+
+	kfree(bo->kfd_bo);
+}
+
+/**
+ * create_dmamap_sg_bo() - Creates a amdgpu_bo object to reflect information
+ * about USERPTR or DOOREBELL or MMIO BO.
+ *
+ * @adev: Device for which dmamap BO is being created
+ * @mem: BO of peer device that is being DMA mapped. Provides parameters
+ *	 in building the dmamap BO
+ * @bo_out: Output parameter updated with handle of dmamap BO
+ */
+static int
+create_dmamap_sg_bo(struct amdgpu_device *adev,
+		 struct kgd_mem *mem, struct amdgpu_bo **bo_out)
+{
+	struct drm_gem_object *gem_obj;
+	int ret;
+	uint64_t flags = 0;
+
+	ret = amdgpu_bo_reserve(mem->bo, false);
+	if (ret)
+		return ret;
+
+	if (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR)
+		flags |= mem->bo->flags & (AMDGPU_GEM_CREATE_COHERENT |
+					AMDGPU_GEM_CREATE_UNCACHED);
+
+	ret = amdgpu_gem_object_create(adev, mem->bo->tbo.base.size, 1,
+			AMDGPU_GEM_DOMAIN_CPU, AMDGPU_GEM_CREATE_PREEMPTIBLE | flags,
+			ttm_bo_type_sg, mem->bo->tbo.base.resv, &gem_obj, 0);
+
+	amdgpu_bo_unreserve(mem->bo);
+
+	if (ret) {
+		pr_err("Error in creating DMA mappable SG BO on domain: %d\n", ret);
+		return -EINVAL;
+	}
+
+	*bo_out = gem_to_amdgpu_bo(gem_obj);
+	(*bo_out)->parent = amdgpu_bo_ref(mem->bo);
+	return ret;
+}
+
+/* amdgpu_amdkfd_remove_eviction_fence - Removes eviction fence from BO's
+ *  reservation object.
+ *
+ * @bo: [IN] Remove eviction fence(s) from this BO
+ * @ef: [IN] This eviction fence is removed if it
+ *  is present in the shared list.
+ *
+ * NOTE: Must be called with BO reserved i.e. bo->tbo.resv->lock held.
+ */
+static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo,
+					struct amdgpu_amdkfd_fence *ef)
+{
+	struct dma_fence *replacement;
+
+	if (!ef)
+		return -EINVAL;
+
+	/* TODO: Instead of block before we should use the fence of the page
+	 * table update and TLB flush here directly.
+	 */
+	replacement = dma_fence_get_stub();
+	dma_resv_replace_fences(bo->tbo.base.resv, ef->base.context,
+				replacement, DMA_RESV_USAGE_BOOKKEEP);
+	dma_fence_put(replacement);
+	return 0;
+}
+
+/**
+ * amdgpu_amdkfd_remove_all_eviction_fences - Remove all eviction fences
+ * @bo: the BO where to remove the evictions fences from.
+ *
+ * This functions should only be used on release when all references to the BO
+ * are already dropped. We remove the eviction fence from the private copy of
+ * the dma_resv object here since that is what is used during release to
+ * determine of the BO is idle or not.
+ */
+void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo)
+{
+	struct dma_resv *resv = &bo->tbo.base._resv;
+	struct dma_fence *fence, *stub;
+	struct dma_resv_iter cursor;
+
+	dma_resv_assert_held(resv);
+
+	stub = dma_fence_get_stub();
+	dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_BOOKKEEP, fence) {
+		if (!to_amdgpu_amdkfd_fence(fence))
+			continue;
+
+		dma_resv_replace_fences(resv, fence->context, stub,
+					DMA_RESV_USAGE_BOOKKEEP);
+	}
+	dma_fence_put(stub);
+}
+
+static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain,
+				     bool wait)
+{
+	struct ttm_operation_ctx ctx = { false, false };
+	int ret;
+
+	if (WARN(amdgpu_ttm_tt_get_usermm(bo->tbo.ttm),
+		 "Called with userptr BO"))
+		return -EINVAL;
+
+	/* bo has been pinned, not need validate it */
+	if (bo->tbo.pin_count)
+		return 0;
+
+	amdgpu_bo_placement_from_domain(bo, domain);
+
+	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	if (ret)
+		goto validate_fail;
+	if (wait)
+		amdgpu_bo_sync_wait(bo, AMDGPU_FENCE_OWNER_KFD, false);
+
+validate_fail:
+	return ret;
+}
+
+int amdgpu_amdkfd_bo_validate_and_fence(struct amdgpu_bo *bo,
+					uint32_t domain,
+					struct dma_fence *fence)
+{
+	int ret = amdgpu_bo_reserve(bo, false);
+
+	if (ret)
+		return ret;
+
+	ret = amdgpu_amdkfd_bo_validate(bo, domain, true);
+	if (ret)
+		goto unreserve_out;
+
+	ret = dma_resv_reserve_fences(bo->tbo.base.resv, 1);
+	if (ret)
+		goto unreserve_out;
+
+	dma_resv_add_fence(bo->tbo.base.resv, fence,
+			   DMA_RESV_USAGE_BOOKKEEP);
+
+unreserve_out:
+	amdgpu_bo_unreserve(bo);
+
+	return ret;
+}
+
+static int amdgpu_amdkfd_validate_vm_bo(void *_unused, struct amdgpu_bo *bo)
+{
+	return amdgpu_amdkfd_bo_validate(bo, bo->allowed_domains, false);
+}
+
+/* vm_validate_pt_pd_bos - Validate page table and directory BOs
+ *
+ * Page directories are not updated here because huge page handling
+ * during page table updates can invalidate page directory entries
+ * again. Page directories are only updated after updating page
+ * tables.
+ */
+static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm,
+				 struct ww_acquire_ctx *ticket)
+{
+	struct amdgpu_bo *pd = vm->root.bo;
+	struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
+	int ret;
+
+	ret = amdgpu_vm_validate(adev, vm, ticket,
+				 amdgpu_amdkfd_validate_vm_bo, NULL);
+	if (ret) {
+		pr_err("failed to validate PT BOs\n");
+		return ret;
+	}
+
+	vm->pd_phys_addr = amdgpu_gmc_pd_addr(vm->root.bo);
+
+	return 0;
+}
+
+static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync)
+{
+	struct amdgpu_bo *pd = vm->root.bo;
+	struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
+	int ret;
+
+	ret = amdgpu_vm_update_pdes(adev, vm, false);
+	if (ret)
+		return ret;
+
+	return amdgpu_sync_fence(sync, vm->last_update, GFP_KERNEL);
+}
+
+static uint64_t get_pte_flags(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			      struct kgd_mem *mem)
+{
+	uint32_t mapping_flags = AMDGPU_VM_PAGE_READABLE |
+				 AMDGPU_VM_MTYPE_DEFAULT;
+
+	if (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE)
+		mapping_flags |= AMDGPU_VM_PAGE_WRITEABLE;
+	if (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE)
+		mapping_flags |= AMDGPU_VM_PAGE_EXECUTABLE;
+
+	return mapping_flags;
+}
+
+/**
+ * create_sg_table() - Create an sg_table for a contiguous DMA addr range
+ * @addr: The starting address to point to
+ * @size: Size of memory area in bytes being pointed to
+ *
+ * Allocates an instance of sg_table and initializes it to point to memory
+ * area specified by input parameters. The address used to build is assumed
+ * to be DMA mapped, if needed.
+ *
+ * DOORBELL or MMIO BOs use only one scatterlist node in their sg_table
+ * because they are physically contiguous.
+ *
+ * Return: Initialized instance of SG Table or NULL
+ */
+static struct sg_table *create_sg_table(uint64_t addr, uint32_t size)
+{
+	struct sg_table *sg = kmalloc(sizeof(*sg), GFP_KERNEL);
+
+	if (!sg)
+		return NULL;
+	if (sg_alloc_table(sg, 1, GFP_KERNEL)) {
+		kfree(sg);
+		return NULL;
+	}
+	sg_dma_address(sg->sgl) = addr;
+	sg->sgl->length = size;
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+	sg->sgl->dma_length = size;
+#endif
+	return sg;
+}
+
+static int
+kfd_mem_dmamap_userptr(struct kgd_mem *mem,
+		       struct kfd_mem_attachment *attachment)
+{
+	enum dma_data_direction direction =
+		mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+	struct ttm_operation_ctx ctx = {.interruptible = true};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+	struct amdgpu_device *adev = attachment->adev;
+	struct ttm_tt *src_ttm = mem->bo->tbo.ttm;
+	struct ttm_tt *ttm = bo->tbo.ttm;
+	int ret;
+
+	if (WARN_ON(ttm->num_pages != src_ttm->num_pages))
+		return -EINVAL;
+
+	ttm->sg = kmalloc(sizeof(*ttm->sg), GFP_KERNEL);
+	if (unlikely(!ttm->sg))
+		return -ENOMEM;
+
+	/* Same sequence as in amdgpu_ttm_tt_pin_userptr */
+	ret = sg_alloc_table_from_pages(ttm->sg, src_ttm->pages,
+					ttm->num_pages, 0,
+					(u64)ttm->num_pages << PAGE_SHIFT,
+					GFP_KERNEL);
+	if (unlikely(ret))
+		goto free_sg;
+
+	ret = dma_map_sgtable(adev->dev, ttm->sg, direction, 0);
+	if (unlikely(ret))
+		goto release_sg;
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	if (ret)
+		goto unmap_sg;
+
+	return 0;
+
+unmap_sg:
+	dma_unmap_sgtable(adev->dev, ttm->sg, direction, 0);
+release_sg:
+	pr_err("DMA map userptr failed: %d\n", ret);
+	sg_free_table(ttm->sg);
+free_sg:
+	kfree(ttm->sg);
+	ttm->sg = NULL;
+	return ret;
+}
+
+static int
+kfd_mem_dmamap_dmabuf(struct kfd_mem_attachment *attachment)
+{
+	struct ttm_operation_ctx ctx = {.interruptible = true};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+	return ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+}
+
+/**
+ * kfd_mem_dmamap_sg_bo() - Create DMA mapped sg_table to access DOORBELL or MMIO BO
+ * @mem: SG BO of the DOORBELL or MMIO resource on the owning device
+ * @attachment: Virtual address attachment of the BO on accessing device
+ *
+ * An access request from the device that owns DOORBELL does not require DMA mapping.
+ * This is because the request doesn't go through PCIe root complex i.e. it instead
+ * loops back. The need to DMA map arises only when accessing peer device's DOORBELL
+ *
+ * In contrast, all access requests for MMIO need to be DMA mapped without regard to
+ * device ownership. This is because access requests for MMIO go through PCIe root
+ * complex.
+ *
+ * This is accomplished in two steps:
+ *   - Obtain DMA mapped address of DOORBELL or MMIO memory that could be used
+ *         in updating requesting device's page table
+ *   - Signal TTM to mark memory pointed to by requesting device's BO as GPU
+ *         accessible. This allows an update of requesting device's page table
+ *         with entries associated with DOOREBELL or MMIO memory
+ *
+ * This method is invoked in the following contexts:
+ *   - Mapping of DOORBELL or MMIO BO of same or peer device
+ *   - Validating an evicted DOOREBELL or MMIO BO on device seeking access
+ *
+ * Return: ZERO if successful, NON-ZERO otherwise
+ */
+static int
+kfd_mem_dmamap_sg_bo(struct kgd_mem *mem,
+		     struct kfd_mem_attachment *attachment)
+{
+	struct ttm_operation_ctx ctx = {.interruptible = true};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+	struct amdgpu_device *adev = attachment->adev;
+	struct ttm_tt *ttm = bo->tbo.ttm;
+	enum dma_data_direction dir;
+	dma_addr_t dma_addr;
+	bool mmio;
+	int ret;
+
+	/* Expect SG Table of dmapmap BO to be NULL */
+	mmio = (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP);
+	if (unlikely(ttm->sg)) {
+		pr_err("SG Table of %d BO for peer device is UNEXPECTEDLY NON-NULL", mmio);
+		return -EINVAL;
+	}
+
+	dir = mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+			DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+	dma_addr = mem->bo->tbo.sg->sgl->dma_address;
+	pr_debug("%d BO size: %d\n", mmio, mem->bo->tbo.sg->sgl->length);
+	pr_debug("%d BO address before DMA mapping: %llx\n", mmio, dma_addr);
+	dma_addr = dma_map_resource(adev->dev, dma_addr,
+			mem->bo->tbo.sg->sgl->length, dir, DMA_ATTR_SKIP_CPU_SYNC);
+	ret = dma_mapping_error(adev->dev, dma_addr);
+	if (unlikely(ret))
+		return ret;
+	pr_debug("%d BO address after DMA mapping: %llx\n", mmio, dma_addr);
+
+	ttm->sg = create_sg_table(dma_addr, mem->bo->tbo.sg->sgl->length);
+	if (unlikely(!ttm->sg)) {
+		ret = -ENOMEM;
+		goto unmap_sg;
+	}
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	if (unlikely(ret))
+		goto free_sg;
+
+	return ret;
+
+free_sg:
+	sg_free_table(ttm->sg);
+	kfree(ttm->sg);
+	ttm->sg = NULL;
+unmap_sg:
+	dma_unmap_resource(adev->dev, dma_addr, mem->bo->tbo.sg->sgl->length,
+			   dir, DMA_ATTR_SKIP_CPU_SYNC);
+	return ret;
+}
+
+static int
+kfd_mem_dmamap_attachment(struct kgd_mem *mem,
+			  struct kfd_mem_attachment *attachment)
+{
+	switch (attachment->type) {
+	case KFD_MEM_ATT_SHARED:
+		return 0;
+	case KFD_MEM_ATT_USERPTR:
+		return kfd_mem_dmamap_userptr(mem, attachment);
+	case KFD_MEM_ATT_DMABUF:
+		return kfd_mem_dmamap_dmabuf(attachment);
+	case KFD_MEM_ATT_SG:
+		return kfd_mem_dmamap_sg_bo(mem, attachment);
+	default:
+		WARN_ON_ONCE(1);
+	}
+	return -EINVAL;
+}
+
+static void
+kfd_mem_dmaunmap_userptr(struct kgd_mem *mem,
+			 struct kfd_mem_attachment *attachment)
+{
+	enum dma_data_direction direction =
+		mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+	struct ttm_operation_ctx ctx = {.interruptible = false};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+	struct amdgpu_device *adev = attachment->adev;
+	struct ttm_tt *ttm = bo->tbo.ttm;
+
+	if (unlikely(!ttm->sg))
+		return;
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+	(void)ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+
+	dma_unmap_sgtable(adev->dev, ttm->sg, direction, 0);
+	sg_free_table(ttm->sg);
+	kfree(ttm->sg);
+	ttm->sg = NULL;
+}
+
+static void
+kfd_mem_dmaunmap_dmabuf(struct kfd_mem_attachment *attachment)
+{
+	/* This is a no-op. We don't want to trigger eviction fences when
+	 * unmapping DMABufs. Therefore the invalidation (moving to system
+	 * domain) is done in kfd_mem_dmamap_dmabuf.
+	 */
+}
+
+/**
+ * kfd_mem_dmaunmap_sg_bo() - Free DMA mapped sg_table of DOORBELL or MMIO BO
+ * @mem: SG BO of the DOORBELL or MMIO resource on the owning device
+ * @attachment: Virtual address attachment of the BO on accessing device
+ *
+ * The method performs following steps:
+ *   - Signal TTM to mark memory pointed to by BO as GPU inaccessible
+ *   - Free SG Table that is used to encapsulate DMA mapped memory of
+ *          peer device's DOORBELL or MMIO memory
+ *
+ * This method is invoked in the following contexts:
+ *     UNMapping of DOORBELL or MMIO BO on a device having access to its memory
+ *     Eviction of DOOREBELL or MMIO BO on device having access to its memory
+ *
+ * Return: void
+ */
+static void
+kfd_mem_dmaunmap_sg_bo(struct kgd_mem *mem,
+		       struct kfd_mem_attachment *attachment)
+{
+	struct ttm_operation_ctx ctx = {.interruptible = true};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+	struct amdgpu_device *adev = attachment->adev;
+	struct ttm_tt *ttm = bo->tbo.ttm;
+	enum dma_data_direction dir;
+
+	if (unlikely(!ttm->sg)) {
+		pr_debug("SG Table of BO is NULL");
+		return;
+	}
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+	(void)ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+
+	dir = mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+				DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+	dma_unmap_resource(adev->dev, ttm->sg->sgl->dma_address,
+			ttm->sg->sgl->length, dir, DMA_ATTR_SKIP_CPU_SYNC);
+	sg_free_table(ttm->sg);
+	kfree(ttm->sg);
+	ttm->sg = NULL;
+	bo->tbo.sg = NULL;
+}
+
+static void
+kfd_mem_dmaunmap_attachment(struct kgd_mem *mem,
+			    struct kfd_mem_attachment *attachment)
+{
+	switch (attachment->type) {
+	case KFD_MEM_ATT_SHARED:
+		break;
+	case KFD_MEM_ATT_USERPTR:
+		kfd_mem_dmaunmap_userptr(mem, attachment);
+		break;
+	case KFD_MEM_ATT_DMABUF:
+		kfd_mem_dmaunmap_dmabuf(attachment);
+		break;
+	case KFD_MEM_ATT_SG:
+		kfd_mem_dmaunmap_sg_bo(mem, attachment);
+		break;
+	default:
+		WARN_ON_ONCE(1);
+	}
+}
+
+static int kfd_mem_export_dmabuf(struct kgd_mem *mem)
+{
+	if (!mem->dmabuf) {
+		struct amdgpu_device *bo_adev;
+		struct dma_buf *dmabuf;
+
+		bo_adev = amdgpu_ttm_adev(mem->bo->tbo.bdev);
+		dmabuf = drm_gem_prime_handle_to_dmabuf(&bo_adev->ddev, bo_adev->kfd.client.file,
+					       mem->gem_handle,
+			mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+					       DRM_RDWR : 0);
+		if (IS_ERR(dmabuf))
+			return PTR_ERR(dmabuf);
+		mem->dmabuf = dmabuf;
+	}
+
+	return 0;
+}
+
+static int
+kfd_mem_attach_dmabuf(struct amdgpu_device *adev, struct kgd_mem *mem,
+		      struct amdgpu_bo **bo)
+{
+	struct drm_gem_object *gobj;
+	int ret;
+
+	ret = kfd_mem_export_dmabuf(mem);
+	if (ret)
+		return ret;
+
+	gobj = amdgpu_gem_prime_import(adev_to_drm(adev), mem->dmabuf);
+	if (IS_ERR(gobj))
+		return PTR_ERR(gobj);
+
+	*bo = gem_to_amdgpu_bo(gobj);
+	(*bo)->flags |= AMDGPU_GEM_CREATE_PREEMPTIBLE;
+
+	return 0;
+}
+
+/* kfd_mem_attach - Add a BO to a VM
+ *
+ * Everything that needs to bo done only once when a BO is first added
+ * to a VM. It can later be mapped and unmapped many times without
+ * repeating these steps.
+ *
+ * 0. Create BO for DMA mapping, if needed
+ * 1. Allocate and initialize BO VA entry data structure
+ * 2. Add BO to the VM
+ * 3. Determine ASIC-specific PTE flags
+ * 4. Alloc page tables and directories if needed
+ * 4a.  Validate new page tables and directories
+ */
+static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
+		struct amdgpu_vm *vm, bool is_aql)
+{
+	struct amdgpu_device *bo_adev = amdgpu_ttm_adev(mem->bo->tbo.bdev);
+	unsigned long bo_size = mem->bo->tbo.base.size;
+	uint64_t va = mem->va;
+	struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
+	struct amdgpu_bo *bo[2] = {NULL, NULL};
+	struct amdgpu_bo_va *bo_va;
+	bool same_hive = false;
+	int i, ret;
+
+	if (!va) {
+		pr_err("Invalid VA when adding BO to VM\n");
+		return -EINVAL;
+	}
+
+	/* Determine access to VRAM, MMIO and DOORBELL BOs of peer devices
+	 *
+	 * The access path of MMIO and DOORBELL BOs of is always over PCIe.
+	 * In contrast the access path of VRAM BOs depens upon the type of
+	 * link that connects the peer device. Access over PCIe is allowed
+	 * if peer device has large BAR. In contrast, access over xGMI is
+	 * allowed for both small and large BAR configurations of peer device
+	 */
+	if ((adev != bo_adev && !adev->apu_prefer_gtt) &&
+	    ((mem->domain == AMDGPU_GEM_DOMAIN_VRAM) ||
+	     (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL) ||
+	     (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
+		if (mem->domain == AMDGPU_GEM_DOMAIN_VRAM)
+			same_hive = amdgpu_xgmi_same_hive(adev, bo_adev);
+		if (!same_hive && !amdgpu_device_is_peer_accessible(bo_adev, adev))
+			return -EINVAL;
+	}
+
+	for (i = 0; i <= is_aql; i++) {
+		attachment[i] = kzalloc(sizeof(*attachment[i]), GFP_KERNEL);
+		if (unlikely(!attachment[i])) {
+			ret = -ENOMEM;
+			goto unwind;
+		}
+
+		pr_debug("\t add VA 0x%llx - 0x%llx to vm %p\n", va,
+			 va + bo_size, vm);
+
+		if ((adev == bo_adev && !(mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) ||
+		    (amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm) && reuse_dmamap(adev, bo_adev)) ||
+		    (mem->domain == AMDGPU_GEM_DOMAIN_GTT && reuse_dmamap(adev, bo_adev)) ||
+		    same_hive) {
+			/* Mappings on the local GPU, or VRAM mappings in the
+			 * local hive, or userptr, or GTT mapping can reuse dma map
+			 * address space share the original BO
+			 */
+			attachment[i]->type = KFD_MEM_ATT_SHARED;
+			bo[i] = mem->bo;
+			drm_gem_object_get(&bo[i]->tbo.base);
+		} else if (i > 0) {
+			/* Multiple mappings on the same GPU share the BO */
+			attachment[i]->type = KFD_MEM_ATT_SHARED;
+			bo[i] = bo[0];
+			drm_gem_object_get(&bo[i]->tbo.base);
+		} else if (amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm)) {
+			/* Create an SG BO to DMA-map userptrs on other GPUs */
+			attachment[i]->type = KFD_MEM_ATT_USERPTR;
+			ret = create_dmamap_sg_bo(adev, mem, &bo[i]);
+			if (ret)
+				goto unwind;
+		/* Handle DOORBELL BOs of peer devices and MMIO BOs of local and peer devices */
+		} else if (mem->bo->tbo.type == ttm_bo_type_sg) {
+			WARN_ONCE(!(mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL ||
+				    mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP),
+				  "Handing invalid SG BO in ATTACH request");
+			attachment[i]->type = KFD_MEM_ATT_SG;
+			ret = create_dmamap_sg_bo(adev, mem, &bo[i]);
+			if (ret)
+				goto unwind;
+		/* Enable acces to GTT and VRAM BOs of peer devices */
+		} else if (mem->domain == AMDGPU_GEM_DOMAIN_GTT ||
+			   mem->domain == AMDGPU_GEM_DOMAIN_VRAM) {
+			attachment[i]->type = KFD_MEM_ATT_DMABUF;
+			ret = kfd_mem_attach_dmabuf(adev, mem, &bo[i]);
+			if (ret)
+				goto unwind;
+			pr_debug("Employ DMABUF mechanism to enable peer GPU access\n");
+		} else {
+			WARN_ONCE(true, "Handling invalid ATTACH request");
+			ret = -EINVAL;
+			goto unwind;
+		}
+
+		/* Add BO to VM internal data structures */
+		ret = amdgpu_bo_reserve(bo[i], false);
+		if (ret) {
+			pr_debug("Unable to reserve BO during memory attach");
+			goto unwind;
+		}
+		bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+		if (!bo_va)
+			bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+		else
+			++bo_va->ref_count;
+		attachment[i]->bo_va = bo_va;
+		amdgpu_bo_unreserve(bo[i]);
+		if (unlikely(!attachment[i]->bo_va)) {
+			ret = -ENOMEM;
+			pr_err("Failed to add BO object to VM. ret == %d\n",
+			       ret);
+			goto unwind;
+		}
+		attachment[i]->va = va;
+		attachment[i]->pte_flags = get_pte_flags(adev, vm, mem);
+		attachment[i]->adev = adev;
+		list_add(&attachment[i]->list, &mem->attachments);
+
+		va += bo_size;
+	}
+
+	return 0;
+
+unwind:
+	for (; i >= 0; i--) {
+		if (!attachment[i])
+			continue;
+		if (attachment[i]->bo_va) {
+			(void)amdgpu_bo_reserve(bo[i], true);
+			if (--attachment[i]->bo_va->ref_count == 0)
+				amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+			amdgpu_bo_unreserve(bo[i]);
+			list_del(&attachment[i]->list);
+		}
+		if (bo[i])
+			drm_gem_object_put(&bo[i]->tbo.base);
+		kfree(attachment[i]);
+	}
+	return ret;
+}
+
+static void kfd_mem_detach(struct kfd_mem_attachment *attachment)
+{
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+
+	pr_debug("\t remove VA 0x%llx in entry %p\n",
+			attachment->va, attachment);
+	if (--attachment->bo_va->ref_count == 0)
+		amdgpu_vm_bo_del(attachment->adev, attachment->bo_va);
+	drm_gem_object_put(&bo->tbo.base);
+	list_del(&attachment->list);
+	kfree(attachment);
+}
+
+static void add_kgd_mem_to_kfd_bo_list(struct kgd_mem *mem,
+				struct amdkfd_process_info *process_info,
+				bool userptr)
+{
+	mutex_lock(&process_info->lock);
+	if (userptr)
+		list_add_tail(&mem->validate_list,
+			      &process_info->userptr_valid_list);
+	else
+		list_add_tail(&mem->validate_list, &process_info->kfd_bo_list);
+	mutex_unlock(&process_info->lock);
+}
+
+static void remove_kgd_mem_from_kfd_bo_list(struct kgd_mem *mem,
+		struct amdkfd_process_info *process_info)
+{
+	mutex_lock(&process_info->lock);
+	list_del(&mem->validate_list);
+	mutex_unlock(&process_info->lock);
+}
+
+/* Initializes user pages. It registers the MMU notifier and validates
+ * the userptr BO in the GTT domain.
+ *
+ * The BO must already be on the userptr_valid_list. Otherwise an
+ * eviction and restore may happen that leaves the new BO unmapped
+ * with the user mode queues running.
+ *
+ * Takes the process_info->lock to protect against concurrent restore
+ * workers.
+ *
+ * Returns 0 for success, negative errno for errors.
+ */
+static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr,
+			   bool criu_resume)
+{
+	struct amdkfd_process_info *process_info = mem->process_info;
+	struct amdgpu_bo *bo = mem->bo;
+	struct ttm_operation_ctx ctx = { true, false };
+	struct amdgpu_hmm_range *range;
+	int ret = 0;
+
+	mutex_lock(&process_info->lock);
+
+	ret = amdgpu_ttm_tt_set_userptr(&bo->tbo, user_addr, 0);
+	if (ret) {
+		pr_err("%s: Failed to set userptr: %d\n", __func__, ret);
+		goto out;
+	}
+
+	ret = amdgpu_hmm_register(bo, user_addr);
+	if (ret) {
+		pr_err("%s: Failed to register MMU notifier: %d\n",
+		       __func__, ret);
+		goto out;
+	}
+
+	if (criu_resume) {
+		/*
+		 * During a CRIU restore operation, the userptr buffer objects
+		 * will be validated in the restore_userptr_work worker at a
+		 * later stage when it is scheduled by another ioctl called by
+		 * CRIU master process for the target pid for restore.
+		 */
+		mutex_lock(&process_info->notifier_lock);
+		mem->invalid++;
+		mutex_unlock(&process_info->notifier_lock);
+		mutex_unlock(&process_info->lock);
+		return 0;
+	}
+
+	range = amdgpu_hmm_range_alloc(NULL);
+	if (unlikely(!range)) {
+		ret = -ENOMEM;
+		goto unregister_out;
+	}
+
+	ret = amdgpu_ttm_tt_get_user_pages(bo, range);
+	if (ret) {
+		amdgpu_hmm_range_free(range);
+		if (ret == -EAGAIN)
+			pr_debug("Failed to get user pages, try again\n");
+		else
+			pr_err("%s: Failed to get user pages: %d\n", __func__, ret);
+		goto unregister_out;
+	}
+
+	ret = amdgpu_bo_reserve(bo, true);
+	if (ret) {
+		pr_err("%s: Failed to reserve BO\n", __func__);
+		goto release_out;
+	}
+
+	amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, range);
+
+	amdgpu_bo_placement_from_domain(bo, mem->domain);
+	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	if (ret)
+		pr_err("%s: failed to validate BO\n", __func__);
+	amdgpu_bo_unreserve(bo);
+
+release_out:
+	amdgpu_hmm_range_free(range);
+unregister_out:
+	if (ret)
+		amdgpu_hmm_unregister(bo);
+out:
+	mutex_unlock(&process_info->lock);
+	return ret;
+}
+
+/* Reserving a BO and its page table BOs must happen atomically to
+ * avoid deadlocks. Some operations update multiple VMs at once. Track
+ * all the reservation info in a context structure. Optionally a sync
+ * object can track VM updates.
+ */
+struct bo_vm_reservation_context {
+	/* DRM execution context for the reservation */
+	struct drm_exec exec;
+	/* Number of VMs reserved */
+	unsigned int n_vms;
+	/* Pointer to sync object */
+	struct amdgpu_sync *sync;
+};
+
+enum bo_vm_match {
+	BO_VM_NOT_MAPPED = 0,	/* Match VMs where a BO is not mapped */
+	BO_VM_MAPPED,		/* Match VMs where a BO is mapped     */
+	BO_VM_ALL,		/* Match all VMs a BO was added to    */
+};
+
+/**
+ * reserve_bo_and_vm - reserve a BO and a VM unconditionally.
+ * @mem: KFD BO structure.
+ * @vm: the VM to reserve.
+ * @ctx: the struct that will be used in unreserve_bo_and_vms().
+ */
+static int reserve_bo_and_vm(struct kgd_mem *mem,
+			      struct amdgpu_vm *vm,
+			      struct bo_vm_reservation_context *ctx)
+{
+	struct amdgpu_bo *bo = mem->bo;
+	int ret;
+
+	WARN_ON(!vm);
+
+	ctx->n_vms = 1;
+	ctx->sync = &mem->sync;
+	drm_exec_init(&ctx->exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0);
+	drm_exec_until_all_locked(&ctx->exec) {
+		ret = amdgpu_vm_lock_pd(vm, &ctx->exec, 2);
+		drm_exec_retry_on_contention(&ctx->exec);
+		if (unlikely(ret))
+			goto error;
+
+		ret = drm_exec_prepare_obj(&ctx->exec, &bo->tbo.base, 1);
+		drm_exec_retry_on_contention(&ctx->exec);
+		if (unlikely(ret))
+			goto error;
+	}
+	return 0;
+
+error:
+	pr_err("Failed to reserve buffers in ttm.\n");
+	drm_exec_fini(&ctx->exec);
+	return ret;
+}
+
+/**
+ * reserve_bo_and_cond_vms - reserve a BO and some VMs conditionally
+ * @mem: KFD BO structure.
+ * @vm: the VM to reserve. If NULL, then all VMs associated with the BO
+ * is used. Otherwise, a single VM associated with the BO.
+ * @map_type: the mapping status that will be used to filter the VMs.
+ * @ctx: the struct that will be used in unreserve_bo_and_vms().
+ *
+ * Returns 0 for success, negative for failure.
+ */
+static int reserve_bo_and_cond_vms(struct kgd_mem *mem,
+				struct amdgpu_vm *vm, enum bo_vm_match map_type,
+				struct bo_vm_reservation_context *ctx)
+{
+	struct kfd_mem_attachment *entry;
+	struct amdgpu_bo *bo = mem->bo;
+	int ret;
+
+	ctx->sync = &mem->sync;
+	drm_exec_init(&ctx->exec, DRM_EXEC_INTERRUPTIBLE_WAIT |
+		      DRM_EXEC_IGNORE_DUPLICATES, 0);
+	drm_exec_until_all_locked(&ctx->exec) {
+		ctx->n_vms = 0;
+		list_for_each_entry(entry, &mem->attachments, list) {
+			if ((vm && vm != entry->bo_va->base.vm) ||
+				(entry->is_mapped != map_type
+				&& map_type != BO_VM_ALL))
+				continue;
+
+			ret = amdgpu_vm_lock_pd(entry->bo_va->base.vm,
+						&ctx->exec, 2);
+			drm_exec_retry_on_contention(&ctx->exec);
+			if (unlikely(ret))
+				goto error;
+			++ctx->n_vms;
+		}
+
+		ret = drm_exec_prepare_obj(&ctx->exec, &bo->tbo.base, 1);
+		drm_exec_retry_on_contention(&ctx->exec);
+		if (unlikely(ret))
+			goto error;
+	}
+	return 0;
+
+error:
+	pr_err("Failed to reserve buffers in ttm.\n");
+	drm_exec_fini(&ctx->exec);
+	return ret;
+}
+
+/**
+ * unreserve_bo_and_vms - Unreserve BO and VMs from a reservation context
+ * @ctx: Reservation context to unreserve
+ * @wait: Optionally wait for a sync object representing pending VM updates
+ * @intr: Whether the wait is interruptible
+ *
+ * Also frees any resources allocated in
+ * reserve_bo_and_(cond_)vm(s). Returns the status from
+ * amdgpu_sync_wait.
+ */
+static int unreserve_bo_and_vms(struct bo_vm_reservation_context *ctx,
+				 bool wait, bool intr)
+{
+	int ret = 0;
+
+	if (wait)
+		ret = amdgpu_sync_wait(ctx->sync, intr);
+
+	drm_exec_fini(&ctx->exec);
+	ctx->sync = NULL;
+	return ret;
+}
+
+static int unmap_bo_from_gpuvm(struct kgd_mem *mem,
+				struct kfd_mem_attachment *entry,
+				struct amdgpu_sync *sync)
+{
+	struct amdgpu_bo_va *bo_va = entry->bo_va;
+	struct amdgpu_device *adev = entry->adev;
+	struct amdgpu_vm *vm = bo_va->base.vm;
+
+	if (bo_va->queue_refcount) {
+		pr_debug("bo_va->queue_refcount %d\n", bo_va->queue_refcount);
+		return -EBUSY;
+	}
+
+	(void)amdgpu_vm_bo_unmap(adev, bo_va, entry->va);
+
+	/* VM entity stopped if process killed, don't clear freed pt bo */
+	if (!amdgpu_vm_ready(vm))
+		return 0;
+
+	(void)amdgpu_vm_clear_freed(adev, vm, &bo_va->last_pt_update);
+
+	(void)amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL);
+
+	return 0;
+}
+
+static int update_gpuvm_pte(struct kgd_mem *mem,
+			    struct kfd_mem_attachment *entry,
+			    struct amdgpu_sync *sync)
+{
+	struct amdgpu_bo_va *bo_va = entry->bo_va;
+	struct amdgpu_device *adev = entry->adev;
+	int ret;
+
+	ret = kfd_mem_dmamap_attachment(mem, entry);
+	if (ret)
+		return ret;
+
+	/* Update the page tables  */
+	ret = amdgpu_vm_bo_update(adev, bo_va, false);
+	if (ret) {
+		pr_err("amdgpu_vm_bo_update failed\n");
+		return ret;
+	}
+
+	return amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL);
+}
+
+static int map_bo_to_gpuvm(struct kgd_mem *mem,
+			   struct kfd_mem_attachment *entry,
+			   struct amdgpu_sync *sync,
+			   bool no_update_pte)
+{
+	int ret;
+
+	/* Set virtual address for the allocation */
+	ret = amdgpu_vm_bo_map(entry->adev, entry->bo_va, entry->va, 0,
+			       amdgpu_bo_size(entry->bo_va->base.bo),
+			       entry->pte_flags);
+	if (ret) {
+		pr_err("Failed to map VA 0x%llx in vm. ret %d\n",
+				entry->va, ret);
+		return ret;
+	}
+
+	if (no_update_pte)
+		return 0;
+
+	ret = update_gpuvm_pte(mem, entry, sync);
+	if (ret) {
+		pr_err("update_gpuvm_pte() failed\n");
+		goto update_gpuvm_pte_failed;
+	}
+
+	return 0;
+
+update_gpuvm_pte_failed:
+	unmap_bo_from_gpuvm(mem, entry, sync);
+	kfd_mem_dmaunmap_attachment(mem, entry);
+	return ret;
+}
+
+static int process_validate_vms(struct amdkfd_process_info *process_info,
+				struct ww_acquire_ctx *ticket)
+{
+	struct amdgpu_vm *peer_vm;
+	int ret;
+
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			    vm_list_node) {
+		ret = vm_validate_pt_pd_bos(peer_vm, ticket);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static int process_sync_pds_resv(struct amdkfd_process_info *process_info,
+				 struct amdgpu_sync *sync)
+{
+	struct amdgpu_vm *peer_vm;
+	int ret;
+
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			    vm_list_node) {
+		struct amdgpu_bo *pd = peer_vm->root.bo;
+
+		ret = amdgpu_sync_resv(NULL, sync, pd->tbo.base.resv,
+				       AMDGPU_SYNC_NE_OWNER,
+				       AMDGPU_FENCE_OWNER_KFD);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static int process_update_pds(struct amdkfd_process_info *process_info,
+			      struct amdgpu_sync *sync)
+{
+	struct amdgpu_vm *peer_vm;
+	int ret;
+
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			    vm_list_node) {
+		ret = vm_update_pds(peer_vm, sync);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
+		       struct dma_fence **ef)
+{
+	struct amdkfd_process_info *info = NULL;
+	int ret;
+
+	if (!*process_info) {
+		info = kzalloc(sizeof(*info), GFP_KERNEL);
+		if (!info)
+			return -ENOMEM;
+
+		mutex_init(&info->lock);
+		mutex_init(&info->notifier_lock);
+		INIT_LIST_HEAD(&info->vm_list_head);
+		INIT_LIST_HEAD(&info->kfd_bo_list);
+		INIT_LIST_HEAD(&info->userptr_valid_list);
+		INIT_LIST_HEAD(&info->userptr_inval_list);
+
+		info->eviction_fence =
+			amdgpu_amdkfd_fence_create(dma_fence_context_alloc(1),
+						   current->mm,
+						   NULL);
+		if (!info->eviction_fence) {
+			pr_err("Failed to create eviction fence\n");
+			ret = -ENOMEM;
+			goto create_evict_fence_fail;
+		}
+
+		info->pid = get_task_pid(current->group_leader, PIDTYPE_PID);
+		INIT_DELAYED_WORK(&info->restore_userptr_work,
+				  amdgpu_amdkfd_restore_userptr_worker);
+
+		*process_info = info;
+	}
+
+	vm->process_info = *process_info;
+
+	/* Validate page directory and attach eviction fence */
+	ret = amdgpu_bo_reserve(vm->root.bo, true);
+	if (ret)
+		goto reserve_pd_fail;
+	ret = vm_validate_pt_pd_bos(vm, NULL);
+	if (ret) {
+		pr_err("validate_pt_pd_bos() failed\n");
+		goto validate_pd_fail;
+	}
+	ret = amdgpu_bo_sync_wait(vm->root.bo,
+				  AMDGPU_FENCE_OWNER_KFD, false);
+	if (ret)
+		goto wait_pd_fail;
+	ret = dma_resv_reserve_fences(vm->root.bo->tbo.base.resv, 1);
+	if (ret)
+		goto reserve_shared_fail;
+	dma_resv_add_fence(vm->root.bo->tbo.base.resv,
+			   &vm->process_info->eviction_fence->base,
+			   DMA_RESV_USAGE_BOOKKEEP);
+	amdgpu_bo_unreserve(vm->root.bo);
+
+	/* Update process info */
+	mutex_lock(&vm->process_info->lock);
+	list_add_tail(&vm->vm_list_node,
+			&(vm->process_info->vm_list_head));
+	vm->process_info->n_vms++;
+	if (ef)
+		*ef = dma_fence_get(&vm->process_info->eviction_fence->base);
+	mutex_unlock(&vm->process_info->lock);
+
+	return 0;
+
+reserve_shared_fail:
+wait_pd_fail:
+validate_pd_fail:
+	amdgpu_bo_unreserve(vm->root.bo);
+reserve_pd_fail:
+	vm->process_info = NULL;
+	if (info) {
+		dma_fence_put(&info->eviction_fence->base);
+		*process_info = NULL;
+		put_pid(info->pid);
+create_evict_fence_fail:
+		mutex_destroy(&info->lock);
+		mutex_destroy(&info->notifier_lock);
+		kfree(info);
+	}
+	return ret;
+}
+
+/**
+ * amdgpu_amdkfd_gpuvm_pin_bo() - Pins a BO using following criteria
+ * @bo: Handle of buffer object being pinned
+ * @domain: Domain into which BO should be pinned
+ *
+ *   - USERPTR BOs are UNPINNABLE and will return error
+ *   - All other BO types (GTT, VRAM, MMIO and DOORBELL) will have their
+ *     PIN count incremented. It is valid to PIN a BO multiple times
+ *
+ * Return: ZERO if successful in pinning, Non-Zero in case of error.
+ */
+static int amdgpu_amdkfd_gpuvm_pin_bo(struct amdgpu_bo *bo, u32 domain)
+{
+	int ret = 0;
+
+	ret = amdgpu_bo_reserve(bo, false);
+	if (unlikely(ret))
+		return ret;
+
+	if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) {
+		/*
+		 * If bo is not contiguous on VRAM, move to system memory first to ensure
+		 * we can get contiguous VRAM space after evicting other BOs.
+		 */
+		if (!(bo->tbo.resource->placement & TTM_PL_FLAG_CONTIGUOUS)) {
+			struct ttm_operation_ctx ctx = { true, false };
+
+			amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+			ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+			if (unlikely(ret)) {
+				pr_debug("validate bo 0x%p to GTT failed %d\n", &bo->tbo, ret);
+				goto out;
+			}
+		}
+	}
+
+	ret = amdgpu_bo_pin(bo, domain);
+	if (ret)
+		pr_err("Error in Pinning BO to domain: %d\n", domain);
+
+	amdgpu_bo_sync_wait(bo, AMDGPU_FENCE_OWNER_KFD, false);
+out:
+	amdgpu_bo_unreserve(bo);
+	return ret;
+}
+
+/**
+ * amdgpu_amdkfd_gpuvm_unpin_bo() - Unpins BO using following criteria
+ * @bo: Handle of buffer object being unpinned
+ *
+ *   - Is a illegal request for USERPTR BOs and is ignored
+ *   - All other BO types (GTT, VRAM, MMIO and DOORBELL) will have their
+ *     PIN count decremented. Calls to UNPIN must balance calls to PIN
+ */
+static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo)
+{
+	int ret = 0;
+
+	ret = amdgpu_bo_reserve(bo, false);
+	if (unlikely(ret))
+		return;
+
+	amdgpu_bo_unpin(bo);
+	amdgpu_bo_unreserve(bo);
+}
+
+int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,
+					   struct amdgpu_vm *avm,
+					   void **process_info,
+					   struct dma_fence **ef)
+{
+	int ret;
+
+	/* Already a compute VM? */
+	if (avm->process_info)
+		return -EINVAL;
+
+	/* Convert VM into a compute VM */
+	ret = amdgpu_vm_make_compute(adev, avm);
+	if (ret)
+		return ret;
+
+	/* Initialize KFD part of the VM and process info */
+	ret = init_kfd_vm(avm, process_info, ef);
+	if (ret)
+		return ret;
+
+	amdgpu_vm_set_task_info(avm);
+
+	return 0;
+}
+
+void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
+				    struct amdgpu_vm *vm)
+{
+	struct amdkfd_process_info *process_info = vm->process_info;
+
+	if (!process_info)
+		return;
+
+	/* Update process info */
+	mutex_lock(&process_info->lock);
+	process_info->n_vms--;
+	list_del(&vm->vm_list_node);
+	mutex_unlock(&process_info->lock);
+
+	vm->process_info = NULL;
+
+	/* Release per-process resources when last compute VM is destroyed */
+	if (!process_info->n_vms) {
+		WARN_ON(!list_empty(&process_info->kfd_bo_list));
+		WARN_ON(!list_empty(&process_info->userptr_valid_list));
+		WARN_ON(!list_empty(&process_info->userptr_inval_list));
+
+		dma_fence_put(&process_info->eviction_fence->base);
+		cancel_delayed_work_sync(&process_info->restore_userptr_work);
+		put_pid(process_info->pid);
+		mutex_destroy(&process_info->lock);
+		mutex_destroy(&process_info->notifier_lock);
+		kfree(process_info);
+	}
+}
+
+uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv)
+{
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
+	struct amdgpu_bo *pd = avm->root.bo;
+	struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
+
+	if (adev->asic_type < CHIP_VEGA10)
+		return avm->pd_phys_addr >> AMDGPU_GPU_PAGE_SHIFT;
+	return avm->pd_phys_addr;
+}
+
+void amdgpu_amdkfd_block_mmu_notifications(void *p)
+{
+	struct amdkfd_process_info *pinfo = (struct amdkfd_process_info *)p;
+
+	mutex_lock(&pinfo->lock);
+	WRITE_ONCE(pinfo->block_mmu_notifications, true);
+	mutex_unlock(&pinfo->lock);
+}
+
+int amdgpu_amdkfd_criu_resume(void *p)
+{
+	int ret = 0;
+	struct amdkfd_process_info *pinfo = (struct amdkfd_process_info *)p;
+
+	mutex_lock(&pinfo->lock);
+	pr_debug("scheduling work\n");
+	mutex_lock(&pinfo->notifier_lock);
+	pinfo->evicted_bos++;
+	mutex_unlock(&pinfo->notifier_lock);
+	if (!READ_ONCE(pinfo->block_mmu_notifications)) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+	WRITE_ONCE(pinfo->block_mmu_notifications, false);
+	queue_delayed_work(system_freezable_wq,
+			   &pinfo->restore_userptr_work, 0);
+
+out_unlock:
+	mutex_unlock(&pinfo->lock);
+	return ret;
+}
+
+size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev,
+					  uint8_t xcp_id)
+{
+	uint64_t reserved_for_pt =
+		ESTIMATE_PT_SIZE(amdgpu_amdkfd_total_mem_size);
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	uint64_t reserved_for_ras = (con ? con->reserved_pages_in_bytes : 0);
+	ssize_t available;
+	uint64_t vram_available, system_mem_available, ttm_mem_available;
+
+	spin_lock(&kfd_mem_limit.mem_limit_lock);
+	if (adev->apu_prefer_gtt && !adev->gmc.is_app_apu)
+		vram_available = KFD_XCP_MEMORY_SIZE(adev, xcp_id)
+			- adev->kfd.vram_used_aligned[xcp_id];
+	else
+		vram_available = KFD_XCP_MEMORY_SIZE(adev, xcp_id)
+			- adev->kfd.vram_used_aligned[xcp_id]
+			- atomic64_read(&adev->vram_pin_size)
+			- reserved_for_pt
+			- reserved_for_ras;
+
+	if (adev->apu_prefer_gtt) {
+		system_mem_available = no_system_mem_limit ?
+					kfd_mem_limit.max_system_mem_limit :
+					kfd_mem_limit.max_system_mem_limit -
+					kfd_mem_limit.system_mem_used;
+
+		ttm_mem_available = kfd_mem_limit.max_ttm_mem_limit -
+				kfd_mem_limit.ttm_mem_used;
+
+		available = min3(system_mem_available, ttm_mem_available,
+				 vram_available);
+		available = ALIGN_DOWN(available, PAGE_SIZE);
+	} else {
+		available = ALIGN_DOWN(vram_available, VRAM_AVAILABLITY_ALIGN);
+	}
+
+	spin_unlock(&kfd_mem_limit.mem_limit_lock);
+
+	if (available < 0)
+		available = 0;
+
+	return available;
+}
+
+int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
+		struct amdgpu_device *adev, uint64_t va, uint64_t size,
+		void *drm_priv, struct kgd_mem **mem,
+		uint64_t *offset, uint32_t flags, bool criu_resume)
+{
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
+	struct amdgpu_fpriv *fpriv = container_of(avm, struct amdgpu_fpriv, vm);
+	enum ttm_bo_type bo_type = ttm_bo_type_device;
+	struct sg_table *sg = NULL;
+	uint64_t user_addr = 0;
+	struct amdgpu_bo *bo;
+	struct drm_gem_object *gobj = NULL;
+	u32 domain, alloc_domain;
+	uint64_t aligned_size;
+	int8_t xcp_id = -1;
+	u64 alloc_flags;
+	int ret;
+
+	/*
+	 * Check on which domain to allocate BO
+	 */
+	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
+		domain = alloc_domain = AMDGPU_GEM_DOMAIN_VRAM;
+
+		if (adev->apu_prefer_gtt) {
+			domain = AMDGPU_GEM_DOMAIN_GTT;
+			alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
+			alloc_flags = 0;
+		} else {
+			alloc_flags = AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE;
+			alloc_flags |= (flags & KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC) ?
+			AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED : 0;
+
+			/* For contiguous VRAM allocation */
+			if (flags & KFD_IOC_ALLOC_MEM_FLAGS_CONTIGUOUS)
+				alloc_flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+		}
+		xcp_id = fpriv->xcp_id == AMDGPU_XCP_NO_PARTITION ?
+					0 : fpriv->xcp_id;
+	} else if (flags & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
+		domain = alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
+		alloc_flags = 0;
+	} else {
+		domain = AMDGPU_GEM_DOMAIN_GTT;
+		alloc_domain = AMDGPU_GEM_DOMAIN_CPU;
+		alloc_flags = AMDGPU_GEM_CREATE_PREEMPTIBLE;
+
+		if (flags & KFD_IOC_ALLOC_MEM_FLAGS_USERPTR) {
+			if (!offset || !*offset)
+				return -EINVAL;
+			user_addr = untagged_addr(*offset);
+		} else if (flags & (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+				    KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
+			bo_type = ttm_bo_type_sg;
+			if (size > UINT_MAX)
+				return -EINVAL;
+			sg = create_sg_table(*offset, size);
+			if (!sg)
+				return -ENOMEM;
+		} else {
+			return -EINVAL;
+		}
+	}
+
+	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_COHERENT)
+		alloc_flags |= AMDGPU_GEM_CREATE_COHERENT;
+	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT)
+		alloc_flags |= AMDGPU_GEM_CREATE_EXT_COHERENT;
+	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED)
+		alloc_flags |= AMDGPU_GEM_CREATE_UNCACHED;
+
+	*mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	if (!*mem) {
+		ret = -ENOMEM;
+		goto err;
+	}
+	INIT_LIST_HEAD(&(*mem)->attachments);
+	mutex_init(&(*mem)->lock);
+	(*mem)->aql_queue = !!(flags & KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM);
+
+	/* Workaround for AQL queue wraparound bug. Map the same
+	 * memory twice. That means we only actually allocate half
+	 * the memory.
+	 */
+	if ((*mem)->aql_queue)
+		size >>= 1;
+	aligned_size = PAGE_ALIGN(size);
+
+	(*mem)->alloc_flags = flags;
+
+	amdgpu_sync_create(&(*mem)->sync);
+
+	ret = amdgpu_amdkfd_reserve_mem_limit(adev, aligned_size, flags,
+					      xcp_id);
+	if (ret) {
+		pr_debug("Insufficient memory\n");
+		goto err_reserve_limit;
+	}
+
+	pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s xcp_id %d\n",
+		 va, (*mem)->aql_queue ? size << 1 : size,
+		 domain_string(alloc_domain), xcp_id);
+
+	ret = amdgpu_gem_object_create(adev, aligned_size, 1, alloc_domain, alloc_flags,
+				       bo_type, NULL, &gobj, xcp_id + 1);
+	if (ret) {
+		pr_debug("Failed to create BO on domain %s. ret %d\n",
+			 domain_string(alloc_domain), ret);
+		goto err_bo_create;
+	}
+	ret = drm_vma_node_allow(&gobj->vma_node, drm_priv);
+	if (ret) {
+		pr_debug("Failed to allow vma node access. ret %d\n", ret);
+		goto err_node_allow;
+	}
+	ret = drm_gem_handle_create(adev->kfd.client.file, gobj, &(*mem)->gem_handle);
+	if (ret)
+		goto err_gem_handle_create;
+	bo = gem_to_amdgpu_bo(gobj);
+	if (bo_type == ttm_bo_type_sg) {
+		bo->tbo.sg = sg;
+		bo->tbo.ttm->sg = sg;
+	}
+	bo->kfd_bo = *mem;
+	(*mem)->bo = bo;
+	if (user_addr)
+		bo->flags |= AMDGPU_AMDKFD_CREATE_USERPTR_BO;
+
+	(*mem)->va = va;
+	(*mem)->domain = domain;
+	(*mem)->mapped_to_gpu_memory = 0;
+	(*mem)->process_info = avm->process_info;
+
+	add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, user_addr);
+
+	if (user_addr) {
+		pr_debug("creating userptr BO for user_addr = %llx\n", user_addr);
+		ret = init_user_pages(*mem, user_addr, criu_resume);
+		if (ret)
+			goto allocate_init_user_pages_failed;
+	} else  if (flags & (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+				KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
+		ret = amdgpu_amdkfd_gpuvm_pin_bo(bo, AMDGPU_GEM_DOMAIN_GTT);
+		if (ret) {
+			pr_err("Pinning MMIO/DOORBELL BO during ALLOC FAILED\n");
+			goto err_pin_bo;
+		}
+		bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
+		bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
+	} else {
+		mutex_lock(&avm->process_info->lock);
+		if (avm->process_info->eviction_fence &&
+		    !dma_fence_is_signaled(&avm->process_info->eviction_fence->base))
+			ret = amdgpu_amdkfd_bo_validate_and_fence(bo, domain,
+				&avm->process_info->eviction_fence->base);
+		mutex_unlock(&avm->process_info->lock);
+		if (ret)
+			goto err_validate_bo;
+	}
+
+	if (offset)
+		*offset = amdgpu_bo_mmap_offset(bo);
+
+	return 0;
+
+allocate_init_user_pages_failed:
+err_pin_bo:
+err_validate_bo:
+	remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info);
+	drm_gem_handle_delete(adev->kfd.client.file, (*mem)->gem_handle);
+err_gem_handle_create:
+	drm_vma_node_revoke(&gobj->vma_node, drm_priv);
+err_node_allow:
+	/* Don't unreserve system mem limit twice */
+	goto err_reserve_limit;
+err_bo_create:
+	amdgpu_amdkfd_unreserve_mem_limit(adev, aligned_size, flags, xcp_id);
+err_reserve_limit:
+	amdgpu_sync_free(&(*mem)->sync);
+	mutex_destroy(&(*mem)->lock);
+	if (gobj)
+		drm_gem_object_put(gobj);
+	else
+		kfree(*mem);
+err:
+	if (sg) {
+		sg_free_table(sg);
+		kfree(sg);
+	}
+	return ret;
+}
+
+int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
+		struct amdgpu_device *adev, struct kgd_mem *mem, void *drm_priv,
+		uint64_t *size)
+{
+	struct amdkfd_process_info *process_info = mem->process_info;
+	unsigned long bo_size = mem->bo->tbo.base.size;
+	bool use_release_notifier = (mem->bo->kfd_bo == mem);
+	struct kfd_mem_attachment *entry, *tmp;
+	struct bo_vm_reservation_context ctx;
+	unsigned int mapped_to_gpu_memory;
+	int ret;
+	bool is_imported = false;
+
+	mutex_lock(&mem->lock);
+
+	/* Unpin MMIO/DOORBELL BO's that were pinned during allocation */
+	if (mem->alloc_flags &
+	    (KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL |
+	     KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP)) {
+		amdgpu_amdkfd_gpuvm_unpin_bo(mem->bo);
+	}
+
+	mapped_to_gpu_memory = mem->mapped_to_gpu_memory;
+	is_imported = mem->is_imported;
+	mutex_unlock(&mem->lock);
+	/* lock is not needed after this, since mem is unused and will
+	 * be freed anyway
+	 */
+
+	if (mapped_to_gpu_memory > 0) {
+		pr_debug("BO VA 0x%llx size 0x%lx is still mapped.\n",
+				mem->va, bo_size);
+		return -EBUSY;
+	}
+
+	/* Make sure restore workers don't access the BO any more */
+	mutex_lock(&process_info->lock);
+	list_del(&mem->validate_list);
+	mutex_unlock(&process_info->lock);
+
+	/* Cleanup user pages and MMU notifiers */
+	if (amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm)) {
+		amdgpu_hmm_unregister(mem->bo);
+		mutex_lock(&process_info->notifier_lock);
+		amdgpu_hmm_range_free(mem->range);
+		mutex_unlock(&process_info->notifier_lock);
+	}
+
+	ret = reserve_bo_and_cond_vms(mem, NULL, BO_VM_ALL, &ctx);
+	if (unlikely(ret))
+		return ret;
+
+	amdgpu_amdkfd_remove_eviction_fence(mem->bo,
+					process_info->eviction_fence);
+	pr_debug("Release VA 0x%llx - 0x%llx\n", mem->va,
+		mem->va + bo_size * (1 + mem->aql_queue));
+
+	/* Remove from VM internal data structures */
+	list_for_each_entry_safe(entry, tmp, &mem->attachments, list) {
+		kfd_mem_dmaunmap_attachment(mem, entry);
+		kfd_mem_detach(entry);
+	}
+
+	ret = unreserve_bo_and_vms(&ctx, false, false);
+
+	/* Free the sync object */
+	amdgpu_sync_free(&mem->sync);
+
+	/* If the SG is not NULL, it's one we created for a doorbell or mmio
+	 * remap BO. We need to free it.
+	 */
+	if (mem->bo->tbo.sg) {
+		sg_free_table(mem->bo->tbo.sg);
+		kfree(mem->bo->tbo.sg);
+	}
+
+	/* Update the size of the BO being freed if it was allocated from
+	 * VRAM and is not imported. For APP APU VRAM allocations are done
+	 * in GTT domain
+	 */
+	if (size) {
+		if (!is_imported &&
+		   mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM)
+			*size = bo_size;
+		else
+			*size = 0;
+	}
+
+	/* Free the BO*/
+	drm_vma_node_revoke(&mem->bo->tbo.base.vma_node, drm_priv);
+	drm_gem_handle_delete(adev->kfd.client.file, mem->gem_handle);
+	if (mem->dmabuf) {
+		dma_buf_put(mem->dmabuf);
+		mem->dmabuf = NULL;
+	}
+	mutex_destroy(&mem->lock);
+
+	/* If this releases the last reference, it will end up calling
+	 * amdgpu_amdkfd_release_notify and kfree the mem struct. That's why
+	 * this needs to be the last call here.
+	 */
+	drm_gem_object_put(&mem->bo->tbo.base);
+
+	/*
+	 * For kgd_mem allocated in amdgpu_amdkfd_gpuvm_import_dmabuf(),
+	 * explicitly free it here.
+	 */
+	if (!use_release_notifier)
+		kfree(mem);
+
+	return ret;
+}
+
+int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
+		struct amdgpu_device *adev, struct kgd_mem *mem,
+		void *drm_priv)
+{
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
+	int ret;
+	struct amdgpu_bo *bo;
+	uint32_t domain;
+	struct kfd_mem_attachment *entry;
+	struct bo_vm_reservation_context ctx;
+	unsigned long bo_size;
+	bool is_invalid_userptr = false;
+
+	bo = mem->bo;
+	if (!bo) {
+		pr_err("Invalid BO when mapping memory to GPU\n");
+		return -EINVAL;
+	}
+
+	/* Make sure restore is not running concurrently. Since we
+	 * don't map invalid userptr BOs, we rely on the next restore
+	 * worker to do the mapping
+	 */
+	mutex_lock(&mem->process_info->lock);
+
+	/* Lock notifier lock. If we find an invalid userptr BO, we can be
+	 * sure that the MMU notifier is no longer running
+	 * concurrently and the queues are actually stopped
+	 */
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) {
+		mutex_lock(&mem->process_info->notifier_lock);
+		is_invalid_userptr = !!mem->invalid;
+		mutex_unlock(&mem->process_info->notifier_lock);
+	}
+
+	mutex_lock(&mem->lock);
+
+	domain = mem->domain;
+	bo_size = bo->tbo.base.size;
+
+	pr_debug("Map VA 0x%llx - 0x%llx to vm %p domain %s\n",
+			mem->va,
+			mem->va + bo_size * (1 + mem->aql_queue),
+			avm, domain_string(domain));
+
+	if (!kfd_mem_is_attached(avm, mem)) {
+		ret = kfd_mem_attach(adev, mem, avm, mem->aql_queue);
+		if (ret)
+			goto out;
+	}
+
+	ret = reserve_bo_and_vm(mem, avm, &ctx);
+	if (unlikely(ret))
+		goto out;
+
+	/* Userptr can be marked as "not invalid", but not actually be
+	 * validated yet (still in the system domain). In that case
+	 * the queues are still stopped and we can leave mapping for
+	 * the next restore worker
+	 */
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) &&
+	    bo->tbo.resource->mem_type == TTM_PL_SYSTEM)
+		is_invalid_userptr = true;
+
+	ret = vm_validate_pt_pd_bos(avm, NULL);
+	if (unlikely(ret))
+		goto out_unreserve;
+
+	list_for_each_entry(entry, &mem->attachments, list) {
+		if (entry->bo_va->base.vm != avm || entry->is_mapped)
+			continue;
+
+		pr_debug("\t map VA 0x%llx - 0x%llx in entry %p\n",
+			 entry->va, entry->va + bo_size, entry);
+
+		ret = map_bo_to_gpuvm(mem, entry, ctx.sync,
+				      is_invalid_userptr);
+		if (ret) {
+			pr_err("Failed to map bo to gpuvm\n");
+			goto out_unreserve;
+		}
+
+		ret = vm_update_pds(avm, ctx.sync);
+		if (ret) {
+			pr_err("Failed to update page directories\n");
+			goto out_unreserve;
+		}
+
+		entry->is_mapped = true;
+		mem->mapped_to_gpu_memory++;
+		pr_debug("\t INC mapping count %d\n",
+			 mem->mapped_to_gpu_memory);
+	}
+
+	ret = unreserve_bo_and_vms(&ctx, false, false);
+
+	goto out;
+
+out_unreserve:
+	unreserve_bo_and_vms(&ctx, false, false);
+out:
+	mutex_unlock(&mem->process_info->lock);
+	mutex_unlock(&mem->lock);
+	return ret;
+}
+
+int amdgpu_amdkfd_gpuvm_dmaunmap_mem(struct kgd_mem *mem, void *drm_priv)
+{
+	struct kfd_mem_attachment *entry;
+	struct amdgpu_vm *vm;
+	int ret;
+
+	vm = drm_priv_to_vm(drm_priv);
+
+	mutex_lock(&mem->lock);
+
+	ret = amdgpu_bo_reserve(mem->bo, true);
+	if (ret)
+		goto out;
+
+	list_for_each_entry(entry, &mem->attachments, list) {
+		if (entry->bo_va->base.vm != vm)
+			continue;
+		if (entry->bo_va->base.bo->tbo.ttm &&
+		    !entry->bo_va->base.bo->tbo.ttm->sg)
+			continue;
+
+		kfd_mem_dmaunmap_attachment(mem, entry);
+	}
+
+	amdgpu_bo_unreserve(mem->bo);
+out:
+	mutex_unlock(&mem->lock);
+
+	return ret;
+}
+
+int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
+		struct amdgpu_device *adev, struct kgd_mem *mem, void *drm_priv)
+{
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
+	unsigned long bo_size = mem->bo->tbo.base.size;
+	struct kfd_mem_attachment *entry;
+	struct bo_vm_reservation_context ctx;
+	int ret;
+
+	mutex_lock(&mem->lock);
+
+	ret = reserve_bo_and_cond_vms(mem, avm, BO_VM_MAPPED, &ctx);
+	if (unlikely(ret))
+		goto out;
+	/* If no VMs were reserved, it means the BO wasn't actually mapped */
+	if (ctx.n_vms == 0) {
+		ret = -EINVAL;
+		goto unreserve_out;
+	}
+
+	ret = vm_validate_pt_pd_bos(avm, NULL);
+	if (unlikely(ret))
+		goto unreserve_out;
+
+	pr_debug("Unmap VA 0x%llx - 0x%llx from vm %p\n",
+		mem->va,
+		mem->va + bo_size * (1 + mem->aql_queue),
+		avm);
+
+	list_for_each_entry(entry, &mem->attachments, list) {
+		if (entry->bo_va->base.vm != avm || !entry->is_mapped)
+			continue;
+
+		pr_debug("\t unmap VA 0x%llx - 0x%llx from entry %p\n",
+			 entry->va, entry->va + bo_size, entry);
+
+		ret = unmap_bo_from_gpuvm(mem, entry, ctx.sync);
+		if (ret)
+			goto unreserve_out;
+
+		entry->is_mapped = false;
+
+		mem->mapped_to_gpu_memory--;
+		pr_debug("\t DEC mapping count %d\n",
+			 mem->mapped_to_gpu_memory);
+	}
+
+unreserve_out:
+	unreserve_bo_and_vms(&ctx, false, false);
+out:
+	mutex_unlock(&mem->lock);
+	return ret;
+}
+
+int amdgpu_amdkfd_gpuvm_sync_memory(
+		struct amdgpu_device *adev, struct kgd_mem *mem, bool intr)
+{
+	struct amdgpu_sync sync;
+	int ret;
+
+	amdgpu_sync_create(&sync);
+
+	mutex_lock(&mem->lock);
+	amdgpu_sync_clone(&mem->sync, &sync);
+	mutex_unlock(&mem->lock);
+
+	ret = amdgpu_sync_wait(&sync, intr);
+	amdgpu_sync_free(&sync);
+	return ret;
+}
+
+/**
+ * amdgpu_amdkfd_map_gtt_bo_to_gart - Map BO to GART and increment reference count
+ * @bo: Buffer object to be mapped
+ * @bo_gart: Return bo reference
+ *
+ * Before return, bo reference count is incremented. To release the reference and unpin/
+ * unmap the BO, call amdgpu_amdkfd_free_gtt_mem.
+ */
+int amdgpu_amdkfd_map_gtt_bo_to_gart(struct amdgpu_bo *bo, struct amdgpu_bo **bo_gart)
+{
+	int ret;
+
+	ret = amdgpu_bo_reserve(bo, true);
+	if (ret) {
+		pr_err("Failed to reserve bo. ret %d\n", ret);
+		goto err_reserve_bo_failed;
+	}
+
+	ret = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT);
+	if (ret) {
+		pr_err("Failed to pin bo. ret %d\n", ret);
+		goto err_pin_bo_failed;
+	}
+
+	ret = amdgpu_ttm_alloc_gart(&bo->tbo);
+	if (ret) {
+		pr_err("Failed to bind bo to GART. ret %d\n", ret);
+		goto err_map_bo_gart_failed;
+	}
+
+	amdgpu_amdkfd_remove_eviction_fence(
+		bo, bo->vm_bo->vm->process_info->eviction_fence);
+
+	amdgpu_bo_unreserve(bo);
+
+	*bo_gart = amdgpu_bo_ref(bo);
+
+	return 0;
+
+err_map_bo_gart_failed:
+	amdgpu_bo_unpin(bo);
+err_pin_bo_failed:
+	amdgpu_bo_unreserve(bo);
+err_reserve_bo_failed:
+
+	return ret;
+}
+
+/** amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel() - Map a GTT BO for kernel CPU access
+ *
+ * @mem: Buffer object to be mapped for CPU access
+ * @kptr[out]: pointer in kernel CPU address space
+ * @size[out]: size of the buffer
+ *
+ * Pins the BO and maps it for kernel CPU access. The eviction fence is removed
+ * from the BO, since pinned BOs cannot be evicted. The bo must remain on the
+ * validate_list, so the GPU mapping can be restored after a page table was
+ * evicted.
+ *
+ * Return: 0 on success, error code on failure
+ */
+int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_mem *mem,
+					     void **kptr, uint64_t *size)
+{
+	int ret;
+	struct amdgpu_bo *bo = mem->bo;
+
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) {
+		pr_err("userptr can't be mapped to kernel\n");
+		return -EINVAL;
+	}
+
+	mutex_lock(&mem->process_info->lock);
+
+	ret = amdgpu_bo_reserve(bo, true);
+	if (ret) {
+		pr_err("Failed to reserve bo. ret %d\n", ret);
+		goto bo_reserve_failed;
+	}
+
+	ret = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT);
+	if (ret) {
+		pr_err("Failed to pin bo. ret %d\n", ret);
+		goto pin_failed;
+	}
+
+	ret = amdgpu_bo_kmap(bo, kptr);
+	if (ret) {
+		pr_err("Failed to map bo to kernel. ret %d\n", ret);
+		goto kmap_failed;
+	}
+
+	amdgpu_amdkfd_remove_eviction_fence(
+		bo, mem->process_info->eviction_fence);
+
+	if (size)
+		*size = amdgpu_bo_size(bo);
+
+	amdgpu_bo_unreserve(bo);
+
+	mutex_unlock(&mem->process_info->lock);
+	return 0;
+
+kmap_failed:
+	amdgpu_bo_unpin(bo);
+pin_failed:
+	amdgpu_bo_unreserve(bo);
+bo_reserve_failed:
+	mutex_unlock(&mem->process_info->lock);
+
+	return ret;
+}
+
+/** amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel() - Unmap a GTT BO for kernel CPU access
+ *
+ * @mem: Buffer object to be unmapped for CPU access
+ *
+ * Removes the kernel CPU mapping and unpins the BO. It does not restore the
+ * eviction fence, so this function should only be used for cleanup before the
+ * BO is destroyed.
+ */
+void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_mem *mem)
+{
+	struct amdgpu_bo *bo = mem->bo;
+
+	(void)amdgpu_bo_reserve(bo, true);
+	amdgpu_bo_kunmap(bo);
+	amdgpu_bo_unpin(bo);
+	amdgpu_bo_unreserve(bo);
+}
+
+int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
+					  struct kfd_vm_fault_info *mem)
+{
+	if (atomic_read_acquire(&adev->gmc.vm_fault_info_updated) == 1) {
+		*mem = *adev->gmc.vm_fault_info;
+		atomic_set_release(&adev->gmc.vm_fault_info_updated, 0);
+	}
+	return 0;
+}
+
+static int import_obj_create(struct amdgpu_device *adev,
+			     struct dma_buf *dma_buf,
+			     struct drm_gem_object *obj,
+			     uint64_t va, void *drm_priv,
+			     struct kgd_mem **mem, uint64_t *size,
+			     uint64_t *mmap_offset)
+{
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
+	struct amdgpu_bo *bo;
+	int ret;
+
+	bo = gem_to_amdgpu_bo(obj);
+	if (!(bo->preferred_domains & (AMDGPU_GEM_DOMAIN_VRAM |
+				    AMDGPU_GEM_DOMAIN_GTT)))
+		/* Only VRAM and GTT BOs are supported */
+		return -EINVAL;
+
+	*mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	if (!*mem)
+		return -ENOMEM;
+
+	ret = drm_vma_node_allow(&obj->vma_node, drm_priv);
+	if (ret)
+		goto err_free_mem;
+
+	if (size)
+		*size = amdgpu_bo_size(bo);
+
+	if (mmap_offset)
+		*mmap_offset = amdgpu_bo_mmap_offset(bo);
+
+	INIT_LIST_HEAD(&(*mem)->attachments);
+	mutex_init(&(*mem)->lock);
+
+	(*mem)->alloc_flags =
+		((bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) ?
+		KFD_IOC_ALLOC_MEM_FLAGS_VRAM : KFD_IOC_ALLOC_MEM_FLAGS_GTT)
+		| KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE
+		| KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE;
+
+	get_dma_buf(dma_buf);
+	(*mem)->dmabuf = dma_buf;
+	(*mem)->bo = bo;
+	(*mem)->va = va;
+	(*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) &&
+			 !adev->apu_prefer_gtt ?
+			 AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT;
+
+	(*mem)->mapped_to_gpu_memory = 0;
+	(*mem)->process_info = avm->process_info;
+	add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, false);
+	amdgpu_sync_create(&(*mem)->sync);
+	(*mem)->is_imported = true;
+
+	mutex_lock(&avm->process_info->lock);
+	if (avm->process_info->eviction_fence &&
+	    !dma_fence_is_signaled(&avm->process_info->eviction_fence->base))
+		ret = amdgpu_amdkfd_bo_validate_and_fence(bo, (*mem)->domain,
+				&avm->process_info->eviction_fence->base);
+	mutex_unlock(&avm->process_info->lock);
+	if (ret)
+		goto err_remove_mem;
+
+	return 0;
+
+err_remove_mem:
+	remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info);
+	drm_vma_node_revoke(&obj->vma_node, drm_priv);
+err_free_mem:
+	kfree(*mem);
+	return ret;
+}
+
+int amdgpu_amdkfd_gpuvm_import_dmabuf_fd(struct amdgpu_device *adev, int fd,
+					 uint64_t va, void *drm_priv,
+					 struct kgd_mem **mem, uint64_t *size,
+					 uint64_t *mmap_offset)
+{
+	struct drm_gem_object *obj;
+	uint32_t handle;
+	int ret;
+
+	ret = drm_gem_prime_fd_to_handle(&adev->ddev, adev->kfd.client.file, fd,
+					 &handle);
+	if (ret)
+		return ret;
+	obj = drm_gem_object_lookup(adev->kfd.client.file, handle);
+	if (!obj) {
+		ret = -EINVAL;
+		goto err_release_handle;
+	}
+
+	ret = import_obj_create(adev, obj->dma_buf, obj, va, drm_priv, mem, size,
+				mmap_offset);
+	if (ret)
+		goto err_put_obj;
+
+	(*mem)->gem_handle = handle;
+
+	return 0;
+
+err_put_obj:
+	drm_gem_object_put(obj);
+err_release_handle:
+	drm_gem_handle_delete(adev->kfd.client.file, handle);
+	return ret;
+}
+
+int amdgpu_amdkfd_gpuvm_export_dmabuf(struct kgd_mem *mem,
+				      struct dma_buf **dma_buf)
+{
+	int ret;
+
+	mutex_lock(&mem->lock);
+	ret = kfd_mem_export_dmabuf(mem);
+	if (ret)
+		goto out;
+
+	get_dma_buf(mem->dmabuf);
+	*dma_buf = mem->dmabuf;
+out:
+	mutex_unlock(&mem->lock);
+	return ret;
+}
+
+/* Evict a userptr BO by stopping the queues if necessary
+ *
+ * Runs in MMU notifier, may be in RECLAIM_FS context. This means it
+ * cannot do any memory allocations, and cannot take any locks that
+ * are held elsewhere while allocating memory.
+ *
+ * It doesn't do anything to the BO itself. The real work happens in
+ * restore, where we get updated page addresses. This function only
+ * ensures that GPU access to the BO is stopped.
+ */
+int amdgpu_amdkfd_evict_userptr(struct mmu_interval_notifier *mni,
+				unsigned long cur_seq, struct kgd_mem *mem)
+{
+	struct amdkfd_process_info *process_info = mem->process_info;
+	int r = 0;
+
+	/* Do not process MMU notifications during CRIU restore until
+	 * KFD_CRIU_OP_RESUME IOCTL is received
+	 */
+	if (READ_ONCE(process_info->block_mmu_notifications))
+		return 0;
+
+	mutex_lock(&process_info->notifier_lock);
+	mmu_interval_set_seq(mni, cur_seq);
+
+	mem->invalid++;
+	if (++process_info->evicted_bos == 1) {
+		/* First eviction, stop the queues */
+		r = kgd2kfd_quiesce_mm(mni->mm,
+				       KFD_QUEUE_EVICTION_TRIGGER_USERPTR);
+
+		if (r && r != -ESRCH)
+			pr_err("Failed to quiesce KFD\n");
+
+		if (r != -ESRCH)
+			queue_delayed_work(system_freezable_wq,
+				&process_info->restore_userptr_work,
+				msecs_to_jiffies(AMDGPU_USERPTR_RESTORE_DELAY_MS));
+	}
+	mutex_unlock(&process_info->notifier_lock);
+
+	return r;
+}
+
+/* Update invalid userptr BOs
+ *
+ * Moves invalidated (evicted) userptr BOs from userptr_valid_list to
+ * userptr_inval_list and updates user pages for all BOs that have
+ * been invalidated since their last update.
+ */
+static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
+				     struct mm_struct *mm)
+{
+	struct kgd_mem *mem, *tmp_mem;
+	struct amdgpu_bo *bo;
+	struct ttm_operation_ctx ctx = { false, false };
+	uint32_t invalid;
+	int ret = 0;
+
+	mutex_lock(&process_info->notifier_lock);
+
+	/* Move all invalidated BOs to the userptr_inval_list */
+	list_for_each_entry_safe(mem, tmp_mem,
+				 &process_info->userptr_valid_list,
+				 validate_list)
+		if (mem->invalid)
+			list_move_tail(&mem->validate_list,
+				       &process_info->userptr_inval_list);
+
+	/* Go through userptr_inval_list and update any invalid user_pages */
+	list_for_each_entry(mem, &process_info->userptr_inval_list,
+			    validate_list) {
+		invalid = mem->invalid;
+		if (!invalid)
+			/* BO hasn't been invalidated since the last
+			 * revalidation attempt. Keep its page list.
+			 */
+			continue;
+
+		bo = mem->bo;
+
+		amdgpu_hmm_range_free(mem->range);
+		mem->range = NULL;
+
+		/* BO reservations and getting user pages (hmm_range_fault)
+		 * must happen outside the notifier lock
+		 */
+		mutex_unlock(&process_info->notifier_lock);
+
+		/* Move the BO to system (CPU) domain if necessary to unmap
+		 * and free the SG table
+		 */
+		if (bo->tbo.resource->mem_type != TTM_PL_SYSTEM) {
+			if (amdgpu_bo_reserve(bo, true))
+				return -EAGAIN;
+			amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+			ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+			amdgpu_bo_unreserve(bo);
+			if (ret) {
+				pr_err("%s: Failed to invalidate userptr BO\n",
+				       __func__);
+				return -EAGAIN;
+			}
+		}
+
+		mem->range = amdgpu_hmm_range_alloc(NULL);
+		if (unlikely(!mem->range))
+			return -ENOMEM;
+		/* Get updated user pages */
+		ret = amdgpu_ttm_tt_get_user_pages(bo, mem->range);
+		if (ret) {
+			amdgpu_hmm_range_free(mem->range);
+			mem->range = NULL;
+			pr_debug("Failed %d to get user pages\n", ret);
+
+			/* Return -EFAULT bad address error as success. It will
+			 * fail later with a VM fault if the GPU tries to access
+			 * it. Better than hanging indefinitely with stalled
+			 * user mode queues.
+			 *
+			 * Return other error -EBUSY or -ENOMEM to retry restore
+			 */
+			if (ret != -EFAULT)
+				return ret;
+
+			/* If applications unmap memory before destroying the userptr
+			 * from the KFD, trigger a segmentation fault in VM debug mode.
+			 */
+			if (amdgpu_ttm_adev(bo->tbo.bdev)->debug_vm_userptr) {
+				struct kfd_process *p;
+
+				pr_err("Pid %d unmapped memory before destroying userptr at GPU addr 0x%llx\n",
+								pid_nr(process_info->pid), mem->va);
+
+				// Send GPU VM fault to user space
+				p = kfd_lookup_process_by_pid(process_info->pid);
+				if (p) {
+					kfd_signal_vm_fault_event_with_userptr(p, mem->va);
+					kfd_unref_process(p);
+				}
+			}
+
+			ret = 0;
+		}
+
+		amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, mem->range);
+
+		mutex_lock(&process_info->notifier_lock);
+
+		/* Mark the BO as valid unless it was invalidated
+		 * again concurrently.
+		 */
+		if (mem->invalid != invalid) {
+			ret = -EAGAIN;
+			goto unlock_out;
+		}
+		 /* set mem valid if mem has hmm range associated */
+		if (mem->range)
+			mem->invalid = 0;
+	}
+
+unlock_out:
+	mutex_unlock(&process_info->notifier_lock);
+
+	return ret;
+}
+
+/* Validate invalid userptr BOs
+ *
+ * Validates BOs on the userptr_inval_list. Also updates GPUVM page tables
+ * with new page addresses and waits for the page table updates to complete.
+ */
+static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
+{
+	struct ttm_operation_ctx ctx = { false, false };
+	struct amdgpu_sync sync;
+	struct drm_exec exec;
+
+	struct amdgpu_vm *peer_vm;
+	struct kgd_mem *mem, *tmp_mem;
+	struct amdgpu_bo *bo;
+	int ret;
+
+	amdgpu_sync_create(&sync);
+
+	drm_exec_init(&exec, 0, 0);
+	/* Reserve all BOs and page tables for validation */
+	drm_exec_until_all_locked(&exec) {
+		/* Reserve all the page directories */
+		list_for_each_entry(peer_vm, &process_info->vm_list_head,
+				    vm_list_node) {
+			ret = amdgpu_vm_lock_pd(peer_vm, &exec, 2);
+			drm_exec_retry_on_contention(&exec);
+			if (unlikely(ret))
+				goto unreserve_out;
+		}
+
+		/* Reserve the userptr_inval_list entries to resv_list */
+		list_for_each_entry(mem, &process_info->userptr_inval_list,
+				    validate_list) {
+			struct drm_gem_object *gobj;
+
+			gobj = &mem->bo->tbo.base;
+			ret = drm_exec_prepare_obj(&exec, gobj, 1);
+			drm_exec_retry_on_contention(&exec);
+			if (unlikely(ret))
+				goto unreserve_out;
+		}
+	}
+
+	ret = process_validate_vms(process_info, NULL);
+	if (ret)
+		goto unreserve_out;
+
+	/* Validate BOs and update GPUVM page tables */
+	list_for_each_entry_safe(mem, tmp_mem,
+				 &process_info->userptr_inval_list,
+				 validate_list) {
+		struct kfd_mem_attachment *attachment;
+
+		bo = mem->bo;
+
+		/* Validate the BO if we got user pages */
+		if (bo->tbo.ttm->pages[0]) {
+			amdgpu_bo_placement_from_domain(bo, mem->domain);
+			ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+			if (ret) {
+				pr_err("%s: failed to validate BO\n", __func__);
+				goto unreserve_out;
+			}
+		}
+
+		/* Update mapping. If the BO was not validated
+		 * (because we couldn't get user pages), this will
+		 * clear the page table entries, which will result in
+		 * VM faults if the GPU tries to access the invalid
+		 * memory.
+		 */
+		list_for_each_entry(attachment, &mem->attachments, list) {
+			if (!attachment->is_mapped)
+				continue;
+
+			kfd_mem_dmaunmap_attachment(mem, attachment);
+			ret = update_gpuvm_pte(mem, attachment, &sync);
+			if (ret) {
+				pr_err("%s: update PTE failed\n", __func__);
+				/* make sure this gets validated again */
+				mutex_lock(&process_info->notifier_lock);
+				mem->invalid++;
+				mutex_unlock(&process_info->notifier_lock);
+				goto unreserve_out;
+			}
+		}
+	}
+
+	/* Update page directories */
+	ret = process_update_pds(process_info, &sync);
+
+unreserve_out:
+	drm_exec_fini(&exec);
+	amdgpu_sync_wait(&sync, false);
+	amdgpu_sync_free(&sync);
+
+	return ret;
+}
+
+/* Confirm that all user pages are valid while holding the notifier lock
+ *
+ * Moves valid BOs from the userptr_inval_list back to userptr_val_list.
+ */
+static int confirm_valid_user_pages_locked(struct amdkfd_process_info *process_info)
+{
+	struct kgd_mem *mem, *tmp_mem;
+	int ret = 0;
+
+	list_for_each_entry_safe(mem, tmp_mem,
+				 &process_info->userptr_inval_list,
+				 validate_list) {
+		bool valid;
+
+		/* keep mem without hmm range at userptr_inval_list */
+		if (!mem->range)
+			continue;
+
+		/* Only check mem with hmm range associated */
+		valid = amdgpu_hmm_range_valid(mem->range);
+		amdgpu_hmm_range_free(mem->range);
+
+		mem->range = NULL;
+		if (!valid) {
+			WARN(!mem->invalid, "Invalid BO not marked invalid");
+			ret = -EAGAIN;
+			continue;
+		}
+
+		if (mem->invalid) {
+			WARN(1, "Valid BO is marked invalid");
+			ret = -EAGAIN;
+			continue;
+		}
+
+		list_move_tail(&mem->validate_list,
+			       &process_info->userptr_valid_list);
+	}
+
+	return ret;
+}
+
+/* Worker callback to restore evicted userptr BOs
+ *
+ * Tries to update and validate all userptr BOs. If successful and no
+ * concurrent evictions happened, the queues are restarted. Otherwise,
+ * reschedule for another attempt later.
+ */
+static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work)
+{
+	struct delayed_work *dwork = to_delayed_work(work);
+	struct amdkfd_process_info *process_info =
+		container_of(dwork, struct amdkfd_process_info,
+			     restore_userptr_work);
+	struct task_struct *usertask;
+	struct mm_struct *mm;
+	uint32_t evicted_bos;
+
+	mutex_lock(&process_info->notifier_lock);
+	evicted_bos = process_info->evicted_bos;
+	mutex_unlock(&process_info->notifier_lock);
+	if (!evicted_bos)
+		return;
+
+	/* Reference task and mm in case of concurrent process termination */
+	usertask = get_pid_task(process_info->pid, PIDTYPE_PID);
+	if (!usertask)
+		return;
+	mm = get_task_mm(usertask);
+	if (!mm) {
+		put_task_struct(usertask);
+		return;
+	}
+
+	mutex_lock(&process_info->lock);
+
+	if (update_invalid_user_pages(process_info, mm))
+		goto unlock_out;
+	/* userptr_inval_list can be empty if all evicted userptr BOs
+	 * have been freed. In that case there is nothing to validate
+	 * and we can just restart the queues.
+	 */
+	if (!list_empty(&process_info->userptr_inval_list)) {
+		if (validate_invalid_user_pages(process_info))
+			goto unlock_out;
+	}
+	/* Final check for concurrent evicton and atomic update. If
+	 * another eviction happens after successful update, it will
+	 * be a first eviction that calls quiesce_mm. The eviction
+	 * reference counting inside KFD will handle this case.
+	 */
+	mutex_lock(&process_info->notifier_lock);
+	if (process_info->evicted_bos != evicted_bos)
+		goto unlock_notifier_out;
+
+	if (confirm_valid_user_pages_locked(process_info)) {
+		WARN(1, "User pages unexpectedly invalid");
+		goto unlock_notifier_out;
+	}
+
+	process_info->evicted_bos = evicted_bos = 0;
+
+	if (kgd2kfd_resume_mm(mm)) {
+		pr_err("%s: Failed to resume KFD\n", __func__);
+		/* No recovery from this failure. Probably the CP is
+		 * hanging. No point trying again.
+		 */
+	}
+
+unlock_notifier_out:
+	mutex_unlock(&process_info->notifier_lock);
+unlock_out:
+	mutex_unlock(&process_info->lock);
+
+	/* If validation failed, reschedule another attempt */
+	if (evicted_bos) {
+		queue_delayed_work(system_freezable_wq,
+			&process_info->restore_userptr_work,
+			msecs_to_jiffies(AMDGPU_USERPTR_RESTORE_DELAY_MS));
+
+		kfd_smi_event_queue_restore_rescheduled(mm);
+	}
+	mmput(mm);
+	put_task_struct(usertask);
+}
+
+static void replace_eviction_fence(struct dma_fence __rcu **ef,
+				   struct dma_fence *new_ef)
+{
+	struct dma_fence *old_ef = rcu_replace_pointer(*ef, new_ef, true
+		/* protected by process_info->lock */);
+
+	/* If we're replacing an unsignaled eviction fence, that fence will
+	 * never be signaled, and if anyone is still waiting on that fence,
+	 * they will hang forever. This should never happen. We should only
+	 * replace the fence in restore_work that only gets scheduled after
+	 * eviction work signaled the fence.
+	 */
+	WARN_ONCE(!dma_fence_is_signaled(old_ef),
+		  "Replacing unsignaled eviction fence");
+	dma_fence_put(old_ef);
+}
+
+/** amdgpu_amdkfd_gpuvm_restore_process_bos - Restore all BOs for the given
+ *   KFD process identified by process_info
+ *
+ * @process_info: amdkfd_process_info of the KFD process
+ *
+ * After memory eviction, restore thread calls this function. The function
+ * should be called when the Process is still valid. BO restore involves -
+ *
+ * 1.  Release old eviction fence and create new one
+ * 2.  Get two copies of PD BO list from all the VMs. Keep one copy as pd_list.
+ * 3   Use the second PD list and kfd_bo_list to create a list (ctx.list) of
+ *     BOs that need to be reserved.
+ * 4.  Reserve all the BOs
+ * 5.  Validate of PD and PT BOs.
+ * 6.  Validate all KFD BOs using kfd_bo_list and Map them and add new fence
+ * 7.  Add fence to all PD and PT BOs.
+ * 8.  Unreserve all BOs
+ */
+int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu **ef)
+{
+	struct amdkfd_process_info *process_info = info;
+	struct amdgpu_vm *peer_vm;
+	struct kgd_mem *mem;
+	struct list_head duplicate_save;
+	struct amdgpu_sync sync_obj;
+	unsigned long failed_size = 0;
+	unsigned long total_size = 0;
+	struct drm_exec exec;
+	int ret;
+
+	INIT_LIST_HEAD(&duplicate_save);
+
+	mutex_lock(&process_info->lock);
+
+	drm_exec_init(&exec, DRM_EXEC_IGNORE_DUPLICATES, 0);
+	drm_exec_until_all_locked(&exec) {
+		list_for_each_entry(peer_vm, &process_info->vm_list_head,
+				    vm_list_node) {
+			ret = amdgpu_vm_lock_pd(peer_vm, &exec, 2);
+			drm_exec_retry_on_contention(&exec);
+			if (unlikely(ret)) {
+				pr_err("Locking VM PD failed, ret: %d\n", ret);
+				goto ttm_reserve_fail;
+			}
+		}
+
+		/* Reserve all BOs and page tables/directory. Add all BOs from
+		 * kfd_bo_list to ctx.list
+		 */
+		list_for_each_entry(mem, &process_info->kfd_bo_list,
+				    validate_list) {
+			struct drm_gem_object *gobj;
+
+			gobj = &mem->bo->tbo.base;
+			ret = drm_exec_prepare_obj(&exec, gobj, 1);
+			drm_exec_retry_on_contention(&exec);
+			if (unlikely(ret)) {
+				pr_err("drm_exec_prepare_obj failed, ret: %d\n", ret);
+				goto ttm_reserve_fail;
+			}
+		}
+	}
+
+	amdgpu_sync_create(&sync_obj);
+
+	/* Validate BOs managed by KFD */
+	list_for_each_entry(mem, &process_info->kfd_bo_list,
+			    validate_list) {
+
+		struct amdgpu_bo *bo = mem->bo;
+		uint32_t domain = mem->domain;
+		struct dma_resv_iter cursor;
+		struct dma_fence *fence;
+
+		total_size += amdgpu_bo_size(bo);
+
+		ret = amdgpu_amdkfd_bo_validate(bo, domain, false);
+		if (ret) {
+			pr_debug("Memory eviction: Validate BOs failed\n");
+			failed_size += amdgpu_bo_size(bo);
+			ret = amdgpu_amdkfd_bo_validate(bo,
+						AMDGPU_GEM_DOMAIN_GTT, false);
+			if (ret) {
+				pr_debug("Memory eviction: Try again\n");
+				goto validate_map_fail;
+			}
+		}
+		dma_resv_for_each_fence(&cursor, bo->tbo.base.resv,
+					DMA_RESV_USAGE_KERNEL, fence) {
+			ret = amdgpu_sync_fence(&sync_obj, fence, GFP_KERNEL);
+			if (ret) {
+				pr_debug("Memory eviction: Sync BO fence failed. Try again\n");
+				goto validate_map_fail;
+			}
+		}
+	}
+
+	if (failed_size)
+		pr_debug("0x%lx/0x%lx in system\n", failed_size, total_size);
+
+	/* Validate PDs, PTs and evicted DMABuf imports last. Otherwise BO
+	 * validations above would invalidate DMABuf imports again.
+	 */
+	ret = process_validate_vms(process_info, &exec.ticket);
+	if (ret) {
+		pr_debug("Validating VMs failed, ret: %d\n", ret);
+		goto validate_map_fail;
+	}
+
+	/* Update mappings managed by KFD. */
+	list_for_each_entry(mem, &process_info->kfd_bo_list,
+			    validate_list) {
+		struct kfd_mem_attachment *attachment;
+
+		list_for_each_entry(attachment, &mem->attachments, list) {
+			if (!attachment->is_mapped)
+				continue;
+
+			kfd_mem_dmaunmap_attachment(mem, attachment);
+			ret = update_gpuvm_pte(mem, attachment, &sync_obj);
+			if (ret) {
+				pr_debug("Memory eviction: update PTE failed. Try again\n");
+				goto validate_map_fail;
+			}
+		}
+	}
+
+	/* Update mappings not managed by KFD */
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			vm_list_node) {
+		struct amdgpu_device *adev = amdgpu_ttm_adev(
+			peer_vm->root.bo->tbo.bdev);
+
+		struct amdgpu_fpriv *fpriv =
+			container_of(peer_vm, struct amdgpu_fpriv, vm);
+
+		ret = amdgpu_vm_bo_update(adev, fpriv->prt_va, false);
+		if (ret) {
+			dev_dbg(adev->dev,
+				"Memory eviction: handle PRT moved failed, pid %8d. Try again.\n",
+				pid_nr(process_info->pid));
+			goto validate_map_fail;
+		}
+
+		ret = amdgpu_vm_handle_moved(adev, peer_vm, &exec.ticket);
+		if (ret) {
+			dev_dbg(adev->dev,
+				"Memory eviction: handle moved failed, pid %8d. Try again.\n",
+				pid_nr(process_info->pid));
+			goto validate_map_fail;
+		}
+	}
+
+	/* Update page directories */
+	ret = process_update_pds(process_info, &sync_obj);
+	if (ret) {
+		pr_debug("Memory eviction: update PDs failed. Try again\n");
+		goto validate_map_fail;
+	}
+
+	/* Sync with fences on all the page tables. They implicitly depend on any
+	 * move fences from amdgpu_vm_handle_moved above.
+	 */
+	ret = process_sync_pds_resv(process_info, &sync_obj);
+	if (ret) {
+		pr_debug("Memory eviction: Failed to sync to PD BO moving fence. Try again\n");
+		goto validate_map_fail;
+	}
+
+	/* Wait for validate and PT updates to finish */
+	amdgpu_sync_wait(&sync_obj, false);
+
+	/* The old eviction fence may be unsignaled if restore happens
+	 * after a GPU reset or suspend/resume. Keep the old fence in that
+	 * case. Otherwise release the old eviction fence and create new
+	 * one, because fence only goes from unsignaled to signaled once
+	 * and cannot be reused. Use context and mm from the old fence.
+	 *
+	 * If an old eviction fence signals after this check, that's OK.
+	 * Anyone signaling an eviction fence must stop the queues first
+	 * and schedule another restore worker.
+	 */
+	if (dma_fence_is_signaled(&process_info->eviction_fence->base)) {
+		struct amdgpu_amdkfd_fence *new_fence =
+			amdgpu_amdkfd_fence_create(
+				process_info->eviction_fence->base.context,
+				process_info->eviction_fence->mm,
+				NULL);
+
+		if (!new_fence) {
+			pr_err("Failed to create eviction fence\n");
+			ret = -ENOMEM;
+			goto validate_map_fail;
+		}
+		dma_fence_put(&process_info->eviction_fence->base);
+		process_info->eviction_fence = new_fence;
+		replace_eviction_fence(ef, dma_fence_get(&new_fence->base));
+	} else {
+		WARN_ONCE(*ef != &process_info->eviction_fence->base,
+			  "KFD eviction fence doesn't match KGD process_info");
+	}
+
+	/* Attach new eviction fence to all BOs except pinned ones */
+	list_for_each_entry(mem, &process_info->kfd_bo_list, validate_list) {
+		if (mem->bo->tbo.pin_count)
+			continue;
+
+		dma_resv_add_fence(mem->bo->tbo.base.resv,
+				   &process_info->eviction_fence->base,
+				   DMA_RESV_USAGE_BOOKKEEP);
+	}
+	/* Attach eviction fence to PD / PT BOs and DMABuf imports */
+	list_for_each_entry(peer_vm, &process_info->vm_list_head,
+			    vm_list_node) {
+		struct amdgpu_bo *bo = peer_vm->root.bo;
+
+		dma_resv_add_fence(bo->tbo.base.resv,
+				   &process_info->eviction_fence->base,
+				   DMA_RESV_USAGE_BOOKKEEP);
+	}
+
+validate_map_fail:
+	amdgpu_sync_free(&sync_obj);
+ttm_reserve_fail:
+	drm_exec_fini(&exec);
+	mutex_unlock(&process_info->lock);
+	return ret;
+}
+
+int amdgpu_amdkfd_add_gws_to_process(void *info, void *gws, struct kgd_mem **mem)
+{
+	struct amdkfd_process_info *process_info = (struct amdkfd_process_info *)info;
+	struct amdgpu_bo *gws_bo = (struct amdgpu_bo *)gws;
+	int ret;
+
+	if (!info || !gws)
+		return -EINVAL;
+
+	*mem = kzalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	if (!*mem)
+		return -ENOMEM;
+
+	mutex_init(&(*mem)->lock);
+	INIT_LIST_HEAD(&(*mem)->attachments);
+	(*mem)->bo = amdgpu_bo_ref(gws_bo);
+	(*mem)->domain = AMDGPU_GEM_DOMAIN_GWS;
+	(*mem)->process_info = process_info;
+	add_kgd_mem_to_kfd_bo_list(*mem, process_info, false);
+	amdgpu_sync_create(&(*mem)->sync);
+
+
+	/* Validate gws bo the first time it is added to process */
+	mutex_lock(&(*mem)->process_info->lock);
+	ret = amdgpu_bo_reserve(gws_bo, false);
+	if (unlikely(ret)) {
+		pr_err("Reserve gws bo failed %d\n", ret);
+		goto bo_reservation_failure;
+	}
+
+	ret = amdgpu_amdkfd_bo_validate(gws_bo, AMDGPU_GEM_DOMAIN_GWS, true);
+	if (ret) {
+		pr_err("GWS BO validate failed %d\n", ret);
+		goto bo_validation_failure;
+	}
+	/* GWS resource is shared b/t amdgpu and amdkfd
+	 * Add process eviction fence to bo so they can
+	 * evict each other.
+	 */
+	ret = dma_resv_reserve_fences(gws_bo->tbo.base.resv, 1);
+	if (ret)
+		goto reserve_shared_fail;
+	dma_resv_add_fence(gws_bo->tbo.base.resv,
+			   &process_info->eviction_fence->base,
+			   DMA_RESV_USAGE_BOOKKEEP);
+	amdgpu_bo_unreserve(gws_bo);
+	mutex_unlock(&(*mem)->process_info->lock);
+
+	return ret;
+
+reserve_shared_fail:
+bo_validation_failure:
+	amdgpu_bo_unreserve(gws_bo);
+bo_reservation_failure:
+	mutex_unlock(&(*mem)->process_info->lock);
+	amdgpu_sync_free(&(*mem)->sync);
+	remove_kgd_mem_from_kfd_bo_list(*mem, process_info);
+	amdgpu_bo_unref(&gws_bo);
+	mutex_destroy(&(*mem)->lock);
+	kfree(*mem);
+	*mem = NULL;
+	return ret;
+}
+
+int amdgpu_amdkfd_remove_gws_from_process(void *info, void *mem)
+{
+	int ret;
+	struct amdkfd_process_info *process_info = (struct amdkfd_process_info *)info;
+	struct kgd_mem *kgd_mem = (struct kgd_mem *)mem;
+	struct amdgpu_bo *gws_bo = kgd_mem->bo;
+
+	/* Remove BO from process's validate list so restore worker won't touch
+	 * it anymore
+	 */
+	remove_kgd_mem_from_kfd_bo_list(kgd_mem, process_info);
+
+	ret = amdgpu_bo_reserve(gws_bo, false);
+	if (unlikely(ret)) {
+		pr_err("Reserve gws bo failed %d\n", ret);
+		//TODO add BO back to validate_list?
+		return ret;
+	}
+	amdgpu_amdkfd_remove_eviction_fence(gws_bo,
+			process_info->eviction_fence);
+	amdgpu_bo_unreserve(gws_bo);
+	amdgpu_sync_free(&kgd_mem->sync);
+	amdgpu_bo_unref(&gws_bo);
+	mutex_destroy(&kgd_mem->lock);
+	kfree(mem);
+	return 0;
+}
+
+/* Returns GPU-specific tiling mode information */
+int amdgpu_amdkfd_get_tile_config(struct amdgpu_device *adev,
+				struct tile_config *config)
+{
+	config->gb_addr_config = adev->gfx.config.gb_addr_config;
+	config->tile_config_ptr = adev->gfx.config.tile_mode_array;
+	config->num_tile_configs =
+			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
+	config->macro_tile_config_ptr =
+			adev->gfx.config.macrotile_mode_array;
+	config->num_macro_tile_configs =
+			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
+
+	/* Those values are not set from GFX9 onwards */
+	config->num_banks = adev->gfx.config.num_banks;
+	config->num_ranks = adev->gfx.config.num_ranks;
+
+	return 0;
+}
+
+bool amdgpu_amdkfd_bo_mapped_to_dev(void *drm_priv, struct kgd_mem *mem)
+{
+	struct amdgpu_vm *vm = drm_priv_to_vm(drm_priv);
+	struct kfd_mem_attachment *entry;
+
+	list_for_each_entry(entry, &mem->attachments, list) {
+		if (entry->is_mapped && entry->bo_va->base.vm == vm)
+			return true;
+	}
+	return false;
+}
+
+#if defined(CONFIG_DEBUG_FS)
+
+int kfd_debugfs_kfd_mem_limits(struct seq_file *m, void *data)
+{
+
+	spin_lock(&kfd_mem_limit.mem_limit_lock);
+	seq_printf(m, "System mem used %lldM out of %lluM\n",
+		  (kfd_mem_limit.system_mem_used >> 20),
+		  (kfd_mem_limit.max_system_mem_limit >> 20));
+	seq_printf(m, "TTM mem used %lldM out of %lluM\n",
+		  (kfd_mem_limit.ttm_mem_used >> 20),
+		  (kfd_mem_limit.max_ttm_mem_limit >> 20));
+	spin_unlock(&kfd_mem_limit.mem_limit_lock);
+
+	return 0;
+}
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c
index 1cf78f4dd339..763f2b8dcf13 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c
@@ -23,24 +23,19 @@
  * Authors: Dave Airlie
  *          Alex Deucher
  */
-#include <drm/drmP.h>
+
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 #include "amdgpu_atombios.h"
+#include "amdgpu_atomfirmware.h"
 #include "amdgpu_i2c.h"
+#include "amdgpu_display.h"
 
 #include "atom.h"
 #include "atom-bits.h"
 #include "atombios_encoders.h"
 #include "bif/bif_4_1_d.h"
 
-static void amdgpu_atombios_lookup_i2c_gpio_quirks(struct amdgpu_device *adev,
-					  ATOM_GPIO_I2C_ASSIGMENT *gpio,
-					  u8 index)
-{
-
-}
-
 static struct amdgpu_i2c_bus_rec amdgpu_atombios_get_bus_rec_for_i2c_gpio(ATOM_GPIO_I2C_ASSIGMENT *gpio)
 {
 	struct amdgpu_i2c_bus_rec i2c;
@@ -106,9 +101,6 @@ struct amdgpu_i2c_bus_rec amdgpu_atombios_lookup_i2c_gpio(struct amdgpu_device *
 
 		gpio = &i2c_info->asGPIO_Info[0];
 		for (i = 0; i < num_indices; i++) {
-
-			amdgpu_atombios_lookup_i2c_gpio_quirks(adev, gpio, i);
-
 			if (gpio->sucI2cId.ucAccess == id) {
 				i2c = amdgpu_atombios_get_bus_rec_for_i2c_gpio(gpio);
 				break;
@@ -140,13 +132,43 @@ void amdgpu_atombios_i2c_init(struct amdgpu_device *adev)
 
 		gpio = &i2c_info->asGPIO_Info[0];
 		for (i = 0; i < num_indices; i++) {
-			amdgpu_atombios_lookup_i2c_gpio_quirks(adev, gpio, i);
-
 			i2c = amdgpu_atombios_get_bus_rec_for_i2c_gpio(gpio);
 
 			if (i2c.valid) {
 				sprintf(stmp, "0x%x", i2c.i2c_id);
-				adev->i2c_bus[i] = amdgpu_i2c_create(adev->ddev, &i2c, stmp);
+				adev->i2c_bus[i] = amdgpu_i2c_create(adev_to_drm(adev), &i2c, stmp);
+			}
+			gpio = (ATOM_GPIO_I2C_ASSIGMENT *)
+				((u8 *)gpio + sizeof(ATOM_GPIO_I2C_ASSIGMENT));
+		}
+	}
+}
+
+void amdgpu_atombios_oem_i2c_init(struct amdgpu_device *adev, u8 i2c_id)
+{
+	struct atom_context *ctx = adev->mode_info.atom_context;
+	ATOM_GPIO_I2C_ASSIGMENT *gpio;
+	struct amdgpu_i2c_bus_rec i2c;
+	int index = GetIndexIntoMasterTable(DATA, GPIO_I2C_Info);
+	struct _ATOM_GPIO_I2C_INFO *i2c_info;
+	uint16_t data_offset, size;
+	int i, num_indices;
+	char stmp[32];
+
+	if (amdgpu_atom_parse_data_header(ctx, index, &size, NULL, NULL, &data_offset)) {
+		i2c_info = (struct _ATOM_GPIO_I2C_INFO *)(ctx->bios + data_offset);
+
+		num_indices = (size - sizeof(ATOM_COMMON_TABLE_HEADER)) /
+			sizeof(ATOM_GPIO_I2C_ASSIGMENT);
+
+		gpio = &i2c_info->asGPIO_Info[0];
+		for (i = 0; i < num_indices; i++) {
+			i2c = amdgpu_atombios_get_bus_rec_for_i2c_gpio(gpio);
+
+			if (i2c.valid && i2c.i2c_id == i2c_id) {
+				sprintf(stmp, "OEM 0x%x", i2c.i2c_id);
+				adev->i2c_bus[i] = amdgpu_i2c_create(adev_to_drm(adev), &i2c, stmp);
+				break;
 			}
 			gpio = (ATOM_GPIO_I2C_ASSIGMENT *)
 				((u8 *)gpio + sizeof(ATOM_GPIO_I2C_ASSIGMENT));
@@ -336,17 +358,9 @@ bool amdgpu_atombios_get_connector_info_from_object_table(struct amdgpu_device *
 		path_size += le16_to_cpu(path->usSize);
 
 		if (device_support & le16_to_cpu(path->usDeviceTag)) {
-			uint8_t con_obj_id, con_obj_num, con_obj_type;
-
-			con_obj_id =
+			uint8_t con_obj_id =
 			    (le16_to_cpu(path->usConnObjectId) & OBJECT_ID_MASK)
 			    >> OBJECT_ID_SHIFT;
-			con_obj_num =
-			    (le16_to_cpu(path->usConnObjectId) & ENUM_ID_MASK)
-			    >> ENUM_ID_SHIFT;
-			con_obj_type =
-			    (le16_to_cpu(path->usConnObjectId) &
-			     OBJECT_TYPE_MASK) >> OBJECT_TYPE_SHIFT;
 
 			/* Skip TV/CV support */
 			if ((le16_to_cpu(path->usDeviceTag) ==
@@ -371,15 +385,7 @@ bool amdgpu_atombios_get_connector_info_from_object_table(struct amdgpu_device *
 			router.ddc_valid = false;
 			router.cd_valid = false;
 			for (j = 0; j < ((le16_to_cpu(path->usSize) - 8) / 2); j++) {
-				uint8_t grph_obj_id, grph_obj_num, grph_obj_type;
-
-				grph_obj_id =
-				    (le16_to_cpu(path->usGraphicObjIds[j]) &
-				     OBJECT_ID_MASK) >> OBJECT_ID_SHIFT;
-				grph_obj_num =
-				    (le16_to_cpu(path->usGraphicObjIds[j]) &
-				     ENUM_ID_MASK) >> ENUM_ID_SHIFT;
-				grph_obj_type =
+				uint8_t grph_obj_type =
 				    (le16_to_cpu(path->usGraphicObjIds[j]) &
 				     OBJECT_TYPE_MASK) >> OBJECT_TYPE_SHIFT;
 
@@ -555,7 +561,7 @@ bool amdgpu_atombios_get_connector_info_from_object_table(struct amdgpu_device *
 		}
 	}
 
-	amdgpu_link_encoder_connector(adev->ddev);
+	amdgpu_link_encoder_connector(adev_to_drm(adev));
 
 	return true;
 }
@@ -690,13 +696,16 @@ int amdgpu_atombios_get_clock_info(struct amdgpu_device *adev)
 			le32_to_cpu(firmware_info->info_21.ulDefaultDispEngineClkFreq);
 		/* set a reasonable default for DP */
 		if (adev->clock.default_dispclk < 53900) {
-			DRM_INFO("Changing default dispclk from %dMhz to 600Mhz\n",
-				 adev->clock.default_dispclk / 100);
+			DRM_DEBUG("Changing default dispclk from %dMhz to 600Mhz\n",
+				  adev->clock.default_dispclk / 100);
 			adev->clock.default_dispclk = 60000;
+		} else if (adev->clock.default_dispclk <= 60000) {
+			DRM_DEBUG("Changing default dispclk from %dMhz to 625Mhz\n",
+				  adev->clock.default_dispclk / 100);
+			adev->clock.default_dispclk = 62500;
 		}
 		adev->clock.dp_extclk =
 			le16_to_cpu(firmware_info->info_21.usUniphyDPModeExtClkFreq);
-		adev->clock.current_dispclk = adev->clock.default_dispclk;
 
 		adev->clock.max_pixel_clock = le16_to_cpu(firmware_info->info.usMaxPixelClock);
 		if (adev->clock.max_pixel_clock == 0)
@@ -1028,7 +1037,9 @@ int amdgpu_atombios_get_clock_dividers(struct amdgpu_device *adev,
 		if (clock_type == COMPUTE_ENGINE_PLL_PARAM) {
 			args.v3.ulClockParams = cpu_to_le32((clock_type << 24) | clock);
 
-			amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+			if (amdgpu_atom_execute_table(adev->mode_info.atom_context,
+			    index, (uint32_t *)&args, sizeof(args)))
+				return -EINVAL;
 
 			dividers->post_div = args.v3.ucPostDiv;
 			dividers->enable_post_div = (args.v3.ucCntlFlag &
@@ -1048,7 +1059,9 @@ int amdgpu_atombios_get_clock_dividers(struct amdgpu_device *adev,
 			if (strobe_mode)
 				args.v5.ucInputFlag = ATOM_PLL_INPUT_FLAG_PLL_STROBE_MODE_EN;
 
-			amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+			if (amdgpu_atom_execute_table(adev->mode_info.atom_context,
+			    index, (uint32_t *)&args, sizeof(args)))
+				return -EINVAL;
 
 			dividers->post_div = args.v5.ucPostDiv;
 			dividers->enable_post_div = (args.v5.ucCntlFlag &
@@ -1066,7 +1079,9 @@ int amdgpu_atombios_get_clock_dividers(struct amdgpu_device *adev,
 		/* fusion */
 		args.v4.ulClock = cpu_to_le32(clock);	/* 10 khz */
 
-		amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+		if (amdgpu_atom_execute_table(adev->mode_info.atom_context,
+		    index, (uint32_t *)&args, sizeof(args)))
+			return -EINVAL;
 
 		dividers->post_divider = dividers->post_div = args.v4.ucPostDiv;
 		dividers->real_clock = le32_to_cpu(args.v4.ulClock);
@@ -1077,7 +1092,9 @@ int amdgpu_atombios_get_clock_dividers(struct amdgpu_device *adev,
 		args.v6_in.ulClock.ulComputeClockFlag = clock_type;
 		args.v6_in.ulClock.ulClockFreq = cpu_to_le32(clock);	/* 10 khz */
 
-		amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+		if (amdgpu_atom_execute_table(adev->mode_info.atom_context,
+		    index, (uint32_t *)&args, sizeof(args)))
+			return -EINVAL;
 
 		dividers->whole_fb_div = le16_to_cpu(args.v6_out.ulFbDiv.usFbDiv);
 		dividers->frac_fb_div = le16_to_cpu(args.v6_out.ulFbDiv.usFbDivFrac);
@@ -1093,6 +1110,7 @@ int amdgpu_atombios_get_clock_dividers(struct amdgpu_device *adev,
 	return 0;
 }
 
+#ifdef CONFIG_DRM_AMDGPU_SI
 int amdgpu_atombios_get_memory_pll_dividers(struct amdgpu_device *adev,
 					    u32 clock,
 					    bool strobe_mode,
@@ -1118,7 +1136,9 @@ int amdgpu_atombios_get_memory_pll_dividers(struct amdgpu_device *adev,
 			if (strobe_mode)
 				args.ucInputFlag |= MPLL_INPUT_FLAG_STROBE_MODE_EN;
 
-			amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+			if (amdgpu_atom_execute_table(adev->mode_info.atom_context,
+			    index, (uint32_t *)&args, sizeof(args)))
+				return -EINVAL;
 
 			mpll_param->clkfrac = le16_to_cpu(args.ulFbDiv.usFbDivFrac);
 			mpll_param->clkf = le16_to_cpu(args.ulFbDiv.usFbDiv);
@@ -1144,8 +1164,8 @@ int amdgpu_atombios_get_memory_pll_dividers(struct amdgpu_device *adev,
 	return 0;
 }
 
-void amdgpu_atombios_set_engine_dram_timings(struct amdgpu_device *adev,
-					     u32 eng_clock, u32 mem_clock)
+int amdgpu_atombios_set_engine_dram_timings(struct amdgpu_device *adev,
+					    u32 eng_clock, u32 mem_clock)
 {
 	SET_ENGINE_CLOCK_PS_ALLOCATION args;
 	int index = GetIndexIntoMasterTable(COMMAND, DynamicMemorySettings);
@@ -1160,7 +1180,8 @@ void amdgpu_atombios_set_engine_dram_timings(struct amdgpu_device *adev,
 	if (mem_clock)
 		args.sReserved.ulClock = cpu_to_le32(mem_clock & SET_CLOCK_FREQ_MASK);
 
-	amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+	return amdgpu_atom_execute_table(adev->mode_info.atom_context, index,
+					 (uint32_t *)&args, sizeof(args));
 }
 
 void amdgpu_atombios_get_default_voltages(struct amdgpu_device *adev,
@@ -1214,7 +1235,9 @@ int amdgpu_atombios_get_max_vddc(struct amdgpu_device *adev, u8 voltage_type,
 		args.v2.ucVoltageMode = 0;
 		args.v2.usVoltageLevel = 0;
 
-		amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+		if (amdgpu_atom_execute_table(adev->mode_info.atom_context,
+		    index, (uint32_t *)&args, sizeof(args)))
+			return -EINVAL;
 
 		*voltage = le16_to_cpu(args.v2.usVoltageLevel);
 		break;
@@ -1223,7 +1246,9 @@ int amdgpu_atombios_get_max_vddc(struct amdgpu_device *adev, u8 voltage_type,
 		args.v3.ucVoltageMode = ATOM_GET_VOLTAGE_LEVEL;
 		args.v3.usVoltageLevel = cpu_to_le16(voltage_id);
 
-		amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
+		if (amdgpu_atom_execute_table(adev->mode_info.atom_context,
+		    index, (uint32_t *)&args, sizeof(args)))
+			return -EINVAL;
 
 		*voltage = le16_to_cpu(args.v3.usVoltageLevel);
 		break;
@@ -1242,157 +1267,6 @@ int amdgpu_atombios_get_leakage_vddc_based_on_leakage_idx(struct amdgpu_device *
 	return amdgpu_atombios_get_max_vddc(adev, VOLTAGE_TYPE_VDDC, leakage_idx, voltage);
 }
 
-int amdgpu_atombios_get_leakage_id_from_vbios(struct amdgpu_device *adev,
-					      u16 *leakage_id)
-{
-	union set_voltage args;
-	int index = GetIndexIntoMasterTable(COMMAND, SetVoltage);
-	u8 frev, crev;
-
-	if (!amdgpu_atom_parse_cmd_header(adev->mode_info.atom_context, index, &frev, &crev))
-		return -EINVAL;
-
-	switch (crev) {
-	case 3:
-	case 4:
-		args.v3.ucVoltageType = 0;
-		args.v3.ucVoltageMode = ATOM_GET_LEAKAGE_ID;
-		args.v3.usVoltageLevel = 0;
-
-		amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
-
-		*leakage_id = le16_to_cpu(args.v3.usVoltageLevel);
-		break;
-	default:
-		DRM_ERROR("Unknown table version %d, %d\n", frev, crev);
-		return -EINVAL;
-	}
-
-	return 0;
-}
-
-int amdgpu_atombios_get_leakage_vddc_based_on_leakage_params(struct amdgpu_device *adev,
-							     u16 *vddc, u16 *vddci,
-							     u16 virtual_voltage_id,
-							     u16 vbios_voltage_id)
-{
-	int index = GetIndexIntoMasterTable(DATA, ASIC_ProfilingInfo);
-	u8 frev, crev;
-	u16 data_offset, size;
-	int i, j;
-	ATOM_ASIC_PROFILING_INFO_V2_1 *profile;
-	u16 *leakage_bin, *vddc_id_buf, *vddc_buf, *vddci_id_buf, *vddci_buf;
-
-	*vddc = 0;
-	*vddci = 0;
-
-	if (!amdgpu_atom_parse_data_header(adev->mode_info.atom_context, index, &size,
-				    &frev, &crev, &data_offset))
-		return -EINVAL;
-
-	profile = (ATOM_ASIC_PROFILING_INFO_V2_1 *)
-		(adev->mode_info.atom_context->bios + data_offset);
-
-	switch (frev) {
-	case 1:
-		return -EINVAL;
-	case 2:
-		switch (crev) {
-		case 1:
-			if (size < sizeof(ATOM_ASIC_PROFILING_INFO_V2_1))
-				return -EINVAL;
-			leakage_bin = (u16 *)
-				(adev->mode_info.atom_context->bios + data_offset +
-				 le16_to_cpu(profile->usLeakageBinArrayOffset));
-			vddc_id_buf = (u16 *)
-				(adev->mode_info.atom_context->bios + data_offset +
-				 le16_to_cpu(profile->usElbVDDC_IdArrayOffset));
-			vddc_buf = (u16 *)
-				(adev->mode_info.atom_context->bios + data_offset +
-				 le16_to_cpu(profile->usElbVDDC_LevelArrayOffset));
-			vddci_id_buf = (u16 *)
-				(adev->mode_info.atom_context->bios + data_offset +
-				 le16_to_cpu(profile->usElbVDDCI_IdArrayOffset));
-			vddci_buf = (u16 *)
-				(adev->mode_info.atom_context->bios + data_offset +
-				 le16_to_cpu(profile->usElbVDDCI_LevelArrayOffset));
-
-			if (profile->ucElbVDDC_Num > 0) {
-				for (i = 0; i < profile->ucElbVDDC_Num; i++) {
-					if (vddc_id_buf[i] == virtual_voltage_id) {
-						for (j = 0; j < profile->ucLeakageBinNum; j++) {
-							if (vbios_voltage_id <= leakage_bin[j]) {
-								*vddc = vddc_buf[j * profile->ucElbVDDC_Num + i];
-								break;
-							}
-						}
-						break;
-					}
-				}
-			}
-			if (profile->ucElbVDDCI_Num > 0) {
-				for (i = 0; i < profile->ucElbVDDCI_Num; i++) {
-					if (vddci_id_buf[i] == virtual_voltage_id) {
-						for (j = 0; j < profile->ucLeakageBinNum; j++) {
-							if (vbios_voltage_id <= leakage_bin[j]) {
-								*vddci = vddci_buf[j * profile->ucElbVDDCI_Num + i];
-								break;
-							}
-						}
-						break;
-					}
-				}
-			}
-			break;
-		default:
-			DRM_ERROR("Unknown table version %d, %d\n", frev, crev);
-			return -EINVAL;
-		}
-		break;
-	default:
-		DRM_ERROR("Unknown table version %d, %d\n", frev, crev);
-		return -EINVAL;
-	}
-
-	return 0;
-}
-
-union get_voltage_info {
-	struct _GET_VOLTAGE_INFO_INPUT_PARAMETER_V1_2 in;
-	struct _GET_EVV_VOLTAGE_INFO_OUTPUT_PARAMETER_V1_2 evv_out;
-};
-
-int amdgpu_atombios_get_voltage_evv(struct amdgpu_device *adev,
-				    u16 virtual_voltage_id,
-				    u16 *voltage)
-{
-	int index = GetIndexIntoMasterTable(COMMAND, GetVoltageInfo);
-	u32 entry_id;
-	u32 count = adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.count;
-	union get_voltage_info args;
-
-	for (entry_id = 0; entry_id < count; entry_id++) {
-		if (adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[entry_id].v ==
-		    virtual_voltage_id)
-			break;
-	}
-
-	if (entry_id >= count)
-		return -EINVAL;
-
-	args.in.ucVoltageType = VOLTAGE_TYPE_VDDC;
-	args.in.ucVoltageMode = ATOM_GET_VOLTAGE_EVV_VOLTAGE;
-	args.in.usVoltageLevel = cpu_to_le16(virtual_voltage_id);
-	args.in.ulSCLKFreq =
-		cpu_to_le32(adev->pm.dpm.dyn_state.vddc_dependency_on_sclk.entries[entry_id].clk);
-
-	amdgpu_atom_execute_table(adev->mode_info.atom_context, index, (uint32_t *)&args);
-
-	*voltage = le16_to_cpu(args.evv_out.usVoltageLevel);
-
-	return 0;
-}
-
 union voltage_object_info {
 	struct _ATOM_VOLTAGE_OBJECT_INFO v1;
 	struct _ATOM_VOLTAGE_OBJECT_INFO_V2 v2;
@@ -1411,7 +1285,7 @@ static ATOM_VOLTAGE_OBJECT_V3 *amdgpu_atombios_lookup_voltage_object_v3(ATOM_VOL
 {
 	u32 size = le16_to_cpu(v3->sHeader.usStructureSize);
 	u32 offset = offsetof(ATOM_VOLTAGE_OBJECT_INFO_V3_1, asVoltageObj[0]);
-	u8 *start = (u8*)v3;
+	u8 *start = (u8 *)v3;
 
 	while (offset < size) {
 		ATOM_VOLTAGE_OBJECT_V3 *vo = (ATOM_VOLTAGE_OBJECT_V3 *)(start + offset);
@@ -1636,6 +1510,8 @@ int amdgpu_atombios_init_mc_reg_table(struct amdgpu_device *adev,
 										(u32)le32_to_cpu(*((u32 *)reg_data + j));
 									j++;
 								} else if ((reg_table->mc_reg_address[i].pre_reg_data & LOW_NIBBLE_MASK) == DATA_EQU_PREV) {
+									if (i == 0)
+										continue;
 									reg_table->mc_reg_table_entry[num_ranges].mc_data[i] =
 										reg_table->mc_reg_table_entry[num_ranges].mc_data[i - 1];
 								}
@@ -1664,6 +1540,7 @@ int amdgpu_atombios_init_mc_reg_table(struct amdgpu_device *adev,
 	}
 	return -EINVAL;
 }
+#endif
 
 bool amdgpu_atombios_has_gpu_virtualization_table(struct amdgpu_device *adev)
 {
@@ -1682,7 +1559,7 @@ void amdgpu_atombios_scratch_regs_lock(struct amdgpu_device *adev, bool lock)
 {
 	uint32_t bios_6_scratch;
 
-	bios_6_scratch = RREG32(mmBIOS_SCRATCH_6);
+	bios_6_scratch = RREG32(adev->bios_scratch_reg_offset + 6);
 
 	if (lock) {
 		bios_6_scratch |= ATOM_S6_CRITICAL_STATE;
@@ -1692,15 +1569,17 @@ void amdgpu_atombios_scratch_regs_lock(struct amdgpu_device *adev, bool lock)
 		bios_6_scratch |= ATOM_S6_ACC_MODE;
 	}
 
-	WREG32(mmBIOS_SCRATCH_6, bios_6_scratch);
+	WREG32(adev->bios_scratch_reg_offset + 6, bios_6_scratch);
 }
 
-void amdgpu_atombios_scratch_regs_init(struct amdgpu_device *adev)
+static void amdgpu_atombios_scratch_regs_init(struct amdgpu_device *adev)
 {
 	uint32_t bios_2_scratch, bios_6_scratch;
 
-	bios_2_scratch = RREG32(mmBIOS_SCRATCH_2);
-	bios_6_scratch = RREG32(mmBIOS_SCRATCH_6);
+	adev->bios_scratch_reg_offset = mmBIOS_SCRATCH_0;
+
+	bios_2_scratch = RREG32(adev->bios_scratch_reg_offset + 2);
+	bios_6_scratch = RREG32(adev->bios_scratch_reg_offset + 6);
 
 	/* let the bios control the backlight */
 	bios_2_scratch &= ~ATOM_S2_VRI_BRIGHT_ENABLE;
@@ -1711,86 +1590,86 @@ void amdgpu_atombios_scratch_regs_init(struct amdgpu_device *adev)
 	/* clear the vbios dpms state */
 	bios_2_scratch &= ~ATOM_S2_DEVICE_DPMS_STATE;
 
-	WREG32(mmBIOS_SCRATCH_2, bios_2_scratch);
-	WREG32(mmBIOS_SCRATCH_6, bios_6_scratch);
+	WREG32(adev->bios_scratch_reg_offset + 2, bios_2_scratch);
+	WREG32(adev->bios_scratch_reg_offset + 6, bios_6_scratch);
 }
 
-void amdgpu_atombios_scratch_regs_save(struct amdgpu_device *adev)
+void amdgpu_atombios_scratch_regs_engine_hung(struct amdgpu_device *adev,
+					      bool hung)
 {
-	int i;
+	u32 tmp = RREG32(adev->bios_scratch_reg_offset + 3);
+
+	if (hung)
+		tmp |= ATOM_S3_ASIC_GUI_ENGINE_HUNG;
+	else
+		tmp &= ~ATOM_S3_ASIC_GUI_ENGINE_HUNG;
 
-	for (i = 0; i < AMDGPU_BIOS_NUM_SCRATCH; i++)
-		adev->bios_scratch[i] = RREG32(mmBIOS_SCRATCH_0 + i);
+	WREG32(adev->bios_scratch_reg_offset + 3, tmp);
 }
 
-void amdgpu_atombios_scratch_regs_restore(struct amdgpu_device *adev)
+void amdgpu_atombios_scratch_regs_set_backlight_level(struct amdgpu_device *adev,
+						      u32 backlight_level)
 {
-	int i;
+	u32 tmp = RREG32(adev->bios_scratch_reg_offset + 2);
 
-	/*
-	 * VBIOS will check ASIC_INIT_COMPLETE bit to decide if
-	 * execute ASIC_Init posting via driver
-	 */
-	adev->bios_scratch[7] &= ~ATOM_S7_ASIC_INIT_COMPLETE_MASK;
+	tmp &= ~ATOM_S2_CURRENT_BL_LEVEL_MASK;
+	tmp |= (backlight_level << ATOM_S2_CURRENT_BL_LEVEL_SHIFT) &
+		ATOM_S2_CURRENT_BL_LEVEL_MASK;
 
-	for (i = 0; i < AMDGPU_BIOS_NUM_SCRATCH; i++)
-		WREG32(mmBIOS_SCRATCH_0 + i, adev->bios_scratch[i]);
+	WREG32(adev->bios_scratch_reg_offset + 2, tmp);
 }
 
-void amdgpu_atombios_scratch_regs_engine_hung(struct amdgpu_device *adev,
-					      bool hung)
+bool amdgpu_atombios_scratch_need_asic_init(struct amdgpu_device *adev)
 {
-	u32 tmp = RREG32(mmBIOS_SCRATCH_3);
+	u32 tmp = RREG32(adev->bios_scratch_reg_offset + 7);
 
-	if (hung)
-		tmp |= ATOM_S3_ASIC_GUI_ENGINE_HUNG;
+	if (tmp & ATOM_S7_ASIC_INIT_COMPLETE_MASK)
+		return false;
 	else
-		tmp &= ~ATOM_S3_ASIC_GUI_ENGINE_HUNG;
-
-	WREG32(mmBIOS_SCRATCH_3, tmp);
+		return true;
 }
 
-/* Atom needs data in little endian format
- * so swap as appropriate when copying data to
- * or from atom. Note that atom operates on
- * dw units.
+/* Atom needs data in little endian format so swap as appropriate when copying
+ * data to or from atom. Note that atom operates on dw units.
+ *
+ * Use to_le=true when sending data to atom and provide at least
+ * ALIGN(num_bytes,4) bytes in the dst buffer.
+ *
+ * Use to_le=false when receiving data from atom and provide ALIGN(num_bytes,4)
+ * byes in the src buffer.
  */
 void amdgpu_atombios_copy_swap(u8 *dst, u8 *src, u8 num_bytes, bool to_le)
 {
 #ifdef __BIG_ENDIAN
-	u8 src_tmp[20], dst_tmp[20]; /* used for byteswapping */
-	u32 *dst32, *src32;
+	u32 src_tmp[5], dst_tmp[5];
 	int i;
+	u8 align_num_bytes = ALIGN(num_bytes, 4);
 
-	memcpy(src_tmp, src, num_bytes);
-	src32 = (u32 *)src_tmp;
-	dst32 = (u32 *)dst_tmp;
 	if (to_le) {
-		for (i = 0; i < ((num_bytes + 3) / 4); i++)
-			dst32[i] = cpu_to_le32(src32[i]);
-		memcpy(dst, dst_tmp, num_bytes);
+		memcpy(src_tmp, src, num_bytes);
+		for (i = 0; i < align_num_bytes / 4; i++)
+			dst_tmp[i] = cpu_to_le32(src_tmp[i]);
+		memcpy(dst, dst_tmp, align_num_bytes);
 	} else {
-		u8 dws = num_bytes & ~3;
-		for (i = 0; i < ((num_bytes + 3) / 4); i++)
-			dst32[i] = le32_to_cpu(src32[i]);
-		memcpy(dst, dst_tmp, dws);
-		if (num_bytes % 4) {
-			for (i = 0; i < (num_bytes % 4); i++)
-				dst[dws+i] = dst_tmp[dws+i];
-		}
+		memcpy(src_tmp, src, align_num_bytes);
+		for (i = 0; i < align_num_bytes / 4; i++)
+			dst_tmp[i] = le32_to_cpu(src_tmp[i]);
+		memcpy(dst, dst_tmp, num_bytes);
 	}
 #else
 	memcpy(dst, src, num_bytes);
 #endif
 }
 
-int amdgpu_atombios_allocate_fb_scratch(struct amdgpu_device *adev)
+static int amdgpu_atombios_allocate_fb_scratch(struct amdgpu_device *adev)
 {
 	struct atom_context *ctx = adev->mode_info.atom_context;
 	int index = GetIndexIntoMasterTable(DATA, VRAM_UsageByFirmware);
 	uint16_t data_offset;
 	int usage_bytes = 0;
 	struct _ATOM_VRAM_USAGE_BY_FIRMWARE *firmware_usage;
+	u64 start_addr;
+	u64 size;
 
 	if (amdgpu_atom_parse_data_header(ctx, index, NULL, NULL, NULL, &data_offset)) {
 		firmware_usage = (struct _ATOM_VRAM_USAGE_BY_FIRMWARE *)(ctx->bios + data_offset);
@@ -1799,7 +1678,21 @@ int amdgpu_atombios_allocate_fb_scratch(struct amdgpu_device *adev)
 			  le32_to_cpu(firmware_usage->asFirmwareVramReserveInfo[0].ulStartAddrUsedByFirmware),
 			  le16_to_cpu(firmware_usage->asFirmwareVramReserveInfo[0].usFirmwareUseInKb));
 
-		usage_bytes = le16_to_cpu(firmware_usage->asFirmwareVramReserveInfo[0].usFirmwareUseInKb) * 1024;
+		start_addr = firmware_usage->asFirmwareVramReserveInfo[0].ulStartAddrUsedByFirmware;
+		size = firmware_usage->asFirmwareVramReserveInfo[0].usFirmwareUseInKb;
+
+		if ((uint32_t)(start_addr & ATOM_VRAM_OPERATION_FLAGS_MASK) ==
+			(uint32_t)(ATOM_VRAM_BLOCK_SRIOV_MSG_SHARE_RESERVATION <<
+			ATOM_VRAM_OPERATION_FLAGS_SHIFT)) {
+			/* Firmware request VRAM reservation for SR-IOV */
+			adev->mman.fw_vram_usage_start_offset = (start_addr &
+				(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
+			adev->mman.fw_vram_usage_size = size << 10;
+			/* Use the default scratch size */
+			usage_bytes = 0;
+		} else {
+			usage_bytes = le16_to_cpu(firmware_usage->asFirmwareVramReserveInfo[0].usFirmwareUseInKb) * 1024;
+		}
 	}
 	ctx->scratch_size_bytes = 0;
 	if (usage_bytes == 0)
@@ -1811,3 +1704,248 @@ int amdgpu_atombios_allocate_fb_scratch(struct amdgpu_device *adev)
 	ctx->scratch_size_bytes = usage_bytes;
 	return 0;
 }
+
+/* ATOM accessor methods */
+/*
+ * ATOM is an interpreted byte code stored in tables in the vbios.  The
+ * driver registers callbacks to access registers and the interpreter
+ * in the driver parses the tables and executes then to program specific
+ * actions (set display modes, asic init, etc.).  See amdgpu_atombios.c,
+ * atombios.h, and atom.c
+ */
+
+/**
+ * cail_pll_read - read PLL register
+ *
+ * @info: atom card_info pointer
+ * @reg: PLL register offset
+ *
+ * Provides a PLL register accessor for the atom interpreter (r4xx+).
+ * Returns the value of the PLL register.
+ */
+static uint32_t cail_pll_read(struct card_info *info, uint32_t reg)
+{
+	return 0;
+}
+
+/**
+ * cail_pll_write - write PLL register
+ *
+ * @info: atom card_info pointer
+ * @reg: PLL register offset
+ * @val: value to write to the pll register
+ *
+ * Provides a PLL register accessor for the atom interpreter (r4xx+).
+ */
+static void cail_pll_write(struct card_info *info, uint32_t reg, uint32_t val)
+{
+
+}
+
+/**
+ * cail_mc_read - read MC (Memory Controller) register
+ *
+ * @info: atom card_info pointer
+ * @reg: MC register offset
+ *
+ * Provides an MC register accessor for the atom interpreter (r4xx+).
+ * Returns the value of the MC register.
+ */
+static uint32_t cail_mc_read(struct card_info *info, uint32_t reg)
+{
+	return 0;
+}
+
+/**
+ * cail_mc_write - write MC (Memory Controller) register
+ *
+ * @info: atom card_info pointer
+ * @reg: MC register offset
+ * @val: value to write to the pll register
+ *
+ * Provides a MC register accessor for the atom interpreter (r4xx+).
+ */
+static void cail_mc_write(struct card_info *info, uint32_t reg, uint32_t val)
+{
+
+}
+
+/**
+ * cail_reg_write - write MMIO register
+ *
+ * @info: atom card_info pointer
+ * @reg: MMIO register offset
+ * @val: value to write to the pll register
+ *
+ * Provides a MMIO register accessor for the atom interpreter (r4xx+).
+ */
+static void cail_reg_write(struct card_info *info, uint32_t reg, uint32_t val)
+{
+	struct amdgpu_device *adev = drm_to_adev(info->dev);
+
+	WREG32(reg, val);
+}
+
+/**
+ * cail_reg_read - read MMIO register
+ *
+ * @info: atom card_info pointer
+ * @reg: MMIO register offset
+ *
+ * Provides an MMIO register accessor for the atom interpreter (r4xx+).
+ * Returns the value of the MMIO register.
+ */
+static uint32_t cail_reg_read(struct card_info *info, uint32_t reg)
+{
+	struct amdgpu_device *adev = drm_to_adev(info->dev);
+	uint32_t r;
+
+	r = RREG32(reg);
+	return r;
+}
+
+static ssize_t amdgpu_atombios_get_vbios_version(struct device *dev,
+						 struct device_attribute *attr,
+						 char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct atom_context *ctx = adev->mode_info.atom_context;
+
+	return sysfs_emit(buf, "%s\n", ctx->vbios_pn);
+}
+
+static ssize_t amdgpu_atombios_get_vbios_build(struct device *dev,
+					       struct device_attribute *attr,
+					       char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct atom_context *ctx = adev->mode_info.atom_context;
+
+	return sysfs_emit(buf, "%s\n", ctx->build_num);
+}
+
+static DEVICE_ATTR(vbios_version, 0444, amdgpu_atombios_get_vbios_version,
+		   NULL);
+static DEVICE_ATTR(vbios_build, 0444, amdgpu_atombios_get_vbios_build, NULL);
+
+static struct attribute *amdgpu_vbios_version_attrs[] = {
+	&dev_attr_vbios_version.attr, &dev_attr_vbios_build.attr, NULL
+};
+
+static umode_t amdgpu_vbios_version_attrs_is_visible(struct kobject *kobj,
+						     struct attribute *attr,
+						     int index)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct atom_context *ctx = adev->mode_info.atom_context;
+
+	if (attr == &dev_attr_vbios_build.attr && !strlen(ctx->build_num))
+		return 0;
+
+	return attr->mode;
+}
+
+const struct attribute_group amdgpu_vbios_version_attr_group = {
+	.attrs = amdgpu_vbios_version_attrs,
+	.is_visible = amdgpu_vbios_version_attrs_is_visible,
+};
+
+int amdgpu_atombios_sysfs_init(struct amdgpu_device *adev)
+{
+	if (adev->mode_info.atom_context)
+		return devm_device_add_group(adev->dev,
+					     &amdgpu_vbios_version_attr_group);
+
+	return 0;
+}
+
+/**
+ * amdgpu_atombios_fini - free the driver info and callbacks for atombios
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Frees the driver info and register access callbacks for the ATOM
+ * interpreter (r4xx+).
+ * Called at driver shutdown.
+ */
+void amdgpu_atombios_fini(struct amdgpu_device *adev)
+{
+	if (adev->mode_info.atom_context) {
+		kfree(adev->mode_info.atom_context->scratch);
+		kfree(adev->mode_info.atom_context->iio);
+	}
+	kfree(adev->mode_info.atom_context);
+	adev->mode_info.atom_context = NULL;
+	kfree(adev->mode_info.atom_card_info);
+	adev->mode_info.atom_card_info = NULL;
+}
+
+/**
+ * amdgpu_atombios_init - init the driver info and callbacks for atombios
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Initializes the driver info and register access callbacks for the
+ * ATOM interpreter (r4xx+).
+ * Returns 0 on sucess, -ENOMEM on failure.
+ * Called at driver startup.
+ */
+int amdgpu_atombios_init(struct amdgpu_device *adev)
+{
+	struct card_info *atom_card_info =
+	    kzalloc(sizeof(struct card_info), GFP_KERNEL);
+
+	if (!atom_card_info)
+		return -ENOMEM;
+
+	adev->mode_info.atom_card_info = atom_card_info;
+	atom_card_info->dev = adev_to_drm(adev);
+	atom_card_info->reg_read = cail_reg_read;
+	atom_card_info->reg_write = cail_reg_write;
+	atom_card_info->mc_read = cail_mc_read;
+	atom_card_info->mc_write = cail_mc_write;
+	atom_card_info->pll_read = cail_pll_read;
+	atom_card_info->pll_write = cail_pll_write;
+
+	adev->mode_info.atom_context = amdgpu_atom_parse(atom_card_info, adev->bios);
+	if (!adev->mode_info.atom_context) {
+		amdgpu_atombios_fini(adev);
+		return -ENOMEM;
+	}
+
+	mutex_init(&adev->mode_info.atom_context->mutex);
+	if (adev->is_atom_fw) {
+		amdgpu_atomfirmware_scratch_regs_init(adev);
+		amdgpu_atomfirmware_allocate_fb_scratch(adev);
+		/* cached firmware_flags for further usage */
+		adev->mode_info.firmware_flags =
+			amdgpu_atomfirmware_query_firmware_capability(adev);
+	} else {
+		amdgpu_atombios_scratch_regs_init(adev);
+		amdgpu_atombios_allocate_fb_scratch(adev);
+	}
+
+	return 0;
+}
+
+int amdgpu_atombios_get_data_table(struct amdgpu_device *adev,
+				   uint32_t table,
+				   uint16_t *size,
+				   uint8_t *frev,
+				   uint8_t *crev,
+				   uint8_t **addr)
+{
+	uint16_t data_start;
+
+	if (!amdgpu_atom_parse_data_header(adev->mode_info.atom_context, table,
+					   size, frev, crev, &data_start))
+		return -EINVAL;
+
+	*addr = (uint8_t *)adev->mode_info.atom_context->bios + data_start;
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h
index 38d0fe32e5cd..867bc5c5ce67 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h
@@ -89,8 +89,7 @@ struct atom_memory_info {
 
 #define MAX_AC_TIMING_ENTRIES 16
 
-struct atom_memory_clock_range_table
-{
+struct atom_memory_clock_range_table {
 	u8 num_entries;
 	u8 rsv[3];
 	u32 mclk[MAX_AC_TIMING_ENTRIES];
@@ -118,14 +117,12 @@ struct atom_mc_reg_table {
 
 #define MAX_VOLTAGE_ENTRIES 32
 
-struct atom_voltage_table_entry
-{
+struct atom_voltage_table_entry {
 	u16 value;
 	u32 smio_low;
 };
 
-struct atom_voltage_table
-{
+struct atom_voltage_table {
 	u32 count;
 	u32 mask_low;
 	u32 phase_delay;
@@ -139,6 +136,7 @@ amdgpu_atombios_lookup_gpio(struct amdgpu_device *adev,
 struct amdgpu_i2c_bus_rec amdgpu_atombios_lookup_i2c_gpio(struct amdgpu_device *adev,
 							  uint8_t id);
 void amdgpu_atombios_i2c_init(struct amdgpu_device *adev);
+void amdgpu_atombios_oem_i2c_init(struct amdgpu_device *adev, u8 i2c_id);
 
 bool amdgpu_atombios_has_dce_engine_info(struct amdgpu_device *adev);
 
@@ -160,25 +158,14 @@ int amdgpu_atombios_get_clock_dividers(struct amdgpu_device *adev,
 				       bool strobe_mode,
 				       struct atom_clock_dividers *dividers);
 
+#ifdef CONFIG_DRM_AMDGPU_SI
 int amdgpu_atombios_get_memory_pll_dividers(struct amdgpu_device *adev,
 					    u32 clock,
 					    bool strobe_mode,
 					    struct atom_mpll_param *mpll_param);
 
-void amdgpu_atombios_set_engine_dram_timings(struct amdgpu_device *adev,
-					     u32 eng_clock, u32 mem_clock);
-
-int amdgpu_atombios_get_leakage_id_from_vbios(struct amdgpu_device *adev,
-					      u16 *leakage_id);
-
-int amdgpu_atombios_get_leakage_vddc_based_on_leakage_params(struct amdgpu_device *adev,
-							     u16 *vddc, u16 *vddci,
-							     u16 virtual_voltage_id,
-							     u16 vbios_voltage_id);
-
-int amdgpu_atombios_get_voltage_evv(struct amdgpu_device *adev,
-				    u16 virtual_voltage_id,
-				    u16 *voltage);
+int amdgpu_atombios_set_engine_dram_timings(struct amdgpu_device *adev,
+					    u32 eng_clock, u32 mem_clock);
 
 bool
 amdgpu_atombios_is_voltage_gpio(struct amdgpu_device *adev,
@@ -191,33 +178,43 @@ int amdgpu_atombios_get_voltage_table(struct amdgpu_device *adev,
 int amdgpu_atombios_init_mc_reg_table(struct amdgpu_device *adev,
 				      u8 module_index,
 				      struct atom_mc_reg_table *reg_table);
+int amdgpu_atombios_get_max_vddc(struct amdgpu_device *adev, u8 voltage_type,
+			     u16 voltage_id, u16 *voltage);
+int amdgpu_atombios_get_leakage_vddc_based_on_leakage_idx(struct amdgpu_device *adev,
+						      u16 *voltage,
+						      u16 leakage_idx);
+void amdgpu_atombios_get_default_voltages(struct amdgpu_device *adev,
+					  u16 *vddc, u16 *vddci, u16 *mvdd);
+int amdgpu_atombios_get_svi2_info(struct amdgpu_device *adev,
+			      u8 voltage_type,
+			      u8 *svd_gpio_id, u8 *svc_gpio_id);
+#endif
 
 bool amdgpu_atombios_has_gpu_virtualization_table(struct amdgpu_device *adev);
 
 void amdgpu_atombios_scratch_regs_lock(struct amdgpu_device *adev, bool lock);
-void amdgpu_atombios_scratch_regs_init(struct amdgpu_device *adev);
-void amdgpu_atombios_scratch_regs_save(struct amdgpu_device *adev);
-void amdgpu_atombios_scratch_regs_restore(struct amdgpu_device *adev);
 void amdgpu_atombios_scratch_regs_engine_hung(struct amdgpu_device *adev,
 					      bool hung);
+void amdgpu_atombios_scratch_regs_set_backlight_level(struct amdgpu_device *adev,
+						      u32 backlight_level);
+bool amdgpu_atombios_scratch_need_asic_init(struct amdgpu_device *adev);
 
 void amdgpu_atombios_copy_swap(u8 *dst, u8 *src, u8 num_bytes, bool to_le);
-int amdgpu_atombios_get_max_vddc(struct amdgpu_device *adev, u8 voltage_type,
-			     u16 voltage_id, u16 *voltage);
-int amdgpu_atombios_get_leakage_vddc_based_on_leakage_idx(struct amdgpu_device *adev,
-						      u16 *voltage,
-						      u16 leakage_idx);
-void amdgpu_atombios_get_default_voltages(struct amdgpu_device *adev,
-					  u16 *vddc, u16 *vddci, u16 *mvdd);
 int amdgpu_atombios_get_clock_dividers(struct amdgpu_device *adev,
 				       u8 clock_type,
 				       u32 clock,
 				       bool strobe_mode,
 				       struct atom_clock_dividers *dividers);
-int amdgpu_atombios_get_svi2_info(struct amdgpu_device *adev,
-			      u8 voltage_type,
-			      u8 *svd_gpio_id, u8 *svc_gpio_id);
 
-int amdgpu_atombios_allocate_fb_scratch(struct amdgpu_device *adev);
+int amdgpu_atombios_get_data_table(struct amdgpu_device *adev,
+				   uint32_t table,
+				   uint16_t *size,
+				   uint8_t *frev,
+				   uint8_t *crev,
+				   uint8_t **addr);
+
+void amdgpu_atombios_fini(struct amdgpu_device *adev);
+int amdgpu_atombios_init(struct amdgpu_device *adev);
+int amdgpu_atombios_sysfs_init(struct amdgpu_device *adev);
 
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c
index 4bdda56fccee..636385c80f64 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c
@@ -20,33 +20,69 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  *
  */
-#include <drm/drmP.h>
+
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 #include "atomfirmware.h"
 #include "amdgpu_atomfirmware.h"
 #include "atom.h"
 #include "atombios.h"
+#include "soc15_hw_ip.h"
 
-#define get_index_into_master_table(master_table, table_name) (offsetof(struct master_table, table_name) / sizeof(uint16_t))
+union firmware_info {
+	struct atom_firmware_info_v3_1 v31;
+	struct atom_firmware_info_v3_2 v32;
+	struct atom_firmware_info_v3_3 v33;
+	struct atom_firmware_info_v3_4 v34;
+	struct atom_firmware_info_v3_5 v35;
+};
 
-bool amdgpu_atomfirmware_gpu_supports_virtualization(struct amdgpu_device *adev)
+/*
+ * Helper function to query firmware capability
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Return firmware_capability in firmwareinfo table on success or 0 if not
+ */
+uint32_t amdgpu_atomfirmware_query_firmware_capability(struct amdgpu_device *adev)
 {
-	int index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
-						firmwareinfo);
-	uint16_t data_offset;
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	int index;
+	u16 data_offset, size;
+	union firmware_info *firmware_info;
+	u8 frev, crev;
+	u32 fw_cap = 0;
 
-	if (amdgpu_atom_parse_data_header(adev->mode_info.atom_context, index, NULL,
-					  NULL, NULL, &data_offset)) {
-		struct atom_firmware_info_v3_1 *firmware_info =
-			(struct atom_firmware_info_v3_1 *)(adev->mode_info.atom_context->bios +
-							   data_offset);
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+			firmwareinfo);
 
-		if (le32_to_cpu(firmware_info->firmware_capability) &
-		    ATOM_FIRMWARE_CAP_GPU_VIRTUALIZATION)
-			return true;
+	if (amdgpu_atom_parse_data_header(adev->mode_info.atom_context,
+				index, &size, &frev, &crev, &data_offset)) {
+		/* support firmware_info 3.1 + */
+		if ((frev == 3 && crev >= 1) || (frev > 3)) {
+			firmware_info = (union firmware_info *)
+				(mode_info->atom_context->bios + data_offset);
+			fw_cap = le32_to_cpu(firmware_info->v31.firmware_capability);
+		}
 	}
-	return false;
+
+	return fw_cap;
+}
+
+/*
+ * Helper function to query gpu virtualizaiton capability
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Return true if gpu virtualization is supported or false if not
+ */
+bool amdgpu_atomfirmware_gpu_virtualization_supported(struct amdgpu_device *adev)
+{
+	u32 fw_cap;
+
+	fw_cap = adev->mode_info.firmware_flags;
+
+	return (fw_cap & ATOM_FIRMWARE_CAP_GPU_VIRTUALIZATION) ? true : false;
 }
 
 void amdgpu_atomfirmware_scratch_regs_init(struct amdgpu_device *adev)
@@ -66,39 +102,72 @@ void amdgpu_atomfirmware_scratch_regs_init(struct amdgpu_device *adev)
 	}
 }
 
-void amdgpu_atomfirmware_scratch_regs_save(struct amdgpu_device *adev)
+static int amdgpu_atomfirmware_allocate_fb_v2_1(struct amdgpu_device *adev,
+	struct vram_usagebyfirmware_v2_1 *fw_usage, int *usage_bytes)
 {
-	int i;
+	u32 start_addr, fw_size, drv_size;
+
+	start_addr = le32_to_cpu(fw_usage->start_address_in_kb);
+	fw_size = le16_to_cpu(fw_usage->used_by_firmware_in_kb);
+	drv_size = le16_to_cpu(fw_usage->used_by_driver_in_kb);
+
+	DRM_DEBUG("atom firmware v2_1 requested %08x %dkb fw %dkb drv\n",
+			  start_addr,
+			  fw_size,
+			  drv_size);
 
-	for (i = 0; i < AMDGPU_BIOS_NUM_SCRATCH; i++)
-		adev->bios_scratch[i] = RREG32(adev->bios_scratch_reg_offset + i);
+	if ((start_addr & ATOM_VRAM_OPERATION_FLAGS_MASK) ==
+		(u32)(ATOM_VRAM_BLOCK_SRIOV_MSG_SHARE_RESERVATION <<
+		ATOM_VRAM_OPERATION_FLAGS_SHIFT)) {
+		/* Firmware request VRAM reservation for SR-IOV */
+		adev->mman.fw_vram_usage_start_offset = (start_addr &
+			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
+		adev->mman.fw_vram_usage_size = fw_size << 10;
+		/* Use the default scratch size */
+		*usage_bytes = 0;
+	} else {
+		*usage_bytes = drv_size << 10;
+	}
+	return 0;
 }
 
-void amdgpu_atomfirmware_scratch_regs_restore(struct amdgpu_device *adev)
+static int amdgpu_atomfirmware_allocate_fb_v2_2(struct amdgpu_device *adev,
+		struct vram_usagebyfirmware_v2_2 *fw_usage, int *usage_bytes)
 {
-	int i;
+	u32 fw_start_addr, fw_size, drv_start_addr, drv_size;
 
-	/*
-	 * VBIOS will check ASIC_INIT_COMPLETE bit to decide if
-	 * execute ASIC_Init posting via driver
-	 */
-	adev->bios_scratch[7] &= ~ATOM_S7_ASIC_INIT_COMPLETE_MASK;
+	fw_start_addr = le32_to_cpu(fw_usage->fw_region_start_address_in_kb);
+	fw_size = le16_to_cpu(fw_usage->used_by_firmware_in_kb);
 
-	for (i = 0; i < AMDGPU_BIOS_NUM_SCRATCH; i++)
-		WREG32(adev->bios_scratch_reg_offset + i, adev->bios_scratch[i]);
-}
+	drv_start_addr = le32_to_cpu(fw_usage->driver_region0_start_address_in_kb);
+	drv_size = le32_to_cpu(fw_usage->used_by_driver_region0_in_kb);
 
-void amdgpu_atomfirmware_scratch_regs_engine_hung(struct amdgpu_device *adev,
-						  bool hung)
-{
-	u32 tmp = RREG32(adev->bios_scratch_reg_offset + 3);
+	DRM_DEBUG("atom requested fw start at %08x %dkb and drv start at %08x %dkb\n",
+			  fw_start_addr,
+			  fw_size,
+			  drv_start_addr,
+			  drv_size);
+
+	if (amdgpu_sriov_vf(adev) &&
+	    ((fw_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION <<
+		ATOM_VRAM_OPERATION_FLAGS_SHIFT)) == 0)) {
+		/* Firmware request VRAM reservation for SR-IOV */
+		adev->mman.fw_vram_usage_start_offset = (fw_start_addr &
+			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
+		adev->mman.fw_vram_usage_size = fw_size << 10;
+	}
 
-	if (hung)
-		tmp |= ATOM_S3_ASIC_GUI_ENGINE_HUNG;
-	else
-		tmp &= ~ATOM_S3_ASIC_GUI_ENGINE_HUNG;
+	if (amdgpu_sriov_vf(adev) &&
+	    ((drv_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION <<
+		ATOM_VRAM_OPERATION_FLAGS_SHIFT)) == 0)) {
+		/* driver request VRAM reservation for SR-IOV */
+		adev->mman.drv_vram_usage_start_offset = (drv_start_addr &
+			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
+		adev->mman.drv_vram_usage_size = drv_size << 10;
+	}
 
-	WREG32(adev->bios_scratch_reg_offset + 3, tmp);
+	*usage_bytes = 0;
+	return 0;
 }
 
 int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev)
@@ -106,20 +175,31 @@ int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev)
 	struct atom_context *ctx = adev->mode_info.atom_context;
 	int index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
 						vram_usagebyfirmware);
-	uint16_t data_offset;
+	struct vram_usagebyfirmware_v2_1 *fw_usage_v2_1;
+	struct vram_usagebyfirmware_v2_2 *fw_usage_v2_2;
+	u16 data_offset;
+	u8 frev, crev;
 	int usage_bytes = 0;
 
-	if (amdgpu_atom_parse_data_header(ctx, index, NULL, NULL, NULL, &data_offset)) {
-		struct vram_usagebyfirmware_v2_1 *firmware_usage =
-			(struct vram_usagebyfirmware_v2_1 *)(ctx->bios + data_offset);
-
-		DRM_DEBUG("atom firmware requested %08x %dkb fw %dkb drv\n",
-			  le32_to_cpu(firmware_usage->start_address_in_kb),
-			  le16_to_cpu(firmware_usage->used_by_firmware_in_kb),
-			  le16_to_cpu(firmware_usage->used_by_driver_in_kb));
-
-		usage_bytes = le16_to_cpu(firmware_usage->used_by_driver_in_kb) * 1024;
+	/* Skip atomfirmware allocation for SRIOV VFs when dynamic crit regn is enabled */
+	if (!(amdgpu_sriov_vf(adev) && adev->virt.is_dynamic_crit_regn_enabled)) {
+		if (amdgpu_atom_parse_data_header(ctx, index, NULL, &frev, &crev, &data_offset)) {
+			if (frev == 2 && crev == 1) {
+				fw_usage_v2_1 =
+					(struct vram_usagebyfirmware_v2_1 *)(ctx->bios + data_offset);
+				amdgpu_atomfirmware_allocate_fb_v2_1(adev,
+						fw_usage_v2_1,
+						&usage_bytes);
+			} else if (frev >= 2 && crev >= 2) {
+				fw_usage_v2_2 =
+					(struct vram_usagebyfirmware_v2_2 *)(ctx->bios + data_offset);
+				amdgpu_atomfirmware_allocate_fb_v2_2(adev,
+						fw_usage_v2_2,
+						&usage_bytes);
+			}
+		}
 	}
+
 	ctx->scratch_size_bytes = 0;
 	if (usage_bytes == 0)
 		usage_bytes = 20 * 1024;
@@ -130,3 +210,802 @@ int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev)
 	ctx->scratch_size_bytes = usage_bytes;
 	return 0;
 }
+
+union igp_info {
+	struct atom_integrated_system_info_v1_11 v11;
+	struct atom_integrated_system_info_v1_12 v12;
+	struct atom_integrated_system_info_v2_1 v21;
+	struct atom_integrated_system_info_v2_3 v23;
+};
+
+union umc_info {
+	struct atom_umc_info_v3_1 v31;
+	struct atom_umc_info_v3_2 v32;
+	struct atom_umc_info_v3_3 v33;
+	struct atom_umc_info_v4_0 v40;
+};
+
+union vram_info {
+	struct atom_vram_info_header_v2_3 v23;
+	struct atom_vram_info_header_v2_4 v24;
+	struct atom_vram_info_header_v2_5 v25;
+	struct atom_vram_info_header_v2_6 v26;
+	struct atom_vram_info_header_v3_0 v30;
+};
+
+union vram_module {
+	struct atom_vram_module_v9 v9;
+	struct atom_vram_module_v10 v10;
+	struct atom_vram_module_v11 v11;
+	struct atom_vram_module_v3_0 v30;
+};
+
+static int convert_atom_mem_type_to_vram_type(struct amdgpu_device *adev,
+					      int atom_mem_type)
+{
+	int vram_type;
+
+	if (adev->flags & AMD_IS_APU) {
+		switch (atom_mem_type) {
+		case Ddr2MemType:
+		case LpDdr2MemType:
+			vram_type = AMDGPU_VRAM_TYPE_DDR2;
+			break;
+		case Ddr3MemType:
+		case LpDdr3MemType:
+			vram_type = AMDGPU_VRAM_TYPE_DDR3;
+			break;
+		case Ddr4MemType:
+			vram_type = AMDGPU_VRAM_TYPE_DDR4;
+			break;
+		case LpDdr4MemType:
+			vram_type = AMDGPU_VRAM_TYPE_LPDDR4;
+			break;
+		case Ddr5MemType:
+			vram_type = AMDGPU_VRAM_TYPE_DDR5;
+			break;
+		case LpDdr5MemType:
+			vram_type = AMDGPU_VRAM_TYPE_LPDDR5;
+			break;
+		default:
+			vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
+			break;
+		}
+	} else {
+		switch (atom_mem_type) {
+		case ATOM_DGPU_VRAM_TYPE_GDDR5:
+			vram_type = AMDGPU_VRAM_TYPE_GDDR5;
+			break;
+		case ATOM_DGPU_VRAM_TYPE_HBM2:
+		case ATOM_DGPU_VRAM_TYPE_HBM2E:
+		case ATOM_DGPU_VRAM_TYPE_HBM3:
+			vram_type = AMDGPU_VRAM_TYPE_HBM;
+			break;
+		case ATOM_DGPU_VRAM_TYPE_GDDR6:
+			vram_type = AMDGPU_VRAM_TYPE_GDDR6;
+			break;
+		case ATOM_DGPU_VRAM_TYPE_HBM3E:
+			vram_type = AMDGPU_VRAM_TYPE_HBM3E;
+			break;
+		default:
+			vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
+			break;
+		}
+	}
+
+	return vram_type;
+}
+
+int
+amdgpu_atomfirmware_get_vram_info(struct amdgpu_device *adev,
+				  int *vram_width, int *vram_type,
+				  int *vram_vendor)
+{
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	int index, i = 0;
+	u16 data_offset, size;
+	union igp_info *igp_info;
+	union vram_info *vram_info;
+	union umc_info *umc_info;
+	union vram_module *vram_module;
+	u8 frev, crev;
+	u8 mem_type;
+	u8 mem_vendor;
+	u32 mem_channel_number;
+	u32 mem_channel_width;
+	u32 module_id;
+
+	if (adev->flags & AMD_IS_APU)
+		index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+						    integratedsysteminfo);
+	else {
+		switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+		case IP_VERSION(12, 0, 0):
+		case IP_VERSION(12, 0, 1):
+			index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1, umc_info);
+			break;
+		default:
+			index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1, vram_info);
+		}
+	}
+	if (amdgpu_atom_parse_data_header(mode_info->atom_context,
+					  index, &size,
+					  &frev, &crev, &data_offset)) {
+		if (adev->flags & AMD_IS_APU) {
+			igp_info = (union igp_info *)
+				(mode_info->atom_context->bios + data_offset);
+			switch (frev) {
+			case 1:
+				switch (crev) {
+				case 11:
+				case 12:
+					mem_channel_number = igp_info->v11.umachannelnumber;
+					if (!mem_channel_number)
+						mem_channel_number = 1;
+					mem_type = igp_info->v11.memorytype;
+					if (mem_type == LpDdr5MemType)
+						mem_channel_width = 32;
+					else
+						mem_channel_width = 64;
+					if (vram_width)
+						*vram_width = mem_channel_number * mem_channel_width;
+					if (vram_type)
+						*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+					break;
+				default:
+					return -EINVAL;
+				}
+				break;
+			case 2:
+				switch (crev) {
+				case 1:
+				case 2:
+					mem_channel_number = igp_info->v21.umachannelnumber;
+					if (!mem_channel_number)
+						mem_channel_number = 1;
+					mem_type = igp_info->v21.memorytype;
+					if (mem_type == LpDdr5MemType)
+						mem_channel_width = 32;
+					else
+						mem_channel_width = 64;
+					if (vram_width)
+						*vram_width = mem_channel_number * mem_channel_width;
+					if (vram_type)
+						*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+					break;
+				case 3:
+					mem_channel_number = igp_info->v23.umachannelnumber;
+					if (!mem_channel_number)
+						mem_channel_number = 1;
+					mem_type = igp_info->v23.memorytype;
+					if (mem_type == LpDdr5MemType)
+						mem_channel_width = 32;
+					else
+						mem_channel_width = 64;
+					if (vram_width)
+						*vram_width = mem_channel_number * mem_channel_width;
+					if (vram_type)
+						*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+					break;
+				default:
+					return -EINVAL;
+				}
+				break;
+			default:
+				return -EINVAL;
+			}
+		} else {
+			switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+			case IP_VERSION(12, 0, 0):
+			case IP_VERSION(12, 0, 1):
+				umc_info = (union umc_info *)(mode_info->atom_context->bios + data_offset);
+
+				if (frev == 4) {
+					switch (crev) {
+					case 0:
+						mem_channel_number = le32_to_cpu(umc_info->v40.channel_num);
+						mem_type = le32_to_cpu(umc_info->v40.vram_type);
+						mem_channel_width = le32_to_cpu(umc_info->v40.channel_width);
+						mem_vendor = RREG32(adev->bios_scratch_reg_offset + 4) & 0xF;
+						if (vram_vendor)
+							*vram_vendor = mem_vendor;
+						if (vram_type)
+							*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+						if (vram_width)
+							*vram_width = mem_channel_number * (1 << mem_channel_width);
+						break;
+					default:
+						return -EINVAL;
+					}
+				} else
+					return -EINVAL;
+				break;
+			default:
+				vram_info = (union vram_info *)
+					(mode_info->atom_context->bios + data_offset);
+
+				module_id = (RREG32(adev->bios_scratch_reg_offset + 4) & 0x00ff0000) >> 16;
+				if (frev == 3) {
+					switch (crev) {
+					/* v30 */
+					case 0:
+						vram_module = (union vram_module *)vram_info->v30.vram_module;
+						mem_vendor = (vram_module->v30.dram_vendor_id) & 0xF;
+						if (vram_vendor)
+							*vram_vendor = mem_vendor;
+						mem_type = vram_info->v30.memory_type;
+						if (vram_type)
+							*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+						mem_channel_number = vram_info->v30.channel_num;
+						mem_channel_width = vram_info->v30.channel_width;
+						if (vram_width)
+							*vram_width = mem_channel_number * 16;
+						break;
+					default:
+						return -EINVAL;
+					}
+				} else if (frev == 2) {
+					switch (crev) {
+					/* v23 */
+					case 3:
+						if (module_id > vram_info->v23.vram_module_num)
+							module_id = 0;
+						vram_module = (union vram_module *)vram_info->v23.vram_module;
+						while (i < module_id) {
+							vram_module = (union vram_module *)
+								((u8 *)vram_module + vram_module->v9.vram_module_size);
+							i++;
+						}
+						mem_type = vram_module->v9.memory_type;
+						if (vram_type)
+							*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+						mem_channel_number = vram_module->v9.channel_num;
+						mem_channel_width = vram_module->v9.channel_width;
+						if (vram_width)
+							*vram_width = mem_channel_number * (1 << mem_channel_width);
+						mem_vendor = (vram_module->v9.vender_rev_id) & 0xF;
+						if (vram_vendor)
+							*vram_vendor = mem_vendor;
+						break;
+					/* v24 */
+					case 4:
+						if (module_id > vram_info->v24.vram_module_num)
+							module_id = 0;
+						vram_module = (union vram_module *)vram_info->v24.vram_module;
+						while (i < module_id) {
+							vram_module = (union vram_module *)
+								((u8 *)vram_module + vram_module->v10.vram_module_size);
+							i++;
+						}
+						mem_type = vram_module->v10.memory_type;
+						if (vram_type)
+							*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+						mem_channel_number = vram_module->v10.channel_num;
+						mem_channel_width = vram_module->v10.channel_width;
+						if (vram_width)
+							*vram_width = mem_channel_number * (1 << mem_channel_width);
+						mem_vendor = (vram_module->v10.vender_rev_id) & 0xF;
+						if (vram_vendor)
+							*vram_vendor = mem_vendor;
+						break;
+					/* v25 */
+					case 5:
+						if (module_id > vram_info->v25.vram_module_num)
+							module_id = 0;
+						vram_module = (union vram_module *)vram_info->v25.vram_module;
+						while (i < module_id) {
+							vram_module = (union vram_module *)
+								((u8 *)vram_module + vram_module->v11.vram_module_size);
+							i++;
+						}
+						mem_type = vram_module->v11.memory_type;
+						if (vram_type)
+							*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+						mem_channel_number = vram_module->v11.channel_num;
+						mem_channel_width = vram_module->v11.channel_width;
+						if (vram_width)
+							*vram_width = mem_channel_number * (1 << mem_channel_width);
+						mem_vendor = (vram_module->v11.vender_rev_id) & 0xF;
+						if (vram_vendor)
+							*vram_vendor = mem_vendor;
+						break;
+					/* v26 */
+					case 6:
+						if (module_id > vram_info->v26.vram_module_num)
+							module_id = 0;
+						vram_module = (union vram_module *)vram_info->v26.vram_module;
+						while (i < module_id) {
+							vram_module = (union vram_module *)
+								((u8 *)vram_module + vram_module->v9.vram_module_size);
+							i++;
+						}
+						mem_type = vram_module->v9.memory_type;
+						if (vram_type)
+							*vram_type = convert_atom_mem_type_to_vram_type(adev, mem_type);
+						mem_channel_number = vram_module->v9.channel_num;
+						mem_channel_width = vram_module->v9.channel_width;
+						if (vram_width)
+							*vram_width = mem_channel_number * (1 << mem_channel_width);
+						mem_vendor = (vram_module->v9.vender_rev_id) & 0xF;
+						if (vram_vendor)
+							*vram_vendor = mem_vendor;
+						break;
+					default:
+						return -EINVAL;
+					}
+				} else {
+					/* invalid frev */
+					return -EINVAL;
+				}
+			}
+		}
+	}
+
+	return 0;
+}
+
+/*
+ * Return true if vbios enabled ecc by default, if umc info table is available
+ * or false if ecc is not enabled or umc info table is not available
+ */
+bool amdgpu_atomfirmware_mem_ecc_supported(struct amdgpu_device *adev)
+{
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	int index;
+	u16 data_offset, size;
+	union umc_info *umc_info;
+	u8 frev, crev;
+	bool mem_ecc_enabled = false;
+	u8 umc_config;
+	u32 umc_config1;
+	adev->ras_default_ecc_enabled = false;
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+			umc_info);
+
+	if (amdgpu_atom_parse_data_header(mode_info->atom_context,
+				index, &size, &frev, &crev, &data_offset)) {
+		umc_info = (union umc_info *)(mode_info->atom_context->bios + data_offset);
+		if (frev == 3) {
+			switch (crev) {
+			case 1:
+				umc_config = le32_to_cpu(umc_info->v31.umc_config);
+				mem_ecc_enabled =
+					(umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ? true : false;
+				break;
+			case 2:
+				umc_config = le32_to_cpu(umc_info->v32.umc_config);
+				mem_ecc_enabled =
+					(umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ? true : false;
+				break;
+			case 3:
+				umc_config = le32_to_cpu(umc_info->v33.umc_config);
+				umc_config1 = le32_to_cpu(umc_info->v33.umc_config1);
+				mem_ecc_enabled =
+					((umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ||
+					 (umc_config1 & UMC_CONFIG1__ENABLE_ECC_CAPABLE)) ? true : false;
+				adev->ras_default_ecc_enabled =
+					(umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ? true : false;
+				break;
+			default:
+				/* unsupported crev */
+				return false;
+			}
+		} else if (frev == 4) {
+			switch (crev) {
+			case 0:
+				umc_config = le32_to_cpu(umc_info->v40.umc_config);
+				umc_config1 = le32_to_cpu(umc_info->v40.umc_config1);
+				mem_ecc_enabled =
+					(umc_config1 & UMC_CONFIG1__ENABLE_ECC_CAPABLE) ? true : false;
+				adev->ras_default_ecc_enabled =
+					(umc_config & UMC_CONFIG__DEFAULT_MEM_ECC_ENABLE) ? true : false;
+				break;
+			default:
+				/* unsupported crev */
+				return false;
+			}
+		} else {
+			/* unsupported frev */
+			return false;
+		}
+	}
+
+	return mem_ecc_enabled;
+}
+
+/*
+ * Helper function to query sram ecc capablity
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Return true if vbios supports sram ecc or false if not
+ */
+bool amdgpu_atomfirmware_sram_ecc_supported(struct amdgpu_device *adev)
+{
+	u32 fw_cap;
+
+	fw_cap = adev->mode_info.firmware_flags;
+
+	return (fw_cap & ATOM_FIRMWARE_CAP_SRAM_ECC) ? true : false;
+}
+
+/*
+ * Helper function to query dynamic boot config capability
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Return true if vbios supports dynamic boot config or false if not
+ */
+bool amdgpu_atomfirmware_dynamic_boot_config_supported(struct amdgpu_device *adev)
+{
+	u32 fw_cap;
+
+	fw_cap = adev->mode_info.firmware_flags;
+
+	return (fw_cap & ATOM_FIRMWARE_CAP_DYNAMIC_BOOT_CFG_ENABLE) ? true : false;
+}
+
+/**
+ * amdgpu_atomfirmware_ras_rom_addr -- Get the RAS EEPROM addr from VBIOS
+ * @adev: amdgpu_device pointer
+ * @i2c_address: pointer to u8; if not NULL, will contain
+ *    the RAS EEPROM address if the function returns true
+ *
+ * Return true if VBIOS supports RAS EEPROM address reporting,
+ * else return false. If true and @i2c_address is not NULL,
+ * will contain the RAS ROM address.
+ */
+bool amdgpu_atomfirmware_ras_rom_addr(struct amdgpu_device *adev,
+				      u8 *i2c_address)
+{
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	int index;
+	u16 data_offset, size;
+	union firmware_info *firmware_info;
+	u8 frev, crev;
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+					    firmwareinfo);
+
+	if (amdgpu_atom_parse_data_header(adev->mode_info.atom_context,
+					  index, &size, &frev, &crev,
+					  &data_offset)) {
+		/* support firmware_info 3.4 + */
+		if ((frev == 3 && crev >= 4) || (frev > 3)) {
+			firmware_info = (union firmware_info *)
+				(mode_info->atom_context->bios + data_offset);
+			/* The ras_rom_i2c_slave_addr should ideally
+			 * be a 19-bit EEPROM address, which would be
+			 * used as is by the driver; see top of
+			 * amdgpu_eeprom.c.
+			 *
+			 * When this is the case, 0 is of course a
+			 * valid RAS EEPROM address, in which case,
+			 * we'll drop the first "if (firm...)" and only
+			 * leave the check for the pointer.
+			 *
+			 * The reason this works right now is because
+			 * ras_rom_i2c_slave_addr contains the EEPROM
+			 * device type qualifier 1010b in the top 4
+			 * bits.
+			 */
+			if (firmware_info->v34.ras_rom_i2c_slave_addr) {
+				if (i2c_address)
+					*i2c_address = firmware_info->v34.ras_rom_i2c_slave_addr;
+				return true;
+			}
+		}
+	}
+
+	return false;
+}
+
+
+union smu_info {
+	struct atom_smu_info_v3_1 v31;
+	struct atom_smu_info_v4_0 v40;
+};
+
+union gfx_info {
+	struct atom_gfx_info_v2_2 v22;
+	struct atom_gfx_info_v2_4 v24;
+	struct atom_gfx_info_v2_7 v27;
+	struct atom_gfx_info_v3_0 v30;
+};
+
+int amdgpu_atomfirmware_get_clock_info(struct amdgpu_device *adev)
+{
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	struct amdgpu_pll *spll = &adev->clock.spll;
+	struct amdgpu_pll *mpll = &adev->clock.mpll;
+	uint8_t frev, crev;
+	uint16_t data_offset;
+	int ret = -EINVAL, index;
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+					    firmwareinfo);
+	if (amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset)) {
+		union firmware_info *firmware_info =
+			(union firmware_info *)(mode_info->atom_context->bios +
+						data_offset);
+
+		adev->clock.default_sclk =
+			le32_to_cpu(firmware_info->v31.bootup_sclk_in10khz);
+		adev->clock.default_mclk =
+			le32_to_cpu(firmware_info->v31.bootup_mclk_in10khz);
+
+		adev->pm.current_sclk = adev->clock.default_sclk;
+		adev->pm.current_mclk = adev->clock.default_mclk;
+
+		ret = 0;
+	}
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+					    smu_info);
+	if (amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset)) {
+		union smu_info *smu_info =
+			(union smu_info *)(mode_info->atom_context->bios +
+					   data_offset);
+
+		/* system clock */
+		if (frev == 3)
+			spll->reference_freq = le32_to_cpu(smu_info->v31.core_refclk_10khz);
+		else if (frev == 4)
+			spll->reference_freq = le32_to_cpu(smu_info->v40.core_refclk_10khz);
+
+		spll->reference_div = 0;
+		spll->min_post_div = 1;
+		spll->max_post_div = 1;
+		spll->min_ref_div = 2;
+		spll->max_ref_div = 0xff;
+		spll->min_feedback_div = 4;
+		spll->max_feedback_div = 0xff;
+		spll->best_vco = 0;
+
+		ret = 0;
+	}
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+					    umc_info);
+	if (amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset)) {
+		union umc_info *umc_info =
+			(union umc_info *)(mode_info->atom_context->bios +
+					   data_offset);
+
+		/* memory clock */
+		mpll->reference_freq = le32_to_cpu(umc_info->v31.mem_refclk_10khz);
+
+		mpll->reference_div = 0;
+		mpll->min_post_div = 1;
+		mpll->max_post_div = 1;
+		mpll->min_ref_div = 2;
+		mpll->max_ref_div = 0xff;
+		mpll->min_feedback_div = 4;
+		mpll->max_feedback_div = 0xff;
+		mpll->best_vco = 0;
+
+		ret = 0;
+	}
+
+	/* if asic is Navi+, the rlc reference clock is used for system clock
+	 * from vbios gfx_info table */
+	if (adev->asic_type >= CHIP_NAVI10) {
+		index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+						   gfx_info);
+		if (amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
+					  &frev, &crev, &data_offset)) {
+			union gfx_info *gfx_info = (union gfx_info *)
+				(mode_info->atom_context->bios + data_offset);
+			if ((frev == 3) ||
+			    (frev == 2 && crev == 6)) {
+				spll->reference_freq = le32_to_cpu(gfx_info->v30.golden_tsc_count_lower_refclk);
+				ret = 0;
+			} else if ((frev == 2) &&
+				   (crev >= 2) &&
+				   (crev != 6)) {
+				spll->reference_freq = le32_to_cpu(gfx_info->v22.rlc_gpu_timer_refclk);
+				ret = 0;
+			} else {
+				BUG();
+			}
+		}
+	}
+
+	return ret;
+}
+
+int amdgpu_atomfirmware_get_gfx_info(struct amdgpu_device *adev)
+{
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	int index;
+	uint8_t frev, crev;
+	uint16_t data_offset;
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+					    gfx_info);
+	if (amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset)) {
+		union gfx_info *gfx_info = (union gfx_info *)
+			(mode_info->atom_context->bios + data_offset);
+		if (frev == 2) {
+			switch (crev) {
+			case 4:
+				adev->gfx.config.max_shader_engines = gfx_info->v24.max_shader_engines;
+				adev->gfx.config.max_cu_per_sh = gfx_info->v24.max_cu_per_sh;
+				adev->gfx.config.max_sh_per_se = gfx_info->v24.max_sh_per_se;
+				adev->gfx.config.max_backends_per_se = gfx_info->v24.max_backends_per_se;
+				adev->gfx.config.max_texture_channel_caches = gfx_info->v24.max_texture_channel_caches;
+				adev->gfx.config.max_gprs = le16_to_cpu(gfx_info->v24.gc_num_gprs);
+				adev->gfx.config.max_gs_threads = gfx_info->v24.gc_num_max_gs_thds;
+				adev->gfx.config.gs_vgt_table_depth = gfx_info->v24.gc_gs_table_depth;
+				adev->gfx.config.gs_prim_buffer_depth =
+					le16_to_cpu(gfx_info->v24.gc_gsprim_buff_depth);
+				adev->gfx.config.double_offchip_lds_buf =
+					gfx_info->v24.gc_double_offchip_lds_buffer;
+				adev->gfx.cu_info.wave_front_size = le16_to_cpu(gfx_info->v24.gc_wave_size);
+				adev->gfx.cu_info.max_waves_per_simd = le16_to_cpu(gfx_info->v24.gc_max_waves_per_simd);
+				adev->gfx.cu_info.max_scratch_slots_per_cu = gfx_info->v24.gc_max_scratch_slots_per_cu;
+				adev->gfx.cu_info.lds_size = le16_to_cpu(gfx_info->v24.gc_lds_size);
+				return 0;
+			case 7:
+				adev->gfx.config.max_shader_engines = gfx_info->v27.max_shader_engines;
+				adev->gfx.config.max_cu_per_sh = gfx_info->v27.max_cu_per_sh;
+				adev->gfx.config.max_sh_per_se = gfx_info->v27.max_sh_per_se;
+				adev->gfx.config.max_backends_per_se = gfx_info->v27.max_backends_per_se;
+				adev->gfx.config.max_texture_channel_caches = gfx_info->v27.max_texture_channel_caches;
+				adev->gfx.config.max_gprs = le16_to_cpu(gfx_info->v27.gc_num_gprs);
+				adev->gfx.config.max_gs_threads = gfx_info->v27.gc_num_max_gs_thds;
+				adev->gfx.config.gs_vgt_table_depth = gfx_info->v27.gc_gs_table_depth;
+				adev->gfx.config.gs_prim_buffer_depth = le16_to_cpu(gfx_info->v27.gc_gsprim_buff_depth);
+				adev->gfx.config.double_offchip_lds_buf = gfx_info->v27.gc_double_offchip_lds_buffer;
+				adev->gfx.cu_info.wave_front_size = le16_to_cpu(gfx_info->v27.gc_wave_size);
+				adev->gfx.cu_info.max_waves_per_simd = le16_to_cpu(gfx_info->v27.gc_max_waves_per_simd);
+				adev->gfx.cu_info.max_scratch_slots_per_cu = gfx_info->v27.gc_max_scratch_slots_per_cu;
+				adev->gfx.cu_info.lds_size = le16_to_cpu(gfx_info->v27.gc_lds_size);
+				return 0;
+			default:
+				return -EINVAL;
+			}
+		} else if (frev == 3) {
+			switch (crev) {
+			case 0:
+				adev->gfx.config.max_shader_engines = gfx_info->v30.max_shader_engines;
+				adev->gfx.config.max_cu_per_sh = gfx_info->v30.max_cu_per_sh;
+				adev->gfx.config.max_sh_per_se = gfx_info->v30.max_sh_per_se;
+				adev->gfx.config.max_backends_per_se = gfx_info->v30.max_backends_per_se;
+				adev->gfx.config.max_texture_channel_caches = gfx_info->v30.max_texture_channel_caches;
+				return 0;
+			default:
+				return -EINVAL;
+			}
+		} else {
+			return -EINVAL;
+		}
+
+	}
+	return -EINVAL;
+}
+
+/*
+ * Helper function to query two stage mem training capability
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Return true if two stage mem training is supported or false if not
+ */
+bool amdgpu_atomfirmware_mem_training_supported(struct amdgpu_device *adev)
+{
+	u32 fw_cap;
+
+	fw_cap = adev->mode_info.firmware_flags;
+
+	return (fw_cap & ATOM_FIRMWARE_CAP_ENABLE_2STAGE_BIST_TRAINING) ? true : false;
+}
+
+int amdgpu_atomfirmware_get_fw_reserved_fb_size(struct amdgpu_device *adev)
+{
+	struct atom_context *ctx = adev->mode_info.atom_context;
+	union firmware_info *firmware_info;
+	int index;
+	u16 data_offset, size;
+	u8 frev, crev;
+	int fw_reserved_fb_size;
+
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+			firmwareinfo);
+
+	if (!amdgpu_atom_parse_data_header(ctx, index, &size,
+				&frev, &crev, &data_offset))
+		/* fail to parse data_header */
+		return 0;
+
+	firmware_info = (union firmware_info *)(ctx->bios + data_offset);
+
+	if (frev != 3)
+		return -EINVAL;
+
+	switch (crev) {
+	case 4:
+		fw_reserved_fb_size =
+			(firmware_info->v34.fw_reserved_size_in_kb << 10);
+		break;
+	case 5:
+		fw_reserved_fb_size =
+			(firmware_info->v35.fw_reserved_size_in_kb << 10);
+		break;
+	default:
+		fw_reserved_fb_size = 0;
+		break;
+	}
+
+	return fw_reserved_fb_size;
+}
+
+/*
+ * Helper function to execute asic_init table
+ *
+ * @adev: amdgpu_device pointer
+ * @fb_reset: flag to indicate whether fb is reset or not
+ *
+ * Return 0 if succeed, otherwise failed
+ */
+int amdgpu_atomfirmware_asic_init(struct amdgpu_device *adev, bool fb_reset)
+{
+	struct amdgpu_mode_info *mode_info = &adev->mode_info;
+	struct atom_context *ctx;
+	uint8_t frev, crev;
+	uint16_t data_offset;
+	uint32_t bootup_sclk_in10khz, bootup_mclk_in10khz;
+	struct asic_init_ps_allocation_v2_1 asic_init_ps_v2_1;
+	int index;
+
+	if (!mode_info)
+		return -EINVAL;
+
+	ctx = mode_info->atom_context;
+	if (!ctx)
+		return -EINVAL;
+
+	/* query bootup sclk/mclk from firmware_info table */
+	index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
+					    firmwareinfo);
+	if (amdgpu_atom_parse_data_header(ctx, index, NULL,
+				&frev, &crev, &data_offset)) {
+		union firmware_info *firmware_info =
+			(union firmware_info *)(ctx->bios +
+						data_offset);
+
+		bootup_sclk_in10khz =
+			le32_to_cpu(firmware_info->v31.bootup_sclk_in10khz);
+		bootup_mclk_in10khz =
+			le32_to_cpu(firmware_info->v31.bootup_mclk_in10khz);
+	} else {
+		return -EINVAL;
+	}
+
+	index = get_index_into_master_table(atom_master_list_of_command_functions_v2_1,
+					asic_init);
+	if (amdgpu_atom_parse_cmd_header(mode_info->atom_context, index, &frev, &crev)) {
+		if (frev == 2 && crev >= 1) {
+			memset(&asic_init_ps_v2_1, 0, sizeof(asic_init_ps_v2_1));
+			asic_init_ps_v2_1.param.engineparam.sclkfreqin10khz = bootup_sclk_in10khz;
+			asic_init_ps_v2_1.param.memparam.mclkfreqin10khz = bootup_mclk_in10khz;
+			asic_init_ps_v2_1.param.engineparam.engineflag = b3NORMAL_ENGINE_INIT;
+			if (!fb_reset)
+				asic_init_ps_v2_1.param.memparam.memflag = b3DRAM_SELF_REFRESH_EXIT;
+			else
+				asic_init_ps_v2_1.param.memparam.memflag = 0;
+		} else {
+			return -EINVAL;
+		}
+	} else {
+		return -EINVAL;
+	}
+
+	return amdgpu_atom_execute_table(ctx, ATOM_CMD_INIT, (uint32_t *)&asic_init_ps_v2_1,
+		sizeof(asic_init_ps_v2_1));
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.h
index a2c3ebe22c71..649b5530d8ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.h
@@ -24,12 +24,22 @@
 #ifndef __AMDGPU_ATOMFIRMWARE_H__
 #define __AMDGPU_ATOMFIRMWARE_H__
 
-bool amdgpu_atomfirmware_gpu_supports_virtualization(struct amdgpu_device *adev);
+#define get_index_into_master_table(master_table, table_name) (offsetof(struct master_table, table_name) / sizeof(uint16_t))
+
+uint32_t amdgpu_atomfirmware_query_firmware_capability(struct amdgpu_device *adev);
+bool amdgpu_atomfirmware_gpu_virtualization_supported(struct amdgpu_device *adev);
 void amdgpu_atomfirmware_scratch_regs_init(struct amdgpu_device *adev);
-void amdgpu_atomfirmware_scratch_regs_save(struct amdgpu_device *adev);
-void amdgpu_atomfirmware_scratch_regs_restore(struct amdgpu_device *adev);
-void amdgpu_atomfirmware_scratch_regs_engine_hung(struct amdgpu_device *adev,
-						  bool hung);
 int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev);
+int amdgpu_atomfirmware_get_vram_info(struct amdgpu_device *adev,
+	int *vram_width, int *vram_type, int *vram_vendor);
+int amdgpu_atomfirmware_get_clock_info(struct amdgpu_device *adev);
+int amdgpu_atomfirmware_get_gfx_info(struct amdgpu_device *adev);
+bool amdgpu_atomfirmware_mem_ecc_supported(struct amdgpu_device *adev);
+bool amdgpu_atomfirmware_sram_ecc_supported(struct amdgpu_device *adev);
+bool amdgpu_atomfirmware_ras_rom_addr(struct amdgpu_device *adev, uint8_t *i2c_address);
+bool amdgpu_atomfirmware_mem_training_supported(struct amdgpu_device *adev);
+bool amdgpu_atomfirmware_dynamic_boot_config_supported(struct amdgpu_device *adev);
+int amdgpu_atomfirmware_get_fw_reserved_fb_size(struct amdgpu_device *adev);
+int amdgpu_atomfirmware_asic_init(struct amdgpu_device *adev, bool fb_reset);
 
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c
index c13c51af0b68..3893e6fc2f03 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c
@@ -1,9 +1,8 @@
+// SPDX-License-Identifier: GPL-2.0-only
 /*
  * Copyright (c) 2010 Red Hat Inc.
  * Author : Dave Airlie <airlied@redhat.com>
  *
- * Licensed under GPLv2
- *
  * ATPX support for both Intel/ATI
  */
 #include <linux/vga_switcheroo.h>
@@ -12,8 +11,19 @@
 #include <linux/pci.h>
 #include <linux/delay.h>
 
+#include "amdgpu.h"
 #include "amd_acpi.h"
 
+#define AMDGPU_PX_QUIRK_FORCE_ATPX  (1 << 0)
+
+struct amdgpu_px_quirk {
+	u32 chip_vendor;
+	u32 chip_device;
+	u32 subsys_vendor;
+	u32 subsys_device;
+	u32 px_quirk_flags;
+};
+
 struct amdgpu_atpx_functions {
 	bool px_params;
 	bool power_cntl;
@@ -22,7 +32,7 @@ struct amdgpu_atpx_functions {
 	bool switch_start;
 	bool switch_end;
 	bool disp_connectors_mapping;
-	bool disp_detetion_ports;
+	bool disp_detection_ports;
 };
 
 struct amdgpu_atpx {
@@ -35,6 +45,7 @@ struct amdgpu_atpx {
 static struct amdgpu_atpx_priv {
 	bool atpx_detected;
 	bool bridge_pm_usable;
+	unsigned int quirks;
 	/* handle for device - and atpx */
 	acpi_handle dhandle;
 	acpi_handle other_handle;
@@ -63,22 +74,21 @@ struct atpx_mux {
 	u16 mux;
 } __packed;
 
-bool amdgpu_has_atpx(void) {
+bool amdgpu_has_atpx(void)
+{
 	return amdgpu_atpx_priv.atpx_detected;
 }
 
-bool amdgpu_has_atpx_dgpu_power_cntl(void) {
+bool amdgpu_has_atpx_dgpu_power_cntl(void)
+{
 	return amdgpu_atpx_priv.atpx.functions.power_cntl;
 }
 
-bool amdgpu_is_atpx_hybrid(void) {
+bool amdgpu_is_atpx_hybrid(void)
+{
 	return amdgpu_atpx_priv.atpx.is_hybrid;
 }
 
-bool amdgpu_atpx_dgpu_req_power_for_displays(void) {
-	return amdgpu_atpx_priv.atpx.dgpu_req_power_for_displays;
-}
-
 /**
  * amdgpu_atpx_call - call an ATPX method
  *
@@ -117,7 +127,7 @@ static union acpi_object *amdgpu_atpx_call(acpi_handle handle, int function,
 
 	/* Fail only if calling the method fails and ATPX is supported */
 	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND) {
-		printk("failed to evaluate ATPX got %s\n",
+		pr_err("failed to evaluate ATPX got %s\n",
 		       acpi_format_exception(status));
 		kfree(buffer.pointer);
 		return NULL;
@@ -145,11 +155,11 @@ static void amdgpu_atpx_parse_functions(struct amdgpu_atpx_functions *f, u32 mas
 	f->switch_start = mask & ATPX_GRAPHICS_DEVICE_SWITCH_START_NOTIFICATION_SUPPORTED;
 	f->switch_end = mask & ATPX_GRAPHICS_DEVICE_SWITCH_END_NOTIFICATION_SUPPORTED;
 	f->disp_connectors_mapping = mask & ATPX_GET_DISPLAY_CONNECTORS_MAPPING_SUPPORTED;
-	f->disp_detetion_ports = mask & ATPX_GET_DISPLAY_DETECTION_PORTS_SUPPORTED;
+	f->disp_detection_ports = mask & ATPX_GET_DISPLAY_DETECTION_PORTS_SUPPORTED;
 }
 
 /**
- * amdgpu_atpx_validate_functions - validate ATPX functions
+ * amdgpu_atpx_validate - validate ATPX functions
  *
  * @atpx: amdgpu atpx struct
  *
@@ -173,7 +183,7 @@ static int amdgpu_atpx_validate(struct amdgpu_atpx *atpx)
 
 		size = *(u16 *) info->buffer.pointer;
 		if (size < 10) {
-			printk("ATPX buffer is too small: %zu\n", size);
+			pr_err("ATPX buffer is too small: %zu\n", size);
 			kfree(info);
 			return -EINVAL;
 		}
@@ -205,13 +215,19 @@ static int amdgpu_atpx_validate(struct amdgpu_atpx *atpx)
 
 	atpx->is_hybrid = false;
 	if (valid_bits & ATPX_MS_HYBRID_GFX_SUPPORTED) {
-		printk("ATPX Hybrid Graphics\n");
-		/*
-		 * Disable legacy PM methods only when pcie port PM is usable,
-		 * otherwise the device might fail to power off or power on.
-		 */
-		atpx->functions.power_cntl = !amdgpu_atpx_priv.bridge_pm_usable;
-		atpx->is_hybrid = true;
+		if (amdgpu_atpx_priv.quirks & AMDGPU_PX_QUIRK_FORCE_ATPX) {
+			pr_warn("ATPX Hybrid Graphics, forcing to ATPX\n");
+			atpx->functions.power_cntl = true;
+			atpx->is_hybrid = false;
+		} else {
+			pr_notice("ATPX Hybrid Graphics\n");
+			/*
+			 * Disable legacy PM methods only when pcie port PM is usable,
+			 * otherwise the device might fail to power off or power on.
+			 */
+			atpx->functions.power_cntl = !amdgpu_atpx_priv.bridge_pm_usable;
+			atpx->is_hybrid = true;
+		}
 	}
 
 	atpx->dgpu_req_power_for_displays = false;
@@ -246,7 +262,7 @@ static int amdgpu_atpx_verify_interface(struct amdgpu_atpx *atpx)
 
 	size = *(u16 *) info->buffer.pointer;
 	if (size < 8) {
-		printk("ATPX buffer is too small: %zu\n", size);
+		pr_err("ATPX buffer is too small: %zu\n", size);
 		err = -EINVAL;
 		goto out;
 	}
@@ -255,8 +271,8 @@ static int amdgpu_atpx_verify_interface(struct amdgpu_atpx *atpx)
 	memcpy(&output, info->buffer.pointer, size);
 
 	/* TODO: check version? */
-	printk("ATPX version %u, functions 0x%08x\n",
-	       output.version, output.function_bits);
+	pr_notice("ATPX version %u, functions 0x%08x\n",
+		  output.version, output.function_bits);
 
 	amdgpu_atpx_parse_functions(&atpx->functions, output.function_bits);
 
@@ -533,7 +549,7 @@ static int amdgpu_atpx_init(void)
  * look up whether we are the integrated or discrete GPU (all asics).
  * Returns the client id.
  */
-static int amdgpu_atpx_get_client_id(struct pci_dev *pdev)
+static enum vga_switcheroo_client_id amdgpu_atpx_get_client_id(struct pci_dev *pdev)
 {
 	if (amdgpu_atpx_priv.dhandle == ACPI_HANDLE(&pdev->dev))
 		return VGA_SWITCHEROO_IGD;
@@ -547,6 +563,34 @@ static const struct vga_switcheroo_handler amdgpu_atpx_handler = {
 	.get_client_id = amdgpu_atpx_get_client_id,
 };
 
+static const struct amdgpu_px_quirk amdgpu_px_quirk_list[] = {
+	/* HG _PR3 doesn't seem to work on this A+A weston board */
+	{ 0x1002, 0x6900, 0x1002, 0x0124, AMDGPU_PX_QUIRK_FORCE_ATPX },
+	{ 0x1002, 0x6900, 0x1028, 0x0812, AMDGPU_PX_QUIRK_FORCE_ATPX },
+	{ 0x1002, 0x6900, 0x1028, 0x0813, AMDGPU_PX_QUIRK_FORCE_ATPX },
+	{ 0x1002, 0x699f, 0x1028, 0x0814, AMDGPU_PX_QUIRK_FORCE_ATPX },
+	{ 0x1002, 0x6900, 0x1025, 0x125A, AMDGPU_PX_QUIRK_FORCE_ATPX },
+	{ 0x1002, 0x6900, 0x17AA, 0x3806, AMDGPU_PX_QUIRK_FORCE_ATPX },
+	{ 0, 0, 0, 0, 0 },
+};
+
+static void amdgpu_atpx_get_quirks(struct pci_dev *pdev)
+{
+	const struct amdgpu_px_quirk *p = amdgpu_px_quirk_list;
+
+	/* Apply PX quirks */
+	while (p && p->chip_device != 0) {
+		if (pdev->vendor == p->chip_vendor &&
+		    pdev->device == p->chip_device &&
+		    pdev->subsystem_vendor == p->subsys_vendor &&
+		    pdev->subsystem_device == p->subsys_device) {
+			amdgpu_atpx_priv.quirks |= p->px_quirk_flags;
+			break;
+		}
+		++p;
+	}
+}
+
 /**
  * amdgpu_atpx_detect - detect whether we have PX
  *
@@ -566,19 +610,21 @@ static bool amdgpu_atpx_detect(void)
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL) {
 		vga_count++;
 
-		has_atpx |= (amdgpu_atpx_pci_probe_handle(pdev) == true);
+		has_atpx |= amdgpu_atpx_pci_probe_handle(pdev);
 
 		parent_pdev = pci_upstream_bridge(pdev);
 		d3_supported |= parent_pdev && parent_pdev->bridge_d3;
+		amdgpu_atpx_get_quirks(pdev);
 	}
 
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_OTHER << 8, pdev)) != NULL) {
 		vga_count++;
 
-		has_atpx |= (amdgpu_atpx_pci_probe_handle(pdev) == true);
+		has_atpx |= amdgpu_atpx_pci_probe_handle(pdev);
 
 		parent_pdev = pci_upstream_bridge(pdev);
 		d3_supported |= parent_pdev && parent_pdev->bridge_d3;
+		amdgpu_atpx_get_quirks(pdev);
 	}
 
 	if (has_atpx && vga_count == 2) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c
index 1beae5b930d0..199693369c7c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c
@@ -21,7 +21,7 @@
  *
  * Authors: Jerome Glisse
  */
-#include <drm/drmP.h>
+
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 
@@ -29,120 +29,99 @@
 #define AMDGPU_BENCHMARK_COMMON_MODES_N 17
 
 static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,
-				    uint64_t saddr, uint64_t daddr, int n)
+				    uint64_t saddr, uint64_t daddr, int n, s64 *time_ms)
 {
-	unsigned long start_jiffies;
-	unsigned long end_jiffies;
-	struct dma_fence *fence = NULL;
+	ktime_t stime, etime;
+	struct dma_fence *fence;
 	int i, r;
 
-	start_jiffies = jiffies;
+	stime = ktime_get();
 	for (i = 0; i < n; i++) {
 		struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
 		r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence,
-				       false);
+				       false, false, 0);
 		if (r)
 			goto exit_do_move;
 		r = dma_fence_wait(fence, false);
+		dma_fence_put(fence);
 		if (r)
 			goto exit_do_move;
-		dma_fence_put(fence);
 	}
-	end_jiffies = jiffies;
-	r = jiffies_to_msecs(end_jiffies - start_jiffies);
 
 exit_do_move:
-	if (fence)
-		dma_fence_put(fence);
+	etime = ktime_get();
+	*time_ms = ktime_ms_delta(etime, stime);
+
 	return r;
 }
 
 
-static void amdgpu_benchmark_log_results(int n, unsigned size,
-					 unsigned int time,
+static void amdgpu_benchmark_log_results(struct amdgpu_device *adev,
+					 int n, unsigned size,
+					 s64 time_ms,
 					 unsigned sdomain, unsigned ddomain,
 					 char *kind)
 {
-	unsigned int throughput = (n * (size >> 10)) / time;
-	DRM_INFO("amdgpu: %s %u bo moves of %u kB from"
-		 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n",
-		 kind, n, size >> 10, sdomain, ddomain, time,
+	s64 throughput = (n * (size >> 10));
+
+	throughput = div64_s64(throughput, time_ms);
+
+	dev_info(adev->dev, "amdgpu: %s %u bo moves of %u kB from"
+		 " %d to %d in %lld ms, throughput: %lld Mb/s or %lld MB/s\n",
+		 kind, n, size >> 10, sdomain, ddomain, time_ms,
 		 throughput * 8, throughput);
 }
 
-static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size,
-				  unsigned sdomain, unsigned ddomain)
+static int amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size,
+				 unsigned sdomain, unsigned ddomain)
 {
 	struct amdgpu_bo *dobj = NULL;
 	struct amdgpu_bo *sobj = NULL;
 	uint64_t saddr, daddr;
+	s64 time_ms;
 	int r, n;
-	int time;
 
 	n = AMDGPU_BENCHMARK_ITERATIONS;
-	r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, sdomain, 0, NULL,
-			     NULL, &sobj);
-	if (r) {
-		goto out_cleanup;
-	}
-	r = amdgpu_bo_reserve(sobj, false);
-	if (unlikely(r != 0))
-		goto out_cleanup;
-	r = amdgpu_bo_pin(sobj, sdomain, &saddr);
-	amdgpu_bo_unreserve(sobj);
-	if (r) {
-		goto out_cleanup;
-	}
-	r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, ddomain, 0, NULL,
-			     NULL, &dobj);
-	if (r) {
-		goto out_cleanup;
-	}
-	r = amdgpu_bo_reserve(dobj, false);
-	if (unlikely(r != 0))
+
+	r = amdgpu_bo_create_kernel(adev, size,
+				    PAGE_SIZE, sdomain,
+				    &sobj,
+				    &saddr,
+				    NULL);
+	if (r)
 		goto out_cleanup;
-	r = amdgpu_bo_pin(dobj, ddomain, &daddr);
-	amdgpu_bo_unreserve(dobj);
-	if (r) {
+	r = amdgpu_bo_create_kernel(adev, size,
+				    PAGE_SIZE, ddomain,
+				    &dobj,
+				    &daddr,
+				    NULL);
+	if (r)
 		goto out_cleanup;
-	}
 
 	if (adev->mman.buffer_funcs) {
-		time = amdgpu_benchmark_do_move(adev, size, saddr, daddr, n);
-		if (time < 0)
+		r = amdgpu_benchmark_do_move(adev, size, saddr, daddr, n, &time_ms);
+		if (r)
 			goto out_cleanup;
-		if (time > 0)
-			amdgpu_benchmark_log_results(n, size, time,
+		else
+			amdgpu_benchmark_log_results(adev, n, size, time_ms,
 						     sdomain, ddomain, "dma");
 	}
 
 out_cleanup:
 	/* Check error value now. The value can be overwritten when clean up.*/
-	if (r) {
-		DRM_ERROR("Error while benchmarking BO move.\n");
-	}
+	if (r < 0)
+		dev_info(adev->dev, "Error while benchmarking BO move.\n");
 
-	if (sobj) {
-		r = amdgpu_bo_reserve(sobj, true);
-		if (likely(r == 0)) {
-			amdgpu_bo_unpin(sobj);
-			amdgpu_bo_unreserve(sobj);
-		}
-		amdgpu_bo_unref(&sobj);
-	}
-	if (dobj) {
-		r = amdgpu_bo_reserve(dobj, true);
-		if (likely(r == 0)) {
-			amdgpu_bo_unpin(dobj);
-			amdgpu_bo_unreserve(dobj);
-		}
-		amdgpu_bo_unref(&dobj);
-	}
+	if (sobj)
+		amdgpu_bo_free_kernel(&sobj, &saddr, NULL);
+	if (dobj)
+		amdgpu_bo_free_kernel(&dobj, &daddr, NULL);
+	return r;
 }
 
-void amdgpu_benchmark(struct amdgpu_device *adev, int test_number)
+int amdgpu_benchmark(struct amdgpu_device *adev, int test_number)
 {
-	int i;
+	int i, r;
 	static const int common_modes[AMDGPU_BENCHMARK_COMMON_MODES_N] = {
 		640 * 480 * 4,
 		720 * 480 * 4,
@@ -163,63 +142,119 @@ void amdgpu_benchmark(struct amdgpu_device *adev, int test_number)
 		1920 * 1200 * 4
 	};
 
+	mutex_lock(&adev->benchmark_mutex);
 	switch (test_number) {
 	case 1:
+		dev_info(adev->dev,
+			 "benchmark test: %d (simple test, VRAM to GTT and GTT to VRAM)\n",
+			 test_number);
 		/* simple test, VRAM to GTT and GTT to VRAM */
-		amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_GTT,
-				      AMDGPU_GEM_DOMAIN_VRAM);
-		amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
-				      AMDGPU_GEM_DOMAIN_GTT);
+		r = amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_GTT,
+					  AMDGPU_GEM_DOMAIN_VRAM);
+		if (r)
+			goto done;
+		r = amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
+					  AMDGPU_GEM_DOMAIN_GTT);
+		if (r)
+			goto done;
 		break;
 	case 2:
+		dev_info(adev->dev,
+			 "benchmark test: %d (simple test, VRAM to VRAM)\n",
+			 test_number);
 		/* simple test, VRAM to VRAM */
-		amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
-				      AMDGPU_GEM_DOMAIN_VRAM);
+		r = amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
+					  AMDGPU_GEM_DOMAIN_VRAM);
+		if (r)
+			goto done;
 		break;
 	case 3:
+		dev_info(adev->dev,
+			 "benchmark test: %d (GTT to VRAM, buffer size sweep, powers of 2)\n",
+			 test_number);
 		/* GTT to VRAM, buffer size sweep, powers of 2 */
-		for (i = 1; i <= 16384; i <<= 1)
-			amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
-					      AMDGPU_GEM_DOMAIN_GTT,
-					      AMDGPU_GEM_DOMAIN_VRAM);
+		for (i = 1; i <= 16384; i <<= 1) {
+			r = amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
+						  AMDGPU_GEM_DOMAIN_GTT,
+						  AMDGPU_GEM_DOMAIN_VRAM);
+			if (r)
+				goto done;
+		}
 		break;
 	case 4:
+		dev_info(adev->dev,
+			 "benchmark test: %d (VRAM to GTT, buffer size sweep, powers of 2)\n",
+			 test_number);
 		/* VRAM to GTT, buffer size sweep, powers of 2 */
-		for (i = 1; i <= 16384; i <<= 1)
-			amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
-					      AMDGPU_GEM_DOMAIN_VRAM,
-					      AMDGPU_GEM_DOMAIN_GTT);
+		for (i = 1; i <= 16384; i <<= 1) {
+			r = amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
+						  AMDGPU_GEM_DOMAIN_VRAM,
+						  AMDGPU_GEM_DOMAIN_GTT);
+			if (r)
+				goto done;
+		}
 		break;
 	case 5:
+		dev_info(adev->dev,
+			 "benchmark test: %d (VRAM to VRAM, buffer size sweep, powers of 2)\n",
+			 test_number);
 		/* VRAM to VRAM, buffer size sweep, powers of 2 */
-		for (i = 1; i <= 16384; i <<= 1)
-			amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
-					      AMDGPU_GEM_DOMAIN_VRAM,
-					      AMDGPU_GEM_DOMAIN_VRAM);
+		for (i = 1; i <= 16384; i <<= 1) {
+			r = amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
+						  AMDGPU_GEM_DOMAIN_VRAM,
+						  AMDGPU_GEM_DOMAIN_VRAM);
+			if (r)
+				goto done;
+		}
 		break;
 	case 6:
+		dev_info(adev->dev,
+			 "benchmark test: %d (GTT to VRAM, buffer size sweep, common modes)\n",
+			 test_number);
 		/* GTT to VRAM, buffer size sweep, common modes */
-		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
-			amdgpu_benchmark_move(adev, common_modes[i],
-					      AMDGPU_GEM_DOMAIN_GTT,
-					      AMDGPU_GEM_DOMAIN_VRAM);
+		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) {
+			r = amdgpu_benchmark_move(adev, common_modes[i],
+						  AMDGPU_GEM_DOMAIN_GTT,
+						  AMDGPU_GEM_DOMAIN_VRAM);
+			if (r)
+				goto done;
+		}
 		break;
 	case 7:
+		dev_info(adev->dev,
+			 "benchmark test: %d (VRAM to GTT, buffer size sweep, common modes)\n",
+			 test_number);
 		/* VRAM to GTT, buffer size sweep, common modes */
-		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
-			amdgpu_benchmark_move(adev, common_modes[i],
-					      AMDGPU_GEM_DOMAIN_VRAM,
-					      AMDGPU_GEM_DOMAIN_GTT);
+		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) {
+			r = amdgpu_benchmark_move(adev, common_modes[i],
+						  AMDGPU_GEM_DOMAIN_VRAM,
+						  AMDGPU_GEM_DOMAIN_GTT);
+			if (r)
+				goto done;
+		}
 		break;
 	case 8:
+		dev_info(adev->dev,
+			 "benchmark test: %d (VRAM to VRAM, buffer size sweep, common modes)\n",
+			 test_number);
 		/* VRAM to VRAM, buffer size sweep, common modes */
-		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
-			amdgpu_benchmark_move(adev, common_modes[i],
+		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) {
+			r = amdgpu_benchmark_move(adev, common_modes[i],
 					      AMDGPU_GEM_DOMAIN_VRAM,
 					      AMDGPU_GEM_DOMAIN_VRAM);
+			if (r)
+				goto done;
+		}
 		break;
 
 	default:
-		DRM_ERROR("Unknown benchmark\n");
+		dev_info(adev->dev, "Unknown benchmark %d\n", test_number);
+		r = -EINVAL;
+		break;
 	}
+
+done:
+	mutex_unlock(&adev->benchmark_mutex);
+
+	return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c
index 365e735f6647..35d04e69aec0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c
@@ -25,10 +25,12 @@
  *          Alex Deucher
  *          Jerome Glisse
  */
-#include <drm/drmP.h>
+
 #include "amdgpu.h"
 #include "atom.h"
 
+#include <linux/device.h>
+#include <linux/pci.h>
 #include <linux/slab.h>
 #include <linux/acpi.h>
 /*
@@ -45,58 +47,48 @@
 /* Check if current bios is an ATOM BIOS.
  * Return true if it is ATOM BIOS. Otherwise, return false.
  */
-static bool check_atom_bios(uint8_t *bios, size_t size)
+static bool check_atom_bios(struct amdgpu_device *adev, size_t size)
 {
 	uint16_t tmp, bios_header_start;
+	uint8_t *bios = adev->bios;
 
 	if (!bios || size < 0x49) {
-		DRM_INFO("vbios mem is null or mem size is wrong\n");
+		dev_dbg(adev->dev, "VBIOS mem is null or mem size is wrong\n");
 		return false;
 	}
 
 	if (!AMD_IS_VALID_VBIOS(bios)) {
-		DRM_INFO("BIOS signature incorrect %x %x\n", bios[0], bios[1]);
-		return false;
-	}
-
-	tmp = bios[0x18] | (bios[0x19] << 8);
-	if (bios[tmp + 0x14] != 0x0) {
-		DRM_INFO("Not an x86 BIOS ROM\n");
+		dev_dbg(adev->dev, "VBIOS signature incorrect %x %x\n", bios[0],
+			bios[1]);
 		return false;
 	}
 
 	bios_header_start = bios[0x48] | (bios[0x49] << 8);
 	if (!bios_header_start) {
-		DRM_INFO("Can't locate bios header\n");
+		dev_dbg(adev->dev, "Can't locate VBIOS header\n");
 		return false;
 	}
 
 	tmp = bios_header_start + 4;
 	if (size < tmp) {
-		DRM_INFO("BIOS header is broken\n");
+		dev_dbg(adev->dev, "VBIOS header is broken\n");
 		return false;
 	}
 
 	if (!memcmp(bios + tmp, "ATOM", 4) ||
 	    !memcmp(bios + tmp, "MOTA", 4)) {
-		DRM_DEBUG("ATOMBIOS detected\n");
+		dev_dbg(adev->dev, "ATOMBIOS detected\n");
 		return true;
 	}
 
 	return false;
 }
 
-static bool is_atom_fw(uint8_t *bios)
+void amdgpu_bios_release(struct amdgpu_device *adev)
 {
-	uint16_t bios_header_start = bios[0x48] | (bios[0x49] << 8);
-	uint8_t frev = bios[bios_header_start + 2];
-	uint8_t crev = bios[bios_header_start + 3];
-
-	if ((frev < 3) ||
-	    ((frev == 3) && (crev < 3)))
-		return false;
-
-	return true;
+	kfree(adev->bios);
+	adev->bios = NULL;
+	adev->bios_size = 0;
 }
 
 /* If you boot an IGP board with a discrete card as the primary,
@@ -104,35 +96,55 @@ static bool is_atom_fw(uint8_t *bios)
  * part of the system bios.  On boot, the system bios puts a
  * copy of the igp rom at the start of vram if a discrete card is
  * present.
+ * For SR-IOV, if dynamic critical region is not enabled,
+ * the vbios image is also put at the start of VRAM in the VF.
  */
-static bool igp_read_bios_from_vram(struct amdgpu_device *adev)
+static bool amdgpu_read_bios_from_vram(struct amdgpu_device *adev)
 {
-	uint8_t __iomem *bios;
+	uint8_t __iomem *bios = NULL;
 	resource_size_t vram_base;
-	resource_size_t size = 256 * 1024; /* ??? */
+	u32 size = 256U * 1024U; /* ??? */
 
 	if (!(adev->flags & AMD_IS_APU))
-		if (amdgpu_need_post(adev))
+		if (amdgpu_device_need_post(adev))
 			return false;
 
+	/* FB BAR not enabled */
+	if (pci_resource_len(adev->pdev, 0) == 0)
+		return false;
+
 	adev->bios = NULL;
 	vram_base = pci_resource_start(adev->pdev, 0);
-	bios = ioremap(vram_base, size);
-	if (!bios) {
-		return false;
-	}
 
 	adev->bios = kmalloc(size, GFP_KERNEL);
-	if (!adev->bios) {
-		iounmap(bios);
+	if (!adev->bios)
 		return false;
+
+	/* For SRIOV with dynamic critical region is enabled,
+	 * the vbios image is put at a dynamic offset of VRAM in the VF.
+	 * If dynamic critical region is disabled, follow the existing logic as on baremetal.
+	 */
+	if (amdgpu_sriov_vf(adev) && adev->virt.is_dynamic_crit_regn_enabled) {
+		if (amdgpu_virt_get_dynamic_data_info(adev,
+				AMD_SRIOV_MSG_VBIOS_IMG_TABLE_ID, adev->bios, &size)) {
+			amdgpu_bios_release(adev);
+			return false;
+		}
+	} else {
+		bios = ioremap_wc(vram_base, size);
+		if (!bios) {
+			amdgpu_bios_release(adev);
+			return false;
+		}
+
+		memcpy_fromio(adev->bios, bios, size);
+		iounmap(bios);
 	}
+
 	adev->bios_size = size;
-	memcpy_fromio(adev->bios, bios, size);
-	iounmap(bios);
 
-	if (!check_atom_bios(adev->bios, size)) {
-		kfree(adev->bios);
+	if (!check_atom_bios(adev, size)) {
+		amdgpu_bios_release(adev);
 		return false;
 	}
 
@@ -147,9 +159,8 @@ bool amdgpu_read_bios(struct amdgpu_device *adev)
 	adev->bios = NULL;
 	/* XXX: some cards may return 0 for rom size? ddx has a workaround */
 	bios = pci_map_rom(adev->pdev, &size);
-	if (!bios) {
+	if (!bios)
 		return false;
-	}
 
 	adev->bios = kzalloc(size, GFP_KERNEL);
 	if (adev->bios == NULL) {
@@ -160,8 +171,8 @@ bool amdgpu_read_bios(struct amdgpu_device *adev)
 	memcpy_fromio(adev->bios, bios, size);
 	pci_unmap_rom(adev->pdev, bios);
 
-	if (!check_atom_bios(adev->bios, size)) {
-		kfree(adev->bios);
+	if (!check_atom_bios(adev, size)) {
+		amdgpu_bios_release(adev);
 		return false;
 	}
 
@@ -173,7 +184,7 @@ static bool amdgpu_read_bios_from_rom(struct amdgpu_device *adev)
 	u8 header[AMD_VBIOS_SIGNATURE_END+1] = {0};
 	int len;
 
-	if (!adev->asic_funcs->read_bios_from_rom)
+	if (!adev->asic_funcs || !adev->asic_funcs->read_bios_from_rom)
 		return false;
 
 	/* validate VBIOS signature */
@@ -182,9 +193,9 @@ static bool amdgpu_read_bios_from_rom(struct amdgpu_device *adev)
 	header[AMD_VBIOS_SIGNATURE_END] = 0;
 
 	if ((!AMD_IS_VALID_VBIOS(header)) ||
-	    0 != memcmp((char *)&header[AMD_VBIOS_SIGNATURE_OFFSET],
-			AMD_VBIOS_SIGNATURE,
-			strlen(AMD_VBIOS_SIGNATURE)))
+		memcmp((char *)&header[AMD_VBIOS_SIGNATURE_OFFSET],
+		       AMD_VBIOS_SIGNATURE,
+		       strlen(AMD_VBIOS_SIGNATURE)) != 0)
 		return false;
 
 	/* valid vbios, go on */
@@ -200,8 +211,8 @@ static bool amdgpu_read_bios_from_rom(struct amdgpu_device *adev)
 	/* read complete BIOS */
 	amdgpu_asic_read_bios_from_rom(adev, adev->bios, len);
 
-	if (!check_atom_bios(adev->bios, len)) {
-		kfree(adev->bios);
+	if (!check_atom_bios(adev, len)) {
+		amdgpu_bios_release(adev);
 		return false;
 	}
 
@@ -210,30 +221,36 @@ static bool amdgpu_read_bios_from_rom(struct amdgpu_device *adev)
 
 static bool amdgpu_read_platform_bios(struct amdgpu_device *adev)
 {
-	uint8_t __iomem *bios;
-	size_t size;
+	phys_addr_t rom = adev->pdev->rom;
+	size_t romlen = adev->pdev->romlen;
+	void __iomem *bios;
 
 	adev->bios = NULL;
 
-	bios = pci_platform_rom(adev->pdev, &size);
-	if (!bios) {
+	if (!rom || romlen == 0)
 		return false;
-	}
 
-	adev->bios = kzalloc(size, GFP_KERNEL);
-	if (adev->bios == NULL)
+	adev->bios = kzalloc(romlen, GFP_KERNEL);
+	if (!adev->bios)
 		return false;
 
-	memcpy_fromio(adev->bios, bios, size);
+	bios = ioremap(rom, romlen);
+	if (!bios)
+		goto free_bios;
 
-	if (!check_atom_bios(adev->bios, size)) {
-		kfree(adev->bios);
-		return false;
-	}
+	memcpy_fromio(adev->bios, bios, romlen);
+	iounmap(bios);
 
-	adev->bios_size = size;
+	if (!check_atom_bios(adev, romlen))
+		goto free_bios;
+
+	adev->bios_size = romlen;
 
 	return true;
+free_bios:
+	amdgpu_bios_release(adev);
+
+	return false;
 }
 
 #ifdef CONFIG_ACPI
@@ -273,7 +290,7 @@ static int amdgpu_atrm_call(acpi_handle atrm_handle, uint8_t *bios,
 
 	status = acpi_evaluate_object(atrm_handle, NULL, &atrm_arg, &buffer);
 	if (ACPI_FAILURE(status)) {
-		printk("failed to evaluate ATRM got %s\n", acpi_format_exception(status));
+		DRM_ERROR("failed to evaluate ATRM got %s\n", acpi_format_exception(status));
 		return -ENODEV;
 	}
 
@@ -294,42 +311,33 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
 	acpi_status status;
 	bool found = false;
 
-	/* ATRM is for the discrete card only */
-	if (adev->flags & AMD_IS_APU)
+	/* ATRM is for on-platform devices only */
+	if (dev_is_removable(&adev->pdev->dev))
 		return false;
 
-	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL) {
+	while ((pdev = pci_get_base_class(PCI_BASE_CLASS_DISPLAY, pdev))) {
+		if ((pdev->class != PCI_CLASS_DISPLAY_VGA << 8) &&
+		    (pdev->class != PCI_CLASS_DISPLAY_OTHER << 8))
+			continue;
+
 		dhandle = ACPI_HANDLE(&pdev->dev);
 		if (!dhandle)
 			continue;
 
 		status = acpi_get_handle(dhandle, "ATRM", &atrm_handle);
-		if (!ACPI_FAILURE(status)) {
+		if (ACPI_SUCCESS(status)) {
 			found = true;
 			break;
 		}
 	}
 
-	if (!found) {
-		while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_OTHER << 8, pdev)) != NULL) {
-			dhandle = ACPI_HANDLE(&pdev->dev);
-			if (!dhandle)
-				continue;
-
-			status = acpi_get_handle(dhandle, "ATRM", &atrm_handle);
-			if (!ACPI_FAILURE(status)) {
-				found = true;
-				break;
-			}
-		}
-	}
-
 	if (!found)
 		return false;
+	pci_dev_put(pdev);
 
 	adev->bios = kmalloc(size, GFP_KERNEL);
 	if (!adev->bios) {
-		DRM_ERROR("Unable to allocate bios\n");
+		dev_err(adev->dev, "Unable to allocate bios\n");
 		return false;
 	}
 
@@ -342,8 +350,8 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
 			break;
 	}
 
-	if (!check_atom_bios(adev->bios, size)) {
-		kfree(adev->bios);
+	if (!check_atom_bios(adev, size)) {
+		amdgpu_bios_release(adev);
 		return false;
 	}
 	adev->bios_size = size;
@@ -358,10 +366,8 @@ static inline bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
 
 static bool amdgpu_read_disabled_bios(struct amdgpu_device *adev)
 {
-	if (adev->flags & AMD_IS_APU)
-		return igp_read_bios_from_vram(adev);
-	else
-		return amdgpu_asic_read_disabled_bios(adev);
+	return (!adev->asic_funcs || !adev->asic_funcs->read_disabled_bios) ?
+		false : amdgpu_asic_read_disabled_bios(adev);
 }
 
 #ifdef CONFIG_ACPI
@@ -370,13 +376,13 @@ static bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev)
 	struct acpi_table_header *hdr;
 	acpi_size tbl_size;
 	UEFI_ACPI_VFCT *vfct;
-	unsigned offset;
+	unsigned int offset;
 
 	if (!ACPI_SUCCESS(acpi_get_table("VFCT", 1, &hdr)))
 		return false;
 	tbl_size = hdr->length;
 	if (tbl_size < sizeof(UEFI_ACPI_VFCT)) {
-		DRM_ERROR("ACPI VFCT table present but broken (too short #1)\n");
+		dev_info(adev->dev, "ACPI VFCT table present but broken (too short #1),skipping\n");
 		return false;
 	}
 
@@ -389,13 +395,13 @@ static bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev)
 
 		offset += sizeof(VFCT_IMAGE_HEADER);
 		if (offset > tbl_size) {
-			DRM_ERROR("ACPI VFCT image header truncated\n");
+			dev_info(adev->dev, "ACPI VFCT image header truncated,skipping\n");
 			return false;
 		}
 
 		offset += vhdr->ImageLength;
 		if (offset > tbl_size) {
-			DRM_ERROR("ACPI VFCT image truncated\n");
+			dev_info(adev->dev, "ACPI VFCT image truncated,skipping\n");
 			return false;
 		}
 
@@ -409,8 +415,8 @@ static bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev)
 					     vhdr->ImageLength,
 					     GFP_KERNEL);
 
-			if (!check_atom_bios(adev->bios, vhdr->ImageLength)) {
-				kfree(adev->bios);
+			if (!check_atom_bios(adev, vhdr->ImageLength)) {
+				amdgpu_bios_release(adev);
 				return false;
 			}
 			adev->bios_size = vhdr->ImageLength;
@@ -418,7 +424,7 @@ static bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev)
 		}
 	}
 
-	DRM_ERROR("ACPI VFCT table present but broken (too short #2)\n");
+	dev_info(adev->dev, "ACPI VFCT table present but broken (too short #2),skipping\n");
 	return false;
 }
 #else
@@ -428,33 +434,158 @@ static inline bool amdgpu_acpi_vfct_bios(struct amdgpu_device *adev)
 }
 #endif
 
-bool amdgpu_get_bios(struct amdgpu_device *adev)
+static bool amdgpu_get_bios_apu(struct amdgpu_device *adev)
 {
-	if (amdgpu_atrm_get_bios(adev))
+	if (amdgpu_acpi_vfct_bios(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from VFCT\n");
+		goto success;
+	}
+
+	if (amdgpu_read_bios_from_vram(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from VRAM BAR\n");
+		goto success;
+	}
+
+	if (amdgpu_read_bios(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from ROM BAR\n");
 		goto success;
+	}
 
-	if (amdgpu_acpi_vfct_bios(adev))
+	if (amdgpu_read_platform_bios(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from platform\n");
 		goto success;
+	}
+
+	dev_err(adev->dev, "Unable to locate a BIOS ROM\n");
+	return false;
+
+success:
+	return true;
+}
+
+static bool amdgpu_prefer_rom_resource(struct amdgpu_device *adev)
+{
+	struct resource *res = &adev->pdev->resource[PCI_ROM_RESOURCE];
+
+	return (res->flags & IORESOURCE_ROM_SHADOW);
+}
 
-	if (igp_read_bios_from_vram(adev))
+static bool amdgpu_get_bios_dgpu(struct amdgpu_device *adev)
+{
+	if (amdgpu_atrm_get_bios(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from ATRM\n");
 		goto success;
+	}
 
-	if (amdgpu_read_bios(adev))
+	if (amdgpu_acpi_vfct_bios(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from VFCT\n");
 		goto success;
+	}
 
-	if (amdgpu_read_bios_from_rom(adev))
+	/* this is required for SR-IOV */
+	if (amdgpu_read_bios_from_vram(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from VRAM BAR\n");
 		goto success;
+	}
+
+	if (amdgpu_prefer_rom_resource(adev)) {
+		if (amdgpu_read_bios(adev)) {
+			dev_info(adev->dev, "Fetched VBIOS from ROM BAR\n");
+			goto success;
+		}
+
+		if (amdgpu_read_platform_bios(adev)) {
+			dev_info(adev->dev, "Fetched VBIOS from platform\n");
+			goto success;
+		}
+
+	} else {
+		if (amdgpu_read_platform_bios(adev)) {
+			dev_info(adev->dev, "Fetched VBIOS from platform\n");
+			goto success;
+		}
+
+		if (amdgpu_read_bios(adev)) {
+			dev_info(adev->dev, "Fetched VBIOS from ROM BAR\n");
+			goto success;
+		}
+	}
 
-	if (amdgpu_read_disabled_bios(adev))
+	if (amdgpu_read_bios_from_rom(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from ROM\n");
 		goto success;
+	}
 
-	if (amdgpu_read_platform_bios(adev))
+	if (amdgpu_read_disabled_bios(adev)) {
+		dev_info(adev->dev, "Fetched VBIOS from disabled ROM BAR\n");
 		goto success;
+	}
 
-	DRM_ERROR("Unable to locate a BIOS ROM\n");
+	dev_err(adev->dev, "Unable to locate a BIOS ROM\n");
 	return false;
 
 success:
-	adev->is_atom_fw = is_atom_fw(adev->bios);
+	return true;
+}
+
+bool amdgpu_get_bios(struct amdgpu_device *adev)
+{
+	bool found;
+
+	if (adev->flags & AMD_IS_APU)
+		found = amdgpu_get_bios_apu(adev);
+	else
+		found = amdgpu_get_bios_dgpu(adev);
+
+	if (found)
+		adev->is_atom_fw = adev->asic_type >= CHIP_VEGA10;
+
+	return found;
+}
+
+/* helper function for soc15 and onwards to read bios from rom */
+bool amdgpu_soc15_read_bios_from_rom(struct amdgpu_device *adev,
+				     u8 *bios, u32 length_bytes)
+{
+	u32 *dw_ptr;
+	u32 i, length_dw;
+	u32 rom_offset;
+	u32 rom_index_offset;
+	u32 rom_data_offset;
+
+	if (bios == NULL)
+		return false;
+	if (length_bytes == 0)
+		return false;
+	/* APU vbios image is part of sbios image */
+	if (adev->flags & AMD_IS_APU)
+		return false;
+	if (!adev->smuio.funcs ||
+	    !adev->smuio.funcs->get_rom_index_offset ||
+	    !adev->smuio.funcs->get_rom_data_offset)
+		return false;
+
+	dw_ptr = (u32 *)bios;
+	length_dw = ALIGN(length_bytes, 4) / 4;
+
+	rom_index_offset =
+		adev->smuio.funcs->get_rom_index_offset(adev);
+	rom_data_offset =
+		adev->smuio.funcs->get_rom_data_offset(adev);
+
+	if (adev->nbio.funcs &&
+	    adev->nbio.funcs->get_rom_offset) {
+		rom_offset = adev->nbio.funcs->get_rom_offset(adev);
+		rom_offset = rom_offset << 17;
+	} else {
+		rom_offset = 0;
+	}
+
+	/* set rom index to rom_offset */
+	WREG32(rom_index_offset, rom_offset);
+	/* read out the rom data */
+	for (i = 0; i < length_dw; i++)
+		dw_ptr[i] = RREG32(rom_data_offset);
+
 	return true;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index a6649874e6ce..66fb37b64388 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -28,78 +28,64 @@
  *    Christian König <deathsimple@vodafone.de>
  */
 
-#include <drm/drmP.h>
+#include <linux/sort.h>
+#include <linux/uaccess.h>
+
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
 
 #define AMDGPU_BO_LIST_MAX_PRIORITY	32u
 #define AMDGPU_BO_LIST_NUM_BUCKETS	(AMDGPU_BO_LIST_MAX_PRIORITY + 1)
 
-static int amdgpu_bo_list_create(struct amdgpu_fpriv *fpriv,
-				 struct amdgpu_bo_list **result,
-				 int *id)
+static void amdgpu_bo_list_free_rcu(struct rcu_head *rcu)
 {
-	int r;
-
-	*result = kzalloc(sizeof(struct amdgpu_bo_list), GFP_KERNEL);
-	if (!*result)
-		return -ENOMEM;
-
-	mutex_lock(&fpriv->bo_list_lock);
-	r = idr_alloc(&fpriv->bo_list_handles, *result,
-		      1, 0, GFP_KERNEL);
-	if (r < 0) {
-		mutex_unlock(&fpriv->bo_list_lock);
-		kfree(*result);
-		return r;
-	}
-	*id = r;
-
-	mutex_init(&(*result)->lock);
-	(*result)->num_entries = 0;
-	(*result)->array = NULL;
+	struct amdgpu_bo_list *list = container_of(rcu, struct amdgpu_bo_list,
+						   rhead);
+	mutex_destroy(&list->bo_list_mutex);
+	kvfree(list);
+}
 
-	mutex_lock(&(*result)->lock);
-	mutex_unlock(&fpriv->bo_list_lock);
+static void amdgpu_bo_list_free(struct kref *ref)
+{
+	struct amdgpu_bo_list *list = container_of(ref, struct amdgpu_bo_list,
+						   refcount);
+	struct amdgpu_bo_list_entry *e;
 
-	return 0;
+	amdgpu_bo_list_for_each_entry(e, list)
+		amdgpu_bo_unref(&e->bo);
+	call_rcu(&list->rhead, amdgpu_bo_list_free_rcu);
 }
 
-static void amdgpu_bo_list_destroy(struct amdgpu_fpriv *fpriv, int id)
+static int amdgpu_bo_list_entry_cmp(const void *_a, const void *_b)
 {
-	struct amdgpu_bo_list *list;
+	const struct amdgpu_bo_list_entry *a = _a, *b = _b;
 
-	mutex_lock(&fpriv->bo_list_lock);
-	list = idr_remove(&fpriv->bo_list_handles, id);
-	if (list) {
-		/* Another user may have a reference to this list still */
-		mutex_lock(&list->lock);
-		mutex_unlock(&list->lock);
-		amdgpu_bo_list_free(list);
-	}
-	mutex_unlock(&fpriv->bo_list_lock);
+	if (a->priority > b->priority)
+		return 1;
+	if (a->priority < b->priority)
+		return -1;
+	return 0;
 }
 
-static int amdgpu_bo_list_set(struct amdgpu_device *adev,
-				     struct drm_file *filp,
-				     struct amdgpu_bo_list *list,
-				     struct drm_amdgpu_bo_list_entry *info,
-				     unsigned num_entries)
+int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
+			  struct drm_amdgpu_bo_list_entry *info,
+			  size_t num_entries, struct amdgpu_bo_list **result)
 {
-	struct amdgpu_bo_list_entry *array;
-	struct amdgpu_bo *gds_obj = adev->gds.gds_gfx_bo;
-	struct amdgpu_bo *gws_obj = adev->gds.gws_gfx_bo;
-	struct amdgpu_bo *oa_obj = adev->gds.oa_gfx_bo;
-
 	unsigned last_entry = 0, first_userptr = num_entries;
+	struct amdgpu_bo_list_entry *array;
+	struct amdgpu_bo_list *list;
+	uint64_t total_size = 0;
 	unsigned i;
 	int r;
-	unsigned long total_size = 0;
 
-	array = drm_malloc_ab(num_entries, sizeof(struct amdgpu_bo_list_entry));
-	if (!array)
+	list = kvzalloc(struct_size(list, entries, num_entries), GFP_KERNEL);
+	if (!list)
 		return -ENOMEM;
-	memset(array, 0, num_entries * sizeof(struct amdgpu_bo_list_entry));
+
+	kref_init(&list->refcount);
+
+	list->num_entries = num_entries;
+	array = list->entries;
 
 	for (i = 0; i < num_entries; ++i) {
 		struct amdgpu_bo_list_entry *entry;
@@ -114,7 +100,7 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 		}
 
 		bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
-		drm_gem_object_unreference_unlocked(gobj);
+		drm_gem_object_put(gobj);
 
 		usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
 		if (usermm) {
@@ -128,161 +114,138 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev,
 			entry = &array[last_entry++];
 		}
 
-		entry->robj = bo;
 		entry->priority = min(info[i].bo_priority,
 				      AMDGPU_BO_LIST_MAX_PRIORITY);
-		entry->tv.bo = &entry->robj->tbo;
-		entry->tv.shared = !entry->robj->prime_shared_count;
-
-		if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GDS)
-			gds_obj = entry->robj;
-		if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GWS)
-			gws_obj = entry->robj;
-		if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_OA)
-			oa_obj = entry->robj;
-
-		total_size += amdgpu_bo_size(entry->robj);
-		trace_amdgpu_bo_list_set(list, entry->robj);
-	}
+		entry->bo = bo;
 
-	for (i = 0; i < list->num_entries; ++i)
-		amdgpu_bo_unref(&list->array[i].robj);
+		if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GDS)
+			list->gds_obj = bo;
+		if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_GWS)
+			list->gws_obj = bo;
+		if (bo->preferred_domains == AMDGPU_GEM_DOMAIN_OA)
+			list->oa_obj = bo;
 
-	drm_free_large(list->array);
+		total_size += amdgpu_bo_size(bo);
+		trace_amdgpu_bo_list_set(list, bo);
+	}
 
-	list->gds_obj = gds_obj;
-	list->gws_obj = gws_obj;
-	list->oa_obj = oa_obj;
 	list->first_userptr = first_userptr;
-	list->array = array;
-	list->num_entries = num_entries;
+	sort(array, last_entry, sizeof(struct amdgpu_bo_list_entry),
+	     amdgpu_bo_list_entry_cmp, NULL);
 
 	trace_amdgpu_cs_bo_status(list->num_entries, total_size);
+
+	mutex_init(&list->bo_list_mutex);
+	*result = list;
 	return 0;
 
 error_free:
-	while (i--)
-		amdgpu_bo_unref(&array[i].robj);
-	drm_free_large(array);
+	for (i = 0; i < last_entry; ++i)
+		amdgpu_bo_unref(&array[i].bo);
+	for (i = first_userptr; i < num_entries; ++i)
+		amdgpu_bo_unref(&array[i].bo);
+	kvfree(list);
 	return r;
+
 }
 
-struct amdgpu_bo_list *
-amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id)
+static void amdgpu_bo_list_destroy(struct amdgpu_fpriv *fpriv, int id)
 {
-	struct amdgpu_bo_list *result;
+	struct amdgpu_bo_list *list;
 
 	mutex_lock(&fpriv->bo_list_lock);
-	result = idr_find(&fpriv->bo_list_handles, id);
-	if (result)
-		mutex_lock(&result->lock);
+	list = idr_remove(&fpriv->bo_list_handles, id);
 	mutex_unlock(&fpriv->bo_list_lock);
-	return result;
+	if (list)
+		kref_put(&list->refcount, amdgpu_bo_list_free);
 }
 
-void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
-			     struct list_head *validated)
+int amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id,
+		       struct amdgpu_bo_list **result)
 {
-	/* This is based on the bucket sort with O(n) time complexity.
-	 * An item with priority "i" is added to bucket[i]. The lists are then
-	 * concatenated in descending order.
-	 */
-	struct list_head bucket[AMDGPU_BO_LIST_NUM_BUCKETS];
-	unsigned i;
-
-	for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
-		INIT_LIST_HEAD(&bucket[i]);
-
-	/* Since buffers which appear sooner in the relocation list are
-	 * likely to be used more often than buffers which appear later
-	 * in the list, the sort mustn't change the ordering of buffers
-	 * with the same priority, i.e. it must be stable.
-	 */
-	for (i = 0; i < list->num_entries; i++) {
-		unsigned priority = list->array[i].priority;
+	rcu_read_lock();
+	*result = idr_find(&fpriv->bo_list_handles, id);
 
-		list_add_tail(&list->array[i].tv.head,
-			      &bucket[priority]);
-		list->array[i].user_pages = NULL;
+	if (*result && kref_get_unless_zero(&(*result)->refcount)) {
+		rcu_read_unlock();
+		return 0;
 	}
 
-	/* Connect the sorted buckets in the output list. */
-	for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++)
-		list_splice(&bucket[i], validated);
+	rcu_read_unlock();
+	*result = NULL;
+	return -ENOENT;
 }
 
 void amdgpu_bo_list_put(struct amdgpu_bo_list *list)
 {
-	mutex_unlock(&list->lock);
+	kref_put(&list->refcount, amdgpu_bo_list_free);
 }
 
-void amdgpu_bo_list_free(struct amdgpu_bo_list *list)
+int amdgpu_bo_create_list_entry_array(struct drm_amdgpu_bo_list_in *in,
+				      struct drm_amdgpu_bo_list_entry **info_param)
 {
-	unsigned i;
+	const uint32_t info_size = sizeof(struct drm_amdgpu_bo_list_entry);
+	const void __user *uptr = u64_to_user_ptr(in->bo_info_ptr);
+	const uint32_t bo_info_size = in->bo_info_size;
+	const uint32_t bo_number = in->bo_number;
+	struct drm_amdgpu_bo_list_entry *info;
+
+	/* copy the handle array from userspace to a kernel buffer */
+	if (likely(info_size == bo_info_size)) {
+		info = vmemdup_array_user(uptr, bo_number, info_size);
+		if (IS_ERR(info))
+			return PTR_ERR(info);
+	} else {
+		const uint32_t bytes = min(bo_info_size, info_size);
+		unsigned i;
 
-	for (i = 0; i < list->num_entries; ++i)
-		amdgpu_bo_unref(&list->array[i].robj);
+		info = kvmalloc_array(bo_number, info_size, GFP_KERNEL);
+		if (!info)
+			return -ENOMEM;
 
-	mutex_destroy(&list->lock);
-	drm_free_large(list->array);
-	kfree(list);
+		memset(info, 0, bo_number * info_size);
+		for (i = 0; i < bo_number; ++i, uptr += bo_info_size) {
+			if (copy_from_user(&info[i], uptr, bytes)) {
+				kvfree(info);
+				return -EFAULT;
+			}
+		}
+	}
+
+	*info_param = info;
+	return 0;
 }
 
 int amdgpu_bo_list_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *filp)
 {
-	const uint32_t info_size = sizeof(struct drm_amdgpu_bo_list_entry);
-
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	union drm_amdgpu_bo_list *args = data;
 	uint32_t handle = args->in.list_handle;
-	const void __user *uptr = (const void*)(uintptr_t)args->in.bo_info_ptr;
-
-	struct drm_amdgpu_bo_list_entry *info;
-	struct amdgpu_bo_list *list;
-
+	struct drm_amdgpu_bo_list_entry *info = NULL;
+	struct amdgpu_bo_list *list, *old;
 	int r;
 
-	info = drm_malloc_ab(args->in.bo_number,
-			     sizeof(struct drm_amdgpu_bo_list_entry));
-	if (!info)
-		return -ENOMEM;
-
-	/* copy the handle array from userspace to a kernel buffer */
-	r = -EFAULT;
-	if (likely(info_size == args->in.bo_info_size)) {
-		unsigned long bytes = args->in.bo_number *
-			args->in.bo_info_size;
-
-		if (copy_from_user(info, uptr, bytes))
-			goto error_free;
-
-	} else {
-		unsigned long bytes = min(args->in.bo_info_size, info_size);
-		unsigned i;
-
-		memset(info, 0, args->in.bo_number * info_size);
-		for (i = 0; i < args->in.bo_number; ++i) {
-			if (copy_from_user(&info[i], uptr, bytes))
-				goto error_free;
-
-			uptr += args->in.bo_info_size;
-		}
-	}
+	r = amdgpu_bo_create_list_entry_array(&args->in, &info);
+	if (r)
+		return r;
 
 	switch (args->in.operation) {
 	case AMDGPU_BO_LIST_OP_CREATE:
-		r = amdgpu_bo_list_create(fpriv, &list, &handle);
+		r = amdgpu_bo_list_create(adev, filp, info, args->in.bo_number,
+					  &list);
 		if (r)
 			goto error_free;
 
-		r = amdgpu_bo_list_set(adev, filp, list, info,
-					      args->in.bo_number);
-		amdgpu_bo_list_put(list);
-		if (r)
-			goto error_free;
+		mutex_lock(&fpriv->bo_list_lock);
+		r = idr_alloc(&fpriv->bo_list_handles, list, 1, 0, GFP_KERNEL);
+		mutex_unlock(&fpriv->bo_list_lock);
+		if (r < 0) {
+			goto error_put_list;
+		}
 
+		handle = r;
 		break;
 
 	case AMDGPU_BO_LIST_OP_DESTROY:
@@ -291,17 +254,21 @@ int amdgpu_bo_list_ioctl(struct drm_device *dev, void *data,
 		break;
 
 	case AMDGPU_BO_LIST_OP_UPDATE:
-		r = -ENOENT;
-		list = amdgpu_bo_list_get(fpriv, handle);
-		if (!list)
-			goto error_free;
-
-		r = amdgpu_bo_list_set(adev, filp, list, info,
-					      args->in.bo_number);
-		amdgpu_bo_list_put(list);
+		r = amdgpu_bo_list_create(adev, filp, info, args->in.bo_number,
+					  &list);
 		if (r)
 			goto error_free;
 
+		mutex_lock(&fpriv->bo_list_lock);
+		old = idr_replace(&fpriv->bo_list_handles, list, handle);
+		mutex_unlock(&fpriv->bo_list_lock);
+
+		if (IS_ERR(old)) {
+			r = PTR_ERR(old);
+			goto error_put_list;
+		}
+
+		amdgpu_bo_list_put(old);
 		break;
 
 	default:
@@ -311,11 +278,14 @@ int amdgpu_bo_list_ioctl(struct drm_device *dev, void *data,
 
 	memset(args, 0, sizeof(*args));
 	args->out.list_handle = handle;
-	drm_free_large(info);
+	kvfree(info);
 
 	return 0;
 
+error_put_list:
+	amdgpu_bo_list_put(list);
+
 error_free:
-	drm_free_large(info);
+	kvfree(info);
 	return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
new file mode 100644
index 000000000000..2b5e7c46a39d
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
@@ -0,0 +1,83 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_BO_LIST_H__
+#define __AMDGPU_BO_LIST_H__
+
+#include <drm/amdgpu_drm.h>
+
+struct hmm_range;
+
+struct drm_file;
+
+struct amdgpu_device;
+struct amdgpu_bo;
+struct amdgpu_bo_va;
+struct amdgpu_fpriv;
+
+struct amdgpu_bo_list_entry {
+	struct amdgpu_bo		*bo;
+	struct amdgpu_bo_va		*bo_va;
+	uint32_t			priority;
+	struct amdgpu_hmm_range		*range;
+	bool				user_invalidated;
+};
+
+struct amdgpu_bo_list {
+	struct rcu_head rhead;
+	struct kref refcount;
+	struct amdgpu_bo *gds_obj;
+	struct amdgpu_bo *gws_obj;
+	struct amdgpu_bo *oa_obj;
+	unsigned first_userptr;
+	unsigned num_entries;
+
+	/* Protect access during command submission.
+	 */
+	struct mutex bo_list_mutex;
+
+	struct amdgpu_bo_list_entry entries[] __counted_by(num_entries);
+};
+
+int amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id,
+		       struct amdgpu_bo_list **result);
+void amdgpu_bo_list_put(struct amdgpu_bo_list *list);
+int amdgpu_bo_create_list_entry_array(struct drm_amdgpu_bo_list_in *in,
+				      struct drm_amdgpu_bo_list_entry **info_param);
+
+int amdgpu_bo_list_create(struct amdgpu_device *adev,
+				 struct drm_file *filp,
+				 struct drm_amdgpu_bo_list_entry *info,
+				 size_t num_entries,
+				 struct amdgpu_bo_list **list);
+
+#define amdgpu_bo_list_for_each_entry(e, list) \
+	for (e = list->entries; \
+	     e != &list->entries[list->num_entries]; \
+	     ++e)
+
+#define amdgpu_bo_list_for_each_userptr_entry(e, list) \
+	for (e = &list->entries[list->first_userptr]; \
+	     e != &list->entries[list->num_entries]; \
+	     ++e)
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c
index c6dba1eaefbd..004a6a9d6b9f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c
@@ -22,14 +22,12 @@
  *
  */
 #include <linux/list.h>
-#include <linux/slab.h>
 #include <linux/pci.h>
-#include <linux/acpi.h>
-#include <drm/drmP.h>
+#include <linux/slab.h>
+
 #include <linux/firmware.h>
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
-#include "cgs_linux.h"
 #include "atom.h"
 #include "amdgpu_ucode.h"
 
@@ -42,180 +40,14 @@ struct amdgpu_cgs_device {
 	struct amdgpu_device *adev =					\
 		((struct amdgpu_cgs_device *)cgs_device)->adev
 
-static int amdgpu_cgs_alloc_gpu_mem(struct cgs_device *cgs_device,
-				    enum cgs_gpu_mem_type type,
-				    uint64_t size, uint64_t align,
-				    uint64_t min_offset, uint64_t max_offset,
-				    cgs_handle_t *handle)
-{
-	CGS_FUNC_ADEV;
-	uint16_t flags = 0;
-	int ret = 0;
-	uint32_t domain = 0;
-	struct amdgpu_bo *obj;
-	struct ttm_placement placement;
-	struct ttm_place place;
-
-	if (min_offset > max_offset) {
-		BUG_ON(1);
-		return -EINVAL;
-	}
-
-	/* fail if the alignment is not a power of 2 */
-	if (((align != 1) && (align & (align - 1)))
-	    || size == 0 || align == 0)
-		return -EINVAL;
-
-
-	switch(type) {
-	case CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB:
-	case CGS_GPU_MEM_TYPE__VISIBLE_FB:
-		flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-			AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-		domain = AMDGPU_GEM_DOMAIN_VRAM;
-		if (max_offset > adev->mc.real_vram_size)
-			return -EINVAL;
-		place.fpfn = min_offset >> PAGE_SHIFT;
-		place.lpfn = max_offset >> PAGE_SHIFT;
-		place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
-			TTM_PL_FLAG_VRAM;
-		break;
-	case CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB:
-	case CGS_GPU_MEM_TYPE__INVISIBLE_FB:
-		flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
-			AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-		domain = AMDGPU_GEM_DOMAIN_VRAM;
-		if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
-			place.fpfn =
-				max(min_offset, adev->mc.visible_vram_size) >> PAGE_SHIFT;
-			place.lpfn =
-				min(max_offset, adev->mc.real_vram_size) >> PAGE_SHIFT;
-			place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
-				TTM_PL_FLAG_VRAM;
-		}
-
-		break;
-	case CGS_GPU_MEM_TYPE__GART_CACHEABLE:
-		domain = AMDGPU_GEM_DOMAIN_GTT;
-		place.fpfn = min_offset >> PAGE_SHIFT;
-		place.lpfn = max_offset >> PAGE_SHIFT;
-		place.flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
-		break;
-	case CGS_GPU_MEM_TYPE__GART_WRITECOMBINE:
-		flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-		domain = AMDGPU_GEM_DOMAIN_GTT;
-		place.fpfn = min_offset >> PAGE_SHIFT;
-		place.lpfn = max_offset >> PAGE_SHIFT;
-		place.flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_TT |
-			TTM_PL_FLAG_UNCACHED;
-		break;
-	default:
-		return -EINVAL;
-	}
 
-
-	*handle = 0;
-
-	placement.placement = &place;
-	placement.num_placement = 1;
-	placement.busy_placement = &place;
-	placement.num_busy_placement = 1;
-
-	ret = amdgpu_bo_create_restricted(adev, size, PAGE_SIZE,
-					  true, domain, flags,
-					  NULL, &placement, NULL,
-					  &obj);
-	if (ret) {
-		DRM_ERROR("(%d) bo create failed\n", ret);
-		return ret;
-	}
-	*handle = (cgs_handle_t)obj;
-
-	return ret;
-}
-
-static int amdgpu_cgs_free_gpu_mem(struct cgs_device *cgs_device, cgs_handle_t handle)
-{
-	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
-
-	if (obj) {
-		int r = amdgpu_bo_reserve(obj, true);
-		if (likely(r == 0)) {
-			amdgpu_bo_kunmap(obj);
-			amdgpu_bo_unpin(obj);
-			amdgpu_bo_unreserve(obj);
-		}
-		amdgpu_bo_unref(&obj);
-
-	}
-	return 0;
-}
-
-static int amdgpu_cgs_gmap_gpu_mem(struct cgs_device *cgs_device, cgs_handle_t handle,
-				   uint64_t *mcaddr)
-{
-	int r;
-	u64 min_offset, max_offset;
-	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
-
-	WARN_ON_ONCE(obj->placement.num_placement > 1);
-
-	min_offset = obj->placements[0].fpfn << PAGE_SHIFT;
-	max_offset = obj->placements[0].lpfn << PAGE_SHIFT;
-
-	r = amdgpu_bo_reserve(obj, true);
-	if (unlikely(r != 0))
-		return r;
-	r = amdgpu_bo_pin_restricted(obj, obj->prefered_domains,
-				     min_offset, max_offset, mcaddr);
-	amdgpu_bo_unreserve(obj);
-	return r;
-}
-
-static int amdgpu_cgs_gunmap_gpu_mem(struct cgs_device *cgs_device, cgs_handle_t handle)
-{
-	int r;
-	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
-	r = amdgpu_bo_reserve(obj, true);
-	if (unlikely(r != 0))
-		return r;
-	r = amdgpu_bo_unpin(obj);
-	amdgpu_bo_unreserve(obj);
-	return r;
-}
-
-static int amdgpu_cgs_kmap_gpu_mem(struct cgs_device *cgs_device, cgs_handle_t handle,
-				   void **map)
-{
-	int r;
-	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
-	r = amdgpu_bo_reserve(obj, true);
-	if (unlikely(r != 0))
-		return r;
-	r = amdgpu_bo_kmap(obj, map);
-	amdgpu_bo_unreserve(obj);
-	return r;
-}
-
-static int amdgpu_cgs_kunmap_gpu_mem(struct cgs_device *cgs_device, cgs_handle_t handle)
-{
-	int r;
-	struct amdgpu_bo *obj = (struct amdgpu_bo *)handle;
-	r = amdgpu_bo_reserve(obj, true);
-	if (unlikely(r != 0))
-		return r;
-	amdgpu_bo_kunmap(obj);
-	amdgpu_bo_unreserve(obj);
-	return r;
-}
-
-static uint32_t amdgpu_cgs_read_register(struct cgs_device *cgs_device, unsigned offset)
+static uint32_t amdgpu_cgs_read_register(struct cgs_device *cgs_device, unsigned int offset)
 {
 	CGS_FUNC_ADEV;
 	return RREG32(offset);
 }
 
-static void amdgpu_cgs_write_register(struct cgs_device *cgs_device, unsigned offset,
+static void amdgpu_cgs_write_register(struct cgs_device *cgs_device, unsigned int offset,
 				      uint32_t value)
 {
 	CGS_FUNC_ADEV;
@@ -224,12 +56,10 @@ static void amdgpu_cgs_write_register(struct cgs_device *cgs_device, unsigned of
 
 static uint32_t amdgpu_cgs_read_ind_register(struct cgs_device *cgs_device,
 					     enum cgs_ind_reg space,
-					     unsigned index)
+					     unsigned int index)
 {
 	CGS_FUNC_ADEV;
 	switch (space) {
-	case CGS_IND_REG__MMIO:
-		return RREG32_IDX(index);
 	case CGS_IND_REG__PCIE:
 		return RREG32_PCIE(index);
 	case CGS_IND_REG__SMC:
@@ -240,9 +70,13 @@ static uint32_t amdgpu_cgs_read_ind_register(struct cgs_device *cgs_device,
 		return RREG32_DIDT(index);
 	case CGS_IND_REG_GC_CAC:
 		return RREG32_GC_CAC(index);
+	case CGS_IND_REG_SE_CAC:
+		return RREG32_SE_CAC(index);
 	case CGS_IND_REG__AUDIO_ENDPT:
 		DRM_ERROR("audio endpt register access not implemented.\n");
 		return 0;
+	default:
+		BUG();
 	}
 	WARN(1, "Invalid indirect register space");
 	return 0;
@@ -250,12 +84,10 @@ static uint32_t amdgpu_cgs_read_ind_register(struct cgs_device *cgs_device,
 
 static void amdgpu_cgs_write_ind_register(struct cgs_device *cgs_device,
 					  enum cgs_ind_reg space,
-					  unsigned index, uint32_t value)
+					  unsigned int index, uint32_t value)
 {
 	CGS_FUNC_ADEV;
 	switch (space) {
-	case CGS_IND_REG__MMIO:
-		return WREG32_IDX(index, value);
 	case CGS_IND_REG__PCIE:
 		return WREG32_PCIE(index, value);
 	case CGS_IND_REG__SMC:
@@ -266,231 +98,17 @@ static void amdgpu_cgs_write_ind_register(struct cgs_device *cgs_device,
 		return WREG32_DIDT(index, value);
 	case CGS_IND_REG_GC_CAC:
 		return WREG32_GC_CAC(index, value);
+	case CGS_IND_REG_SE_CAC:
+		return WREG32_SE_CAC(index, value);
 	case CGS_IND_REG__AUDIO_ENDPT:
 		DRM_ERROR("audio endpt register access not implemented.\n");
 		return;
-	}
-	WARN(1, "Invalid indirect register space");
-}
-
-static int amdgpu_cgs_get_pci_resource(struct cgs_device *cgs_device,
-				       enum cgs_resource_type resource_type,
-				       uint64_t size,
-				       uint64_t offset,
-				       uint64_t *resource_base)
-{
-	CGS_FUNC_ADEV;
-
-	if (resource_base == NULL)
-		return -EINVAL;
-
-	switch (resource_type) {
-	case CGS_RESOURCE_TYPE_MMIO:
-		if (adev->rmmio_size == 0)
-			return -ENOENT;
-		if ((offset + size) > adev->rmmio_size)
-			return -EINVAL;
-		*resource_base = adev->rmmio_base;
-		return 0;
-	case CGS_RESOURCE_TYPE_DOORBELL:
-		if (adev->doorbell.size == 0)
-			return -ENOENT;
-		if ((offset + size) > adev->doorbell.size)
-			return -EINVAL;
-		*resource_base = adev->doorbell.base;
-		return 0;
-	case CGS_RESOURCE_TYPE_FB:
-	case CGS_RESOURCE_TYPE_IO:
-	case CGS_RESOURCE_TYPE_ROM:
 	default:
-		return -EINVAL;
-	}
-}
-
-static const void *amdgpu_cgs_atom_get_data_table(struct cgs_device *cgs_device,
-						  unsigned table, uint16_t *size,
-						  uint8_t *frev, uint8_t *crev)
-{
-	CGS_FUNC_ADEV;
-	uint16_t data_start;
-
-	if (amdgpu_atom_parse_data_header(
-		    adev->mode_info.atom_context, table, size,
-		    frev, crev, &data_start))
-		return (uint8_t*)adev->mode_info.atom_context->bios +
-			data_start;
-
-	return NULL;
-}
-
-static int amdgpu_cgs_atom_get_cmd_table_revs(struct cgs_device *cgs_device, unsigned table,
-					      uint8_t *frev, uint8_t *crev)
-{
-	CGS_FUNC_ADEV;
-
-	if (amdgpu_atom_parse_cmd_header(
-		    adev->mode_info.atom_context, table,
-		    frev, crev))
-		return 0;
-
-	return -EINVAL;
-}
-
-static int amdgpu_cgs_atom_exec_cmd_table(struct cgs_device *cgs_device, unsigned table,
-					  void *args)
-{
-	CGS_FUNC_ADEV;
-
-	return amdgpu_atom_execute_table(
-		adev->mode_info.atom_context, table, args);
-}
-
-struct cgs_irq_params {
-	unsigned src_id;
-	cgs_irq_source_set_func_t set;
-	cgs_irq_handler_func_t handler;
-	void *private_data;
-};
-
-static int cgs_set_irq_state(struct amdgpu_device *adev,
-			     struct amdgpu_irq_src *src,
-			     unsigned type,
-			     enum amdgpu_interrupt_state state)
-{
-	struct cgs_irq_params *irq_params =
-		(struct cgs_irq_params *)src->data;
-	if (!irq_params)
-		return -EINVAL;
-	if (!irq_params->set)
-		return -EINVAL;
-	return irq_params->set(irq_params->private_data,
-			       irq_params->src_id,
-			       type,
-			       (int)state);
-}
-
-static int cgs_process_irq(struct amdgpu_device *adev,
-			   struct amdgpu_irq_src *source,
-			   struct amdgpu_iv_entry *entry)
-{
-	struct cgs_irq_params *irq_params =
-		(struct cgs_irq_params *)source->data;
-	if (!irq_params)
-		return -EINVAL;
-	if (!irq_params->handler)
-		return -EINVAL;
-	return irq_params->handler(irq_params->private_data,
-				   irq_params->src_id,
-				   entry->iv_entry);
-}
-
-static const struct amdgpu_irq_src_funcs cgs_irq_funcs = {
-	.set = cgs_set_irq_state,
-	.process = cgs_process_irq,
-};
-
-static int amdgpu_cgs_add_irq_source(void *cgs_device,
-				     unsigned client_id,
-				     unsigned src_id,
-				     unsigned num_types,
-				     cgs_irq_source_set_func_t set,
-				     cgs_irq_handler_func_t handler,
-				     void *private_data)
-{
-	CGS_FUNC_ADEV;
-	int ret = 0;
-	struct cgs_irq_params *irq_params;
-	struct amdgpu_irq_src *source =
-		kzalloc(sizeof(struct amdgpu_irq_src), GFP_KERNEL);
-	if (!source)
-		return -ENOMEM;
-	irq_params =
-		kzalloc(sizeof(struct cgs_irq_params), GFP_KERNEL);
-	if (!irq_params) {
-		kfree(source);
-		return -ENOMEM;
+		BUG();
 	}
-	source->num_types = num_types;
-	source->funcs = &cgs_irq_funcs;
-	irq_params->src_id = src_id;
-	irq_params->set = set;
-	irq_params->handler = handler;
-	irq_params->private_data = private_data;
-	source->data = (void *)irq_params;
-	ret = amdgpu_irq_add_id(adev, client_id, src_id, source);
-	if (ret) {
-		kfree(irq_params);
-		kfree(source);
-	}
-
-	return ret;
-}
-
-static int amdgpu_cgs_irq_get(void *cgs_device, unsigned client_id,
-			      unsigned src_id, unsigned type)
-{
-	CGS_FUNC_ADEV;
-
-	if (!adev->irq.client[client_id].sources)
-		return -EINVAL;
-
-	return amdgpu_irq_get(adev, adev->irq.client[client_id].sources[src_id], type);
-}
-
-static int amdgpu_cgs_irq_put(void *cgs_device, unsigned client_id,
-			      unsigned src_id, unsigned type)
-{
-	CGS_FUNC_ADEV;
-
-	if (!adev->irq.client[client_id].sources)
-		return -EINVAL;
-
-	return amdgpu_irq_put(adev, adev->irq.client[client_id].sources[src_id], type);
-}
-
-static int amdgpu_cgs_set_clockgating_state(struct cgs_device *cgs_device,
-				  enum amd_ip_block_type block_type,
-				  enum amd_clockgating_state state)
-{
-	CGS_FUNC_ADEV;
-	int i, r = -1;
-
-	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
-			continue;
-
-		if (adev->ip_blocks[i].version->type == block_type) {
-			r = adev->ip_blocks[i].version->funcs->set_clockgating_state(
-								(void *)adev,
-									state);
-			break;
-		}
-	}
-	return r;
-}
-
-static int amdgpu_cgs_set_powergating_state(struct cgs_device *cgs_device,
-				  enum amd_ip_block_type block_type,
-				  enum amd_powergating_state state)
-{
-	CGS_FUNC_ADEV;
-	int i, r = -1;
-
-	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
-			continue;
-
-		if (adev->ip_blocks[i].version->type == block_type) {
-			r = adev->ip_blocks[i].version->funcs->set_powergating_state(
-								(void *)adev,
-									state);
-			break;
-		}
-	}
-	return r;
+	WARN(1, "Invalid indirect register space");
 }
 
-
 static uint32_t fw_type_convert(struct cgs_device *cgs_device, uint32_t fw_type)
 {
 	CGS_FUNC_ADEV;
@@ -538,18 +156,6 @@ static uint32_t fw_type_convert(struct cgs_device *cgs_device, uint32_t fw_type)
 	return result;
 }
 
-static int amdgpu_cgs_rel_firmware(struct cgs_device *cgs_device, enum cgs_ucode_id type)
-{
-	CGS_FUNC_ADEV;
-	if ((CGS_UCODE_ID_SMU == type) || (CGS_UCODE_ID_SMU_SK == type)) {
-		release_firmware(adev->pm.fw);
-		adev->pm.fw = NULL;
-		return 0;
-	}
-	/* cannot release other firmware because they are not created by cgs */
-	return -EINVAL;
-}
-
 static uint16_t amdgpu_get_firmware_version(struct cgs_device *cgs_device,
 					enum cgs_ucode_id type)
 {
@@ -557,66 +163,49 @@ static uint16_t amdgpu_get_firmware_version(struct cgs_device *cgs_device,
 	uint16_t fw_version = 0;
 
 	switch (type) {
-		case CGS_UCODE_ID_SDMA0:
-			fw_version = adev->sdma.instance[0].fw_version;
-			break;
-		case CGS_UCODE_ID_SDMA1:
-			fw_version = adev->sdma.instance[1].fw_version;
-			break;
-		case CGS_UCODE_ID_CP_CE:
-			fw_version = adev->gfx.ce_fw_version;
-			break;
-		case CGS_UCODE_ID_CP_PFP:
-			fw_version = adev->gfx.pfp_fw_version;
-			break;
-		case CGS_UCODE_ID_CP_ME:
-			fw_version = adev->gfx.me_fw_version;
-			break;
-		case CGS_UCODE_ID_CP_MEC:
-			fw_version = adev->gfx.mec_fw_version;
-			break;
-		case CGS_UCODE_ID_CP_MEC_JT1:
-			fw_version = adev->gfx.mec_fw_version;
-			break;
-		case CGS_UCODE_ID_CP_MEC_JT2:
-			fw_version = adev->gfx.mec_fw_version;
-			break;
-		case CGS_UCODE_ID_RLC_G:
-			fw_version = adev->gfx.rlc_fw_version;
-			break;
-		case CGS_UCODE_ID_STORAGE:
-			break;
-		default:
-			DRM_ERROR("firmware type %d do not have version\n", type);
-			break;
+	case CGS_UCODE_ID_SDMA0:
+		fw_version = adev->sdma.instance[0].fw_version;
+		break;
+	case CGS_UCODE_ID_SDMA1:
+		fw_version = adev->sdma.instance[1].fw_version;
+		break;
+	case CGS_UCODE_ID_CP_CE:
+		fw_version = adev->gfx.ce_fw_version;
+		break;
+	case CGS_UCODE_ID_CP_PFP:
+		fw_version = adev->gfx.pfp_fw_version;
+		break;
+	case CGS_UCODE_ID_CP_ME:
+		fw_version = adev->gfx.me_fw_version;
+		break;
+	case CGS_UCODE_ID_CP_MEC:
+		fw_version = adev->gfx.mec_fw_version;
+		break;
+	case CGS_UCODE_ID_CP_MEC_JT1:
+		fw_version = adev->gfx.mec_fw_version;
+		break;
+	case CGS_UCODE_ID_CP_MEC_JT2:
+		fw_version = adev->gfx.mec_fw_version;
+		break;
+	case CGS_UCODE_ID_RLC_G:
+		fw_version = adev->gfx.rlc_fw_version;
+		break;
+	case CGS_UCODE_ID_STORAGE:
+		break;
+	default:
+		DRM_ERROR("firmware type %d do not have version\n", type);
+		break;
 	}
 	return fw_version;
 }
 
-static int amdgpu_cgs_enter_safe_mode(struct cgs_device *cgs_device,
-					bool en)
-{
-	CGS_FUNC_ADEV;
-
-	if (adev->gfx.rlc.funcs->enter_safe_mode == NULL ||
-		adev->gfx.rlc.funcs->exit_safe_mode == NULL)
-		return 0;
-
-	if (en)
-		adev->gfx.rlc.funcs->enter_safe_mode(adev);
-	else
-		adev->gfx.rlc.funcs->exit_safe_mode(adev);
-
-	return 0;
-}
-
 static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 					enum cgs_ucode_id type,
 					struct cgs_firmware_info *info)
 {
 	CGS_FUNC_ADEV;
 
-	if ((CGS_UCODE_ID_SMU != type) && (CGS_UCODE_ID_SMU_SK != type)) {
+	if (type != CGS_UCODE_ID_SMU && type != CGS_UCODE_ID_SMU_SK) {
 		uint64_t gpu_addr;
 		uint32_t data_size;
 		const struct gfx_firmware_header_v1_0 *header;
@@ -624,6 +213,9 @@ static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 		struct amdgpu_firmware_info *ucode;
 
 		id = fw_type_convert(cgs_device, type);
+		if (id >= AMDGPU_UCODE_ID_MAXIMUM)
+			return -EINVAL;
+
 		ucode = &adev->firmware.ucode[id];
 		if (ucode->fw == NULL)
 			return -EINVAL;
@@ -643,8 +235,8 @@ static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 		info->mc_addr = gpu_addr;
 		info->version = (uint16_t)le32_to_cpu(header->header.ucode_version);
 
-		if (CGS_UCODE_ID_CP_MEC == type)
-			info->image_size = (header->jt_offset) << 2;
+		if (type == CGS_UCODE_ID_CP_MEC)
+			info->image_size = le32_to_cpu(header->jt_offset) << 2;
 
 		info->fw_version = amdgpu_get_firmware_version(cgs_device, type);
 		info->feature_version = (uint16_t)le32_to_cpu(header->ucode_feature_version);
@@ -660,83 +252,113 @@ static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 
 		if (!adev->pm.fw) {
 			switch (adev->asic_type) {
+			case CHIP_BONAIRE:
+				if ((adev->pdev->revision == 0x80) ||
+					(adev->pdev->revision == 0x81) ||
+					(adev->pdev->device == 0x665f)) {
+					info->is_kicker = true;
+					strscpy(fw_name, "amdgpu/bonaire_k_smc.bin");
+				} else {
+					strscpy(fw_name, "amdgpu/bonaire_smc.bin");
+				}
+				break;
+			case CHIP_HAWAII:
+				if (adev->pdev->revision == 0x80) {
+					info->is_kicker = true;
+					strscpy(fw_name, "amdgpu/hawaii_k_smc.bin");
+				} else {
+					strscpy(fw_name, "amdgpu/hawaii_smc.bin");
+				}
+				break;
 			case CHIP_TOPAZ:
 				if (((adev->pdev->device == 0x6900) && (adev->pdev->revision == 0x81)) ||
 				    ((adev->pdev->device == 0x6900) && (adev->pdev->revision == 0x83)) ||
-				    ((adev->pdev->device == 0x6907) && (adev->pdev->revision == 0x87))) {
+				    ((adev->pdev->device == 0x6907) && (adev->pdev->revision == 0x87)) ||
+				    ((adev->pdev->device == 0x6900) && (adev->pdev->revision == 0xD1)) ||
+				    ((adev->pdev->device == 0x6900) && (adev->pdev->revision == 0xD3))) {
 					info->is_kicker = true;
-					strcpy(fw_name, "amdgpu/topaz_k_smc.bin");
+					strscpy(fw_name, "amdgpu/topaz_k_smc.bin");
 				} else
-					strcpy(fw_name, "amdgpu/topaz_smc.bin");
+					strscpy(fw_name, "amdgpu/topaz_smc.bin");
 				break;
 			case CHIP_TONGA:
 				if (((adev->pdev->device == 0x6939) && (adev->pdev->revision == 0xf1)) ||
 				    ((adev->pdev->device == 0x6938) && (adev->pdev->revision == 0xf1))) {
 					info->is_kicker = true;
-					strcpy(fw_name, "amdgpu/tonga_k_smc.bin");
+					strscpy(fw_name, "amdgpu/tonga_k_smc.bin");
 				} else
-					strcpy(fw_name, "amdgpu/tonga_smc.bin");
+					strscpy(fw_name, "amdgpu/tonga_smc.bin");
 				break;
 			case CHIP_FIJI:
-				strcpy(fw_name, "amdgpu/fiji_smc.bin");
+				strscpy(fw_name, "amdgpu/fiji_smc.bin");
 				break;
 			case CHIP_POLARIS11:
 				if (type == CGS_UCODE_ID_SMU) {
-					if (((adev->pdev->device == 0x67ef) &&
-					     ((adev->pdev->revision == 0xe0) ||
-					      (adev->pdev->revision == 0xe2) ||
-					      (adev->pdev->revision == 0xe5))) ||
-					    ((adev->pdev->device == 0x67ff) &&
-					     ((adev->pdev->revision == 0xcf) ||
-					      (adev->pdev->revision == 0xef) ||
-					      (adev->pdev->revision == 0xff)))) {
+					if (ASICID_IS_P21(adev->pdev->device, adev->pdev->revision)) {
 						info->is_kicker = true;
-						strcpy(fw_name, "amdgpu/polaris11_k_smc.bin");
-					} else
-						strcpy(fw_name, "amdgpu/polaris11_smc.bin");
+						strscpy(fw_name, "amdgpu/polaris11_k_smc.bin");
+					} else if (ASICID_IS_P31(adev->pdev->device, adev->pdev->revision)) {
+						info->is_kicker = true;
+						strscpy(fw_name, "amdgpu/polaris11_k2_smc.bin");
+					} else {
+						strscpy(fw_name, "amdgpu/polaris11_smc.bin");
+					}
 				} else if (type == CGS_UCODE_ID_SMU_SK) {
-					strcpy(fw_name, "amdgpu/polaris11_smc_sk.bin");
+					strscpy(fw_name, "amdgpu/polaris11_smc_sk.bin");
 				}
 				break;
 			case CHIP_POLARIS10:
 				if (type == CGS_UCODE_ID_SMU) {
-					if ((adev->pdev->device == 0x67df) &&
-					    ((adev->pdev->revision == 0xe0) ||
-					     (adev->pdev->revision == 0xe3) ||
-					     (adev->pdev->revision == 0xe4) ||
-					     (adev->pdev->revision == 0xe5) ||
-					     (adev->pdev->revision == 0xe7) ||
-					     (adev->pdev->revision == 0xef))) {
+					if (ASICID_IS_P20(adev->pdev->device, adev->pdev->revision)) {
+						info->is_kicker = true;
+						strscpy(fw_name, "amdgpu/polaris10_k_smc.bin");
+					} else if (ASICID_IS_P30(adev->pdev->device, adev->pdev->revision)) {
 						info->is_kicker = true;
-						strcpy(fw_name, "amdgpu/polaris10_k_smc.bin");
-					} else
-						strcpy(fw_name, "amdgpu/polaris10_smc.bin");
+						strscpy(fw_name, "amdgpu/polaris10_k2_smc.bin");
+					} else {
+						strscpy(fw_name, "amdgpu/polaris10_smc.bin");
+					}
 				} else if (type == CGS_UCODE_ID_SMU_SK) {
-					strcpy(fw_name, "amdgpu/polaris10_smc_sk.bin");
+					strscpy(fw_name, "amdgpu/polaris10_smc_sk.bin");
 				}
 				break;
 			case CHIP_POLARIS12:
-				strcpy(fw_name, "amdgpu/polaris12_smc.bin");
+				if (ASICID_IS_P23(adev->pdev->device, adev->pdev->revision)) {
+					info->is_kicker = true;
+					strscpy(fw_name, "amdgpu/polaris12_k_smc.bin");
+				} else {
+					strscpy(fw_name, "amdgpu/polaris12_smc.bin");
+				}
+				break;
+			case CHIP_VEGAM:
+				strscpy(fw_name, "amdgpu/vegam_smc.bin");
 				break;
 			case CHIP_VEGA10:
-				strcpy(fw_name, "amdgpu/vega10_smc.bin");
+				if ((adev->pdev->device == 0x687f) &&
+					((adev->pdev->revision == 0xc0) ||
+					(adev->pdev->revision == 0xc1) ||
+					(adev->pdev->revision == 0xc3)))
+					strscpy(fw_name, "amdgpu/vega10_acg_smc.bin");
+				else
+					strscpy(fw_name, "amdgpu/vega10_smc.bin");
+				break;
+			case CHIP_VEGA12:
+				strscpy(fw_name, "amdgpu/vega12_smc.bin");
+				break;
+			case CHIP_VEGA20:
+				strscpy(fw_name, "amdgpu/vega20_smc.bin");
 				break;
 			default:
 				DRM_ERROR("SMC firmware not supported\n");
 				return -EINVAL;
 			}
 
-			err = request_firmware(&adev->pm.fw, fw_name, adev->dev);
-			if (err) {
-				DRM_ERROR("Failed to request firmware\n");
-				return err;
-			}
-
-			err = amdgpu_ucode_validate(adev->pm.fw);
+			err = amdgpu_ucode_request(adev, &adev->pm.fw,
+						   AMDGPU_UCODE_REQUIRED,
+						   "%s", fw_name);
 			if (err) {
 				DRM_ERROR("Failed to load firmware \"%s\"", fw_name);
-				release_firmware(adev->pm.fw);
-				adev->pm.fw = NULL;
+				amdgpu_ucode_release(&adev->pm.fw);
 				return err;
 			}
 
@@ -766,357 +388,12 @@ static int amdgpu_cgs_get_firmware_info(struct cgs_device *cgs_device,
 	return 0;
 }
 
-static int amdgpu_cgs_is_virtualization_enabled(void *cgs_device)
-{
-	CGS_FUNC_ADEV;
-	return amdgpu_sriov_vf(adev);
-}
-
-static int amdgpu_cgs_query_system_info(struct cgs_device *cgs_device,
-					struct cgs_system_info *sys_info)
-{
-	CGS_FUNC_ADEV;
-
-	if (NULL == sys_info)
-		return -ENODEV;
-
-	if (sizeof(struct cgs_system_info) != sys_info->size)
-		return -ENODEV;
-
-	switch (sys_info->info_id) {
-	case CGS_SYSTEM_INFO_ADAPTER_BDF_ID:
-		sys_info->value = adev->pdev->devfn | (adev->pdev->bus->number << 8);
-		break;
-	case CGS_SYSTEM_INFO_PCIE_GEN_INFO:
-		sys_info->value = adev->pm.pcie_gen_mask;
-		break;
-	case CGS_SYSTEM_INFO_PCIE_MLW:
-		sys_info->value = adev->pm.pcie_mlw_mask;
-		break;
-	case CGS_SYSTEM_INFO_PCIE_DEV:
-		sys_info->value = adev->pdev->device;
-		break;
-	case CGS_SYSTEM_INFO_PCIE_REV:
-		sys_info->value = adev->pdev->revision;
-		break;
-	case CGS_SYSTEM_INFO_CG_FLAGS:
-		sys_info->value = adev->cg_flags;
-		break;
-	case CGS_SYSTEM_INFO_PG_FLAGS:
-		sys_info->value = adev->pg_flags;
-		break;
-	case CGS_SYSTEM_INFO_GFX_CU_INFO:
-		sys_info->value = adev->gfx.cu_info.number;
-		break;
-	case CGS_SYSTEM_INFO_GFX_SE_INFO:
-		sys_info->value = adev->gfx.config.max_shader_engines;
-		break;
-	case CGS_SYSTEM_INFO_PCIE_SUB_SYS_ID:
-		sys_info->value = adev->pdev->subsystem_device;
-		break;
-	case CGS_SYSTEM_INFO_PCIE_SUB_SYS_VENDOR_ID:
-		sys_info->value = adev->pdev->subsystem_vendor;
-		break;
-	default:
-		return -ENODEV;
-	}
-
-	return 0;
-}
-
-static int amdgpu_cgs_get_active_displays_info(struct cgs_device *cgs_device,
-					  struct cgs_display_info *info)
-{
-	CGS_FUNC_ADEV;
-	struct amdgpu_crtc *amdgpu_crtc;
-	struct drm_device *ddev = adev->ddev;
-	struct drm_crtc *crtc;
-	uint32_t line_time_us, vblank_lines;
-	struct cgs_mode_info *mode_info;
-
-	if (info == NULL)
-		return -EINVAL;
-
-	mode_info = info->mode_info;
-
-	if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) {
-		list_for_each_entry(crtc,
-				&ddev->mode_config.crtc_list, head) {
-			amdgpu_crtc = to_amdgpu_crtc(crtc);
-			if (crtc->enabled) {
-				info->active_display_mask |= (1 << amdgpu_crtc->crtc_id);
-				info->display_count++;
-			}
-			if (mode_info != NULL &&
-				crtc->enabled && amdgpu_crtc->enabled &&
-				amdgpu_crtc->hw_mode.clock) {
-				line_time_us = (amdgpu_crtc->hw_mode.crtc_htotal * 1000) /
-							amdgpu_crtc->hw_mode.clock;
-				vblank_lines = amdgpu_crtc->hw_mode.crtc_vblank_end -
-							amdgpu_crtc->hw_mode.crtc_vdisplay +
-							(amdgpu_crtc->v_border * 2);
-				mode_info->vblank_time_us = vblank_lines * line_time_us;
-				mode_info->refresh_rate = drm_mode_vrefresh(&amdgpu_crtc->hw_mode);
-				mode_info->ref_clock = adev->clock.spll.reference_freq;
-				mode_info = NULL;
-			}
-		}
-	}
-
-	return 0;
-}
-
-
-static int amdgpu_cgs_notify_dpm_enabled(struct cgs_device *cgs_device, bool enabled)
-{
-	CGS_FUNC_ADEV;
-
-	adev->pm.dpm_enabled = enabled;
-
-	return 0;
-}
-
-/** \brief evaluate acpi namespace object, handle or pathname must be valid
- *  \param cgs_device
- *  \param info input/output arguments for the control method
- *  \return status
- */
-
-#if defined(CONFIG_ACPI)
-static int amdgpu_cgs_acpi_eval_object(struct cgs_device *cgs_device,
-				    struct cgs_acpi_method_info *info)
-{
-	CGS_FUNC_ADEV;
-	acpi_handle handle;
-	struct acpi_object_list input;
-	struct acpi_buffer output = { ACPI_ALLOCATE_BUFFER, NULL };
-	union acpi_object *params, *obj;
-	uint8_t name[5] = {'\0'};
-	struct cgs_acpi_method_argument *argument;
-	uint32_t i, count;
-	acpi_status status;
-	int result;
-
-	handle = ACPI_HANDLE(&adev->pdev->dev);
-	if (!handle)
-		return -ENODEV;
-
-	memset(&input, 0, sizeof(struct acpi_object_list));
-
-	/* validate input info */
-	if (info->size != sizeof(struct cgs_acpi_method_info))
-		return -EINVAL;
-
-	input.count = info->input_count;
-	if (info->input_count > 0) {
-		if (info->pinput_argument == NULL)
-			return -EINVAL;
-		argument = info->pinput_argument;
-		for (i = 0; i < info->input_count; i++) {
-			if (((argument->type == ACPI_TYPE_STRING) ||
-			     (argument->type == ACPI_TYPE_BUFFER)) &&
-			    (argument->pointer == NULL))
-				return -EINVAL;
-			argument++;
-		}
-	}
-
-	if (info->output_count > 0) {
-		if (info->poutput_argument == NULL)
-			return -EINVAL;
-		argument = info->poutput_argument;
-		for (i = 0; i < info->output_count; i++) {
-			if (((argument->type == ACPI_TYPE_STRING) ||
-				(argument->type == ACPI_TYPE_BUFFER))
-				&& (argument->pointer == NULL))
-				return -EINVAL;
-			argument++;
-		}
-	}
-
-	/* The path name passed to acpi_evaluate_object should be null terminated */
-	if ((info->field & CGS_ACPI_FIELD_METHOD_NAME) != 0) {
-		strncpy(name, (char *)&(info->name), sizeof(uint32_t));
-		name[4] = '\0';
-	}
-
-	/* parse input parameters */
-	if (input.count > 0) {
-		input.pointer = params =
-				kzalloc(sizeof(union acpi_object) * input.count, GFP_KERNEL);
-		if (params == NULL)
-			return -EINVAL;
-
-		argument = info->pinput_argument;
-
-		for (i = 0; i < input.count; i++) {
-			params->type = argument->type;
-			switch (params->type) {
-			case ACPI_TYPE_INTEGER:
-				params->integer.value = argument->value;
-				break;
-			case ACPI_TYPE_STRING:
-				params->string.length = argument->data_length;
-				params->string.pointer = argument->pointer;
-				break;
-			case ACPI_TYPE_BUFFER:
-				params->buffer.length = argument->data_length;
-				params->buffer.pointer = argument->pointer;
-				break;
-			default:
-				break;
-			}
-			params++;
-			argument++;
-		}
-	}
-
-	/* parse output info */
-	count = info->output_count;
-	argument = info->poutput_argument;
-
-	/* evaluate the acpi method */
-	status = acpi_evaluate_object(handle, name, &input, &output);
-
-	if (ACPI_FAILURE(status)) {
-		result = -EIO;
-		goto free_input;
-	}
-
-	/* return the output info */
-	obj = output.pointer;
-
-	if (count > 1) {
-		if ((obj->type != ACPI_TYPE_PACKAGE) ||
-			(obj->package.count != count)) {
-			result = -EIO;
-			goto free_obj;
-		}
-		params = obj->package.elements;
-	} else
-		params = obj;
-
-	if (params == NULL) {
-		result = -EIO;
-		goto free_obj;
-	}
-
-	for (i = 0; i < count; i++) {
-		if (argument->type != params->type) {
-			result = -EIO;
-			goto free_obj;
-		}
-		switch (params->type) {
-		case ACPI_TYPE_INTEGER:
-			argument->value = params->integer.value;
-			break;
-		case ACPI_TYPE_STRING:
-			if ((params->string.length != argument->data_length) ||
-				(params->string.pointer == NULL)) {
-				result = -EIO;
-				goto free_obj;
-			}
-			strncpy(argument->pointer,
-				params->string.pointer,
-				params->string.length);
-			break;
-		case ACPI_TYPE_BUFFER:
-			if (params->buffer.pointer == NULL) {
-				result = -EIO;
-				goto free_obj;
-			}
-			memcpy(argument->pointer,
-				params->buffer.pointer,
-				argument->data_length);
-			break;
-		default:
-			break;
-		}
-		argument++;
-		params++;
-	}
-
-	result = 0;
-free_obj:
-	kfree(obj);
-free_input:
-	kfree((void *)input.pointer);
-	return result;
-}
-#else
-static int amdgpu_cgs_acpi_eval_object(struct cgs_device *cgs_device,
-				struct cgs_acpi_method_info *info)
-{
-	return -EIO;
-}
-#endif
-
-static int amdgpu_cgs_call_acpi_method(struct cgs_device *cgs_device,
-					uint32_t acpi_method,
-					uint32_t acpi_function,
-					void *pinput, void *poutput,
-					uint32_t output_count,
-					uint32_t input_size,
-					uint32_t output_size)
-{
-	struct cgs_acpi_method_argument acpi_input[2] = { {0}, {0} };
-	struct cgs_acpi_method_argument acpi_output = {0};
-	struct cgs_acpi_method_info info = {0};
-
-	acpi_input[0].type = CGS_ACPI_TYPE_INTEGER;
-	acpi_input[0].data_length = sizeof(uint32_t);
-	acpi_input[0].value = acpi_function;
-
-	acpi_input[1].type = CGS_ACPI_TYPE_BUFFER;
-	acpi_input[1].data_length = input_size;
-	acpi_input[1].pointer = pinput;
-
-	acpi_output.type = CGS_ACPI_TYPE_BUFFER;
-	acpi_output.data_length = output_size;
-	acpi_output.pointer = poutput;
-
-	info.size = sizeof(struct cgs_acpi_method_info);
-	info.field = CGS_ACPI_FIELD_METHOD_NAME | CGS_ACPI_FIELD_INPUT_ARGUMENT_COUNT;
-	info.input_count = 2;
-	info.name = acpi_method;
-	info.pinput_argument = acpi_input;
-	info.output_count = output_count;
-	info.poutput_argument = &acpi_output;
-
-	return amdgpu_cgs_acpi_eval_object(cgs_device, &info);
-}
-
 static const struct cgs_ops amdgpu_cgs_ops = {
-	.alloc_gpu_mem = amdgpu_cgs_alloc_gpu_mem,
-	.free_gpu_mem = amdgpu_cgs_free_gpu_mem,
-	.gmap_gpu_mem = amdgpu_cgs_gmap_gpu_mem,
-	.gunmap_gpu_mem = amdgpu_cgs_gunmap_gpu_mem,
-	.kmap_gpu_mem = amdgpu_cgs_kmap_gpu_mem,
-	.kunmap_gpu_mem = amdgpu_cgs_kunmap_gpu_mem,
 	.read_register = amdgpu_cgs_read_register,
 	.write_register = amdgpu_cgs_write_register,
 	.read_ind_register = amdgpu_cgs_read_ind_register,
 	.write_ind_register = amdgpu_cgs_write_ind_register,
-	.get_pci_resource = amdgpu_cgs_get_pci_resource,
-	.atom_get_data_table = amdgpu_cgs_atom_get_data_table,
-	.atom_get_cmd_table_revs = amdgpu_cgs_atom_get_cmd_table_revs,
-	.atom_exec_cmd_table = amdgpu_cgs_atom_exec_cmd_table,
 	.get_firmware_info = amdgpu_cgs_get_firmware_info,
-	.rel_firmware = amdgpu_cgs_rel_firmware,
-	.set_powergating_state = amdgpu_cgs_set_powergating_state,
-	.set_clockgating_state = amdgpu_cgs_set_clockgating_state,
-	.get_active_displays_info = amdgpu_cgs_get_active_displays_info,
-	.notify_dpm_enabled = amdgpu_cgs_notify_dpm_enabled,
-	.call_acpi_method = amdgpu_cgs_call_acpi_method,
-	.query_system_info = amdgpu_cgs_query_system_info,
-	.is_virtualization_enabled = amdgpu_cgs_is_virtualization_enabled,
-	.enter_safe_mode = amdgpu_cgs_enter_safe_mode,
-};
-
-static const struct cgs_os_ops amdgpu_cgs_os_ops = {
-	.add_irq_source = amdgpu_cgs_add_irq_source,
-	.irq_get = amdgpu_cgs_irq_get,
-	.irq_put = amdgpu_cgs_irq_put
 };
 
 struct cgs_device *amdgpu_cgs_create_device(struct amdgpu_device *adev)
@@ -1130,7 +407,6 @@ struct cgs_device *amdgpu_cgs_create_device(struct amdgpu_device *adev)
 	}
 
 	cgs_device->base.ops = &amdgpu_cgs_ops;
-	cgs_device->base.os_ops = &amdgpu_cgs_os_ops;
 	cgs_device->adev = adev;
 
 	return (struct cgs_device *)cgs_device;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c
index 8d1cf2d3e663..9f96d568acf2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c
@@ -23,10 +23,12 @@
  * Authors: Dave Airlie
  *          Alex Deucher
  */
-#include <drm/drmP.h>
-#include <drm/drm_edid.h>
+
+#include <drm/display/drm_dp_helper.h>
 #include <drm/drm_crtc_helper.h>
-#include <drm/drm_fb_helper.h>
+#include <drm/drm_edid.h>
+#include <drm/drm_modeset_helper_vtables.h>
+#include <drm/drm_probe_helper.h>
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 #include "atom.h"
@@ -34,13 +36,14 @@
 #include "atombios_dp.h"
 #include "amdgpu_connectors.h"
 #include "amdgpu_i2c.h"
+#include "amdgpu_display.h"
 
 #include <linux/pm_runtime.h>
 
 void amdgpu_connector_hotplug(struct drm_connector *connector)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 
 	/* bail if the connector does not have hpd pin, e.g.,
@@ -69,25 +72,18 @@ void amdgpu_connector_hotplug(struct drm_connector *connector)
 		/* don't do anything if sink is not display port, i.e.,
 		 * passive dp->(dvi|hdmi) adaptor
 		 */
-		if (dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_DISPLAYPORT) {
-			int saved_dpms = connector->dpms;
-			/* Only turn off the display if it's physically disconnected */
-			if (!amdgpu_display_hpd_sense(adev, amdgpu_connector->hpd.hpd)) {
-				drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
-			} else if (amdgpu_atombios_dp_needs_link_train(amdgpu_connector)) {
-				/* Don't try to start link training before we
-				 * have the dpcd */
-				if (amdgpu_atombios_dp_get_dpcd(amdgpu_connector))
-					return;
-
-				/* set it to OFF so that drm_helper_connector_dpms()
-				 * won't return immediately since the current state
-				 * is ON at this point.
-				 */
-				connector->dpms = DRM_MODE_DPMS_OFF;
-				drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON);
-			}
-			connector->dpms = saved_dpms;
+		if (dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_DISPLAYPORT &&
+		    amdgpu_display_hpd_sense(adev, amdgpu_connector->hpd.hpd) &&
+		    amdgpu_atombios_dp_needs_link_train(amdgpu_connector)) {
+			/* Don't start link training before we have the DPCD */
+			if (amdgpu_atombios_dp_get_dpcd(amdgpu_connector))
+				return;
+
+			/* Turn the connector off and back on immediately, which
+			 * will trigger link training
+			 */
+			drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
+			drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON);
 		}
 	}
 }
@@ -107,13 +103,13 @@ int amdgpu_connector_get_monitor_bpc(struct drm_connector *connector)
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 	struct amdgpu_connector_atom_dig *dig_connector;
 	int bpc = 8;
-	unsigned mode_clock, max_tmds_clock;
+	unsigned int mode_clock, max_tmds_clock;
 
 	switch (connector->connector_type) {
 	case DRM_MODE_CONNECTOR_DVII:
 	case DRM_MODE_CONNECTOR_HDMIB:
 		if (amdgpu_connector->use_digital) {
-			if (drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
+			if (connector->display_info.is_hdmi) {
 				if (connector->display_info.bpc)
 					bpc = connector->display_info.bpc;
 			}
@@ -121,7 +117,7 @@ int amdgpu_connector_get_monitor_bpc(struct drm_connector *connector)
 		break;
 	case DRM_MODE_CONNECTOR_DVID:
 	case DRM_MODE_CONNECTOR_HDMIA:
-		if (drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
+		if (connector->display_info.is_hdmi) {
 			if (connector->display_info.bpc)
 				bpc = connector->display_info.bpc;
 		}
@@ -130,7 +126,7 @@ int amdgpu_connector_get_monitor_bpc(struct drm_connector *connector)
 		dig_connector = amdgpu_connector->con_priv;
 		if ((dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_DISPLAYPORT) ||
 		    (dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_eDP) ||
-		    drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
+		    connector->display_info.is_hdmi) {
 			if (connector->display_info.bpc)
 				bpc = connector->display_info.bpc;
 		}
@@ -154,7 +150,7 @@ int amdgpu_connector_get_monitor_bpc(struct drm_connector *connector)
 		break;
 	}
 
-	if (drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
+	if (connector->display_info.is_hdmi) {
 		/*
 		 * Pre DCE-8 hw can't handle > 12 bpc, and more than 12 bpc doesn't make
 		 * much sense without support for > 12 bpc framebuffers. RGB 4:4:4 at
@@ -180,7 +176,7 @@ int amdgpu_connector_get_monitor_bpc(struct drm_connector *connector)
 
 			/* Check if bpc is within clock limit. Try to degrade gracefully otherwise */
 			if ((bpc == 12) && (mode_clock * 3/2 > max_tmds_clock)) {
-				if ((connector->display_info.edid_hdmi_dc_modes & DRM_EDID_HDMI_DC_30) &&
+				if ((connector->display_info.edid_hdmi_rgb444_dc_modes & DRM_EDID_HDMI_DC_30) &&
 				    (mode_clock * 5/4 <= max_tmds_clock))
 					bpc = 10;
 				else
@@ -219,30 +215,20 @@ static void
 amdgpu_connector_update_scratch_regs(struct drm_connector *connector,
 				      enum drm_connector_status status)
 {
-	struct drm_encoder *best_encoder = NULL;
-	struct drm_encoder *encoder = NULL;
+	struct drm_encoder *best_encoder;
+	struct drm_encoder *encoder;
 	const struct drm_connector_helper_funcs *connector_funcs = connector->helper_private;
 	bool connected;
-	int i;
 
 	best_encoder = connector_funcs->best_encoder(connector);
 
-	for (i = 0; i < DRM_CONNECTOR_MAX_ENCODER; i++) {
-		if (connector->encoder_ids[i] == 0)
-			break;
-
-		encoder = drm_encoder_find(connector->dev,
-					connector->encoder_ids[i]);
-		if (!encoder)
-			continue;
-
+	drm_connector_for_each_possible_encoder(connector, encoder) {
 		if ((encoder == best_encoder) && (status == connector_status_connected))
 			connected = true;
 		else
 			connected = false;
 
 		amdgpu_atombios_encoder_set_bios_scratch_regs(connector, encoder, connected);
-
 	}
 }
 
@@ -251,58 +237,25 @@ amdgpu_connector_find_encoder(struct drm_connector *connector,
 			       int encoder_type)
 {
 	struct drm_encoder *encoder;
-	int i;
-
-	for (i = 0; i < DRM_CONNECTOR_MAX_ENCODER; i++) {
-		if (connector->encoder_ids[i] == 0)
-			break;
-		encoder = drm_encoder_find(connector->dev,
-					connector->encoder_ids[i]);
-		if (!encoder)
-			continue;
 
+	drm_connector_for_each_possible_encoder(connector, encoder) {
 		if (encoder->encoder_type == encoder_type)
 			return encoder;
 	}
-	return NULL;
-}
-
-struct edid *amdgpu_connector_edid(struct drm_connector *connector)
-{
-	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
-	struct drm_property_blob *edid_blob = connector->edid_blob_ptr;
 
-	if (amdgpu_connector->edid) {
-		return amdgpu_connector->edid;
-	} else if (edid_blob) {
-		struct edid *edid = kmemdup(edid_blob->data, edid_blob->length, GFP_KERNEL);
-		if (edid)
-			amdgpu_connector->edid = edid;
-	}
-	return amdgpu_connector->edid;
+	return NULL;
 }
 
 static struct edid *
 amdgpu_connector_get_hardcoded_edid(struct amdgpu_device *adev)
 {
-	struct edid *edid;
-
-	if (adev->mode_info.bios_hardcoded_edid) {
-		edid = kmalloc(adev->mode_info.bios_hardcoded_edid_size, GFP_KERNEL);
-		if (edid) {
-			memcpy((unsigned char *)edid,
-			       (unsigned char *)adev->mode_info.bios_hardcoded_edid,
-			       adev->mode_info.bios_hardcoded_edid_size);
-			return edid;
-		}
-	}
-	return NULL;
+	return drm_edid_duplicate(drm_edid_raw(adev->mode_info.bios_hardcoded_edid));
 }
 
 static void amdgpu_connector_get_edid(struct drm_connector *connector)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 
 	if (amdgpu_connector->edid)
@@ -337,8 +290,10 @@ static void amdgpu_connector_get_edid(struct drm_connector *connector)
 	if (!amdgpu_connector->edid) {
 		/* some laptops provide a hardcoded edid in rom for LCDs */
 		if (((connector->connector_type == DRM_MODE_CONNECTOR_LVDS) ||
-		     (connector->connector_type == DRM_MODE_CONNECTOR_eDP)))
+		     (connector->connector_type == DRM_MODE_CONNECTOR_eDP))) {
 			amdgpu_connector->edid = amdgpu_connector_get_hardcoded_edid(adev);
+			drm_connector_update_edid_property(connector, amdgpu_connector->edid);
+		}
 	}
 }
 
@@ -346,10 +301,8 @@ static void amdgpu_connector_free_edid(struct drm_connector *connector)
 {
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 
-	if (amdgpu_connector->edid) {
-		kfree(amdgpu_connector->edid);
-		amdgpu_connector->edid = NULL;
-	}
+	kfree(amdgpu_connector->edid);
+	amdgpu_connector->edid = NULL;
 }
 
 static int amdgpu_connector_ddc_get_modes(struct drm_connector *connector)
@@ -358,23 +311,23 @@ static int amdgpu_connector_ddc_get_modes(struct drm_connector *connector)
 	int ret;
 
 	if (amdgpu_connector->edid) {
-		drm_mode_connector_update_edid_property(connector, amdgpu_connector->edid);
+		drm_connector_update_edid_property(connector, amdgpu_connector->edid);
 		ret = drm_add_edid_modes(connector, amdgpu_connector->edid);
-		drm_edid_to_eld(connector, amdgpu_connector->edid);
 		return ret;
 	}
-	drm_mode_connector_update_edid_property(connector, NULL);
+	drm_connector_update_edid_property(connector, NULL);
 	return 0;
 }
 
 static struct drm_encoder *
 amdgpu_connector_best_single_encoder(struct drm_connector *connector)
 {
-	int enc_id = connector->encoder_ids[0];
+	struct drm_encoder *encoder;
+
+	/* pick the first one */
+	drm_connector_for_each_possible_encoder(connector, encoder)
+		return encoder;
 
-	/* pick the encoder ids */
-	if (enc_id)
-		return drm_encoder_find(connector->dev, enc_id);
 	return NULL;
 }
 
@@ -411,6 +364,9 @@ amdgpu_connector_lcd_native_mode(struct drm_encoder *encoder)
 	    native_mode->vdisplay != 0 &&
 	    native_mode->clock != 0) {
 		mode = drm_mode_duplicate(dev, native_mode);
+		if (!mode)
+			return NULL;
+
 		mode->type = DRM_MODE_TYPE_PREFERRED | DRM_MODE_TYPE_DRIVER;
 		drm_mode_set_name(mode);
 
@@ -425,6 +381,9 @@ amdgpu_connector_lcd_native_mode(struct drm_encoder *encoder)
 		 * simpler.
 		 */
 		mode = drm_cvt_mode(dev, native_mode->hdisplay, native_mode->vdisplay, 60, true, false, false);
+		if (!mode)
+			return NULL;
+
 		mode->type = DRM_MODE_TYPE_PREFERRED | DRM_MODE_TYPE_DRIVER;
 		DRM_DEBUG_KMS("Adding cvt approximation of native panel mode %s\n", mode->name);
 	}
@@ -439,30 +398,28 @@ static void amdgpu_connector_add_common_modes(struct drm_encoder *encoder,
 	struct drm_display_mode *mode = NULL;
 	struct drm_display_mode *native_mode = &amdgpu_encoder->native_mode;
 	int i;
-	static const struct mode_size {
+	int n;
+	struct mode_size {
+		char name[DRM_DISPLAY_MODE_LEN];
 		int w;
 		int h;
-	} common_modes[17] = {
-		{ 640,  480},
-		{ 720,  480},
-		{ 800,  600},
-		{ 848,  480},
-		{1024,  768},
-		{1152,  768},
-		{1280,  720},
-		{1280,  800},
-		{1280,  854},
-		{1280,  960},
-		{1280, 1024},
-		{1440,  900},
-		{1400, 1050},
-		{1680, 1050},
-		{1600, 1200},
-		{1920, 1080},
-		{1920, 1200}
+	} common_modes[] = {
+		{  "640x480",  640,  480},
+		{  "800x600",  800,  600},
+		{ "1024x768", 1024,  768},
+		{ "1280x720", 1280,  720},
+		{ "1280x800", 1280,  800},
+		{"1280x1024", 1280, 1024},
+		{ "1440x900", 1440,  900},
+		{"1680x1050", 1680, 1050},
+		{"1600x1200", 1600, 1200},
+		{"1920x1080", 1920, 1080},
+		{"1920x1200", 1920, 1200}
 	};
 
-	for (i = 0; i < 17; i++) {
+	n = ARRAY_SIZE(common_modes);
+
+	for (i = 0; i < n; i++) {
 		if (amdgpu_encoder->devices & (ATOM_DEVICE_TV_SUPPORT)) {
 			if (common_modes[i].w > 1024 ||
 			    common_modes[i].h > 768)
@@ -475,10 +432,12 @@ static void amdgpu_connector_add_common_modes(struct drm_encoder *encoder,
 			     common_modes[i].h == native_mode->vdisplay))
 				continue;
 		}
-		if (common_modes[i].w < 320 || common_modes[i].h < 200)
-			continue;
 
 		mode = drm_cvt_mode(dev, common_modes[i].w, common_modes[i].h, 60, false, false, false);
+		if (!mode)
+			return;
+		strscpy(mode->name, common_modes[i].name, DRM_DISPLAY_MODE_LEN);
+
 		drm_mode_probed_add(connector, mode);
 	}
 }
@@ -488,7 +447,7 @@ static int amdgpu_connector_set_property(struct drm_connector *connector,
 					  uint64_t val)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct drm_encoder *encoder;
 	struct amdgpu_encoder *amdgpu_encoder;
 
@@ -603,16 +562,26 @@ static int amdgpu_connector_set_property(struct drm_connector *connector,
 			amdgpu_encoder = to_amdgpu_encoder(connector->encoder);
 		} else {
 			const struct drm_connector_helper_funcs *connector_funcs = connector->helper_private;
+
 			amdgpu_encoder = to_amdgpu_encoder(connector_funcs->best_encoder(connector));
 		}
 
 		switch (val) {
 		default:
-		case DRM_MODE_SCALE_NONE: rmx_type = RMX_OFF; break;
-		case DRM_MODE_SCALE_CENTER: rmx_type = RMX_CENTER; break;
-		case DRM_MODE_SCALE_ASPECT: rmx_type = RMX_ASPECT; break;
-		case DRM_MODE_SCALE_FULLSCREEN: rmx_type = RMX_FULL; break;
+		case DRM_MODE_SCALE_NONE:
+			rmx_type = RMX_OFF;
+			break;
+		case DRM_MODE_SCALE_CENTER:
+			rmx_type = RMX_CENTER;
+			break;
+		case DRM_MODE_SCALE_ASPECT:
+			rmx_type = RMX_ASPECT;
+			break;
+		case DRM_MODE_SCALE_FULLSCREEN:
+			rmx_type = RMX_FULL;
+			break;
 		}
+
 		if (amdgpu_encoder->rmx_type == rmx_type)
 			return 0;
 
@@ -641,7 +610,7 @@ amdgpu_connector_fixup_lcd_native_mode(struct drm_encoder *encoder,
 		if (mode->type & DRM_MODE_TYPE_PREFERRED) {
 			if (mode->hdisplay != native_mode->hdisplay ||
 			    mode->vdisplay != native_mode->vdisplay)
-				memcpy(native_mode, mode, sizeof(*mode));
+				drm_mode_copy(native_mode, mode);
 		}
 	}
 
@@ -650,7 +619,7 @@ amdgpu_connector_fixup_lcd_native_mode(struct drm_encoder *encoder,
 		list_for_each_entry_safe(mode, t, &connector->probed_modes, head) {
 			if (mode->hdisplay == native_mode->hdisplay &&
 			    mode->vdisplay == native_mode->vdisplay) {
-				*native_mode = *mode;
+				drm_mode_copy(native_mode, mode);
 				drm_mode_set_crtcinfo(native_mode, CRTC_INTERLACE_HALVE_V);
 				DRM_DEBUG_KMS("Determined LVDS native mode details from EDID\n");
 				break;
@@ -701,8 +670,8 @@ static int amdgpu_connector_lvds_get_modes(struct drm_connector *connector)
 	return ret;
 }
 
-static int amdgpu_connector_lvds_mode_valid(struct drm_connector *connector,
-					     struct drm_display_mode *mode)
+static enum drm_mode_status amdgpu_connector_lvds_mode_valid(struct drm_connector *connector,
+					     const struct drm_display_mode *mode)
 {
 	struct drm_encoder *encoder = amdgpu_connector_best_single_encoder(connector);
 
@@ -739,9 +708,13 @@ amdgpu_connector_lvds_detect(struct drm_connector *connector, bool force)
 	enum drm_connector_status ret = connector_status_disconnected;
 	int r;
 
-	r = pm_runtime_get_sync(connector->dev->dev);
-	if (r < 0)
-		return connector_status_disconnected;
+	if (!drm_kms_helper_is_poll_worker()) {
+		r = pm_runtime_get_sync(connector->dev->dev);
+		if (r < 0) {
+			pm_runtime_put_autosuspend(connector->dev->dev);
+			return connector_status_disconnected;
+		}
+	}
 
 	if (encoder) {
 		struct amdgpu_encoder *amdgpu_encoder = to_amdgpu_encoder(encoder);
@@ -760,8 +733,10 @@ amdgpu_connector_lvds_detect(struct drm_connector *connector, bool force)
 	/* check acpi lid status ??? */
 
 	amdgpu_connector_update_scratch_regs(connector, ret);
-	pm_runtime_mark_last_busy(connector->dev->dev);
-	pm_runtime_put_autosuspend(connector->dev->dev);
+
+	if (!drm_kms_helper_is_poll_worker())
+		pm_runtime_put_autosuspend(connector->dev->dev);
+
 	return ret;
 }
 
@@ -802,16 +777,26 @@ static int amdgpu_connector_set_lcd_property(struct drm_connector *connector,
 		amdgpu_encoder = to_amdgpu_encoder(connector->encoder);
 	else {
 		const struct drm_connector_helper_funcs *connector_funcs = connector->helper_private;
+
 		amdgpu_encoder = to_amdgpu_encoder(connector_funcs->best_encoder(connector));
 	}
 
 	switch (value) {
-	case DRM_MODE_SCALE_NONE: rmx_type = RMX_OFF; break;
-	case DRM_MODE_SCALE_CENTER: rmx_type = RMX_CENTER; break;
-	case DRM_MODE_SCALE_ASPECT: rmx_type = RMX_ASPECT; break;
+	case DRM_MODE_SCALE_NONE:
+		rmx_type = RMX_OFF;
+		break;
+	case DRM_MODE_SCALE_CENTER:
+		rmx_type = RMX_CENTER;
+		break;
+	case DRM_MODE_SCALE_ASPECT:
+		rmx_type = RMX_ASPECT;
+		break;
 	default:
-	case DRM_MODE_SCALE_FULLSCREEN: rmx_type = RMX_FULL; break;
+	case DRM_MODE_SCALE_FULLSCREEN:
+		rmx_type = RMX_FULL;
+		break;
 	}
+
 	if (amdgpu_encoder->rmx_type == rmx_type)
 		return 0;
 
@@ -843,15 +828,16 @@ static int amdgpu_connector_vga_get_modes(struct drm_connector *connector)
 
 	amdgpu_connector_get_edid(connector);
 	ret = amdgpu_connector_ddc_get_modes(connector);
+	amdgpu_get_native_mode(connector);
 
 	return ret;
 }
 
-static int amdgpu_connector_vga_mode_valid(struct drm_connector *connector,
-					    struct drm_display_mode *mode)
+static enum drm_mode_status amdgpu_connector_vga_mode_valid(struct drm_connector *connector,
+					    const struct drm_display_mode *mode)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 
 	/* XXX check mode bandwidth */
 
@@ -871,16 +857,20 @@ amdgpu_connector_vga_detect(struct drm_connector *connector, bool force)
 	enum drm_connector_status ret = connector_status_disconnected;
 	int r;
 
-	r = pm_runtime_get_sync(connector->dev->dev);
-	if (r < 0)
-		return connector_status_disconnected;
+	if (!drm_kms_helper_is_poll_worker()) {
+		r = pm_runtime_get_sync(connector->dev->dev);
+		if (r < 0) {
+			pm_runtime_put_autosuspend(connector->dev->dev);
+			return connector_status_disconnected;
+		}
+	}
 
 	encoder = amdgpu_connector_best_single_encoder(connector);
 	if (!encoder)
 		ret = connector_status_disconnected;
 
 	if (amdgpu_connector->ddc_bus)
-		dret = amdgpu_ddc_probe(amdgpu_connector, false);
+		dret = amdgpu_display_ddc_probe(amdgpu_connector, false);
 	if (dret) {
 		amdgpu_connector->detected_by_load = false;
 		amdgpu_connector_free_edid(connector);
@@ -927,8 +917,8 @@ amdgpu_connector_vga_detect(struct drm_connector *connector, bool force)
 	amdgpu_connector_update_scratch_regs(connector, ret);
 
 out:
-	pm_runtime_mark_last_busy(connector->dev->dev);
-	pm_runtime_put_autosuspend(connector->dev->dev);
+	if (!drm_kms_helper_is_poll_worker())
+		pm_runtime_put_autosuspend(connector->dev->dev);
 
 	return ret;
 }
@@ -952,7 +942,7 @@ static bool
 amdgpu_connector_check_hpd_status_unchanged(struct drm_connector *connector)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 	enum drm_connector_status status;
 
@@ -968,6 +958,41 @@ amdgpu_connector_check_hpd_status_unchanged(struct drm_connector *connector)
 	return false;
 }
 
+static void amdgpu_connector_shared_ddc(enum drm_connector_status *status,
+					struct drm_connector *connector,
+					struct amdgpu_connector *amdgpu_connector)
+{
+	struct drm_connector *list_connector;
+	struct drm_connector_list_iter iter;
+	struct amdgpu_connector *list_amdgpu_connector;
+	struct drm_device *dev = connector->dev;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+
+	if (amdgpu_connector->shared_ddc && *status == connector_status_connected) {
+		drm_connector_list_iter_begin(dev, &iter);
+		drm_for_each_connector_iter(list_connector,
+					    &iter) {
+			if (connector == list_connector)
+				continue;
+			list_amdgpu_connector = to_amdgpu_connector(list_connector);
+			if (list_amdgpu_connector->shared_ddc &&
+			    list_amdgpu_connector->ddc_bus->rec.i2c_id ==
+			     amdgpu_connector->ddc_bus->rec.i2c_id) {
+				/* cases where both connectors are digital */
+				if (list_connector->connector_type != DRM_MODE_CONNECTOR_VGA) {
+					/* hpd is our only option in this case */
+					if (!amdgpu_display_hpd_sense(adev,
+								      amdgpu_connector->hpd.hpd)) {
+						amdgpu_connector_free_edid(connector);
+						*status = connector_status_disconnected;
+					}
+				}
+			}
+		}
+		drm_connector_list_iter_end(&iter);
+	}
+}
+
 /*
  * DVI is complicated
  * Do a DDC probe, if DDC probe passes, get the full EDID so
@@ -983,25 +1008,48 @@ static enum drm_connector_status
 amdgpu_connector_dvi_detect(struct drm_connector *connector, bool force)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
-	struct drm_encoder *encoder = NULL;
 	const struct drm_encoder_helper_funcs *encoder_funcs;
-	int i, r;
+	int r;
 	enum drm_connector_status ret = connector_status_disconnected;
 	bool dret = false, broken_edid = false;
 
-	r = pm_runtime_get_sync(connector->dev->dev);
-	if (r < 0)
-		return connector_status_disconnected;
+	if (!drm_kms_helper_is_poll_worker()) {
+		r = pm_runtime_get_sync(connector->dev->dev);
+		if (r < 0) {
+			pm_runtime_put_autosuspend(connector->dev->dev);
+			return connector_status_disconnected;
+		}
+	}
+
+	if (amdgpu_connector->detected_hpd_without_ddc) {
+		force = true;
+		amdgpu_connector->detected_hpd_without_ddc = false;
+	}
 
 	if (!force && amdgpu_connector_check_hpd_status_unchanged(connector)) {
 		ret = connector->status;
 		goto exit;
 	}
 
-	if (amdgpu_connector->ddc_bus)
-		dret = amdgpu_ddc_probe(amdgpu_connector, false);
+	if (amdgpu_connector->ddc_bus) {
+		dret = amdgpu_display_ddc_probe(amdgpu_connector, false);
+
+		/* Sometimes the pins required for the DDC probe on DVI
+		 * connectors don't make contact at the same time that the ones
+		 * for HPD do. If the DDC probe fails even though we had an HPD
+		 * signal, try again later
+		 */
+		if (!dret && !force &&
+		    amdgpu_display_hpd_sense(adev, amdgpu_connector->hpd.hpd)) {
+			DRM_DEBUG_KMS("hpd detected without ddc, retrying in 1 second\n");
+			amdgpu_connector->detected_hpd_without_ddc = true;
+			schedule_delayed_work(&adev->hotplug_work,
+					      msecs_to_jiffies(1000));
+			goto exit;
+		}
+	}
 	if (dret) {
 		amdgpu_connector->detected_by_load = false;
 		amdgpu_connector_free_edid(connector);
@@ -1031,27 +1079,7 @@ amdgpu_connector_dvi_detect(struct drm_connector *connector, bool force)
 			 * DDC line.  The latter is more complex because with DVI<->HDMI adapters
 			 * you don't really know what's connected to which port as both are digital.
 			 */
-			if (amdgpu_connector->shared_ddc && (ret == connector_status_connected)) {
-				struct drm_connector *list_connector;
-				struct amdgpu_connector *list_amdgpu_connector;
-				list_for_each_entry(list_connector, &dev->mode_config.connector_list, head) {
-					if (connector == list_connector)
-						continue;
-					list_amdgpu_connector = to_amdgpu_connector(list_connector);
-					if (list_amdgpu_connector->shared_ddc &&
-					    (list_amdgpu_connector->ddc_bus->rec.i2c_id ==
-					     amdgpu_connector->ddc_bus->rec.i2c_id)) {
-						/* cases where both connectors are digital */
-						if (list_connector->connector_type != DRM_MODE_CONNECTOR_VGA) {
-							/* hpd is our only option in this case */
-							if (!amdgpu_display_hpd_sense(adev, amdgpu_connector->hpd.hpd)) {
-								amdgpu_connector_free_edid(connector);
-								ret = connector_status_disconnected;
-							}
-						}
-					}
-				}
-			}
+			amdgpu_connector_shared_ddc(&ret, connector, amdgpu_connector);
 		}
 	}
 
@@ -1075,14 +1103,9 @@ amdgpu_connector_dvi_detect(struct drm_connector *connector, bool force)
 
 	/* find analog encoder */
 	if (amdgpu_connector->dac_load_detect) {
-		for (i = 0; i < DRM_CONNECTOR_MAX_ENCODER; i++) {
-			if (connector->encoder_ids[i] == 0)
-				break;
-
-			encoder = drm_encoder_find(connector->dev, connector->encoder_ids[i]);
-			if (!encoder)
-				continue;
+		struct drm_encoder *encoder;
 
+		drm_connector_for_each_possible_encoder(connector, encoder) {
 			if (encoder->encoder_type != DRM_MODE_ENCODER_DAC &&
 			    encoder->encoder_type != DRM_MODE_ENCODER_TVDAC)
 				continue;
@@ -1104,7 +1127,8 @@ amdgpu_connector_dvi_detect(struct drm_connector *connector, bool force)
 					/* assume digital unless load detected otherwise */
 					amdgpu_connector->use_digital = true;
 					lret = encoder_funcs->detect(encoder, connector);
-					DRM_DEBUG_KMS("load_detect %x returned: %x\n",encoder->encoder_type,lret);
+					DRM_DEBUG_KMS("load_detect %x returned: %x\n",
+						      encoder->encoder_type, lret);
 					if (lret == connector_status_connected)
 						amdgpu_connector->use_digital = false;
 				}
@@ -1118,8 +1142,8 @@ out:
 	amdgpu_connector_update_scratch_regs(connector, ret);
 
 exit:
-	pm_runtime_mark_last_busy(connector->dev->dev);
-	pm_runtime_put_autosuspend(connector->dev->dev);
+	if (!drm_kms_helper_is_poll_worker())
+		pm_runtime_put_autosuspend(connector->dev->dev);
 
 	return ret;
 }
@@ -1128,18 +1152,10 @@ exit:
 static struct drm_encoder *
 amdgpu_connector_dvi_encoder(struct drm_connector *connector)
 {
-	int enc_id = connector->encoder_ids[0];
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 	struct drm_encoder *encoder;
-	int i;
-	for (i = 0; i < DRM_CONNECTOR_MAX_ENCODER; i++) {
-		if (connector->encoder_ids[i] == 0)
-			break;
-
-		encoder = drm_encoder_find(connector->dev, connector->encoder_ids[i]);
-		if (!encoder)
-			continue;
 
+	drm_connector_for_each_possible_encoder(connector, encoder) {
 		if (amdgpu_connector->use_digital == true) {
 			if (encoder->encoder_type == DRM_MODE_ENCODER_TMDS)
 				return encoder;
@@ -1154,43 +1170,85 @@ amdgpu_connector_dvi_encoder(struct drm_connector *connector)
 
 	/* then check use digitial */
 	/* pick the first one */
-	if (enc_id)
-		return drm_encoder_find(connector->dev, enc_id);
+	drm_connector_for_each_possible_encoder(connector, encoder)
+		return encoder;
+
 	return NULL;
 }
 
 static void amdgpu_connector_dvi_force(struct drm_connector *connector)
 {
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
+
 	if (connector->force == DRM_FORCE_ON)
 		amdgpu_connector->use_digital = false;
 	if (connector->force == DRM_FORCE_ON_DIGITAL)
 		amdgpu_connector->use_digital = true;
 }
 
-static int amdgpu_connector_dvi_mode_valid(struct drm_connector *connector,
-					    struct drm_display_mode *mode)
+/**
+ * amdgpu_max_hdmi_pixel_clock - Return max supported HDMI (TMDS) pixel clock
+ * @adev: pointer to amdgpu_device
+ *
+ * Return: maximum supported HDMI (TMDS) pixel clock in KHz.
+ */
+static int amdgpu_max_hdmi_pixel_clock(const struct amdgpu_device *adev)
+{
+	if (adev->asic_type >= CHIP_POLARIS10)
+		return 600000;
+	else if (adev->asic_type >= CHIP_TONGA)
+		return 300000;
+	else
+		return 297000;
+}
+
+/**
+ * amdgpu_connector_dvi_mode_valid - Validate a mode on DVI/HDMI connectors
+ * @connector: DRM connector to validate the mode on
+ * @mode: display mode to validate
+ *
+ * Validate the given display mode on DVI and HDMI connectors, including
+ * analog signals on DVI-I.
+ *
+ * Return: drm_mode_status indicating whether the mode is valid.
+ */
+static enum drm_mode_status amdgpu_connector_dvi_mode_valid(struct drm_connector *connector,
+					    const struct drm_display_mode *mode)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
+	const int max_hdmi_pixel_clock = amdgpu_max_hdmi_pixel_clock(adev);
+	const int max_dvi_single_link_pixel_clock = 165000;
+	int max_digital_pixel_clock_khz;
 
 	/* XXX check mode bandwidth */
 
-	if (amdgpu_connector->use_digital && (mode->clock > 165000)) {
-		if ((amdgpu_connector->connector_object_id == CONNECTOR_OBJECT_ID_DUAL_LINK_DVI_I) ||
-		    (amdgpu_connector->connector_object_id == CONNECTOR_OBJECT_ID_DUAL_LINK_DVI_D) ||
-		    (amdgpu_connector->connector_object_id == CONNECTOR_OBJECT_ID_HDMI_TYPE_B)) {
-			return MODE_OK;
-		} else if (drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
-			/* HDMI 1.3+ supports max clock of 340 Mhz */
-			if (mode->clock > 340000)
-				return MODE_CLOCK_HIGH;
-			else
-				return MODE_OK;
-		} else {
-			return MODE_CLOCK_HIGH;
+	if (amdgpu_connector->use_digital) {
+		switch (amdgpu_connector->connector_object_id) {
+		case CONNECTOR_OBJECT_ID_HDMI_TYPE_A:
+			max_digital_pixel_clock_khz = max_hdmi_pixel_clock;
+			break;
+		case CONNECTOR_OBJECT_ID_SINGLE_LINK_DVI_I:
+		case CONNECTOR_OBJECT_ID_SINGLE_LINK_DVI_D:
+			max_digital_pixel_clock_khz = max_dvi_single_link_pixel_clock;
+			break;
+		case CONNECTOR_OBJECT_ID_DUAL_LINK_DVI_I:
+		case CONNECTOR_OBJECT_ID_DUAL_LINK_DVI_D:
+		case CONNECTOR_OBJECT_ID_HDMI_TYPE_B:
+			max_digital_pixel_clock_khz = max_dvi_single_link_pixel_clock * 2;
+			break;
 		}
+
+		/* When the display EDID claims that it's an HDMI display,
+		 * we use the HDMI encoder mode of the display HW,
+		 * so we should verify against the max HDMI clock here.
+		 */
+		if (connector->display_info.is_hdmi)
+			max_digital_pixel_clock_khz = max_hdmi_pixel_clock;
+
+		if (mode->clock > max_digital_pixel_clock_khz)
+			return MODE_CLOCK_HIGH;
 	}
 
 	/* check against the max pixel clock */
@@ -1290,17 +1348,8 @@ u16 amdgpu_connector_encoder_get_dp_bridge_encoder_id(struct drm_connector *conn
 {
 	struct drm_encoder *encoder;
 	struct amdgpu_encoder *amdgpu_encoder;
-	int i;
-
-	for (i = 0; i < DRM_CONNECTOR_MAX_ENCODER; i++) {
-		if (connector->encoder_ids[i] == 0)
-			break;
-
-		encoder = drm_encoder_find(connector->dev,
-					connector->encoder_ids[i]);
-		if (!encoder)
-			continue;
 
+	drm_connector_for_each_possible_encoder(connector, encoder) {
 		amdgpu_encoder = to_amdgpu_encoder(encoder);
 
 		switch (amdgpu_encoder->encoder_id) {
@@ -1319,17 +1368,9 @@ static bool amdgpu_connector_encoder_is_hbr2(struct drm_connector *connector)
 {
 	struct drm_encoder *encoder;
 	struct amdgpu_encoder *amdgpu_encoder;
-	int i;
 	bool found = false;
 
-	for (i = 0; i < DRM_CONNECTOR_MAX_ENCODER; i++) {
-		if (connector->encoder_ids[i] == 0)
-			break;
-		encoder = drm_encoder_find(connector->dev,
-					connector->encoder_ids[i]);
-		if (!encoder)
-			continue;
-
+	drm_connector_for_each_possible_encoder(connector, encoder) {
 		amdgpu_encoder = to_amdgpu_encoder(encoder);
 		if (amdgpu_encoder->caps & ATOM_ENCODER_CAP_RECORD_HBR2)
 			found = true;
@@ -1341,7 +1382,7 @@ static bool amdgpu_connector_encoder_is_hbr2(struct drm_connector *connector)
 bool amdgpu_connector_is_dp12_capable(struct drm_connector *connector)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 
 	if ((adev->clock.default_dispclk >= 53900) &&
 	    amdgpu_connector_encoder_is_hbr2(connector)) {
@@ -1355,16 +1396,20 @@ static enum drm_connector_status
 amdgpu_connector_dp_detect(struct drm_connector *connector, bool force)
 {
 	struct drm_device *dev = connector->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 	enum drm_connector_status ret = connector_status_disconnected;
 	struct amdgpu_connector_atom_dig *amdgpu_dig_connector = amdgpu_connector->con_priv;
 	struct drm_encoder *encoder = amdgpu_connector_best_single_encoder(connector);
 	int r;
 
-	r = pm_runtime_get_sync(connector->dev->dev);
-	if (r < 0)
-		return connector_status_disconnected;
+	if (!drm_kms_helper_is_poll_worker()) {
+		r = pm_runtime_get_sync(connector->dev->dev);
+		if (r < 0) {
+			pm_runtime_put_autosuspend(connector->dev->dev);
+			return connector_status_disconnected;
+		}
+	}
 
 	if (!force && amdgpu_connector_check_hpd_status_unchanged(connector)) {
 		ret = connector->status;
@@ -1404,10 +1449,12 @@ amdgpu_connector_dp_detect(struct drm_connector *connector, bool force)
 			/* setup ddc on the bridge */
 			amdgpu_atombios_encoder_setup_ext_encoder_ddc(encoder);
 			/* bridge chips are always aux */
-			if (amdgpu_ddc_probe(amdgpu_connector, true)) /* try DDC */
+			/* try DDC */
+			if (amdgpu_display_ddc_probe(amdgpu_connector, true))
 				ret = connector_status_connected;
 			else if (amdgpu_connector->dac_load_detect) { /* try load detection */
 				const struct drm_encoder_helper_funcs *encoder_funcs = encoder->helper_private;
+
 				ret = encoder_funcs->detect(encoder, connector);
 			}
 		}
@@ -1424,7 +1471,8 @@ amdgpu_connector_dp_detect(struct drm_connector *connector, bool force)
 					ret = connector_status_connected;
 			} else {
 				/* try non-aux ddc (DP to DVI/HDMI/etc. adapter) */
-				if (amdgpu_ddc_probe(amdgpu_connector, false))
+				if (amdgpu_display_ddc_probe(amdgpu_connector,
+							     false))
 					ret = connector_status_connected;
 			}
 		}
@@ -1432,14 +1480,20 @@ amdgpu_connector_dp_detect(struct drm_connector *connector, bool force)
 
 	amdgpu_connector_update_scratch_regs(connector, ret);
 out:
-	pm_runtime_mark_last_busy(connector->dev->dev);
-	pm_runtime_put_autosuspend(connector->dev->dev);
-
+	if (!drm_kms_helper_is_poll_worker())
+		pm_runtime_put_autosuspend(connector->dev->dev);
+
+	if (connector->connector_type == DRM_MODE_CONNECTOR_DisplayPort ||
+	    connector->connector_type == DRM_MODE_CONNECTOR_eDP)
+		drm_dp_set_subconnector_property(&amdgpu_connector->base,
+						 ret,
+						 amdgpu_dig_connector->dpcd,
+						 amdgpu_dig_connector->downstream_ports);
 	return ret;
 }
 
-static int amdgpu_connector_dp_mode_valid(struct drm_connector *connector,
-					   struct drm_display_mode *mode)
+static enum drm_mode_status amdgpu_connector_dp_mode_valid(struct drm_connector *connector,
+					   const struct drm_display_mode *mode)
 {
 	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
 	struct amdgpu_connector_atom_dig *amdgpu_dig_connector = amdgpu_connector->con_priv;
@@ -1477,7 +1531,7 @@ static int amdgpu_connector_dp_mode_valid(struct drm_connector *connector,
 		    (amdgpu_dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_eDP)) {
 			return amdgpu_atombios_dp_mode_valid_helper(connector, mode);
 		} else {
-			if (drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
+			if (connector->display_info.is_hdmi) {
 				/* HDMI 1.3+ supports max clock of 340 Mhz */
 				if (mode->clock > 340000)
 					return MODE_CLOCK_HIGH;
@@ -1491,6 +1545,20 @@ static int amdgpu_connector_dp_mode_valid(struct drm_connector *connector,
 	return MODE_OK;
 }
 
+static int
+amdgpu_connector_late_register(struct drm_connector *connector)
+{
+	struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
+	int r = 0;
+
+	if (amdgpu_connector->ddc_bus->has_aux) {
+		amdgpu_connector->ddc_bus->aux.dev = amdgpu_connector->base.kdev;
+		r = drm_dp_aux_register(&amdgpu_connector->ddc_bus->aux);
+	}
+
+	return r;
+}
+
 static const struct drm_connector_helper_funcs amdgpu_connector_dp_helper_funcs = {
 	.get_modes = amdgpu_connector_dp_get_modes,
 	.mode_valid = amdgpu_connector_dp_mode_valid,
@@ -1505,6 +1573,7 @@ static const struct drm_connector_funcs amdgpu_connector_dp_funcs = {
 	.early_unregister = amdgpu_connector_unregister,
 	.destroy = amdgpu_connector_destroy,
 	.force = amdgpu_connector_dvi_force,
+	.late_register = amdgpu_connector_late_register,
 };
 
 static const struct drm_connector_funcs amdgpu_connector_edp_funcs = {
@@ -1515,6 +1584,7 @@ static const struct drm_connector_funcs amdgpu_connector_edp_funcs = {
 	.early_unregister = amdgpu_connector_unregister,
 	.destroy = amdgpu_connector_destroy,
 	.force = amdgpu_connector_dvi_force,
+	.late_register = amdgpu_connector_late_register,
 };
 
 void
@@ -1527,12 +1597,14 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 		      struct amdgpu_hpd *hpd,
 		      struct amdgpu_router *router)
 {
-	struct drm_device *dev = adev->ddev;
+	struct drm_device *dev = adev_to_drm(adev);
 	struct drm_connector *connector;
+	struct drm_connector_list_iter iter;
 	struct amdgpu_connector *amdgpu_connector;
 	struct amdgpu_connector_atom_dig *amdgpu_dig_connector;
 	struct drm_encoder *encoder;
 	struct amdgpu_encoder *amdgpu_encoder;
+	struct i2c_adapter *ddc = NULL;
 	uint32_t subpixel_order = SubPixelNone;
 	bool shared_ddc = false;
 	bool is_dp_bridge = false;
@@ -1542,10 +1614,12 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 		return;
 
 	/* see if we already added it */
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
+	drm_connector_list_iter_begin(dev, &iter);
+	drm_for_each_connector_iter(connector, &iter) {
 		amdgpu_connector = to_amdgpu_connector(connector);
 		if (amdgpu_connector->connector_id == connector_id) {
 			amdgpu_connector->devices |= supported_device;
+			drm_connector_list_iter_end(&iter);
 			return;
 		}
 		if (amdgpu_connector->ddc_bus && i2c_bus->valid) {
@@ -1560,6 +1634,7 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			}
 		}
 	}
+	drm_connector_list_iter_end(&iter);
 
 	/* check if it's a dp bridge */
 	list_for_each_entry(encoder, &dev->mode_config.encoder_list, head) {
@@ -1602,17 +1677,21 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 		amdgpu_connector->con_priv = amdgpu_dig_connector;
 		if (i2c_bus->valid) {
 			amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
-			if (amdgpu_connector->ddc_bus)
+			if (amdgpu_connector->ddc_bus) {
 				has_aux = true;
-			else
+				ddc = &amdgpu_connector->ddc_bus->adapter;
+			} else {
 				DRM_ERROR("DP: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+			}
 		}
 		switch (connector_type) {
 		case DRM_MODE_CONNECTOR_VGA:
 		case DRM_MODE_CONNECTOR_DVIA:
 		default:
-			drm_connector_init(dev, &amdgpu_connector->base,
-					   &amdgpu_connector_dp_funcs, connector_type);
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_dp_funcs,
+						    connector_type,
+						    ddc);
 			drm_connector_helper_add(&amdgpu_connector->base,
 						 &amdgpu_connector_dp_helper_funcs);
 			connector->interlace_allowed = true;
@@ -1630,8 +1709,10 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 		case DRM_MODE_CONNECTOR_HDMIA:
 		case DRM_MODE_CONNECTOR_HDMIB:
 		case DRM_MODE_CONNECTOR_DisplayPort:
-			drm_connector_init(dev, &amdgpu_connector->base,
-					   &amdgpu_connector_dp_funcs, connector_type);
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_dp_funcs,
+						    connector_type,
+						    ddc);
 			drm_connector_helper_add(&amdgpu_connector->base,
 						 &amdgpu_connector_dp_helper_funcs);
 			drm_object_attach_property(&amdgpu_connector->base.base,
@@ -1652,10 +1733,12 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 						   adev->mode_info.dither_property,
 						   AMDGPU_FMT_DITHER_DISABLE);
 
-			if (amdgpu_audio != 0)
+			if (amdgpu_audio != 0) {
 				drm_object_attach_property(&amdgpu_connector->base.base,
 							   adev->mode_info.audio_property,
 							   AMDGPU_AUDIO_AUTO);
+				amdgpu_connector->audio = AMDGPU_AUDIO_AUTO;
+			}
 
 			subpixel_order = SubPixelHorizontalRGB;
 			connector->interlace_allowed = true;
@@ -1672,8 +1755,10 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			break;
 		case DRM_MODE_CONNECTOR_LVDS:
 		case DRM_MODE_CONNECTOR_eDP:
-			drm_connector_init(dev, &amdgpu_connector->base,
-					   &amdgpu_connector_edp_funcs, connector_type);
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_edp_funcs,
+						    connector_type,
+						    ddc);
 			drm_connector_helper_add(&amdgpu_connector->base,
 						 &amdgpu_connector_dp_helper_funcs);
 			drm_object_attach_property(&amdgpu_connector->base.base,
@@ -1687,13 +1772,18 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 	} else {
 		switch (connector_type) {
 		case DRM_MODE_CONNECTOR_VGA:
-			drm_connector_init(dev, &amdgpu_connector->base, &amdgpu_connector_vga_funcs, connector_type);
-			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_vga_helper_funcs);
 			if (i2c_bus->valid) {
 				amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
 				if (!amdgpu_connector->ddc_bus)
 					DRM_ERROR("VGA: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+				else
+					ddc = &amdgpu_connector->ddc_bus->adapter;
 			}
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_vga_funcs,
+						    connector_type,
+						    ddc);
+			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_vga_helper_funcs);
 			amdgpu_connector->dac_load_detect = true;
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						      adev->mode_info.load_detect_property,
@@ -1707,13 +1797,18 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			connector->doublescan_allowed = true;
 			break;
 		case DRM_MODE_CONNECTOR_DVIA:
-			drm_connector_init(dev, &amdgpu_connector->base, &amdgpu_connector_vga_funcs, connector_type);
-			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_vga_helper_funcs);
 			if (i2c_bus->valid) {
 				amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
 				if (!amdgpu_connector->ddc_bus)
 					DRM_ERROR("DVIA: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+				else
+					ddc = &amdgpu_connector->ddc_bus->adapter;
 			}
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_vga_funcs,
+						    connector_type,
+						    ddc);
+			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_vga_helper_funcs);
 			amdgpu_connector->dac_load_detect = true;
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						      adev->mode_info.load_detect_property,
@@ -1732,13 +1827,18 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			if (!amdgpu_dig_connector)
 				goto failed;
 			amdgpu_connector->con_priv = amdgpu_dig_connector;
-			drm_connector_init(dev, &amdgpu_connector->base, &amdgpu_connector_dvi_funcs, connector_type);
-			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dvi_helper_funcs);
 			if (i2c_bus->valid) {
 				amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
 				if (!amdgpu_connector->ddc_bus)
 					DRM_ERROR("DVI: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+				else
+					ddc = &amdgpu_connector->ddc_bus->adapter;
 			}
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_dvi_funcs,
+						    connector_type,
+						    ddc);
+			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dvi_helper_funcs);
 			subpixel_order = SubPixelHorizontalRGB;
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						      adev->mode_info.coherent_mode_property,
@@ -1760,6 +1860,7 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 				drm_object_attach_property(&amdgpu_connector->base.base,
 							   adev->mode_info.audio_property,
 							   AMDGPU_AUDIO_AUTO);
+				amdgpu_connector->audio = AMDGPU_AUDIO_AUTO;
 			}
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						   adev->mode_info.dither_property,
@@ -1782,13 +1883,18 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			if (!amdgpu_dig_connector)
 				goto failed;
 			amdgpu_connector->con_priv = amdgpu_dig_connector;
-			drm_connector_init(dev, &amdgpu_connector->base, &amdgpu_connector_dvi_funcs, connector_type);
-			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dvi_helper_funcs);
 			if (i2c_bus->valid) {
 				amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
 				if (!amdgpu_connector->ddc_bus)
 					DRM_ERROR("HDMI: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+				else
+					ddc = &amdgpu_connector->ddc_bus->adapter;
 			}
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_dvi_funcs,
+						    connector_type,
+						    ddc);
+			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dvi_helper_funcs);
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						      adev->mode_info.coherent_mode_property,
 						      1);
@@ -1808,6 +1914,7 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 				drm_object_attach_property(&amdgpu_connector->base.base,
 							   adev->mode_info.audio_property,
 							   AMDGPU_AUDIO_AUTO);
+				amdgpu_connector->audio = AMDGPU_AUDIO_AUTO;
 			}
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						   adev->mode_info.dither_property,
@@ -1824,15 +1931,20 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			if (!amdgpu_dig_connector)
 				goto failed;
 			amdgpu_connector->con_priv = amdgpu_dig_connector;
-			drm_connector_init(dev, &amdgpu_connector->base, &amdgpu_connector_dp_funcs, connector_type);
-			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dp_helper_funcs);
 			if (i2c_bus->valid) {
 				amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
-				if (amdgpu_connector->ddc_bus)
+				if (amdgpu_connector->ddc_bus) {
 					has_aux = true;
-				else
+					ddc = &amdgpu_connector->ddc_bus->adapter;
+				} else {
 					DRM_ERROR("DP: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+				}
 			}
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_dp_funcs,
+						    connector_type,
+						    ddc);
+			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dp_helper_funcs);
 			subpixel_order = SubPixelHorizontalRGB;
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						      adev->mode_info.coherent_mode_property,
@@ -1853,6 +1965,7 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 				drm_object_attach_property(&amdgpu_connector->base.base,
 							   adev->mode_info.audio_property,
 							   AMDGPU_AUDIO_AUTO);
+				amdgpu_connector->audio = AMDGPU_AUDIO_AUTO;
 			}
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						   adev->mode_info.dither_property,
@@ -1866,15 +1979,20 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			if (!amdgpu_dig_connector)
 				goto failed;
 			amdgpu_connector->con_priv = amdgpu_dig_connector;
-			drm_connector_init(dev, &amdgpu_connector->base, &amdgpu_connector_edp_funcs, connector_type);
-			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dp_helper_funcs);
 			if (i2c_bus->valid) {
 				amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
-				if (amdgpu_connector->ddc_bus)
+				if (amdgpu_connector->ddc_bus) {
 					has_aux = true;
-				else
+					ddc = &amdgpu_connector->ddc_bus->adapter;
+				} else {
 					DRM_ERROR("DP: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+				}
 			}
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_edp_funcs,
+						    connector_type,
+						    ddc);
+			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_dp_helper_funcs);
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						      dev->mode_config.scaling_mode_property,
 						      DRM_MODE_SCALE_FULLSCREEN);
@@ -1887,13 +2005,18 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 			if (!amdgpu_dig_connector)
 				goto failed;
 			amdgpu_connector->con_priv = amdgpu_dig_connector;
-			drm_connector_init(dev, &amdgpu_connector->base, &amdgpu_connector_lvds_funcs, connector_type);
-			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_lvds_helper_funcs);
 			if (i2c_bus->valid) {
 				amdgpu_connector->ddc_bus = amdgpu_i2c_lookup(adev, i2c_bus);
 				if (!amdgpu_connector->ddc_bus)
 					DRM_ERROR("LVDS: Failed to assign ddc bus! Check dmesg for i2c errors.\n");
+				else
+					ddc = &amdgpu_connector->ddc_bus->adapter;
 			}
+			drm_connector_init_with_ddc(dev, &amdgpu_connector->base,
+						    &amdgpu_connector_lvds_funcs,
+						    connector_type,
+						    ddc);
+			drm_connector_helper_add(&amdgpu_connector->base, &amdgpu_connector_lvds_helper_funcs);
 			drm_object_attach_property(&amdgpu_connector->base.base,
 						      dev->mode_config.scaling_mode_property,
 						      DRM_MODE_SCALE_FULLSCREEN);
@@ -1907,17 +2030,21 @@ amdgpu_connector_add(struct amdgpu_device *adev,
 	if (amdgpu_connector->hpd.hpd == AMDGPU_HPD_NONE) {
 		if (i2c_bus->valid) {
 			connector->polled = DRM_CONNECTOR_POLL_CONNECT |
-			                    DRM_CONNECTOR_POLL_DISCONNECT;
+						DRM_CONNECTOR_POLL_DISCONNECT;
 		}
 	} else
 		connector->polled = DRM_CONNECTOR_POLL_HPD;
 
 	connector->display_info.subpixel_order = subpixel_order;
-	drm_connector_register(connector);
 
 	if (has_aux)
 		amdgpu_atombios_dp_aux_init(amdgpu_connector);
 
+	if (connector_type == DRM_MODE_CONNECTOR_DisplayPort ||
+	    connector_type == DRM_MODE_CONNECTOR_eDP) {
+		drm_connector_attach_dp_subconnector_property(&amdgpu_connector->base);
+	}
+
 	return;
 
 failed:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.h
index 61fcef15ad72..eff833b6ed31 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.h
@@ -24,7 +24,6 @@
 #ifndef __AMDGPU_CONNECTORS_H__
 #define __AMDGPU_CONNECTORS_H__
 
-struct edid *amdgpu_connector_edid(struct drm_connector *connector);
 void amdgpu_connector_hotplug(struct drm_connector *connector);
 int amdgpu_connector_get_monitor_bpc(struct drm_connector *connector);
 u16 amdgpu_connector_encoder_get_dp_bridge_encoder_id(struct drm_connector *connector);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cper.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cper.c
new file mode 100644
index 000000000000..425a3e564360
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cper.c
@@ -0,0 +1,591 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2025 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include <linux/list.h>
+#include "amdgpu.h"
+
+static const guid_t MCE			= CPER_NOTIFY_MCE;
+static const guid_t CMC			= CPER_NOTIFY_CMC;
+static const guid_t BOOT		= BOOT_TYPE;
+
+static const guid_t CRASHDUMP		= AMD_CRASHDUMP;
+static const guid_t RUNTIME		= AMD_GPU_NONSTANDARD_ERROR;
+
+static void __inc_entry_length(struct cper_hdr *hdr, uint32_t size)
+{
+	hdr->record_length += size;
+}
+
+static void amdgpu_cper_get_timestamp(struct cper_timestamp *timestamp)
+{
+	struct tm tm;
+	time64_t now = ktime_get_real_seconds();
+
+	time64_to_tm(now, 0, &tm);
+	timestamp->seconds = tm.tm_sec;
+	timestamp->minutes = tm.tm_min;
+	timestamp->hours = tm.tm_hour;
+	timestamp->flag = 0;
+	timestamp->day = tm.tm_mday;
+	timestamp->month = 1 + tm.tm_mon;
+	timestamp->year = (1900 + tm.tm_year) % 100;
+	timestamp->century = (1900 + tm.tm_year) / 100;
+}
+
+void amdgpu_cper_entry_fill_hdr(struct amdgpu_device *adev,
+				struct cper_hdr *hdr,
+				enum amdgpu_cper_type type,
+				enum cper_error_severity sev)
+{
+	char record_id[16];
+
+	hdr->signature[0]		= 'C';
+	hdr->signature[1]		= 'P';
+	hdr->signature[2]		= 'E';
+	hdr->signature[3]		= 'R';
+	hdr->revision			= CPER_HDR_REV_1;
+	hdr->signature_end		= 0xFFFFFFFF;
+	hdr->error_severity		= sev;
+
+	hdr->valid_bits.platform_id	= 1;
+	hdr->valid_bits.timestamp	= 1;
+
+	amdgpu_cper_get_timestamp(&hdr->timestamp);
+
+	snprintf(record_id, 9, "%d:%X",
+		 (adev->smuio.funcs && adev->smuio.funcs->get_socket_id) ?
+			 adev->smuio.funcs->get_socket_id(adev) :
+			 0,
+		 atomic_inc_return(&adev->cper.unique_id));
+	memcpy(hdr->record_id, record_id, 8);
+
+	snprintf(hdr->platform_id, 16, "0x%04X:0x%04X",
+		 adev->pdev->vendor, adev->pdev->device);
+	/* pmfw version should be part of creator_id according to CPER spec */
+	snprintf(hdr->creator_id, 16, "%s", CPER_CREATOR_ID_AMDGPU);
+
+	switch (type) {
+	case AMDGPU_CPER_TYPE_BOOT:
+		hdr->notify_type = BOOT;
+		break;
+	case AMDGPU_CPER_TYPE_FATAL:
+	case AMDGPU_CPER_TYPE_BP_THRESHOLD:
+		hdr->notify_type = MCE;
+		break;
+	case AMDGPU_CPER_TYPE_RUNTIME:
+		if (sev == CPER_SEV_NON_FATAL_CORRECTED)
+			hdr->notify_type = CMC;
+		else
+			hdr->notify_type = MCE;
+		break;
+	default:
+		dev_err(adev->dev, "Unknown CPER Type\n");
+		break;
+	}
+
+	__inc_entry_length(hdr, HDR_LEN);
+}
+
+static int amdgpu_cper_entry_fill_section_desc(struct amdgpu_device *adev,
+					       struct cper_sec_desc *section_desc,
+					       bool bp_threshold,
+					       bool poison,
+					       enum cper_error_severity sev,
+					       guid_t sec_type,
+					       uint32_t section_length,
+					       uint32_t section_offset)
+{
+	section_desc->revision_minor		= CPER_SEC_MINOR_REV_1;
+	section_desc->revision_major		= CPER_SEC_MAJOR_REV_22;
+	section_desc->sec_offset		= section_offset;
+	section_desc->sec_length		= section_length;
+	section_desc->valid_bits.fru_text	= 1;
+	section_desc->flag_bits.primary		= 1;
+	section_desc->severity			= sev;
+	section_desc->sec_type			= sec_type;
+
+	snprintf(section_desc->fru_text, 20, "OAM%d",
+		 (adev->smuio.funcs && adev->smuio.funcs->get_socket_id) ?
+			 adev->smuio.funcs->get_socket_id(adev) :
+			 0);
+
+	if (bp_threshold)
+		section_desc->flag_bits.exceed_err_threshold = 1;
+	if (poison)
+		section_desc->flag_bits.latent_err = 1;
+
+	return 0;
+}
+
+int amdgpu_cper_entry_fill_fatal_section(struct amdgpu_device *adev,
+					 struct cper_hdr *hdr,
+					 uint32_t idx,
+					 struct cper_sec_crashdump_reg_data reg_data)
+{
+	struct cper_sec_desc *section_desc;
+	struct cper_sec_crashdump_fatal *section;
+
+	section_desc = (struct cper_sec_desc *)((uint8_t *)hdr + SEC_DESC_OFFSET(idx));
+	section = (struct cper_sec_crashdump_fatal *)((uint8_t *)hdr +
+		   FATAL_SEC_OFFSET(hdr->sec_cnt, idx));
+
+	amdgpu_cper_entry_fill_section_desc(adev, section_desc, false, false,
+					    CPER_SEV_FATAL, CRASHDUMP, FATAL_SEC_LEN,
+					    FATAL_SEC_OFFSET(hdr->sec_cnt, idx));
+
+	section->body.reg_ctx_type = CPER_CTX_TYPE_CRASH;
+	section->body.reg_arr_size = sizeof(reg_data);
+	section->body.data = reg_data;
+
+	__inc_entry_length(hdr, SEC_DESC_LEN + FATAL_SEC_LEN);
+
+	return 0;
+}
+
+int amdgpu_cper_entry_fill_runtime_section(struct amdgpu_device *adev,
+					   struct cper_hdr *hdr,
+					   uint32_t idx,
+					   enum cper_error_severity sev,
+					   uint32_t *reg_dump,
+					   uint32_t reg_count)
+{
+	struct cper_sec_desc *section_desc;
+	struct cper_sec_nonstd_err *section;
+	bool poison;
+
+	poison = sev != CPER_SEV_NON_FATAL_CORRECTED;
+	section_desc = (struct cper_sec_desc *)((uint8_t *)hdr + SEC_DESC_OFFSET(idx));
+	section = (struct cper_sec_nonstd_err *)((uint8_t *)hdr +
+		   NONSTD_SEC_OFFSET(hdr->sec_cnt, idx));
+
+	amdgpu_cper_entry_fill_section_desc(adev, section_desc, false, poison,
+					    sev, RUNTIME, NONSTD_SEC_LEN,
+					    NONSTD_SEC_OFFSET(hdr->sec_cnt, idx));
+
+	reg_count = umin(reg_count, CPER_ACA_REG_COUNT);
+
+	section->hdr.valid_bits.err_info_cnt = 1;
+	section->hdr.valid_bits.err_context_cnt = 1;
+
+	section->info.error_type = RUNTIME;
+	section->info.ms_chk_bits.err_type_valid = 1;
+	section->ctx.reg_ctx_type = CPER_CTX_TYPE_CRASH;
+	section->ctx.reg_arr_size = sizeof(section->ctx.reg_dump);
+
+	memcpy(section->ctx.reg_dump, reg_dump, reg_count * sizeof(uint32_t));
+
+	__inc_entry_length(hdr, SEC_DESC_LEN + NONSTD_SEC_LEN);
+
+	return 0;
+}
+
+int amdgpu_cper_entry_fill_bad_page_threshold_section(struct amdgpu_device *adev,
+						      struct cper_hdr *hdr,
+						      uint32_t idx)
+{
+	struct cper_sec_desc *section_desc;
+	struct cper_sec_nonstd_err *section;
+	uint32_t socket_id;
+
+	section_desc = (struct cper_sec_desc *)((uint8_t *)hdr + SEC_DESC_OFFSET(idx));
+	section = (struct cper_sec_nonstd_err *)((uint8_t *)hdr +
+		   NONSTD_SEC_OFFSET(hdr->sec_cnt, idx));
+
+	amdgpu_cper_entry_fill_section_desc(adev, section_desc, true, false,
+					    CPER_SEV_FATAL, RUNTIME, NONSTD_SEC_LEN,
+					    NONSTD_SEC_OFFSET(hdr->sec_cnt, idx));
+
+	section->hdr.valid_bits.err_info_cnt = 1;
+	section->hdr.valid_bits.err_context_cnt = 1;
+
+	section->info.error_type = RUNTIME;
+	section->info.valid_bits.ms_chk = 1;
+	section->info.ms_chk_bits.err_type_valid = 1;
+	section->info.ms_chk_bits.err_type = 1;
+	section->info.ms_chk_bits.pcc = 1;
+	section->ctx.reg_ctx_type = CPER_CTX_TYPE_CRASH;
+	section->ctx.reg_arr_size = sizeof(section->ctx.reg_dump);
+
+	/* Hardcoded Reg dump for bad page threshold CPER */
+	socket_id = (adev->smuio.funcs && adev->smuio.funcs->get_socket_id) ?
+				adev->smuio.funcs->get_socket_id(adev) :
+				0;
+	section->ctx.reg_dump[CPER_ACA_REG_CTL_LO]    = 0x1;
+	section->ctx.reg_dump[CPER_ACA_REG_CTL_HI]    = 0x0;
+	section->ctx.reg_dump[CPER_ACA_REG_STATUS_LO] = 0x137;
+	section->ctx.reg_dump[CPER_ACA_REG_STATUS_HI] = 0xB0000000;
+	section->ctx.reg_dump[CPER_ACA_REG_ADDR_LO]   = 0x0;
+	section->ctx.reg_dump[CPER_ACA_REG_ADDR_HI]   = 0x0;
+	section->ctx.reg_dump[CPER_ACA_REG_MISC0_LO]  = 0x0;
+	section->ctx.reg_dump[CPER_ACA_REG_MISC0_HI]  = 0x0;
+	section->ctx.reg_dump[CPER_ACA_REG_CONFIG_LO] = 0x2;
+	section->ctx.reg_dump[CPER_ACA_REG_CONFIG_HI] = 0x1ff;
+	section->ctx.reg_dump[CPER_ACA_REG_IPID_LO]   = (socket_id / 4) & 0x01;
+	section->ctx.reg_dump[CPER_ACA_REG_IPID_HI]   = 0x096 | (((socket_id % 4) & 0x3) << 12);
+	section->ctx.reg_dump[CPER_ACA_REG_SYND_LO]   = 0x0;
+	section->ctx.reg_dump[CPER_ACA_REG_SYND_HI]   = 0x0;
+
+	__inc_entry_length(hdr, SEC_DESC_LEN + NONSTD_SEC_LEN);
+
+	return 0;
+}
+
+struct cper_hdr *amdgpu_cper_alloc_entry(struct amdgpu_device *adev,
+					 enum amdgpu_cper_type type,
+					 uint16_t section_count)
+{
+	struct cper_hdr *hdr;
+	uint32_t size = 0;
+
+	size += HDR_LEN;
+	size += (SEC_DESC_LEN * section_count);
+
+	switch (type) {
+	case AMDGPU_CPER_TYPE_RUNTIME:
+	case AMDGPU_CPER_TYPE_BP_THRESHOLD:
+		size += (NONSTD_SEC_LEN * section_count);
+		break;
+	case AMDGPU_CPER_TYPE_FATAL:
+		size += (FATAL_SEC_LEN * section_count);
+		break;
+	case AMDGPU_CPER_TYPE_BOOT:
+		size += (BOOT_SEC_LEN * section_count);
+		break;
+	default:
+		dev_err(adev->dev, "Unknown CPER Type!\n");
+		return NULL;
+	}
+
+	hdr = kzalloc(size, GFP_KERNEL);
+	if (!hdr)
+		return NULL;
+
+	/* Save this early */
+	hdr->sec_cnt = section_count;
+
+	return hdr;
+}
+
+int amdgpu_cper_generate_ue_record(struct amdgpu_device *adev,
+				   struct aca_bank *bank)
+{
+	struct cper_hdr *fatal = NULL;
+	struct cper_sec_crashdump_reg_data reg_data = { 0 };
+	struct amdgpu_ring *ring = &adev->cper.ring_buf;
+	int ret;
+
+	fatal = amdgpu_cper_alloc_entry(adev, AMDGPU_CPER_TYPE_FATAL, 1);
+	if (!fatal) {
+		dev_err(adev->dev, "fail to alloc cper entry for ue record\n");
+		return -ENOMEM;
+	}
+
+	reg_data.status_lo = lower_32_bits(bank->regs[ACA_REG_IDX_STATUS]);
+	reg_data.status_hi = upper_32_bits(bank->regs[ACA_REG_IDX_STATUS]);
+	reg_data.addr_lo   = lower_32_bits(bank->regs[ACA_REG_IDX_ADDR]);
+	reg_data.addr_hi   = upper_32_bits(bank->regs[ACA_REG_IDX_ADDR]);
+	reg_data.ipid_lo   = lower_32_bits(bank->regs[ACA_REG_IDX_IPID]);
+	reg_data.ipid_hi   = upper_32_bits(bank->regs[ACA_REG_IDX_IPID]);
+	reg_data.synd_lo   = lower_32_bits(bank->regs[ACA_REG_IDX_SYND]);
+	reg_data.synd_hi   = upper_32_bits(bank->regs[ACA_REG_IDX_SYND]);
+
+	amdgpu_cper_entry_fill_hdr(adev, fatal, AMDGPU_CPER_TYPE_FATAL, CPER_SEV_FATAL);
+	ret = amdgpu_cper_entry_fill_fatal_section(adev, fatal, 0, reg_data);
+	if (ret)
+		return ret;
+
+	amdgpu_cper_ring_write(ring, fatal, fatal->record_length);
+	kfree(fatal);
+
+	return 0;
+}
+
+int amdgpu_cper_generate_bp_threshold_record(struct amdgpu_device *adev)
+{
+	struct cper_hdr *bp_threshold = NULL;
+	struct amdgpu_ring *ring = &adev->cper.ring_buf;
+	int ret;
+
+	bp_threshold = amdgpu_cper_alloc_entry(adev, AMDGPU_CPER_TYPE_BP_THRESHOLD, 1);
+	if (!bp_threshold) {
+		dev_err(adev->dev, "fail to alloc cper entry for bad page threshold record\n");
+		return -ENOMEM;
+	}
+
+	amdgpu_cper_entry_fill_hdr(adev, bp_threshold,
+				   AMDGPU_CPER_TYPE_BP_THRESHOLD,
+				   CPER_SEV_FATAL);
+	ret = amdgpu_cper_entry_fill_bad_page_threshold_section(adev, bp_threshold, 0);
+	if (ret)
+		return ret;
+
+	amdgpu_cper_ring_write(ring, bp_threshold, bp_threshold->record_length);
+	kfree(bp_threshold);
+
+	return 0;
+}
+
+static enum cper_error_severity amdgpu_aca_err_type_to_cper_sev(struct amdgpu_device *adev,
+								enum aca_error_type aca_err_type)
+{
+	switch (aca_err_type) {
+	case ACA_ERROR_TYPE_UE:
+		return CPER_SEV_FATAL;
+	case ACA_ERROR_TYPE_CE:
+		return CPER_SEV_NON_FATAL_CORRECTED;
+	case ACA_ERROR_TYPE_DEFERRED:
+		return CPER_SEV_NON_FATAL_UNCORRECTED;
+	default:
+		dev_err(adev->dev, "Unknown ACA error type!\n");
+		return CPER_SEV_FATAL;
+	}
+}
+
+int amdgpu_cper_generate_ce_records(struct amdgpu_device *adev,
+				    struct aca_banks *banks,
+				    uint16_t bank_count)
+{
+	struct cper_hdr *corrected = NULL;
+	enum cper_error_severity sev = CPER_SEV_NON_FATAL_CORRECTED;
+	struct amdgpu_ring *ring = &adev->cper.ring_buf;
+	uint32_t reg_data[CPER_ACA_REG_COUNT] = { 0 };
+	struct aca_bank_node *node;
+	struct aca_bank *bank;
+	uint32_t i = 0;
+	int ret;
+
+	corrected = amdgpu_cper_alloc_entry(adev, AMDGPU_CPER_TYPE_RUNTIME, bank_count);
+	if (!corrected) {
+		dev_err(adev->dev, "fail to allocate cper entry for ce records\n");
+		return -ENOMEM;
+	}
+
+	/* Raise severity if any DE is detected in the ACA bank list */
+	list_for_each_entry(node, &banks->list, node) {
+		bank = &node->bank;
+		if (bank->aca_err_type == ACA_ERROR_TYPE_DEFERRED) {
+			sev = CPER_SEV_NON_FATAL_UNCORRECTED;
+			break;
+		}
+	}
+
+	amdgpu_cper_entry_fill_hdr(adev, corrected, AMDGPU_CPER_TYPE_RUNTIME, sev);
+
+	/* Combine CE and DE in cper record */
+	list_for_each_entry(node, &banks->list, node) {
+		bank = &node->bank;
+		reg_data[CPER_ACA_REG_CTL_LO]    = lower_32_bits(bank->regs[ACA_REG_IDX_CTL]);
+		reg_data[CPER_ACA_REG_CTL_HI]    = upper_32_bits(bank->regs[ACA_REG_IDX_CTL]);
+		reg_data[CPER_ACA_REG_STATUS_LO] = lower_32_bits(bank->regs[ACA_REG_IDX_STATUS]);
+		reg_data[CPER_ACA_REG_STATUS_HI] = upper_32_bits(bank->regs[ACA_REG_IDX_STATUS]);
+		reg_data[CPER_ACA_REG_ADDR_LO]   = lower_32_bits(bank->regs[ACA_REG_IDX_ADDR]);
+		reg_data[CPER_ACA_REG_ADDR_HI]   = upper_32_bits(bank->regs[ACA_REG_IDX_ADDR]);
+		reg_data[CPER_ACA_REG_MISC0_LO]  = lower_32_bits(bank->regs[ACA_REG_IDX_MISC0]);
+		reg_data[CPER_ACA_REG_MISC0_HI]  = upper_32_bits(bank->regs[ACA_REG_IDX_MISC0]);
+		reg_data[CPER_ACA_REG_CONFIG_LO] = lower_32_bits(bank->regs[ACA_REG_IDX_CONFIG]);
+		reg_data[CPER_ACA_REG_CONFIG_HI] = upper_32_bits(bank->regs[ACA_REG_IDX_CONFIG]);
+		reg_data[CPER_ACA_REG_IPID_LO]   = lower_32_bits(bank->regs[ACA_REG_IDX_IPID]);
+		reg_data[CPER_ACA_REG_IPID_HI]   = upper_32_bits(bank->regs[ACA_REG_IDX_IPID]);
+		reg_data[CPER_ACA_REG_SYND_LO]   = lower_32_bits(bank->regs[ACA_REG_IDX_SYND]);
+		reg_data[CPER_ACA_REG_SYND_HI]   = upper_32_bits(bank->regs[ACA_REG_IDX_SYND]);
+
+		ret = amdgpu_cper_entry_fill_runtime_section(adev, corrected, i++,
+				amdgpu_aca_err_type_to_cper_sev(adev, bank->aca_err_type),
+				reg_data, CPER_ACA_REG_COUNT);
+		if (ret)
+			return ret;
+	}
+
+	amdgpu_cper_ring_write(ring, corrected, corrected->record_length);
+	kfree(corrected);
+
+	return 0;
+}
+
+static bool amdgpu_cper_is_hdr(struct amdgpu_ring *ring, u64 pos)
+{
+	struct cper_hdr *chdr;
+
+	chdr = (struct cper_hdr *)&(ring->ring[pos]);
+	return strcmp(chdr->signature, "CPER") ? false : true;
+}
+
+static u32 amdgpu_cper_ring_get_ent_sz(struct amdgpu_ring *ring, u64 pos)
+{
+	struct cper_hdr *chdr;
+	u64 p;
+	u32 chunk, rec_len = 0;
+
+	chdr = (struct cper_hdr *)&(ring->ring[pos]);
+	chunk = ring->ring_size - (pos << 2);
+
+	if (!strcmp(chdr->signature, "CPER")) {
+		rec_len = chdr->record_length;
+		goto calc;
+	}
+
+	/* ring buffer is not full, no cper data after ring->wptr */
+	if (ring->count_dw)
+		goto calc;
+
+	for (p = pos + 1; p <= ring->buf_mask; p++) {
+		chdr = (struct cper_hdr *)&(ring->ring[p]);
+		if (!strcmp(chdr->signature, "CPER")) {
+			rec_len = (p - pos) << 2;
+			goto calc;
+		}
+	}
+
+calc:
+	if (!rec_len)
+		return chunk;
+	else
+		return umin(rec_len, chunk);
+}
+
+void amdgpu_cper_ring_write(struct amdgpu_ring *ring, void *src, int count)
+{
+	u64 pos, wptr_old, rptr;
+	int rec_cnt_dw = count >> 2;
+	u32 chunk, ent_sz;
+	u8 *s = (u8 *)src;
+
+	if (count >= ring->ring_size - 4) {
+		dev_err(ring->adev->dev,
+			"CPER data size(%d) is larger than ring size(%d)\n",
+			count, ring->ring_size - 4);
+
+		return;
+	}
+
+	mutex_lock(&ring->adev->cper.ring_lock);
+
+	wptr_old = ring->wptr;
+	rptr = *ring->rptr_cpu_addr & ring->ptr_mask;
+
+	while (count) {
+		ent_sz = amdgpu_cper_ring_get_ent_sz(ring, ring->wptr);
+		chunk = umin(ent_sz, count);
+
+		memcpy(&ring->ring[ring->wptr], s, chunk);
+
+		ring->wptr += (chunk >> 2);
+		ring->wptr &= ring->ptr_mask;
+		count -= chunk;
+		s += chunk;
+	}
+
+	if (ring->count_dw < rec_cnt_dw)
+		ring->count_dw = 0;
+
+	/* the buffer is overflow, adjust rptr */
+	if (((wptr_old < rptr) && (rptr <= ring->wptr)) ||
+	    ((ring->wptr < wptr_old) && (wptr_old < rptr)) ||
+	    ((rptr <= ring->wptr) && (ring->wptr < wptr_old))) {
+		pos = (ring->wptr + 1) & ring->ptr_mask;
+
+		do {
+			ent_sz = amdgpu_cper_ring_get_ent_sz(ring, pos);
+
+			rptr += (ent_sz >> 2);
+			rptr &= ring->ptr_mask;
+			*ring->rptr_cpu_addr = rptr;
+
+			pos = rptr;
+		} while (!amdgpu_cper_is_hdr(ring, rptr));
+	}
+
+	if (ring->count_dw >= rec_cnt_dw)
+		ring->count_dw -= rec_cnt_dw;
+	mutex_unlock(&ring->adev->cper.ring_lock);
+}
+
+static u64 amdgpu_cper_ring_get_rptr(struct amdgpu_ring *ring)
+{
+	return *(ring->rptr_cpu_addr);
+}
+
+static u64 amdgpu_cper_ring_get_wptr(struct amdgpu_ring *ring)
+{
+	return ring->wptr;
+}
+
+static const struct amdgpu_ring_funcs cper_ring_funcs = {
+	.type = AMDGPU_RING_TYPE_CPER,
+	.align_mask = 0xff,
+	.support_64bit_ptrs = false,
+	.get_rptr = amdgpu_cper_ring_get_rptr,
+	.get_wptr = amdgpu_cper_ring_get_wptr,
+};
+
+static int amdgpu_cper_ring_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_ring *ring = &(adev->cper.ring_buf);
+
+	mutex_init(&adev->cper.ring_lock);
+
+	ring->adev = NULL;
+	ring->ring_obj = NULL;
+	ring->use_doorbell = false;
+	ring->no_scheduler = true;
+	ring->funcs = &cper_ring_funcs;
+
+	sprintf(ring->name, "cper");
+	return amdgpu_ring_init(adev, ring, CPER_MAX_RING_SIZE, NULL, 0,
+				AMDGPU_RING_PRIO_DEFAULT, NULL);
+}
+
+int amdgpu_cper_init(struct amdgpu_device *adev)
+{
+	int r;
+
+	if (!amdgpu_aca_is_enabled(adev) && !amdgpu_sriov_ras_cper_en(adev))
+		return 0;
+
+	r = amdgpu_cper_ring_init(adev);
+	if (r) {
+		dev_err(adev->dev, "failed to initialize cper ring, r = %d\n", r);
+		return r;
+	}
+
+	mutex_init(&adev->cper.cper_lock);
+
+	adev->cper.enabled = true;
+	adev->cper.max_count = CPER_MAX_ALLOWED_COUNT;
+
+	return 0;
+}
+
+int amdgpu_cper_fini(struct amdgpu_device *adev)
+{
+	if (!amdgpu_aca_is_enabled(adev) && !amdgpu_sriov_ras_cper_en(adev))
+		return 0;
+
+	adev->cper.enabled = false;
+
+	amdgpu_ring_fini(&(adev->cper.ring_buf));
+	adev->cper.count = 0;
+	adev->cper.wptr = 0;
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cper.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_cper.h
new file mode 100644
index 000000000000..353421807387
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cper.h
@@ -0,0 +1,105 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright 2025 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_CPER_H__
+#define __AMDGPU_CPER_H__
+
+#include "amd_cper.h"
+#include "amdgpu_aca.h"
+
+#define CPER_MAX_ALLOWED_COUNT		0x1000
+#define CPER_MAX_RING_SIZE		0X100000
+#define HDR_LEN				(sizeof(struct cper_hdr))
+#define SEC_DESC_LEN			(sizeof(struct cper_sec_desc))
+
+#define BOOT_SEC_LEN			(sizeof(struct cper_sec_crashdump_boot))
+#define FATAL_SEC_LEN			(sizeof(struct cper_sec_crashdump_fatal))
+#define NONSTD_SEC_LEN			(sizeof(struct cper_sec_nonstd_err))
+
+#define SEC_DESC_OFFSET(idx)		(HDR_LEN + (SEC_DESC_LEN * idx))
+
+#define BOOT_SEC_OFFSET(count, idx)	(HDR_LEN + (SEC_DESC_LEN * count) + (BOOT_SEC_LEN * idx))
+#define FATAL_SEC_OFFSET(count, idx)	(HDR_LEN + (SEC_DESC_LEN * count) + (FATAL_SEC_LEN * idx))
+#define NONSTD_SEC_OFFSET(count, idx)	(HDR_LEN + (SEC_DESC_LEN * count) + (NONSTD_SEC_LEN * idx))
+
+enum amdgpu_cper_type {
+	AMDGPU_CPER_TYPE_RUNTIME,
+	AMDGPU_CPER_TYPE_FATAL,
+	AMDGPU_CPER_TYPE_BOOT,
+	AMDGPU_CPER_TYPE_BP_THRESHOLD,
+};
+
+struct amdgpu_cper {
+	bool enabled;
+
+	atomic_t unique_id;
+	struct mutex cper_lock;
+
+	/* Lifetime CPERs generated */
+	uint32_t count;
+	uint32_t max_count;
+
+	uint32_t wptr;
+
+	void *ring[CPER_MAX_ALLOWED_COUNT];
+	struct amdgpu_ring ring_buf;
+	struct mutex ring_lock;
+};
+
+void amdgpu_cper_entry_fill_hdr(struct amdgpu_device *adev,
+				struct cper_hdr *hdr,
+				enum amdgpu_cper_type type,
+				enum cper_error_severity sev);
+int amdgpu_cper_entry_fill_fatal_section(struct amdgpu_device *adev,
+					 struct cper_hdr *hdr,
+					 uint32_t idx,
+					 struct cper_sec_crashdump_reg_data reg_data);
+int amdgpu_cper_entry_fill_runtime_section(struct amdgpu_device *adev,
+					   struct cper_hdr *hdr,
+					   uint32_t idx,
+					   enum cper_error_severity sev,
+					   uint32_t *reg_dump,
+					   uint32_t reg_count);
+int amdgpu_cper_entry_fill_bad_page_threshold_section(struct amdgpu_device *adev,
+						      struct cper_hdr *hdr,
+						      uint32_t section_idx);
+
+struct cper_hdr *amdgpu_cper_alloc_entry(struct amdgpu_device *adev,
+					 enum amdgpu_cper_type type,
+					 uint16_t section_count);
+/* UE must be encoded into separated cper entries, 1 UE 1 cper */
+int amdgpu_cper_generate_ue_record(struct amdgpu_device *adev,
+				   struct aca_bank *bank);
+/* CEs and DEs are combined into 1 cper entry */
+int amdgpu_cper_generate_ce_records(struct amdgpu_device *adev,
+				    struct aca_banks *banks,
+				    uint16_t bank_count);
+/* Bad page threshold is encoded into separated cper entry */
+int amdgpu_cper_generate_bp_threshold_record(struct amdgpu_device *adev);
+void amdgpu_cper_ring_write(struct amdgpu_ring *ring,
+			void *src, int count);
+int amdgpu_cper_init(struct amdgpu_device *adev);
+int amdgpu_cper_fini(struct amdgpu_device *adev);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 4e6b9501ab0a..ecdfe6cb36cc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -24,87 +24,108 @@
  * Authors:
  *    Jerome Glisse <glisse@freedesktop.org>
  */
+
+#include <linux/file.h>
 #include <linux/pagemap.h>
-#include <drm/drmP.h>
+#include <linux/sync_file.h>
+#include <linux/dma-buf.h>
+
 #include <drm/amdgpu_drm.h>
+#include <drm/drm_syncobj.h>
+#include <drm/ttm/ttm_tt.h>
+
+#include "amdgpu_cs.h"
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
-
-int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
-		       u32 ip_instance, u32 ring,
-		       struct amdgpu_ring **out_ring)
+#include "amdgpu_gmc.h"
+#include "amdgpu_gem.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_hmm.h"
+
+static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p,
+				 struct amdgpu_device *adev,
+				 struct drm_file *filp,
+				 union drm_amdgpu_cs *cs)
 {
-	/* Right now all IPs have only one instance - multiple rings. */
-	if (ip_instance != 0) {
-		DRM_ERROR("invalid ip instance: %d\n", ip_instance);
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
+
+	if (cs->in.num_chunks == 0)
 		return -EINVAL;
-	}
 
-	switch (ip_type) {
-	default:
-		DRM_ERROR("unknown ip type: %d\n", ip_type);
+	memset(p, 0, sizeof(*p));
+	p->adev = adev;
+	p->filp = filp;
+
+	p->ctx = amdgpu_ctx_get(fpriv, cs->in.ctx_id);
+	if (!p->ctx)
 		return -EINVAL;
-	case AMDGPU_HW_IP_GFX:
-		if (ring < adev->gfx.num_gfx_rings) {
-			*out_ring = &adev->gfx.gfx_ring[ring];
-		} else {
-			DRM_ERROR("only %d gfx rings are supported now\n",
-				  adev->gfx.num_gfx_rings);
-			return -EINVAL;
-		}
-		break;
-	case AMDGPU_HW_IP_COMPUTE:
-		if (ring < adev->gfx.num_compute_rings) {
-			*out_ring = &adev->gfx.compute_ring[ring];
-		} else {
-			DRM_ERROR("only %d compute rings are supported now\n",
-				  adev->gfx.num_compute_rings);
-			return -EINVAL;
-		}
-		break;
-	case AMDGPU_HW_IP_DMA:
-		if (ring < adev->sdma.num_instances) {
-			*out_ring = &adev->sdma.instance[ring].ring;
-		} else {
-			DRM_ERROR("only %d SDMA rings are supported\n",
-				  adev->sdma.num_instances);
-			return -EINVAL;
-		}
-		break;
-	case AMDGPU_HW_IP_UVD:
-		*out_ring = &adev->uvd.ring;
-		break;
-	case AMDGPU_HW_IP_VCE:
-		if (ring < adev->vce.num_rings){
-			*out_ring = &adev->vce.ring[ring];
-		} else {
-			DRM_ERROR("only %d VCE rings are supported\n", adev->vce.num_rings);
-			return -EINVAL;
-		}
-		break;
-	case AMDGPU_HW_IP_UVD_ENC:
-		if (ring < adev->uvd.num_enc_rings){
-			*out_ring = &adev->uvd.ring_enc[ring];
-		} else {
-			DRM_ERROR("only %d UVD ENC rings are supported\n",
-				adev->uvd.num_enc_rings);
-			return -EINVAL;
-		}
-		break;
+
+	if (atomic_read(&p->ctx->guilty)) {
+		amdgpu_ctx_put(p->ctx);
+		return -ECANCELED;
 	}
 
-	if (!(*out_ring && (*out_ring)->adev)) {
-		DRM_ERROR("Ring %d is not initialized on IP %d\n",
-			  ring, ip_type);
+	amdgpu_sync_create(&p->sync);
+	drm_exec_init(&p->exec, DRM_EXEC_INTERRUPTIBLE_WAIT |
+		      DRM_EXEC_IGNORE_DUPLICATES, 0);
+	return 0;
+}
+
+static int amdgpu_cs_job_idx(struct amdgpu_cs_parser *p,
+			     struct drm_amdgpu_cs_chunk_ib *chunk_ib)
+{
+	struct drm_sched_entity *entity;
+	unsigned int i;
+	int r;
+
+	r = amdgpu_ctx_get_entity(p->ctx, chunk_ib->ip_type,
+				  chunk_ib->ip_instance,
+				  chunk_ib->ring, &entity);
+	if (r)
+		return r;
+
+	/*
+	 * Abort if there is no run queue associated with this entity.
+	 * Possibly because of disabled HW IP.
+	 */
+	if (entity->rq == NULL)
 		return -EINVAL;
-	}
 
+	/* Check if we can add this IB to some existing job */
+	for (i = 0; i < p->gang_size; ++i)
+		if (p->entities[i] == entity)
+			return i;
+
+	/* If not increase the gang size if possible */
+	if (i == AMDGPU_CS_GANG_SIZE)
+		return -EINVAL;
+
+	p->entities[i] = entity;
+	p->gang_size = i + 1;
+	return i;
+}
+
+static int amdgpu_cs_p1_ib(struct amdgpu_cs_parser *p,
+			   struct drm_amdgpu_cs_chunk_ib *chunk_ib,
+			   unsigned int *num_ibs)
+{
+	int r;
+
+	r = amdgpu_cs_job_idx(p, chunk_ib);
+	if (r < 0)
+		return r;
+
+	if (num_ibs[r] >= amdgpu_ring_max_ibs(chunk_ib->ip_type))
+		return -EINVAL;
+
+	++(num_ibs[r]);
+	p->gang_leader_idx = r;
 	return 0;
 }
 
-static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
-				      struct drm_amdgpu_cs_chunk_fence *data,
-				      uint32_t *offset)
+static int amdgpu_cs_p1_user_fence(struct amdgpu_cs_parser *p,
+				   struct drm_amdgpu_cs_chunk_fence *data,
+				   uint32_t *offset)
 {
 	struct drm_gem_object *gobj;
 	unsigned long size;
@@ -113,75 +134,76 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
 	if (gobj == NULL)
 		return -EINVAL;
 
-	p->uf_entry.robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
-	p->uf_entry.priority = 0;
-	p->uf_entry.tv.bo = &p->uf_entry.robj->tbo;
-	p->uf_entry.tv.shared = true;
-	p->uf_entry.user_pages = NULL;
+	p->uf_bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
+	drm_gem_object_put(gobj);
 
-	size = amdgpu_bo_size(p->uf_entry.robj);
-	if (size != PAGE_SIZE || (data->offset + 8) > size)
+	size = amdgpu_bo_size(p->uf_bo);
+	if (size != PAGE_SIZE || data->offset > (size - 8))
+		return -EINVAL;
+
+	if (amdgpu_ttm_tt_get_usermm(p->uf_bo->tbo.ttm))
 		return -EINVAL;
 
 	*offset = data->offset;
+	return 0;
+}
+
+static int amdgpu_cs_p1_bo_handles(struct amdgpu_cs_parser *p,
+				   struct drm_amdgpu_bo_list_in *data)
+{
+	struct drm_amdgpu_bo_list_entry *info;
+	int r;
 
-	drm_gem_object_unreference_unlocked(gobj);
+	r = amdgpu_bo_create_list_entry_array(data, &info);
+	if (r)
+		return r;
 
-	if (amdgpu_ttm_tt_get_usermm(p->uf_entry.robj->tbo.ttm)) {
-		amdgpu_bo_unref(&p->uf_entry.robj);
-		return -EINVAL;
-	}
+	r = amdgpu_bo_list_create(p->adev, p->filp, info, data->bo_number,
+				  &p->bo_list);
+	if (r)
+		goto error_free;
 
+	kvfree(info);
 	return 0;
+
+error_free:
+	kvfree(info);
+
+	return r;
 }
 
-int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
+/* Copy the data from userspace and go over it the first time */
+static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
+			   union drm_amdgpu_cs *cs)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	unsigned int num_ibs[AMDGPU_CS_GANG_SIZE] = { };
 	struct amdgpu_vm *vm = &fpriv->vm;
-	union drm_amdgpu_cs *cs = data;
-	uint64_t *chunk_array_user;
 	uint64_t *chunk_array;
-	unsigned size, num_ibs = 0;
 	uint32_t uf_offset = 0;
-	int i;
+	size_t size;
 	int ret;
+	int i;
 
-	if (cs->in.num_chunks == 0)
-		return 0;
-
-	chunk_array = kmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
-	if (!chunk_array)
-		return -ENOMEM;
-
-	p->ctx = amdgpu_ctx_get(fpriv, cs->in.ctx_id);
-	if (!p->ctx) {
-		ret = -EINVAL;
-		goto free_chunk;
-	}
-
-	/* get chunks */
-	chunk_array_user = (uint64_t __user *)(uintptr_t)(cs->in.chunks);
-	if (copy_from_user(chunk_array, chunk_array_user,
-			   sizeof(uint64_t)*cs->in.num_chunks)) {
-		ret = -EFAULT;
-		goto put_ctx;
-	}
+	chunk_array = memdup_array_user(u64_to_user_ptr(cs->in.chunks),
+					cs->in.num_chunks,
+					sizeof(uint64_t));
+	if (IS_ERR(chunk_array))
+		return PTR_ERR(chunk_array);
 
 	p->nchunks = cs->in.num_chunks;
-	p->chunks = kmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
+	p->chunks = kvmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
 			    GFP_KERNEL);
 	if (!p->chunks) {
 		ret = -ENOMEM;
-		goto put_ctx;
+		goto free_chunk;
 	}
 
 	for (i = 0; i < p->nchunks; i++) {
-		struct drm_amdgpu_cs_chunk __user **chunk_ptr = NULL;
+		struct drm_amdgpu_cs_chunk __user *chunk_ptr = NULL;
 		struct drm_amdgpu_cs_chunk user_chunk;
-		uint32_t __user *cdata;
 
-		chunk_ptr = (void __user *)(uintptr_t)chunk_array[i];
+		chunk_ptr = u64_to_user_ptr(chunk_array[i]);
 		if (copy_from_user(&user_chunk, chunk_ptr,
 				       sizeof(struct drm_amdgpu_cs_chunk))) {
 			ret = -EFAULT;
@@ -192,73 +214,452 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		p->chunks[i].length_dw = user_chunk.length_dw;
 
 		size = p->chunks[i].length_dw;
-		cdata = (void __user *)(uintptr_t)user_chunk.chunk_data;
 
-		p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
-		if (p->chunks[i].kdata == NULL) {
-			ret = -ENOMEM;
+		p->chunks[i].kdata = vmemdup_array_user(u64_to_user_ptr(user_chunk.chunk_data),
+							size,
+							sizeof(uint32_t));
+		if (IS_ERR(p->chunks[i].kdata)) {
+			ret = PTR_ERR(p->chunks[i].kdata);
 			i--;
 			goto free_partial_kdata;
 		}
 		size *= sizeof(uint32_t);
-		if (copy_from_user(p->chunks[i].kdata, cdata, size)) {
-			ret = -EFAULT;
-			goto free_partial_kdata;
-		}
 
+		/* Assume the worst on the following checks */
+		ret = -EINVAL;
 		switch (p->chunks[i].chunk_id) {
 		case AMDGPU_CHUNK_ID_IB:
-			++num_ibs;
+			if (size < sizeof(struct drm_amdgpu_cs_chunk_ib))
+				goto free_partial_kdata;
+
+			ret = amdgpu_cs_p1_ib(p, p->chunks[i].kdata, num_ibs);
+			if (ret)
+				goto free_partial_kdata;
 			break;
 
 		case AMDGPU_CHUNK_ID_FENCE:
-			size = sizeof(struct drm_amdgpu_cs_chunk_fence);
-			if (p->chunks[i].length_dw * sizeof(uint32_t) < size) {
-				ret = -EINVAL;
+			if (size < sizeof(struct drm_amdgpu_cs_chunk_fence))
 				goto free_partial_kdata;
-			}
 
-			ret = amdgpu_cs_user_fence_chunk(p, p->chunks[i].kdata,
-							 &uf_offset);
+			ret = amdgpu_cs_p1_user_fence(p, p->chunks[i].kdata,
+						      &uf_offset);
 			if (ret)
 				goto free_partial_kdata;
+			break;
+
+		case AMDGPU_CHUNK_ID_BO_HANDLES:
+			if (size < sizeof(struct drm_amdgpu_bo_list_in))
+				goto free_partial_kdata;
 
+			/* Only a single BO list is allowed to simplify handling. */
+			if (p->bo_list)
+				goto free_partial_kdata;
+
+			ret = amdgpu_cs_p1_bo_handles(p, p->chunks[i].kdata);
+			if (ret)
+				goto free_partial_kdata;
 			break;
 
 		case AMDGPU_CHUNK_ID_DEPENDENCIES:
+		case AMDGPU_CHUNK_ID_SYNCOBJ_IN:
+		case AMDGPU_CHUNK_ID_SYNCOBJ_OUT:
+		case AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES:
+		case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_WAIT:
+		case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_SIGNAL:
+		case AMDGPU_CHUNK_ID_CP_GFX_SHADOW:
 			break;
 
 		default:
-			ret = -EINVAL;
 			goto free_partial_kdata;
 		}
 	}
 
-	ret = amdgpu_job_alloc(p->adev, num_ibs, &p->job, vm);
-	if (ret)
+	if (!p->gang_size || (amdgpu_sriov_vf(p->adev) && p->gang_size > 1)) {
+		ret = -EINVAL;
+		goto free_all_kdata;
+	}
+
+	for (i = 0; i < p->gang_size; ++i) {
+		ret = amdgpu_job_alloc(p->adev, vm, p->entities[i], vm,
+				       num_ibs[i], &p->jobs[i],
+				       p->filp->client_id);
+		if (ret)
+			goto free_all_kdata;
+		switch (p->adev->enforce_isolation[fpriv->xcp_id]) {
+		case AMDGPU_ENFORCE_ISOLATION_DISABLE:
+		default:
+			p->jobs[i]->enforce_isolation = false;
+			p->jobs[i]->run_cleaner_shader = false;
+			break;
+		case AMDGPU_ENFORCE_ISOLATION_ENABLE:
+			p->jobs[i]->enforce_isolation = true;
+			p->jobs[i]->run_cleaner_shader = true;
+			break;
+		case AMDGPU_ENFORCE_ISOLATION_ENABLE_LEGACY:
+			p->jobs[i]->enforce_isolation = true;
+			p->jobs[i]->run_cleaner_shader = false;
+			break;
+		case AMDGPU_ENFORCE_ISOLATION_NO_CLEANER_SHADER:
+			p->jobs[i]->enforce_isolation = true;
+			p->jobs[i]->run_cleaner_shader = false;
+			break;
+		}
+	}
+	p->gang_leader = p->jobs[p->gang_leader_idx];
+
+	if (p->ctx->generation != p->gang_leader->generation) {
+		ret = -ECANCELED;
 		goto free_all_kdata;
+	}
+
+	if (p->uf_bo)
+		p->gang_leader->uf_addr = uf_offset;
+	kvfree(chunk_array);
+
+	/* Use this opportunity to fill in task info for the vm */
+	amdgpu_vm_set_task_info(vm);
 
-	if (p->uf_entry.robj)
-		p->job->uf_addr = uf_offset;
-	kfree(chunk_array);
 	return 0;
 
 free_all_kdata:
 	i = p->nchunks - 1;
 free_partial_kdata:
 	for (; i >= 0; i--)
-		drm_free_large(p->chunks[i].kdata);
-	kfree(p->chunks);
+		kvfree(p->chunks[i].kdata);
+	kvfree(p->chunks);
 	p->chunks = NULL;
 	p->nchunks = 0;
-put_ctx:
-	amdgpu_ctx_put(p->ctx);
 free_chunk:
-	kfree(chunk_array);
+	kvfree(chunk_array);
 
 	return ret;
 }
 
+static int amdgpu_cs_p2_ib(struct amdgpu_cs_parser *p,
+			   struct amdgpu_cs_chunk *chunk,
+			   unsigned int *ce_preempt,
+			   unsigned int *de_preempt)
+{
+	struct drm_amdgpu_cs_chunk_ib *chunk_ib = chunk->kdata;
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct amdgpu_vm *vm = &fpriv->vm;
+	struct amdgpu_ring *ring;
+	struct amdgpu_job *job;
+	struct amdgpu_ib *ib;
+	int r;
+
+	r = amdgpu_cs_job_idx(p, chunk_ib);
+	if (r < 0)
+		return r;
+
+	job = p->jobs[r];
+	ring = amdgpu_job_ring(job);
+	ib = &job->ibs[job->num_ibs++];
+
+	/* submissions to kernel queues are disabled */
+	if (ring->no_user_submission)
+		return -EINVAL;
+
+	/* MM engine doesn't support user fences */
+	if (p->uf_bo && ring->funcs->no_user_fence)
+		return -EINVAL;
+
+	if (!p->adev->debug_enable_ce_cs &&
+	    chunk_ib->flags & AMDGPU_IB_FLAG_CE) {
+		dev_err_ratelimited(p->adev->dev, "CE CS is blocked, use debug=0x400 to override\n");
+		return -EINVAL;
+	}
+
+	if (chunk_ib->ip_type == AMDGPU_HW_IP_GFX &&
+	    chunk_ib->flags & AMDGPU_IB_FLAG_PREEMPT) {
+		if (chunk_ib->flags & AMDGPU_IB_FLAG_CE)
+			(*ce_preempt)++;
+		else
+			(*de_preempt)++;
+
+		/* Each GFX command submit allows only 1 IB max
+		 * preemptible for CE & DE */
+		if (*ce_preempt > 1 || *de_preempt > 1)
+			return -EINVAL;
+	}
+
+	if (chunk_ib->flags & AMDGPU_IB_FLAG_PREAMBLE)
+		job->preamble_status |= AMDGPU_PREAMBLE_IB_PRESENT;
+
+	r =  amdgpu_ib_get(p->adev, vm, ring->funcs->parse_cs ?
+			   chunk_ib->ib_bytes : 0,
+			   AMDGPU_IB_POOL_DELAYED, ib);
+	if (r) {
+		drm_err(adev_to_drm(p->adev), "Failed to get ib !\n");
+		return r;
+	}
+
+	ib->gpu_addr = chunk_ib->va_start;
+	ib->length_dw = chunk_ib->ib_bytes / 4;
+	ib->flags = chunk_ib->flags;
+	return 0;
+}
+
+static int amdgpu_cs_p2_dependencies(struct amdgpu_cs_parser *p,
+				     struct amdgpu_cs_chunk *chunk)
+{
+	struct drm_amdgpu_cs_chunk_dep *deps = chunk->kdata;
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	unsigned int num_deps;
+	int i, r;
+
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_dep);
+
+	for (i = 0; i < num_deps; ++i) {
+		struct amdgpu_ctx *ctx;
+		struct drm_sched_entity *entity;
+		struct dma_fence *fence;
+
+		ctx = amdgpu_ctx_get(fpriv, deps[i].ctx_id);
+		if (ctx == NULL)
+			return -EINVAL;
+
+		r = amdgpu_ctx_get_entity(ctx, deps[i].ip_type,
+					  deps[i].ip_instance,
+					  deps[i].ring, &entity);
+		if (r) {
+			amdgpu_ctx_put(ctx);
+			return r;
+		}
+
+		fence = amdgpu_ctx_get_fence(ctx, entity, deps[i].handle);
+		amdgpu_ctx_put(ctx);
+
+		if (IS_ERR(fence))
+			return PTR_ERR(fence);
+		else if (!fence)
+			continue;
+
+		if (chunk->chunk_id == AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES) {
+			struct drm_sched_fence *s_fence;
+			struct dma_fence *old = fence;
+
+			s_fence = to_drm_sched_fence(fence);
+			fence = dma_fence_get(&s_fence->scheduled);
+			dma_fence_put(old);
+		}
+
+		r = amdgpu_sync_fence(&p->sync, fence, GFP_KERNEL);
+		dma_fence_put(fence);
+		if (r)
+			return r;
+	}
+	return 0;
+}
+
+static int amdgpu_syncobj_lookup_and_add(struct amdgpu_cs_parser *p,
+					 uint32_t handle, u64 point,
+					 u64 flags)
+{
+	struct dma_fence *fence;
+	int r;
+
+	r = drm_syncobj_find_fence(p->filp, handle, point, flags, &fence);
+	if (r) {
+		drm_err(adev_to_drm(p->adev), "syncobj %u failed to find fence @ %llu (%d)!\n",
+			  handle, point, r);
+		return r;
+	}
+
+	r = amdgpu_sync_fence(&p->sync, fence, GFP_KERNEL);
+	dma_fence_put(fence);
+	return r;
+}
+
+static int amdgpu_cs_p2_syncobj_in(struct amdgpu_cs_parser *p,
+				   struct amdgpu_cs_chunk *chunk)
+{
+	struct drm_amdgpu_cs_chunk_sem *deps = chunk->kdata;
+	unsigned int num_deps;
+	int i, r;
+
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_sem);
+	for (i = 0; i < num_deps; ++i) {
+		r = amdgpu_syncobj_lookup_and_add(p, deps[i].handle, 0, 0);
+		if (r)
+			return r;
+	}
+
+	return 0;
+}
+
+static int amdgpu_cs_p2_syncobj_timeline_wait(struct amdgpu_cs_parser *p,
+					      struct amdgpu_cs_chunk *chunk)
+{
+	struct drm_amdgpu_cs_chunk_syncobj *syncobj_deps = chunk->kdata;
+	unsigned int num_deps;
+	int i, r;
+
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_syncobj);
+	for (i = 0; i < num_deps; ++i) {
+		r = amdgpu_syncobj_lookup_and_add(p, syncobj_deps[i].handle,
+						  syncobj_deps[i].point,
+						  syncobj_deps[i].flags);
+		if (r)
+			return r;
+	}
+
+	return 0;
+}
+
+static int amdgpu_cs_p2_syncobj_out(struct amdgpu_cs_parser *p,
+				    struct amdgpu_cs_chunk *chunk)
+{
+	struct drm_amdgpu_cs_chunk_sem *deps = chunk->kdata;
+	unsigned int num_deps;
+	int i;
+
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_sem);
+
+	if (p->post_deps)
+		return -EINVAL;
+
+	p->post_deps = kmalloc_array(num_deps, sizeof(*p->post_deps),
+				     GFP_KERNEL);
+	p->num_post_deps = 0;
+
+	if (!p->post_deps)
+		return -ENOMEM;
+
+
+	for (i = 0; i < num_deps; ++i) {
+		p->post_deps[i].syncobj =
+			drm_syncobj_find(p->filp, deps[i].handle);
+		if (!p->post_deps[i].syncobj)
+			return -EINVAL;
+		p->post_deps[i].chain = NULL;
+		p->post_deps[i].point = 0;
+		p->num_post_deps++;
+	}
+
+	return 0;
+}
+
+static int amdgpu_cs_p2_syncobj_timeline_signal(struct amdgpu_cs_parser *p,
+						struct amdgpu_cs_chunk *chunk)
+{
+	struct drm_amdgpu_cs_chunk_syncobj *syncobj_deps = chunk->kdata;
+	unsigned int num_deps;
+	int i;
+
+	num_deps = chunk->length_dw * 4 /
+		sizeof(struct drm_amdgpu_cs_chunk_syncobj);
+
+	if (p->post_deps)
+		return -EINVAL;
+
+	p->post_deps = kmalloc_array(num_deps, sizeof(*p->post_deps),
+				     GFP_KERNEL);
+	p->num_post_deps = 0;
+
+	if (!p->post_deps)
+		return -ENOMEM;
+
+	for (i = 0; i < num_deps; ++i) {
+		struct amdgpu_cs_post_dep *dep = &p->post_deps[i];
+
+		dep->chain = NULL;
+		if (syncobj_deps[i].point) {
+			dep->chain = dma_fence_chain_alloc();
+			if (!dep->chain)
+				return -ENOMEM;
+		}
+
+		dep->syncobj = drm_syncobj_find(p->filp,
+						syncobj_deps[i].handle);
+		if (!dep->syncobj) {
+			dma_fence_chain_free(dep->chain);
+			return -EINVAL;
+		}
+		dep->point = syncobj_deps[i].point;
+		p->num_post_deps++;
+	}
+
+	return 0;
+}
+
+static int amdgpu_cs_p2_shadow(struct amdgpu_cs_parser *p,
+			       struct amdgpu_cs_chunk *chunk)
+{
+	struct drm_amdgpu_cs_chunk_cp_gfx_shadow *shadow = chunk->kdata;
+	int i;
+
+	if (shadow->flags & ~AMDGPU_CS_CHUNK_CP_GFX_SHADOW_FLAGS_INIT_SHADOW)
+		return -EINVAL;
+
+	for (i = 0; i < p->gang_size; ++i) {
+		p->jobs[i]->shadow_va = shadow->shadow_va;
+		p->jobs[i]->csa_va = shadow->csa_va;
+		p->jobs[i]->gds_va = shadow->gds_va;
+		p->jobs[i]->init_shadow =
+			shadow->flags & AMDGPU_CS_CHUNK_CP_GFX_SHADOW_FLAGS_INIT_SHADOW;
+	}
+
+	return 0;
+}
+
+static int amdgpu_cs_pass2(struct amdgpu_cs_parser *p)
+{
+	unsigned int ce_preempt = 0, de_preempt = 0;
+	int i, r;
+
+	for (i = 0; i < p->nchunks; ++i) {
+		struct amdgpu_cs_chunk *chunk;
+
+		chunk = &p->chunks[i];
+
+		switch (chunk->chunk_id) {
+		case AMDGPU_CHUNK_ID_IB:
+			r = amdgpu_cs_p2_ib(p, chunk, &ce_preempt, &de_preempt);
+			if (r)
+				return r;
+			break;
+		case AMDGPU_CHUNK_ID_DEPENDENCIES:
+		case AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES:
+			r = amdgpu_cs_p2_dependencies(p, chunk);
+			if (r)
+				return r;
+			break;
+		case AMDGPU_CHUNK_ID_SYNCOBJ_IN:
+			r = amdgpu_cs_p2_syncobj_in(p, chunk);
+			if (r)
+				return r;
+			break;
+		case AMDGPU_CHUNK_ID_SYNCOBJ_OUT:
+			r = amdgpu_cs_p2_syncobj_out(p, chunk);
+			if (r)
+				return r;
+			break;
+		case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_WAIT:
+			r = amdgpu_cs_p2_syncobj_timeline_wait(p, chunk);
+			if (r)
+				return r;
+			break;
+		case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_SIGNAL:
+			r = amdgpu_cs_p2_syncobj_timeline_signal(p, chunk);
+			if (r)
+				return r;
+			break;
+		case AMDGPU_CHUNK_ID_CP_GFX_SHADOW:
+			r = amdgpu_cs_p2_shadow(p, chunk);
+			if (r)
+				return r;
+			break;
+		}
+	}
+
+	return 0;
+}
+
 /* Convert microseconds to bytes. */
 static u64 us_to_bytes(struct amdgpu_device *adev, s64 us)
 {
@@ -292,12 +693,12 @@ static s64 bytes_to_us(struct amdgpu_device *adev, u64 bytes)
  * ticks. The accumulated microseconds (us) are converted to bytes and
  * returned.
  */
-static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
+static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev,
+					      u64 *max_bytes,
+					      u64 *max_vis_bytes)
 {
 	s64 time_us, increment_us;
-	u64 max_bytes;
 	u64 free_vram, total_vram, used_vram;
-
 	/* Allow a maximum of 200 accumulated ms. This is basically per-IB
 	 * throttling.
 	 *
@@ -307,11 +708,14 @@ static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
 	 */
 	const s64 us_upper_bound = 200000;
 
-	if (!adev->mm_stats.log2_max_MBps)
-		return 0;
+	if ((!adev->mm_stats.log2_max_MBps) || !ttm_resource_manager_used(&adev->mman.vram_mgr.manager)) {
+		*max_bytes = 0;
+		*max_vis_bytes = 0;
+		return;
+	}
 
-	total_vram = adev->mc.real_vram_size - adev->vram_pin_size;
-	used_vram = atomic64_read(&adev->vram_usage);
+	total_vram = adev->gmc.real_vram_size - atomic64_read(&adev->vram_pin_size);
+	used_vram = ttm_resource_manager_usage(&adev->mman.vram_mgr.manager);
 	free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;
 
 	spin_lock(&adev->mm_stats.lock);
@@ -321,7 +725,7 @@ static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
 	increment_us = time_us - adev->mm_stats.last_update_us;
 	adev->mm_stats.last_update_us = time_us;
 	adev->mm_stats.accum_us = min(adev->mm_stats.accum_us + increment_us,
-                                      us_upper_bound);
+				      us_upper_bound);
 
 	/* This prevents the short period of low performance when the VRAM
 	 * usage is low and the driver is in debt or doesn't have enough
@@ -338,7 +742,7 @@ static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
 	if (free_vram >= 128 * 1024 * 1024 || free_vram >= total_vram / 8) {
 		s64 min_us;
 
-		/* Be more aggresive on dGPUs. Try to fill a portion of free
+		/* Be more aggressive on dGPUs. Try to fill a portion of free
 		 * VRAM now.
 		 */
 		if (!(adev->flags & AMD_IS_APU))
@@ -349,51 +753,96 @@ static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
 		adev->mm_stats.accum_us = max(min_us, adev->mm_stats.accum_us);
 	}
 
-	/* This returns 0 if the driver is in debt to disallow (optional)
+	/* This is set to 0 if the driver is in debt to disallow (optional)
 	 * buffer moves.
 	 */
-	max_bytes = us_to_bytes(adev, adev->mm_stats.accum_us);
+	*max_bytes = us_to_bytes(adev, adev->mm_stats.accum_us);
+
+	/* Do the same for visible VRAM if half of it is free */
+	if (!amdgpu_gmc_vram_full_visible(&adev->gmc)) {
+		u64 total_vis_vram = adev->gmc.visible_vram_size;
+		u64 used_vis_vram =
+		  amdgpu_vram_mgr_vis_usage(&adev->mman.vram_mgr);
+
+		if (used_vis_vram < total_vis_vram) {
+			u64 free_vis_vram = total_vis_vram - used_vis_vram;
+
+			adev->mm_stats.accum_us_vis = min(adev->mm_stats.accum_us_vis +
+							  increment_us, us_upper_bound);
+
+			if (free_vis_vram >= total_vis_vram / 2)
+				adev->mm_stats.accum_us_vis =
+					max(bytes_to_us(adev, free_vis_vram / 2),
+					    adev->mm_stats.accum_us_vis);
+		}
+
+		*max_vis_bytes = us_to_bytes(adev, adev->mm_stats.accum_us_vis);
+	} else {
+		*max_vis_bytes = 0;
+	}
 
 	spin_unlock(&adev->mm_stats.lock);
-	return max_bytes;
 }
 
 /* Report how many bytes have really been moved for the last command
  * submission. This can result in a debt that can stop buffer migrations
  * temporarily.
  */
-void amdgpu_cs_report_moved_bytes(struct amdgpu_device *adev, u64 num_bytes)
+void amdgpu_cs_report_moved_bytes(struct amdgpu_device *adev, u64 num_bytes,
+				  u64 num_vis_bytes)
 {
 	spin_lock(&adev->mm_stats.lock);
 	adev->mm_stats.accum_us -= bytes_to_us(adev, num_bytes);
+	adev->mm_stats.accum_us_vis -= bytes_to_us(adev, num_vis_bytes);
 	spin_unlock(&adev->mm_stats.lock);
 }
 
-static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
-				 struct amdgpu_bo *bo)
+static int amdgpu_cs_bo_validate(void *param, struct amdgpu_bo *bo)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	u64 initial_bytes_moved;
+	struct amdgpu_cs_parser *p = param;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = false,
+		.resv = bo->tbo.base.resv
+	};
 	uint32_t domain;
 	int r;
 
-	if (bo->pin_count)
+	if (bo->tbo.pin_count)
 		return 0;
 
 	/* Don't move this buffer if we have depleted our allowance
 	 * to move it. Don't move anything if the threshold is zero.
 	 */
-	if (p->bytes_moved < p->bytes_moved_threshold)
-		domain = bo->prefered_domains;
-	else
+	if (p->bytes_moved < p->bytes_moved_threshold &&
+	    (!bo->tbo.base.dma_buf ||
+	    list_empty(&bo->tbo.base.dma_buf->attachments))) {
+		if (!amdgpu_gmc_vram_full_visible(&adev->gmc) &&
+		    (bo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)) {
+			/* And don't move a CPU_ACCESS_REQUIRED BO to limited
+			 * visible VRAM if we've depleted our allowance to do
+			 * that.
+			 */
+			if (p->bytes_moved_vis < p->bytes_moved_vis_threshold)
+				domain = bo->preferred_domains;
+			else
+				domain = bo->allowed_domains;
+		} else {
+			domain = bo->preferred_domains;
+		}
+	} else {
 		domain = bo->allowed_domains;
+	}
 
 retry:
-	amdgpu_ttm_placement_from_domain(bo, domain);
-	initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
-	p->bytes_moved += atomic64_read(&adev->num_bytes_moved) -
-		initial_bytes_moved;
+	amdgpu_bo_placement_from_domain(bo, domain);
+	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+
+	p->bytes_moved += ctx.bytes_moved;
+	if (!amdgpu_gmc_vram_full_visible(&adev->gmc) &&
+	    amdgpu_res_cpu_visible(adev, bo->tbo.resource))
+		p->bytes_moved_vis += ctx.bytes_moved;
 
 	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
 		domain = bo->allowed_domains;
@@ -403,387 +852,272 @@ retry:
 	return r;
 }
 
-/* Last resort, try to evict something from the current working set */
-static bool amdgpu_cs_try_evict(struct amdgpu_cs_parser *p,
-				struct amdgpu_bo *validated)
-{
-	uint32_t domain = validated->allowed_domains;
-	int r;
-
-	if (!p->evictable)
-		return false;
-
-	for (;&p->evictable->tv.head != &p->validated;
-	     p->evictable = list_prev_entry(p->evictable, tv.head)) {
-
-		struct amdgpu_bo_list_entry *candidate = p->evictable;
-		struct amdgpu_bo *bo = candidate->robj;
-		struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-		u64 initial_bytes_moved;
-		uint32_t other;
-
-		/* If we reached our current BO we can forget it */
-		if (candidate->robj == validated)
-			break;
-
-		other = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-
-		/* Check if this BO is in one of the domains we need space for */
-		if (!(other & domain))
-			continue;
-
-		/* Check if we can move this BO somewhere else */
-		other = bo->allowed_domains & ~domain;
-		if (!other)
-			continue;
-
-		/* Good we can try to move this BO somewhere else */
-		amdgpu_ttm_placement_from_domain(bo, other);
-		initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
-		r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
-		p->bytes_moved += atomic64_read(&adev->num_bytes_moved) -
-			initial_bytes_moved;
-
-		if (unlikely(r))
-			break;
-
-		p->evictable = list_prev_entry(p->evictable, tv.head);
-		list_move(&candidate->tv.head, &p->validated);
-
-		return true;
-	}
-
-	return false;
-}
-
-static int amdgpu_cs_validate(void *param, struct amdgpu_bo *bo)
-{
-	struct amdgpu_cs_parser *p = param;
-	int r;
-
-	do {
-		r = amdgpu_cs_bo_validate(p, bo);
-	} while (r == -ENOMEM && amdgpu_cs_try_evict(p, bo));
-	if (r)
-		return r;
-
-	if (bo->shadow)
-		r = amdgpu_cs_bo_validate(p, bo->shadow);
-
-	return r;
-}
-
-static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
-			    struct list_head *validated)
-{
-	struct amdgpu_bo_list_entry *lobj;
-	int r;
-
-	list_for_each_entry(lobj, validated, tv.head) {
-		struct amdgpu_bo *bo = lobj->robj;
-		bool binding_userptr = false;
-		struct mm_struct *usermm;
-
-		usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
-		if (usermm && usermm != current->mm)
-			return -EPERM;
-
-		/* Check if we have user pages and nobody bound the BO already */
-		if (lobj->user_pages && bo->tbo.ttm->state != tt_bound) {
-			size_t size = sizeof(struct page *);
-
-			size *= bo->tbo.ttm->num_pages;
-			memcpy(bo->tbo.ttm->pages, lobj->user_pages, size);
-			binding_userptr = true;
-		}
-
-		if (p->evictable == lobj)
-			p->evictable = NULL;
-
-		r = amdgpu_cs_validate(p, bo);
-		if (r)
-			return r;
-
-		if (binding_userptr) {
-			drm_free_large(lobj->user_pages);
-			lobj->user_pages = NULL;
-		}
-	}
-	return 0;
-}
-
 static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 				union drm_amdgpu_cs *cs)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct ttm_operation_ctx ctx = { true, false };
+	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_list_entry *e;
-	struct list_head duplicates;
-	bool need_mmap_lock = false;
-	unsigned i, tries = 10;
+	struct drm_gem_object *obj;
+	unsigned long index;
+	unsigned int i;
 	int r;
 
-	INIT_LIST_HEAD(&p->validated);
+	/* p->bo_list could already be assigned if AMDGPU_CHUNK_ID_BO_HANDLES is present */
+	if (cs->in.bo_list_handle) {
+		if (p->bo_list)
+			return -EINVAL;
 
-	p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
-	if (p->bo_list) {
-		need_mmap_lock = p->bo_list->first_userptr !=
-			p->bo_list->num_entries;
-		amdgpu_bo_list_get_list(p->bo_list, &p->validated);
+		r = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle,
+				       &p->bo_list);
+		if (r)
+			return r;
+	} else if (!p->bo_list) {
+		/* Create a empty bo_list when no handle is provided */
+		r = amdgpu_bo_list_create(p->adev, p->filp, NULL, 0,
+					  &p->bo_list);
+		if (r)
+			return r;
 	}
 
-	INIT_LIST_HEAD(&duplicates);
-	amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
+	mutex_lock(&p->bo_list->bo_list_mutex);
 
-	if (p->uf_entry.robj)
-		list_add(&p->uf_entry.tv.head, &p->validated);
+	/* Get userptr backing pages. If pages are updated after registered
+	 * in amdgpu_gem_userptr_ioctl(), amdgpu_cs_list_validate() will do
+	 * amdgpu_ttm_backend_bind() to flush and invalidate new pages
+	 */
+	amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
+		bool userpage_invalidated = false;
+		struct amdgpu_bo *bo = e->bo;
 
-	if (need_mmap_lock)
-		down_read(&current->mm->mmap_sem);
+		e->range = amdgpu_hmm_range_alloc(NULL);
+		if (unlikely(!e->range))
+			return -ENOMEM;
 
-	while (1) {
-		struct list_head need_pages;
-		unsigned i;
+		r = amdgpu_ttm_tt_get_user_pages(bo, e->range);
+		if (r)
+			goto out_free_user_pages;
 
-		r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
-					   &duplicates);
-		if (unlikely(r != 0)) {
-			if (r != -ERESTARTSYS)
-				DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
-			goto error_free_pages;
+		for (i = 0; i < bo->tbo.ttm->num_pages; i++) {
+			if (bo->tbo.ttm->pages[i] !=
+				hmm_pfn_to_page(e->range->hmm_range.hmm_pfns[i])) {
+				userpage_invalidated = true;
+				break;
+			}
 		}
+		e->user_invalidated = userpage_invalidated;
+	}
 
-		/* Without a BO list we don't have userptr BOs */
-		if (!p->bo_list)
-			break;
-
-		INIT_LIST_HEAD(&need_pages);
-		for (i = p->bo_list->first_userptr;
-		     i < p->bo_list->num_entries; ++i) {
-
-			e = &p->bo_list->array[i];
-
-			if (amdgpu_ttm_tt_userptr_invalidated(e->robj->tbo.ttm,
-				 &e->user_invalidated) && e->user_pages) {
-
-				/* We acquired a page array, but somebody
-				 * invalidated it. Free it an try again
-				 */
-				release_pages(e->user_pages,
-					      e->robj->tbo.ttm->num_pages,
-					      false);
-				drm_free_large(e->user_pages);
-				e->user_pages = NULL;
-			}
+	drm_exec_until_all_locked(&p->exec) {
+		r = amdgpu_vm_lock_pd(&fpriv->vm, &p->exec, 1 + p->gang_size);
+		drm_exec_retry_on_contention(&p->exec);
+		if (unlikely(r))
+			goto out_free_user_pages;
 
-			if (e->robj->tbo.ttm->state != tt_bound &&
-			    !e->user_pages) {
-				list_del(&e->tv.head);
-				list_add(&e->tv.head, &need_pages);
+		amdgpu_bo_list_for_each_entry(e, p->bo_list) {
+			/* One fence for TTM and one for each CS job */
+			r = drm_exec_prepare_obj(&p->exec, &e->bo->tbo.base,
+						 1 + p->gang_size);
+			drm_exec_retry_on_contention(&p->exec);
+			if (unlikely(r))
+				goto out_free_user_pages;
 
-				amdgpu_bo_unreserve(e->robj);
-			}
+			e->bo_va = amdgpu_vm_bo_find(vm, e->bo);
 		}
 
-		if (list_empty(&need_pages))
-			break;
+		if (p->uf_bo) {
+			r = drm_exec_prepare_obj(&p->exec, &p->uf_bo->tbo.base,
+						 1 + p->gang_size);
+			drm_exec_retry_on_contention(&p->exec);
+			if (unlikely(r))
+				goto out_free_user_pages;
+		}
+	}
 
-		/* Unreserve everything again. */
-		ttm_eu_backoff_reservation(&p->ticket, &p->validated);
+	amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
+		struct mm_struct *usermm;
 
-		/* We tried too many times, just abort */
-		if (!--tries) {
-			r = -EDEADLK;
-			DRM_ERROR("deadlock in %s\n", __func__);
-			goto error_free_pages;
+		usermm = amdgpu_ttm_tt_get_usermm(e->bo->tbo.ttm);
+		if (usermm && usermm != current->mm) {
+			r = -EPERM;
+			goto out_free_user_pages;
 		}
 
-		/* Fill the page arrays for all useptrs. */
-		list_for_each_entry(e, &need_pages, tv.head) {
-			struct ttm_tt *ttm = e->robj->tbo.ttm;
-
-			e->user_pages = drm_calloc_large(ttm->num_pages,
-							 sizeof(struct page*));
-			if (!e->user_pages) {
-				r = -ENOMEM;
-				DRM_ERROR("calloc failure in %s\n", __func__);
-				goto error_free_pages;
-			}
+		if (amdgpu_ttm_tt_is_userptr(e->bo->tbo.ttm) &&
+		    e->user_invalidated) {
+			amdgpu_bo_placement_from_domain(e->bo,
+							AMDGPU_GEM_DOMAIN_CPU);
+			r = ttm_bo_validate(&e->bo->tbo, &e->bo->placement,
+					    &ctx);
+			if (r)
+				goto out_free_user_pages;
 
-			r = amdgpu_ttm_tt_get_user_pages(ttm, e->user_pages);
-			if (r) {
-				DRM_ERROR("amdgpu_ttm_tt_get_user_pages failed.\n");
-				drm_free_large(e->user_pages);
-				e->user_pages = NULL;
-				goto error_free_pages;
-			}
+			amdgpu_ttm_tt_set_user_pages(e->bo->tbo.ttm,
+						     e->range);
 		}
-
-		/* And try again. */
-		list_splice(&need_pages, &p->validated);
 	}
 
-	p->bytes_moved_threshold = amdgpu_cs_get_threshold_for_moves(p->adev);
+	amdgpu_cs_get_threshold_for_moves(p->adev, &p->bytes_moved_threshold,
+					  &p->bytes_moved_vis_threshold);
 	p->bytes_moved = 0;
-	p->evictable = list_last_entry(&p->validated,
-				       struct amdgpu_bo_list_entry,
-				       tv.head);
+	p->bytes_moved_vis = 0;
 
-	r = amdgpu_vm_validate_pt_bos(p->adev, &fpriv->vm,
-				      amdgpu_cs_validate, p);
+	r = amdgpu_vm_validate(p->adev, &fpriv->vm, NULL,
+			       amdgpu_cs_bo_validate, p);
 	if (r) {
-		DRM_ERROR("amdgpu_vm_validate_pt_bos() failed.\n");
-		goto error_validate;
+		drm_err(adev_to_drm(p->adev), "amdgpu_vm_validate() failed.\n");
+		goto out_free_user_pages;
 	}
 
-	r = amdgpu_cs_list_validate(p, &duplicates);
-	if (r) {
-		DRM_ERROR("amdgpu_cs_list_validate(duplicates) failed.\n");
-		goto error_validate;
+	drm_exec_for_each_locked_object(&p->exec, index, obj) {
+		r = amdgpu_cs_bo_validate(p, gem_to_amdgpu_bo(obj));
+		if (unlikely(r))
+			goto out_free_user_pages;
 	}
 
-	r = amdgpu_cs_list_validate(p, &p->validated);
-	if (r) {
-		DRM_ERROR("amdgpu_cs_list_validate(validated) failed.\n");
-		goto error_validate;
+	if (p->uf_bo) {
+		r = amdgpu_ttm_alloc_gart(&p->uf_bo->tbo);
+		if (unlikely(r))
+			goto out_free_user_pages;
+
+		p->gang_leader->uf_addr += amdgpu_bo_gpu_offset(p->uf_bo);
 	}
 
-	amdgpu_cs_report_moved_bytes(p->adev, p->bytes_moved);
+	amdgpu_cs_report_moved_bytes(p->adev, p->bytes_moved,
+				     p->bytes_moved_vis);
+
+	for (i = 0; i < p->gang_size; ++i)
+		amdgpu_job_set_resources(p->jobs[i], p->bo_list->gds_obj,
+					 p->bo_list->gws_obj,
+					 p->bo_list->oa_obj);
+	return 0;
 
-	fpriv->vm.last_eviction_counter =
-		atomic64_read(&p->adev->num_evictions);
+out_free_user_pages:
+	amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
+		amdgpu_hmm_range_free(e->range);
+		e->range = NULL;
+	}
+	mutex_unlock(&p->bo_list->bo_list_mutex);
+	return r;
+}
 
-	if (p->bo_list) {
-		struct amdgpu_bo *gds = p->bo_list->gds_obj;
-		struct amdgpu_bo *gws = p->bo_list->gws_obj;
-		struct amdgpu_bo *oa = p->bo_list->oa_obj;
-		struct amdgpu_vm *vm = &fpriv->vm;
-		unsigned i;
+static void trace_amdgpu_cs_ibs(struct amdgpu_cs_parser *p)
+{
+	int i, j;
 
-		for (i = 0; i < p->bo_list->num_entries; i++) {
-			struct amdgpu_bo *bo = p->bo_list->array[i].robj;
+	if (!trace_amdgpu_cs_enabled())
+		return;
 
-			p->bo_list->array[i].bo_va = amdgpu_vm_bo_find(vm, bo);
-		}
+	for (i = 0; i < p->gang_size; ++i) {
+		struct amdgpu_job *job = p->jobs[i];
 
-		if (gds) {
-			p->job->gds_base = amdgpu_bo_gpu_offset(gds);
-			p->job->gds_size = amdgpu_bo_size(gds);
-		}
-		if (gws) {
-			p->job->gws_base = amdgpu_bo_gpu_offset(gws);
-			p->job->gws_size = amdgpu_bo_size(gws);
-		}
-		if (oa) {
-			p->job->oa_base = amdgpu_bo_gpu_offset(oa);
-			p->job->oa_size = amdgpu_bo_size(oa);
-		}
+		for (j = 0; j < job->num_ibs; ++j)
+			trace_amdgpu_cs(p, job, &job->ibs[j]);
 	}
+}
 
-	if (!r && p->uf_entry.robj) {
-		struct amdgpu_bo *uf = p->uf_entry.robj;
+static int amdgpu_cs_patch_ibs(struct amdgpu_cs_parser *p,
+			       struct amdgpu_job *job)
+{
+	struct amdgpu_ring *ring = amdgpu_job_ring(job);
+	unsigned int i;
+	int r;
 
-		r = amdgpu_ttm_bind(&uf->tbo, &uf->tbo.mem);
-		p->job->uf_addr += amdgpu_bo_gpu_offset(uf);
-	}
+	/* Only for UVD/VCE VM emulation */
+	if (!ring->funcs->parse_cs && !ring->funcs->patch_cs_in_place)
+		return 0;
 
-error_validate:
-	if (r) {
-		amdgpu_vm_move_pt_bos_in_lru(p->adev, &fpriv->vm);
-		ttm_eu_backoff_reservation(&p->ticket, &p->validated);
-	}
+	for (i = 0; i < job->num_ibs; ++i) {
+		struct amdgpu_ib *ib = &job->ibs[i];
+		struct amdgpu_bo_va_mapping *m;
+		struct amdgpu_bo *aobj;
+		uint64_t va_start;
+		uint8_t *kptr;
+
+		va_start = ib->gpu_addr & AMDGPU_GMC_HOLE_MASK;
+		r = amdgpu_cs_find_mapping(p, va_start, &aobj, &m);
+		if (r) {
+			drm_err(adev_to_drm(p->adev), "IB va_start is invalid\n");
+			return r;
+		}
 
-error_free_pages:
+		if ((va_start + ib->length_dw * 4) >
+		    (m->last + 1) * AMDGPU_GPU_PAGE_SIZE) {
+			drm_err(adev_to_drm(p->adev), "IB va_start+ib_bytes is invalid\n");
+			return -EINVAL;
+		}
+
+		/* the IB should be reserved at this point */
+		r = amdgpu_bo_kmap(aobj, (void **)&kptr);
+		if (r)
+			return r;
 
-	if (need_mmap_lock)
-		up_read(&current->mm->mmap_sem);
+		kptr += va_start - (m->start * AMDGPU_GPU_PAGE_SIZE);
 
-	if (p->bo_list) {
-		for (i = p->bo_list->first_userptr;
-		     i < p->bo_list->num_entries; ++i) {
-			e = &p->bo_list->array[i];
+		if (ring->funcs->parse_cs) {
+			memcpy(ib->ptr, kptr, ib->length_dw * 4);
+			amdgpu_bo_kunmap(aobj);
 
-			if (!e->user_pages)
-				continue;
+			r = amdgpu_ring_parse_cs(ring, p, job, ib);
+			if (r)
+				return r;
 
-			release_pages(e->user_pages,
-				      e->robj->tbo.ttm->num_pages,
-				      false);
-			drm_free_large(e->user_pages);
+			if (ib->sa_bo)
+				ib->gpu_addr =  amdgpu_sa_bo_gpu_addr(ib->sa_bo);
+		} else {
+			ib->ptr = (uint32_t *)kptr;
+			r = amdgpu_ring_patch_cs_in_place(ring, p, job, ib);
+			amdgpu_bo_kunmap(aobj);
+			if (r)
+				return r;
 		}
 	}
 
-	return r;
+	return 0;
 }
 
-static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
+static int amdgpu_cs_patch_jobs(struct amdgpu_cs_parser *p)
 {
-	struct amdgpu_bo_list_entry *e;
+	unsigned int i;
 	int r;
 
-	list_for_each_entry(e, &p->validated, tv.head) {
-		struct reservation_object *resv = e->robj->tbo.resv;
-		r = amdgpu_sync_resv(p->adev, &p->job->sync, resv, p->filp);
-
+	for (i = 0; i < p->gang_size; ++i) {
+		r = amdgpu_cs_patch_ibs(p, p->jobs[i]);
 		if (r)
 			return r;
 	}
 	return 0;
 }
 
-/**
- * cs_parser_fini() - clean parser states
- * @parser:	parser structure holding parsing context.
- * @error:	error number
- *
- * If error is set than unvalidate buffer, otherwise just free memory
- * used by parsing context.
- **/
-static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
-{
-	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
-	unsigned i;
-
-	if (!error) {
-		amdgpu_vm_move_pt_bos_in_lru(parser->adev, &fpriv->vm);
-
-		ttm_eu_fence_buffer_objects(&parser->ticket,
-					    &parser->validated,
-					    parser->fence);
-	} else if (backoff) {
-		ttm_eu_backoff_reservation(&parser->ticket,
-					   &parser->validated);
-	}
-	dma_fence_put(parser->fence);
-
-	if (parser->ctx)
-		amdgpu_ctx_put(parser->ctx);
-	if (parser->bo_list)
-		amdgpu_bo_list_put(parser->bo_list);
-
-	for (i = 0; i < parser->nchunks; i++)
-		drm_free_large(parser->chunks[i].kdata);
-	kfree(parser->chunks);
-	if (parser->job)
-		amdgpu_job_free(parser->job);
-	amdgpu_bo_unref(&parser->uf_entry.robj);
-}
-
-static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p)
+static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 {
-	struct amdgpu_device *adev = p->adev;
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct amdgpu_job *job = p->gang_leader;
+	struct amdgpu_device *adev = p->adev;
 	struct amdgpu_vm *vm = &fpriv->vm;
+	struct amdgpu_bo_list_entry *e;
 	struct amdgpu_bo_va *bo_va;
-	struct amdgpu_bo *bo;
-	int i, r;
+	unsigned int i;
+	int r;
 
-	r = amdgpu_vm_update_directories(adev, vm);
-	if (r)
-		return r;
+	/*
+	 * We can't use gang submit on with reserved VMIDs when the VM changes
+	 * can't be invalidated by more than one engine at the same time.
+	 */
+	if (p->gang_size > 1 && !adev->vm_manager.concurrent_flush) {
+		for (i = 0; i < p->gang_size; ++i) {
+			struct drm_sched_entity *entity = p->entities[i];
+			struct drm_gpu_scheduler *sched = entity->rq->sched;
+			struct amdgpu_ring *ring = to_amdgpu_ring(sched);
 
-	r = amdgpu_sync_fence(adev, &p->job->sync, vm->last_dir_update);
-	if (r)
-		return r;
+			if (amdgpu_vmid_uses_reserved(vm, ring->vm_hub))
+				return -EINVAL;
+		}
+	}
+
+	if (!amdgpu_vm_ready(vm))
+		return -EINVAL;
 
 	r = amdgpu_vm_clear_freed(adev, vm, NULL);
 	if (r)
@@ -793,349 +1127,354 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p)
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(adev, &p->job->sync,
-			      fpriv->prt_va->last_pt_update);
+	r = amdgpu_sync_fence(&p->sync, fpriv->prt_va->last_pt_update,
+			      GFP_KERNEL);
 	if (r)
 		return r;
 
-	if (amdgpu_sriov_vf(adev)) {
-		struct dma_fence *f;
-		bo_va = vm->csa_bo_va;
+	if (fpriv->csa_va) {
+		bo_va = fpriv->csa_va;
 		BUG_ON(!bo_va);
 		r = amdgpu_vm_bo_update(adev, bo_va, false);
 		if (r)
 			return r;
 
-		f = bo_va->last_pt_update;
-		r = amdgpu_sync_fence(adev, &p->job->sync, f);
+		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update,
+				      GFP_KERNEL);
 		if (r)
 			return r;
 	}
 
-	if (p->bo_list) {
-		for (i = 0; i < p->bo_list->num_entries; i++) {
-			struct dma_fence *f;
+	/* FIXME: In theory this loop shouldn't be needed any more when
+	 * amdgpu_vm_handle_moved handles all moved BOs that are reserved
+	 * with p->ticket. But removing it caused test regressions, so I'm
+	 * leaving it here for now.
+	 */
+	amdgpu_bo_list_for_each_entry(e, p->bo_list) {
+		bo_va = e->bo_va;
+		if (bo_va == NULL)
+			continue;
 
-			/* ignore duplicates */
-			bo = p->bo_list->array[i].robj;
-			if (!bo)
-				continue;
+		r = amdgpu_vm_bo_update(adev, bo_va, false);
+		if (r)
+			return r;
 
-			bo_va = p->bo_list->array[i].bo_va;
-			if (bo_va == NULL)
-				continue;
+		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update,
+				      GFP_KERNEL);
+		if (r)
+			return r;
+	}
 
-			r = amdgpu_vm_bo_update(adev, bo_va, false);
-			if (r)
-				return r;
+	r = amdgpu_vm_handle_moved(adev, vm, &p->exec.ticket);
+	if (r)
+		return r;
 
-			f = bo_va->last_pt_update;
-			r = amdgpu_sync_fence(adev, &p->job->sync, f);
-			if (r)
-				return r;
-		}
+	r = amdgpu_vm_update_pdes(adev, vm, false);
+	if (r)
+		return r;
 
-	}
+	r = amdgpu_sync_fence(&p->sync, vm->last_update, GFP_KERNEL);
+	if (r)
+		return r;
+
+	for (i = 0; i < p->gang_size; ++i) {
+		job = p->jobs[i];
 
-	r = amdgpu_vm_clear_invalids(adev, vm, &p->job->sync);
+		if (!job->vm)
+			continue;
+
+		job->vm_pd_addr = amdgpu_gmc_pd_addr(vm->root.bo);
+	}
 
-	if (amdgpu_vm_debug && p->bo_list) {
+	if (adev->debug_vm) {
 		/* Invalidate all BOs to test for userspace bugs */
-		for (i = 0; i < p->bo_list->num_entries; i++) {
+		amdgpu_bo_list_for_each_entry(e, p->bo_list) {
+			struct amdgpu_bo *bo = e->bo;
+
 			/* ignore duplicates */
-			bo = p->bo_list->array[i].robj;
 			if (!bo)
 				continue;
 
-			amdgpu_vm_bo_invalidate(adev, bo);
+			amdgpu_vm_bo_invalidate(bo, false);
 		}
 	}
 
-	return r;
+	return 0;
 }
 
-static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
-				 struct amdgpu_cs_parser *p)
+static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	struct amdgpu_vm *vm = &fpriv->vm;
-	struct amdgpu_ring *ring = p->job->ring;
-	int i, r;
+	struct drm_gpu_scheduler *sched;
+	struct drm_gem_object *obj;
+	struct dma_fence *fence;
+	unsigned long index;
+	unsigned int i;
+	int r;
 
-	/* Only for UVD/VCE VM emulation */
-	if (ring->funcs->parse_cs) {
-		for (i = 0; i < p->job->num_ibs; i++) {
-			r = amdgpu_ring_parse_cs(ring, p, i);
-			if (r)
-				return r;
-		}
+	r = amdgpu_ctx_wait_prev_fence(p->ctx, p->entities[p->gang_leader_idx]);
+	if (r) {
+		if (r != -ERESTARTSYS)
+			drm_err(adev_to_drm(p->adev), "amdgpu_ctx_wait_prev_fence failed.\n");
+		return r;
 	}
 
-	if (p->job->vm) {
-		p->job->vm_pd_addr = amdgpu_bo_gpu_offset(vm->root.bo);
+	drm_exec_for_each_locked_object(&p->exec, index, obj) {
+		struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+
+		struct dma_resv *resv = bo->tbo.base.resv;
+		enum amdgpu_sync_mode sync_mode;
 
-		r = amdgpu_bo_vm_update_pte(p);
+		sync_mode = amdgpu_bo_explicit_sync(bo) ?
+			AMDGPU_SYNC_EXPLICIT : AMDGPU_SYNC_NE_OWNER;
+		r = amdgpu_sync_resv(p->adev, &p->sync, resv, sync_mode,
+				     &fpriv->vm);
 		if (r)
 			return r;
 	}
 
-	return amdgpu_cs_sync_rings(p);
-}
-
-static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
-			     struct amdgpu_cs_parser *parser)
-{
-	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
-	struct amdgpu_vm *vm = &fpriv->vm;
-	int i, j;
-	int r, ce_preempt = 0, de_preempt = 0;
-
-	for (i = 0, j = 0; i < parser->nchunks && j < parser->job->num_ibs; i++) {
-		struct amdgpu_cs_chunk *chunk;
-		struct amdgpu_ib *ib;
-		struct drm_amdgpu_cs_chunk_ib *chunk_ib;
-		struct amdgpu_ring *ring;
+	for (i = 0; i < p->gang_size; ++i) {
+		r = amdgpu_sync_push_to_job(&p->sync, p->jobs[i]);
+		if (r)
+			return r;
+	}
 
-		chunk = &parser->chunks[i];
-		ib = &parser->job->ibs[j];
-		chunk_ib = (struct drm_amdgpu_cs_chunk_ib *)chunk->kdata;
+	sched = p->gang_leader->base.entity->rq->sched;
+	while ((fence = amdgpu_sync_get_fence(&p->sync))) {
+		struct drm_sched_fence *s_fence = to_drm_sched_fence(fence);
 
-		if (chunk->chunk_id != AMDGPU_CHUNK_ID_IB)
+		/*
+		 * When we have an dependency it might be necessary to insert a
+		 * pipeline sync to make sure that all caches etc are flushed and the
+		 * next job actually sees the results from the previous one
+		 * before we start executing on the same scheduler ring.
+		 */
+		if (!s_fence || s_fence->sched != sched) {
+			dma_fence_put(fence);
 			continue;
-
-		if (chunk_ib->ip_type == AMDGPU_HW_IP_GFX && amdgpu_sriov_vf(adev)) {
-			if (chunk_ib->flags & AMDGPU_IB_FLAG_PREEMPT) {
-				if (chunk_ib->flags & AMDGPU_IB_FLAG_CE)
-					ce_preempt++;
-				else
-					de_preempt++;
-			}
-
-			/* each GFX command submit allows 0 or 1 IB preemptible for CE & DE */
-			if (ce_preempt > 1 || de_preempt > 1)
-				return -EINVAL;
 		}
 
-		r = amdgpu_cs_get_ring(adev, chunk_ib->ip_type,
-				       chunk_ib->ip_instance, chunk_ib->ring,
-				       &ring);
+		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence,
+				      GFP_KERNEL);
+		dma_fence_put(fence);
 		if (r)
 			return r;
+	}
+	return 0;
+}
 
-		if (chunk_ib->flags & AMDGPU_IB_FLAG_PREAMBLE) {
-			parser->job->preamble_status |= AMDGPU_PREAMBLE_IB_PRESENT;
-			if (!parser->ctx->preamble_presented) {
-				parser->job->preamble_status |= AMDGPU_PREAMBLE_IB_PRESENT_FIRST;
-				parser->ctx->preamble_presented = true;
-			}
-		}
-
-		if (parser->job->ring && parser->job->ring != ring)
-			return -EINVAL;
-
-		parser->job->ring = ring;
-
-		if (ring->funcs->parse_cs) {
-			struct amdgpu_bo_va_mapping *m;
-			struct amdgpu_bo *aobj = NULL;
-			uint64_t offset;
-			uint8_t *kptr;
-
-			m = amdgpu_cs_find_mapping(parser, chunk_ib->va_start,
-						   &aobj);
-			if (!aobj) {
-				DRM_ERROR("IB va_start is invalid\n");
-				return -EINVAL;
-			}
+static void amdgpu_cs_post_dependencies(struct amdgpu_cs_parser *p)
+{
+	int i;
 
-			if ((chunk_ib->va_start + chunk_ib->ib_bytes) >
-			    (m->last + 1) * AMDGPU_GPU_PAGE_SIZE) {
-				DRM_ERROR("IB va_start+ib_bytes is invalid\n");
-				return -EINVAL;
-			}
+	for (i = 0; i < p->num_post_deps; ++i) {
+		if (p->post_deps[i].chain && p->post_deps[i].point) {
+			drm_syncobj_add_point(p->post_deps[i].syncobj,
+					      p->post_deps[i].chain,
+					      p->fence, p->post_deps[i].point);
+			p->post_deps[i].chain = NULL;
+		} else {
+			drm_syncobj_replace_fence(p->post_deps[i].syncobj,
+						  p->fence);
+		}
+	}
+}
 
-			/* the IB should be reserved at this point */
-			r = amdgpu_bo_kmap(aobj, (void **)&kptr);
-			if (r) {
-				return r;
-			}
+static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
+			    union drm_amdgpu_cs *cs)
+{
+	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct amdgpu_job *leader = p->gang_leader;
+	struct amdgpu_bo_list_entry *e;
+	struct drm_gem_object *gobj;
+	unsigned long index;
+	unsigned int i;
+	uint64_t seq;
+	int r;
 
-			offset = m->start * AMDGPU_GPU_PAGE_SIZE;
-			kptr += chunk_ib->va_start - offset;
+	for (i = 0; i < p->gang_size; ++i)
+		drm_sched_job_arm(&p->jobs[i]->base);
 
-			r =  amdgpu_ib_get(adev, vm, chunk_ib->ib_bytes, ib);
-			if (r) {
-				DRM_ERROR("Failed to get ib !\n");
-				return r;
-			}
+	for (i = 0; i < p->gang_size; ++i) {
+		struct dma_fence *fence;
 
-			memcpy(ib->ptr, kptr, chunk_ib->ib_bytes);
-			amdgpu_bo_kunmap(aobj);
-		} else {
-			r =  amdgpu_ib_get(adev, vm, 0, ib);
-			if (r) {
-				DRM_ERROR("Failed to get ib !\n");
-				return r;
-			}
+		if (p->jobs[i] == leader)
+			continue;
 
+		fence = &p->jobs[i]->base.s_fence->scheduled;
+		dma_fence_get(fence);
+		r = drm_sched_job_add_dependency(&leader->base, fence);
+		if (r) {
+			dma_fence_put(fence);
+			return r;
 		}
-
-		ib->gpu_addr = chunk_ib->va_start;
-		ib->length_dw = chunk_ib->ib_bytes / 4;
-		ib->flags = chunk_ib->flags;
-		j++;
 	}
 
-	/* UVD & VCE fw doesn't support user fences */
-	if (parser->job->uf_addr && (
-	    parser->job->ring->funcs->type == AMDGPU_RING_TYPE_UVD ||
-	    parser->job->ring->funcs->type == AMDGPU_RING_TYPE_VCE))
-		return -EINVAL;
+	if (p->gang_size > 1) {
+		for (i = 0; i < p->gang_size; ++i)
+			amdgpu_job_set_gang_leader(p->jobs[i], leader);
+	}
 
-	return 0;
-}
+	/* No memory allocation is allowed while holding the notifier lock.
+	 * The lock is held until amdgpu_cs_submit is finished and fence is
+	 * added to BOs.
+	 */
+	mutex_lock(&p->adev->notifier_lock);
 
-static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
-				  struct amdgpu_cs_parser *p)
-{
-	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	int i, j, r;
+	/* If userptr are invalidated after amdgpu_cs_parser_bos(), return
+	 * -EAGAIN, drmIoctl in libdrm will restart the amdgpu_cs_ioctl.
+	 */
+	r = 0;
+	amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
+		r |= !amdgpu_hmm_range_valid(e->range);
+		amdgpu_hmm_range_free(e->range);
+		e->range = NULL;
+	}
+	if (r) {
+		r = -EAGAIN;
+		mutex_unlock(&p->adev->notifier_lock);
+		return r;
+	}
 
-	for (i = 0; i < p->nchunks; ++i) {
-		struct drm_amdgpu_cs_chunk_dep *deps;
-		struct amdgpu_cs_chunk *chunk;
-		unsigned num_deps;
+	p->fence = dma_fence_get(&leader->base.s_fence->finished);
+	drm_exec_for_each_locked_object(&p->exec, index, gobj) {
 
-		chunk = &p->chunks[i];
+		ttm_bo_move_to_lru_tail_unlocked(&gem_to_amdgpu_bo(gobj)->tbo);
 
-		if (chunk->chunk_id != AMDGPU_CHUNK_ID_DEPENDENCIES)
-			continue;
+		/* Everybody except for the gang leader uses READ */
+		for (i = 0; i < p->gang_size; ++i) {
+			if (p->jobs[i] == leader)
+				continue;
 
-		deps = (struct drm_amdgpu_cs_chunk_dep *)chunk->kdata;
-		num_deps = chunk->length_dw * 4 /
-			sizeof(struct drm_amdgpu_cs_chunk_dep);
+			dma_resv_add_fence(gobj->resv,
+					   &p->jobs[i]->base.s_fence->finished,
+					   DMA_RESV_USAGE_READ);
+		}
 
-		for (j = 0; j < num_deps; ++j) {
-			struct amdgpu_ring *ring;
-			struct amdgpu_ctx *ctx;
-			struct dma_fence *fence;
+		/* The gang leader as remembered as writer */
+		dma_resv_add_fence(gobj->resv, p->fence, DMA_RESV_USAGE_WRITE);
+	}
 
-			r = amdgpu_cs_get_ring(adev, deps[j].ip_type,
-					       deps[j].ip_instance,
-					       deps[j].ring, &ring);
-			if (r)
-				return r;
+	seq = amdgpu_ctx_add_fence(p->ctx, p->entities[p->gang_leader_idx],
+				   p->fence);
+	amdgpu_cs_post_dependencies(p);
 
-			ctx = amdgpu_ctx_get(fpriv, deps[j].ctx_id);
-			if (ctx == NULL)
-				return -EINVAL;
+	if ((leader->preamble_status & AMDGPU_PREAMBLE_IB_PRESENT) &&
+	    !p->ctx->preamble_presented) {
+		leader->preamble_status |= AMDGPU_PREAMBLE_IB_PRESENT_FIRST;
+		p->ctx->preamble_presented = true;
+	}
 
-			fence = amdgpu_ctx_get_fence(ctx, ring,
-						     deps[j].handle);
-			if (IS_ERR(fence)) {
-				r = PTR_ERR(fence);
-				amdgpu_ctx_put(ctx);
-				return r;
+	cs->out.handle = seq;
+	leader->uf_sequence = seq;
 
-			} else if (fence) {
-				r = amdgpu_sync_fence(adev, &p->job->sync,
-						      fence);
-				dma_fence_put(fence);
-				amdgpu_ctx_put(ctx);
-				if (r)
-					return r;
-			}
-		}
+	amdgpu_vm_bo_trace_cs(&fpriv->vm, &p->exec.ticket);
+	for (i = 0; i < p->gang_size; ++i) {
+		amdgpu_job_free_resources(p->jobs[i]);
+		trace_amdgpu_cs_ioctl(p->jobs[i]);
+		drm_sched_entity_push_job(&p->jobs[i]->base);
+		p->jobs[i] = NULL;
 	}
 
+	amdgpu_vm_move_to_lru_tail(p->adev, &fpriv->vm);
+
+	mutex_unlock(&p->adev->notifier_lock);
+	mutex_unlock(&p->bo_list->bo_list_mutex);
 	return 0;
 }
 
-static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
-			    union drm_amdgpu_cs *cs)
+/* Cleanup the parser structure */
+static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser)
 {
-	struct amdgpu_ring *ring = p->job->ring;
-	struct amd_sched_entity *entity = &p->ctx->rings[ring->idx].entity;
-	struct amdgpu_job *job;
-	int r;
+	unsigned int i;
 
-	job = p->job;
-	p->job = NULL;
+	amdgpu_sync_free(&parser->sync);
+	drm_exec_fini(&parser->exec);
 
-	r = amd_sched_job_init(&job->base, &ring->sched, entity, p->filp);
-	if (r) {
-		amdgpu_job_free(job);
-		return r;
+	for (i = 0; i < parser->num_post_deps; i++) {
+		drm_syncobj_put(parser->post_deps[i].syncobj);
+		kfree(parser->post_deps[i].chain);
 	}
+	kfree(parser->post_deps);
 
-	job->owner = p->filp;
-	job->fence_ctx = entity->fence_context;
-	p->fence = dma_fence_get(&job->base.s_fence->finished);
-	cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, p->fence);
-	job->uf_sequence = cs->out.handle;
-	amdgpu_job_free_resources(job);
-	amdgpu_cs_parser_fini(p, 0, true);
+	dma_fence_put(parser->fence);
 
-	trace_amdgpu_cs_ioctl(job);
-	amd_sched_entity_push_job(&job->base);
+	if (parser->ctx)
+		amdgpu_ctx_put(parser->ctx);
+	if (parser->bo_list)
+		amdgpu_bo_list_put(parser->bo_list);
 
-	return 0;
+	for (i = 0; i < parser->nchunks; i++)
+		kvfree(parser->chunks[i].kdata);
+	kvfree(parser->chunks);
+	for (i = 0; i < parser->gang_size; ++i) {
+		if (parser->jobs[i])
+			amdgpu_job_free(parser->jobs[i]);
+	}
+	amdgpu_bo_unref(&parser->uf_bo);
 }
 
 int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
-	struct amdgpu_device *adev = dev->dev_private;
-	union drm_amdgpu_cs *cs = data;
-	struct amdgpu_cs_parser parser = {};
-	bool reserved_buffers = false;
-	int i, r;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	struct amdgpu_cs_parser parser;
+	int r;
+
+	if (amdgpu_ras_intr_triggered())
+		return -EHWPOISON;
 
 	if (!adev->accel_working)
 		return -EBUSY;
 
-	parser.adev = adev;
-	parser.filp = filp;
-
-	r = amdgpu_cs_parser_init(&parser, data);
+	r = amdgpu_cs_parser_init(&parser, adev, filp, data);
 	if (r) {
-		DRM_ERROR("Failed to initialize parser !\n");
-		goto out;
+		drm_err_ratelimited(dev, "Failed to initialize parser %d!\n", r);
+		return r;
 	}
 
+	r = amdgpu_cs_pass1(&parser, data);
+	if (r)
+		goto error_fini;
+
+	r = amdgpu_cs_pass2(&parser);
+	if (r)
+		goto error_fini;
+
 	r = amdgpu_cs_parser_bos(&parser, data);
 	if (r) {
 		if (r == -ENOMEM)
-			DRM_ERROR("Not enough memory for command submission!\n");
-		else if (r != -ERESTARTSYS)
-			DRM_ERROR("Failed to process the buffer list %d!\n", r);
-		goto out;
+			drm_err(dev, "Not enough memory for command submission!\n");
+		else if (r != -ERESTARTSYS && r != -EAGAIN)
+			drm_dbg(dev, "Failed to process the buffer list %d!\n", r);
+		goto error_fini;
 	}
 
-	reserved_buffers = true;
-	r = amdgpu_cs_ib_fill(adev, &parser);
+	r = amdgpu_cs_patch_jobs(&parser);
 	if (r)
-		goto out;
+		goto error_backoff;
 
-	r = amdgpu_cs_dependencies(adev, &parser);
-	if (r) {
-		DRM_ERROR("Failed in the dependencies handling %d!\n", r);
-		goto out;
-	}
-
-	for (i = 0; i < parser.job->num_ibs; i++)
-		trace_amdgpu_cs(&parser, i);
+	r = amdgpu_cs_vm_handling(&parser);
+	if (r)
+		goto error_backoff;
 
-	r = amdgpu_cs_ib_vm_chunk(adev, &parser);
+	r = amdgpu_cs_sync_rings(&parser);
 	if (r)
-		goto out;
+		goto error_backoff;
+
+	trace_amdgpu_cs_ibs(&parser);
 
-	r = amdgpu_cs_submit(&parser, cs);
+	r = amdgpu_cs_submit(&parser, data);
 	if (r)
-		goto out;
+		goto error_backoff;
 
+	amdgpu_cs_parser_fini(&parser);
 	return 0;
-out:
-	amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
+
+error_backoff:
+	mutex_unlock(&parser.bo_list->bo_list_mutex);
+
+error_fini:
+	amdgpu_cs_parser_fini(&parser);
 	return r;
 }
 
@@ -1152,27 +1491,30 @@ int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
 			 struct drm_file *filp)
 {
 	union drm_amdgpu_wait_cs *wait = data;
-	struct amdgpu_device *adev = dev->dev_private;
 	unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
-	struct amdgpu_ring *ring = NULL;
+	struct drm_sched_entity *entity;
 	struct amdgpu_ctx *ctx;
 	struct dma_fence *fence;
 	long r;
 
-	r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
-			       wait->in.ring, &ring);
-	if (r)
-		return r;
-
 	ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
 	if (ctx == NULL)
 		return -EINVAL;
 
-	fence = amdgpu_ctx_get_fence(ctx, ring, wait->in.handle);
+	r = amdgpu_ctx_get_entity(ctx, wait->in.ip_type, wait->in.ip_instance,
+				  wait->in.ring, &entity);
+	if (r) {
+		amdgpu_ctx_put(ctx);
+		return r;
+	}
+
+	fence = amdgpu_ctx_get_fence(ctx, entity, wait->in.handle);
 	if (IS_ERR(fence))
 		r = PTR_ERR(fence);
 	else if (fence) {
 		r = dma_fence_wait_timeout(fence, true, timeout);
+		if (r > 0 && fence->error)
+			r = fence->error;
 		dma_fence_put(fence);
 	} else
 		r = 1;
@@ -1198,28 +1540,90 @@ static struct dma_fence *amdgpu_cs_get_fence(struct amdgpu_device *adev,
 					     struct drm_file *filp,
 					     struct drm_amdgpu_fence *user)
 {
-	struct amdgpu_ring *ring;
+	struct drm_sched_entity *entity;
 	struct amdgpu_ctx *ctx;
 	struct dma_fence *fence;
 	int r;
 
-	r = amdgpu_cs_get_ring(adev, user->ip_type, user->ip_instance,
-			       user->ring, &ring);
-	if (r)
-		return ERR_PTR(r);
-
 	ctx = amdgpu_ctx_get(filp->driver_priv, user->ctx_id);
 	if (ctx == NULL)
 		return ERR_PTR(-EINVAL);
 
-	fence = amdgpu_ctx_get_fence(ctx, ring, user->seq_no);
+	r = amdgpu_ctx_get_entity(ctx, user->ip_type, user->ip_instance,
+				  user->ring, &entity);
+	if (r) {
+		amdgpu_ctx_put(ctx);
+		return ERR_PTR(r);
+	}
+
+	fence = amdgpu_ctx_get_fence(ctx, entity, user->seq_no);
 	amdgpu_ctx_put(ctx);
 
 	return fence;
 }
 
+int amdgpu_cs_fence_to_handle_ioctl(struct drm_device *dev, void *data,
+				    struct drm_file *filp)
+{
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	union drm_amdgpu_fence_to_handle *info = data;
+	struct dma_fence *fence;
+	struct drm_syncobj *syncobj;
+	struct sync_file *sync_file;
+	int fd, r;
+
+	fence = amdgpu_cs_get_fence(adev, filp, &info->in.fence);
+	if (IS_ERR(fence))
+		return PTR_ERR(fence);
+
+	if (!fence)
+		fence = dma_fence_get_stub();
+
+	switch (info->in.what) {
+	case AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ:
+		r = drm_syncobj_create(&syncobj, 0, fence);
+		dma_fence_put(fence);
+		if (r)
+			return r;
+		r = drm_syncobj_get_handle(filp, syncobj, &info->out.handle);
+		drm_syncobj_put(syncobj);
+		return r;
+
+	case AMDGPU_FENCE_TO_HANDLE_GET_SYNCOBJ_FD:
+		r = drm_syncobj_create(&syncobj, 0, fence);
+		dma_fence_put(fence);
+		if (r)
+			return r;
+		r = drm_syncobj_get_fd(syncobj, (int *)&info->out.handle);
+		drm_syncobj_put(syncobj);
+		return r;
+
+	case AMDGPU_FENCE_TO_HANDLE_GET_SYNC_FILE_FD:
+		fd = get_unused_fd_flags(O_CLOEXEC);
+		if (fd < 0) {
+			dma_fence_put(fence);
+			return fd;
+		}
+
+		sync_file = sync_file_create(fence);
+		dma_fence_put(fence);
+		if (!sync_file) {
+			put_unused_fd(fd);
+			return -ENOMEM;
+		}
+
+		fd_install(fd, sync_file->file);
+		info->out.handle = fd;
+		return 0;
+
+	default:
+		dma_fence_put(fence);
+		return -EINVAL;
+	}
+}
+
 /**
- * amdgpu_cs_wait_all_fence - wait on all fences to signal
+ * amdgpu_cs_wait_all_fences - wait on all fences to signal
  *
  * @adev: amdgpu device
  * @filp: file private
@@ -1246,6 +1650,9 @@ static int amdgpu_cs_wait_all_fences(struct amdgpu_device *adev,
 			continue;
 
 		r = dma_fence_wait_timeout(fence, true, timeout);
+		if (r > 0 && fence->error)
+			r = fence->error;
+
 		dma_fence_put(fence);
 		if (r < 0)
 			return r;
@@ -1297,6 +1704,7 @@ static int amdgpu_cs_wait_any_fence(struct amdgpu_device *adev,
 			array[i] = fence;
 		} else { /* NULL, the fence has been already signaled */
 			r = 1;
+			first = i;
 			goto out;
 		}
 	}
@@ -1310,8 +1718,11 @@ out:
 	memset(wait, 0, sizeof(*wait));
 	wait->out.status = (r > 0);
 	wait->out.first_signaled = first;
-	/* set return value 0 to indicate success */
-	r = 0;
+
+	if (first < fence_count && array[first])
+		r = array[first]->error;
+	else
+		r = 0;
 
 err_free_fence_array:
 	for (i = 0; i < fence_count; i++)
@@ -1331,120 +1742,75 @@ err_free_fence_array:
 int amdgpu_cs_wait_fences_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *filp)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	union drm_amdgpu_wait_fences *wait = data;
-	uint32_t fence_count = wait->in.fence_count;
-	struct drm_amdgpu_fence *fences_user;
 	struct drm_amdgpu_fence *fences;
 	int r;
 
 	/* Get the fences from userspace */
-	fences = kmalloc_array(fence_count, sizeof(struct drm_amdgpu_fence),
-			GFP_KERNEL);
-	if (fences == NULL)
-		return -ENOMEM;
-
-	fences_user = (void __user *)(uintptr_t)(wait->in.fences);
-	if (copy_from_user(fences, fences_user,
-		sizeof(struct drm_amdgpu_fence) * fence_count)) {
-		r = -EFAULT;
-		goto err_free_fences;
-	}
+	fences = memdup_array_user(u64_to_user_ptr(wait->in.fences),
+				   wait->in.fence_count,
+				   sizeof(struct drm_amdgpu_fence));
+	if (IS_ERR(fences))
+		return PTR_ERR(fences);
 
 	if (wait->in.wait_all)
 		r = amdgpu_cs_wait_all_fences(adev, filp, wait, fences);
 	else
 		r = amdgpu_cs_wait_any_fence(adev, filp, wait, fences);
 
-err_free_fences:
 	kfree(fences);
 
 	return r;
 }
 
 /**
- * amdgpu_cs_find_bo_va - find bo_va for VM address
+ * amdgpu_cs_find_mapping - find bo_va for VM address
  *
  * @parser: command submission parser context
  * @addr: VM address
  * @bo: resulting BO of the mapping found
+ * @map: Placeholder to return found BO mapping
  *
  * Search the buffer objects in the command submission context for a certain
  * virtual memory address. Returns allocation structure when found, NULL
  * otherwise.
  */
-struct amdgpu_bo_va_mapping *
-amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
-		       uint64_t addr, struct amdgpu_bo **bo)
+int amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
+			   uint64_t addr, struct amdgpu_bo **bo,
+			   struct amdgpu_bo_va_mapping **map)
 {
+	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
+	struct ttm_operation_ctx ctx = { false, false };
+	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_va_mapping *mapping;
-	unsigned i;
-
-	if (!parser->bo_list)
-		return NULL;
+	int i, r;
 
 	addr /= AMDGPU_GPU_PAGE_SIZE;
 
-	for (i = 0; i < parser->bo_list->num_entries; i++) {
-		struct amdgpu_bo_list_entry *lobj;
-
-		lobj = &parser->bo_list->array[i];
-		if (!lobj->bo_va)
-			continue;
-
-		list_for_each_entry(mapping, &lobj->bo_va->valids, list) {
-			if (mapping->start > addr ||
-			    addr > mapping->last)
-				continue;
-
-			*bo = lobj->bo_va->bo;
-			return mapping;
-		}
-
-		list_for_each_entry(mapping, &lobj->bo_va->invalids, list) {
-			if (mapping->start > addr ||
-			    addr > mapping->last)
-				continue;
-
-			*bo = lobj->bo_va->bo;
-			return mapping;
-		}
-	}
-
-	return NULL;
-}
-
-/**
- * amdgpu_cs_sysvm_access_required - make BOs accessible by the system VM
- *
- * @parser: command submission parser context
- *
- * Helper for UVD/VCE VM emulation, make sure BOs are accessible by the system VM.
- */
-int amdgpu_cs_sysvm_access_required(struct amdgpu_cs_parser *parser)
-{
-	unsigned i;
-	int r;
-
-	if (!parser->bo_list)
-		return 0;
+	mapping = amdgpu_vm_bo_lookup_mapping(vm, addr);
+	if (!mapping || !mapping->bo_va || !mapping->bo_va->base.bo)
+		return -EINVAL;
 
-	for (i = 0; i < parser->bo_list->num_entries; i++) {
-		struct amdgpu_bo *bo = parser->bo_list->array[i].robj;
+	*bo = mapping->bo_va->base.bo;
+	*map = mapping;
 
-		r = amdgpu_ttm_bind(&bo->tbo, &bo->tbo.mem);
-		if (unlikely(r))
-			return r;
+	/* Double check that the BO is reserved by this CS */
+	if (dma_resv_locking_ctx((*bo)->tbo.base.resv) != &parser->exec.ticket)
+		return -EINVAL;
 
-		if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
-			continue;
+	/* Make sure VRAM is allocated contigiously */
+	(*bo)->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+	if ((*bo)->tbo.resource->mem_type == TTM_PL_VRAM &&
+	    !((*bo)->tbo.resource->placement & TTM_PL_FLAG_CONTIGUOUS)) {
 
-		bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-		amdgpu_ttm_placement_from_domain(bo, bo->allowed_domains);
-		r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-		if (unlikely(r))
+		amdgpu_bo_placement_from_domain(*bo, (*bo)->allowed_domains);
+		for (i = 0; i < (*bo)->placement.num_placement; i++)
+			(*bo)->placements[i].flags |= TTM_PL_FLAG_CONTIGUOUS;
+		r = ttm_bo_validate(&(*bo)->tbo, &(*bo)->placement, &ctx);
+		if (r)
 			return r;
 	}
 
-	return 0;
+	return amdgpu_ttm_alloc_gart(&(*bo)->tbo);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
new file mode 100644
index 000000000000..39c33ad100cb
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.h
@@ -0,0 +1,88 @@
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_CS_H__
+#define __AMDGPU_CS_H__
+
+#include <linux/ww_mutex.h>
+#include <drm/drm_exec.h>
+
+#include "amdgpu_job.h"
+#include "amdgpu_bo_list.h"
+#include "amdgpu_ring.h"
+
+#define AMDGPU_CS_GANG_SIZE	4
+
+struct amdgpu_bo_va_mapping;
+
+struct amdgpu_cs_chunk {
+	uint32_t		chunk_id;
+	uint32_t		length_dw;
+	void			*kdata;
+};
+
+struct amdgpu_cs_post_dep {
+	struct drm_syncobj *syncobj;
+	struct dma_fence_chain *chain;
+	u64 point;
+};
+
+struct amdgpu_cs_parser {
+	struct amdgpu_device	*adev;
+	struct drm_file		*filp;
+	struct amdgpu_ctx	*ctx;
+
+	/* chunks */
+	unsigned		nchunks;
+	struct amdgpu_cs_chunk	*chunks;
+
+	/* scheduler job objects */
+	unsigned int		gang_size;
+	unsigned int		gang_leader_idx;
+	struct drm_sched_entity	*entities[AMDGPU_CS_GANG_SIZE];
+	struct amdgpu_job	*jobs[AMDGPU_CS_GANG_SIZE];
+	struct amdgpu_job	*gang_leader;
+
+	/* buffer objects */
+	struct drm_exec			exec;
+	struct amdgpu_bo_list		*bo_list;
+	struct amdgpu_mn		*mn;
+	struct dma_fence		*fence;
+	uint64_t			bytes_moved_threshold;
+	uint64_t			bytes_moved_vis_threshold;
+	uint64_t			bytes_moved;
+	uint64_t			bytes_moved_vis;
+
+	/* user fence */
+	struct amdgpu_bo		*uf_bo;
+
+	unsigned			num_post_deps;
+	struct amdgpu_cs_post_dep	*post_deps;
+
+	struct amdgpu_sync		sync;
+};
+
+int amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
+			   uint64_t addr, struct amdgpu_bo **bo,
+			   struct amdgpu_bo_va_mapping **mapping);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
new file mode 100644
index 000000000000..02138aa55793
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.c
@@ -0,0 +1,135 @@
+/*
+ * Copyright 2016 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+
+ * * Author: Monk.liu@amd.com
+ */
+
+#include <drm/drm_exec.h>
+
+#include "amdgpu.h"
+
+uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev)
+{
+	uint64_t addr = AMDGPU_VA_RESERVED_CSA_START(adev);
+
+	addr = amdgpu_gmc_sign_extend(addr);
+
+	return addr;
+}
+
+int amdgpu_allocate_static_csa(struct amdgpu_device *adev, struct amdgpu_bo **bo,
+				u32 domain, uint32_t size)
+{
+	void *ptr;
+
+	amdgpu_bo_create_kernel(adev, size, PAGE_SIZE,
+				domain, bo,
+				NULL, &ptr);
+	if (!*bo)
+		return -ENOMEM;
+
+	memset(ptr, 0, size);
+	adev->virt.csa_cpu_addr = ptr;
+	return 0;
+}
+
+void amdgpu_free_static_csa(struct amdgpu_bo **bo)
+{
+	amdgpu_bo_free_kernel(bo, NULL, NULL);
+}
+
+/*
+ * amdgpu_map_static_csa should be called during amdgpu_vm_init
+ * it maps virtual address amdgpu_csa_vaddr() to this VM, and each command
+ * submission of GFX should use this virtual address within META_DATA init
+ * package to support SRIOV gfx preemption.
+ */
+int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			  struct amdgpu_bo *bo, struct amdgpu_bo_va **bo_va,
+			  uint64_t csa_addr, uint32_t size)
+{
+	struct drm_exec exec;
+	int r;
+
+	drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0);
+	drm_exec_until_all_locked(&exec) {
+		r = amdgpu_vm_lock_pd(vm, &exec, 0);
+		if (likely(!r))
+			r = drm_exec_lock_obj(&exec, &bo->tbo.base);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(r)) {
+			DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r);
+			goto error;
+		}
+	}
+
+	*bo_va = amdgpu_vm_bo_add(adev, vm, bo);
+	if (!*bo_va) {
+		r = -ENOMEM;
+		goto error;
+	}
+
+	r = amdgpu_vm_bo_map(adev, *bo_va, csa_addr, 0, size,
+			     AMDGPU_PTE_READABLE | AMDGPU_PTE_WRITEABLE |
+			     AMDGPU_PTE_EXECUTABLE);
+
+	if (r) {
+		DRM_ERROR("failed to do bo_map on static CSA, err=%d\n", r);
+		amdgpu_vm_bo_del(adev, *bo_va);
+		goto error;
+	}
+
+error:
+	drm_exec_fini(&exec);
+	return r;
+}
+
+int amdgpu_unmap_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			    struct amdgpu_bo *bo, struct amdgpu_bo_va *bo_va,
+			    uint64_t csa_addr)
+{
+	struct drm_exec exec;
+	int r;
+
+	drm_exec_init(&exec, 0, 0);
+	drm_exec_until_all_locked(&exec) {
+		r = amdgpu_vm_lock_pd(vm, &exec, 0);
+		if (likely(!r))
+			r = drm_exec_lock_obj(&exec, &bo->tbo.base);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(r)) {
+			DRM_ERROR("failed to reserve CSA,PD BOs: err=%d\n", r);
+			goto error;
+		}
+	}
+
+	r = amdgpu_vm_bo_unmap(adev, bo_va, csa_addr);
+	if (r) {
+		DRM_ERROR("failed to do bo_unmap on static CSA, err=%d\n", r);
+		goto error;
+	}
+
+	amdgpu_vm_bo_del(adev, bo_va);
+
+error:
+	drm_exec_fini(&exec);
+	return r;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.h
new file mode 100644
index 000000000000..7dfc1f2012eb
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_csa.h
@@ -0,0 +1,42 @@
+/*
+ * Copyright 2016 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Author: Monk.liu@amd.com
+ */
+
+#ifndef AMDGPU_CSA_MANAGER_H
+#define AMDGPU_CSA_MANAGER_H
+
+#define AMDGPU_CSA_SIZE		(128 * 1024)
+
+uint32_t amdgpu_get_total_csa_size(struct amdgpu_device *adev);
+uint64_t amdgpu_csa_vaddr(struct amdgpu_device *adev);
+int amdgpu_allocate_static_csa(struct amdgpu_device *adev, struct amdgpu_bo **bo,
+				u32 domain, uint32_t size);
+int amdgpu_map_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			  struct amdgpu_bo *bo, struct amdgpu_bo_va **bo_va,
+			  uint64_t csa_addr, uint32_t size);
+int amdgpu_unmap_static_csa(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			    struct amdgpu_bo *bo, struct amdgpu_bo_va *bo_va,
+			    uint64_t csa_addr);
+void amdgpu_free_static_csa(struct amdgpu_bo **bo);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 90d1ac8a80f8..afedea02188d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -22,74 +22,458 @@
  * Authors: monk liu <monk.liu@amd.com>
  */
 
-#include <drm/drmP.h>
+#include <drm/drm_auth.h>
+#include <drm/drm_drv.h>
 #include "amdgpu.h"
+#include "amdgpu_sched.h"
+#include "amdgpu_ras.h"
+#include <linux/nospec.h>
+
+#define to_amdgpu_ctx_entity(e)	\
+	container_of((e), struct amdgpu_ctx_entity, entity)
+
+const unsigned int amdgpu_ctx_num_entities[AMDGPU_HW_IP_NUM] = {
+	[AMDGPU_HW_IP_GFX]	=	1,
+	[AMDGPU_HW_IP_COMPUTE]	=	4,
+	[AMDGPU_HW_IP_DMA]	=	2,
+	[AMDGPU_HW_IP_UVD]	=	1,
+	[AMDGPU_HW_IP_VCE]	=	1,
+	[AMDGPU_HW_IP_UVD_ENC]	=	1,
+	[AMDGPU_HW_IP_VCN_DEC]	=	1,
+	[AMDGPU_HW_IP_VCN_ENC]	=	1,
+	[AMDGPU_HW_IP_VCN_JPEG]	=	1,
+	[AMDGPU_HW_IP_VPE]	=	1,
+};
+
+bool amdgpu_ctx_priority_is_valid(int32_t ctx_prio)
+{
+	switch (ctx_prio) {
+	case AMDGPU_CTX_PRIORITY_VERY_LOW:
+	case AMDGPU_CTX_PRIORITY_LOW:
+	case AMDGPU_CTX_PRIORITY_NORMAL:
+	case AMDGPU_CTX_PRIORITY_HIGH:
+	case AMDGPU_CTX_PRIORITY_VERY_HIGH:
+		return true;
+	default:
+	case AMDGPU_CTX_PRIORITY_UNSET:
+		/* UNSET priority is not valid and we don't carry that
+		 * around, but set it to NORMAL in the only place this
+		 * function is called, amdgpu_ctx_ioctl().
+		 */
+		return false;
+	}
+}
 
-static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ctx *ctx)
+static enum drm_sched_priority
+amdgpu_ctx_to_drm_sched_prio(int32_t ctx_prio)
 {
-	unsigned i, j;
+	switch (ctx_prio) {
+	case AMDGPU_CTX_PRIORITY_UNSET:
+		pr_warn_once("AMD-->DRM context priority value UNSET-->NORMAL");
+		return DRM_SCHED_PRIORITY_NORMAL;
+
+	case AMDGPU_CTX_PRIORITY_VERY_LOW:
+		return DRM_SCHED_PRIORITY_LOW;
+
+	case AMDGPU_CTX_PRIORITY_LOW:
+		return DRM_SCHED_PRIORITY_LOW;
+
+	case AMDGPU_CTX_PRIORITY_NORMAL:
+		return DRM_SCHED_PRIORITY_NORMAL;
+
+	case AMDGPU_CTX_PRIORITY_HIGH:
+		return DRM_SCHED_PRIORITY_HIGH;
+
+	case AMDGPU_CTX_PRIORITY_VERY_HIGH:
+		return DRM_SCHED_PRIORITY_HIGH;
+
+	/* This should not happen as we sanitized userspace provided priority
+	 * already, WARN if this happens.
+	 */
+	default:
+		WARN(1, "Invalid context priority %d\n", ctx_prio);
+		return DRM_SCHED_PRIORITY_NORMAL;
+	}
+
+}
+
+static int amdgpu_ctx_priority_permit(struct drm_file *filp,
+				      int32_t priority)
+{
+	/* NORMAL and below are accessible by everyone */
+	if (priority <= AMDGPU_CTX_PRIORITY_NORMAL)
+		return 0;
+
+	if (capable(CAP_SYS_NICE))
+		return 0;
+
+	if (drm_is_current_master(filp))
+		return 0;
+
+	return -EACCES;
+}
+
+static enum amdgpu_gfx_pipe_priority amdgpu_ctx_prio_to_gfx_pipe_prio(int32_t prio)
+{
+	switch (prio) {
+	case AMDGPU_CTX_PRIORITY_HIGH:
+	case AMDGPU_CTX_PRIORITY_VERY_HIGH:
+		return AMDGPU_GFX_PIPE_PRIO_HIGH;
+	default:
+		return AMDGPU_GFX_PIPE_PRIO_NORMAL;
+	}
+}
+
+static enum amdgpu_ring_priority_level amdgpu_ctx_sched_prio_to_ring_prio(int32_t prio)
+{
+	switch (prio) {
+	case AMDGPU_CTX_PRIORITY_HIGH:
+		return AMDGPU_RING_PRIO_1;
+	case AMDGPU_CTX_PRIORITY_VERY_HIGH:
+		return AMDGPU_RING_PRIO_2;
+	default:
+		return AMDGPU_RING_PRIO_0;
+	}
+}
+
+static unsigned int amdgpu_ctx_get_hw_prio(struct amdgpu_ctx *ctx, u32 hw_ip)
+{
+	struct amdgpu_device *adev = ctx->mgr->adev;
+	unsigned int hw_prio;
+	int32_t ctx_prio;
+
+	ctx_prio = (ctx->override_priority == AMDGPU_CTX_PRIORITY_UNSET) ?
+			ctx->init_priority : ctx->override_priority;
+
+	switch (hw_ip) {
+	case AMDGPU_HW_IP_GFX:
+	case AMDGPU_HW_IP_COMPUTE:
+		hw_prio = amdgpu_ctx_prio_to_gfx_pipe_prio(ctx_prio);
+		break;
+	case AMDGPU_HW_IP_VCE:
+	case AMDGPU_HW_IP_VCN_ENC:
+		hw_prio = amdgpu_ctx_sched_prio_to_ring_prio(ctx_prio);
+		break;
+	default:
+		hw_prio = AMDGPU_RING_PRIO_DEFAULT;
+		break;
+	}
+
+	hw_ip = array_index_nospec(hw_ip, AMDGPU_HW_IP_NUM);
+	if (adev->gpu_sched[hw_ip][hw_prio].num_scheds == 0)
+		hw_prio = AMDGPU_RING_PRIO_DEFAULT;
+
+	return hw_prio;
+}
+
+/* Calculate the time spend on the hw */
+static ktime_t amdgpu_ctx_fence_time(struct dma_fence *fence)
+{
+	struct drm_sched_fence *s_fence;
+
+	if (!fence)
+		return ns_to_ktime(0);
+
+	/* When the fence is not even scheduled it can't have spend time */
+	s_fence = to_drm_sched_fence(fence);
+	if (!test_bit(DMA_FENCE_FLAG_TIMESTAMP_BIT, &s_fence->scheduled.flags))
+		return ns_to_ktime(0);
+
+	/* When it is still running account how much already spend */
+	if (!test_bit(DMA_FENCE_FLAG_TIMESTAMP_BIT, &s_fence->finished.flags))
+		return ktime_sub(ktime_get(), s_fence->scheduled.timestamp);
+
+	return ktime_sub(s_fence->finished.timestamp,
+			 s_fence->scheduled.timestamp);
+}
+
+static ktime_t amdgpu_ctx_entity_time(struct amdgpu_ctx *ctx,
+				      struct amdgpu_ctx_entity *centity)
+{
+	ktime_t res = ns_to_ktime(0);
+	uint32_t i;
+
+	spin_lock(&ctx->ring_lock);
+	for (i = 0; i < amdgpu_sched_jobs; i++) {
+		res = ktime_add(res, amdgpu_ctx_fence_time(centity->fences[i]));
+	}
+	spin_unlock(&ctx->ring_lock);
+	return res;
+}
+
+static int amdgpu_ctx_init_entity(struct amdgpu_ctx *ctx, u32 hw_ip,
+				  const u32 ring)
+{
+	struct drm_gpu_scheduler **scheds = NULL, *sched = NULL;
+	struct amdgpu_device *adev = ctx->mgr->adev;
+	struct amdgpu_ctx_entity *entity;
+	enum drm_sched_priority drm_prio;
+	unsigned int hw_prio, num_scheds;
+	int32_t ctx_prio;
+	int r;
+
+	entity = kzalloc(struct_size(entity, fences, amdgpu_sched_jobs),
+			 GFP_KERNEL);
+	if (!entity)
+		return  -ENOMEM;
+
+	ctx_prio = (ctx->override_priority == AMDGPU_CTX_PRIORITY_UNSET) ?
+			ctx->init_priority : ctx->override_priority;
+	entity->hw_ip = hw_ip;
+	entity->sequence = 1;
+	hw_prio = amdgpu_ctx_get_hw_prio(ctx, hw_ip);
+	drm_prio = amdgpu_ctx_to_drm_sched_prio(ctx_prio);
+
+	hw_ip = array_index_nospec(hw_ip, AMDGPU_HW_IP_NUM);
+
+	if (!(adev)->xcp_mgr) {
+		scheds = adev->gpu_sched[hw_ip][hw_prio].sched;
+		num_scheds = adev->gpu_sched[hw_ip][hw_prio].num_scheds;
+	} else {
+		struct amdgpu_fpriv *fpriv;
+
+		fpriv = container_of(ctx->ctx_mgr, struct amdgpu_fpriv, ctx_mgr);
+		r = amdgpu_xcp_select_scheds(adev, hw_ip, hw_prio, fpriv,
+						&num_scheds, &scheds);
+		if (r)
+			goto error_free_entity;
+	}
+
+	/* disable load balance if the hw engine retains context among dependent jobs */
+	if (hw_ip == AMDGPU_HW_IP_VCN_ENC ||
+	    hw_ip == AMDGPU_HW_IP_VCN_DEC ||
+	    hw_ip == AMDGPU_HW_IP_UVD_ENC ||
+	    hw_ip == AMDGPU_HW_IP_UVD) {
+		sched = drm_sched_pick_best(scheds, num_scheds);
+		scheds = &sched;
+		num_scheds = 1;
+	}
+
+	r = drm_sched_entity_init(&entity->entity, drm_prio, scheds, num_scheds,
+				  &ctx->guilty);
+	if (r)
+		goto error_free_entity;
+
+	/* It's not an error if we fail to install the new entity */
+	if (cmpxchg(&ctx->entities[hw_ip][ring], NULL, entity))
+		goto cleanup_entity;
+
+	return 0;
+
+cleanup_entity:
+	drm_sched_entity_fini(&entity->entity);
+
+error_free_entity:
+	kfree(entity);
+
+	return r;
+}
+
+static ktime_t amdgpu_ctx_fini_entity(struct amdgpu_device *adev,
+				  struct amdgpu_ctx_entity *entity)
+{
+	ktime_t res = ns_to_ktime(0);
+	int i;
+
+	if (!entity)
+		return res;
+
+	for (i = 0; i < amdgpu_sched_jobs; ++i) {
+		res = ktime_add(res, amdgpu_ctx_fence_time(entity->fences[i]));
+		dma_fence_put(entity->fences[i]);
+	}
+
+	amdgpu_xcp_release_sched(adev, entity);
+
+	kfree(entity);
+	return res;
+}
+
+static int amdgpu_ctx_get_stable_pstate(struct amdgpu_ctx *ctx,
+					u32 *stable_pstate)
+{
+	struct amdgpu_device *adev = ctx->mgr->adev;
+	enum amd_dpm_forced_level current_level;
+
+	current_level = amdgpu_dpm_get_performance_level(adev);
+
+	switch (current_level) {
+	case AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_STANDARD;
+		break;
+	case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK;
+		break;
+	case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK;
+		break;
+	case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_PEAK;
+		break;
+	default:
+		*stable_pstate = AMDGPU_CTX_STABLE_PSTATE_NONE;
+		break;
+	}
+	return 0;
+}
+
+static int amdgpu_ctx_init(struct amdgpu_ctx_mgr *mgr, int32_t priority,
+			   struct drm_file *filp, struct amdgpu_ctx *ctx)
+{
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
+	u32 current_stable_pstate;
 	int r;
 
+	r = amdgpu_ctx_priority_permit(filp, priority);
+	if (r)
+		return r;
+
 	memset(ctx, 0, sizeof(*ctx));
-	ctx->adev = adev;
+
 	kref_init(&ctx->refcount);
+	ctx->mgr = mgr;
 	spin_lock_init(&ctx->ring_lock);
-	ctx->fences = kcalloc(amdgpu_sched_jobs * AMDGPU_MAX_RINGS,
-			      sizeof(struct dma_fence*), GFP_KERNEL);
-	if (!ctx->fences)
-		return -ENOMEM;
 
-	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		ctx->rings[i].sequence = 1;
-		ctx->rings[i].fences = &ctx->fences[amdgpu_sched_jobs * i];
-	}
+	ctx->reset_counter = atomic_read(&mgr->adev->gpu_reset_counter);
+	ctx->reset_counter_query = ctx->reset_counter;
+	ctx->generation = amdgpu_vm_generation(mgr->adev, &fpriv->vm);
+	ctx->init_priority = priority;
+	ctx->override_priority = AMDGPU_CTX_PRIORITY_UNSET;
+
+	r = amdgpu_ctx_get_stable_pstate(ctx, &current_stable_pstate);
+	if (r)
+		return r;
 
-	ctx->reset_counter = atomic_read(&adev->gpu_reset_counter);
+	if (mgr->adev->pm.stable_pstate_ctx)
+		ctx->stable_pstate = mgr->adev->pm.stable_pstate_ctx->stable_pstate;
+	else
+		ctx->stable_pstate = current_stable_pstate;
 
-	/* create context entity for each ring */
-	for (i = 0; i < adev->num_rings; i++) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		struct amd_sched_rq *rq;
+	ctx->ctx_mgr = &(fpriv->ctx_mgr);
+	return 0;
+}
 
-		rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL];
-		r = amd_sched_entity_init(&ring->sched, &ctx->rings[i].entity,
-					  rq, amdgpu_sched_jobs);
-		if (r)
-			goto failed;
+static int amdgpu_ctx_set_stable_pstate(struct amdgpu_ctx *ctx,
+					u32 stable_pstate)
+{
+	struct amdgpu_device *adev = ctx->mgr->adev;
+	enum amd_dpm_forced_level level;
+	u32 current_stable_pstate;
+	int r;
+
+	mutex_lock(&adev->pm.stable_pstate_ctx_lock);
+	if (adev->pm.stable_pstate_ctx && adev->pm.stable_pstate_ctx != ctx) {
+		r = -EBUSY;
+		goto done;
 	}
 
-	return 0;
+	r = amdgpu_ctx_get_stable_pstate(ctx, &current_stable_pstate);
+	if (r || (stable_pstate == current_stable_pstate))
+		goto done;
+
+	switch (stable_pstate) {
+	case AMDGPU_CTX_STABLE_PSTATE_NONE:
+		level = AMD_DPM_FORCED_LEVEL_AUTO;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_STANDARD:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_MIN_SCLK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_MIN_MCLK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK;
+		break;
+	case AMDGPU_CTX_STABLE_PSTATE_PEAK:
+		level = AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
+		break;
+	default:
+		r = -EINVAL;
+		goto done;
+	}
+
+	r = amdgpu_dpm_force_performance_level(adev, level);
+
+	if (level == AMD_DPM_FORCED_LEVEL_AUTO)
+		adev->pm.stable_pstate_ctx = NULL;
+	else
+		adev->pm.stable_pstate_ctx = ctx;
+done:
+	mutex_unlock(&adev->pm.stable_pstate_ctx_lock);
 
-failed:
-	for (j = 0; j < i; j++)
-		amd_sched_entity_fini(&adev->rings[j]->sched,
-				      &ctx->rings[j].entity);
-	kfree(ctx->fences);
-	ctx->fences = NULL;
 	return r;
 }
 
-static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
+static void amdgpu_ctx_fini(struct kref *ref)
 {
-	struct amdgpu_device *adev = ctx->adev;
-	unsigned i, j;
+	struct amdgpu_ctx *ctx = container_of(ref, struct amdgpu_ctx, refcount);
+	struct amdgpu_ctx_mgr *mgr = ctx->mgr;
+	struct amdgpu_device *adev = mgr->adev;
+	unsigned i, j, idx;
 
 	if (!adev)
 		return;
 
-	for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
-		for (j = 0; j < amdgpu_sched_jobs; ++j)
-			dma_fence_put(ctx->rings[i].fences[j]);
-	kfree(ctx->fences);
-	ctx->fences = NULL;
+	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
+		for (j = 0; j < AMDGPU_MAX_ENTITY_NUM; ++j) {
+			ktime_t spend;
 
-	for (i = 0; i < adev->num_rings; i++)
-		amd_sched_entity_fini(&adev->rings[i]->sched,
-				      &ctx->rings[i].entity);
+			spend = amdgpu_ctx_fini_entity(adev, ctx->entities[i][j]);
+			atomic64_add(ktime_to_ns(spend), &mgr->time_spend[i]);
+		}
+	}
+
+	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
+		amdgpu_ctx_set_stable_pstate(ctx, ctx->stable_pstate);
+		drm_dev_exit(idx);
+	}
+
+	kfree(ctx);
+}
+
+int amdgpu_ctx_get_entity(struct amdgpu_ctx *ctx, u32 hw_ip, u32 instance,
+			  u32 ring, struct drm_sched_entity **entity)
+{
+	int r;
+	struct drm_sched_entity *ctx_entity;
+
+	if (hw_ip >= AMDGPU_HW_IP_NUM) {
+		DRM_ERROR("unknown HW IP type: %d\n", hw_ip);
+		return -EINVAL;
+	}
+
+	/* Right now all IPs have only one instance - multiple rings. */
+	if (instance != 0) {
+		DRM_DEBUG("invalid ip instance: %d\n", instance);
+		return -EINVAL;
+	}
+
+	if (ring >= amdgpu_ctx_num_entities[hw_ip]) {
+		DRM_DEBUG("invalid ring: %d %d\n", hw_ip, ring);
+		return -EINVAL;
+	}
+
+	if (ctx->entities[hw_ip][ring] == NULL) {
+		r = amdgpu_ctx_init_entity(ctx, hw_ip, ring);
+		if (r)
+			return r;
+	}
+
+	ctx_entity = &ctx->entities[hw_ip][ring]->entity;
+	r = drm_sched_entity_error(ctx_entity);
+	if (r) {
+		DRM_DEBUG("error entity %p\n", ctx_entity);
+		return r;
+	}
+
+	*entity = ctx_entity;
+	return 0;
 }
 
 static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
 			    struct amdgpu_fpriv *fpriv,
+			    struct drm_file *filp,
+			    int32_t priority,
 			    uint32_t *id)
 {
 	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
@@ -101,14 +485,15 @@ static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
 		return -ENOMEM;
 
 	mutex_lock(&mgr->lock);
-	r = idr_alloc(&mgr->ctx_handles, ctx, 1, 0, GFP_KERNEL);
+	r = idr_alloc(&mgr->ctx_handles, ctx, 1, AMDGPU_VM_MAX_NUM_CTX, GFP_KERNEL);
 	if (r < 0) {
 		mutex_unlock(&mgr->lock);
 		kfree(ctx);
 		return r;
 	}
+
 	*id = (uint32_t)r;
-	r = amdgpu_ctx_init(adev, ctx);
+	r = amdgpu_ctx_init(mgr, priority, filp, ctx);
 	if (r) {
 		idr_remove(&mgr->ctx_handles, *id);
 		*id = 0;
@@ -121,12 +506,19 @@ static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
 static void amdgpu_ctx_do_release(struct kref *ref)
 {
 	struct amdgpu_ctx *ctx;
+	u32 i, j;
 
 	ctx = container_of(ref, struct amdgpu_ctx, refcount);
+	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
+		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) {
+			if (!ctx->entities[i][j])
+				continue;
 
-	amdgpu_ctx_fini(ctx);
+			drm_sched_entity_destroy(&ctx->entities[i][j]->entity);
+		}
+	}
 
-	kfree(ctx);
+	amdgpu_ctx_fini(ref);
 }
 
 static int amdgpu_ctx_free(struct amdgpu_fpriv *fpriv, uint32_t id)
@@ -168,40 +560,166 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
 	/* determine if a GPU reset has occured since the last call */
 	reset_counter = atomic_read(&adev->gpu_reset_counter);
 	/* TODO: this should ideally return NO, GUILTY, or INNOCENT. */
-	if (ctx->reset_counter == reset_counter)
+	if (ctx->reset_counter_query == reset_counter)
 		out->state.reset_status = AMDGPU_CTX_NO_RESET;
 	else
 		out->state.reset_status = AMDGPU_CTX_UNKNOWN_RESET;
-	ctx->reset_counter = reset_counter;
+	ctx->reset_counter_query = reset_counter;
+
+	mutex_unlock(&mgr->lock);
+	return 0;
+}
+
+#define AMDGPU_RAS_COUNTE_DELAY_MS 3000
+
+static int amdgpu_ctx_query2(struct amdgpu_device *adev,
+			     struct amdgpu_fpriv *fpriv, uint32_t id,
+			     union drm_amdgpu_ctx_out *out)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct amdgpu_ctx *ctx;
+	struct amdgpu_ctx_mgr *mgr;
+
+	if (!fpriv)
+		return -EINVAL;
+
+	mgr = &fpriv->ctx_mgr;
+	mutex_lock(&mgr->lock);
+	ctx = idr_find(&mgr->ctx_handles, id);
+	if (!ctx) {
+		mutex_unlock(&mgr->lock);
+		return -EINVAL;
+	}
+
+	out->state.flags = 0x0;
+	out->state.hangs = 0x0;
+
+	if (ctx->reset_counter != atomic_read(&adev->gpu_reset_counter))
+		out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_RESET;
+
+	if (ctx->generation != amdgpu_vm_generation(adev, &fpriv->vm))
+		out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_VRAMLOST;
+
+	if (atomic_read(&ctx->guilty))
+		out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_GUILTY;
+
+	if (amdgpu_in_reset(adev))
+		out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS;
+
+	if (adev->ras_enabled && con) {
+		/* Return the cached values in O(1),
+		 * and schedule delayed work to cache
+		 * new vaues.
+		 */
+		int ce_count, ue_count;
+
+		ce_count = atomic_read(&con->ras_ce_count);
+		ue_count = atomic_read(&con->ras_ue_count);
+
+		if (ce_count != ctx->ras_counter_ce) {
+			ctx->ras_counter_ce = ce_count;
+			out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_RAS_CE;
+		}
+
+		if (ue_count != ctx->ras_counter_ue) {
+			ctx->ras_counter_ue = ue_count;
+			out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_RAS_UE;
+		}
+
+		schedule_delayed_work(&con->ras_counte_delay_work,
+				      msecs_to_jiffies(AMDGPU_RAS_COUNTE_DELAY_MS));
+	}
 
 	mutex_unlock(&mgr->lock);
 	return 0;
 }
 
+static int amdgpu_ctx_stable_pstate(struct amdgpu_device *adev,
+				    struct amdgpu_fpriv *fpriv, uint32_t id,
+				    bool set, u32 *stable_pstate)
+{
+	struct amdgpu_ctx *ctx;
+	struct amdgpu_ctx_mgr *mgr;
+	int r;
+
+	if (!fpriv)
+		return -EINVAL;
+
+	mgr = &fpriv->ctx_mgr;
+	mutex_lock(&mgr->lock);
+	ctx = idr_find(&mgr->ctx_handles, id);
+	if (!ctx) {
+		mutex_unlock(&mgr->lock);
+		return -EINVAL;
+	}
+
+	if (set)
+		r = amdgpu_ctx_set_stable_pstate(ctx, *stable_pstate);
+	else
+		r = amdgpu_ctx_get_stable_pstate(ctx, stable_pstate);
+
+	mutex_unlock(&mgr->lock);
+	return r;
+}
+
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp)
 {
 	int r;
-	uint32_t id;
+	uint32_t id, stable_pstate;
+	int32_t priority;
 
 	union drm_amdgpu_ctx *args = data;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 
-	r = 0;
 	id = args->in.ctx_id;
+	priority = args->in.priority;
+
+	/* For backwards compatibility, we need to accept ioctls with garbage
+	 * in the priority field. Garbage values in the priority field, result
+	 * in the priority being set to NORMAL.
+	 */
+	if (!amdgpu_ctx_priority_is_valid(priority))
+		priority = AMDGPU_CTX_PRIORITY_NORMAL;
 
 	switch (args->in.op) {
 	case AMDGPU_CTX_OP_ALLOC_CTX:
-		r = amdgpu_ctx_alloc(adev, fpriv, &id);
+		if (args->in.flags)
+			return -EINVAL;
+		r = amdgpu_ctx_alloc(adev, fpriv, filp, priority, &id);
 		args->out.alloc.ctx_id = id;
 		break;
 	case AMDGPU_CTX_OP_FREE_CTX:
+		if (args->in.flags)
+			return -EINVAL;
 		r = amdgpu_ctx_free(fpriv, id);
 		break;
 	case AMDGPU_CTX_OP_QUERY_STATE:
+		if (args->in.flags)
+			return -EINVAL;
 		r = amdgpu_ctx_query(adev, fpriv, id, &args->out);
 		break;
+	case AMDGPU_CTX_OP_QUERY_STATE2:
+		if (args->in.flags)
+			return -EINVAL;
+		r = amdgpu_ctx_query2(adev, fpriv, id, &args->out);
+		break;
+	case AMDGPU_CTX_OP_GET_STABLE_PSTATE:
+		if (args->in.flags)
+			return -EINVAL;
+		r = amdgpu_ctx_stable_pstate(adev, fpriv, id, false, &stable_pstate);
+		if (!r)
+			args->out.pstate.flags = stable_pstate;
+		break;
+	case AMDGPU_CTX_OP_SET_STABLE_PSTATE:
+		if (args->in.flags & ~AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK)
+			return -EINVAL;
+		stable_pstate = args->in.flags & AMDGPU_CTX_STABLE_PSTATE_FLAGS_MASK;
+		if (stable_pstate > AMDGPU_CTX_STABLE_PSTATE_PEAK)
+			return -EINVAL;
+		r = amdgpu_ctx_stable_pstate(adev, fpriv, id, true, &stable_pstate);
+		break;
 	default:
 		return -EINVAL;
 	}
@@ -236,82 +754,239 @@ int amdgpu_ctx_put(struct amdgpu_ctx *ctx)
 	return 0;
 }
 
-uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
+uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx,
+			      struct drm_sched_entity *entity,
 			      struct dma_fence *fence)
 {
-	struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
-	uint64_t seq = cring->sequence;
-	unsigned idx = 0;
+	struct amdgpu_ctx_entity *centity = to_amdgpu_ctx_entity(entity);
+	uint64_t seq = centity->sequence;
 	struct dma_fence *other = NULL;
+	unsigned idx = 0;
 
 	idx = seq & (amdgpu_sched_jobs - 1);
-	other = cring->fences[idx];
-	if (other) {
-		signed long r;
-		r = dma_fence_wait_timeout(other, false, MAX_SCHEDULE_TIMEOUT);
-		if (r < 0)
-			DRM_ERROR("Error (%ld) waiting for fence!\n", r);
-	}
+	other = centity->fences[idx];
+	WARN_ON(other && !dma_fence_is_signaled(other));
 
 	dma_fence_get(fence);
 
 	spin_lock(&ctx->ring_lock);
-	cring->fences[idx] = fence;
-	cring->sequence++;
+	centity->fences[idx] = fence;
+	centity->sequence++;
 	spin_unlock(&ctx->ring_lock);
 
-	dma_fence_put(other);
+	atomic64_add(ktime_to_ns(amdgpu_ctx_fence_time(other)),
+		     &ctx->mgr->time_spend[centity->hw_ip]);
 
+	dma_fence_put(other);
 	return seq;
 }
 
 struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
-				       struct amdgpu_ring *ring, uint64_t seq)
+				       struct drm_sched_entity *entity,
+				       uint64_t seq)
 {
-	struct amdgpu_ctx_ring *cring = & ctx->rings[ring->idx];
+	struct amdgpu_ctx_entity *centity = to_amdgpu_ctx_entity(entity);
 	struct dma_fence *fence;
 
 	spin_lock(&ctx->ring_lock);
 
 	if (seq == ~0ull)
-		seq = ctx->rings[ring->idx].sequence - 1;
+		seq = centity->sequence - 1;
 
-	if (seq >= cring->sequence) {
+	if (seq >= centity->sequence) {
 		spin_unlock(&ctx->ring_lock);
 		return ERR_PTR(-EINVAL);
 	}
 
 
-	if (seq + amdgpu_sched_jobs < cring->sequence) {
+	if (seq + amdgpu_sched_jobs < centity->sequence) {
 		spin_unlock(&ctx->ring_lock);
 		return NULL;
 	}
 
-	fence = dma_fence_get(cring->fences[seq & (amdgpu_sched_jobs - 1)]);
+	fence = dma_fence_get(centity->fences[seq & (amdgpu_sched_jobs - 1)]);
 	spin_unlock(&ctx->ring_lock);
 
 	return fence;
 }
 
-void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr)
+static void amdgpu_ctx_set_entity_priority(struct amdgpu_ctx *ctx,
+					   struct amdgpu_ctx_entity *aentity,
+					   int hw_ip,
+					   int32_t priority)
+{
+	struct amdgpu_device *adev = ctx->mgr->adev;
+	unsigned int hw_prio;
+	struct drm_gpu_scheduler **scheds = NULL;
+	unsigned num_scheds;
+
+	/* set sw priority */
+	drm_sched_entity_set_priority(&aentity->entity,
+				      amdgpu_ctx_to_drm_sched_prio(priority));
+
+	/* set hw priority */
+	if (hw_ip == AMDGPU_HW_IP_COMPUTE || hw_ip == AMDGPU_HW_IP_GFX) {
+		hw_prio = amdgpu_ctx_get_hw_prio(ctx, hw_ip);
+		hw_prio = array_index_nospec(hw_prio, AMDGPU_RING_PRIO_MAX);
+		scheds = adev->gpu_sched[hw_ip][hw_prio].sched;
+		num_scheds = adev->gpu_sched[hw_ip][hw_prio].num_scheds;
+		drm_sched_entity_modify_sched(&aentity->entity, scheds,
+					      num_scheds);
+	}
+}
+
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
+				  int32_t priority)
 {
+	int32_t ctx_prio;
+	unsigned i, j;
+
+	ctx->override_priority = priority;
+
+	ctx_prio = (ctx->override_priority == AMDGPU_CTX_PRIORITY_UNSET) ?
+			ctx->init_priority : ctx->override_priority;
+	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
+		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) {
+			if (!ctx->entities[i][j])
+				continue;
+
+			amdgpu_ctx_set_entity_priority(ctx, ctx->entities[i][j],
+						       i, ctx_prio);
+		}
+	}
+}
+
+int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx,
+			       struct drm_sched_entity *entity)
+{
+	struct amdgpu_ctx_entity *centity = to_amdgpu_ctx_entity(entity);
+	struct dma_fence *other;
+	unsigned idx;
+	long r;
+
+	spin_lock(&ctx->ring_lock);
+	idx = centity->sequence & (amdgpu_sched_jobs - 1);
+	other = dma_fence_get(centity->fences[idx]);
+	spin_unlock(&ctx->ring_lock);
+
+	if (!other)
+		return 0;
+
+	r = dma_fence_wait(other, true);
+	if (r < 0 && r != -ERESTARTSYS)
+		DRM_ERROR("Error (%ld) waiting for fence!\n", r);
+
+	dma_fence_put(other);
+	return r;
+}
+
+void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr,
+			 struct amdgpu_device *adev)
+{
+	unsigned int i;
+
+	mgr->adev = adev;
 	mutex_init(&mgr->lock);
-	idr_init(&mgr->ctx_handles);
+	idr_init_base(&mgr->ctx_handles, 1);
+
+	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i)
+		atomic64_set(&mgr->time_spend[i], 0);
 }
 
-void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
+long amdgpu_ctx_mgr_entity_flush(struct amdgpu_ctx_mgr *mgr, long timeout)
 {
 	struct amdgpu_ctx *ctx;
 	struct idr *idp;
-	uint32_t id;
+	uint32_t id, i, j;
 
 	idp = &mgr->ctx_handles;
 
+	mutex_lock(&mgr->lock);
 	idr_for_each_entry(idp, ctx, id) {
-		if (kref_put(&ctx->refcount, amdgpu_ctx_do_release) != 1)
+		for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
+			for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) {
+				struct drm_sched_entity *entity;
+
+				if (!ctx->entities[i][j])
+					continue;
+
+				entity = &ctx->entities[i][j]->entity;
+				timeout = drm_sched_entity_flush(entity, timeout);
+			}
+		}
+	}
+	mutex_unlock(&mgr->lock);
+	return timeout;
+}
+
+static void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr)
+{
+	struct amdgpu_ctx *ctx;
+	struct idr *idp;
+	uint32_t id, i, j;
+
+	idp = &mgr->ctx_handles;
+
+	idr_for_each_entry(idp, ctx, id) {
+		if (kref_read(&ctx->refcount) != 1) {
 			DRM_ERROR("ctx %p is still alive\n", ctx);
+			continue;
+		}
+
+		for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
+			for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) {
+				struct drm_sched_entity *entity;
+
+				if (!ctx->entities[i][j])
+					continue;
+
+				entity = &ctx->entities[i][j]->entity;
+				drm_sched_entity_fini(entity);
+			}
+		}
+		kref_put(&ctx->refcount, amdgpu_ctx_fini);
 	}
+}
 
+void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
+{
+	amdgpu_ctx_mgr_entity_fini(mgr);
 	idr_destroy(&mgr->ctx_handles);
 	mutex_destroy(&mgr->lock);
 }
+
+void amdgpu_ctx_mgr_usage(struct amdgpu_ctx_mgr *mgr,
+			  ktime_t usage[AMDGPU_HW_IP_NUM])
+{
+	struct amdgpu_ctx *ctx;
+	unsigned int hw_ip, i;
+	uint32_t id;
+
+	/*
+	 * This is a little bit racy because it can be that a ctx or a fence are
+	 * destroyed just in the moment we try to account them. But that is ok
+	 * since exactly that case is explicitely allowed by the interface.
+	 */
+	mutex_lock(&mgr->lock);
+	for (hw_ip = 0; hw_ip < AMDGPU_HW_IP_NUM; ++hw_ip) {
+		uint64_t ns = atomic64_read(&mgr->time_spend[hw_ip]);
+
+		usage[hw_ip] = ns_to_ktime(ns);
+	}
+
+	idr_for_each_entry(&mgr->ctx_handles, ctx, id) {
+		for (hw_ip = 0; hw_ip < AMDGPU_HW_IP_NUM; ++hw_ip) {
+			for (i = 0; i < amdgpu_ctx_num_entities[hw_ip]; ++i) {
+				struct amdgpu_ctx_entity *centity;
+				ktime_t spend;
+
+				centity = ctx->entities[hw_ip][i];
+				if (!centity)
+					continue;
+				spend = amdgpu_ctx_entity_time(ctx, centity);
+				usage[hw_ip] = ktime_add(usage[hw_ip], spend);
+			}
+		}
+	}
+	mutex_unlock(&mgr->lock);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
new file mode 100644
index 000000000000..090dfe86f75b
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
@@ -0,0 +1,100 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_CTX_H__
+#define __AMDGPU_CTX_H__
+
+#include <linux/ktime.h>
+#include <linux/types.h>
+
+#include "amdgpu_ring.h"
+
+struct drm_device;
+struct drm_file;
+struct amdgpu_fpriv;
+struct amdgpu_ctx_mgr;
+
+#define AMDGPU_MAX_ENTITY_NUM 4
+
+struct amdgpu_ctx_entity {
+	uint32_t		hw_ip;
+	uint64_t		sequence;
+	struct drm_sched_entity	entity;
+	struct dma_fence	*fences[];
+};
+
+struct amdgpu_ctx {
+	struct kref			refcount;
+	struct amdgpu_ctx_mgr		*mgr;
+	unsigned			reset_counter;
+	unsigned			reset_counter_query;
+	uint64_t			generation;
+	spinlock_t			ring_lock;
+	struct amdgpu_ctx_entity	*entities[AMDGPU_HW_IP_NUM][AMDGPU_MAX_ENTITY_NUM];
+	bool				preamble_presented;
+	int32_t				init_priority;
+	int32_t				override_priority;
+	atomic_t			guilty;
+	unsigned long			ras_counter_ce;
+	unsigned long			ras_counter_ue;
+	uint32_t			stable_pstate;
+	struct amdgpu_ctx_mgr		*ctx_mgr;
+};
+
+struct amdgpu_ctx_mgr {
+	struct amdgpu_device	*adev;
+	struct mutex		lock;
+	/* protected by lock */
+	struct idr		ctx_handles;
+	atomic64_t		time_spend[AMDGPU_HW_IP_NUM];
+};
+
+extern const unsigned int amdgpu_ctx_num_entities[AMDGPU_HW_IP_NUM];
+
+struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id);
+int amdgpu_ctx_put(struct amdgpu_ctx *ctx);
+
+int amdgpu_ctx_get_entity(struct amdgpu_ctx *ctx, u32 hw_ip, u32 instance,
+			  u32 ring, struct drm_sched_entity **entity);
+uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx,
+			      struct drm_sched_entity *entity,
+			      struct dma_fence *fence);
+struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
+				       struct drm_sched_entity *entity,
+				       uint64_t seq);
+bool amdgpu_ctx_priority_is_valid(int32_t ctx_prio);
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx, int32_t ctx_prio);
+
+int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
+		     struct drm_file *filp);
+
+int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx,
+			       struct drm_sched_entity *entity);
+
+void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr,
+			 struct amdgpu_device *adev);
+long amdgpu_ctx_mgr_entity_flush(struct amdgpu_ctx_mgr *mgr, long timeout);
+void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
+void amdgpu_ctx_mgr_usage(struct amdgpu_ctx_mgr *mgr,
+			  ktime_t usage[AMDGPU_HW_IP_NUM]);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
new file mode 100644
index 000000000000..62d43b8cbe58
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -0,0 +1,2175 @@
+/*
+ * Copyright 2008 Advanced Micro Devices, Inc.
+ * Copyright 2008 Red Hat Inc.
+ * Copyright 2009 Jerome Glisse.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/kthread.h>
+#include <linux/pci.h>
+#include <linux/uaccess.h>
+#include <linux/pm_runtime.h>
+
+#include "amdgpu.h"
+#include "amdgpu_pm.h"
+#include "amdgpu_dm_debugfs.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_rap.h"
+#include "amdgpu_securedisplay.h"
+#include "amdgpu_fw_attestation.h"
+#include "amdgpu_umr.h"
+
+#include "amdgpu_reset.h"
+#include "amdgpu_psp_ta.h"
+
+#if defined(CONFIG_DEBUG_FS)
+
+/**
+ * amdgpu_debugfs_process_reg_op - Handle MMIO register reads/writes
+ *
+ * @read: True if reading
+ * @f: open file handle
+ * @buf: User buffer to write/read to
+ * @size: Number of bytes to write/read
+ * @pos:  Offset to seek to
+ *
+ * This debugfs entry has special meaning on the offset being sought.
+ * Various bits have different meanings:
+ *
+ * Bit 62:  Indicates a GRBM bank switch is needed
+ * Bit 61:  Indicates a SRBM bank switch is needed (implies bit 62 is
+ *	    zero)
+ * Bits 24..33: The SE or ME selector if needed
+ * Bits 34..43: The SH (or SA) or PIPE selector if needed
+ * Bits 44..53: The INSTANCE (or CU/WGP) or QUEUE selector if needed
+ *
+ * Bit 23:  Indicates that the PM power gating lock should be held
+ *	    This is necessary to read registers that might be
+ *	    unreliable during a power gating transistion.
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
+static int  amdgpu_debugfs_process_reg_op(bool read, struct file *f,
+		char __user *buf, size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+	bool pm_pg_lock, use_bank, use_ring;
+	unsigned int instance_bank, sh_bank, se_bank, me, pipe, queue, vmid;
+
+	pm_pg_lock = use_bank = use_ring = false;
+	instance_bank = sh_bank = se_bank = me = pipe = queue = vmid = 0;
+
+	if (size & 0x3 || *pos & 0x3 ||
+			((*pos & (1ULL << 62)) && (*pos & (1ULL << 61))))
+		return -EINVAL;
+
+	/* are we reading registers for which a PG lock is necessary? */
+	pm_pg_lock = (*pos >> 23) & 1;
+
+	if (*pos & (1ULL << 62)) {
+		se_bank = (*pos & GENMASK_ULL(33, 24)) >> 24;
+		sh_bank = (*pos & GENMASK_ULL(43, 34)) >> 34;
+		instance_bank = (*pos & GENMASK_ULL(53, 44)) >> 44;
+
+		if (se_bank == 0x3FF)
+			se_bank = 0xFFFFFFFF;
+		if (sh_bank == 0x3FF)
+			sh_bank = 0xFFFFFFFF;
+		if (instance_bank == 0x3FF)
+			instance_bank = 0xFFFFFFFF;
+		use_bank = true;
+	} else if (*pos & (1ULL << 61)) {
+
+		me = (*pos & GENMASK_ULL(33, 24)) >> 24;
+		pipe = (*pos & GENMASK_ULL(43, 34)) >> 34;
+		queue = (*pos & GENMASK_ULL(53, 44)) >> 44;
+		vmid = (*pos & GENMASK_ULL(58, 54)) >> 54;
+
+		use_ring = true;
+	} else {
+		use_bank = use_ring = false;
+	}
+
+	*pos &= (1UL << 22) - 1;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	if (use_bank) {
+		if ((sh_bank != 0xFFFFFFFF && sh_bank >= adev->gfx.config.max_sh_per_se) ||
+		    (se_bank != 0xFFFFFFFF && se_bank >= adev->gfx.config.max_shader_engines)) {
+			pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+			amdgpu_virt_disable_access_debugfs(adev);
+			return -EINVAL;
+		}
+		mutex_lock(&adev->grbm_idx_mutex);
+		amdgpu_gfx_select_se_sh(adev, se_bank,
+					sh_bank, instance_bank, 0);
+	} else if (use_ring) {
+		mutex_lock(&adev->srbm_mutex);
+		amdgpu_gfx_select_me_pipe_q(adev, me, pipe, queue, vmid, 0);
+	}
+
+	if (pm_pg_lock)
+		mutex_lock(&adev->pm.mutex);
+
+	while (size) {
+		uint32_t value;
+
+		if (read) {
+			value = RREG32(*pos >> 2);
+			r = put_user(value, (uint32_t *)buf);
+		} else {
+			r = get_user(value, (uint32_t *)buf);
+			if (!r)
+				amdgpu_mm_wreg_mmio_rlc(adev, *pos >> 2, value, 0);
+		}
+		if (r) {
+			result = r;
+			goto end;
+		}
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+end:
+	if (use_bank) {
+		amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
+		mutex_unlock(&adev->grbm_idx_mutex);
+	} else if (use_ring) {
+		amdgpu_gfx_select_me_pipe_q(adev, 0, 0, 0, 0, 0);
+		mutex_unlock(&adev->srbm_mutex);
+	}
+
+	if (pm_pg_lock)
+		mutex_unlock(&adev->pm.mutex);
+
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	amdgpu_virt_disable_access_debugfs(adev);
+	return result;
+}
+
+/*
+ * amdgpu_debugfs_regs_read - Callback for reading MMIO registers
+ */
+static ssize_t amdgpu_debugfs_regs_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	return amdgpu_debugfs_process_reg_op(true, f, buf, size, pos);
+}
+
+/*
+ * amdgpu_debugfs_regs_write - Callback for writing MMIO registers
+ */
+static ssize_t amdgpu_debugfs_regs_write(struct file *f, const char __user *buf,
+					 size_t size, loff_t *pos)
+{
+	return amdgpu_debugfs_process_reg_op(false, f, (char __user *)buf, size, pos);
+}
+
+static int amdgpu_debugfs_regs2_open(struct inode *inode, struct file *file)
+{
+	struct amdgpu_debugfs_regs2_data *rd;
+
+	rd = kzalloc(sizeof(*rd), GFP_KERNEL);
+	if (!rd)
+		return -ENOMEM;
+	rd->adev = file_inode(file)->i_private;
+	file->private_data = rd;
+	mutex_init(&rd->lock);
+
+	return 0;
+}
+
+static int amdgpu_debugfs_regs2_release(struct inode *inode, struct file *file)
+{
+	struct amdgpu_debugfs_regs2_data *rd = file->private_data;
+
+	mutex_destroy(&rd->lock);
+	kfree(file->private_data);
+	return 0;
+}
+
+static ssize_t amdgpu_debugfs_regs2_op(struct file *f, char __user *buf, u32 offset, size_t size, int write_en)
+{
+	struct amdgpu_debugfs_regs2_data *rd = f->private_data;
+	struct amdgpu_device *adev = rd->adev;
+	ssize_t result = 0;
+	int r;
+	uint32_t value;
+
+	if (size & 0x3 || offset & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	mutex_lock(&rd->lock);
+
+	if (rd->id.use_grbm) {
+		if ((rd->id.grbm.sh != 0xFFFFFFFF && rd->id.grbm.sh >= adev->gfx.config.max_sh_per_se) ||
+		    (rd->id.grbm.se != 0xFFFFFFFF && rd->id.grbm.se >= adev->gfx.config.max_shader_engines)) {
+			pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+			amdgpu_virt_disable_access_debugfs(adev);
+			mutex_unlock(&rd->lock);
+			return -EINVAL;
+		}
+		mutex_lock(&adev->grbm_idx_mutex);
+		amdgpu_gfx_select_se_sh(adev, rd->id.grbm.se,
+						  rd->id.grbm.sh,
+						  rd->id.grbm.instance, rd->id.xcc_id);
+	}
+
+	if (rd->id.use_srbm) {
+		mutex_lock(&adev->srbm_mutex);
+		amdgpu_gfx_select_me_pipe_q(adev, rd->id.srbm.me, rd->id.srbm.pipe,
+					    rd->id.srbm.queue, rd->id.srbm.vmid, rd->id.xcc_id);
+	}
+
+	if (rd->id.pg_lock)
+		mutex_lock(&adev->pm.mutex);
+
+	while (size) {
+		if (!write_en) {
+			value = RREG32(offset >> 2);
+			r = put_user(value, (uint32_t *)buf);
+		} else {
+			r = get_user(value, (uint32_t *)buf);
+			if (!r)
+				amdgpu_mm_wreg_mmio_rlc(adev, offset >> 2, value, rd->id.xcc_id);
+		}
+		if (r) {
+			result = r;
+			goto end;
+		}
+		offset += 4;
+		size -= 4;
+		result += 4;
+		buf += 4;
+	}
+end:
+	if (rd->id.use_grbm) {
+		amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, rd->id.xcc_id);
+		mutex_unlock(&adev->grbm_idx_mutex);
+	}
+
+	if (rd->id.use_srbm) {
+		amdgpu_gfx_select_me_pipe_q(adev, 0, 0, 0, 0, rd->id.xcc_id);
+		mutex_unlock(&adev->srbm_mutex);
+	}
+
+	if (rd->id.pg_lock)
+		mutex_unlock(&adev->pm.mutex);
+
+	mutex_unlock(&rd->lock);
+
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	amdgpu_virt_disable_access_debugfs(adev);
+	return result;
+}
+
+static long amdgpu_debugfs_regs2_ioctl(struct file *f, unsigned int cmd, unsigned long data)
+{
+	struct amdgpu_debugfs_regs2_data *rd = f->private_data;
+	struct amdgpu_debugfs_regs2_iocdata v1_data;
+	int r;
+
+	mutex_lock(&rd->lock);
+
+	switch (cmd) {
+	case AMDGPU_DEBUGFS_REGS2_IOC_SET_STATE_V2:
+		r = copy_from_user(&rd->id, (struct amdgpu_debugfs_regs2_iocdata_v2 *)data,
+				   sizeof(rd->id));
+		if (r)
+			r = -EINVAL;
+		goto done;
+	case AMDGPU_DEBUGFS_REGS2_IOC_SET_STATE:
+		r = copy_from_user(&v1_data, (struct amdgpu_debugfs_regs2_iocdata *)data,
+				   sizeof(v1_data));
+		if (r) {
+			r = -EINVAL;
+			goto done;
+		}
+		goto v1_copy;
+	default:
+		r = -EINVAL;
+		goto done;
+	}
+
+v1_copy:
+	rd->id.use_srbm = v1_data.use_srbm;
+	rd->id.use_grbm = v1_data.use_grbm;
+	rd->id.pg_lock = v1_data.pg_lock;
+	rd->id.grbm.se = v1_data.grbm.se;
+	rd->id.grbm.sh = v1_data.grbm.sh;
+	rd->id.grbm.instance = v1_data.grbm.instance;
+	rd->id.srbm.me = v1_data.srbm.me;
+	rd->id.srbm.pipe = v1_data.srbm.pipe;
+	rd->id.srbm.queue = v1_data.srbm.queue;
+	rd->id.xcc_id = 0;
+done:
+	mutex_unlock(&rd->lock);
+	return r;
+}
+
+static ssize_t amdgpu_debugfs_regs2_read(struct file *f, char __user *buf, size_t size, loff_t *pos)
+{
+	return amdgpu_debugfs_regs2_op(f, buf, *pos, size, 0);
+}
+
+static ssize_t amdgpu_debugfs_regs2_write(struct file *f, const char __user *buf, size_t size, loff_t *pos)
+{
+	return amdgpu_debugfs_regs2_op(f, (char __user *)buf, *pos, size, 1);
+}
+
+static int amdgpu_debugfs_gprwave_open(struct inode *inode, struct file *file)
+{
+	struct amdgpu_debugfs_gprwave_data *rd;
+
+	rd = kzalloc(sizeof(*rd), GFP_KERNEL);
+	if (!rd)
+		return -ENOMEM;
+	rd->adev = file_inode(file)->i_private;
+	file->private_data = rd;
+	mutex_init(&rd->lock);
+
+	return 0;
+}
+
+static int amdgpu_debugfs_gprwave_release(struct inode *inode, struct file *file)
+{
+	struct amdgpu_debugfs_gprwave_data *rd = file->private_data;
+
+	mutex_destroy(&rd->lock);
+	kfree(file->private_data);
+	return 0;
+}
+
+static ssize_t amdgpu_debugfs_gprwave_read(struct file *f, char __user *buf, size_t size, loff_t *pos)
+{
+	struct amdgpu_debugfs_gprwave_data *rd = f->private_data;
+	struct amdgpu_device *adev = rd->adev;
+	ssize_t result = 0;
+	int r;
+	uint32_t *data, x;
+
+	if (size > 4096 || size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	data = kcalloc(1024, sizeof(*data), GFP_KERNEL);
+	if (!data) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		amdgpu_virt_disable_access_debugfs(adev);
+		return -ENOMEM;
+	}
+
+	/* switch to the specific se/sh/cu */
+	mutex_lock(&adev->grbm_idx_mutex);
+	amdgpu_gfx_select_se_sh(adev, rd->id.se, rd->id.sh, rd->id.cu, rd->id.xcc_id);
+
+	if (!rd->id.gpr_or_wave) {
+		x = 0;
+		if (adev->gfx.funcs->read_wave_data)
+			adev->gfx.funcs->read_wave_data(adev, rd->id.xcc_id, rd->id.simd, rd->id.wave, data, &x);
+	} else {
+		x = size >> 2;
+		if (rd->id.gpr.vpgr_or_sgpr) {
+			if (adev->gfx.funcs->read_wave_vgprs)
+				adev->gfx.funcs->read_wave_vgprs(adev, rd->id.xcc_id, rd->id.simd, rd->id.wave, rd->id.gpr.thread, *pos, size>>2, data);
+		} else {
+			if (adev->gfx.funcs->read_wave_sgprs)
+				adev->gfx.funcs->read_wave_sgprs(adev, rd->id.xcc_id, rd->id.simd, rd->id.wave, *pos, size>>2, data);
+		}
+	}
+
+	amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, rd->id.xcc_id);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	if (!x) {
+		result = -EINVAL;
+		goto done;
+	}
+
+	while (size && (*pos < x * 4)) {
+		uint32_t value;
+
+		value = data[*pos >> 2];
+		r = put_user(value, (uint32_t *)buf);
+		if (r) {
+			result = r;
+			goto done;
+		}
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+done:
+	amdgpu_virt_disable_access_debugfs(adev);
+	kfree(data);
+	return result;
+}
+
+static long amdgpu_debugfs_gprwave_ioctl(struct file *f, unsigned int cmd, unsigned long data)
+{
+	struct amdgpu_debugfs_gprwave_data *rd = f->private_data;
+	int r = 0;
+
+	mutex_lock(&rd->lock);
+
+	switch (cmd) {
+	case AMDGPU_DEBUGFS_GPRWAVE_IOC_SET_STATE:
+		if (copy_from_user(&rd->id,
+				   (struct amdgpu_debugfs_gprwave_iocdata *)data,
+				   sizeof(rd->id)))
+			r = -EFAULT;
+		goto done;
+	default:
+		r = -EINVAL;
+		goto done;
+	}
+
+done:
+	mutex_unlock(&rd->lock);
+	return r;
+}
+
+
+
+
+/**
+ * amdgpu_debugfs_regs_pcie_read - Read from a PCIE register
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
+static ssize_t amdgpu_debugfs_regs_pcie_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		if (upper_32_bits(*pos))
+			value = RREG32_PCIE_EXT(*pos);
+		else
+			value = RREG32_PCIE(*pos);
+
+		r = put_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+	amdgpu_virt_disable_access_debugfs(adev);
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_regs_pcie_write - Write to a PCIE register
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to write.  This
+ * allows writing multiple registers in a single call and having
+ * the returned size reflect that.
+ */
+static ssize_t amdgpu_debugfs_regs_pcie_write(struct file *f, const char __user *buf,
+					 size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		r = get_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		if (upper_32_bits(*pos))
+			WREG32_PCIE_EXT(*pos, value);
+		else
+			WREG32_PCIE(*pos, value);
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+	amdgpu_virt_disable_access_debugfs(adev);
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_regs_didt_read - Read from a DIDT register
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
+static ssize_t amdgpu_debugfs_regs_didt_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	if (!adev->didt_rreg)
+		return -EOPNOTSUPP;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		value = RREG32_DIDT(*pos >> 2);
+		r = put_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+	amdgpu_virt_disable_access_debugfs(adev);
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_regs_didt_write - Write to a DIDT register
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to write.  This
+ * allows writing multiple registers in a single call and having
+ * the returned size reflect that.
+ */
+static ssize_t amdgpu_debugfs_regs_didt_write(struct file *f, const char __user *buf,
+					 size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	if (!adev->didt_wreg)
+		return -EOPNOTSUPP;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		r = get_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		WREG32_DIDT(*pos >> 2, value);
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+	amdgpu_virt_disable_access_debugfs(adev);
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_regs_smc_read - Read from a SMC register
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to read.  This
+ * allows reading multiple registers in a single call and having
+ * the returned size reflect that.
+ */
+static ssize_t amdgpu_debugfs_regs_smc_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (!adev->smc_rreg)
+		return -EOPNOTSUPP;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		value = RREG32_SMC(*pos);
+		r = put_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+	amdgpu_virt_disable_access_debugfs(adev);
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_regs_smc_write - Write to a SMC register
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * The lower bits are the BYTE offset of the register to write.  This
+ * allows writing multiple registers in a single call and having
+ * the returned size reflect that.
+ */
+static ssize_t amdgpu_debugfs_regs_smc_write(struct file *f, const char __user *buf,
+					 size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (!adev->smc_wreg)
+		return -EOPNOTSUPP;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		r = get_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		WREG32_SMC(*pos, value);
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+	amdgpu_virt_disable_access_debugfs(adev);
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_gca_config_read - Read from gfx config data
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * This file is used to access configuration data in a somewhat
+ * stable fashion.  The format is a series of DWORDs with the first
+ * indicating which revision it is.  New content is appended to the
+ * end so that older software can still read the data.
+ */
+
+static ssize_t amdgpu_debugfs_gca_config_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+	uint32_t *config, no_regs = 0;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	config = kmalloc_array(256, sizeof(*config), GFP_KERNEL);
+	if (!config)
+		return -ENOMEM;
+
+	/* version, increment each time something is added */
+	config[no_regs++] = 5;
+	config[no_regs++] = adev->gfx.config.max_shader_engines;
+	config[no_regs++] = adev->gfx.config.max_tile_pipes;
+	config[no_regs++] = adev->gfx.config.max_cu_per_sh;
+	config[no_regs++] = adev->gfx.config.max_sh_per_se;
+	config[no_regs++] = adev->gfx.config.max_backends_per_se;
+	config[no_regs++] = adev->gfx.config.max_texture_channel_caches;
+	config[no_regs++] = adev->gfx.config.max_gprs;
+	config[no_regs++] = adev->gfx.config.max_gs_threads;
+	config[no_regs++] = adev->gfx.config.max_hw_contexts;
+	config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_frontend;
+	config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_backend;
+	config[no_regs++] = adev->gfx.config.sc_hiz_tile_fifo_size;
+	config[no_regs++] = adev->gfx.config.sc_earlyz_tile_fifo_size;
+	config[no_regs++] = adev->gfx.config.num_tile_pipes;
+	config[no_regs++] = adev->gfx.config.backend_enable_mask;
+	config[no_regs++] = adev->gfx.config.mem_max_burst_length_bytes;
+	config[no_regs++] = adev->gfx.config.mem_row_size_in_kb;
+	config[no_regs++] = adev->gfx.config.shader_engine_tile_size;
+	config[no_regs++] = adev->gfx.config.num_gpus;
+	config[no_regs++] = adev->gfx.config.multi_gpu_tile_size;
+	config[no_regs++] = adev->gfx.config.mc_arb_ramcfg;
+	config[no_regs++] = adev->gfx.config.gb_addr_config;
+	config[no_regs++] = adev->gfx.config.num_rbs;
+
+	/* rev==1 */
+	config[no_regs++] = adev->rev_id;
+	config[no_regs++] = adev->pg_flags;
+	config[no_regs++] = lower_32_bits(adev->cg_flags);
+
+	/* rev==2 */
+	config[no_regs++] = adev->family;
+	config[no_regs++] = adev->external_rev_id;
+
+	/* rev==3 */
+	config[no_regs++] = adev->pdev->device;
+	config[no_regs++] = adev->pdev->revision;
+	config[no_regs++] = adev->pdev->subsystem_device;
+	config[no_regs++] = adev->pdev->subsystem_vendor;
+
+	/* rev==4 APU flag */
+	config[no_regs++] = adev->flags & AMD_IS_APU ? 1 : 0;
+
+	/* rev==5 PG/CG flag upper 32bit */
+	config[no_regs++] = 0;
+	config[no_regs++] = upper_32_bits(adev->cg_flags);
+
+	while (size && (*pos < no_regs * 4)) {
+		uint32_t value;
+
+		value = config[*pos >> 2];
+		r = put_user(value, (uint32_t *)buf);
+		if (r) {
+			kfree(config);
+			return r;
+		}
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	kfree(config);
+	return result;
+}
+
+/**
+ * amdgpu_debugfs_sensor_read - Read from the powerplay sensors
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The offset is treated as the BYTE address of one of the sensors
+ * enumerated in amd/include/kgd_pp_interface.h under the
+ * 'amd_pp_sensors' enumeration.  For instance to read the UVD VCLK
+ * you would use the offset 3 * 4 = 12.
+ */
+static ssize_t amdgpu_debugfs_sensor_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	int idx, x, outsize, r, valuesize;
+	uint32_t values[16];
+
+	if (size & 3 || *pos & 0x3)
+		return -EINVAL;
+
+	if (!adev->pm.dpm_enabled)
+		return -EINVAL;
+
+	/* convert offset to sensor number */
+	idx = *pos >> 2;
+
+	valuesize = sizeof(values);
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_dpm_read_sensor(adev, idx, &values[0], &valuesize);
+
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	if (r) {
+		amdgpu_virt_disable_access_debugfs(adev);
+		return r;
+	}
+
+	if (size > valuesize) {
+		amdgpu_virt_disable_access_debugfs(adev);
+		return -EINVAL;
+	}
+
+	outsize = 0;
+	x = 0;
+	if (!r) {
+		while (size) {
+			r = put_user(values[x++], (int32_t *)buf);
+			buf += 4;
+			size -= 4;
+			outsize += 4;
+		}
+	}
+
+	amdgpu_virt_disable_access_debugfs(adev);
+	return !r ? outsize : r;
+}
+
+/** amdgpu_debugfs_wave_read - Read WAVE STATUS data
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The offset being sought changes which wave that the status data
+ * will be returned for.  The bits are used as follows:
+ *
+ * Bits 0..6:	Byte offset into data
+ * Bits 7..14:	SE selector
+ * Bits 15..22:	SH/SA selector
+ * Bits 23..30: CU/{WGP+SIMD} selector
+ * Bits 31..36: WAVE ID selector
+ * Bits 37..44: SIMD ID selector
+ *
+ * The returned data begins with one DWORD of version information
+ * Followed by WAVE STATUS registers relevant to the GFX IP version
+ * being used.  See gfx_v8_0_read_wave_data() for an example output.
+ */
+static ssize_t amdgpu_debugfs_wave_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = f->f_inode->i_private;
+	int r, x;
+	ssize_t result = 0;
+	uint32_t offset, se, sh, cu, wave, simd, data[32];
+
+	if (size & 3 || *pos & 3)
+		return -EINVAL;
+
+	/* decode offset */
+	offset = (*pos & GENMASK_ULL(6, 0));
+	se = (*pos & GENMASK_ULL(14, 7)) >> 7;
+	sh = (*pos & GENMASK_ULL(22, 15)) >> 15;
+	cu = (*pos & GENMASK_ULL(30, 23)) >> 23;
+	wave = (*pos & GENMASK_ULL(36, 31)) >> 31;
+	simd = (*pos & GENMASK_ULL(44, 37)) >> 37;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	/* switch to the specific se/sh/cu */
+	mutex_lock(&adev->grbm_idx_mutex);
+	amdgpu_gfx_select_se_sh(adev, se, sh, cu, 0);
+
+	x = 0;
+	if (adev->gfx.funcs->read_wave_data)
+		adev->gfx.funcs->read_wave_data(adev, 0, simd, wave, data, &x);
+
+	amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	if (!x) {
+		amdgpu_virt_disable_access_debugfs(adev);
+		return -EINVAL;
+	}
+
+	while (size && (offset < x * 4)) {
+		uint32_t value;
+
+		value = data[offset >> 2];
+		r = put_user(value, (uint32_t *)buf);
+		if (r) {
+			amdgpu_virt_disable_access_debugfs(adev);
+			return r;
+		}
+
+		result += 4;
+		buf += 4;
+		offset += 4;
+		size -= 4;
+	}
+
+	amdgpu_virt_disable_access_debugfs(adev);
+	return result;
+}
+
+/** amdgpu_debugfs_gpr_read - Read wave gprs
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * The offset being sought changes which wave that the status data
+ * will be returned for.  The bits are used as follows:
+ *
+ * Bits 0..11:	Byte offset into data
+ * Bits 12..19:	SE selector
+ * Bits 20..27:	SH/SA selector
+ * Bits 28..35: CU/{WGP+SIMD} selector
+ * Bits 36..43: WAVE ID selector
+ * Bits 37..44: SIMD ID selector
+ * Bits 52..59: Thread selector
+ * Bits 60..61: Bank selector (VGPR=0,SGPR=1)
+ *
+ * The return data comes from the SGPR or VGPR register bank for
+ * the selected operational unit.
+ */
+static ssize_t amdgpu_debugfs_gpr_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = f->f_inode->i_private;
+	int r;
+	ssize_t result = 0;
+	uint32_t offset, se, sh, cu, wave, simd, thread, bank, *data;
+
+	if (size > 4096 || size & 3 || *pos & 3)
+		return -EINVAL;
+
+	/* decode offset */
+	offset = (*pos & GENMASK_ULL(11, 0)) >> 2;
+	se = (*pos & GENMASK_ULL(19, 12)) >> 12;
+	sh = (*pos & GENMASK_ULL(27, 20)) >> 20;
+	cu = (*pos & GENMASK_ULL(35, 28)) >> 28;
+	wave = (*pos & GENMASK_ULL(43, 36)) >> 36;
+	simd = (*pos & GENMASK_ULL(51, 44)) >> 44;
+	thread = (*pos & GENMASK_ULL(59, 52)) >> 52;
+	bank = (*pos & GENMASK_ULL(61, 60)) >> 60;
+
+	data = kcalloc(1024, sizeof(*data), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0)
+		goto err;
+
+	r = amdgpu_virt_enable_access_debugfs(adev);
+	if (r < 0)
+		goto err;
+
+	/* switch to the specific se/sh/cu */
+	mutex_lock(&adev->grbm_idx_mutex);
+	amdgpu_gfx_select_se_sh(adev, se, sh, cu, 0);
+
+	if (bank == 0) {
+		if (adev->gfx.funcs->read_wave_vgprs)
+			adev->gfx.funcs->read_wave_vgprs(adev, 0, simd, wave, thread, offset, size>>2, data);
+	} else {
+		if (adev->gfx.funcs->read_wave_sgprs)
+			adev->gfx.funcs->read_wave_sgprs(adev, 0, simd, wave, offset, size>>2, data);
+	}
+
+	amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0);
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	while (size) {
+		uint32_t value;
+
+		value = data[result >> 2];
+		r = put_user(value, (uint32_t *)buf);
+		if (r) {
+			amdgpu_virt_disable_access_debugfs(adev);
+			goto err;
+		}
+
+		result += 4;
+		buf += 4;
+		size -= 4;
+	}
+
+	kfree(data);
+	amdgpu_virt_disable_access_debugfs(adev);
+	return result;
+
+err:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+	kfree(data);
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_gfxoff_residency_read - Read GFXOFF residency
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * Read the last residency value logged. It doesn't auto update, one needs to
+ * stop logging before getting the current value.
+ */
+static ssize_t amdgpu_debugfs_gfxoff_residency_read(struct file *f, char __user *buf,
+						    size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		r = amdgpu_get_gfx_off_residency(adev, &value);
+		if (r)
+			goto out;
+
+		r = put_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_gfxoff_residency_write - Log GFXOFF Residency
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * Write a 32-bit non-zero to start logging; write a 32-bit zero to stop
+ */
+static ssize_t amdgpu_debugfs_gfxoff_residency_write(struct file *f, const char __user *buf,
+						     size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		u32 value;
+
+		r = get_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		amdgpu_set_gfx_off_residency(adev, value ? true : false);
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+
+/**
+ * amdgpu_debugfs_gfxoff_count_read - Read GFXOFF entry count
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ */
+static ssize_t amdgpu_debugfs_gfxoff_count_read(struct file *f, char __user *buf,
+						size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		u64 value = 0;
+
+		r = amdgpu_get_gfx_off_entrycount(adev, &value);
+		if (r)
+			goto out;
+
+		r = put_user(value, (u64 *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_gfxoff_write - Enable/disable GFXOFF
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * Write a 32-bit zero to disable or a 32-bit non-zero to enable
+ */
+static ssize_t amdgpu_debugfs_gfxoff_write(struct file *f, const char __user *buf,
+					 size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		r = get_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		amdgpu_gfx_off_ctrl(adev, value ? true : false);
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+
+/**
+ * amdgpu_debugfs_gfxoff_read - read gfxoff status
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ */
+static ssize_t amdgpu_debugfs_gfxoff_read(struct file *f, char __user *buf,
+					 size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		u32 value = adev->gfx.gfx_off_state;
+
+		r = put_user(value, (u32 *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+static ssize_t amdgpu_debugfs_gfxoff_status_read(struct file *f, char __user *buf,
+						 size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		u32 value;
+
+		r = amdgpu_get_gfx_off_status(adev, &value);
+		if (r)
+			goto out;
+
+		r = put_user(value, (u32 *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+static const struct file_operations amdgpu_debugfs_regs2_fops = {
+	.owner = THIS_MODULE,
+	.unlocked_ioctl = amdgpu_debugfs_regs2_ioctl,
+	.read = amdgpu_debugfs_regs2_read,
+	.write = amdgpu_debugfs_regs2_write,
+	.open = amdgpu_debugfs_regs2_open,
+	.release = amdgpu_debugfs_regs2_release,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_gprwave_fops = {
+	.owner = THIS_MODULE,
+	.unlocked_ioctl = amdgpu_debugfs_gprwave_ioctl,
+	.read = amdgpu_debugfs_gprwave_read,
+	.open = amdgpu_debugfs_gprwave_open,
+	.release = amdgpu_debugfs_gprwave_release,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_regs_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_regs_read,
+	.write = amdgpu_debugfs_regs_write,
+	.llseek = default_llseek
+};
+static const struct file_operations amdgpu_debugfs_regs_didt_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_regs_didt_read,
+	.write = amdgpu_debugfs_regs_didt_write,
+	.llseek = default_llseek
+};
+static const struct file_operations amdgpu_debugfs_regs_pcie_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_regs_pcie_read,
+	.write = amdgpu_debugfs_regs_pcie_write,
+	.llseek = default_llseek
+};
+static const struct file_operations amdgpu_debugfs_regs_smc_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_regs_smc_read,
+	.write = amdgpu_debugfs_regs_smc_write,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_gca_config_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gca_config_read,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_sensors_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_sensor_read,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_wave_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_wave_read,
+	.llseek = default_llseek
+};
+static const struct file_operations amdgpu_debugfs_gpr_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gpr_read,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_gfxoff_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gfxoff_read,
+	.write = amdgpu_debugfs_gfxoff_write,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_gfxoff_status_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gfxoff_status_read,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_gfxoff_count_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gfxoff_count_read,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_gfxoff_residency_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gfxoff_residency_read,
+	.write = amdgpu_debugfs_gfxoff_residency_write,
+	.llseek = default_llseek
+};
+
+static const struct file_operations *debugfs_regs[] = {
+	&amdgpu_debugfs_regs_fops,
+	&amdgpu_debugfs_regs2_fops,
+	&amdgpu_debugfs_gprwave_fops,
+	&amdgpu_debugfs_regs_didt_fops,
+	&amdgpu_debugfs_regs_pcie_fops,
+	&amdgpu_debugfs_regs_smc_fops,
+	&amdgpu_debugfs_gca_config_fops,
+	&amdgpu_debugfs_sensors_fops,
+	&amdgpu_debugfs_wave_fops,
+	&amdgpu_debugfs_gpr_fops,
+	&amdgpu_debugfs_gfxoff_fops,
+	&amdgpu_debugfs_gfxoff_status_fops,
+	&amdgpu_debugfs_gfxoff_count_fops,
+	&amdgpu_debugfs_gfxoff_residency_fops,
+};
+
+static const char * const debugfs_regs_names[] = {
+	"amdgpu_regs",
+	"amdgpu_regs2",
+	"amdgpu_gprwave",
+	"amdgpu_regs_didt",
+	"amdgpu_regs_pcie",
+	"amdgpu_regs_smc",
+	"amdgpu_gca_config",
+	"amdgpu_sensors",
+	"amdgpu_wave",
+	"amdgpu_gpr",
+	"amdgpu_gfxoff",
+	"amdgpu_gfxoff_status",
+	"amdgpu_gfxoff_count",
+	"amdgpu_gfxoff_residency",
+};
+
+/**
+ * amdgpu_debugfs_regs_init -	Initialize debugfs entries that provide
+ *				register access.
+ *
+ * @adev: The device to attach the debugfs entries to
+ */
+int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
+{
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *ent, *root = minor->debugfs_root;
+	unsigned int i;
+
+	for (i = 0; i < ARRAY_SIZE(debugfs_regs); i++) {
+		ent = debugfs_create_file(debugfs_regs_names[i],
+					  S_IFREG | 0400, root,
+					  adev, debugfs_regs[i]);
+		if (!i && !IS_ERR_OR_NULL(ent))
+			i_size_write(ent->d_inode, adev->rmmio_size);
+	}
+
+	return 0;
+}
+
+static int amdgpu_debugfs_test_ib_show(struct seq_file *m, void *unused)
+{
+	struct amdgpu_device *adev = m->private;
+	struct drm_device *dev = adev_to_drm(adev);
+	int r = 0, i;
+
+	r = pm_runtime_get_sync(dev->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(dev->dev);
+		return r;
+	}
+
+	/* Avoid accidently unparking the sched thread during GPU reset */
+	r = down_write_killable(&adev->reset_domain->sem);
+	if (r)
+		return r;
+
+	/* hold on the scheduler */
+	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
+		struct amdgpu_ring *ring = adev->rings[i];
+
+		if (!amdgpu_ring_sched_ready(ring))
+			continue;
+		drm_sched_wqueue_stop(&ring->sched);
+	}
+
+	seq_puts(m, "run ib test:\n");
+	r = amdgpu_ib_ring_tests(adev);
+	if (r)
+		seq_printf(m, "ib ring tests failed (%d).\n", r);
+	else
+		seq_puts(m, "ib ring tests passed.\n");
+
+	/* go on the scheduler */
+	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
+		struct amdgpu_ring *ring = adev->rings[i];
+
+		if (!amdgpu_ring_sched_ready(ring))
+			continue;
+		drm_sched_wqueue_start(&ring->sched);
+	}
+
+	up_write(&adev->reset_domain->sem);
+
+	pm_runtime_put_autosuspend(dev->dev);
+
+	return 0;
+}
+
+static int amdgpu_debugfs_evict_vram(void *data, u64 *val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	struct drm_device *dev = adev_to_drm(adev);
+	int r;
+
+	r = pm_runtime_get_sync(dev->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(dev->dev);
+		return r;
+	}
+
+	*val = amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM);
+
+	pm_runtime_put_autosuspend(dev->dev);
+
+	return 0;
+}
+
+
+static int amdgpu_debugfs_evict_gtt(void *data, u64 *val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	struct drm_device *dev = adev_to_drm(adev);
+	int r;
+
+	r = pm_runtime_get_sync(dev->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(dev->dev);
+		return r;
+	}
+
+	*val = amdgpu_ttm_evict_resources(adev, TTM_PL_TT);
+
+	pm_runtime_put_autosuspend(dev->dev);
+
+	return 0;
+}
+
+static int amdgpu_debugfs_benchmark(void *data, u64 val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	struct drm_device *dev = adev_to_drm(adev);
+	int r;
+
+	r = pm_runtime_get_sync(dev->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(dev->dev);
+		return r;
+	}
+
+	r = amdgpu_benchmark(adev, val);
+
+	pm_runtime_put_autosuspend(dev->dev);
+
+	return r;
+}
+
+static int amdgpu_debugfs_vm_info_show(struct seq_file *m, void *unused)
+{
+	struct amdgpu_device *adev = m->private;
+	struct drm_device *dev = adev_to_drm(adev);
+	struct drm_file *file;
+	int r;
+
+	r = mutex_lock_interruptible(&dev->filelist_mutex);
+	if (r)
+		return r;
+
+	list_for_each_entry(file, &dev->filelist, lhead) {
+		struct amdgpu_fpriv *fpriv = file->driver_priv;
+		struct amdgpu_vm *vm = &fpriv->vm;
+		struct amdgpu_task_info *ti;
+
+		ti = amdgpu_vm_get_task_info_vm(vm);
+		if (ti) {
+			seq_printf(m, "pid:%d\tProcess:%s ----------\n", ti->task.pid, ti->process_name);
+			amdgpu_vm_put_task_info(ti);
+		}
+
+		r = amdgpu_bo_reserve(vm->root.bo, true);
+		if (r)
+			break;
+		amdgpu_debugfs_vm_bo_info(vm, m);
+		amdgpu_bo_unreserve(vm->root.bo);
+	}
+
+	mutex_unlock(&dev->filelist_mutex);
+
+	return r;
+}
+
+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_test_ib);
+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_vm_info);
+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_evict_vram_fops, amdgpu_debugfs_evict_vram,
+			 NULL, "%lld\n");
+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_evict_gtt_fops, amdgpu_debugfs_evict_gtt,
+			 NULL, "%lld\n");
+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_benchmark_fops, NULL, amdgpu_debugfs_benchmark,
+			 "%lld\n");
+
+static void amdgpu_ib_preempt_fences_swap(struct amdgpu_ring *ring,
+					  struct dma_fence **fences)
+{
+	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	uint32_t sync_seq, last_seq;
+
+	last_seq = atomic_read(&ring->fence_drv.last_seq);
+	sync_seq = ring->fence_drv.sync_seq;
+
+	last_seq &= drv->num_fences_mask;
+	sync_seq &= drv->num_fences_mask;
+
+	do {
+		struct dma_fence *fence, **ptr;
+
+		++last_seq;
+		last_seq &= drv->num_fences_mask;
+		ptr = &drv->fences[last_seq];
+
+		fence = rcu_dereference_protected(*ptr, 1);
+		RCU_INIT_POINTER(*ptr, NULL);
+
+		if (!fence)
+			continue;
+
+		fences[last_seq] = fence;
+
+	} while (last_seq != sync_seq);
+}
+
+static void amdgpu_ib_preempt_signal_fences(struct dma_fence **fences,
+					    int length)
+{
+	int i;
+	struct dma_fence *fence;
+
+	for (i = 0; i < length; i++) {
+		fence = fences[i];
+		if (!fence)
+			continue;
+		dma_fence_signal(fence);
+		dma_fence_put(fence);
+	}
+}
+
+static void amdgpu_ib_preempt_job_recovery(struct drm_gpu_scheduler *sched)
+{
+	struct drm_sched_job *s_job;
+	struct dma_fence *fence;
+
+	spin_lock(&sched->job_list_lock);
+	list_for_each_entry(s_job, &sched->pending_list, list) {
+		fence = sched->ops->run_job(s_job);
+		dma_fence_put(fence);
+	}
+	spin_unlock(&sched->job_list_lock);
+}
+
+static void amdgpu_ib_preempt_mark_partial_job(struct amdgpu_ring *ring)
+{
+	struct amdgpu_job *job;
+	struct drm_sched_job *s_job, *tmp;
+	uint32_t preempt_seq;
+	struct dma_fence *fence, **ptr;
+	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	struct drm_gpu_scheduler *sched = &ring->sched;
+	bool preempted = true;
+
+	if (ring->funcs->type != AMDGPU_RING_TYPE_GFX)
+		return;
+
+	preempt_seq = le32_to_cpu(*(drv->cpu_addr + 2));
+	if (preempt_seq <= atomic_read(&drv->last_seq)) {
+		preempted = false;
+		goto no_preempt;
+	}
+
+	preempt_seq &= drv->num_fences_mask;
+	ptr = &drv->fences[preempt_seq];
+	fence = rcu_dereference_protected(*ptr, 1);
+
+no_preempt:
+	spin_lock(&sched->job_list_lock);
+	list_for_each_entry_safe(s_job, tmp, &sched->pending_list, list) {
+		if (dma_fence_is_signaled(&s_job->s_fence->finished)) {
+			/* remove job from ring_mirror_list */
+			list_del_init(&s_job->list);
+			sched->ops->free_job(s_job);
+			continue;
+		}
+		job = to_amdgpu_job(s_job);
+		if (preempted && (&job->hw_fence->base) == fence)
+			/* mark the job as preempted */
+			job->preemption_status |= AMDGPU_IB_PREEMPTED;
+	}
+	spin_unlock(&sched->job_list_lock);
+}
+
+static int amdgpu_debugfs_ib_preempt(void *data, u64 val)
+{
+	int r, length;
+	struct amdgpu_ring *ring;
+	struct dma_fence **fences = NULL;
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+
+	if (val >= AMDGPU_MAX_RINGS)
+		return -EINVAL;
+
+	ring = adev->rings[val];
+
+	if (!amdgpu_ring_sched_ready(ring) ||
+	    !ring->funcs->preempt_ib)
+		return -EINVAL;
+
+	/* the last preemption failed */
+	if (ring->trail_seq != le32_to_cpu(*ring->trail_fence_cpu_addr))
+		return -EBUSY;
+
+	length = ring->fence_drv.num_fences_mask + 1;
+	fences = kcalloc(length, sizeof(void *), GFP_KERNEL);
+	if (!fences)
+		return -ENOMEM;
+
+	/* Avoid accidently unparking the sched thread during GPU reset */
+	r = down_read_killable(&adev->reset_domain->sem);
+	if (r)
+		goto pro_end;
+
+	/* stop the scheduler */
+	drm_sched_wqueue_stop(&ring->sched);
+
+	/* preempt the IB */
+	r = amdgpu_ring_preempt_ib(ring);
+	if (r) {
+		DRM_WARN("failed to preempt ring %d\n", ring->idx);
+		goto failure;
+	}
+
+	amdgpu_fence_process(ring);
+
+	if (atomic_read(&ring->fence_drv.last_seq) !=
+	    ring->fence_drv.sync_seq) {
+		DRM_INFO("ring %d was preempted\n", ring->idx);
+
+		amdgpu_ib_preempt_mark_partial_job(ring);
+
+		/* swap out the old fences */
+		amdgpu_ib_preempt_fences_swap(ring, fences);
+
+		amdgpu_fence_driver_force_completion(ring);
+
+		/* resubmit unfinished jobs */
+		amdgpu_ib_preempt_job_recovery(&ring->sched);
+
+		/* wait for jobs finished */
+		amdgpu_fence_wait_empty(ring);
+
+		/* signal the old fences */
+		amdgpu_ib_preempt_signal_fences(fences, length);
+	}
+
+failure:
+	/* restart the scheduler */
+	drm_sched_wqueue_start(&ring->sched);
+
+	up_read(&adev->reset_domain->sem);
+
+pro_end:
+	kfree(fences);
+
+	return r;
+}
+
+static int amdgpu_debugfs_sclk_set(void *data, u64 val)
+{
+	int ret = 0;
+	uint32_t max_freq, min_freq;
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+
+	if (amdgpu_sriov_multi_vf_mode(adev))
+		return -EINVAL;
+
+	ret = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (ret < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return ret;
+	}
+
+	ret = amdgpu_dpm_get_dpm_freq_range(adev, PP_SCLK, &min_freq, &max_freq);
+	if (ret == -EOPNOTSUPP) {
+		ret = 0;
+		goto out;
+	}
+	if (ret || val > max_freq || val < min_freq) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = amdgpu_dpm_set_soft_freq_range(adev, PP_SCLK, (uint32_t)val, (uint32_t)val);
+	if (ret)
+		ret = -EINVAL;
+
+out:
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return ret;
+}
+
+DEFINE_DEBUGFS_ATTRIBUTE(fops_ib_preempt, NULL,
+			amdgpu_debugfs_ib_preempt, "%llu\n");
+
+DEFINE_DEBUGFS_ATTRIBUTE(fops_sclk_set, NULL,
+			amdgpu_debugfs_sclk_set, "%llu\n");
+
+int amdgpu_debugfs_init(struct amdgpu_device *adev)
+{
+	struct dentry *root = adev_to_drm(adev)->primary->debugfs_root;
+	struct dentry *ent;
+	int r, i;
+
+	if (!debugfs_initialized())
+		return 0;
+
+	debugfs_create_x32("amdgpu_smu_debug", 0600, root,
+			   &adev->pm.smu_debug_mask);
+
+	ent = debugfs_create_file("amdgpu_preempt_ib", 0600, root, adev,
+				  &fops_ib_preempt);
+	if (IS_ERR(ent)) {
+		DRM_ERROR("unable to create amdgpu_preempt_ib debugsfs file\n");
+		return PTR_ERR(ent);
+	}
+
+	ent = debugfs_create_file("amdgpu_force_sclk", 0200, root, adev,
+				  &fops_sclk_set);
+	if (IS_ERR(ent)) {
+		DRM_ERROR("unable to create amdgpu_set_sclk debugsfs file\n");
+		return PTR_ERR(ent);
+	}
+
+	/* Register debugfs entries for amdgpu_ttm */
+	amdgpu_ttm_debugfs_init(adev);
+	amdgpu_debugfs_pm_init(adev);
+	amdgpu_debugfs_sa_init(adev);
+	amdgpu_debugfs_fence_init(adev);
+	amdgpu_debugfs_gem_init(adev);
+
+	r = amdgpu_debugfs_regs_init(adev);
+	if (r)
+		DRM_ERROR("registering register debugfs failed (%d).\n", r);
+
+	amdgpu_debugfs_firmware_init(adev);
+	amdgpu_ta_if_debugfs_init(adev);
+
+	amdgpu_debugfs_mes_event_log_init(adev);
+
+#if defined(CONFIG_DRM_AMD_DC)
+	if (adev->dc_enabled)
+		dtn_debugfs_init(adev);
+#endif
+
+	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
+		struct amdgpu_ring *ring = adev->rings[i];
+
+		if (!ring)
+			continue;
+
+		amdgpu_debugfs_ring_init(adev, ring);
+	}
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		if (!amdgpu_vcnfw_log)
+			break;
+
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+
+		amdgpu_debugfs_vcn_fwlog_init(adev, i, &adev->vcn.inst[i]);
+	}
+
+	if (amdgpu_umsch_mm & amdgpu_umsch_mm_fwlog)
+		amdgpu_debugfs_umsch_fwlog_init(adev, &adev->umsch_mm);
+
+	amdgpu_debugfs_vcn_sched_mask_init(adev);
+	amdgpu_debugfs_jpeg_sched_mask_init(adev);
+	amdgpu_debugfs_gfx_sched_mask_init(adev);
+	amdgpu_debugfs_compute_sched_mask_init(adev);
+	amdgpu_debugfs_sdma_sched_mask_init(adev);
+
+	amdgpu_ras_debugfs_create_all(adev);
+	amdgpu_rap_debugfs_init(adev);
+	amdgpu_securedisplay_debugfs_init(adev);
+	amdgpu_fw_attestation_debugfs_init(adev);
+	amdgpu_psp_debugfs_init(adev);
+
+	debugfs_create_file("amdgpu_evict_vram", 0400, root, adev,
+			    &amdgpu_evict_vram_fops);
+	debugfs_create_file("amdgpu_evict_gtt", 0400, root, adev,
+			    &amdgpu_evict_gtt_fops);
+	debugfs_create_file("amdgpu_test_ib", 0400, root, adev,
+			    &amdgpu_debugfs_test_ib_fops);
+	debugfs_create_file("amdgpu_vm_info", 0444, root, adev,
+			    &amdgpu_debugfs_vm_info_fops);
+	debugfs_create_file("amdgpu_benchmark", 0200, root, adev,
+			    &amdgpu_benchmark_fops);
+
+	adev->debugfs_vbios_blob.data = adev->bios;
+	adev->debugfs_vbios_blob.size = adev->bios_size;
+	debugfs_create_blob("amdgpu_vbios", 0444, root,
+			    &adev->debugfs_vbios_blob);
+
+	if (adev->discovery.debugfs_blob.size)
+		debugfs_create_blob("amdgpu_discovery", 0444, root,
+				    &adev->discovery.debugfs_blob);
+
+	return 0;
+}
+
+static int amdgpu_pt_info_read(struct seq_file *m, void *unused)
+{
+	struct drm_file *file;
+	struct amdgpu_fpriv *fpriv;
+	struct amdgpu_bo *root_bo;
+	struct amdgpu_device *adev;
+	int r;
+
+	file = m->private;
+	if (!file)
+		return -EINVAL;
+
+	adev = drm_to_adev(file->minor->dev);
+	fpriv = file->driver_priv;
+	if (!fpriv || !fpriv->vm.root.bo)
+		return -ENODEV;
+
+	root_bo = amdgpu_bo_ref(fpriv->vm.root.bo);
+	r = amdgpu_bo_reserve(root_bo, true);
+	if (r) {
+		amdgpu_bo_unref(&root_bo);
+		return -EINVAL;
+	}
+
+	seq_printf(m, "pd_address: 0x%llx\n", amdgpu_gmc_pd_addr(fpriv->vm.root.bo));
+	seq_printf(m, "max_pfn: 0x%llx\n", adev->vm_manager.max_pfn);
+	seq_printf(m, "num_level: 0x%x\n", adev->vm_manager.num_level);
+	seq_printf(m, "block_size: 0x%x\n", adev->vm_manager.block_size);
+	seq_printf(m, "fragment_size: 0x%x\n", adev->vm_manager.fragment_size);
+
+	amdgpu_bo_unreserve(root_bo);
+	amdgpu_bo_unref(&root_bo);
+
+	return 0;
+}
+
+static int amdgpu_pt_info_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, amdgpu_pt_info_read, inode->i_private);
+}
+
+static const struct file_operations amdgpu_pt_info_fops = {
+	.owner = THIS_MODULE,
+	.open = amdgpu_pt_info_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = single_release,
+};
+
+void amdgpu_debugfs_vm_init(struct drm_file *file)
+{
+	debugfs_create_file("vm_pagetable_info", 0444, file->debugfs_client, file,
+			    &amdgpu_pt_info_fops);
+}
+
+#else
+int amdgpu_debugfs_init(struct amdgpu_device *adev)
+{
+	return 0;
+}
+int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
+{
+	return 0;
+}
+void amdgpu_debugfs_vm_init(struct drm_file *file)
+{
+}
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
new file mode 100644
index 000000000000..e7b3c38e5186
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
@@ -0,0 +1,37 @@
+/*
+ * Copyright 2008 Advanced Micro Devices, Inc.
+ * Copyright 2008 Red Hat Inc.
+ * Copyright 2009 Jerome Glisse.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+/*
+ * Debugfs
+ */
+
+int amdgpu_debugfs_regs_init(struct amdgpu_device *adev);
+int amdgpu_debugfs_init(struct amdgpu_device *adev);
+void amdgpu_debugfs_fini(struct amdgpu_device *adev);
+void amdgpu_debugfs_fence_init(struct amdgpu_device *adev);
+void amdgpu_debugfs_firmware_init(struct amdgpu_device *adev);
+void amdgpu_debugfs_gem_init(struct amdgpu_device *adev);
+void amdgpu_debugfs_mes_event_log_init(struct amdgpu_device *adev);
+void amdgpu_debugfs_vm_init(struct drm_file *file);
+
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
new file mode 100644
index 000000000000..4e2fe6674db8
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.c
@@ -0,0 +1,371 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2024 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <generated/utsrelease.h>
+#include <linux/devcoredump.h>
+#include "amdgpu_dev_coredump.h"
+#include "atom.h"
+
+#ifndef CONFIG_DEV_COREDUMP
+void amdgpu_coredump(struct amdgpu_device *adev, bool skip_vram_check,
+		     bool vram_lost, struct amdgpu_job *job)
+{
+}
+#else
+
+const char *hw_ip_names[MAX_HWIP] = {
+	[GC_HWIP]		= "GC",
+	[HDP_HWIP]		= "HDP",
+	[SDMA0_HWIP]		= "SDMA0",
+	[SDMA1_HWIP]		= "SDMA1",
+	[SDMA2_HWIP]		= "SDMA2",
+	[SDMA3_HWIP]		= "SDMA3",
+	[SDMA4_HWIP]		= "SDMA4",
+	[SDMA5_HWIP]		= "SDMA5",
+	[SDMA6_HWIP]		= "SDMA6",
+	[SDMA7_HWIP]		= "SDMA7",
+	[LSDMA_HWIP]		= "LSDMA",
+	[MMHUB_HWIP]		= "MMHUB",
+	[ATHUB_HWIP]		= "ATHUB",
+	[NBIO_HWIP]		= "NBIO",
+	[MP0_HWIP]		= "MP0",
+	[MP1_HWIP]		= "MP1",
+	[UVD_HWIP]		= "UVD/JPEG/VCN",
+	[VCN1_HWIP]		= "VCN1",
+	[VCE_HWIP]		= "VCE",
+	[VPE_HWIP]		= "VPE",
+	[DF_HWIP]		= "DF",
+	[DCE_HWIP]		= "DCE",
+	[OSSSYS_HWIP]		= "OSSSYS",
+	[SMUIO_HWIP]		= "SMUIO",
+	[PWR_HWIP]		= "PWR",
+	[NBIF_HWIP]		= "NBIF",
+	[THM_HWIP]		= "THM",
+	[CLK_HWIP]		= "CLK",
+	[UMC_HWIP]		= "UMC",
+	[RSMU_HWIP]		= "RSMU",
+	[XGMI_HWIP]		= "XGMI",
+	[DCI_HWIP]		= "DCI",
+	[PCIE_HWIP]		= "PCIE",
+};
+
+static void amdgpu_devcoredump_fw_info(struct amdgpu_device *adev,
+				       struct drm_printer *p)
+{
+	uint32_t version;
+	uint32_t feature;
+	uint8_t smu_program, smu_major, smu_minor, smu_debug;
+	struct atom_context *ctx = adev->mode_info.atom_context;
+
+	drm_printf(p, "VCE feature version: %u, fw version: 0x%08x\n",
+		   adev->vce.fb_version, adev->vce.fw_version);
+	drm_printf(p, "UVD feature version: %u, fw version: 0x%08x\n", 0,
+		   adev->uvd.fw_version);
+	drm_printf(p, "GMC feature version: %u, fw version: 0x%08x\n", 0,
+		   adev->gmc.fw_version);
+	drm_printf(p, "ME feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.me_feature_version, adev->gfx.me_fw_version);
+	drm_printf(p, "PFP feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.pfp_feature_version, adev->gfx.pfp_fw_version);
+	drm_printf(p, "CE feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.ce_feature_version, adev->gfx.ce_fw_version);
+	drm_printf(p, "RLC feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.rlc_feature_version, adev->gfx.rlc_fw_version);
+
+	drm_printf(p, "RLC SRLC feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.rlc_srlc_feature_version,
+		   adev->gfx.rlc_srlc_fw_version);
+	drm_printf(p, "RLC SRLG feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.rlc_srlg_feature_version,
+		   adev->gfx.rlc_srlg_fw_version);
+	drm_printf(p, "RLC SRLS feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.rlc_srls_feature_version,
+		   adev->gfx.rlc_srls_fw_version);
+	drm_printf(p, "RLCP feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.rlcp_ucode_feature_version,
+		   adev->gfx.rlcp_ucode_version);
+	drm_printf(p, "RLCV feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.rlcv_ucode_feature_version,
+		   adev->gfx.rlcv_ucode_version);
+	drm_printf(p, "MEC feature version: %u, fw version: 0x%08x\n",
+		   adev->gfx.mec_feature_version, adev->gfx.mec_fw_version);
+
+	if (adev->gfx.mec2_fw)
+		drm_printf(p, "MEC2 feature version: %u, fw version: 0x%08x\n",
+			   adev->gfx.mec2_feature_version,
+			   adev->gfx.mec2_fw_version);
+
+	drm_printf(p, "IMU feature version: %u, fw version: 0x%08x\n", 0,
+		   adev->gfx.imu_fw_version);
+	drm_printf(p, "PSP SOS feature version: %u, fw version: 0x%08x\n",
+		   adev->psp.sos.feature_version, adev->psp.sos.fw_version);
+	drm_printf(p, "PSP ASD feature version: %u, fw version: 0x%08x\n",
+		   adev->psp.asd_context.bin_desc.feature_version,
+		   adev->psp.asd_context.bin_desc.fw_version);
+
+	drm_printf(p, "TA XGMI feature version: 0x%08x, fw version: 0x%08x\n",
+		   adev->psp.xgmi_context.context.bin_desc.feature_version,
+		   adev->psp.xgmi_context.context.bin_desc.fw_version);
+	drm_printf(p, "TA RAS feature version: 0x%08x, fw version: 0x%08x\n",
+		   adev->psp.ras_context.context.bin_desc.feature_version,
+		   adev->psp.ras_context.context.bin_desc.fw_version);
+	drm_printf(p, "TA HDCP feature version: 0x%08x, fw version: 0x%08x\n",
+		   adev->psp.hdcp_context.context.bin_desc.feature_version,
+		   adev->psp.hdcp_context.context.bin_desc.fw_version);
+	drm_printf(p, "TA DTM feature version: 0x%08x, fw version: 0x%08x\n",
+		   adev->psp.dtm_context.context.bin_desc.feature_version,
+		   adev->psp.dtm_context.context.bin_desc.fw_version);
+	drm_printf(p, "TA RAP feature version: 0x%08x, fw version: 0x%08x\n",
+		   adev->psp.rap_context.context.bin_desc.feature_version,
+		   adev->psp.rap_context.context.bin_desc.fw_version);
+	drm_printf(p,
+		   "TA SECURE DISPLAY feature version: 0x%08x, fw version: 0x%08x\n",
+		   adev->psp.securedisplay_context.context.bin_desc.feature_version,
+		   adev->psp.securedisplay_context.context.bin_desc.fw_version);
+
+	/* SMC firmware */
+	version = adev->pm.fw_version;
+
+	smu_program = (version >> 24) & 0xff;
+	smu_major = (version >> 16) & 0xff;
+	smu_minor = (version >> 8) & 0xff;
+	smu_debug = (version >> 0) & 0xff;
+	drm_printf(p,
+		   "SMC feature version: %u, program: %d, fw version: 0x%08x (%d.%d.%d)\n",
+		   0, smu_program, version, smu_major, smu_minor, smu_debug);
+
+	/* SDMA firmware */
+	for (int i = 0; i < adev->sdma.num_instances; i++) {
+		drm_printf(p,
+			   "SDMA%d feature version: %u, firmware version: 0x%08x\n",
+			   i, adev->sdma.instance[i].feature_version,
+			   adev->sdma.instance[i].fw_version);
+	}
+
+	drm_printf(p, "VCN feature version: %u, fw version: 0x%08x\n", 0,
+		   adev->vcn.fw_version);
+	drm_printf(p, "DMCU feature version: %u, fw version: 0x%08x\n", 0,
+		   adev->dm.dmcu_fw_version);
+	drm_printf(p, "DMCUB feature version: %u, fw version: 0x%08x\n", 0,
+		   adev->dm.dmcub_fw_version);
+	drm_printf(p, "PSP TOC feature version: %u, fw version: 0x%08x\n",
+		   adev->psp.toc.feature_version, adev->psp.toc.fw_version);
+
+	version = adev->mes.kiq_version & AMDGPU_MES_VERSION_MASK;
+	feature = (adev->mes.kiq_version & AMDGPU_MES_FEAT_VERSION_MASK) >>
+		  AMDGPU_MES_FEAT_VERSION_SHIFT;
+	drm_printf(p, "MES_KIQ feature version: %u, fw version: 0x%08x\n",
+		   feature, version);
+
+	version = adev->mes.sched_version & AMDGPU_MES_VERSION_MASK;
+	feature = (adev->mes.sched_version & AMDGPU_MES_FEAT_VERSION_MASK) >>
+		  AMDGPU_MES_FEAT_VERSION_SHIFT;
+	drm_printf(p, "MES feature version: %u, fw version: 0x%08x\n", feature,
+		   version);
+
+	drm_printf(p, "VPE feature version: %u, fw version: 0x%08x\n",
+		   adev->vpe.feature_version, adev->vpe.fw_version);
+
+	drm_printf(p, "\nVBIOS Information\n");
+	drm_printf(p, "vbios name       : %s\n", ctx->name);
+	drm_printf(p, "vbios pn         : %s\n", ctx->vbios_pn);
+	drm_printf(p, "vbios version    : %d\n", ctx->version);
+	drm_printf(p, "vbios ver_str    : %s\n", ctx->vbios_ver_str);
+	drm_printf(p, "vbios date       : %s\n", ctx->date);
+}
+
+static ssize_t
+amdgpu_devcoredump_read(char *buffer, loff_t offset, size_t count,
+			void *data, size_t datalen)
+{
+	struct drm_printer p;
+	struct amdgpu_coredump_info *coredump = data;
+	struct drm_print_iterator iter;
+	struct amdgpu_vm_fault_info *fault_info;
+	struct amdgpu_ip_block *ip_block;
+	int ver;
+
+	iter.data = buffer;
+	iter.offset = 0;
+	iter.start = offset;
+	iter.remain = count;
+
+	p = drm_coredump_printer(&iter);
+
+	drm_printf(&p, "**** AMDGPU Device Coredump ****\n");
+	drm_printf(&p, "version: " AMDGPU_COREDUMP_VERSION "\n");
+	drm_printf(&p, "kernel: " UTS_RELEASE "\n");
+	drm_printf(&p, "module: " KBUILD_MODNAME "\n");
+	drm_printf(&p, "time: %ptSp\n", &coredump->reset_time);
+
+	if (coredump->reset_task_info.task.pid)
+		drm_printf(&p, "process_name: %s PID: %d\n",
+			   coredump->reset_task_info.process_name,
+			   coredump->reset_task_info.task.pid);
+
+	/* SOC Information */
+	drm_printf(&p, "\nSOC Information\n");
+	drm_printf(&p, "SOC Device id: %d\n", coredump->adev->pdev->device);
+	drm_printf(&p, "SOC PCI Revision id: %d\n", coredump->adev->pdev->revision);
+	drm_printf(&p, "SOC Family: %d\n", coredump->adev->family);
+	drm_printf(&p, "SOC Revision id: %d\n", coredump->adev->rev_id);
+	drm_printf(&p, "SOC External Revision id: %d\n", coredump->adev->external_rev_id);
+
+	/* Memory Information */
+	drm_printf(&p, "\nSOC Memory Information\n");
+	drm_printf(&p, "real vram size: %llu\n", coredump->adev->gmc.real_vram_size);
+	drm_printf(&p, "visible vram size: %llu\n", coredump->adev->gmc.visible_vram_size);
+	drm_printf(&p, "gtt size: %llu\n", coredump->adev->mman.gtt_mgr.manager.size);
+
+	/* GDS Config */
+	drm_printf(&p, "\nGDS Config\n");
+	drm_printf(&p, "gds: total size: %d\n", coredump->adev->gds.gds_size);
+	drm_printf(&p, "gds: compute partition size: %d\n", coredump->adev->gds.gds_size);
+	drm_printf(&p, "gds: gws per compute partition: %d\n", coredump->adev->gds.gws_size);
+	drm_printf(&p, "gds: os per compute partition: %d\n", coredump->adev->gds.oa_size);
+
+	/* HWIP Version Information */
+	drm_printf(&p, "\nHW IP Version Information\n");
+	for (int i = 1; i < MAX_HWIP; i++) {
+		for (int j = 0; j < HWIP_MAX_INSTANCE; j++) {
+			ver = coredump->adev->ip_versions[i][j];
+			if (ver)
+				drm_printf(&p, "HWIP: %s[%d][%d]: v%d.%d.%d.%d.%d\n",
+					   hw_ip_names[i], i, j,
+					   IP_VERSION_MAJ(ver),
+					   IP_VERSION_MIN(ver),
+					   IP_VERSION_REV(ver),
+					   IP_VERSION_VARIANT(ver),
+					   IP_VERSION_SUBREV(ver));
+		}
+	}
+
+	/* IP firmware information */
+	drm_printf(&p, "\nIP Firmwares\n");
+	amdgpu_devcoredump_fw_info(coredump->adev, &p);
+
+	if (coredump->ring) {
+		drm_printf(&p, "\nRing timed out details\n");
+		drm_printf(&p, "IP Type: %d Ring Name: %s\n",
+			   coredump->ring->funcs->type,
+			   coredump->ring->name);
+	}
+
+	/* Add page fault information */
+	fault_info = &coredump->adev->vm_manager.fault_info;
+	drm_printf(&p, "\n[%s] Page fault observed\n",
+		   fault_info->vmhub ? "mmhub" : "gfxhub");
+	drm_printf(&p, "Faulty page starting at address: 0x%016llx\n", fault_info->addr);
+	drm_printf(&p, "Protection fault status register: 0x%x\n\n", fault_info->status);
+
+	/* dump the ip state for each ip */
+	drm_printf(&p, "IP Dump\n");
+	for (int i = 0; i < coredump->adev->num_ip_blocks; i++) {
+		ip_block = &coredump->adev->ip_blocks[i];
+		if (ip_block->version->funcs->print_ip_state) {
+			drm_printf(&p, "IP: %s\n", ip_block->version->funcs->name);
+			ip_block->version->funcs->print_ip_state(ip_block, &p);
+			drm_printf(&p, "\n");
+		}
+	}
+
+	/* Add ring buffer information */
+	drm_printf(&p, "Ring buffer information\n");
+	for (int i = 0; i < coredump->adev->num_rings; i++) {
+		int j = 0;
+		struct amdgpu_ring *ring = coredump->adev->rings[i];
+
+		drm_printf(&p, "ring name: %s\n", ring->name);
+		drm_printf(&p, "Rptr: 0x%llx Wptr: 0x%llx RB mask: %x\n",
+			   amdgpu_ring_get_rptr(ring),
+			   amdgpu_ring_get_wptr(ring),
+			   ring->buf_mask);
+		drm_printf(&p, "Ring size in dwords: %d\n",
+			   ring->ring_size / 4);
+		drm_printf(&p, "Ring contents\n");
+		drm_printf(&p, "Offset \t Value\n");
+
+		while (j < ring->ring_size) {
+			drm_printf(&p, "0x%x \t 0x%x\n", j, ring->ring[j / 4]);
+			j += 4;
+		}
+	}
+
+	if (coredump->skip_vram_check)
+		drm_printf(&p, "VRAM lost check is skipped!\n");
+	else if (coredump->reset_vram_lost)
+		drm_printf(&p, "VRAM is lost due to GPU reset!\n");
+
+	return count - iter.remain;
+}
+
+static void amdgpu_devcoredump_free(void *data)
+{
+	kfree(data);
+}
+
+void amdgpu_coredump(struct amdgpu_device *adev, bool skip_vram_check,
+		     bool vram_lost, struct amdgpu_job *job)
+{
+	struct drm_device *dev = adev_to_drm(adev);
+	struct amdgpu_coredump_info *coredump;
+	struct drm_sched_job *s_job;
+
+	coredump = kzalloc(sizeof(*coredump), GFP_NOWAIT);
+
+	if (!coredump) {
+		DRM_ERROR("%s: failed to allocate memory for coredump\n", __func__);
+		return;
+	}
+
+	coredump->skip_vram_check = skip_vram_check;
+	coredump->reset_vram_lost = vram_lost;
+
+	if (job && job->pasid) {
+		struct amdgpu_task_info *ti;
+
+		ti = amdgpu_vm_get_task_info_pasid(adev, job->pasid);
+		if (ti) {
+			coredump->reset_task_info = *ti;
+			amdgpu_vm_put_task_info(ti);
+		}
+	}
+
+	if (job) {
+		s_job = &job->base;
+		coredump->ring = to_amdgpu_ring(s_job->sched);
+	}
+
+	coredump->adev = adev;
+
+	ktime_get_ts64(&coredump->reset_time);
+
+	dev_coredumpm(dev->dev, THIS_MODULE, coredump, 0, GFP_NOWAIT,
+		      amdgpu_devcoredump_read, amdgpu_devcoredump_free);
+
+	drm_info(dev, "AMDGPU device coredump file has been created\n");
+	drm_info(dev, "Check your /sys/class/drm/card%d/device/devcoredump/data\n",
+		 dev->primary->index);
+}
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h
new file mode 100644
index 000000000000..ef9772c6bcc9
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dev_coredump.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright 2024 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_DEV_COREDUMP_H__
+#define __AMDGPU_DEV_COREDUMP_H__
+
+#include "amdgpu.h"
+
+#ifdef CONFIG_DEV_COREDUMP
+
+#define AMDGPU_COREDUMP_VERSION "1"
+
+struct amdgpu_coredump_info {
+	struct amdgpu_device            *adev;
+	struct amdgpu_task_info         reset_task_info;
+	struct timespec64               reset_time;
+	bool                            skip_vram_check;
+	bool                            reset_vram_lost;
+	struct amdgpu_ring              *ring;
+};
+#endif
+
+void amdgpu_coredump(struct amdgpu_device *adev, bool skip_vram_check,
+		     bool vram_lost, struct amdgpu_job *job);
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 43ca16b6eee2..58c3ffe707d1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -25,13 +25,24 @@
  *          Alex Deucher
  *          Jerome Glisse
  */
+
+#include <linux/aperture.h>
+#include <linux/power_supply.h>
 #include <linux/kthread.h>
+#include <linux/module.h>
 #include <linux/console.h>
 #include <linux/slab.h>
-#include <linux/debugfs.h>
-#include <drm/drmP.h>
+#include <linux/iommu.h>
+#include <linux/pci.h>
+#include <linux/pci-p2pdma.h>
+#include <linux/apple-gmux.h>
+
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_client_event.h>
 #include <drm/drm_crtc_helper.h>
+#include <drm/drm_probe_helper.h>
 #include <drm/amdgpu_drm.h>
+#include <linux/device.h>
 #include <linux/vgaarb.h>
 #include <linux/vga_switcheroo.h>
 #include <linux/efi.h>
@@ -50,14 +61,56 @@
 #endif
 #include "vi.h"
 #include "soc15.h"
+#include "nv.h"
 #include "bif/bif_4_1_d.h"
-#include <linux/pci.h>
 #include <linux/firmware.h>
+#include "amdgpu_vf_error.h"
+
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_pm.h"
+
+#include "amdgpu_xgmi.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_ras_mgr.h"
+#include "amdgpu_pmu.h"
+#include "amdgpu_fru_eeprom.h"
+#include "amdgpu_reset.h"
+#include "amdgpu_virt.h"
+#include "amdgpu_dev_coredump.h"
+
+#include <linux/suspend.h>
+#include <drm/task_barrier.h>
+#include <linux/pm_runtime.h>
+
+#include <drm/drm_drv.h>
+
+#if IS_ENABLED(CONFIG_X86)
+#include <asm/intel-family.h>
+#include <asm/cpu_device_id.h>
+#endif
+
+MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
+MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin");
+MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin");
+MODULE_FIRMWARE("amdgpu/picasso_gpu_info.bin");
+MODULE_FIRMWARE("amdgpu/raven2_gpu_info.bin");
+MODULE_FIRMWARE("amdgpu/arcturus_gpu_info.bin");
+MODULE_FIRMWARE("amdgpu/navi12_gpu_info.bin");
+MODULE_FIRMWARE("amdgpu/cyan_skillfish_gpu_info.bin");
 
-static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev);
-static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev);
+#define AMDGPU_RESUME_MS		2000
+#define AMDGPU_MAX_RETRY_LIMIT		2
+#define AMDGPU_RETRY_SRIOV_RESET(r) ((r) == -EBUSY || (r) == -ETIMEDOUT || (r) == -EINVAL)
+#define AMDGPU_PCIE_INDEX_FALLBACK (0x38 >> 2)
+#define AMDGPU_PCIE_INDEX_HI_FALLBACK (0x44 >> 2)
+#define AMDGPU_PCIE_DATA_FALLBACK (0x3C >> 2)
 
-static const char *amdgpu_asic_name[] = {
+#define AMDGPU_VBIOS_SKIP (1U << 0)
+#define AMDGPU_VBIOS_OPTIONAL (1U << 1)
+
+static const struct drm_driver amdgpu_kms_driver;
+
+const char *amdgpu_asic_name[] = {
 	"TAHITI",
 	"PITCAIRN",
 	"VERDE",
@@ -76,169 +129,1157 @@ static const char *amdgpu_asic_name[] = {
 	"POLARIS10",
 	"POLARIS11",
 	"POLARIS12",
+	"VEGAM",
 	"VEGA10",
+	"VEGA12",
+	"VEGA20",
+	"RAVEN",
+	"ARCTURUS",
+	"RENOIR",
+	"ALDEBARAN",
+	"NAVI10",
+	"CYAN_SKILLFISH",
+	"NAVI14",
+	"NAVI12",
+	"SIENNA_CICHLID",
+	"NAVY_FLOUNDER",
+	"VANGOGH",
+	"DIMGREY_CAVEFISH",
+	"BEIGE_GOBY",
+	"YELLOW_CARP",
+	"IP DISCOVERY",
 	"LAST",
 };
 
-bool amdgpu_device_is_px(struct drm_device *dev)
+#define AMDGPU_IP_BLK_MASK_ALL GENMASK(AMD_IP_BLOCK_TYPE_NUM  - 1, 0)
+/*
+ * Default init level where all blocks are expected to be initialized. This is
+ * the level of initialization expected by default and also after a full reset
+ * of the device.
+ */
+struct amdgpu_init_level amdgpu_init_default = {
+	.level = AMDGPU_INIT_LEVEL_DEFAULT,
+	.hwini_ip_block_mask = AMDGPU_IP_BLK_MASK_ALL,
+};
+
+struct amdgpu_init_level amdgpu_init_recovery = {
+	.level = AMDGPU_INIT_LEVEL_RESET_RECOVERY,
+	.hwini_ip_block_mask = AMDGPU_IP_BLK_MASK_ALL,
+};
+
+/*
+ * Minimal blocks needed to be initialized before a XGMI hive can be reset. This
+ * is used for cases like reset on initialization where the entire hive needs to
+ * be reset before first use.
+ */
+struct amdgpu_init_level amdgpu_init_minimal_xgmi = {
+	.level = AMDGPU_INIT_LEVEL_MINIMAL_XGMI,
+	.hwini_ip_block_mask =
+		BIT(AMD_IP_BLOCK_TYPE_GMC) | BIT(AMD_IP_BLOCK_TYPE_SMC) |
+		BIT(AMD_IP_BLOCK_TYPE_COMMON) | BIT(AMD_IP_BLOCK_TYPE_IH) |
+		BIT(AMD_IP_BLOCK_TYPE_PSP)
+};
+
+static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev);
+static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev);
+static int amdgpu_device_ip_resume_phase3(struct amdgpu_device *adev);
+
+static void amdgpu_device_load_switch_state(struct amdgpu_device *adev);
+
+static inline bool amdgpu_ip_member_of_hwini(struct amdgpu_device *adev,
+					     enum amd_ip_block_type block)
+{
+	return (adev->init_lvl->hwini_ip_block_mask & (1U << block)) != 0;
+}
+
+void amdgpu_set_init_level(struct amdgpu_device *adev,
+			   enum amdgpu_init_lvl_id lvl)
+{
+	switch (lvl) {
+	case AMDGPU_INIT_LEVEL_MINIMAL_XGMI:
+		adev->init_lvl = &amdgpu_init_minimal_xgmi;
+		break;
+	case AMDGPU_INIT_LEVEL_RESET_RECOVERY:
+		adev->init_lvl = &amdgpu_init_recovery;
+		break;
+	case AMDGPU_INIT_LEVEL_DEFAULT:
+		fallthrough;
+	default:
+		adev->init_lvl = &amdgpu_init_default;
+		break;
+	}
+}
+
+static inline void amdgpu_device_stop_pending_resets(struct amdgpu_device *adev);
+static int amdgpu_device_pm_notifier(struct notifier_block *nb, unsigned long mode,
+				     void *data);
+
+/**
+ * DOC: pcie_replay_count
+ *
+ * The amdgpu driver provides a sysfs API for reporting the total number
+ * of PCIe replays (NAKs).
+ * The file pcie_replay_count is used for this and returns the total
+ * number of replays as a sum of the NAKs generated and NAKs received.
+ */
+
+static ssize_t amdgpu_device_get_pcie_replay_count(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	uint64_t cnt = amdgpu_asic_get_pcie_replay_count(adev);
+
+	return sysfs_emit(buf, "%llu\n", cnt);
+}
+
+static DEVICE_ATTR(pcie_replay_count, 0444,
+		amdgpu_device_get_pcie_replay_count, NULL);
+
+static int amdgpu_device_attr_sysfs_init(struct amdgpu_device *adev)
+{
+	int ret = 0;
+
+	if (amdgpu_nbio_is_replay_cnt_supported(adev))
+		ret = sysfs_create_file(&adev->dev->kobj,
+					&dev_attr_pcie_replay_count.attr);
+
+	return ret;
+}
+
+static void amdgpu_device_attr_sysfs_fini(struct amdgpu_device *adev)
+{
+	if (amdgpu_nbio_is_replay_cnt_supported(adev))
+		sysfs_remove_file(&adev->dev->kobj,
+				  &dev_attr_pcie_replay_count.attr);
+}
+
+static ssize_t amdgpu_sysfs_reg_state_get(struct file *f, struct kobject *kobj,
+					  const struct bin_attribute *attr, char *buf,
+					  loff_t ppos, size_t count)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	ssize_t bytes_read;
+
+	switch (ppos) {
+	case AMDGPU_SYS_REG_STATE_XGMI:
+		bytes_read = amdgpu_asic_get_reg_state(
+			adev, AMDGPU_REG_STATE_TYPE_XGMI, buf, count);
+		break;
+	case AMDGPU_SYS_REG_STATE_WAFL:
+		bytes_read = amdgpu_asic_get_reg_state(
+			adev, AMDGPU_REG_STATE_TYPE_WAFL, buf, count);
+		break;
+	case AMDGPU_SYS_REG_STATE_PCIE:
+		bytes_read = amdgpu_asic_get_reg_state(
+			adev, AMDGPU_REG_STATE_TYPE_PCIE, buf, count);
+		break;
+	case AMDGPU_SYS_REG_STATE_USR:
+		bytes_read = amdgpu_asic_get_reg_state(
+			adev, AMDGPU_REG_STATE_TYPE_USR, buf, count);
+		break;
+	case AMDGPU_SYS_REG_STATE_USR_1:
+		bytes_read = amdgpu_asic_get_reg_state(
+			adev, AMDGPU_REG_STATE_TYPE_USR_1, buf, count);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return bytes_read;
+}
+
+static const BIN_ATTR(reg_state, 0444, amdgpu_sysfs_reg_state_get, NULL,
+		      AMDGPU_SYS_REG_STATE_END);
+
+int amdgpu_reg_state_sysfs_init(struct amdgpu_device *adev)
+{
+	int ret;
+
+	if (!amdgpu_asic_get_reg_state_supported(adev))
+		return 0;
+
+	ret = sysfs_create_bin_file(&adev->dev->kobj, &bin_attr_reg_state);
+
+	return ret;
+}
+
+void amdgpu_reg_state_sysfs_fini(struct amdgpu_device *adev)
+{
+	if (!amdgpu_asic_get_reg_state_supported(adev))
+		return;
+	sysfs_remove_bin_file(&adev->dev->kobj, &bin_attr_reg_state);
+}
+
+int amdgpu_ip_block_suspend(struct amdgpu_ip_block *ip_block)
+{
+	int r;
+
+	if (ip_block->version->funcs->suspend) {
+		r = ip_block->version->funcs->suspend(ip_block);
+		if (r) {
+			dev_err(ip_block->adev->dev,
+				"suspend of IP block <%s> failed %d\n",
+				ip_block->version->funcs->name, r);
+			return r;
+		}
+	}
+
+	ip_block->status.hw = false;
+	return 0;
+}
+
+int amdgpu_ip_block_resume(struct amdgpu_ip_block *ip_block)
+{
+	int r;
+
+	if (ip_block->version->funcs->resume) {
+		r = ip_block->version->funcs->resume(ip_block);
+		if (r) {
+			dev_err(ip_block->adev->dev,
+				"resume of IP block <%s> failed %d\n",
+				ip_block->version->funcs->name, r);
+			return r;
+		}
+	}
+
+	ip_block->status.hw = true;
+	return 0;
+}
+
+/**
+ * DOC: board_info
+ *
+ * The amdgpu driver provides a sysfs API for giving board related information.
+ * It provides the form factor information in the format
+ *
+ *   type : form factor
+ *
+ * Possible form factor values
+ *
+ * - "cem"		- PCIE CEM card
+ * - "oam"		- Open Compute Accelerator Module
+ * - "unknown"	- Not known
+ *
+ */
+
+static ssize_t amdgpu_device_get_board_info(struct device *dev,
+					    struct device_attribute *attr,
+					    char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	enum amdgpu_pkg_type pkg_type = AMDGPU_PKG_TYPE_CEM;
+	const char *pkg;
+
+	if (adev->smuio.funcs && adev->smuio.funcs->get_pkg_type)
+		pkg_type = adev->smuio.funcs->get_pkg_type(adev);
+
+	switch (pkg_type) {
+	case AMDGPU_PKG_TYPE_CEM:
+		pkg = "cem";
+		break;
+	case AMDGPU_PKG_TYPE_OAM:
+		pkg = "oam";
+		break;
+	default:
+		pkg = "unknown";
+		break;
+	}
+
+	return sysfs_emit(buf, "%s : %s\n", "type", pkg);
+}
+
+static DEVICE_ATTR(board_info, 0444, amdgpu_device_get_board_info, NULL);
+
+static struct attribute *amdgpu_board_attrs[] = {
+	&dev_attr_board_info.attr,
+	NULL,
+};
+
+static umode_t amdgpu_board_attrs_is_visible(struct kobject *kobj,
+					     struct attribute *attr, int n)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	if (adev->flags & AMD_IS_APU)
+		return 0;
+
+	return attr->mode;
+}
+
+static const struct attribute_group amdgpu_board_attrs_group = {
+	.attrs = amdgpu_board_attrs,
+	.is_visible = amdgpu_board_attrs_is_visible
+};
+
+static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev);
+
+/**
+ * amdgpu_device_supports_px - Is the device a dGPU with ATPX power control
+ *
+ * @adev: amdgpu device pointer
+ *
+ * Returns true if the device is a dGPU with ATPX power control,
+ * otherwise return false.
+ */
+bool amdgpu_device_supports_px(struct amdgpu_device *adev)
+{
+	if ((adev->flags & AMD_IS_PX) && !amdgpu_is_atpx_hybrid())
+		return true;
+	return false;
+}
+
+/**
+ * amdgpu_device_supports_boco - Is the device a dGPU with ACPI power resources
+ *
+ * @adev: amdgpu device pointer
+ *
+ * Returns true if the device is a dGPU with ACPI power control,
+ * otherwise return false.
+ */
+bool amdgpu_device_supports_boco(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	if (!IS_ENABLED(CONFIG_HOTPLUG_PCI_PCIE))
+		return false;
 
-	if (adev->flags & AMD_IS_PX)
+	if (adev->has_pr3 ||
+	    ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid()))
 		return true;
 	return false;
 }
 
+/**
+ * amdgpu_device_supports_baco - Does the device support BACO
+ *
+ * @adev: amdgpu device pointer
+ *
+ * Return:
+ * 1 if the device supports BACO;
+ * 3 if the device supports MACO (only works if BACO is supported)
+ * otherwise return 0.
+ */
+int amdgpu_device_supports_baco(struct amdgpu_device *adev)
+{
+	return amdgpu_asic_supports_baco(adev);
+}
+
+void amdgpu_device_detect_runtime_pm_mode(struct amdgpu_device *adev)
+{
+	int bamaco_support;
+
+	adev->pm.rpm_mode = AMDGPU_RUNPM_NONE;
+	bamaco_support = amdgpu_device_supports_baco(adev);
+
+	switch (amdgpu_runtime_pm) {
+	case 2:
+		if (bamaco_support & MACO_SUPPORT) {
+			adev->pm.rpm_mode = AMDGPU_RUNPM_BAMACO;
+			dev_info(adev->dev, "Forcing BAMACO for runtime pm\n");
+		} else if (bamaco_support == BACO_SUPPORT) {
+			adev->pm.rpm_mode = AMDGPU_RUNPM_BACO;
+			dev_info(adev->dev, "Requested mode BAMACO not available,fallback to use BACO\n");
+		}
+		break;
+	case 1:
+		if (bamaco_support & BACO_SUPPORT) {
+			adev->pm.rpm_mode = AMDGPU_RUNPM_BACO;
+			dev_info(adev->dev, "Forcing BACO for runtime pm\n");
+		}
+		break;
+	case -1:
+	case -2:
+		if (amdgpu_device_supports_px(adev)) {
+			/* enable PX as runtime mode */
+			adev->pm.rpm_mode = AMDGPU_RUNPM_PX;
+			dev_info(adev->dev, "Using ATPX for runtime pm\n");
+		} else if (amdgpu_device_supports_boco(adev)) {
+			/* enable boco as runtime mode */
+			adev->pm.rpm_mode = AMDGPU_RUNPM_BOCO;
+			dev_info(adev->dev, "Using BOCO for runtime pm\n");
+		} else {
+			if (!bamaco_support)
+				goto no_runtime_pm;
+
+			switch (adev->asic_type) {
+			case CHIP_VEGA20:
+			case CHIP_ARCTURUS:
+				/* BACO are not supported on vega20 and arctrus */
+				break;
+			case CHIP_VEGA10:
+				/* enable BACO as runpm mode if noretry=0 */
+				if (!adev->gmc.noretry && !amdgpu_passthrough(adev))
+					adev->pm.rpm_mode = AMDGPU_RUNPM_BACO;
+				break;
+			default:
+				/* enable BACO as runpm mode on CI+ */
+				if (!amdgpu_passthrough(adev))
+					adev->pm.rpm_mode = AMDGPU_RUNPM_BACO;
+				break;
+			}
+
+			if (adev->pm.rpm_mode == AMDGPU_RUNPM_BACO) {
+				if (bamaco_support & MACO_SUPPORT) {
+					adev->pm.rpm_mode = AMDGPU_RUNPM_BAMACO;
+					dev_info(adev->dev, "Using BAMACO for runtime pm\n");
+				} else {
+					dev_info(adev->dev, "Using BACO for runtime pm\n");
+				}
+			}
+		}
+		break;
+	case 0:
+		dev_info(adev->dev, "runtime pm is manually disabled\n");
+		break;
+	default:
+		break;
+	}
+
+no_runtime_pm:
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_NONE)
+		dev_info(adev->dev, "Runtime PM not available\n");
+}
+/**
+ * amdgpu_device_supports_smart_shift - Is the device dGPU with
+ * smart shift support
+ *
+ * @adev: amdgpu device pointer
+ *
+ * Returns true if the device is a dGPU with Smart Shift support,
+ * otherwise returns false.
+ */
+bool amdgpu_device_supports_smart_shift(struct amdgpu_device *adev)
+{
+	return (amdgpu_device_supports_boco(adev) &&
+		amdgpu_acpi_is_power_shift_control_supported());
+}
+
 /*
- * MMIO register access helper functions.
+ * VRAM access helper functions
  */
-uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg,
-			uint32_t acc_flags)
+
+/**
+ * amdgpu_device_mm_access - access vram by MM_INDEX/MM_DATA
+ *
+ * @adev: amdgpu_device pointer
+ * @pos: offset of the buffer in vram
+ * @buf: virtual address of the buffer in system memory
+ * @size: read/write size, sizeof(@buf) must > @size
+ * @write: true - write to vram, otherwise - read from vram
+ */
+void amdgpu_device_mm_access(struct amdgpu_device *adev, loff_t pos,
+			     void *buf, size_t size, bool write)
 {
-	uint32_t ret;
+	unsigned long flags;
+	uint32_t hi = ~0, tmp = 0;
+	uint32_t *data = buf;
+	uint64_t last;
+	int idx;
 
-	if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) {
-		BUG_ON(in_interrupt());
-		return amdgpu_virt_kiq_rreg(adev, reg);
+	if (!drm_dev_enter(adev_to_drm(adev), &idx))
+		return;
+
+	BUG_ON(!IS_ALIGNED(pos, 4) || !IS_ALIGNED(size, 4));
+
+	spin_lock_irqsave(&adev->mmio_idx_lock, flags);
+	for (last = pos + size; pos < last; pos += 4) {
+		tmp = pos >> 31;
+
+		WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)pos) | 0x80000000);
+		if (tmp != hi) {
+			WREG32_NO_KIQ(mmMM_INDEX_HI, tmp);
+			hi = tmp;
+		}
+		if (write)
+			WREG32_NO_KIQ(mmMM_DATA, *data++);
+		else
+			*data++ = RREG32_NO_KIQ(mmMM_DATA);
 	}
 
-	if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
-		ret = readl(((void __iomem *)adev->rmmio) + (reg * 4));
-	else {
-		unsigned long flags;
+	spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
+	drm_dev_exit(idx);
+}
+
+/**
+ * amdgpu_device_aper_access - access vram by vram aperture
+ *
+ * @adev: amdgpu_device pointer
+ * @pos: offset of the buffer in vram
+ * @buf: virtual address of the buffer in system memory
+ * @size: read/write size, sizeof(@buf) must > @size
+ * @write: true - write to vram, otherwise - read from vram
+ *
+ * The return value means how many bytes have been transferred.
+ */
+size_t amdgpu_device_aper_access(struct amdgpu_device *adev, loff_t pos,
+				 void *buf, size_t size, bool write)
+{
+#ifdef CONFIG_64BIT
+	void __iomem *addr;
+	size_t count = 0;
+	uint64_t last;
+
+	if (!adev->mman.aper_base_kaddr)
+		return 0;
+
+	last = min(pos + size, adev->gmc.visible_vram_size);
+	if (last > pos) {
+		addr = adev->mman.aper_base_kaddr + pos;
+		count = last - pos;
+
+		if (write) {
+			memcpy_toio(addr, buf, count);
+			/* Make sure HDP write cache flush happens without any reordering
+			 * after the system memory contents are sent over PCIe device
+			 */
+			mb();
+			amdgpu_device_flush_hdp(adev, NULL);
+		} else {
+			amdgpu_device_invalidate_hdp(adev, NULL);
+			/* Make sure HDP read cache is invalidated before issuing a read
+			 * to the PCIe device
+			 */
+			mb();
+			memcpy_fromio(buf, addr, count);
+		}
 
-		spin_lock_irqsave(&adev->mmio_idx_lock, flags);
-		writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4));
-		ret = readl(((void __iomem *)adev->rmmio) + (mmMM_DATA * 4));
-		spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
 	}
-	trace_amdgpu_mm_rreg(adev->pdev->device, reg, ret);
-	return ret;
+
+	return count;
+#else
+	return 0;
+#endif
 }
 
-void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v,
-		    uint32_t acc_flags)
+/**
+ * amdgpu_device_vram_access - read/write a buffer in vram
+ *
+ * @adev: amdgpu_device pointer
+ * @pos: offset of the buffer in vram
+ * @buf: virtual address of the buffer in system memory
+ * @size: read/write size, sizeof(@buf) must > @size
+ * @write: true - write to vram, otherwise - read from vram
+ */
+void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos,
+			       void *buf, size_t size, bool write)
 {
-	trace_amdgpu_mm_wreg(adev->pdev->device, reg, v);
+	size_t count;
 
-	if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) {
-		BUG_ON(in_interrupt());
-		return amdgpu_virt_kiq_wreg(adev, reg, v);
+	/* try to using vram apreature to access vram first */
+	count = amdgpu_device_aper_access(adev, pos, buf, size, write);
+	size -= count;
+	if (size) {
+		/* using MM to access rest vram */
+		pos += count;
+		buf += count;
+		amdgpu_device_mm_access(adev, pos, buf, size, write);
 	}
+}
 
-	if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
-		writel(v, ((void __iomem *)adev->rmmio) + (reg * 4));
-	else {
-		unsigned long flags;
+/*
+ * register access helper functions.
+ */
+
+/* Check if hw access should be skipped because of hotplug or device error */
+bool amdgpu_device_skip_hw_access(struct amdgpu_device *adev)
+{
+	if (adev->no_hw_access)
+		return true;
 
-		spin_lock_irqsave(&adev->mmio_idx_lock, flags);
-		writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4));
-		writel(v, ((void __iomem *)adev->rmmio) + (mmMM_DATA * 4));
-		spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
+#ifdef CONFIG_LOCKDEP
+	/*
+	 * This is a bit complicated to understand, so worth a comment. What we assert
+	 * here is that the GPU reset is not running on another thread in parallel.
+	 *
+	 * For this we trylock the read side of the reset semaphore, if that succeeds
+	 * we know that the reset is not running in parallel.
+	 *
+	 * If the trylock fails we assert that we are either already holding the read
+	 * side of the lock or are the reset thread itself and hold the write side of
+	 * the lock.
+	 */
+	if (in_task()) {
+		if (down_read_trylock(&adev->reset_domain->sem))
+			up_read(&adev->reset_domain->sem);
+		else
+			lockdep_assert_held(&adev->reset_domain->sem);
 	}
+#endif
+	return false;
 }
 
-u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg)
+/**
+ * amdgpu_device_rreg - read a memory mapped IO or indirect register
+ *
+ * @adev: amdgpu_device pointer
+ * @reg: dword aligned register offset
+ * @acc_flags: access flags which require special behavior
+ *
+ * Returns the 32 bit value from the offset specified.
+ */
+uint32_t amdgpu_device_rreg(struct amdgpu_device *adev,
+			    uint32_t reg, uint32_t acc_flags)
 {
-	if ((reg * 4) < adev->rio_mem_size)
-		return ioread32(adev->rio_mem + (reg * 4));
-	else {
-		iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
-		return ioread32(adev->rio_mem + (mmMM_DATA * 4));
+	uint32_t ret;
+
+	if (amdgpu_device_skip_hw_access(adev))
+		return 0;
+
+	if ((reg * 4) < adev->rmmio_size) {
+		if (!(acc_flags & AMDGPU_REGS_NO_KIQ) &&
+		    amdgpu_sriov_runtime(adev) &&
+		    down_read_trylock(&adev->reset_domain->sem)) {
+			ret = amdgpu_kiq_rreg(adev, reg, 0);
+			up_read(&adev->reset_domain->sem);
+		} else {
+			ret = readl(((void __iomem *)adev->rmmio) + (reg * 4));
+		}
+	} else {
+		ret = adev->pcie_rreg(adev, reg * 4);
 	}
+
+	trace_amdgpu_device_rreg(adev->pdev->device, reg, ret);
+
+	return ret;
 }
 
-void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
+/*
+ * MMIO register read with bytes helper functions
+ * @offset:bytes offset from MMIO start
+ */
+
+/**
+ * amdgpu_mm_rreg8 - read a memory mapped IO register
+ *
+ * @adev: amdgpu_device pointer
+ * @offset: byte aligned register offset
+ *
+ * Returns the 8 bit value from the offset specified.
+ */
+uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset)
 {
+	if (amdgpu_device_skip_hw_access(adev))
+		return 0;
 
-	if ((reg * 4) < adev->rio_mem_size)
-		iowrite32(v, adev->rio_mem + (reg * 4));
-	else {
-		iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
-		iowrite32(v, adev->rio_mem + (mmMM_DATA * 4));
+	if (offset < adev->rmmio_size)
+		return (readb(adev->rmmio + offset));
+	BUG();
+}
+
+
+/**
+ * amdgpu_device_xcc_rreg - read a memory mapped IO or indirect register with specific XCC
+ *
+ * @adev: amdgpu_device pointer
+ * @reg: dword aligned register offset
+ * @acc_flags: access flags which require special behavior
+ * @xcc_id: xcc accelerated compute core id
+ *
+ * Returns the 32 bit value from the offset specified.
+ */
+uint32_t amdgpu_device_xcc_rreg(struct amdgpu_device *adev,
+				uint32_t reg, uint32_t acc_flags,
+				uint32_t xcc_id)
+{
+	uint32_t ret, rlcg_flag;
+
+	if (amdgpu_device_skip_hw_access(adev))
+		return 0;
+
+	if ((reg * 4) < adev->rmmio_size) {
+		if (amdgpu_sriov_vf(adev) &&
+		    !amdgpu_sriov_runtime(adev) &&
+		    adev->gfx.rlc.rlcg_reg_access_supported &&
+		    amdgpu_virt_get_rlcg_reg_access_flag(adev, acc_flags,
+							 GC_HWIP, false,
+							 &rlcg_flag)) {
+			ret = amdgpu_virt_rlcg_reg_rw(adev, reg, 0, rlcg_flag, GET_INST(GC, xcc_id));
+		} else if (!(acc_flags & AMDGPU_REGS_NO_KIQ) &&
+		    amdgpu_sriov_runtime(adev) &&
+		    down_read_trylock(&adev->reset_domain->sem)) {
+			ret = amdgpu_kiq_rreg(adev, reg, xcc_id);
+			up_read(&adev->reset_domain->sem);
+		} else {
+			ret = readl(((void __iomem *)adev->rmmio) + (reg * 4));
+		}
+	} else {
+		ret = adev->pcie_rreg(adev, reg * 4);
 	}
+
+	return ret;
+}
+
+/*
+ * MMIO register write with bytes helper functions
+ * @offset:bytes offset from MMIO start
+ * @value: the value want to be written to the register
+ */
+
+/**
+ * amdgpu_mm_wreg8 - read a memory mapped IO register
+ *
+ * @adev: amdgpu_device pointer
+ * @offset: byte aligned register offset
+ * @value: 8 bit value to write
+ *
+ * Writes the value specified to the offset specified.
+ */
+void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value)
+{
+	if (amdgpu_device_skip_hw_access(adev))
+		return;
+
+	if (offset < adev->rmmio_size)
+		writeb(value, adev->rmmio + offset);
+	else
+		BUG();
 }
 
 /**
- * amdgpu_mm_rdoorbell - read a doorbell dword
+ * amdgpu_device_wreg - write to a memory mapped IO or indirect register
  *
  * @adev: amdgpu_device pointer
- * @index: doorbell index
+ * @reg: dword aligned register offset
+ * @v: 32 bit value to write to the register
+ * @acc_flags: access flags which require special behavior
  *
- * Returns the value in the doorbell aperture at the
- * requested doorbell index (CIK).
+ * Writes the value specified to the offset specified.
  */
-u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index)
+void amdgpu_device_wreg(struct amdgpu_device *adev,
+			uint32_t reg, uint32_t v,
+			uint32_t acc_flags)
 {
-	if (index < adev->doorbell.num_doorbells) {
-		return readl(adev->doorbell.ptr + index);
+	if (amdgpu_device_skip_hw_access(adev))
+		return;
+
+	if ((reg * 4) < adev->rmmio_size) {
+		if (!(acc_flags & AMDGPU_REGS_NO_KIQ) &&
+		    amdgpu_sriov_runtime(adev) &&
+		    down_read_trylock(&adev->reset_domain->sem)) {
+			amdgpu_kiq_wreg(adev, reg, v, 0);
+			up_read(&adev->reset_domain->sem);
+		} else {
+			writel(v, ((void __iomem *)adev->rmmio) + (reg * 4));
+		}
 	} else {
-		DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
-		return 0;
+		adev->pcie_wreg(adev, reg * 4, v);
 	}
+
+	trace_amdgpu_device_wreg(adev->pdev->device, reg, v);
 }
 
 /**
- * amdgpu_mm_wdoorbell - write a doorbell dword
+ * amdgpu_mm_wreg_mmio_rlc -  write register either with direct/indirect mmio or with RLC path if in range
  *
  * @adev: amdgpu_device pointer
- * @index: doorbell index
+ * @reg: mmio/rlc register
  * @v: value to write
+ * @xcc_id: xcc accelerated compute core id
  *
- * Writes @v to the doorbell aperture at the
- * requested doorbell index (CIK).
+ * this function is invoked only for the debugfs register access
  */
-void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v)
+void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev,
+			     uint32_t reg, uint32_t v,
+			     uint32_t xcc_id)
 {
-	if (index < adev->doorbell.num_doorbells) {
-		writel(v, adev->doorbell.ptr + index);
+	if (amdgpu_device_skip_hw_access(adev))
+		return;
+
+	if (amdgpu_sriov_fullaccess(adev) &&
+	    adev->gfx.rlc.funcs &&
+	    adev->gfx.rlc.funcs->is_rlcg_access_range) {
+		if (adev->gfx.rlc.funcs->is_rlcg_access_range(adev, reg))
+			return amdgpu_sriov_wreg(adev, reg, v, 0, 0, xcc_id);
+	} else if ((reg * 4) >= adev->rmmio_size) {
+		adev->pcie_wreg(adev, reg * 4, v);
 	} else {
-		DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
+		writel(v, ((void __iomem *)adev->rmmio) + (reg * 4));
 	}
 }
 
 /**
- * amdgpu_mm_rdoorbell64 - read a doorbell Qword
+ * amdgpu_device_xcc_wreg - write to a memory mapped IO or indirect register with specific XCC
  *
  * @adev: amdgpu_device pointer
- * @index: doorbell index
+ * @reg: dword aligned register offset
+ * @v: 32 bit value to write to the register
+ * @acc_flags: access flags which require special behavior
+ * @xcc_id: xcc accelerated compute core id
  *
- * Returns the value in the doorbell aperture at the
- * requested doorbell index (VEGA10+).
+ * Writes the value specified to the offset specified.
  */
-u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index)
+void amdgpu_device_xcc_wreg(struct amdgpu_device *adev,
+			uint32_t reg, uint32_t v,
+			uint32_t acc_flags, uint32_t xcc_id)
 {
-	if (index < adev->doorbell.num_doorbells) {
-		return atomic64_read((atomic64_t *)(adev->doorbell.ptr + index));
+	uint32_t rlcg_flag;
+
+	if (amdgpu_device_skip_hw_access(adev))
+		return;
+
+	if ((reg * 4) < adev->rmmio_size) {
+		if (amdgpu_sriov_vf(adev) &&
+		    !amdgpu_sriov_runtime(adev) &&
+		    adev->gfx.rlc.rlcg_reg_access_supported &&
+		    amdgpu_virt_get_rlcg_reg_access_flag(adev, acc_flags,
+							 GC_HWIP, true,
+							 &rlcg_flag)) {
+			amdgpu_virt_rlcg_reg_rw(adev, reg, v, rlcg_flag, GET_INST(GC, xcc_id));
+		} else if (!(acc_flags & AMDGPU_REGS_NO_KIQ) &&
+		    amdgpu_sriov_runtime(adev) &&
+		    down_read_trylock(&adev->reset_domain->sem)) {
+			amdgpu_kiq_wreg(adev, reg, v, xcc_id);
+			up_read(&adev->reset_domain->sem);
+		} else {
+			writel(v, ((void __iomem *)adev->rmmio) + (reg * 4));
+		}
 	} else {
-		DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
-		return 0;
+		adev->pcie_wreg(adev, reg * 4, v);
 	}
 }
 
 /**
- * amdgpu_mm_wdoorbell64 - write a doorbell Qword
+ * amdgpu_device_indirect_rreg - read an indirect register
  *
  * @adev: amdgpu_device pointer
- * @index: doorbell index
- * @v: value to write
+ * @reg_addr: indirect register address to read from
  *
- * Writes @v to the doorbell aperture at the
- * requested doorbell index (VEGA10+).
+ * Returns the value of indirect register @reg_addr
  */
-void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v)
+u32 amdgpu_device_indirect_rreg(struct amdgpu_device *adev,
+				u32 reg_addr)
+{
+	unsigned long flags, pcie_index, pcie_data;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_data_offset;
+	u32 r;
+
+	pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+	pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	r = readl(pcie_data_offset);
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+
+	return r;
+}
+
+u32 amdgpu_device_indirect_rreg_ext(struct amdgpu_device *adev,
+				    u64 reg_addr)
 {
-	if (index < adev->doorbell.num_doorbells) {
-		atomic64_set((atomic64_t *)(adev->doorbell.ptr + index), v);
+	unsigned long flags, pcie_index, pcie_index_hi, pcie_data;
+	u32 r;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_index_hi_offset;
+	void __iomem *pcie_data_offset;
+
+	if (unlikely(!adev->nbio.funcs)) {
+		pcie_index = AMDGPU_PCIE_INDEX_FALLBACK;
+		pcie_data = AMDGPU_PCIE_DATA_FALLBACK;
+	} else {
+		pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+		pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+	}
+
+	if (reg_addr >> 32) {
+		if (unlikely(!adev->nbio.funcs))
+			pcie_index_hi = AMDGPU_PCIE_INDEX_HI_FALLBACK;
+		else
+			pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev);
 	} else {
-		DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
+		pcie_index_hi = 0;
+	}
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+	if (pcie_index_hi != 0)
+		pcie_index_hi_offset = (void __iomem *)adev->rmmio +
+				pcie_index_hi * 4;
+
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	if (pcie_index_hi != 0) {
+		writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+	r = readl(pcie_data_offset);
+
+	/* clear the high bits */
+	if (pcie_index_hi != 0) {
+		writel(0, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+
+	return r;
+}
+
+/**
+ * amdgpu_device_indirect_rreg64 - read a 64bits indirect register
+ *
+ * @adev: amdgpu_device pointer
+ * @reg_addr: indirect register address to read from
+ *
+ * Returns the value of indirect register @reg_addr
+ */
+u64 amdgpu_device_indirect_rreg64(struct amdgpu_device *adev,
+				  u32 reg_addr)
+{
+	unsigned long flags, pcie_index, pcie_data;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_data_offset;
+	u64 r;
+
+	pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+	pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+
+	/* read low 32 bits */
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	r = readl(pcie_data_offset);
+	/* read high 32 bits */
+	writel(reg_addr + 4, pcie_index_offset);
+	readl(pcie_index_offset);
+	r |= ((u64)readl(pcie_data_offset) << 32);
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+
+	return r;
+}
+
+u64 amdgpu_device_indirect_rreg64_ext(struct amdgpu_device *adev,
+				  u64 reg_addr)
+{
+	unsigned long flags, pcie_index, pcie_data;
+	unsigned long pcie_index_hi = 0;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_index_hi_offset;
+	void __iomem *pcie_data_offset;
+	u64 r;
+
+	pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+	pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+	if ((reg_addr >> 32) && (adev->nbio.funcs->get_pcie_index_hi_offset))
+		pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev);
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+	if (pcie_index_hi != 0)
+		pcie_index_hi_offset = (void __iomem *)adev->rmmio +
+			pcie_index_hi * 4;
+
+	/* read low 32 bits */
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	if (pcie_index_hi != 0) {
+		writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+	r = readl(pcie_data_offset);
+	/* read high 32 bits */
+	writel(reg_addr + 4, pcie_index_offset);
+	readl(pcie_index_offset);
+	if (pcie_index_hi != 0) {
+		writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+	r |= ((u64)readl(pcie_data_offset) << 32);
+
+	/* clear the high bits */
+	if (pcie_index_hi != 0) {
+		writel(0, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+
+	return r;
+}
+
+/**
+ * amdgpu_device_indirect_wreg - write an indirect register address
+ *
+ * @adev: amdgpu_device pointer
+ * @reg_addr: indirect register offset
+ * @reg_data: indirect register data
+ *
+ */
+void amdgpu_device_indirect_wreg(struct amdgpu_device *adev,
+				 u32 reg_addr, u32 reg_data)
+{
+	unsigned long flags, pcie_index, pcie_data;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_data_offset;
+
+	pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+	pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	writel(reg_data, pcie_data_offset);
+	readl(pcie_data_offset);
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+}
+
+void amdgpu_device_indirect_wreg_ext(struct amdgpu_device *adev,
+				     u64 reg_addr, u32 reg_data)
+{
+	unsigned long flags, pcie_index, pcie_index_hi, pcie_data;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_index_hi_offset;
+	void __iomem *pcie_data_offset;
+
+	pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+	pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+	if ((reg_addr >> 32) && (adev->nbio.funcs->get_pcie_index_hi_offset))
+		pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev);
+	else
+		pcie_index_hi = 0;
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+	if (pcie_index_hi != 0)
+		pcie_index_hi_offset = (void __iomem *)adev->rmmio +
+				pcie_index_hi * 4;
+
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	if (pcie_index_hi != 0) {
+		writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+	writel(reg_data, pcie_data_offset);
+	readl(pcie_data_offset);
+
+	/* clear the high bits */
+	if (pcie_index_hi != 0) {
+		writel(0, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
 	}
+
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+}
+
+/**
+ * amdgpu_device_indirect_wreg64 - write a 64bits indirect register address
+ *
+ * @adev: amdgpu_device pointer
+ * @reg_addr: indirect register offset
+ * @reg_data: indirect register data
+ *
+ */
+void amdgpu_device_indirect_wreg64(struct amdgpu_device *adev,
+				   u32 reg_addr, u64 reg_data)
+{
+	unsigned long flags, pcie_index, pcie_data;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_data_offset;
+
+	pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+	pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+
+	/* write low 32 bits */
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	writel((u32)(reg_data & 0xffffffffULL), pcie_data_offset);
+	readl(pcie_data_offset);
+	/* write high 32 bits */
+	writel(reg_addr + 4, pcie_index_offset);
+	readl(pcie_index_offset);
+	writel((u32)(reg_data >> 32), pcie_data_offset);
+	readl(pcie_data_offset);
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+}
+
+void amdgpu_device_indirect_wreg64_ext(struct amdgpu_device *adev,
+				   u64 reg_addr, u64 reg_data)
+{
+	unsigned long flags, pcie_index, pcie_data;
+	unsigned long pcie_index_hi = 0;
+	void __iomem *pcie_index_offset;
+	void __iomem *pcie_index_hi_offset;
+	void __iomem *pcie_data_offset;
+
+	pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev);
+	pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev);
+	if ((reg_addr >> 32) && (adev->nbio.funcs->get_pcie_index_hi_offset))
+		pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev);
+
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4;
+	pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4;
+	if (pcie_index_hi != 0)
+		pcie_index_hi_offset = (void __iomem *)adev->rmmio +
+				pcie_index_hi * 4;
+
+	/* write low 32 bits */
+	writel(reg_addr, pcie_index_offset);
+	readl(pcie_index_offset);
+	if (pcie_index_hi != 0) {
+		writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+	writel((u32)(reg_data & 0xffffffffULL), pcie_data_offset);
+	readl(pcie_data_offset);
+	/* write high 32 bits */
+	writel(reg_addr + 4, pcie_index_offset);
+	readl(pcie_index_offset);
+	if (pcie_index_hi != 0) {
+		writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+	writel((u32)(reg_data >> 32), pcie_data_offset);
+	readl(pcie_data_offset);
+
+	/* clear the high bits */
+	if (pcie_index_hi != 0) {
+		writel(0, pcie_index_hi_offset);
+		readl(pcie_index_hi_offset);
+	}
+
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+}
+
+/**
+ * amdgpu_device_get_rev_id - query device rev_id
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Return device rev_id
+ */
+u32 amdgpu_device_get_rev_id(struct amdgpu_device *adev)
+{
+	return adev->nbio.funcs->get_rev_id(adev);
 }
 
 /**
  * amdgpu_invalid_rreg - dummy reg read function
  *
- * @adev: amdgpu device pointer
+ * @adev: amdgpu_device pointer
  * @reg: offset of register
  *
  * Dummy register read function.  Used for register blocks
@@ -247,7 +1288,14 @@ void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v)
  */
 static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg)
 {
-	DRM_ERROR("Invalid callback to read register 0x%04X\n", reg);
+	dev_err(adev->dev, "Invalid callback to read register 0x%04X\n", reg);
+	BUG();
+	return 0;
+}
+
+static uint32_t amdgpu_invalid_rreg_ext(struct amdgpu_device *adev, uint64_t reg)
+{
+	dev_err(adev->dev, "Invalid callback to read register 0x%llX\n", reg);
 	BUG();
 	return 0;
 }
@@ -255,7 +1303,7 @@ static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg)
 /**
  * amdgpu_invalid_wreg - dummy reg write function
  *
- * @adev: amdgpu device pointer
+ * @adev: amdgpu_device pointer
  * @reg: offset of register
  * @v: value to write to the register
  *
@@ -264,15 +1312,75 @@ static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg)
  */
 static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v)
 {
-	DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n",
-		  reg, v);
+	dev_err(adev->dev,
+		"Invalid callback to write register 0x%04X with 0x%08X\n", reg,
+		v);
+	BUG();
+}
+
+static void amdgpu_invalid_wreg_ext(struct amdgpu_device *adev, uint64_t reg, uint32_t v)
+{
+	dev_err(adev->dev,
+		"Invalid callback to write register 0x%llX with 0x%08X\n", reg,
+		v);
+	BUG();
+}
+
+/**
+ * amdgpu_invalid_rreg64 - dummy 64 bit reg read function
+ *
+ * @adev: amdgpu_device pointer
+ * @reg: offset of register
+ *
+ * Dummy register read function.  Used for register blocks
+ * that certain asics don't have (all asics).
+ * Returns the value in the register.
+ */
+static uint64_t amdgpu_invalid_rreg64(struct amdgpu_device *adev, uint32_t reg)
+{
+	dev_err(adev->dev, "Invalid callback to read 64 bit register 0x%04X\n",
+		reg);
+	BUG();
+	return 0;
+}
+
+static uint64_t amdgpu_invalid_rreg64_ext(struct amdgpu_device *adev, uint64_t reg)
+{
+	dev_err(adev->dev, "Invalid callback to read register 0x%llX\n", reg);
+	BUG();
+	return 0;
+}
+
+/**
+ * amdgpu_invalid_wreg64 - dummy reg write function
+ *
+ * @adev: amdgpu_device pointer
+ * @reg: offset of register
+ * @v: value to write to the register
+ *
+ * Dummy register read function.  Used for register blocks
+ * that certain asics don't have (all asics).
+ */
+static void amdgpu_invalid_wreg64(struct amdgpu_device *adev, uint32_t reg, uint64_t v)
+{
+	dev_err(adev->dev,
+		"Invalid callback to write 64 bit register 0x%04X with 0x%08llX\n",
+		reg, v);
+	BUG();
+}
+
+static void amdgpu_invalid_wreg64_ext(struct amdgpu_device *adev, uint64_t reg, uint64_t v)
+{
+	dev_err(adev->dev,
+		"Invalid callback to write 64 bit register 0x%llX with 0x%08llX\n",
+		reg, v);
 	BUG();
 }
 
 /**
  * amdgpu_block_invalid_rreg - dummy reg read function
  *
- * @adev: amdgpu device pointer
+ * @adev: amdgpu_device pointer
  * @block: offset of instance
  * @reg: offset of register
  *
@@ -283,8 +1391,9 @@ static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32
 static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev,
 					  uint32_t block, uint32_t reg)
 {
-	DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n",
-		  reg, block);
+	dev_err(adev->dev,
+		"Invalid callback to read register 0x%04X in block 0x%04X\n",
+		reg, block);
 	BUG();
 	return 0;
 }
@@ -292,7 +1401,7 @@ static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev,
 /**
  * amdgpu_block_invalid_wreg - dummy reg write function
  *
- * @adev: amdgpu device pointer
+ * @adev: amdgpu_device pointer
  * @block: offset of instance
  * @reg: offset of register
  * @v: value to write to the register
@@ -304,73 +1413,103 @@ static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev,
 				      uint32_t block,
 				      uint32_t reg, uint32_t v)
 {
-	DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n",
-		  reg, block, v);
+	dev_err(adev->dev,
+		"Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n",
+		reg, block, v);
 	BUG();
 }
 
-static int amdgpu_vram_scratch_init(struct amdgpu_device *adev)
+static uint32_t amdgpu_device_get_vbios_flags(struct amdgpu_device *adev)
 {
-	int r;
+	if (hweight32(adev->aid_mask) && (adev->flags & AMD_IS_APU))
+		return AMDGPU_VBIOS_SKIP;
 
-	if (adev->vram_scratch.robj == NULL) {
-		r = amdgpu_bo_create(adev, AMDGPU_GPU_PAGE_SIZE,
-				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
-				     NULL, NULL, &adev->vram_scratch.robj);
-		if (r) {
-			return r;
-		}
-	}
+	if (hweight32(adev->aid_mask) && amdgpu_passthrough(adev))
+		return AMDGPU_VBIOS_OPTIONAL;
 
-	r = amdgpu_bo_reserve(adev->vram_scratch.robj, false);
-	if (unlikely(r != 0))
-		return r;
-	r = amdgpu_bo_pin(adev->vram_scratch.robj,
-			  AMDGPU_GEM_DOMAIN_VRAM, &adev->vram_scratch.gpu_addr);
-	if (r) {
-		amdgpu_bo_unreserve(adev->vram_scratch.robj);
-		return r;
+	return 0;
+}
+
+/**
+ * amdgpu_device_asic_init - Wrapper for atom asic_init
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Does any asic specific work and then calls atom asic init.
+ */
+static int amdgpu_device_asic_init(struct amdgpu_device *adev)
+{
+	uint32_t flags;
+	bool optional;
+	int ret;
+
+	amdgpu_asic_pre_asic_init(adev);
+	flags = amdgpu_device_get_vbios_flags(adev);
+	optional = !!(flags & (AMDGPU_VBIOS_OPTIONAL | AMDGPU_VBIOS_SKIP));
+
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 5, 0) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0)) {
+		amdgpu_psp_wait_for_bootloader(adev);
+		if (optional && !adev->bios)
+			return 0;
+
+		ret = amdgpu_atomfirmware_asic_init(adev, true);
+		return ret;
+	} else {
+		if (optional && !adev->bios)
+			return 0;
+
+		return amdgpu_atom_asic_init(adev->mode_info.atom_context);
 	}
-	r = amdgpu_bo_kmap(adev->vram_scratch.robj,
-				(void **)&adev->vram_scratch.ptr);
-	if (r)
-		amdgpu_bo_unpin(adev->vram_scratch.robj);
-	amdgpu_bo_unreserve(adev->vram_scratch.robj);
 
-	return r;
+	return 0;
 }
 
-static void amdgpu_vram_scratch_fini(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_mem_scratch_init - allocate the VRAM scratch page
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Allocates a scratch page of VRAM for use by various things in the
+ * driver.
+ */
+static int amdgpu_device_mem_scratch_init(struct amdgpu_device *adev)
 {
-	int r;
+	return amdgpu_bo_create_kernel(adev, AMDGPU_GPU_PAGE_SIZE, PAGE_SIZE,
+				       AMDGPU_GEM_DOMAIN_VRAM |
+				       AMDGPU_GEM_DOMAIN_GTT,
+				       &adev->mem_scratch.robj,
+				       &adev->mem_scratch.gpu_addr,
+				       (void **)&adev->mem_scratch.ptr);
+}
 
-	if (adev->vram_scratch.robj == NULL) {
-		return;
-	}
-	r = amdgpu_bo_reserve(adev->vram_scratch.robj, true);
-	if (likely(r == 0)) {
-		amdgpu_bo_kunmap(adev->vram_scratch.robj);
-		amdgpu_bo_unpin(adev->vram_scratch.robj);
-		amdgpu_bo_unreserve(adev->vram_scratch.robj);
-	}
-	amdgpu_bo_unref(&adev->vram_scratch.robj);
+/**
+ * amdgpu_device_mem_scratch_fini - Free the VRAM scratch page
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Frees the VRAM scratch page.
+ */
+static void amdgpu_device_mem_scratch_fini(struct amdgpu_device *adev)
+{
+	amdgpu_bo_free_kernel(&adev->mem_scratch.robj, NULL, NULL);
 }
 
 /**
- * amdgpu_program_register_sequence - program an array of registers.
+ * amdgpu_device_program_register_sequence - program an array of registers.
  *
  * @adev: amdgpu_device pointer
  * @registers: pointer to the register array
  * @array_size: size of the register array
  *
- * Programs an array or registers with and and or masks.
+ * Programs an array or registers with and or masks.
  * This is a helper for setting golden registers.
  */
-void amdgpu_program_register_sequence(struct amdgpu_device *adev,
-				      const u32 *registers,
-				      const u32 array_size)
+void amdgpu_device_program_register_sequence(struct amdgpu_device *adev,
+					     const u32 *registers,
+					     const u32 array_size)
 {
 	u32 tmp, reg, and_mask, or_mask;
 	int i;
@@ -378,7 +1517,7 @@ void amdgpu_program_register_sequence(struct amdgpu_device *adev,
 	if (array_size % 3)
 		return;
 
-	for (i = 0; i < array_size; i +=3) {
+	for (i = 0; i < array_size; i += 3) {
 		reg = registers[i + 0];
 		and_mask = registers[i + 1];
 		or_mask = registers[i + 2];
@@ -388,110 +1527,55 @@ void amdgpu_program_register_sequence(struct amdgpu_device *adev,
 		} else {
 			tmp = RREG32(reg);
 			tmp &= ~and_mask;
-			tmp |= or_mask;
+			if (adev->family >= AMDGPU_FAMILY_AI)
+				tmp |= (or_mask & and_mask);
+			else
+				tmp |= or_mask;
 		}
 		WREG32(reg, tmp);
 	}
 }
 
-void amdgpu_pci_config_reset(struct amdgpu_device *adev)
-{
-	pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA);
-}
-
-/*
- * GPU doorbell aperture helpers function.
- */
-/**
- * amdgpu_doorbell_init - Init doorbell driver information.
- *
- * @adev: amdgpu_device pointer
- *
- * Init doorbell driver information (CIK)
- * Returns 0 on success, error on failure.
- */
-static int amdgpu_doorbell_init(struct amdgpu_device *adev)
-{
-	/* doorbell bar mapping */
-	adev->doorbell.base = pci_resource_start(adev->pdev, 2);
-	adev->doorbell.size = pci_resource_len(adev->pdev, 2);
-
-	adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32),
-					     AMDGPU_DOORBELL_MAX_ASSIGNMENT+1);
-	if (adev->doorbell.num_doorbells == 0)
-		return -EINVAL;
-
-	adev->doorbell.ptr = ioremap(adev->doorbell.base,
-				     adev->doorbell.num_doorbells *
-				     sizeof(u32));
-	if (adev->doorbell.ptr == NULL)
-		return -ENOMEM;
-
-	return 0;
-}
-
 /**
- * amdgpu_doorbell_fini - Tear down doorbell driver information.
+ * amdgpu_device_pci_config_reset - reset the GPU
  *
  * @adev: amdgpu_device pointer
  *
- * Tear down doorbell driver information (CIK)
+ * Resets the GPU using the pci config reset sequence.
+ * Only applicable to asics prior to vega10.
  */
-static void amdgpu_doorbell_fini(struct amdgpu_device *adev)
+void amdgpu_device_pci_config_reset(struct amdgpu_device *adev)
 {
-	iounmap(adev->doorbell.ptr);
-	adev->doorbell.ptr = NULL;
+	pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA);
 }
 
 /**
- * amdgpu_doorbell_get_kfd_info - Report doorbell configuration required to
- *                                setup amdkfd
+ * amdgpu_device_pci_reset - reset the GPU using generic PCI means
  *
  * @adev: amdgpu_device pointer
- * @aperture_base: output returning doorbell aperture base physical address
- * @aperture_size: output returning doorbell aperture size in bytes
- * @start_offset: output returning # of doorbell bytes reserved for amdgpu.
  *
- * amdgpu and amdkfd share the doorbell aperture. amdgpu sets it up,
- * takes doorbells required for its own rings and reports the setup to amdkfd.
- * amdgpu reserved doorbells are at the start of the doorbell aperture.
+ * Resets the GPU using generic pci reset interfaces (FLR, SBR, etc.).
  */
-void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev,
-				phys_addr_t *aperture_base,
-				size_t *aperture_size,
-				size_t *start_offset)
+int amdgpu_device_pci_reset(struct amdgpu_device *adev)
 {
-	/*
-	 * The first num_doorbells are used by amdgpu.
-	 * amdkfd takes whatever's left in the aperture.
-	 */
-	if (adev->doorbell.size > adev->doorbell.num_doorbells * sizeof(u32)) {
-		*aperture_base = adev->doorbell.base;
-		*aperture_size = adev->doorbell.size;
-		*start_offset = adev->doorbell.num_doorbells * sizeof(u32);
-	} else {
-		*aperture_base = 0;
-		*aperture_size = 0;
-		*start_offset = 0;
-	}
+	return pci_reset_function(adev->pdev);
 }
 
 /*
- * amdgpu_wb_*()
- * Writeback is the the method by which the the GPU updates special pages
- * in memory with the status of certain GPU events (fences, ring pointers,
- * etc.).
+ * amdgpu_device_wb_*()
+ * Writeback is the method by which the GPU updates special pages in memory
+ * with the status of certain GPU events (fences, ring pointers,etc.).
  */
 
 /**
- * amdgpu_wb_fini - Disable Writeback and free memory
+ * amdgpu_device_wb_fini - Disable Writeback and free memory
  *
  * @adev: amdgpu_device pointer
  *
  * Disables Writeback and frees the Writeback memory (all asics).
  * Used at driver shutdown.
  */
-static void amdgpu_wb_fini(struct amdgpu_device *adev)
+static void amdgpu_device_wb_fini(struct amdgpu_device *adev)
 {
 	if (adev->wb.wb_obj) {
 		amdgpu_bo_free_kernel(&adev->wb.wb_obj,
@@ -502,20 +1586,21 @@ static void amdgpu_wb_fini(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_wb_init- Init Writeback driver info and allocate memory
+ * amdgpu_device_wb_init - Init Writeback driver info and allocate memory
  *
  * @adev: amdgpu_device pointer
  *
- * Disables Writeback and frees the Writeback memory (all asics).
+ * Initializes writeback and allocates writeback memory (all asics).
  * Used at driver startup.
  * Returns 0 on success or an -error on failure.
  */
-static int amdgpu_wb_init(struct amdgpu_device *adev)
+static int amdgpu_device_wb_init(struct amdgpu_device *adev)
 {
 	int r;
 
 	if (adev->wb.wb_obj == NULL) {
-		r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t),
+		/* AMDGPU_MAX_WB * sizeof(uint32_t) * 8 = AMDGPU_MAX_WB 256bit slots */
+		r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t) * 8,
 					    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
 					    &adev->wb.wb_obj, &adev->wb.gpu_addr,
 					    (void **)&adev->wb.wb);
@@ -528,14 +1613,14 @@ static int amdgpu_wb_init(struct amdgpu_device *adev)
 		memset(&adev->wb.used, 0, sizeof(adev->wb.used));
 
 		/* clear wb memory */
-		memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t));
+		memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t) * 8);
 	}
 
 	return 0;
 }
 
 /**
- * amdgpu_wb_get - Allocate a wb entry
+ * amdgpu_device_wb_get - Allocate a wb entry
  *
  * @adev: amdgpu_device pointer
  * @wb: wb index
@@ -543,162 +1628,143 @@ static int amdgpu_wb_init(struct amdgpu_device *adev)
  * Allocate a wb slot for use by the driver (all asics).
  * Returns 0 on success or -EINVAL on failure.
  */
-int amdgpu_wb_get(struct amdgpu_device *adev, u32 *wb)
+int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb)
 {
-	unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb);
-	if (offset < adev->wb.num_wb) {
-		__set_bit(offset, adev->wb.used);
-		*wb = offset;
-		return 0;
-	} else {
-		return -EINVAL;
-	}
-}
+	unsigned long flags, offset;
 
-/**
- * amdgpu_wb_get_64bit - Allocate a wb entry
- *
- * @adev: amdgpu_device pointer
- * @wb: wb index
- *
- * Allocate a wb slot for use by the driver (all asics).
- * Returns 0 on success or -EINVAL on failure.
- */
-int amdgpu_wb_get_64bit(struct amdgpu_device *adev, u32 *wb)
-{
-	unsigned long offset = bitmap_find_next_zero_area_off(adev->wb.used,
-				adev->wb.num_wb, 0, 2, 7, 0);
-	if ((offset + 1) < adev->wb.num_wb) {
+	spin_lock_irqsave(&adev->wb.lock, flags);
+	offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb);
+	if (offset < adev->wb.num_wb) {
 		__set_bit(offset, adev->wb.used);
-		__set_bit(offset + 1, adev->wb.used);
-		*wb = offset;
+		spin_unlock_irqrestore(&adev->wb.lock, flags);
+		*wb = offset << 3; /* convert to dw offset */
 		return 0;
 	} else {
+		spin_unlock_irqrestore(&adev->wb.lock, flags);
 		return -EINVAL;
 	}
 }
 
 /**
- * amdgpu_wb_free - Free a wb entry
+ * amdgpu_device_wb_free - Free a wb entry
  *
  * @adev: amdgpu_device pointer
  * @wb: wb index
  *
  * Free a wb slot allocated for use by the driver (all asics)
  */
-void amdgpu_wb_free(struct amdgpu_device *adev, u32 wb)
+void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb)
 {
+	unsigned long flags;
+
+	wb >>= 3;
+	spin_lock_irqsave(&adev->wb.lock, flags);
 	if (wb < adev->wb.num_wb)
 		__clear_bit(wb, adev->wb.used);
+	spin_unlock_irqrestore(&adev->wb.lock, flags);
 }
 
 /**
- * amdgpu_wb_free_64bit - Free a wb entry
+ * amdgpu_device_resize_fb_bar - try to resize FB BAR
  *
  * @adev: amdgpu_device pointer
- * @wb: wb index
  *
- * Free a wb slot allocated for use by the driver (all asics)
+ * Try to resize FB BAR to make all VRAM CPU accessible. We try very hard not
+ * to fail, but if any of the BARs is not accessible after the size we abort
+ * driver loading by returning -ENODEV.
  */
-void amdgpu_wb_free_64bit(struct amdgpu_device *adev, u32 wb)
+int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev)
 {
-	if ((wb + 1) < adev->wb.num_wb) {
-		__clear_bit(wb, adev->wb.used);
-		__clear_bit(wb + 1, adev->wb.used);
-	}
-}
+	int rbar_size = pci_rebar_bytes_to_size(adev->gmc.real_vram_size);
+	struct pci_bus *root;
+	struct resource *res;
+	int max_size, r;
+	unsigned int i;
+	u16 cmd;
+
+	if (!IS_ENABLED(CONFIG_PHYS_ADDR_T_64BIT))
+		return 0;
 
-/**
- * amdgpu_vram_location - try to find VRAM location
- * @adev: amdgpu device structure holding all necessary informations
- * @mc: memory controller structure holding memory informations
- * @base: base address at which to put VRAM
- *
- * Function will place try to place VRAM at base address provided
- * as parameter (which is so far either PCI aperture address or
- * for IGP TOM base address).
- *
- * If there is not enough space to fit the unvisible VRAM in the 32bits
- * address space then we limit the VRAM size to the aperture.
- *
- * Note: We don't explicitly enforce VRAM start to be aligned on VRAM size,
- * this shouldn't be a problem as we are using the PCI aperture as a reference.
- * Otherwise this would be needed for rv280, all r3xx, and all r4xx, but
- * not IGP.
- *
- * Note: we use mc_vram_size as on some board we need to program the mc to
- * cover the whole aperture even if VRAM size is inferior to aperture size
- * Novell bug 204882 + along with lots of ubuntu ones
- *
- * Note: when limiting vram it's safe to overwritte real_vram_size because
- * we are not in case where real_vram_size is inferior to mc_vram_size (ie
- * note afected by bogus hw of Novell bug 204882 + along with lots of ubuntu
- * ones)
- *
- * Note: IGP TOM addr should be the same as the aperture addr, we don't
- * explicitly check for that thought.
- *
- * FIXME: when reducing VRAM size align new size on power of 2.
- */
-void amdgpu_vram_location(struct amdgpu_device *adev, struct amdgpu_mc *mc, u64 base)
-{
-	uint64_t limit = (uint64_t)amdgpu_vram_limit << 20;
+	/* Bypass for VF */
+	if (amdgpu_sriov_vf(adev))
+		return 0;
 
-	mc->vram_start = base;
-	if (mc->mc_vram_size > (adev->mc.mc_mask - base + 1)) {
-		dev_warn(adev->dev, "limiting VRAM to PCI aperture size\n");
-		mc->real_vram_size = mc->aper_size;
-		mc->mc_vram_size = mc->aper_size;
-	}
-	mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
-	if (limit && limit < mc->real_vram_size)
-		mc->real_vram_size = limit;
-	dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
-			mc->mc_vram_size >> 20, mc->vram_start,
-			mc->vram_end, mc->real_vram_size >> 20);
-}
+	if (!amdgpu_rebar)
+		return 0;
 
-/**
- * amdgpu_gtt_location - try to find GTT location
- * @adev: amdgpu device structure holding all necessary informations
- * @mc: memory controller structure holding memory informations
- *
- * Function will place try to place GTT before or after VRAM.
- *
- * If GTT size is bigger than space left then we ajust GTT size.
- * Thus function will never fails.
- *
- * FIXME: when reducing GTT size align new size on power of 2.
- */
-void amdgpu_gtt_location(struct amdgpu_device *adev, struct amdgpu_mc *mc)
-{
-	u64 size_af, size_bf;
+	/* resizing on Dell G5 SE platforms causes problems with runtime pm */
+	if ((amdgpu_runtime_pm != 0) &&
+	    adev->pdev->vendor == PCI_VENDOR_ID_ATI &&
+	    adev->pdev->device == 0x731f &&
+	    adev->pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)
+		return 0;
 
-	size_af = ((adev->mc.mc_mask - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align;
-	size_bf = mc->vram_start & ~mc->gtt_base_align;
-	if (size_bf > size_af) {
-		if (mc->gtt_size > size_bf) {
-			dev_warn(adev->dev, "limiting GTT\n");
-			mc->gtt_size = size_bf;
-		}
-		mc->gtt_start = 0;
-	} else {
-		if (mc->gtt_size > size_af) {
-			dev_warn(adev->dev, "limiting GTT\n");
-			mc->gtt_size = size_af;
-		}
-		mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align;
+	/* PCI_EXT_CAP_ID_VNDR extended capability is located at 0x100 */
+	if (!pci_find_ext_capability(adev->pdev, PCI_EXT_CAP_ID_VNDR))
+		dev_warn(
+			adev->dev,
+			"System can't access extended configuration space, please check!!\n");
+
+	/* skip if the bios has already enabled large BAR */
+	if (adev->gmc.real_vram_size &&
+	    (pci_resource_len(adev->pdev, 0) >= adev->gmc.real_vram_size))
+		return 0;
+
+	/* Check if the root BUS has 64bit memory resources */
+	root = adev->pdev->bus;
+	while (root->parent)
+		root = root->parent;
+
+	pci_bus_for_each_resource(root, res, i) {
+		if (res && res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) &&
+		    res->start > 0x100000000ull)
+			break;
 	}
-	mc->gtt_end = mc->gtt_start + mc->gtt_size - 1;
-	dev_info(adev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n",
-			mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end);
+
+	/* Trying to resize is pointless without a root hub window above 4GB */
+	if (!res)
+		return 0;
+
+	/* Limit the BAR size to what is available */
+	max_size = pci_rebar_get_max_size(adev->pdev, 0);
+	if (max_size < 0)
+		return 0;
+	rbar_size = min(max_size, rbar_size);
+
+	/* Disable memory decoding while we change the BAR addresses and size */
+	pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd);
+	pci_write_config_word(adev->pdev, PCI_COMMAND,
+			      cmd & ~PCI_COMMAND_MEMORY);
+
+	/* Tear down doorbell as resizing will release BARs */
+	amdgpu_doorbell_fini(adev);
+
+	r = pci_resize_resource(adev->pdev, 0, rbar_size,
+				(adev->asic_type >= CHIP_BONAIRE) ? 1 << 5
+								  : 1 << 2);
+	if (r == -ENOSPC)
+		dev_info(adev->dev,
+			 "Not enough PCI address space for a large BAR.");
+	else if (r && r != -ENOTSUPP)
+		dev_err(adev->dev, "Problem resizing BAR0 (%d).", r);
+
+	/* When the doorbell or fb BAR isn't available we have no chance of
+	 * using the device.
+	 */
+	r = amdgpu_doorbell_init(adev);
+	if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET))
+		return -ENODEV;
+
+	pci_write_config_word(adev->pdev, PCI_COMMAND, cmd);
+
+	return 0;
 }
 
 /*
  * GPU helpers function.
  */
 /**
- * amdgpu_need_post - check if the hw need post or not
+ * amdgpu_device_need_post - check if the hw need post or not
  *
  * @adev: amdgpu_device pointer
  *
@@ -706,27 +1772,17 @@ void amdgpu_gtt_location(struct amdgpu_device *adev, struct amdgpu_mc *mc)
  * or post is needed if  hw reset is performed.
  * Returns true if need or false if not.
  */
-bool amdgpu_need_post(struct amdgpu_device *adev)
+bool amdgpu_device_need_post(struct amdgpu_device *adev)
 {
-	uint32_t reg;
+	uint32_t reg, flags;
 
-	if (adev->has_hw_reset) {
-		adev->has_hw_reset = false;
-		return true;
-	}
-	/* then check MEM_SIZE, in case the crtcs are off */
-	reg = amdgpu_asic_get_config_memsize(adev);
-
-	if ((reg != 0) && (reg != 0xffffffff))
+	if (amdgpu_sriov_vf(adev))
 		return false;
 
-	return true;
-
-}
-
-static bool amdgpu_vpost_needed(struct amdgpu_device *adev)
-{
-	if (amdgpu_sriov_vf(adev))
+	flags = amdgpu_device_get_vbios_flags(adev);
+	if (flags & AMDGPU_VBIOS_SKIP)
+		return false;
+	if ((flags & AMDGPU_VBIOS_OPTIONAL) && !adev->bios)
 		return false;
 
 	if (amdgpu_passthrough(adev)) {
@@ -738,285 +1794,175 @@ static bool amdgpu_vpost_needed(struct amdgpu_device *adev)
 		if (adev->asic_type == CHIP_FIJI) {
 			int err;
 			uint32_t fw_ver;
+
 			err = request_firmware(&adev->pm.fw, "amdgpu/fiji_smc.bin", adev->dev);
-			/* force vPost if error occured */
+			/* force vPost if error occurred */
 			if (err)
 				return true;
 
 			fw_ver = *((uint32_t *)adev->pm.fw->data + 69);
+			release_firmware(adev->pm.fw);
 			if (fw_ver < 0x00160e00)
 				return true;
 		}
 	}
-	return amdgpu_need_post(adev);
-}
 
-/**
- * amdgpu_dummy_page_init - init dummy page used by the driver
- *
- * @adev: amdgpu_device pointer
- *
- * Allocate the dummy page used by the driver (all asics).
- * This dummy page is used by the driver as a filler for gart entries
- * when pages are taken out of the GART
- * Returns 0 on sucess, -ENOMEM on failure.
- */
-int amdgpu_dummy_page_init(struct amdgpu_device *adev)
-{
-	if (adev->dummy_page.page)
-		return 0;
-	adev->dummy_page.page = alloc_page(GFP_DMA32 | GFP_KERNEL | __GFP_ZERO);
-	if (adev->dummy_page.page == NULL)
-		return -ENOMEM;
-	adev->dummy_page.addr = pci_map_page(adev->pdev, adev->dummy_page.page,
-					0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-	if (pci_dma_mapping_error(adev->pdev, adev->dummy_page.addr)) {
-		dev_err(&adev->pdev->dev, "Failed to DMA MAP the dummy page\n");
-		__free_page(adev->dummy_page.page);
-		adev->dummy_page.page = NULL;
-		return -ENOMEM;
+	/* Don't post if we need to reset whole hive on init */
+	if (adev->init_lvl->level == AMDGPU_INIT_LEVEL_MINIMAL_XGMI)
+		return false;
+
+	if (adev->has_hw_reset) {
+		adev->has_hw_reset = false;
+		return true;
 	}
-	return 0;
-}
 
-/**
- * amdgpu_dummy_page_fini - free dummy page used by the driver
- *
- * @adev: amdgpu_device pointer
- *
- * Frees the dummy page used by the driver (all asics).
- */
-void amdgpu_dummy_page_fini(struct amdgpu_device *adev)
-{
-	if (adev->dummy_page.page == NULL)
-		return;
-	pci_unmap_page(adev->pdev, adev->dummy_page.addr,
-			PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-	__free_page(adev->dummy_page.page);
-	adev->dummy_page.page = NULL;
-}
+	/* bios scratch used on CIK+ */
+	if (adev->asic_type >= CHIP_BONAIRE)
+		return amdgpu_atombios_scratch_need_asic_init(adev);
 
+	/* check MEM_SIZE for older asics */
+	reg = amdgpu_asic_get_config_memsize(adev);
 
-/* ATOM accessor methods */
-/*
- * ATOM is an interpreted byte code stored in tables in the vbios.  The
- * driver registers callbacks to access registers and the interpreter
- * in the driver parses the tables and executes then to program specific
- * actions (set display modes, asic init, etc.).  See amdgpu_atombios.c,
- * atombios.h, and atom.c
- */
+	if ((reg != 0) && (reg != 0xffffffff))
+		return false;
 
-/**
- * cail_pll_read - read PLL register
- *
- * @info: atom card_info pointer
- * @reg: PLL register offset
- *
- * Provides a PLL register accessor for the atom interpreter (r4xx+).
- * Returns the value of the PLL register.
- */
-static uint32_t cail_pll_read(struct card_info *info, uint32_t reg)
-{
-	return 0;
+	return true;
 }
 
-/**
- * cail_pll_write - write PLL register
- *
- * @info: atom card_info pointer
- * @reg: PLL register offset
- * @val: value to write to the pll register
+/*
+ * Check whether seamless boot is supported.
  *
- * Provides a PLL register accessor for the atom interpreter (r4xx+).
+ * So far we only support seamless boot on DCE 3.0 or later.
+ * If users report that it works on older ASICS as well, we may
+ * loosen this.
  */
-static void cail_pll_write(struct card_info *info, uint32_t reg, uint32_t val)
+bool amdgpu_device_seamless_boot_supported(struct amdgpu_device *adev)
 {
+	switch (amdgpu_seamless) {
+	case -1:
+		break;
+	case 1:
+		return true;
+	case 0:
+		return false;
+	default:
+		dev_err(adev->dev, "Invalid value for amdgpu.seamless: %d\n",
+			amdgpu_seamless);
+		return false;
+	}
 
-}
-
-/**
- * cail_mc_read - read MC (Memory Controller) register
- *
- * @info: atom card_info pointer
- * @reg: MC register offset
- *
- * Provides an MC register accessor for the atom interpreter (r4xx+).
- * Returns the value of the MC register.
- */
-static uint32_t cail_mc_read(struct card_info *info, uint32_t reg)
-{
-	return 0;
-}
+	if (!(adev->flags & AMD_IS_APU))
+		return false;
 
-/**
- * cail_mc_write - write MC (Memory Controller) register
- *
- * @info: atom card_info pointer
- * @reg: MC register offset
- * @val: value to write to the pll register
- *
- * Provides a MC register accessor for the atom interpreter (r4xx+).
- */
-static void cail_mc_write(struct card_info *info, uint32_t reg, uint32_t val)
-{
+	if (adev->mman.keep_stolen_vga_memory)
+		return false;
 
+	return amdgpu_ip_version(adev, DCE_HWIP, 0) >= IP_VERSION(3, 0, 0);
 }
 
-/**
- * cail_reg_write - write MMIO register
- *
- * @info: atom card_info pointer
- * @reg: MMIO register offset
- * @val: value to write to the pll register
+/*
+ * Intel hosts such as Rocket Lake, Alder Lake, Raptor Lake and Sapphire Rapids
+ * don't support dynamic speed switching. Until we have confirmation from Intel
+ * that a specific host supports it, it's safer that we keep it disabled for all.
  *
- * Provides a MMIO register accessor for the atom interpreter (r4xx+).
+ * https://edc.intel.com/content/www/us/en/design/products/platforms/details/raptor-lake-s/13th-generation-core-processors-datasheet-volume-1-of-2/005/pci-express-support/
+ * https://gitlab.freedesktop.org/drm/amd/-/issues/2663
  */
-static void cail_reg_write(struct card_info *info, uint32_t reg, uint32_t val)
+static bool amdgpu_device_pcie_dynamic_switching_supported(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = info->dev->dev_private;
+#if IS_ENABLED(CONFIG_X86)
+	struct cpuinfo_x86 *c = &cpu_data(0);
 
-	WREG32(reg, val);
-}
-
-/**
- * cail_reg_read - read MMIO register
- *
- * @info: atom card_info pointer
- * @reg: MMIO register offset
- *
- * Provides an MMIO register accessor for the atom interpreter (r4xx+).
- * Returns the value of the MMIO register.
- */
-static uint32_t cail_reg_read(struct card_info *info, uint32_t reg)
-{
-	struct amdgpu_device *adev = info->dev->dev_private;
-	uint32_t r;
+	/* eGPU change speeds based on USB4 fabric conditions */
+	if (dev_is_removable(adev->dev))
+		return true;
 
-	r = RREG32(reg);
-	return r;
+	if (c->x86_vendor == X86_VENDOR_INTEL)
+		return false;
+#endif
+	return true;
 }
 
-/**
- * cail_ioreg_write - write IO register
- *
- * @info: atom card_info pointer
- * @reg: IO register offset
- * @val: value to write to the pll register
- *
- * Provides a IO register accessor for the atom interpreter (r4xx+).
- */
-static void cail_ioreg_write(struct card_info *info, uint32_t reg, uint32_t val)
+static bool amdgpu_device_aspm_support_quirk(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = info->dev->dev_private;
-
-	WREG32_IO(reg, val);
-}
+	/* Enabling ASPM causes randoms hangs on Tahiti and Oland on Zen4.
+	 * It's unclear if this is a platform-specific or GPU-specific issue.
+	 * Disable ASPM on SI for the time being.
+	 */
+	if (adev->family == AMDGPU_FAMILY_SI)
+		return true;
 
-/**
- * cail_ioreg_read - read IO register
- *
- * @info: atom card_info pointer
- * @reg: IO register offset
- *
- * Provides an IO register accessor for the atom interpreter (r4xx+).
- * Returns the value of the IO register.
- */
-static uint32_t cail_ioreg_read(struct card_info *info, uint32_t reg)
-{
-	struct amdgpu_device *adev = info->dev->dev_private;
-	uint32_t r;
+#if IS_ENABLED(CONFIG_X86)
+	struct cpuinfo_x86 *c = &cpu_data(0);
 
-	r = RREG32_IO(reg);
-	return r;
-}
+	if (!(amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(12, 0, 0) ||
+		  amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(12, 0, 1)))
+		return false;
 
-/**
- * amdgpu_atombios_fini - free the driver info and callbacks for atombios
- *
- * @adev: amdgpu_device pointer
- *
- * Frees the driver info and register access callbacks for the ATOM
- * interpreter (r4xx+).
- * Called at driver shutdown.
- */
-static void amdgpu_atombios_fini(struct amdgpu_device *adev)
-{
-	if (adev->mode_info.atom_context) {
-		kfree(adev->mode_info.atom_context->scratch);
-		kfree(adev->mode_info.atom_context->iio);
+	if (c->x86 == 6 &&
+		adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_GEN5) {
+		switch (c->x86_model) {
+		case VFM_MODEL(INTEL_ALDERLAKE):
+		case VFM_MODEL(INTEL_ALDERLAKE_L):
+		case VFM_MODEL(INTEL_RAPTORLAKE):
+		case VFM_MODEL(INTEL_RAPTORLAKE_P):
+		case VFM_MODEL(INTEL_RAPTORLAKE_S):
+			return true;
+		default:
+			return false;
+		}
+	} else {
+		return false;
 	}
-	kfree(adev->mode_info.atom_context);
-	adev->mode_info.atom_context = NULL;
-	kfree(adev->mode_info.atom_card_info);
-	adev->mode_info.atom_card_info = NULL;
+#else
+	return false;
+#endif
 }
 
 /**
- * amdgpu_atombios_init - init the driver info and callbacks for atombios
+ * amdgpu_device_should_use_aspm - check if the device should program ASPM
  *
  * @adev: amdgpu_device pointer
  *
- * Initializes the driver info and register access callbacks for the
- * ATOM interpreter (r4xx+).
- * Returns 0 on sucess, -ENOMEM on failure.
- * Called at driver startup.
+ * Confirm whether the module parameter and pcie bridge agree that ASPM should
+ * be set for this device.
+ *
+ * Returns true if it should be used or false if not.
  */
-static int amdgpu_atombios_init(struct amdgpu_device *adev)
+bool amdgpu_device_should_use_aspm(struct amdgpu_device *adev)
 {
-	struct card_info *atom_card_info =
-	    kzalloc(sizeof(struct card_info), GFP_KERNEL);
-
-	if (!atom_card_info)
-		return -ENOMEM;
-
-	adev->mode_info.atom_card_info = atom_card_info;
-	atom_card_info->dev = adev->ddev;
-	atom_card_info->reg_read = cail_reg_read;
-	atom_card_info->reg_write = cail_reg_write;
-	/* needed for iio ops */
-	if (adev->rio_mem) {
-		atom_card_info->ioreg_read = cail_ioreg_read;
-		atom_card_info->ioreg_write = cail_ioreg_write;
-	} else {
-		DRM_INFO("PCI I/O BAR is not found. Using MMIO to access ATOM BIOS\n");
-		atom_card_info->ioreg_read = cail_reg_read;
-		atom_card_info->ioreg_write = cail_reg_write;
-	}
-	atom_card_info->mc_read = cail_mc_read;
-	atom_card_info->mc_write = cail_mc_write;
-	atom_card_info->pll_read = cail_pll_read;
-	atom_card_info->pll_write = cail_pll_write;
-
-	adev->mode_info.atom_context = amdgpu_atom_parse(atom_card_info, adev->bios);
-	if (!adev->mode_info.atom_context) {
-		amdgpu_atombios_fini(adev);
-		return -ENOMEM;
-	}
-
-	mutex_init(&adev->mode_info.atom_context->mutex);
-	if (adev->is_atom_fw) {
-		amdgpu_atomfirmware_scratch_regs_init(adev);
-		amdgpu_atomfirmware_allocate_fb_scratch(adev);
-	} else {
-		amdgpu_atombios_scratch_regs_init(adev);
-		amdgpu_atombios_allocate_fb_scratch(adev);
+	switch (amdgpu_aspm) {
+	case -1:
+		break;
+	case 0:
+		return false;
+	case 1:
+		return true;
+	default:
+		return false;
 	}
-	return 0;
+	if (adev->flags & AMD_IS_APU)
+		return false;
+	if (amdgpu_device_aspm_support_quirk(adev))
+		return false;
+	return pcie_aspm_enabled(adev->pdev);
 }
 
 /* if we get transitioned to only one device, take VGA back */
 /**
- * amdgpu_vga_set_decode - enable/disable vga decode
+ * amdgpu_device_vga_set_decode - enable/disable vga decode
  *
- * @cookie: amdgpu_device pointer
+ * @pdev: PCI device pointer
  * @state: enable/disable vga decode
  *
  * Enable/disable vga decode (all asics).
  * Returns VGA resource flags.
  */
-static unsigned int amdgpu_vga_set_decode(void *cookie, bool state)
+static unsigned int amdgpu_device_vga_set_decode(struct pci_dev *pdev,
+		bool state)
 {
-	struct amdgpu_device *adev = cookie;
+	struct amdgpu_device *adev = drm_to_adev(pci_get_drvdata(pdev));
+
 	amdgpu_asic_set_vga_state(adev, state);
 	if (state)
 		return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM |
@@ -1026,113 +1972,222 @@ static unsigned int amdgpu_vga_set_decode(void *cookie, bool state)
 }
 
 /**
- * amdgpu_check_pot_argument - check that argument is a power of two
+ * amdgpu_device_check_block_size - validate the vm block size
  *
- * @arg: value to check
+ * @adev: amdgpu_device pointer
  *
- * Validates that a certain argument is a power of two (all asics).
- * Returns true if argument is valid.
+ * Validates the vm block size specified via module parameter.
+ * The vm block size defines number of bits in page table versus page directory,
+ * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
+ * page table and the remaining bits are in the page directory.
  */
-static bool amdgpu_check_pot_argument(int arg)
-{
-	return (arg & (arg - 1)) == 0;
-}
-
-static void amdgpu_check_block_size(struct amdgpu_device *adev)
+static void amdgpu_device_check_block_size(struct amdgpu_device *adev)
 {
 	/* defines number of bits in page table versus page directory,
 	 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
-	 * page table and the remaining bits are in the page directory */
+	 * page table and the remaining bits are in the page directory
+	 */
 	if (amdgpu_vm_block_size == -1)
 		return;
 
 	if (amdgpu_vm_block_size < 9) {
 		dev_warn(adev->dev, "VM page table size (%d) too small\n",
 			 amdgpu_vm_block_size);
-		goto def_value;
-	}
-
-	if (amdgpu_vm_block_size > 24 ||
-	    (amdgpu_vm_size * 1024) < (1ull << amdgpu_vm_block_size)) {
-		dev_warn(adev->dev, "VM page table size (%d) too large\n",
-			 amdgpu_vm_block_size);
-		goto def_value;
+		amdgpu_vm_block_size = -1;
 	}
-
-	return;
-
-def_value:
-	amdgpu_vm_block_size = -1;
 }
 
-static void amdgpu_check_vm_size(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_check_vm_size - validate the vm size
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Validates the vm size in GB specified via module parameter.
+ * The VM size is the size of the GPU virtual memory space in GB.
+ */
+static void amdgpu_device_check_vm_size(struct amdgpu_device *adev)
 {
-	if (!amdgpu_check_pot_argument(amdgpu_vm_size)) {
-		dev_warn(adev->dev, "VM size (%d) must be a power of 2\n",
-			 amdgpu_vm_size);
-		goto def_value;
-	}
+	/* no need to check the default value */
+	if (amdgpu_vm_size == -1)
+		return;
 
 	if (amdgpu_vm_size < 1) {
 		dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n",
 			 amdgpu_vm_size);
-		goto def_value;
+		amdgpu_vm_size = -1;
 	}
+}
 
-	/*
-	 * Max GPUVM size for Cayman, SI, CI VI are 40 bits.
-	 */
-	if (amdgpu_vm_size > 1024) {
-		dev_warn(adev->dev, "VM size (%d) too large, max is 1TB\n",
-			 amdgpu_vm_size);
+static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev)
+{
+	struct sysinfo si;
+	bool is_os_64 = (sizeof(void *) == 8);
+	uint64_t total_memory;
+	uint64_t dram_size_seven_GB = 0x1B8000000;
+	uint64_t dram_size_three_GB = 0xB8000000;
+
+	if (amdgpu_smu_memory_pool_size == 0)
+		return;
+
+	if (!is_os_64) {
+		dev_warn(adev->dev, "Not 64-bit OS, feature not supported\n");
 		goto def_value;
 	}
+	si_meminfo(&si);
+	total_memory = (uint64_t)si.totalram * si.mem_unit;
+
+	if ((amdgpu_smu_memory_pool_size == 1) ||
+		(amdgpu_smu_memory_pool_size == 2)) {
+		if (total_memory < dram_size_three_GB)
+			goto def_value1;
+	} else if ((amdgpu_smu_memory_pool_size == 4) ||
+		(amdgpu_smu_memory_pool_size == 8)) {
+		if (total_memory < dram_size_seven_GB)
+			goto def_value1;
+	} else {
+		dev_warn(adev->dev, "Smu memory pool size not supported\n");
+		goto def_value;
+	}
+	adev->pm.smu_prv_buffer_size = amdgpu_smu_memory_pool_size << 28;
 
 	return;
 
+def_value1:
+	dev_warn(adev->dev, "No enough system memory\n");
 def_value:
-	amdgpu_vm_size = -1;
+	adev->pm.smu_prv_buffer_size = 0;
+}
+
+static int amdgpu_device_init_apu_flags(struct amdgpu_device *adev)
+{
+	if (!(adev->flags & AMD_IS_APU) ||
+	    adev->asic_type < CHIP_RAVEN)
+		return 0;
+
+	switch (adev->asic_type) {
+	case CHIP_RAVEN:
+		if (adev->pdev->device == 0x15dd)
+			adev->apu_flags |= AMD_APU_IS_RAVEN;
+		if (adev->pdev->device == 0x15d8)
+			adev->apu_flags |= AMD_APU_IS_PICASSO;
+		break;
+	case CHIP_RENOIR:
+		if ((adev->pdev->device == 0x1636) ||
+		    (adev->pdev->device == 0x164c))
+			adev->apu_flags |= AMD_APU_IS_RENOIR;
+		else
+			adev->apu_flags |= AMD_APU_IS_GREEN_SARDINE;
+		break;
+	case CHIP_VANGOGH:
+		adev->apu_flags |= AMD_APU_IS_VANGOGH;
+		break;
+	case CHIP_YELLOW_CARP:
+		break;
+	case CHIP_CYAN_SKILLFISH:
+		if ((adev->pdev->device == 0x13FE) ||
+		    (adev->pdev->device == 0x143F))
+			adev->apu_flags |= AMD_APU_IS_CYAN_SKILLFISH2;
+		break;
+	default:
+		break;
+	}
+
+	return 0;
 }
 
 /**
- * amdgpu_check_arguments - validate module params
+ * amdgpu_device_check_arguments - validate module params
  *
  * @adev: amdgpu_device pointer
  *
  * Validates certain module parameters and updates
  * the associated values used by the driver (all asics).
  */
-static void amdgpu_check_arguments(struct amdgpu_device *adev)
+static int amdgpu_device_check_arguments(struct amdgpu_device *adev)
 {
+	int i;
+
 	if (amdgpu_sched_jobs < 4) {
 		dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n",
 			 amdgpu_sched_jobs);
 		amdgpu_sched_jobs = 4;
-	} else if (!amdgpu_check_pot_argument(amdgpu_sched_jobs)){
+	} else if (!is_power_of_2(amdgpu_sched_jobs)) {
 		dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n",
 			 amdgpu_sched_jobs);
 		amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs);
 	}
 
-	if (amdgpu_gart_size != -1) {
+	if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) {
+		/* gart size must be greater or equal to 32M */
+		dev_warn(adev->dev, "gart size (%d) too small\n",
+			 amdgpu_gart_size);
+		amdgpu_gart_size = -1;
+	}
+
+	if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) {
 		/* gtt size must be greater or equal to 32M */
-		if (amdgpu_gart_size < 32) {
-			dev_warn(adev->dev, "gart size (%d) too small\n",
-				 amdgpu_gart_size);
-			amdgpu_gart_size = -1;
-		}
+		dev_warn(adev->dev, "gtt size (%d) too small\n",
+				 amdgpu_gtt_size);
+		amdgpu_gtt_size = -1;
 	}
 
-	amdgpu_check_vm_size(adev);
+	/* valid range is between 4 and 9 inclusive */
+	if (amdgpu_vm_fragment_size != -1 &&
+	    (amdgpu_vm_fragment_size > 9 || amdgpu_vm_fragment_size < 4)) {
+		dev_warn(adev->dev, "valid range is between 4 and 9\n");
+		amdgpu_vm_fragment_size = -1;
+	}
+
+	if (amdgpu_sched_hw_submission < 2) {
+		dev_warn(adev->dev, "sched hw submission jobs (%d) must be at least 2\n",
+			 amdgpu_sched_hw_submission);
+		amdgpu_sched_hw_submission = 2;
+	} else if (!is_power_of_2(amdgpu_sched_hw_submission)) {
+		dev_warn(adev->dev, "sched hw submission jobs (%d) must be a power of 2\n",
+			 amdgpu_sched_hw_submission);
+		amdgpu_sched_hw_submission = roundup_pow_of_two(amdgpu_sched_hw_submission);
+	}
+
+	if (amdgpu_reset_method < -1 || amdgpu_reset_method > 4) {
+		dev_warn(adev->dev, "invalid option for reset method, reverting to default\n");
+		amdgpu_reset_method = -1;
+	}
+
+	amdgpu_device_check_smu_prv_buffer_size(adev);
+
+	amdgpu_device_check_vm_size(adev);
+
+	amdgpu_device_check_block_size(adev);
 
-	amdgpu_check_block_size(adev);
+	adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type);
 
-	if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 ||
-	    !amdgpu_check_pot_argument(amdgpu_vram_page_split))) {
-		dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
-			 amdgpu_vram_page_split);
-		amdgpu_vram_page_split = 1024;
+	for (i = 0; i < MAX_XCP; i++) {
+		switch (amdgpu_enforce_isolation) {
+		case -1:
+		case 0:
+		default:
+			/* disable */
+			adev->enforce_isolation[i] = AMDGPU_ENFORCE_ISOLATION_DISABLE;
+			break;
+		case 1:
+			/* enable */
+			adev->enforce_isolation[i] =
+				AMDGPU_ENFORCE_ISOLATION_ENABLE;
+			break;
+		case 2:
+			/* enable legacy mode */
+			adev->enforce_isolation[i] =
+				AMDGPU_ENFORCE_ISOLATION_ENABLE_LEGACY;
+			break;
+		case 3:
+			/* enable only process isolation without submitting cleaner shader */
+			adev->enforce_isolation[i] =
+				AMDGPU_ENFORCE_ISOLATION_NO_CLEANER_SHADER;
+			break;
+		}
 	}
+
+	return 0;
 }
 
 /**
@@ -1141,34 +2196,42 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
  * @pdev: pci dev pointer
  * @state: vga_switcheroo state
  *
- * Callback for the switcheroo driver.  Suspends or resumes the
+ * Callback for the switcheroo driver.  Suspends or resumes
  * the asics before or after it is powered up using ACPI methods.
  */
-static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_state state)
+static void amdgpu_switcheroo_set_state(struct pci_dev *pdev,
+					enum vga_switcheroo_state state)
 {
 	struct drm_device *dev = pci_get_drvdata(pdev);
+	int r;
 
-	if (amdgpu_device_is_px(dev) && state == VGA_SWITCHEROO_OFF)
+	if (amdgpu_device_supports_px(drm_to_adev(dev)) &&
+	    state == VGA_SWITCHEROO_OFF)
 		return;
 
 	if (state == VGA_SWITCHEROO_ON) {
-		unsigned d3_delay = dev->pdev->d3_delay;
-
-		pr_info("amdgpu: switched on\n");
+		pr_info("switched on\n");
 		/* don't suspend or resume card normally */
 		dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
 
-		amdgpu_device_resume(dev, true, true);
-
-		dev->pdev->d3_delay = d3_delay;
+		pci_set_power_state(pdev, PCI_D0);
+		amdgpu_device_load_pci_state(pdev);
+		r = pci_enable_device(pdev);
+		if (r)
+			dev_warn(&pdev->dev, "pci_enable_device failed (%d)\n",
+				 r);
+		amdgpu_device_resume(dev, true);
 
 		dev->switch_power_state = DRM_SWITCH_POWER_ON;
-		drm_kms_helper_poll_enable(dev);
 	} else {
-		pr_info("amdgpu: switched off\n");
-		drm_kms_helper_poll_disable(dev);
+		dev_info(&pdev->dev, "switched off\n");
 		dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
-		amdgpu_device_suspend(dev, true, true);
+		amdgpu_device_prepare(dev);
+		amdgpu_device_suspend(dev, true);
+		amdgpu_device_cache_pci_state(pdev);
+		/* Shut down the device */
+		pci_disable_device(pdev);
+		pci_set_power_state(pdev, PCI_D3cold);
 		dev->switch_power_state = DRM_SWITCH_POWER_OFF;
 	}
 }
@@ -1186,12 +2249,12 @@ static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev)
 {
 	struct drm_device *dev = pci_get_drvdata(pdev);
 
-	/*
+       /*
 	* FIXME: open_count is protected by drm_global_mutex but that would lead to
 	* locking inversion with the driver load path. And the access here is
 	* completely racy anyway. So don't bother with locking for now.
 	*/
-	return dev->open_count == 0;
+	return atomic_read(&dev->open_count) == 0;
 }
 
 static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = {
@@ -1200,10 +2263,22 @@ static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = {
 	.can_switch = amdgpu_switcheroo_can_switch,
 };
 
-int amdgpu_set_clockgating_state(struct amdgpu_device *adev,
-				  enum amd_ip_block_type block_type,
-				  enum amd_clockgating_state state)
+/**
+ * amdgpu_device_ip_set_clockgating_state - set the CG state
+ *
+ * @dev: amdgpu_device pointer
+ * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
+ * @state: clockgating state (gate or ungate)
+ *
+ * Sets the requested clockgating state for all instances of
+ * the hardware IP specified.
+ * Returns the error code from the last instance.
+ */
+int amdgpu_device_ip_set_clockgating_state(void *dev,
+					   enum amd_ip_block_type block_type,
+					   enum amd_clockgating_state state)
 {
+	struct amdgpu_device *adev = dev;
 	int i, r = 0;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
@@ -1214,18 +2289,31 @@ int amdgpu_set_clockgating_state(struct amdgpu_device *adev,
 		if (!adev->ip_blocks[i].version->funcs->set_clockgating_state)
 			continue;
 		r = adev->ip_blocks[i].version->funcs->set_clockgating_state(
-			(void *)adev, state);
+			&adev->ip_blocks[i], state);
 		if (r)
-			DRM_ERROR("set_clockgating_state of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
+			dev_err(adev->dev,
+				"set_clockgating_state of IP block <%s> failed %d\n",
+				adev->ip_blocks[i].version->funcs->name, r);
 	}
 	return r;
 }
 
-int amdgpu_set_powergating_state(struct amdgpu_device *adev,
-				  enum amd_ip_block_type block_type,
-				  enum amd_powergating_state state)
+/**
+ * amdgpu_device_ip_set_powergating_state - set the PG state
+ *
+ * @dev: amdgpu_device pointer
+ * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
+ * @state: powergating state (gate or ungate)
+ *
+ * Sets the requested powergating state for all instances of
+ * the hardware IP specified.
+ * Returns the error code from the last instance.
+ */
+int amdgpu_device_ip_set_powergating_state(void *dev,
+					   enum amd_ip_block_type block_type,
+					   enum amd_powergating_state state)
 {
+	struct amdgpu_device *adev = dev;
 	int i, r = 0;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
@@ -1236,15 +2324,28 @@ int amdgpu_set_powergating_state(struct amdgpu_device *adev,
 		if (!adev->ip_blocks[i].version->funcs->set_powergating_state)
 			continue;
 		r = adev->ip_blocks[i].version->funcs->set_powergating_state(
-			(void *)adev, state);
+			&adev->ip_blocks[i], state);
 		if (r)
-			DRM_ERROR("set_powergating_state of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
+			dev_err(adev->dev,
+				"set_powergating_state of IP block <%s> failed %d\n",
+				adev->ip_blocks[i].version->funcs->name, r);
 	}
 	return r;
 }
 
-void amdgpu_get_clockgating_state(struct amdgpu_device *adev, u32 *flags)
+/**
+ * amdgpu_device_ip_get_clockgating_state - get the CG state
+ *
+ * @adev: amdgpu_device pointer
+ * @flags: clockgating feature flags
+ *
+ * Walks the list of IPs on the device and updates the clockgating
+ * flags for each IP.
+ * Updates @flags with the feature flags for each hardware IP where
+ * clockgating is enabled.
+ */
+void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev,
+					    u64 *flags)
 {
 	int i;
 
@@ -1252,12 +2353,22 @@ void amdgpu_get_clockgating_state(struct amdgpu_device *adev, u32 *flags)
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
 		if (adev->ip_blocks[i].version->funcs->get_clockgating_state)
-			adev->ip_blocks[i].version->funcs->get_clockgating_state((void *)adev, flags);
+			adev->ip_blocks[i].version->funcs->get_clockgating_state(
+				&adev->ip_blocks[i], flags);
 	}
 }
 
-int amdgpu_wait_for_idle(struct amdgpu_device *adev,
-			 enum amd_ip_block_type block_type)
+/**
+ * amdgpu_device_ip_wait_for_idle - wait for idle
+ *
+ * @adev: amdgpu_device pointer
+ * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
+ *
+ * Waits for the request hardware IP to be idle.
+ * Returns 0 for success or a negative error code on failure.
+ */
+int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev,
+				   enum amd_ip_block_type block_type)
 {
 	int i, r;
 
@@ -1265,9 +2376,12 @@ int amdgpu_wait_for_idle(struct amdgpu_device *adev,
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
 		if (adev->ip_blocks[i].version->type == block_type) {
-			r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev);
-			if (r)
-				return r;
+			if (adev->ip_blocks[i].version->funcs->wait_for_idle) {
+				r = adev->ip_blocks[i].version->funcs->wait_for_idle(
+								&adev->ip_blocks[i]);
+				if (r)
+					return r;
+			}
 			break;
 		}
 	}
@@ -1275,23 +2389,61 @@ int amdgpu_wait_for_idle(struct amdgpu_device *adev,
 
 }
 
-bool amdgpu_is_idle(struct amdgpu_device *adev,
-		    enum amd_ip_block_type block_type)
+/**
+ * amdgpu_device_ip_is_hw - is the hardware IP enabled
+ *
+ * @adev: amdgpu_device pointer
+ * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
+ *
+ * Check if the hardware IP is enable or not.
+ * Returns true if it the IP is enable, false if not.
+ */
+bool amdgpu_device_ip_is_hw(struct amdgpu_device *adev,
+			    enum amd_ip_block_type block_type)
 {
 	int i;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
-			continue;
 		if (adev->ip_blocks[i].version->type == block_type)
-			return adev->ip_blocks[i].version->funcs->is_idle((void *)adev);
+			return adev->ip_blocks[i].status.hw;
 	}
-	return true;
+	return false;
+}
+
+/**
+ * amdgpu_device_ip_is_valid - is the hardware IP valid
+ *
+ * @adev: amdgpu_device pointer
+ * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
+ *
+ * Check if the hardware IP is valid or not.
+ * Returns true if it the IP is valid, false if not.
+ */
+bool amdgpu_device_ip_is_valid(struct amdgpu_device *adev,
+			       enum amd_ip_block_type block_type)
+{
+	int i;
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (adev->ip_blocks[i].version->type == block_type)
+			return adev->ip_blocks[i].status.valid;
+	}
+	return false;
 
 }
 
-struct amdgpu_ip_block * amdgpu_get_ip_block(struct amdgpu_device *adev,
-					     enum amd_ip_block_type type)
+/**
+ * amdgpu_device_ip_get_ip_block - get a hw IP pointer
+ *
+ * @adev: amdgpu_device pointer
+ * @type: Type of hardware IP (SMU, GFX, UVD, etc.)
+ *
+ * Returns a pointer to the hardware IP block structure
+ * if it exists for the asic, otherwise NULL.
+ */
+struct amdgpu_ip_block *
+amdgpu_device_ip_get_ip_block(struct amdgpu_device *adev,
+			      enum amd_ip_block_type type)
 {
 	int i;
 
@@ -1303,7 +2455,7 @@ struct amdgpu_ip_block * amdgpu_get_ip_block(struct amdgpu_device *adev,
 }
 
 /**
- * amdgpu_ip_block_version_cmp
+ * amdgpu_device_ip_block_version_cmp
  *
  * @adev: amdgpu_device pointer
  * @type: enum amd_ip_block_type
@@ -1313,11 +2465,11 @@ struct amdgpu_ip_block * amdgpu_get_ip_block(struct amdgpu_device *adev,
  * return 0 if equal or greater
  * return 1 if smaller or the ip_block doesn't exist
  */
-int amdgpu_ip_block_version_cmp(struct amdgpu_device *adev,
-				enum amd_ip_block_type type,
-				u32 major, u32 minor)
+int amdgpu_device_ip_block_version_cmp(struct amdgpu_device *adev,
+				       enum amd_ip_block_type type,
+				       u32 major, u32 minor)
 {
-	struct amdgpu_ip_block *ip_block = amdgpu_get_ip_block(adev, type);
+	struct amdgpu_ip_block *ip_block = amdgpu_device_ip_get_ip_block(adev, type);
 
 	if (ip_block && ((ip_block->version->major > major) ||
 			((ip_block->version->major == major) &&
@@ -1327,8 +2479,36 @@ int amdgpu_ip_block_version_cmp(struct amdgpu_device *adev,
 	return 1;
 }
 
+static const char *ip_block_names[] = {
+	[AMD_IP_BLOCK_TYPE_COMMON] = "common",
+	[AMD_IP_BLOCK_TYPE_GMC] = "gmc",
+	[AMD_IP_BLOCK_TYPE_IH] = "ih",
+	[AMD_IP_BLOCK_TYPE_SMC] = "smu",
+	[AMD_IP_BLOCK_TYPE_PSP] = "psp",
+	[AMD_IP_BLOCK_TYPE_DCE] = "dce",
+	[AMD_IP_BLOCK_TYPE_GFX] = "gfx",
+	[AMD_IP_BLOCK_TYPE_SDMA] = "sdma",
+	[AMD_IP_BLOCK_TYPE_UVD] = "uvd",
+	[AMD_IP_BLOCK_TYPE_VCE] = "vce",
+	[AMD_IP_BLOCK_TYPE_ACP] = "acp",
+	[AMD_IP_BLOCK_TYPE_VCN] = "vcn",
+	[AMD_IP_BLOCK_TYPE_MES] = "mes",
+	[AMD_IP_BLOCK_TYPE_JPEG] = "jpeg",
+	[AMD_IP_BLOCK_TYPE_VPE] = "vpe",
+	[AMD_IP_BLOCK_TYPE_UMSCH_MM] = "umsch_mm",
+	[AMD_IP_BLOCK_TYPE_ISP] = "isp",
+	[AMD_IP_BLOCK_TYPE_RAS] = "ras",
+};
+
+static const char *ip_block_name(struct amdgpu_device *adev, enum amd_ip_block_type type)
+{
+	int idx = (int)type;
+
+	return idx < ARRAY_SIZE(ip_block_names) ? ip_block_names[idx] : "unknown";
+}
+
 /**
- * amdgpu_ip_block_add
+ * amdgpu_device_ip_block_add
  *
  * @adev: amdgpu_device pointer
  * @ip_block_version: pointer to the IP to add
@@ -1336,24 +2516,58 @@ int amdgpu_ip_block_version_cmp(struct amdgpu_device *adev,
  * Adds the IP block driver information to the collection of IPs
  * on the asic.
  */
-int amdgpu_ip_block_add(struct amdgpu_device *adev,
-			const struct amdgpu_ip_block_version *ip_block_version)
+int amdgpu_device_ip_block_add(struct amdgpu_device *adev,
+			       const struct amdgpu_ip_block_version *ip_block_version)
 {
 	if (!ip_block_version)
 		return -EINVAL;
 
+	switch (ip_block_version->type) {
+	case AMD_IP_BLOCK_TYPE_VCN:
+		if (adev->harvest_ip_mask & AMD_HARVEST_IP_VCN_MASK)
+			return 0;
+		break;
+	case AMD_IP_BLOCK_TYPE_JPEG:
+		if (adev->harvest_ip_mask & AMD_HARVEST_IP_JPEG_MASK)
+			return 0;
+		break;
+	default:
+		break;
+	}
+
+	dev_info(adev->dev, "detected ip block number %d <%s_v%d_%d_%d> (%s)\n",
+		 adev->num_ip_blocks,
+		 ip_block_name(adev, ip_block_version->type),
+		 ip_block_version->major,
+		 ip_block_version->minor,
+		 ip_block_version->rev,
+		 ip_block_version->funcs->name);
+
+	adev->ip_blocks[adev->num_ip_blocks].adev = adev;
+
 	adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version;
 
 	return 0;
 }
 
+/**
+ * amdgpu_device_enable_virtual_display - enable virtual display feature
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Enabled the virtual display feature if the user has enabled it via
+ * the module parameter virtual_display.  This feature provides a virtual
+ * display hardware on headless boards or in virtualized environments.
+ * This function parses and validates the configuration string specified by
+ * the user and configures the virtual display configuration (number of
+ * virtual connectors, crtcs, etc.) specified.
+ */
 static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev)
 {
 	adev->enable_virtual_display = false;
 
 	if (amdgpu_virtual_display) {
-		struct drm_device *ddev = adev->ddev;
-		const char *pci_address_name = pci_name(ddev->pdev);
+		const char *pci_address_name = pci_name(adev->pdev);
 		char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname;
 
 		pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL);
@@ -1384,38 +2598,203 @@ static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev)
 			}
 		}
 
-		DRM_INFO("virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n",
-			 amdgpu_virtual_display, pci_address_name,
-			 adev->enable_virtual_display, adev->mode_info.num_crtc);
+		dev_info(
+			adev->dev,
+			"virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n",
+			amdgpu_virtual_display, pci_address_name,
+			adev->enable_virtual_display, adev->mode_info.num_crtc);
 
 		kfree(pciaddstr);
 	}
 }
 
-static int amdgpu_early_init(struct amdgpu_device *adev)
+void amdgpu_device_set_sriov_virtual_display(struct amdgpu_device *adev)
 {
-	int i, r;
+	if (amdgpu_sriov_vf(adev) && !adev->enable_virtual_display) {
+		adev->mode_info.num_crtc = 1;
+		adev->enable_virtual_display = true;
+		dev_info(adev->dev, "virtual_display:%d, num_crtc:%d\n",
+			 adev->enable_virtual_display,
+			 adev->mode_info.num_crtc);
+	}
+}
 
-	amdgpu_device_enable_virtual_display(adev);
+/**
+ * amdgpu_device_parse_gpu_info_fw - parse gpu info firmware
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Parses the asic configuration parameters specified in the gpu info
+ * firmware and makes them available to the driver for use in configuring
+ * the asic.
+ * Returns 0 on success, -EINVAL on failure.
+ */
+static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev)
+{
+	const char *chip_name;
+	int err;
+	const struct gpu_info_firmware_header_v1_0 *hdr;
+
+	adev->firmware.gpu_info_fw = NULL;
 
 	switch (adev->asic_type) {
-	case CHIP_TOPAZ:
-	case CHIP_TONGA:
-	case CHIP_FIJI:
-	case CHIP_POLARIS11:
-	case CHIP_POLARIS10:
-	case CHIP_POLARIS12:
-	case CHIP_CARRIZO:
-	case CHIP_STONEY:
-		if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY)
-			adev->family = AMDGPU_FAMILY_CZ;
+	default:
+		return 0;
+	case CHIP_VEGA10:
+		chip_name = "vega10";
+		break;
+	case CHIP_VEGA12:
+		chip_name = "vega12";
+		break;
+	case CHIP_RAVEN:
+		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
+			chip_name = "raven2";
+		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
+			chip_name = "picasso";
 		else
-			adev->family = AMDGPU_FAMILY_VI;
+			chip_name = "raven";
+		break;
+	case CHIP_ARCTURUS:
+		chip_name = "arcturus";
+		break;
+	case CHIP_NAVI12:
+		if (adev->discovery.bin)
+			return 0;
+		chip_name = "navi12";
+		break;
+	case CHIP_CYAN_SKILLFISH:
+		if (adev->discovery.bin)
+			return 0;
+		chip_name = "cyan_skillfish";
+		break;
+	}
 
-		r = vi_set_ip_blocks(adev);
+	err = amdgpu_ucode_request(adev, &adev->firmware.gpu_info_fw,
+				   AMDGPU_UCODE_OPTIONAL,
+				   "amdgpu/%s_gpu_info.bin", chip_name);
+	if (err) {
+		dev_err(adev->dev,
+			"Failed to get gpu_info firmware \"%s_gpu_info.bin\"\n",
+			chip_name);
+		goto out;
+	}
+
+	hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data;
+	amdgpu_ucode_print_gpu_info_hdr(&hdr->header);
+
+	switch (hdr->version_major) {
+	case 1:
+	{
+		const struct gpu_info_firmware_v1_0 *gpu_info_fw =
+			(const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data +
+								le32_to_cpu(hdr->header.ucode_array_offset_bytes));
+
+		/*
+		 * Should be dropped when DAL no longer needs it.
+		 */
+		if (adev->asic_type == CHIP_NAVI12)
+			goto parse_soc_bounding_box;
+
+		adev->gfx.config.max_shader_engines = le32_to_cpu(gpu_info_fw->gc_num_se);
+		adev->gfx.config.max_cu_per_sh = le32_to_cpu(gpu_info_fw->gc_num_cu_per_sh);
+		adev->gfx.config.max_sh_per_se = le32_to_cpu(gpu_info_fw->gc_num_sh_per_se);
+		adev->gfx.config.max_backends_per_se = le32_to_cpu(gpu_info_fw->gc_num_rb_per_se);
+		adev->gfx.config.max_texture_channel_caches =
+			le32_to_cpu(gpu_info_fw->gc_num_tccs);
+		adev->gfx.config.max_gprs = le32_to_cpu(gpu_info_fw->gc_num_gprs);
+		adev->gfx.config.max_gs_threads = le32_to_cpu(gpu_info_fw->gc_num_max_gs_thds);
+		adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gpu_info_fw->gc_gs_table_depth);
+		adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gpu_info_fw->gc_gsprim_buff_depth);
+		adev->gfx.config.double_offchip_lds_buf =
+			le32_to_cpu(gpu_info_fw->gc_double_offchip_lds_buffer);
+		adev->gfx.cu_info.wave_front_size = le32_to_cpu(gpu_info_fw->gc_wave_size);
+		adev->gfx.cu_info.max_waves_per_simd =
+			le32_to_cpu(gpu_info_fw->gc_max_waves_per_simd);
+		adev->gfx.cu_info.max_scratch_slots_per_cu =
+			le32_to_cpu(gpu_info_fw->gc_max_scratch_slots_per_cu);
+		adev->gfx.cu_info.lds_size = le32_to_cpu(gpu_info_fw->gc_lds_size);
+		if (hdr->version_minor >= 1) {
+			const struct gpu_info_firmware_v1_1 *gpu_info_fw =
+				(const struct gpu_info_firmware_v1_1 *)(adev->firmware.gpu_info_fw->data +
+									le32_to_cpu(hdr->header.ucode_array_offset_bytes));
+			adev->gfx.config.num_sc_per_sh =
+				le32_to_cpu(gpu_info_fw->num_sc_per_sh);
+			adev->gfx.config.num_packer_per_sc =
+				le32_to_cpu(gpu_info_fw->num_packer_per_sc);
+		}
+
+parse_soc_bounding_box:
+		/*
+		 * soc bounding box info is not integrated in disocovery table,
+		 * we always need to parse it from gpu info firmware if needed.
+		 */
+		if (hdr->version_minor == 2) {
+			const struct gpu_info_firmware_v1_2 *gpu_info_fw =
+				(const struct gpu_info_firmware_v1_2 *)(adev->firmware.gpu_info_fw->data +
+									le32_to_cpu(hdr->header.ucode_array_offset_bytes));
+			adev->dm.soc_bounding_box = &gpu_info_fw->soc_bounding_box;
+		}
+		break;
+	}
+	default:
+		dev_err(adev->dev,
+			"Unsupported gpu_info table %d\n", hdr->header.ucode_version);
+		err = -EINVAL;
+		goto out;
+	}
+out:
+	return err;
+}
+
+static void amdgpu_uid_init(struct amdgpu_device *adev)
+{
+	/* Initialize the UID for the device */
+	adev->uid_info = kzalloc(sizeof(struct amdgpu_uid), GFP_KERNEL);
+	if (!adev->uid_info) {
+		dev_warn(adev->dev, "Failed to allocate memory for UID\n");
+		return;
+	}
+	adev->uid_info->adev = adev;
+}
+
+static void amdgpu_uid_fini(struct amdgpu_device *adev)
+{
+	/* Free the UID memory */
+	kfree(adev->uid_info);
+	adev->uid_info = NULL;
+}
+
+/**
+ * amdgpu_device_ip_early_init - run early init for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Early initialization pass for hardware IPs.  The hardware IPs that make
+ * up each asic are discovered each IP's early_init callback is run.  This
+ * is the first stage in initializing the asic.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_ip_block *ip_block;
+	struct pci_dev *parent;
+	bool total, skip_bios;
+	uint32_t bios_flags;
+	int i, r;
+
+	amdgpu_device_enable_virtual_display(adev);
+
+	if (amdgpu_sriov_vf(adev)) {
+		r = amdgpu_virt_request_full_gpu(adev, true);
 		if (r)
 			return r;
-		break;
+
+		r = amdgpu_virt_init_critical_region(adev);
+		if (r)
+			return r;
+	}
+
+	switch (adev->asic_type) {
 #ifdef CONFIG_DRM_AMDGPU_SI
 	case CHIP_VERDE:
 	case CHIP_TAHITI:
@@ -1434,55 +2813,149 @@ static int amdgpu_early_init(struct amdgpu_device *adev)
 	case CHIP_KAVERI:
 	case CHIP_KABINI:
 	case CHIP_MULLINS:
-		if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII))
-			adev->family = AMDGPU_FAMILY_CI;
-		else
+		if (adev->flags & AMD_IS_APU)
 			adev->family = AMDGPU_FAMILY_KV;
+		else
+			adev->family = AMDGPU_FAMILY_CI;
 
 		r = cik_set_ip_blocks(adev);
 		if (r)
 			return r;
 		break;
 #endif
-	case CHIP_VEGA10:
-		adev->family = AMDGPU_FAMILY_AI;
+	case CHIP_TOPAZ:
+	case CHIP_TONGA:
+	case CHIP_FIJI:
+	case CHIP_POLARIS10:
+	case CHIP_POLARIS11:
+	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
+	case CHIP_CARRIZO:
+	case CHIP_STONEY:
+		if (adev->flags & AMD_IS_APU)
+			adev->family = AMDGPU_FAMILY_CZ;
+		else
+			adev->family = AMDGPU_FAMILY_VI;
 
-		r = soc15_set_ip_blocks(adev);
+		r = vi_set_ip_blocks(adev);
 		if (r)
 			return r;
 		break;
 	default:
-		/* FIXME: not supported yet */
-		return -EINVAL;
+		r = amdgpu_discovery_set_ip_blocks(adev);
+		if (r)
+			return r;
+		break;
+	}
+
+	/* Check for IP version 9.4.3 with A0 hardware */
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) &&
+	    !amdgpu_device_get_rev_id(adev)) {
+		dev_err(adev->dev, "Unsupported A0 hardware\n");
+		return -ENODEV;	/* device unsupported - no device error */
+	}
+
+	if (amdgpu_has_atpx() &&
+	    (amdgpu_is_atpx_hybrid() ||
+	     amdgpu_has_atpx_dgpu_power_cntl()) &&
+	    ((adev->flags & AMD_IS_APU) == 0) &&
+	    !dev_is_removable(&adev->pdev->dev))
+		adev->flags |= AMD_IS_PX;
+
+	if (!(adev->flags & AMD_IS_APU)) {
+		parent = pcie_find_root_port(adev->pdev);
+		adev->has_pr3 = parent ? pci_pr3_present(parent) : false;
 	}
 
+	adev->pm.pp_feature = amdgpu_pp_feature_mask;
+	if (amdgpu_sriov_vf(adev) || sched_policy == KFD_SCHED_POLICY_NO_HWS)
+		adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
+	if (amdgpu_sriov_vf(adev) && adev->asic_type == CHIP_SIENNA_CICHLID)
+		adev->pm.pp_feature &= ~PP_OVERDRIVE_MASK;
+	if (!amdgpu_device_pcie_dynamic_switching_supported(adev))
+		adev->pm.pp_feature &= ~PP_PCIE_DPM_MASK;
+
+	adev->virt.is_xgmi_node_migrate_enabled = false;
 	if (amdgpu_sriov_vf(adev)) {
-		r = amdgpu_virt_request_full_gpu(adev, true);
-		if (r)
-			return r;
+		adev->virt.is_xgmi_node_migrate_enabled =
+			amdgpu_ip_version((adev), GC_HWIP, 0) == IP_VERSION(9, 4, 4);
 	}
 
+	total = true;
 	for (i = 0; i < adev->num_ip_blocks; i++) {
+		ip_block = &adev->ip_blocks[i];
+
 		if ((amdgpu_ip_block_mask & (1 << i)) == 0) {
-			DRM_ERROR("disabled ip block: %d\n", i);
+			dev_warn(adev->dev, "disabled ip block: %d <%s>\n", i,
+				 adev->ip_blocks[i].version->funcs->name);
 			adev->ip_blocks[i].status.valid = false;
-		} else {
-			if (adev->ip_blocks[i].version->funcs->early_init) {
-				r = adev->ip_blocks[i].version->funcs->early_init((void *)adev);
-				if (r == -ENOENT) {
-					adev->ip_blocks[i].status.valid = false;
-				} else if (r) {
-					DRM_ERROR("early_init of IP block <%s> failed %d\n",
-						  adev->ip_blocks[i].version->funcs->name, r);
-					return r;
-				} else {
-					adev->ip_blocks[i].status.valid = true;
-				}
+		} else if (ip_block->version->funcs->early_init) {
+			r = ip_block->version->funcs->early_init(ip_block);
+			if (r == -ENOENT) {
+				adev->ip_blocks[i].status.valid = false;
+			} else if (r) {
+				dev_err(adev->dev,
+					"early_init of IP block <%s> failed %d\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
+				total = false;
 			} else {
 				adev->ip_blocks[i].status.valid = true;
 			}
+		} else {
+			adev->ip_blocks[i].status.valid = true;
+		}
+		/* get the vbios after the asic_funcs are set up */
+		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON) {
+			r = amdgpu_device_parse_gpu_info_fw(adev);
+			if (r)
+				return r;
+
+			bios_flags = amdgpu_device_get_vbios_flags(adev);
+			skip_bios = !!(bios_flags & AMDGPU_VBIOS_SKIP);
+			/* Read BIOS */
+			if (!skip_bios) {
+				bool optional =
+					!!(bios_flags & AMDGPU_VBIOS_OPTIONAL);
+				if (!amdgpu_get_bios(adev) && !optional)
+					return -EINVAL;
+
+				if (optional && !adev->bios)
+					dev_info(
+						adev->dev,
+						"VBIOS image optional, proceeding without VBIOS image");
+
+				if (adev->bios) {
+					r = amdgpu_atombios_init(adev);
+					if (r) {
+						dev_err(adev->dev,
+							"amdgpu_atombios_init failed\n");
+						amdgpu_vf_error_put(
+							adev,
+							AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL,
+							0, 0);
+						return r;
+					}
+				}
+			}
+
+			/*get pf2vf msg info at it's earliest time*/
+			if (amdgpu_sriov_vf(adev))
+				amdgpu_virt_init_data_exchange(adev);
+
 		}
 	}
+	if (!total)
+		return -ENODEV;
+
+	if (adev->gmc.xgmi.supported)
+		amdgpu_xgmi_early_init(adev);
+
+	if (amdgpu_is_multi_aid(adev))
+		amdgpu_uid_init(adev);
+	ip_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_GFX);
+	if (ip_block->status.valid != false)
+		amdgpu_amdkfd_device_probe(adev);
 
 	adev->cg_flags &= amdgpu_cg_mask;
 	adev->pg_flags &= amdgpu_pg_mask;
@@ -1490,93 +2963,463 @@ static int amdgpu_early_init(struct amdgpu_device *adev)
 	return 0;
 }
 
-static int amdgpu_init(struct amdgpu_device *adev)
+static int amdgpu_device_ip_hw_init_phase1(struct amdgpu_device *adev)
 {
 	int i, r;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
+		if (!adev->ip_blocks[i].status.sw)
 			continue;
-		r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev);
-		if (r) {
-			DRM_ERROR("sw_init of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
-			return r;
-		}
-		adev->ip_blocks[i].status.sw = true;
-		/* need to do gmc hw init early so we can allocate gpu mem */
-		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
-			r = amdgpu_vram_scratch_init(adev);
-			if (r) {
-				DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r);
-				return r;
-			}
-			r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev);
-			if (r) {
-				DRM_ERROR("hw_init %d failed %d\n", i, r);
-				return r;
-			}
-			r = amdgpu_wb_init(adev);
+		if (adev->ip_blocks[i].status.hw)
+			continue;
+		if (!amdgpu_ip_member_of_hwini(
+			    adev, adev->ip_blocks[i].version->type))
+			continue;
+		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
+		    (amdgpu_sriov_vf(adev) && (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP)) ||
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) {
+			r = adev->ip_blocks[i].version->funcs->hw_init(&adev->ip_blocks[i]);
 			if (r) {
-				DRM_ERROR("amdgpu_wb_init failed %d\n", r);
+				dev_err(adev->dev,
+					"hw_init of IP block <%s> failed %d\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
 				return r;
 			}
 			adev->ip_blocks[i].status.hw = true;
+		}
+	}
 
-			/* right after GMC hw init, we create CSA */
-			if (amdgpu_sriov_vf(adev)) {
-				r = amdgpu_allocate_static_csa(adev);
+	return 0;
+}
+
+static int amdgpu_device_ip_hw_init_phase2(struct amdgpu_device *adev)
+{
+	int i, r;
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!adev->ip_blocks[i].status.sw)
+			continue;
+		if (adev->ip_blocks[i].status.hw)
+			continue;
+		if (!amdgpu_ip_member_of_hwini(
+			    adev, adev->ip_blocks[i].version->type))
+			continue;
+		r = adev->ip_blocks[i].version->funcs->hw_init(&adev->ip_blocks[i]);
+		if (r) {
+			dev_err(adev->dev,
+				"hw_init of IP block <%s> failed %d\n",
+				adev->ip_blocks[i].version->funcs->name, r);
+			return r;
+		}
+		adev->ip_blocks[i].status.hw = true;
+	}
+
+	return 0;
+}
+
+static int amdgpu_device_fw_loading(struct amdgpu_device *adev)
+{
+	int r = 0;
+	int i;
+	uint32_t smu_version;
+
+	if (adev->asic_type >= CHIP_VEGA10) {
+		for (i = 0; i < adev->num_ip_blocks; i++) {
+			if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_PSP)
+				continue;
+
+			if (!amdgpu_ip_member_of_hwini(adev,
+						       AMD_IP_BLOCK_TYPE_PSP))
+				break;
+
+			if (!adev->ip_blocks[i].status.sw)
+				continue;
+
+			/* no need to do the fw loading again if already done*/
+			if (adev->ip_blocks[i].status.hw == true)
+				break;
+
+			if (amdgpu_in_reset(adev) || adev->in_suspend) {
+				r = amdgpu_ip_block_resume(&adev->ip_blocks[i]);
+				if (r)
+					return r;
+			} else {
+				r = adev->ip_blocks[i].version->funcs->hw_init(&adev->ip_blocks[i]);
 				if (r) {
-					DRM_ERROR("allocate CSA failed %d\n", r);
+					dev_err(adev->dev,
+						"hw_init of IP block <%s> failed %d\n",
+						adev->ip_blocks[i]
+							.version->funcs->name,
+						r);
 					return r;
 				}
+				adev->ip_blocks[i].status.hw = true;
 			}
+			break;
 		}
 	}
 
-	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.sw)
-			continue;
-		/* gmc hw init is done early */
-		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC)
+	if (!amdgpu_sriov_vf(adev) || adev->asic_type == CHIP_TONGA)
+		r = amdgpu_pm_load_smu_firmware(adev, &smu_version);
+
+	return r;
+}
+
+static int amdgpu_device_init_schedulers(struct amdgpu_device *adev)
+{
+	struct drm_sched_init_args args = {
+		.ops = &amdgpu_sched_ops,
+		.num_rqs = DRM_SCHED_PRIORITY_COUNT,
+		.timeout_wq = adev->reset_domain->wq,
+		.dev = adev->dev,
+	};
+	long timeout;
+	int r, i;
+
+	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
+		struct amdgpu_ring *ring = adev->rings[i];
+
+		/* No need to setup the GPU scheduler for rings that don't need it */
+		if (!ring || ring->no_scheduler)
 			continue;
-		r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev);
+
+		switch (ring->funcs->type) {
+		case AMDGPU_RING_TYPE_GFX:
+			timeout = adev->gfx_timeout;
+			break;
+		case AMDGPU_RING_TYPE_COMPUTE:
+			timeout = adev->compute_timeout;
+			break;
+		case AMDGPU_RING_TYPE_SDMA:
+			timeout = adev->sdma_timeout;
+			break;
+		default:
+			timeout = adev->video_timeout;
+			break;
+		}
+
+		args.timeout = timeout;
+		args.credit_limit = ring->num_hw_submission;
+		args.score = ring->sched_score;
+		args.name = ring->name;
+
+		r = drm_sched_init(&ring->sched, &args);
 		if (r) {
-			DRM_ERROR("hw_init of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
+			dev_err(adev->dev,
+				"Failed to create scheduler on ring %s.\n",
+				ring->name);
+			return r;
+		}
+		r = amdgpu_uvd_entity_init(adev, ring);
+		if (r) {
+			dev_err(adev->dev,
+				"Failed to create UVD scheduling entity on ring %s.\n",
+				ring->name);
+			return r;
+		}
+		r = amdgpu_vce_entity_init(adev, ring);
+		if (r) {
+			dev_err(adev->dev,
+				"Failed to create VCE scheduling entity on ring %s.\n",
+				ring->name);
 			return r;
 		}
-		adev->ip_blocks[i].status.hw = true;
 	}
 
+	if (adev->xcp_mgr)
+		amdgpu_xcp_update_partition_sched_list(adev);
+
 	return 0;
 }
 
-static int amdgpu_late_init(struct amdgpu_device *adev)
+
+/**
+ * amdgpu_device_ip_init - run init for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Main initialization pass for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked and the sw_init and hw_init callbacks
+ * are run.  sw_init initializes the software state associated with each IP
+ * and hw_init initializes the hardware associated with each IP.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 {
-	int i = 0, r;
+	bool init_badpage;
+	int i, r;
+
+	r = amdgpu_ras_init(adev);
+	if (r)
+		return r;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
-		if (adev->ip_blocks[i].version->funcs->late_init) {
-			r = adev->ip_blocks[i].version->funcs->late_init((void *)adev);
+		if (adev->ip_blocks[i].version->funcs->sw_init) {
+			r = adev->ip_blocks[i].version->funcs->sw_init(&adev->ip_blocks[i]);
 			if (r) {
-				DRM_ERROR("late_init of IP block <%s> failed %d\n",
-					  adev->ip_blocks[i].version->funcs->name, r);
-				return r;
+				dev_err(adev->dev,
+					"sw_init of IP block <%s> failed %d\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
+				goto init_failed;
+			}
+		}
+		adev->ip_blocks[i].status.sw = true;
+
+		if (!amdgpu_ip_member_of_hwini(
+			    adev, adev->ip_blocks[i].version->type))
+			continue;
+
+		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON) {
+			/* need to do common hw init early so everything is set up for gmc */
+			r = adev->ip_blocks[i].version->funcs->hw_init(&adev->ip_blocks[i]);
+			if (r) {
+				dev_err(adev->dev, "hw_init %d failed %d\n", i,
+					r);
+				goto init_failed;
+			}
+			adev->ip_blocks[i].status.hw = true;
+		} else if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
+			/* need to do gmc hw init early so we can allocate gpu mem */
+			/* Try to reserve bad pages early */
+			if (amdgpu_sriov_vf(adev))
+				amdgpu_virt_exchange_data(adev);
+
+			r = amdgpu_device_mem_scratch_init(adev);
+			if (r) {
+				dev_err(adev->dev,
+					"amdgpu_mem_scratch_init failed %d\n",
+					r);
+				goto init_failed;
+			}
+			r = adev->ip_blocks[i].version->funcs->hw_init(&adev->ip_blocks[i]);
+			if (r) {
+				dev_err(adev->dev, "hw_init %d failed %d\n", i,
+					r);
+				goto init_failed;
+			}
+			r = amdgpu_device_wb_init(adev);
+			if (r) {
+				dev_err(adev->dev,
+					"amdgpu_device_wb_init failed %d\n", r);
+				goto init_failed;
+			}
+			adev->ip_blocks[i].status.hw = true;
+
+			/* right after GMC hw init, we create CSA */
+			if (adev->gfx.mcbp) {
+				r = amdgpu_allocate_static_csa(adev, &adev->virt.csa_obj,
+							       AMDGPU_GEM_DOMAIN_VRAM |
+							       AMDGPU_GEM_DOMAIN_GTT,
+							       AMDGPU_CSA_SIZE);
+				if (r) {
+					dev_err(adev->dev,
+						"allocate CSA failed %d\n", r);
+					goto init_failed;
+				}
+			}
+
+			r = amdgpu_seq64_init(adev);
+			if (r) {
+				dev_err(adev->dev, "allocate seq64 failed %d\n",
+					r);
+				goto init_failed;
+			}
+		}
+	}
+
+	if (amdgpu_sriov_vf(adev))
+		amdgpu_virt_init_data_exchange(adev);
+
+	r = amdgpu_ib_pool_init(adev);
+	if (r) {
+		dev_err(adev->dev, "IB initialization failed (%d).\n", r);
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r);
+		goto init_failed;
+	}
+
+	r = amdgpu_ucode_create_bo(adev); /* create ucode bo when sw_init complete*/
+	if (r)
+		goto init_failed;
+
+	r = amdgpu_device_ip_hw_init_phase1(adev);
+	if (r)
+		goto init_failed;
+
+	r = amdgpu_device_fw_loading(adev);
+	if (r)
+		goto init_failed;
+
+	r = amdgpu_device_ip_hw_init_phase2(adev);
+	if (r)
+		goto init_failed;
+
+	/*
+	 * retired pages will be loaded from eeprom and reserved here,
+	 * it should be called after amdgpu_device_ip_hw_init_phase2  since
+	 * for some ASICs the RAS EEPROM code relies on SMU fully functioning
+	 * for I2C communication which only true at this point.
+	 *
+	 * amdgpu_ras_recovery_init may fail, but the upper only cares the
+	 * failure from bad gpu situation and stop amdgpu init process
+	 * accordingly. For other failed cases, it will still release all
+	 * the resource and print error message, rather than returning one
+	 * negative value to upper level.
+	 *
+	 * Note: theoretically, this should be called before all vram allocations
+	 * to protect retired page from abusing
+	 */
+	init_badpage = (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI);
+	r = amdgpu_ras_recovery_init(adev, init_badpage);
+	if (r)
+		goto init_failed;
+
+	/**
+	 * In case of XGMI grab extra reference for reset domain for this device
+	 */
+	if (adev->gmc.xgmi.num_physical_nodes > 1) {
+		if (amdgpu_xgmi_add_device(adev) == 0) {
+			if (!amdgpu_sriov_vf(adev)) {
+				struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev);
+
+				if (WARN_ON(!hive)) {
+					r = -ENOENT;
+					goto init_failed;
+				}
+
+				if (!hive->reset_domain ||
+				    !amdgpu_reset_get_reset_domain(hive->reset_domain)) {
+					r = -ENOENT;
+					amdgpu_put_xgmi_hive(hive);
+					goto init_failed;
+				}
+
+				/* Drop the early temporary reset domain we created for device */
+				amdgpu_reset_put_reset_domain(adev->reset_domain);
+				adev->reset_domain = hive->reset_domain;
+				amdgpu_put_xgmi_hive(hive);
 			}
-			adev->ip_blocks[i].status.late_initialized = true;
 		}
+	}
+
+	r = amdgpu_device_init_schedulers(adev);
+	if (r)
+		goto init_failed;
+
+	if (adev->mman.buffer_funcs_ring->sched.ready)
+		amdgpu_ttm_set_buffer_funcs_status(adev, true);
+
+	/* Don't init kfd if whole hive need to be reset during init */
+	if (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) {
+		kgd2kfd_init_zone_device(adev);
+		amdgpu_amdkfd_device_init(adev);
+	}
+
+	amdgpu_fru_get_product_info(adev);
+
+	if (!amdgpu_sriov_vf(adev) || amdgpu_sriov_ras_cper_en(adev))
+		r = amdgpu_cper_init(adev);
+
+init_failed:
+
+	return r;
+}
+
+/**
+ * amdgpu_device_fill_reset_magic - writes reset magic to gart pointer
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Writes a reset magic value to the gart pointer in VRAM.  The driver calls
+ * this function before a GPU reset.  If the value is retained after a
+ * GPU reset, VRAM has not been lost. Some GPU resets may destroy VRAM contents.
+ */
+static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev)
+{
+	memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM);
+}
+
+/**
+ * amdgpu_device_check_vram_lost - check if vram is valid
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Checks the reset magic value written to the gart pointer in VRAM.
+ * The driver calls this after a GPU reset to see if the contents of
+ * VRAM is lost or now.
+ * returns true if vram is lost, false if not.
+ */
+static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev)
+{
+	if (memcmp(adev->gart.ptr, adev->reset_magic,
+			AMDGPU_RESET_MAGIC_NUM))
+		return true;
+
+	if (!amdgpu_in_reset(adev))
+		return false;
+
+	/*
+	 * For all ASICs with baco/mode1 reset, the VRAM is
+	 * always assumed to be lost.
+	 */
+	switch (amdgpu_asic_reset_method(adev)) {
+	case AMD_RESET_METHOD_LEGACY:
+	case AMD_RESET_METHOD_LINK:
+	case AMD_RESET_METHOD_BACO:
+	case AMD_RESET_METHOD_MODE1:
+		return true;
+	default:
+		return false;
+	}
+}
+
+/**
+ * amdgpu_device_set_cg_state - set clockgating for amdgpu device
+ *
+ * @adev: amdgpu_device pointer
+ * @state: clockgating state (gate or ungate)
+ *
+ * The list of all the hardware IPs that make up the asic is walked and the
+ * set_clockgating_state callbacks are run.
+ * Late initialization pass enabling clockgating for hardware IPs.
+ * Fini or suspend, pass disabling clockgating for hardware IPs.
+ * Returns 0 on success, negative error code on failure.
+ */
+
+int amdgpu_device_set_cg_state(struct amdgpu_device *adev,
+			       enum amd_clockgating_state state)
+{
+	int i, j, r;
+
+	if (amdgpu_emu_mode == 1)
+		return 0;
+
+	for (j = 0; j < adev->num_ip_blocks; j++) {
+		i = state == AMD_CG_STATE_GATE ? j : adev->num_ip_blocks - j - 1;
+		if (!adev->ip_blocks[i].status.late_initialized)
+			continue;
+		/* skip CG for GFX, SDMA on S0ix */
+		if (adev->in_s0ix &&
+		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX ||
+		     adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SDMA))
+			continue;
 		/* skip CG for VCE/UVD, it's handled specially */
 		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
-		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE) {
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG &&
+		    adev->ip_blocks[i].version->funcs->set_clockgating_state) {
 			/* enable clockgating to save power */
-			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
-										     AMD_CG_STATE_GATE);
+			r = adev->ip_blocks[i].version->funcs->set_clockgating_state(&adev->ip_blocks[i],
+										     state);
 			if (r) {
-				DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n",
-					  adev->ip_blocks[i].version->funcs->name, r);
+				dev_err(adev->dev,
+					"set_clockgating_state(gate) of IP block <%s> failed %d\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
 				return r;
 			}
 		}
@@ -1585,72 +3428,323 @@ static int amdgpu_late_init(struct amdgpu_device *adev)
 	return 0;
 }
 
-static int amdgpu_fini(struct amdgpu_device *adev)
+int amdgpu_device_set_pg_state(struct amdgpu_device *adev,
+			       enum amd_powergating_state state)
 {
-	int i, r;
+	int i, j, r;
 
-	/* need to disable SMC first */
-	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.hw)
+	if (amdgpu_emu_mode == 1)
+		return 0;
+
+	for (j = 0; j < adev->num_ip_blocks; j++) {
+		i = state == AMD_PG_STATE_GATE ? j : adev->num_ip_blocks - j - 1;
+		if (!adev->ip_blocks[i].status.late_initialized)
 			continue;
-		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) {
-			/* ungate blocks before hw fini so that we can shutdown the blocks safely */
-			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
-										     AMD_CG_STATE_UNGATE);
+		/* skip PG for GFX, SDMA on S0ix */
+		if (adev->in_s0ix &&
+		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX ||
+		     adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SDMA))
+			continue;
+		/* skip CG for VCE/UVD/VPE, it's handled specially */
+		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VPE &&
+		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG &&
+		    adev->ip_blocks[i].version->funcs->set_powergating_state) {
+			/* enable powergating to save power */
+			r = adev->ip_blocks[i].version->funcs->set_powergating_state(&adev->ip_blocks[i],
+											state);
 			if (r) {
-				DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n",
-					  adev->ip_blocks[i].version->funcs->name, r);
+				dev_err(adev->dev,
+					"set_powergating_state(gate) of IP block <%s> failed %d\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
 				return r;
 			}
-			r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
-			/* XXX handle errors */
-			if (r) {
-				DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
-					  adev->ip_blocks[i].version->funcs->name, r);
-			}
-			adev->ip_blocks[i].status.hw = false;
-			break;
 		}
 	}
+	return 0;
+}
 
-	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
-		if (!adev->ip_blocks[i].status.hw)
-			continue;
-		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
-			amdgpu_wb_fini(adev);
-			amdgpu_vram_scratch_fini(adev);
+static int amdgpu_device_enable_mgpu_fan_boost(void)
+{
+	struct amdgpu_gpu_instance *gpu_ins;
+	struct amdgpu_device *adev;
+	int i, ret = 0;
+
+	mutex_lock(&mgpu_info.mutex);
+
+	/*
+	 * MGPU fan boost feature should be enabled
+	 * only when there are two or more dGPUs in
+	 * the system
+	 */
+	if (mgpu_info.num_dgpu < 2)
+		goto out;
+
+	for (i = 0; i < mgpu_info.num_dgpu; i++) {
+		gpu_ins = &(mgpu_info.gpu_ins[i]);
+		adev = gpu_ins->adev;
+		if (!(adev->flags & AMD_IS_APU || amdgpu_sriov_multi_vf_mode(adev)) &&
+		    !gpu_ins->mgpu_fan_enabled) {
+			ret = amdgpu_dpm_enable_mgpu_fan_boost(adev);
+			if (ret)
+				break;
+
+			gpu_ins->mgpu_fan_enabled = 1;
 		}
+	}
 
-		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
-			adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE) {
-			/* ungate blocks before hw fini so that we can shutdown the blocks safely */
-			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
-										     AMD_CG_STATE_UNGATE);
+out:
+	mutex_unlock(&mgpu_info.mutex);
+
+	return ret;
+}
+
+/**
+ * amdgpu_device_ip_late_init - run late init for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Late initialization pass for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked and the late_init callbacks are run.
+ * late_init covers any special initialization that an IP requires
+ * after all of the have been initialized or something that needs to happen
+ * late in the init process.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_late_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_gpu_instance *gpu_instance;
+	int i = 0, r;
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!adev->ip_blocks[i].status.hw)
+			continue;
+		if (adev->ip_blocks[i].version->funcs->late_init) {
+			r = adev->ip_blocks[i].version->funcs->late_init(&adev->ip_blocks[i]);
 			if (r) {
-				DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n",
-					  adev->ip_blocks[i].version->funcs->name, r);
+				dev_err(adev->dev,
+					"late_init of IP block <%s> failed %d\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
 				return r;
 			}
 		}
+		adev->ip_blocks[i].status.late_initialized = true;
+	}
+
+	r = amdgpu_ras_late_init(adev);
+	if (r) {
+		dev_err(adev->dev, "amdgpu_ras_late_init failed %d", r);
+		return r;
+	}
+
+	if (!amdgpu_reset_in_recovery(adev))
+		amdgpu_ras_set_error_query_ready(adev, true);
+
+	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE);
+	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE);
+
+	amdgpu_device_fill_reset_magic(adev);
+
+	r = amdgpu_device_enable_mgpu_fan_boost();
+	if (r)
+		dev_err(adev->dev, "enable mgpu fan boost failed (%d).\n", r);
+
+	/* For passthrough configuration on arcturus and aldebaran, enable special handling SBR */
+	if (amdgpu_passthrough(adev) &&
+	    ((adev->asic_type == CHIP_ARCTURUS && adev->gmc.xgmi.num_physical_nodes > 1) ||
+	     adev->asic_type == CHIP_ALDEBARAN))
+		amdgpu_dpm_handle_passthrough_sbr(adev, true);
+
+	if (adev->gmc.xgmi.num_physical_nodes > 1) {
+		mutex_lock(&mgpu_info.mutex);
+
+		/*
+		 * Reset device p-state to low as this was booted with high.
+		 *
+		 * This should be performed only after all devices from the same
+		 * hive get initialized.
+		 *
+		 * However, it's unknown how many device in the hive in advance.
+		 * As this is counted one by one during devices initializations.
+		 *
+		 * So, we wait for all XGMI interlinked devices initialized.
+		 * This may bring some delays as those devices may come from
+		 * different hives. But that should be OK.
+		 */
+		if (mgpu_info.num_dgpu == adev->gmc.xgmi.num_physical_nodes) {
+			for (i = 0; i < mgpu_info.num_gpu; i++) {
+				gpu_instance = &(mgpu_info.gpu_ins[i]);
+				if (gpu_instance->adev->flags & AMD_IS_APU)
+					continue;
+
+				r = amdgpu_xgmi_set_pstate(gpu_instance->adev,
+						AMDGPU_XGMI_PSTATE_MIN);
+				if (r) {
+					dev_err(adev->dev,
+						"pstate setting failed (%d).\n",
+						r);
+					break;
+				}
+			}
+		}
+
+		mutex_unlock(&mgpu_info.mutex);
+	}
+
+	return 0;
+}
+
+static void amdgpu_ip_block_hw_fini(struct amdgpu_ip_block *ip_block)
+{
+	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
-		r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
+	if (!ip_block->version->funcs->hw_fini) {
+		dev_err(adev->dev, "hw_fini of IP block <%s> not defined\n",
+			ip_block->version->funcs->name);
+	} else {
+		r = ip_block->version->funcs->hw_fini(ip_block);
 		/* XXX handle errors */
 		if (r) {
-			DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
+			dev_dbg(adev->dev,
+				"hw_fini of IP block <%s> failed %d\n",
+				ip_block->version->funcs->name, r);
+		}
+	}
+
+	ip_block->status.hw = false;
+}
+
+/**
+ * amdgpu_device_smu_fini_early - smu hw_fini wrapper
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * For ASICs need to disable SMC first
+ */
+static void amdgpu_device_smu_fini_early(struct amdgpu_device *adev)
+{
+	int i;
+
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) > IP_VERSION(9, 0, 0))
+		return;
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!adev->ip_blocks[i].status.hw)
+			continue;
+		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) {
+			amdgpu_ip_block_hw_fini(&adev->ip_blocks[i]);
+			break;
 		}
+	}
+}
+
+static int amdgpu_device_ip_fini_early(struct amdgpu_device *adev)
+{
+	int i, r;
 
-		adev->ip_blocks[i].status.hw = false;
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!adev->ip_blocks[i].version->funcs->early_fini)
+			continue;
+
+		r = adev->ip_blocks[i].version->funcs->early_fini(&adev->ip_blocks[i]);
+		if (r) {
+			dev_dbg(adev->dev,
+				"early_fini of IP block <%s> failed %d\n",
+				adev->ip_blocks[i].version->funcs->name, r);
+		}
 	}
 
+	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
+	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
+
+	amdgpu_amdkfd_suspend(adev, true);
+	amdgpu_userq_suspend(adev);
+
+	/* Workaround for ASICs need to disable SMC first */
+	amdgpu_device_smu_fini_early(adev);
+
+	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
+		if (!adev->ip_blocks[i].status.hw)
+			continue;
+
+		amdgpu_ip_block_hw_fini(&adev->ip_blocks[i]);
+	}
+
+	if (amdgpu_sriov_vf(adev)) {
+		if (amdgpu_virt_release_full_gpu(adev, false))
+			dev_err(adev->dev,
+				"failed to release exclusive mode on fini\n");
+	}
+
+	/*
+	 * Driver reload on the APU can fail due to firmware validation because
+	 * the PSP is always running, as it is shared across the whole SoC.
+	 * This same issue does not occur on dGPU because it has a mechanism
+	 * that checks whether the PSP is running. A solution for those issues
+	 * in the APU is to trigger a GPU reset, but this should be done during
+	 * the unload phase to avoid adding boot latency and screen flicker.
+	 */
+	if ((adev->flags & AMD_IS_APU) && !adev->gmc.is_app_apu) {
+		r = amdgpu_asic_reset(adev);
+		if (r)
+			dev_err(adev->dev, "asic reset on %s failed\n", __func__);
+	}
+
+	return 0;
+}
+
+/**
+ * amdgpu_device_ip_fini - run fini for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Main teardown pass for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked and the hw_fini and sw_fini callbacks
+ * are run.  hw_fini tears down the hardware associated with each IP
+ * and sw_fini tears down any software state associated with each IP.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_fini(struct amdgpu_device *adev)
+{
+	int i, r;
+
+	amdgpu_cper_fini(adev);
+
+	if (amdgpu_sriov_vf(adev) && adev->virt.ras_init_done)
+		amdgpu_virt_release_ras_err_handler_data(adev);
+
+	if (adev->gmc.xgmi.num_physical_nodes > 1)
+		amdgpu_xgmi_remove_device(adev);
+
+	amdgpu_amdkfd_device_fini_sw(adev);
+
 	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
 		if (!adev->ip_blocks[i].status.sw)
 			continue;
-		r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev);
-		/* XXX handle errors */
-		if (r) {
-			DRM_DEBUG("sw_fini of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
+
+		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
+			amdgpu_ucode_free_bo(adev);
+			amdgpu_free_static_csa(&adev->virt.csa_obj);
+			amdgpu_device_wb_fini(adev);
+			amdgpu_device_mem_scratch_fini(adev);
+			amdgpu_ib_pool_fini(adev);
+			amdgpu_seq64_fini(adev);
+			amdgpu_doorbell_fini(adev);
+		}
+		if (adev->ip_blocks[i].version->funcs->sw_fini) {
+			r = adev->ip_blocks[i].version->funcs->sw_fini(&adev->ip_blocks[i]);
+			/* XXX handle errors */
+			if (r) {
+				dev_dbg(adev->dev,
+					"sw_fini of IP block <%s> failed %d\n",
+					adev->ip_blocks[i].version->funcs->name,
+					r);
+			}
 		}
 		adev->ip_blocks[i].status.sw = false;
 		adev->ip_blocks[i].status.valid = false;
@@ -1660,141 +3754,764 @@ static int amdgpu_fini(struct amdgpu_device *adev)
 		if (!adev->ip_blocks[i].status.late_initialized)
 			continue;
 		if (adev->ip_blocks[i].version->funcs->late_fini)
-			adev->ip_blocks[i].version->funcs->late_fini((void *)adev);
+			adev->ip_blocks[i].version->funcs->late_fini(&adev->ip_blocks[i]);
 		adev->ip_blocks[i].status.late_initialized = false;
 	}
 
-	if (amdgpu_sriov_vf(adev)) {
-		amdgpu_bo_free_kernel(&adev->virt.csa_obj, &adev->virt.csa_vmid0_addr, NULL);
-		amdgpu_virt_release_full_gpu(adev, false);
-	}
+	amdgpu_ras_fini(adev);
+	amdgpu_uid_fini(adev);
 
 	return 0;
 }
 
-int amdgpu_suspend(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_delayed_init_work_handler - work handler for IB tests
+ *
+ * @work: work_struct.
+ */
+static void amdgpu_device_delayed_init_work_handler(struct work_struct *work)
+{
+	struct amdgpu_device *adev =
+		container_of(work, struct amdgpu_device, delayed_init_work.work);
+	int r;
+
+	r = amdgpu_ib_ring_tests(adev);
+	if (r)
+		dev_err(adev->dev, "ib ring test failed (%d).\n", r);
+}
+
+static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work)
 {
-	int i, r;
+	struct amdgpu_device *adev =
+		container_of(work, struct amdgpu_device, gfx.gfx_off_delay_work.work);
 
-	if (amdgpu_sriov_vf(adev))
-		amdgpu_virt_request_full_gpu(adev, false);
+	WARN_ON_ONCE(adev->gfx.gfx_off_state);
+	WARN_ON_ONCE(adev->gfx.gfx_off_req_count);
 
-	/* ungate SMC block first */
-	r = amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_SMC,
-					 AMD_CG_STATE_UNGATE);
-	if (r) {
-		DRM_ERROR("set_clockgating_state(ungate) SMC failed %d\n",r);
+	if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true, 0))
+		adev->gfx.gfx_off_state = true;
+}
+
+/**
+ * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1)
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Main suspend function for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked, clockgating is disabled and the
+ * suspend callbacks are run.  suspend puts the hardware and software state
+ * in each IP into a state suitable for suspend.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev)
+{
+	int i, r, rec;
+
+	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
+	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
+
+	/*
+	 * Per PMFW team's suggestion, driver needs to handle gfxoff
+	 * and df cstate features disablement for gpu reset(e.g. Mode1Reset)
+	 * scenario. Add the missing df cstate disablement here.
+	 */
+	if (amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_DISALLOW))
+		dev_warn(adev->dev, "Failed to disallow df cstate");
+
+	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
+		if (!adev->ip_blocks[i].status.valid)
+			continue;
+
+		/* displays are handled separately */
+		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_DCE)
+			continue;
+
+		r = amdgpu_ip_block_suspend(&adev->ip_blocks[i]);
+		if (r)
+			goto unwind;
 	}
 
+	return 0;
+unwind:
+	rec = amdgpu_device_ip_resume_phase3(adev);
+	if (rec)
+		dev_err(adev->dev,
+			"amdgpu_device_ip_resume_phase3 failed during unwind: %d\n",
+			rec);
+
+	amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_ALLOW);
+
+	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE);
+	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE);
+
+	return r;
+}
+
+/**
+ * amdgpu_device_ip_suspend_phase2 - run suspend for hardware IPs (phase 2)
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Main suspend function for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked, clockgating is disabled and the
+ * suspend callbacks are run.  suspend puts the hardware and software state
+ * in each IP into a state suitable for suspend.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev)
+{
+	int i, r, rec;
+
+	if (adev->in_s0ix)
+		amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D3Entry);
+
 	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
-		/* ungate blocks so that suspend can properly shut them down */
-		if (i != AMD_IP_BLOCK_TYPE_SMC) {
-			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
-										     AMD_CG_STATE_UNGATE);
-			if (r) {
-				DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n",
-					  adev->ip_blocks[i].version->funcs->name, r);
-			}
+		/* displays are handled in phase1 */
+		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE)
+			continue;
+		/* PSP lost connection when err_event_athub occurs */
+		if (amdgpu_ras_intr_triggered() &&
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) {
+			adev->ip_blocks[i].status.hw = false;
+			continue;
 		}
-		/* XXX handle errors */
-		r = adev->ip_blocks[i].version->funcs->suspend(adev);
-		/* XXX handle errors */
-		if (r) {
-			DRM_ERROR("suspend of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
+
+		/* skip unnecessary suspend if we do not initialize them yet */
+		if (!amdgpu_ip_member_of_hwini(
+			    adev, adev->ip_blocks[i].version->type))
+			continue;
+
+		/* Since we skip suspend for S0i3, we need to cancel the delayed
+		 * idle work here as the suspend callback never gets called.
+		 */
+		if (adev->in_s0ix &&
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX &&
+		    amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(10, 0, 0))
+			cancel_delayed_work_sync(&adev->gfx.idle_work);
+		/* skip suspend of gfx/mes and psp for S0ix
+		 * gfx is in gfxoff state, so on resume it will exit gfxoff just
+		 * like at runtime. PSP is also part of the always on hardware
+		 * so no need to suspend it.
+		 */
+		if (adev->in_s0ix &&
+		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP ||
+		     adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX ||
+		     adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_MES))
+			continue;
+
+		/* SDMA 5.x+ is part of GFX power domain so it's covered by GFXOFF */
+		if (adev->in_s0ix &&
+		    (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >=
+		     IP_VERSION(5, 0, 0)) &&
+		    (adev->ip_blocks[i].version->type ==
+		     AMD_IP_BLOCK_TYPE_SDMA))
+			continue;
+
+		/* Once swPSP provides the IMU, RLC FW binaries to TOS during cold-boot.
+		 * These are in TMR, hence are expected to be reused by PSP-TOS to reload
+		 * from this location and RLC Autoload automatically also gets loaded
+		 * from here based on PMFW -> PSP message during re-init sequence.
+		 * Therefore, the psp suspend & resume should be skipped to avoid destroy
+		 * the TMR and reload FWs again for IMU enabled APU ASICs.
+		 */
+		if (amdgpu_in_reset(adev) &&
+		    (adev->flags & AMD_IS_APU) && adev->gfx.imu.funcs &&
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP)
+			continue;
+
+		r = amdgpu_ip_block_suspend(&adev->ip_blocks[i]);
+		if (r)
+			goto unwind;
+
+		/* handle putting the SMC in the appropriate state */
+		if (!amdgpu_sriov_vf(adev)) {
+			if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) {
+				r = amdgpu_dpm_set_mp1_state(adev, adev->mp1_state);
+				if (r) {
+					dev_err(adev->dev,
+						"SMC failed to set mp1 state %d, %d\n",
+						adev->mp1_state, r);
+					goto unwind;
+				}
+			}
 		}
 	}
 
+	return 0;
+unwind:
+	/* suspend phase 2 = resume phase 1 + resume phase 2 */
+	rec = amdgpu_device_ip_resume_phase1(adev);
+	if (rec) {
+		dev_err(adev->dev,
+			"amdgpu_device_ip_resume_phase1 failed during unwind: %d\n",
+			rec);
+		return r;
+	}
+
+	rec = amdgpu_device_fw_loading(adev);
+	if (rec) {
+		dev_err(adev->dev,
+			"amdgpu_device_fw_loading failed during unwind: %d\n",
+			rec);
+		return r;
+	}
+
+	rec = amdgpu_device_ip_resume_phase2(adev);
+	if (rec) {
+		dev_err(adev->dev,
+			"amdgpu_device_ip_resume_phase2 failed during unwind: %d\n",
+			rec);
+		return r;
+	}
+
+	return r;
+}
+
+/**
+ * amdgpu_device_ip_suspend - run suspend for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Main suspend function for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked, clockgating is disabled and the
+ * suspend callbacks are run.  suspend puts the hardware and software state
+ * in each IP into a state suitable for suspend.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_suspend(struct amdgpu_device *adev)
+{
+	int r;
+
+	if (amdgpu_sriov_vf(adev)) {
+		amdgpu_virt_fini_data_exchange(adev);
+		amdgpu_virt_request_full_gpu(adev, false);
+	}
+
+	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+
+	r = amdgpu_device_ip_suspend_phase1(adev);
+	if (r)
+		return r;
+	r = amdgpu_device_ip_suspend_phase2(adev);
+
 	if (amdgpu_sriov_vf(adev))
 		amdgpu_virt_release_full_gpu(adev, false);
 
-	return 0;
+	return r;
 }
 
-static int amdgpu_sriov_reinit_early(struct amdgpu_device *adev)
+static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev)
 {
 	int i, r;
 
+	static enum amd_ip_block_type ip_order[] = {
+		AMD_IP_BLOCK_TYPE_COMMON,
+		AMD_IP_BLOCK_TYPE_GMC,
+		AMD_IP_BLOCK_TYPE_PSP,
+		AMD_IP_BLOCK_TYPE_IH,
+	};
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
+		int j;
+		struct amdgpu_ip_block *block;
+
+		block = &adev->ip_blocks[i];
+		block->status.hw = false;
+
+		for (j = 0; j < ARRAY_SIZE(ip_order); j++) {
+
+			if (block->version->type != ip_order[j] ||
+				!block->status.valid)
+				continue;
+
+			r = block->version->funcs->hw_init(&adev->ip_blocks[i]);
+			if (r) {
+				dev_err(adev->dev, "RE-INIT-early: %s failed\n",
+					 block->version->funcs->name);
+				return r;
+			}
+			block->status.hw = true;
+		}
+	}
+
+	return 0;
+}
+
+static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev)
+{
+	struct amdgpu_ip_block *block;
+	int i, r = 0;
+
+	static enum amd_ip_block_type ip_order[] = {
+		AMD_IP_BLOCK_TYPE_SMC,
+		AMD_IP_BLOCK_TYPE_DCE,
+		AMD_IP_BLOCK_TYPE_GFX,
+		AMD_IP_BLOCK_TYPE_SDMA,
+		AMD_IP_BLOCK_TYPE_MES,
+		AMD_IP_BLOCK_TYPE_UVD,
+		AMD_IP_BLOCK_TYPE_VCE,
+		AMD_IP_BLOCK_TYPE_VCN,
+		AMD_IP_BLOCK_TYPE_JPEG
+	};
+
+	for (i = 0; i < ARRAY_SIZE(ip_order); i++) {
+		block = amdgpu_device_ip_get_ip_block(adev, ip_order[i]);
+
+		if (!block)
 			continue;
 
+		if (block->status.valid && !block->status.hw) {
+			if (block->version->type == AMD_IP_BLOCK_TYPE_SMC) {
+				r = amdgpu_ip_block_resume(block);
+			} else {
+				r = block->version->funcs->hw_init(block);
+			}
+
+			if (r) {
+				dev_err(adev->dev, "RE-INIT-late: %s failed\n",
+					 block->version->funcs->name);
+				break;
+			}
+			block->status.hw = true;
+		}
+	}
+
+	return r;
+}
+
+/**
+ * amdgpu_device_ip_resume_phase1 - run resume for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * First resume function for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked and the resume callbacks are run for
+ * COMMON, GMC, and IH.  resume puts the hardware into a functional state
+ * after a suspend and updates the software state as necessary.  This
+ * function is also used for restoring the GPU after a GPU reset.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev)
+{
+	int i, r;
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw)
+			continue;
 		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
-				adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
-				adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH)
-			r = adev->ip_blocks[i].version->funcs->hw_init(adev);
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH ||
+		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP && amdgpu_sriov_vf(adev))) {
 
-		if (r) {
-			DRM_ERROR("resume of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
-			return r;
+			r = amdgpu_ip_block_resume(&adev->ip_blocks[i]);
+			if (r)
+				return r;
 		}
 	}
 
 	return 0;
 }
 
-static int amdgpu_sriov_reinit_late(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_ip_resume_phase2 - run resume for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Second resume function for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked and the resume callbacks are run for
+ * all blocks except COMMON, GMC, and IH.  resume puts the hardware into a
+ * functional state after a suspend and updates the software state as
+ * necessary.  This function is also used for restoring the GPU after a GPU
+ * reset.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev)
 {
 	int i, r;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
+		if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw)
 			continue;
-
 		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
-				adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
-				adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH )
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH ||
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE ||
+		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP)
 			continue;
-
-		r = adev->ip_blocks[i].version->funcs->hw_init(adev);
-		if (r) {
-			DRM_ERROR("resume of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
+		r = amdgpu_ip_block_resume(&adev->ip_blocks[i]);
+		if (r)
 			return r;
-		}
 	}
 
 	return 0;
 }
 
-static int amdgpu_resume(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_ip_resume_phase3 - run resume for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Third resume function for hardware IPs.  The list of all the hardware
+ * IPs that make up the asic is walked and the resume callbacks are run for
+ * all DCE.  resume puts the hardware into a functional state after a suspend
+ * and updates the software state as necessary.  This function is also used
+ * for restoring the GPU after a GPU reset.
+ *
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_resume_phase3(struct amdgpu_device *adev)
 {
 	int i, r;
 
 	for (i = 0; i < adev->num_ip_blocks; i++) {
-		if (!adev->ip_blocks[i].status.valid)
+		if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw)
 			continue;
-		r = adev->ip_blocks[i].version->funcs->resume(adev);
-		if (r) {
-			DRM_ERROR("resume of IP block <%s> failed %d\n",
-				  adev->ip_blocks[i].version->funcs->name, r);
-			return r;
+		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) {
+			r = amdgpu_ip_block_resume(&adev->ip_blocks[i]);
+			if (r)
+				return r;
 		}
 	}
 
 	return 0;
 }
 
+/**
+ * amdgpu_device_ip_resume - run resume for hardware IPs
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Main resume function for hardware IPs.  The hardware IPs
+ * are split into two resume functions because they are
+ * also used in recovering from a GPU reset and some additional
+ * steps need to be take between them.  In this case (S3/S4) they are
+ * run sequentially.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_resume(struct amdgpu_device *adev)
+{
+	int r;
+
+	r = amdgpu_device_ip_resume_phase1(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_device_fw_loading(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_device_ip_resume_phase2(adev);
+
+	if (adev->mman.buffer_funcs_ring->sched.ready)
+		amdgpu_ttm_set_buffer_funcs_status(adev, true);
+
+	if (r)
+		return r;
+
+	amdgpu_fence_driver_hw_init(adev);
+
+	r = amdgpu_device_ip_resume_phase3(adev);
+
+	return r;
+}
+
+/**
+ * amdgpu_device_detect_sriov_bios - determine if the board supports SR-IOV
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Query the VBIOS data tables to determine if the board supports SR-IOV.
+ */
 static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev)
 {
-	if (adev->is_atom_fw) {
-		if (amdgpu_atomfirmware_gpu_supports_virtualization(adev))
-			adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
+	if (amdgpu_sriov_vf(adev)) {
+		if (adev->is_atom_fw) {
+			if (amdgpu_atomfirmware_gpu_virtualization_supported(adev))
+				adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
+		} else {
+			if (amdgpu_atombios_has_gpu_virtualization_table(adev))
+				adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
+		}
+
+		if (!(adev->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS))
+			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0);
+	}
+}
+
+/**
+ * amdgpu_device_asic_has_dc_support - determine if DC supports the asic
+ *
+ * @pdev : pci device context
+ * @asic_type: AMD asic type
+ *
+ * Check if there is DC (new modesetting infrastructre) support for an asic.
+ * returns true if DC has support, false if not.
+ */
+bool amdgpu_device_asic_has_dc_support(struct pci_dev *pdev,
+				       enum amd_asic_type asic_type)
+{
+	switch (asic_type) {
+#ifdef CONFIG_DRM_AMDGPU_SI
+	case CHIP_HAINAN:
+#endif
+	case CHIP_TOPAZ:
+		/* chips with no display hardware */
+		return false;
+#if defined(CONFIG_DRM_AMD_DC)
+	case CHIP_TAHITI:
+	case CHIP_PITCAIRN:
+	case CHIP_VERDE:
+	case CHIP_OLAND:
+		return amdgpu_dc != 0 && IS_ENABLED(CONFIG_DRM_AMD_DC_SI);
+	case CHIP_KAVERI:
+	case CHIP_KABINI:
+	case CHIP_MULLINS:
+		/*
+		 * We have systems in the wild with these ASICs that require
+		 * TRAVIS and NUTMEG support which is not supported with DC.
+		 *
+		 * Fallback to the non-DC driver here by default so as not to
+		 * cause regressions.
+		 */
+		return amdgpu_dc > 0;
+	default:
+		return amdgpu_dc != 0;
+#else
+	default:
+		if (amdgpu_dc > 0)
+			dev_info_once(
+				&pdev->dev,
+				"Display Core has been requested via kernel parameter but isn't supported by ASIC, ignoring\n");
+		return false;
+#endif
+	}
+}
+
+/**
+ * amdgpu_device_has_dc_support - check if dc is supported
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Returns true for supported, false for not supported
+ */
+bool amdgpu_device_has_dc_support(struct amdgpu_device *adev)
+{
+	if (adev->enable_virtual_display ||
+	    (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK))
+		return false;
+
+	return amdgpu_device_asic_has_dc_support(adev->pdev, adev->asic_type);
+}
+
+static void amdgpu_device_xgmi_reset_func(struct work_struct *__work)
+{
+	struct amdgpu_device *adev =
+		container_of(__work, struct amdgpu_device, xgmi_reset_work);
+	struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev);
+
+	/* It's a bug to not have a hive within this function */
+	if (WARN_ON(!hive))
+		return;
+
+	/*
+	 * Use task barrier to synchronize all xgmi reset works across the
+	 * hive. task_barrier_enter and task_barrier_exit will block
+	 * until all the threads running the xgmi reset works reach
+	 * those points. task_barrier_full will do both blocks.
+	 */
+	if (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) {
+
+		task_barrier_enter(&hive->tb);
+		adev->asic_reset_res = amdgpu_device_baco_enter(adev);
+
+		if (adev->asic_reset_res)
+			goto fail;
+
+		task_barrier_exit(&hive->tb);
+		adev->asic_reset_res = amdgpu_device_baco_exit(adev);
+
+		if (adev->asic_reset_res)
+			goto fail;
+
+		amdgpu_ras_reset_error_count(adev, AMDGPU_RAS_BLOCK__MMHUB);
 	} else {
-		if (amdgpu_atombios_has_gpu_virtualization_table(adev))
-			adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
+
+		task_barrier_full(&hive->tb);
+		adev->asic_reset_res =  amdgpu_asic_reset(adev);
+	}
+
+fail:
+	if (adev->asic_reset_res)
+		dev_warn(adev->dev,
+			 "ASIC reset failed with error, %d for drm dev, %s",
+			 adev->asic_reset_res, adev_to_drm(adev)->unique);
+	amdgpu_put_xgmi_hive(hive);
+}
+
+static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)
+{
+	char *input = amdgpu_lockup_timeout;
+	char *timeout_setting = NULL;
+	int index = 0;
+	long timeout;
+	int ret = 0;
+
+	/* By default timeout for all queues is 2 sec */
+	adev->gfx_timeout = adev->compute_timeout = adev->sdma_timeout =
+		adev->video_timeout = msecs_to_jiffies(2000);
+
+	if (!strnlen(input, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH))
+		return 0;
+
+	while ((timeout_setting = strsep(&input, ",")) &&
+	       strnlen(timeout_setting, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH)) {
+		ret = kstrtol(timeout_setting, 0, &timeout);
+		if (ret)
+			return ret;
+
+		if (timeout == 0) {
+			index++;
+			continue;
+		} else if (timeout < 0) {
+			timeout = MAX_SCHEDULE_TIMEOUT;
+			dev_warn(adev->dev, "lockup timeout disabled");
+			add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
+		} else {
+			timeout = msecs_to_jiffies(timeout);
+		}
+
+		switch (index++) {
+		case 0:
+			adev->gfx_timeout = timeout;
+			break;
+		case 1:
+			adev->compute_timeout = timeout;
+			break;
+		case 2:
+			adev->sdma_timeout = timeout;
+			break;
+		case 3:
+			adev->video_timeout = timeout;
+			break;
+		default:
+			break;
+		}
 	}
+
+	/* When only one value specified apply it to all queues. */
+	if (index == 1)
+		adev->gfx_timeout = adev->compute_timeout = adev->sdma_timeout =
+			adev->video_timeout = timeout;
+
+	return ret;
+}
+
+/**
+ * amdgpu_device_check_iommu_direct_map - check if RAM direct mapped to GPU
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * RAM direct mapped to GPU if IOMMU is not enabled or is pass through mode
+ */
+static void amdgpu_device_check_iommu_direct_map(struct amdgpu_device *adev)
+{
+	struct iommu_domain *domain;
+
+	domain = iommu_get_domain_for_dev(adev->dev);
+	if (!domain || domain->type == IOMMU_DOMAIN_IDENTITY)
+		adev->ram_is_direct_mapped = true;
+}
+
+#if defined(CONFIG_HSA_AMD_P2P)
+/**
+ * amdgpu_device_check_iommu_remap - Check if DMA remapping is enabled.
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * return if IOMMU remapping bar address
+ */
+static bool amdgpu_device_check_iommu_remap(struct amdgpu_device *adev)
+{
+	struct iommu_domain *domain;
+
+	domain = iommu_get_domain_for_dev(adev->dev);
+	if (domain && (domain->type == IOMMU_DOMAIN_DMA ||
+		domain->type ==	IOMMU_DOMAIN_DMA_FQ))
+		return true;
+
+	return false;
+}
+#endif
+
+static void amdgpu_device_set_mcbp(struct amdgpu_device *adev)
+{
+	if (amdgpu_mcbp == 1)
+		adev->gfx.mcbp = true;
+	else if (amdgpu_mcbp == 0)
+		adev->gfx.mcbp = false;
+
+	if (amdgpu_sriov_vf(adev))
+		adev->gfx.mcbp = true;
+
+	if (adev->gfx.mcbp)
+		dev_info(adev->dev, "MCBP is enabled\n");
+}
+
+static int amdgpu_device_sys_interface_init(struct amdgpu_device *adev)
+{
+	int r;
+
+	r = amdgpu_atombios_sysfs_init(adev);
+	if (r)
+		drm_err(&adev->ddev,
+			"registering atombios sysfs failed (%d).\n", r);
+
+	r = amdgpu_pm_sysfs_init(adev);
+	if (r)
+		dev_err(adev->dev, "registering pm sysfs failed (%d).\n", r);
+
+	r = amdgpu_ucode_sysfs_init(adev);
+	if (r) {
+		adev->ucode_sysfs_en = false;
+		dev_err(adev->dev, "Creating firmware sysfs failed (%d).\n", r);
+	} else
+		adev->ucode_sysfs_en = true;
+
+	r = amdgpu_device_attr_sysfs_init(adev);
+	if (r)
+		dev_err(adev->dev, "Could not create amdgpu device attr\n");
+
+	r = devm_device_add_group(adev->dev, &amdgpu_board_attrs_group);
+	if (r)
+		dev_err(adev->dev,
+			"Could not create amdgpu board attributes\n");
+
+	amdgpu_fru_sysfs_init(adev);
+	amdgpu_reg_state_sysfs_init(adev);
+	amdgpu_xcp_sysfs_init(adev);
+
+	return r;
+}
+
+static void amdgpu_device_sys_interface_fini(struct amdgpu_device *adev)
+{
+	if (adev->pm.sysfs_initialized)
+		amdgpu_pm_sysfs_fini(adev);
+	if (adev->ucode_sysfs_en)
+		amdgpu_ucode_sysfs_fini(adev);
+	amdgpu_device_attr_sysfs_fini(adev);
+	amdgpu_fru_sysfs_fini(adev);
+
+	amdgpu_reg_state_sysfs_fini(adev);
+	amdgpu_xcp_sysfs_fini(adev);
 }
 
 /**
  * amdgpu_device_init - initialize the driver
  *
  * @adev: amdgpu_device pointer
- * @pdev: drm dev pointer
- * @pdev: pci dev pointer
  * @flags: driver flags
  *
  * Initializes the driver info and hw (all asics).
@@ -1802,37 +4519,50 @@ static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev)
  * Called at driver startup.
  */
 int amdgpu_device_init(struct amdgpu_device *adev,
-		       struct drm_device *ddev,
-		       struct pci_dev *pdev,
 		       uint32_t flags)
 {
+	struct pci_dev *pdev = adev->pdev;
 	int r, i;
-	bool runtime = false;
+	bool px = false;
 	u32 max_MBps;
+	int tmp;
 
 	adev->shutdown = false;
-	adev->dev = &pdev->dev;
-	adev->ddev = ddev;
-	adev->pdev = pdev;
 	adev->flags = flags;
-	adev->asic_type = flags & AMD_ASIC_MASK;
+
+	if (amdgpu_force_asic_type >= 0 && amdgpu_force_asic_type < CHIP_LAST)
+		adev->asic_type = amdgpu_force_asic_type;
+	else
+		adev->asic_type = flags & AMD_ASIC_MASK;
+
 	adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT;
-	adev->mc.gtt_size = 512 * 1024 * 1024;
+	if (amdgpu_emu_mode == 1)
+		adev->usec_timeout *= 10;
+	adev->gmc.gart_size = 512 * 1024 * 1024;
 	adev->accel_working = false;
 	adev->num_rings = 0;
+	RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub());
 	adev->mman.buffer_funcs = NULL;
 	adev->mman.buffer_funcs_ring = NULL;
 	adev->vm_manager.vm_pte_funcs = NULL;
-	adev->vm_manager.vm_pte_num_rings = 0;
-	adev->gart.gart_funcs = NULL;
+	adev->vm_manager.vm_pte_num_scheds = 0;
+	adev->gmc.gmc_funcs = NULL;
+	adev->harvest_ip_mask = 0x0;
 	adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS);
+	bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
 
 	adev->smc_rreg = &amdgpu_invalid_rreg;
 	adev->smc_wreg = &amdgpu_invalid_wreg;
 	adev->pcie_rreg = &amdgpu_invalid_rreg;
 	adev->pcie_wreg = &amdgpu_invalid_wreg;
+	adev->pcie_rreg_ext = &amdgpu_invalid_rreg_ext;
+	adev->pcie_wreg_ext = &amdgpu_invalid_wreg_ext;
 	adev->pciep_rreg = &amdgpu_invalid_rreg;
 	adev->pciep_wreg = &amdgpu_invalid_wreg;
+	adev->pcie_rreg64 = &amdgpu_invalid_rreg64;
+	adev->pcie_wreg64 = &amdgpu_invalid_wreg64;
+	adev->pcie_rreg64_ext = &amdgpu_invalid_rreg64_ext;
+	adev->pcie_wreg64_ext = &amdgpu_invalid_wreg64_ext;
 	adev->uvd_ctx_rreg = &amdgpu_invalid_rreg;
 	adev->uvd_ctx_wreg = &amdgpu_invalid_wreg;
 	adev->didt_rreg = &amdgpu_invalid_rreg;
@@ -1842,41 +4572,112 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg;
 	adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg;
 
-
-	DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n",
-		 amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device,
-		 pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision);
+	dev_info(
+		adev->dev,
+		"initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n",
+		amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device,
+		pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision);
 
 	/* mutex initialization are all done here so we
-	 * can recall function without having locking issues */
-	atomic_set(&adev->irq.ih.lock, 0);
+	 * can recall function without having locking issues
+	 */
 	mutex_init(&adev->firmware.mutex);
 	mutex_init(&adev->pm.mutex);
 	mutex_init(&adev->gfx.gpu_clock_mutex);
 	mutex_init(&adev->srbm_mutex);
+	mutex_init(&adev->gfx.pipe_reserve_mutex);
+	mutex_init(&adev->gfx.gfx_off_mutex);
+	mutex_init(&adev->gfx.partition_mutex);
 	mutex_init(&adev->grbm_idx_mutex);
 	mutex_init(&adev->mn_lock);
+	mutex_init(&adev->virt.vf_errors.lock);
 	hash_init(adev->mn_hash);
+	mutex_init(&adev->psp.mutex);
+	mutex_init(&adev->notifier_lock);
+	mutex_init(&adev->pm.stable_pstate_ctx_lock);
+	mutex_init(&adev->benchmark_mutex);
+	mutex_init(&adev->gfx.reset_sem_mutex);
+	/* Initialize the mutex for cleaner shader isolation between GFX and compute processes */
+	mutex_init(&adev->enforce_isolation_mutex);
+	for (i = 0; i < MAX_XCP; ++i) {
+		adev->isolation[i].spearhead = dma_fence_get_stub();
+		amdgpu_sync_create(&adev->isolation[i].active);
+		amdgpu_sync_create(&adev->isolation[i].prev);
+	}
+	mutex_init(&adev->gfx.userq_sch_mutex);
+	mutex_init(&adev->gfx.workload_profile_mutex);
+	mutex_init(&adev->vcn.workload_profile_mutex);
+
+	amdgpu_device_init_apu_flags(adev);
+
+	r = amdgpu_device_check_arguments(adev);
+	if (r)
+		return r;
 
-	amdgpu_check_arguments(adev);
-
-	/* Registers mapping */
-	/* TODO: block userspace mapping of io register */
 	spin_lock_init(&adev->mmio_idx_lock);
 	spin_lock_init(&adev->smc_idx_lock);
 	spin_lock_init(&adev->pcie_idx_lock);
 	spin_lock_init(&adev->uvd_ctx_idx_lock);
 	spin_lock_init(&adev->didt_idx_lock);
 	spin_lock_init(&adev->gc_cac_idx_lock);
+	spin_lock_init(&adev->se_cac_idx_lock);
 	spin_lock_init(&adev->audio_endpt_idx_lock);
 	spin_lock_init(&adev->mm_stats.lock);
+	spin_lock_init(&adev->virt.rlcg_reg_lock);
+	spin_lock_init(&adev->wb.lock);
+
+	xa_init_flags(&adev->userq_xa, XA_FLAGS_LOCK_IRQ);
+
+	INIT_LIST_HEAD(&adev->reset_list);
+
+	INIT_LIST_HEAD(&adev->ras_list);
+
+	INIT_LIST_HEAD(&adev->pm.od_kobj_list);
 
-	INIT_LIST_HEAD(&adev->shadow_list);
-	mutex_init(&adev->shadow_list_lock);
+	xa_init(&adev->userq_doorbell_xa);
 
-	INIT_LIST_HEAD(&adev->gtt_list);
-	spin_lock_init(&adev->gtt_list_lock);
+	INIT_DELAYED_WORK(&adev->delayed_init_work,
+			  amdgpu_device_delayed_init_work_handler);
+	INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
+			  amdgpu_device_delay_enable_gfx_off);
+	/*
+	 * Initialize the enforce_isolation work structures for each XCP
+	 * partition.  This work handler is responsible for enforcing shader
+	 * isolation on AMD GPUs.  It counts the number of emitted fences for
+	 * each GFX and compute ring.  If there are any fences, it schedules
+	 * the `enforce_isolation_work` to be run after a delay.  If there are
+	 * no fences, it signals the Kernel Fusion Driver (KFD) to resume the
+	 * runqueue.
+	 */
+	for (i = 0; i < MAX_XCP; i++) {
+		INIT_DELAYED_WORK(&adev->gfx.enforce_isolation[i].work,
+				  amdgpu_gfx_enforce_isolation_handler);
+		adev->gfx.enforce_isolation[i].adev = adev;
+		adev->gfx.enforce_isolation[i].xcp_id = i;
+	}
 
+	INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func);
+	INIT_WORK(&adev->userq_reset_work, amdgpu_userq_reset_work);
+
+	adev->gfx.gfx_off_req_count = 1;
+	adev->gfx.gfx_off_residency = 0;
+	adev->gfx.gfx_off_entrycount = 0;
+	adev->pm.ac_power = power_supply_is_system_supplied() > 0;
+
+	atomic_set(&adev->throttling_logging_enabled, 1);
+	/*
+	 * If throttling continues, logging will be performed every minute
+	 * to avoid log flooding. "-1" is subtracted since the thermal
+	 * throttling interrupt comes every second. Thus, the total logging
+	 * interval is 59 seconds(retelimited printk interval) + 1(waiting
+	 * for throttling interrupt) = 60 seconds.
+	 */
+	ratelimit_state_init(&adev->throttling_logging_rs, (60 - 1) * HZ, 1);
+
+	ratelimit_set_flags(&adev->throttling_logging_rs, RATELIMIT_MSG_ON_RELEASE);
+
+	/* Registers mapping */
+	/* TODO: block userspace mapping of io register */
 	if (adev->asic_type >= CHIP_BONAIRE) {
 		adev->rmmio_base = pci_resource_start(adev->pdev, 5);
 		adev->rmmio_size = pci_resource_len(adev->pdev, 5);
@@ -1885,110 +4686,215 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 		adev->rmmio_size = pci_resource_len(adev->pdev, 2);
 	}
 
+	for (i = 0; i < AMD_IP_BLOCK_TYPE_NUM; i++)
+		atomic_set(&adev->pm.pwr_state[i], POWER_STATE_UNKNOWN);
+
 	adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size);
-	if (adev->rmmio == NULL) {
+	if (!adev->rmmio)
 		return -ENOMEM;
-	}
-	DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base);
-	DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size);
 
-	if (adev->asic_type >= CHIP_BONAIRE)
-		/* doorbell bar mapping */
-		amdgpu_doorbell_init(adev);
-
-	/* io port mapping */
-	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
-		if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) {
-			adev->rio_mem_size = pci_resource_len(adev->pdev, i);
-			adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size);
-			break;
-		}
+	dev_info(adev->dev, "register mmio base: 0x%08X\n",
+		 (uint32_t)adev->rmmio_base);
+	dev_info(adev->dev, "register mmio size: %u\n",
+		 (unsigned int)adev->rmmio_size);
+
+	/*
+	 * Reset domain needs to be present early, before XGMI hive discovered
+	 * (if any) and initialized to use reset sem and in_gpu reset flag
+	 * early on during init and before calling to RREG32.
+	 */
+	adev->reset_domain = amdgpu_reset_create_reset_domain(SINGLE_DEVICE, "amdgpu-reset-dev");
+	if (!adev->reset_domain)
+		return -ENOMEM;
+
+	/* detect hw virtualization here */
+	amdgpu_virt_init(adev);
+
+	amdgpu_device_get_pcie_info(adev);
+
+	r = amdgpu_device_get_job_timeout_settings(adev);
+	if (r) {
+		dev_err(adev->dev, "invalid lockup_timeout parameter syntax\n");
+		return r;
 	}
-	if (adev->rio_mem == NULL)
-		DRM_INFO("PCI I/O BAR is not found.\n");
 
+	amdgpu_device_set_mcbp(adev);
+
+	/*
+	 * By default, use default mode where all blocks are expected to be
+	 * initialized. At present a 'swinit' of blocks is required to be
+	 * completed before the need for a different level is detected.
+	 */
+	amdgpu_set_init_level(adev, AMDGPU_INIT_LEVEL_DEFAULT);
 	/* early init functions */
-	r = amdgpu_early_init(adev);
+	r = amdgpu_device_ip_early_init(adev);
 	if (r)
 		return r;
 
-	/* if we have > 1 VGA cards, then disable the amdgpu VGA resources */
-	/* this will fail for cards that aren't VGA class devices, just
-	 * ignore it */
-	vga_client_register(adev->pdev, adev, NULL, amdgpu_vga_set_decode);
-
-	if (amdgpu_runtime_pm == 1)
-		runtime = true;
-	if (amdgpu_device_is_px(ddev))
-		runtime = true;
-	if (!pci_is_thunderbolt_attached(adev->pdev))
-		vga_switcheroo_register_client(adev->pdev,
-					       &amdgpu_switcheroo_ops, runtime);
-	if (runtime)
-		vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain);
+	/*
+	 * No need to remove conflicting FBs for non-display class devices.
+	 * This prevents the sysfb from being freed accidently.
+	 */
+	if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA ||
+	    (pdev->class >> 8) == PCI_CLASS_DISPLAY_OTHER) {
+		/* Get rid of things like offb */
+		r = aperture_remove_conflicting_pci_devices(adev->pdev, amdgpu_kms_driver.name);
+		if (r)
+			return r;
+	}
 
-	/* Read BIOS */
-	if (!amdgpu_get_bios(adev)) {
-		r = -EINVAL;
-		goto failed;
+	/* Enable TMZ based on IP_VERSION */
+	amdgpu_gmc_tmz_set(adev);
+
+	if (amdgpu_sriov_vf(adev) &&
+	    amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(10, 3, 0))
+		/* VF MMIO access (except mailbox range) from CPU
+		 * will be blocked during sriov runtime
+		 */
+		adev->virt.caps |= AMDGPU_VF_MMIO_ACCESS_PROTECT;
+
+	amdgpu_gmc_noretry_set(adev);
+	/* Need to get xgmi info early to decide the reset behavior*/
+	if (adev->gmc.xgmi.supported) {
+		r = adev->gfxhub.funcs->get_xgmi_info(adev);
+		if (r)
+			return r;
 	}
 
-	r = amdgpu_atombios_init(adev);
-	if (r) {
-		dev_err(adev->dev, "amdgpu_atombios_init failed\n");
-		goto failed;
+	/* enable PCIE atomic ops */
+	if (amdgpu_sriov_vf(adev)) {
+		if (adev->virt.fw_reserve.p_pf2vf)
+			adev->have_atomics_support = ((struct amd_sriov_msg_pf2vf_info *)
+						      adev->virt.fw_reserve.p_pf2vf)->pcie_atomic_ops_support_flags ==
+				(PCI_EXP_DEVCAP2_ATOMIC_COMP32 | PCI_EXP_DEVCAP2_ATOMIC_COMP64);
+	/* APUs w/ gfx9 onwards doesn't reply on PCIe atomics, rather it is a
+	 * internal path natively support atomics, set have_atomics_support to true.
+	 */
+	} else if ((adev->flags & AMD_IS_APU) &&
+		   (amdgpu_ip_version(adev, GC_HWIP, 0) >
+		    IP_VERSION(9, 0, 0))) {
+		adev->have_atomics_support = true;
+	} else {
+		adev->have_atomics_support =
+			!pci_enable_atomic_ops_to_root(adev->pdev,
+					  PCI_EXP_DEVCAP2_ATOMIC_COMP32 |
+					  PCI_EXP_DEVCAP2_ATOMIC_COMP64);
 	}
 
+	if (!adev->have_atomics_support)
+		dev_info(adev->dev, "PCIE atomic ops is not supported\n");
+
+	/* doorbell bar mapping and doorbell index init*/
+	amdgpu_doorbell_init(adev);
+
+	if (amdgpu_emu_mode == 1) {
+		/* post the asic on emulation mode */
+		emu_soc_asic_init(adev);
+		goto fence_driver_init;
+	}
+
+	amdgpu_reset_init(adev);
+
 	/* detect if we are with an SRIOV vbios */
-	amdgpu_device_detect_sriov_bios(adev);
+	if (adev->bios)
+		amdgpu_device_detect_sriov_bios(adev);
+
+	/* check if we need to reset the asic
+	 *  E.g., driver was not cleanly unloaded previously, etc.
+	 */
+	if (!amdgpu_sriov_vf(adev) && amdgpu_asic_need_reset_on_init(adev)) {
+		if (adev->gmc.xgmi.num_physical_nodes) {
+			dev_info(adev->dev, "Pending hive reset.\n");
+			amdgpu_set_init_level(adev,
+					      AMDGPU_INIT_LEVEL_MINIMAL_XGMI);
+		} else if (amdgpu_ip_version(adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 10) &&
+				   !amdgpu_device_has_display_hardware(adev)) {
+					r = psp_gpu_reset(adev);
+		} else {
+				tmp = amdgpu_reset_method;
+				/* It should do a default reset when loading or reloading the driver,
+				 * regardless of the module parameter reset_method.
+				 */
+				amdgpu_reset_method = AMD_RESET_METHOD_NONE;
+				r = amdgpu_asic_reset(adev);
+				amdgpu_reset_method = tmp;
+		}
+
+		if (r) {
+		  dev_err(adev->dev, "asic reset on init failed\n");
+		  goto failed;
+		}
+	}
 
 	/* Post card if necessary */
-	if (amdgpu_vpost_needed(adev)) {
+	if (amdgpu_device_need_post(adev)) {
 		if (!adev->bios) {
 			dev_err(adev->dev, "no vBIOS found\n");
 			r = -EINVAL;
 			goto failed;
 		}
-		DRM_INFO("GPU posting now...\n");
-		r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
+		dev_info(adev->dev, "GPU posting now...\n");
+		r = amdgpu_device_asic_init(adev);
 		if (r) {
 			dev_err(adev->dev, "gpu post error!\n");
 			goto failed;
 		}
-	} else {
-		DRM_INFO("GPU post is not needed\n");
 	}
 
-	if (!adev->is_atom_fw) {
-		/* Initialize clocks */
-		r = amdgpu_atombios_get_clock_info(adev);
-		if (r) {
-			dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n");
-			return r;
+	if (adev->bios) {
+		if (adev->is_atom_fw) {
+			/* Initialize clocks */
+			r = amdgpu_atomfirmware_get_clock_info(adev);
+			if (r) {
+				dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n");
+				amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
+				goto failed;
+			}
+		} else {
+			/* Initialize clocks */
+			r = amdgpu_atombios_get_clock_info(adev);
+			if (r) {
+				dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n");
+				amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
+				goto failed;
+			}
+			/* init i2c buses */
+			amdgpu_i2c_init(adev);
 		}
-		/* init i2c buses */
-		amdgpu_atombios_i2c_init(adev);
 	}
 
+fence_driver_init:
 	/* Fence driver */
-	r = amdgpu_fence_driver_init(adev);
+	r = amdgpu_fence_driver_sw_init(adev);
 	if (r) {
-		dev_err(adev->dev, "amdgpu_fence_driver_init failed\n");
+		dev_err(adev->dev, "amdgpu_fence_driver_sw_init failed\n");
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0);
 		goto failed;
 	}
 
 	/* init the mode config */
-	drm_mode_config_init(adev->ddev);
+	drm_mode_config_init(adev_to_drm(adev));
 
-	r = amdgpu_init(adev);
+	r = amdgpu_device_ip_init(adev);
 	if (r) {
-		dev_err(adev->dev, "amdgpu_init failed\n");
-		amdgpu_fini(adev);
-		goto failed;
+		dev_err(adev->dev, "amdgpu_device_ip_init failed\n");
+		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
+		goto release_ras_con;
 	}
 
+	amdgpu_fence_driver_hw_init(adev);
+
+	dev_info(adev->dev,
+		"SE %d, SH per SE %d, CU per SH %d, active_cu_number %d\n",
+			adev->gfx.config.max_shader_engines,
+			adev->gfx.config.max_sh_per_se,
+			adev->gfx.config.max_cu_per_sh,
+			adev->gfx.cu_info.number);
+
 	adev->accel_working = true;
 
+	amdgpu_vm_check_compute_bug(adev);
+
 	/* Initialize the buffer migration limit. */
 	if (amdgpu_moverate >= 0)
 		max_MBps = amdgpu_moverate;
@@ -1997,342 +4903,673 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	/* Get a log2 for easy divisions. */
 	adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps));
 
-	r = amdgpu_ib_pool_init(adev);
-	if (r) {
-		dev_err(adev->dev, "IB initialization failed (%d).\n", r);
-		goto failed;
+	/*
+	 * Register gpu instance before amdgpu_device_enable_mgpu_fan_boost.
+	 * Otherwise the mgpu fan boost feature will be skipped due to the
+	 * gpu instance is counted less.
+	 */
+	amdgpu_register_gpu_instance(adev);
+
+	/* enable clockgating, etc. after ib tests, etc. since some blocks require
+	 * explicit gating rather than handling it automatically.
+	 */
+	if (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) {
+		r = amdgpu_device_ip_late_init(adev);
+		if (r) {
+			dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n");
+			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
+			goto release_ras_con;
+		}
+		/* must succeed. */
+		amdgpu_ras_resume(adev);
+		queue_delayed_work(system_wq, &adev->delayed_init_work,
+				   msecs_to_jiffies(AMDGPU_RESUME_MS));
 	}
 
-	r = amdgpu_ib_ring_tests(adev);
-	if (r)
-		DRM_ERROR("ib ring test failed (%d).\n", r);
+	if (amdgpu_sriov_vf(adev)) {
+		amdgpu_virt_release_full_gpu(adev, true);
+		flush_delayed_work(&adev->delayed_init_work);
+	}
 
-	amdgpu_fbdev_init(adev);
+	if (adev->init_lvl->level == AMDGPU_INIT_LEVEL_MINIMAL_XGMI)
+		amdgpu_xgmi_reset_on_init(adev);
+	/*
+	 * Place those sysfs registering after `late_init`. As some of those
+	 * operations performed in `late_init` might affect the sysfs
+	 * interfaces creating.
+	 */
+	r = amdgpu_device_sys_interface_init(adev);
 
-	r = amdgpu_gem_debugfs_init(adev);
+	if (IS_ENABLED(CONFIG_PERF_EVENTS))
+		r = amdgpu_pmu_init(adev);
 	if (r)
-		DRM_ERROR("registering gem debugfs failed (%d).\n", r);
+		dev_err(adev->dev, "amdgpu_pmu_init failed\n");
 
-	r = amdgpu_debugfs_regs_init(adev);
-	if (r)
-		DRM_ERROR("registering register debugfs failed (%d).\n", r);
+	/* Have stored pci confspace at hand for restore in sudden PCI error */
+	if (amdgpu_device_cache_pci_state(adev->pdev))
+		pci_restore_state(pdev);
 
-	r = amdgpu_debugfs_firmware_init(adev);
-	if (r)
-		DRM_ERROR("registering firmware debugfs failed (%d).\n", r);
+	/* if we have > 1 VGA cards, then disable the amdgpu VGA resources */
+	/* this will fail for cards that aren't VGA class devices, just
+	 * ignore it
+	 */
+	if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA)
+		vga_client_register(adev->pdev, amdgpu_device_vga_set_decode);
 
-	if ((amdgpu_testing & 1)) {
-		if (adev->accel_working)
-			amdgpu_test_moves(adev);
-		else
-			DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n");
-	}
-	if (amdgpu_benchmarking) {
-		if (adev->accel_working)
-			amdgpu_benchmark(adev, amdgpu_benchmarking);
-		else
-			DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n");
-	}
+	px = amdgpu_device_supports_px(adev);
 
-	/* enable clockgating, etc. after ib tests, etc. since some blocks require
-	 * explicit gating rather than handling it automatically.
-	 */
-	r = amdgpu_late_init(adev);
-	if (r) {
-		dev_err(adev->dev, "amdgpu_late_init failed\n");
+	if (px || (!dev_is_removable(&adev->pdev->dev) &&
+				apple_gmux_detect(NULL, NULL)))
+		vga_switcheroo_register_client(adev->pdev,
+					       &amdgpu_switcheroo_ops, px);
+
+	if (px)
+		vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain);
+
+	amdgpu_device_check_iommu_direct_map(adev);
+
+	adev->pm_nb.notifier_call = amdgpu_device_pm_notifier;
+	r = register_pm_notifier(&adev->pm_nb);
+	if (r)
 		goto failed;
-	}
 
 	return 0;
 
+release_ras_con:
+	if (amdgpu_sriov_vf(adev))
+		amdgpu_virt_release_full_gpu(adev, true);
+
+	/* failed in exclusive mode due to timeout */
+	if (amdgpu_sriov_vf(adev) &&
+		!amdgpu_sriov_runtime(adev) &&
+		amdgpu_virt_mmio_blocked(adev) &&
+		!amdgpu_virt_wait_reset(adev)) {
+		dev_err(adev->dev, "VF exclusive mode timeout\n");
+		/* Don't send request since VF is inactive. */
+		adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME;
+		adev->virt.ops = NULL;
+		r = -EAGAIN;
+	}
+	amdgpu_release_ras_context(adev);
+
 failed:
-	if (runtime)
-		vga_switcheroo_fini_domain_pm_ops(adev->dev);
+	amdgpu_vf_error_trans_all(adev);
+
 	return r;
 }
 
+static void amdgpu_device_unmap_mmio(struct amdgpu_device *adev)
+{
+
+	/* Clear all CPU mappings pointing to this device */
+	unmap_mapping_range(adev->ddev.anon_inode->i_mapping, 0, 0, 1);
+
+	/* Unmap all mapped bars - Doorbell, registers and VRAM */
+	amdgpu_doorbell_fini(adev);
+
+	iounmap(adev->rmmio);
+	adev->rmmio = NULL;
+	if (adev->mman.aper_base_kaddr)
+		iounmap(adev->mman.aper_base_kaddr);
+	adev->mman.aper_base_kaddr = NULL;
+
+	/* Memory manager related */
+	if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) {
+		arch_phys_wc_del(adev->gmc.vram_mtrr);
+		arch_io_free_memtype_wc(adev->gmc.aper_base, adev->gmc.aper_size);
+	}
+}
+
 /**
- * amdgpu_device_fini - tear down the driver
+ * amdgpu_device_fini_hw - tear down the driver
  *
  * @adev: amdgpu_device pointer
  *
  * Tear down the driver info (all asics).
  * Called at driver shutdown.
  */
-void amdgpu_device_fini(struct amdgpu_device *adev)
+void amdgpu_device_fini_hw(struct amdgpu_device *adev)
 {
-	int r;
+	dev_info(adev->dev, "amdgpu: finishing device.\n");
+	flush_delayed_work(&adev->delayed_init_work);
 
-	DRM_INFO("amdgpu: finishing device.\n");
+	if (adev->mman.initialized)
+		drain_workqueue(adev->mman.bdev.wq);
 	adev->shutdown = true;
-	if (adev->mode_info.mode_config_initialized)
-		drm_crtc_force_disable_all(adev->ddev);
-	/* evict vram memory */
-	amdgpu_bo_evict_vram(adev);
-	amdgpu_ib_pool_fini(adev);
-	amdgpu_fence_driver_fini(adev);
-	amdgpu_fbdev_fini(adev);
-	r = amdgpu_fini(adev);
+
+	unregister_pm_notifier(&adev->pm_nb);
+
+	/* make sure IB test finished before entering exclusive mode
+	 * to avoid preemption on IB test
+	 */
+	if (amdgpu_sriov_vf(adev)) {
+		amdgpu_virt_request_full_gpu(adev, false);
+		amdgpu_virt_fini_data_exchange(adev);
+	}
+
+	/* disable all interrupts */
+	amdgpu_irq_disable_all(adev);
+	if (adev->mode_info.mode_config_initialized) {
+		if (!drm_drv_uses_atomic_modeset(adev_to_drm(adev)))
+			drm_helper_force_disable_all(adev_to_drm(adev));
+		else
+			drm_atomic_helper_shutdown(adev_to_drm(adev));
+	}
+	amdgpu_fence_driver_hw_fini(adev);
+
+	amdgpu_device_sys_interface_fini(adev);
+
+	/* disable ras feature must before hw fini */
+	amdgpu_ras_pre_fini(adev);
+
+	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+
+	amdgpu_device_ip_fini_early(adev);
+
+	amdgpu_irq_fini_hw(adev);
+
+	if (adev->mman.initialized)
+		ttm_device_clear_dma_mappings(&adev->mman.bdev);
+
+	amdgpu_gart_dummy_page_fini(adev);
+
+	if (drm_dev_is_unplugged(adev_to_drm(adev)))
+		amdgpu_device_unmap_mmio(adev);
+
+}
+
+void amdgpu_device_fini_sw(struct amdgpu_device *adev)
+{
+	int i, idx;
+	bool px;
+
+	amdgpu_device_ip_fini(adev);
+	amdgpu_fence_driver_sw_fini(adev);
+	amdgpu_ucode_release(&adev->firmware.gpu_info_fw);
 	adev->accel_working = false;
+	dma_fence_put(rcu_dereference_protected(adev->gang_submit, true));
+	for (i = 0; i < MAX_XCP; ++i) {
+		dma_fence_put(adev->isolation[i].spearhead);
+		amdgpu_sync_free(&adev->isolation[i].active);
+		amdgpu_sync_free(&adev->isolation[i].prev);
+	}
+
+	amdgpu_reset_fini(adev);
+
 	/* free i2c buses */
 	amdgpu_i2c_fini(adev);
-	amdgpu_atombios_fini(adev);
-	kfree(adev->bios);
-	adev->bios = NULL;
-	if (!pci_is_thunderbolt_attached(adev->pdev))
+
+	if (adev->bios) {
+		if (amdgpu_emu_mode != 1)
+			amdgpu_atombios_fini(adev);
+		amdgpu_bios_release(adev);
+	}
+
+	kfree(adev->fru_info);
+	adev->fru_info = NULL;
+
+	kfree(adev->xcp_mgr);
+	adev->xcp_mgr = NULL;
+
+	px = amdgpu_device_supports_px(adev);
+
+	if (px || (!dev_is_removable(&adev->pdev->dev) &&
+				apple_gmux_detect(NULL, NULL)))
 		vga_switcheroo_unregister_client(adev->pdev);
-	if (adev->flags & AMD_IS_PX)
+
+	if (px)
 		vga_switcheroo_fini_domain_pm_ops(adev->dev);
-	vga_client_register(adev->pdev, NULL, NULL, NULL);
-	if (adev->rio_mem)
-		pci_iounmap(adev->pdev, adev->rio_mem);
-	adev->rio_mem = NULL;
-	iounmap(adev->rmmio);
-	adev->rmmio = NULL;
-	if (adev->asic_type >= CHIP_BONAIRE)
-		amdgpu_doorbell_fini(adev);
-	amdgpu_debugfs_regs_cleanup(adev);
+
+	if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA)
+		vga_client_unregister(adev->pdev);
+
+	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
+
+		iounmap(adev->rmmio);
+		adev->rmmio = NULL;
+		drm_dev_exit(idx);
+	}
+
+	if (IS_ENABLED(CONFIG_PERF_EVENTS))
+		amdgpu_pmu_fini(adev);
+	if (adev->discovery.bin)
+		amdgpu_discovery_fini(adev);
+
+	amdgpu_reset_put_reset_domain(adev->reset_domain);
+	adev->reset_domain = NULL;
+
+	kfree(adev->pci_state);
+	kfree(adev->pcie_reset_ctx.swds_pcistate);
+	kfree(adev->pcie_reset_ctx.swus_pcistate);
 }
 
+/**
+ * amdgpu_device_evict_resources - evict device resources
+ * @adev: amdgpu device object
+ *
+ * Evicts all ttm device resources(vram BOs, gart table) from the lru list
+ * of the vram memory type. Mainly used for evicting device resources
+ * at suspend time.
+ *
+ */
+static int amdgpu_device_evict_resources(struct amdgpu_device *adev)
+{
+	int ret;
+
+	/* No need to evict vram on APUs unless going to S4 */
+	if (!adev->in_s4 && (adev->flags & AMD_IS_APU))
+		return 0;
+
+	/* No need to evict when going to S5 through S4 callbacks */
+	if (system_state == SYSTEM_POWER_OFF)
+		return 0;
+
+	ret = amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM);
+	if (ret) {
+		dev_warn(adev->dev, "evicting device resources failed\n");
+		return ret;
+	}
+
+	if (adev->in_s4) {
+		ret = ttm_device_prepare_hibernation(&adev->mman.bdev);
+		if (ret)
+			dev_err(adev->dev, "prepare hibernation failed, %d\n", ret);
+	}
+	return ret;
+}
 
 /*
  * Suspend & resume.
  */
 /**
- * amdgpu_device_suspend - initiate device suspend
+ * amdgpu_device_pm_notifier - Notification block for Suspend/Hibernate events
+ * @nb: notifier block
+ * @mode: suspend mode
+ * @data: data
+ *
+ * This function is called when the system is about to suspend or hibernate.
+ * It is used to set the appropriate flags so that eviction can be optimized
+ * in the pm prepare callback.
+ */
+static int amdgpu_device_pm_notifier(struct notifier_block *nb, unsigned long mode,
+				     void *data)
+{
+	struct amdgpu_device *adev = container_of(nb, struct amdgpu_device, pm_nb);
+
+	switch (mode) {
+	case PM_HIBERNATION_PREPARE:
+		adev->in_s4 = true;
+		break;
+	case PM_POST_HIBERNATION:
+		adev->in_s4 = false;
+		break;
+	}
+
+	return NOTIFY_DONE;
+}
+
+/**
+ * amdgpu_device_prepare - prepare for device suspend
  *
- * @pdev: drm dev pointer
- * @state: suspend state
+ * @dev: drm dev pointer
  *
- * Puts the hw in the suspend state (all asics).
+ * Prepare to put the hw in the suspend state (all asics).
  * Returns 0 for success or an error on failure.
  * Called at driver suspend.
  */
-int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon)
+int amdgpu_device_prepare(struct drm_device *dev)
 {
-	struct amdgpu_device *adev;
-	struct drm_crtc *crtc;
-	struct drm_connector *connector;
-	int r;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	int i, r;
 
-	if (dev == NULL || dev->dev_private == NULL) {
-		return -ENODEV;
+	if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
+		return 0;
+
+	/* Evict the majority of BOs before starting suspend sequence */
+	r = amdgpu_device_evict_resources(adev);
+	if (r)
+		return r;
+
+	flush_delayed_work(&adev->gfx.gfx_off_delay_work);
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!adev->ip_blocks[i].status.valid)
+			continue;
+		if (!adev->ip_blocks[i].version->funcs->prepare_suspend)
+			continue;
+		r = adev->ip_blocks[i].version->funcs->prepare_suspend(&adev->ip_blocks[i]);
+		if (r)
+			return r;
+	}
+
+	return 0;
+}
+
+/**
+ * amdgpu_device_complete - complete power state transition
+ *
+ * @dev: drm dev pointer
+ *
+ * Undo the changes from amdgpu_device_prepare. This will be
+ * called on all resume transitions, including those that failed.
+ */
+void amdgpu_device_complete(struct drm_device *dev)
+{
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	int i;
+
+	for (i = 0; i < adev->num_ip_blocks; i++) {
+		if (!adev->ip_blocks[i].status.valid)
+			continue;
+		if (!adev->ip_blocks[i].version->funcs->complete)
+			continue;
+		adev->ip_blocks[i].version->funcs->complete(&adev->ip_blocks[i]);
 	}
+}
 
-	adev = dev->dev_private;
+/**
+ * amdgpu_device_suspend - initiate device suspend
+ *
+ * @dev: drm dev pointer
+ * @notify_clients: notify in-kernel DRM clients
+ *
+ * Puts the hw in the suspend state (all asics).
+ * Returns 0 for success or an error on failure.
+ * Called at driver suspend.
+ */
+int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
+{
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	int r, rec;
 
 	if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
 		return 0;
 
-	drm_kms_helper_poll_disable(dev);
+	adev->in_suspend = true;
 
-	/* turn off display hw */
-	drm_modeset_lock_all(dev);
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
-		drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
+	if (amdgpu_sriov_vf(adev)) {
+		if (!adev->in_runpm)
+			amdgpu_amdkfd_suspend_process(adev);
+		amdgpu_virt_fini_data_exchange(adev);
+		r = amdgpu_virt_request_full_gpu(adev, false);
+		if (r)
+			return r;
 	}
-	drm_modeset_unlock_all(dev);
 
-	/* unpin the front buffers and cursors */
-	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
-		struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
-		struct amdgpu_framebuffer *rfb = to_amdgpu_framebuffer(crtc->primary->fb);
-		struct amdgpu_bo *robj;
+	r = amdgpu_acpi_smart_shift_update(adev, AMDGPU_SS_DEV_D3);
+	if (r)
+		goto unwind_sriov;
 
-		if (amdgpu_crtc->cursor_bo) {
-			struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
-			r = amdgpu_bo_reserve(aobj, true);
-			if (r == 0) {
-				amdgpu_bo_unpin(aobj);
-				amdgpu_bo_unreserve(aobj);
-			}
-		}
+	if (notify_clients)
+		drm_client_dev_suspend(adev_to_drm(adev));
 
-		if (rfb == NULL || rfb->obj == NULL) {
-			continue;
-		}
-		robj = gem_to_amdgpu_bo(rfb->obj);
-		/* don't unpin kernel fb objects */
-		if (!amdgpu_fbdev_robj_is_fb(adev, robj)) {
-			r = amdgpu_bo_reserve(robj, true);
-			if (r == 0) {
-				amdgpu_bo_unpin(robj);
-				amdgpu_bo_unreserve(robj);
-			}
+	cancel_delayed_work_sync(&adev->delayed_init_work);
+
+	amdgpu_ras_suspend(adev);
+
+	r = amdgpu_device_ip_suspend_phase1(adev);
+	if (r)
+		goto unwind_smartshift;
+
+	amdgpu_amdkfd_suspend(adev, !amdgpu_sriov_vf(adev) && !adev->in_runpm);
+	r = amdgpu_userq_suspend(adev);
+	if (r)
+		goto unwind_ip_phase1;
+
+	r = amdgpu_device_evict_resources(adev);
+	if (r)
+		goto unwind_userq;
+
+	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+
+	amdgpu_fence_driver_hw_fini(adev);
+
+	r = amdgpu_device_ip_suspend_phase2(adev);
+	if (r)
+		goto unwind_evict;
+
+	if (amdgpu_sriov_vf(adev))
+		amdgpu_virt_release_full_gpu(adev, false);
+
+	return 0;
+
+unwind_evict:
+	if (adev->mman.buffer_funcs_ring->sched.ready)
+		amdgpu_ttm_set_buffer_funcs_status(adev, true);
+	amdgpu_fence_driver_hw_init(adev);
+
+unwind_userq:
+	rec = amdgpu_userq_resume(adev);
+	if (rec) {
+		dev_warn(adev->dev, "failed to re-initialize user queues: %d\n", rec);
+		return r;
+	}
+	rec = amdgpu_amdkfd_resume(adev, !amdgpu_sriov_vf(adev) && !adev->in_runpm);
+	if (rec) {
+		dev_warn(adev->dev, "failed to re-initialize kfd: %d\n", rec);
+		return r;
+	}
+
+unwind_ip_phase1:
+	/* suspend phase 1 = resume phase 3 */
+	rec = amdgpu_device_ip_resume_phase3(adev);
+	if (rec) {
+		dev_warn(adev->dev, "failed to re-initialize IPs phase1: %d\n", rec);
+		return r;
+	}
+
+unwind_smartshift:
+	rec = amdgpu_acpi_smart_shift_update(adev, AMDGPU_SS_DEV_D0);
+	if (rec) {
+		dev_warn(adev->dev, "failed to re-update smart shift: %d\n", rec);
+		return r;
+	}
+
+	if (notify_clients)
+		drm_client_dev_resume(adev_to_drm(adev));
+
+	amdgpu_ras_resume(adev);
+
+unwind_sriov:
+	if (amdgpu_sriov_vf(adev)) {
+		rec = amdgpu_virt_request_full_gpu(adev, true);
+		if (rec) {
+			dev_warn(adev->dev, "failed to reinitialize sriov: %d\n", rec);
+			return r;
 		}
 	}
-	/* evict vram memory */
-	amdgpu_bo_evict_vram(adev);
 
-	amdgpu_fence_driver_suspend(adev);
+	adev->in_suspend = adev->in_s0ix = adev->in_s3 = false;
+
+	return r;
+}
 
-	r = amdgpu_suspend(adev);
+static inline int amdgpu_virt_resume(struct amdgpu_device *adev)
+{
+	int r;
+	unsigned int prev_physical_node_id = adev->gmc.xgmi.physical_node_id;
 
-	/* evict remaining vram memory
-	 * This second call to evict vram is to evict the gart page table
-	 * using the CPU.
+	/* During VM resume, QEMU programming of VF MSIX table (register GFXMSIX_VECT0_ADDR_LO)
+	 * may not work. The access could be blocked by nBIF protection as VF isn't in
+	 * exclusive access mode. Exclusive access is enabled now, disable/enable MSIX
+	 * so that QEMU reprograms MSIX table.
 	 */
-	amdgpu_bo_evict_vram(adev);
+	amdgpu_restore_msix(adev);
 
-	if (adev->is_atom_fw)
-		amdgpu_atomfirmware_scratch_regs_save(adev);
-	else
-		amdgpu_atombios_scratch_regs_save(adev);
-	pci_save_state(dev->pdev);
-	if (suspend) {
-		/* Shut down the device */
-		pci_disable_device(dev->pdev);
-		pci_set_power_state(dev->pdev, PCI_D3hot);
-	} else {
-		r = amdgpu_asic_reset(adev);
-		if (r)
-			DRM_ERROR("amdgpu asic reset failed\n");
-	}
+	r = adev->gfxhub.funcs->get_xgmi_info(adev);
+	if (r)
+		return r;
+
+	dev_info(adev->dev, "xgmi node, old id %d, new id %d\n",
+		prev_physical_node_id, adev->gmc.xgmi.physical_node_id);
+
+	adev->vm_manager.vram_base_offset = adev->gfxhub.funcs->get_mc_fb_offset(adev);
+	adev->vm_manager.vram_base_offset +=
+		adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
 
-	if (fbcon) {
-		console_lock();
-		amdgpu_fbdev_set_suspend(adev, 1);
-		console_unlock();
-	}
 	return 0;
 }
 
 /**
  * amdgpu_device_resume - initiate device resume
  *
- * @pdev: drm dev pointer
+ * @dev: drm dev pointer
+ * @notify_clients: notify in-kernel DRM clients
  *
  * Bring the hw back to operating state (all asics).
  * Returns 0 for success or an error on failure.
  * Called at driver resume.
  */
-int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon)
+int amdgpu_device_resume(struct drm_device *dev, bool notify_clients)
 {
-	struct drm_connector *connector;
-	struct amdgpu_device *adev = dev->dev_private;
-	struct drm_crtc *crtc;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	int r = 0;
 
-	if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
-		return 0;
-
-	if (fbcon)
-		console_lock();
+	if (amdgpu_sriov_vf(adev)) {
+		r = amdgpu_virt_request_full_gpu(adev, true);
+		if (r)
+			return r;
+	}
 
-	if (resume) {
-		pci_set_power_state(dev->pdev, PCI_D0);
-		pci_restore_state(dev->pdev);
-		r = pci_enable_device(dev->pdev);
+	if (amdgpu_virt_xgmi_migrate_enabled(adev)) {
+		r = amdgpu_virt_resume(adev);
 		if (r)
-			goto unlock;
+			goto exit;
 	}
-	if (adev->is_atom_fw)
-		amdgpu_atomfirmware_scratch_regs_restore(adev);
-	else
-		amdgpu_atombios_scratch_regs_restore(adev);
+
+	if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
+		return 0;
+
+	if (adev->in_s0ix)
+		amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D0Entry);
 
 	/* post card */
-	if (amdgpu_need_post(adev)) {
-		r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
+	if (amdgpu_device_need_post(adev)) {
+		r = amdgpu_device_asic_init(adev);
 		if (r)
-			DRM_ERROR("amdgpu asic init failed\n");
+			dev_err(adev->dev, "amdgpu asic init failed\n");
 	}
 
-	r = amdgpu_resume(adev);
+	r = amdgpu_device_ip_resume(adev);
+
 	if (r) {
-		DRM_ERROR("amdgpu_resume failed (%d).\n", r);
-		goto unlock;
+		dev_err(adev->dev, "amdgpu_device_ip_resume failed (%d).\n", r);
+		goto exit;
 	}
-	amdgpu_fence_driver_resume(adev);
 
-	if (resume) {
-		r = amdgpu_ib_ring_tests(adev);
-		if (r)
-			DRM_ERROR("ib ring test failed (%d).\n", r);
-	}
+	r = amdgpu_amdkfd_resume(adev, !amdgpu_sriov_vf(adev) && !adev->in_runpm);
+	if (r)
+		goto exit;
 
-	r = amdgpu_late_init(adev);
+	r = amdgpu_userq_resume(adev);
 	if (r)
-		goto unlock;
-
-	/* pin cursors */
-	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
-		struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
-
-		if (amdgpu_crtc->cursor_bo) {
-			struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
-			r = amdgpu_bo_reserve(aobj, true);
-			if (r == 0) {
-				r = amdgpu_bo_pin(aobj,
-						  AMDGPU_GEM_DOMAIN_VRAM,
-						  &amdgpu_crtc->cursor_addr);
-				if (r != 0)
-					DRM_ERROR("Failed to pin cursor BO (%d)\n", r);
-				amdgpu_bo_unreserve(aobj);
-			}
-		}
-	}
+		goto exit;
 
-	/* blat the mode back in */
-	if (fbcon) {
-		drm_helper_resume_force_mode(dev);
-		/* turn on display hw */
-		drm_modeset_lock_all(dev);
-		list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
-			drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON);
-		}
-		drm_modeset_unlock_all(dev);
+	r = amdgpu_device_ip_late_init(adev);
+	if (r)
+		goto exit;
+
+	queue_delayed_work(system_wq, &adev->delayed_init_work,
+			   msecs_to_jiffies(AMDGPU_RESUME_MS));
+exit:
+	if (amdgpu_sriov_vf(adev)) {
+		amdgpu_virt_init_data_exchange(adev);
+		amdgpu_virt_release_full_gpu(adev, true);
+
+		if (!r && !adev->in_runpm)
+			r = amdgpu_amdkfd_resume_process(adev);
 	}
 
-	drm_kms_helper_poll_enable(dev);
+	if (r)
+		return r;
 
-	/*
-	 * Most of the connector probing functions try to acquire runtime pm
-	 * refs to ensure that the GPU is powered on when connector polling is
-	 * performed. Since we're calling this from a runtime PM callback,
-	 * trying to acquire rpm refs will cause us to deadlock.
-	 *
-	 * Since we're guaranteed to be holding the rpm lock, it's safe to
-	 * temporarily disable the rpm helpers so this doesn't deadlock us.
-	 */
+	/* Make sure IB tests flushed */
+	flush_delayed_work(&adev->delayed_init_work);
+
+	if (notify_clients)
+		drm_client_dev_resume(adev_to_drm(adev));
+
+	amdgpu_ras_resume(adev);
+
+	if (adev->mode_info.num_crtc) {
+		/*
+		 * Most of the connector probing functions try to acquire runtime pm
+		 * refs to ensure that the GPU is powered on when connector polling is
+		 * performed. Since we're calling this from a runtime PM callback,
+		 * trying to acquire rpm refs will cause us to deadlock.
+		 *
+		 * Since we're guaranteed to be holding the rpm lock, it's safe to
+		 * temporarily disable the rpm helpers so this doesn't deadlock us.
+		 */
 #ifdef CONFIG_PM
-	dev->dev->power.disable_depth++;
+		dev->dev->power.disable_depth++;
 #endif
-	drm_helper_hpd_irq_event(dev);
+		if (!adev->dc_enabled)
+			drm_helper_hpd_irq_event(dev);
+		else
+			drm_kms_helper_hotplug_event(dev);
 #ifdef CONFIG_PM
-	dev->dev->power.disable_depth--;
+		dev->dev->power.disable_depth--;
 #endif
+	}
 
-	if (fbcon)
-		amdgpu_fbdev_set_suspend(adev, 0);
+	amdgpu_vram_mgr_clear_reset_blocks(adev);
+	adev->in_suspend = false;
 
-unlock:
-	if (fbcon)
-		console_unlock();
+	if (amdgpu_acpi_smart_shift_update(adev, AMDGPU_SS_DEV_D0))
+		dev_warn(adev->dev, "smart shift update failed\n");
 
-	return r;
+	return 0;
 }
 
-static bool amdgpu_check_soft_reset(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_ip_check_soft_reset - did soft reset succeed
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * The list of all the hardware IPs that make up the asic is walked and
+ * the check_soft_reset callbacks are run.  check_soft_reset determines
+ * if the asic is still hung or not.
+ * Returns true if any of the IPs are still in a hung state, false if not.
+ */
+static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev)
 {
 	int i;
 	bool asic_hang = false;
 
+	if (amdgpu_sriov_vf(adev))
+		return true;
+
+	if (amdgpu_asic_need_full_reset(adev))
+		return true;
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
 		if (adev->ip_blocks[i].version->funcs->check_soft_reset)
 			adev->ip_blocks[i].status.hang =
-				adev->ip_blocks[i].version->funcs->check_soft_reset(adev);
+				adev->ip_blocks[i].version->funcs->check_soft_reset(
+					&adev->ip_blocks[i]);
 		if (adev->ip_blocks[i].status.hang) {
-			DRM_INFO("IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name);
+			dev_info(adev->dev, "IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name);
 			asic_hang = true;
 		}
 	}
 	return asic_hang;
 }
 
-static int amdgpu_pre_soft_reset(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_ip_pre_soft_reset - prepare for soft reset
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * The list of all the hardware IPs that make up the asic is walked and the
+ * pre_soft_reset callbacks are run if the block is hung.  pre_soft_reset
+ * handles any IP specific hardware or software state changes that are
+ * necessary for a soft reset to succeed.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev)
 {
 	int i, r = 0;
 
@@ -2341,7 +5578,7 @@ static int amdgpu_pre_soft_reset(struct amdgpu_device *adev)
 			continue;
 		if (adev->ip_blocks[i].status.hang &&
 		    adev->ip_blocks[i].version->funcs->pre_soft_reset) {
-			r = adev->ip_blocks[i].version->funcs->pre_soft_reset(adev);
+			r = adev->ip_blocks[i].version->funcs->pre_soft_reset(&adev->ip_blocks[i]);
 			if (r)
 				return r;
 		}
@@ -2350,19 +5587,32 @@ static int amdgpu_pre_soft_reset(struct amdgpu_device *adev)
 	return 0;
 }
 
-static bool amdgpu_need_full_reset(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_ip_need_full_reset - check if a full asic reset is needed
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Some hardware IPs cannot be soft reset.  If they are hung, a full gpu
+ * reset is necessary to recover.
+ * Returns true if a full asic reset is required, false if not.
+ */
+static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev)
 {
 	int i;
 
+	if (amdgpu_asic_need_full_reset(adev))
+		return true;
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
 		if ((adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) ||
 		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) ||
 		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_ACP) ||
-		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE)) {
+		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) ||
+		     adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) {
 			if (adev->ip_blocks[i].status.hang) {
-				DRM_INFO("Some block need full reset!\n");
+				dev_info(adev->dev, "Some block need full reset!\n");
 				return true;
 			}
 		}
@@ -2370,7 +5620,18 @@ static bool amdgpu_need_full_reset(struct amdgpu_device *adev)
 	return false;
 }
 
-static int amdgpu_soft_reset(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_ip_soft_reset - do a soft reset
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * The list of all the hardware IPs that make up the asic is walked and the
+ * soft_reset callbacks are run if the block is hung.  soft_reset handles any
+ * IP specific hardware or software state changes that are necessary to soft
+ * reset the IP.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev)
 {
 	int i, r = 0;
 
@@ -2379,7 +5640,7 @@ static int amdgpu_soft_reset(struct amdgpu_device *adev)
 			continue;
 		if (adev->ip_blocks[i].status.hang &&
 		    adev->ip_blocks[i].version->funcs->soft_reset) {
-			r = adev->ip_blocks[i].version->funcs->soft_reset(adev);
+			r = adev->ip_blocks[i].version->funcs->soft_reset(&adev->ip_blocks[i]);
 			if (r)
 				return r;
 		}
@@ -2388,7 +5649,18 @@ static int amdgpu_soft_reset(struct amdgpu_device *adev)
 	return 0;
 }
 
-static int amdgpu_post_soft_reset(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_ip_post_soft_reset - clean up from soft reset
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * The list of all the hardware IPs that make up the asic is walked and the
+ * post_soft_reset callbacks are run if the asic was hung.  post_soft_reset
+ * handles any IP specific hardware or software state changes that are
+ * necessary after the IP has been soft reset.
+ * Returns 0 on success, negative error code on failure.
+ */
+static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev)
 {
 	int i, r = 0;
 
@@ -2397,7 +5669,7 @@ static int amdgpu_post_soft_reset(struct amdgpu_device *adev)
 			continue;
 		if (adev->ip_blocks[i].status.hang &&
 		    adev->ip_blocks[i].version->funcs->post_soft_reset)
-			r = adev->ip_blocks[i].version->funcs->post_soft_reset(adev);
+			r = adev->ip_blocks[i].version->funcs->post_soft_reset(&adev->ip_blocks[i]);
 		if (r)
 			return r;
 	}
@@ -2405,325 +5677,1143 @@ static int amdgpu_post_soft_reset(struct amdgpu_device *adev)
 	return 0;
 }
 
-bool amdgpu_need_backup(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf
+ *
+ * @adev: amdgpu_device pointer
+ * @reset_context: amdgpu reset context pointer
+ *
+ * do VF FLR and reinitialize Asic
+ * return 0 means succeeded otherwise failed
+ */
+static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
+				     struct amdgpu_reset_context *reset_context)
 {
-	if (adev->flags & AMD_IS_APU)
-		return false;
+	int r;
+	struct amdgpu_hive_info *hive = NULL;
 
-	return amdgpu_lockup_timeout > 0 ? true : false;
-}
+	if (test_bit(AMDGPU_HOST_FLR, &reset_context->flags)) {
+		if (!amdgpu_ras_get_fed_status(adev))
+			amdgpu_virt_ready_to_reset(adev);
+		amdgpu_virt_wait_reset(adev);
+		clear_bit(AMDGPU_HOST_FLR, &reset_context->flags);
+		r = amdgpu_virt_request_full_gpu(adev, true);
+	} else {
+		r = amdgpu_virt_reset_gpu(adev);
+	}
+	if (r)
+		return r;
 
-static int amdgpu_recover_vram_from_shadow(struct amdgpu_device *adev,
-					   struct amdgpu_ring *ring,
-					   struct amdgpu_bo *bo,
-					   struct dma_fence **fence)
-{
-	uint32_t domain;
-	int r;
+	amdgpu_ras_clear_err_state(adev);
+	amdgpu_irq_gpu_reset_resume_helper(adev);
 
-	if (!bo->shadow)
-		return 0;
+	/* some sw clean up VF needs to do before recover */
+	amdgpu_virt_post_reset(adev);
 
-	r = amdgpu_bo_reserve(bo, true);
+	/* Resume IP prior to SMC */
+	r = amdgpu_device_ip_reinit_early_sriov(adev);
 	if (r)
 		return r;
-	domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-	/* if bo has been evicted, then no need to recover */
-	if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
-		r = amdgpu_bo_validate(bo->shadow);
-		if (r) {
-			DRM_ERROR("bo validate failed!\n");
-			goto err;
-		}
 
-		r = amdgpu_ttm_bind(&bo->shadow->tbo, &bo->shadow->tbo.mem);
-		if (r) {
-			DRM_ERROR("%p bind failed\n", bo->shadow);
-			goto err;
-		}
+	amdgpu_virt_init_data_exchange(adev);
 
-		r = amdgpu_bo_restore_from_shadow(adev, ring, bo,
-						 NULL, fence, true);
-		if (r) {
-			DRM_ERROR("recover page table failed!\n");
-			goto err;
-		}
+	r = amdgpu_device_fw_loading(adev);
+	if (r)
+		return r;
+
+	/* now we are okay to resume SMC/CP/SDMA */
+	r = amdgpu_device_ip_reinit_late_sriov(adev);
+	if (r)
+		return r;
+
+	hive = amdgpu_get_xgmi_hive(adev);
+	/* Update PSP FW topology after reset */
+	if (hive && adev->gmc.xgmi.num_physical_nodes > 1)
+		r = amdgpu_xgmi_update_topology(hive, adev);
+	if (hive)
+		amdgpu_put_xgmi_hive(hive);
+	if (r)
+		return r;
+
+	r = amdgpu_ib_ring_tests(adev);
+	if (r)
+		return r;
+
+	if (adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST)
+		amdgpu_inc_vram_lost(adev);
+
+	/* need to be called during full access so we can't do it later like
+	 * bare-metal does.
+	 */
+	amdgpu_amdkfd_post_reset(adev);
+	amdgpu_virt_release_full_gpu(adev, true);
+
+	/* Aldebaran and gfx_11_0_3 support ras in SRIOV, so need resume ras during reset */
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 5, 0) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(11, 0, 3))
+		amdgpu_ras_resume(adev);
+
+	amdgpu_virt_ras_telemetry_post_reset(adev);
+
+	return 0;
+}
+
+/**
+ * amdgpu_device_has_job_running - check if there is any unfinished job
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * check if there is any job running on the device when guest driver receives
+ * FLR notification from host driver. If there are still jobs running, then
+ * the guest driver will not respond the FLR reset. Instead, let the job hit
+ * the timeout and guest driver then issue the reset request.
+ */
+bool amdgpu_device_has_job_running(struct amdgpu_device *adev)
+{
+	int i;
+
+	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
+		struct amdgpu_ring *ring = adev->rings[i];
+
+		if (!amdgpu_ring_sched_ready(ring))
+			continue;
+
+		if (amdgpu_fence_count_emitted(ring))
+			return true;
 	}
-err:
-	amdgpu_bo_unreserve(bo);
-	return r;
+	return false;
 }
 
 /**
- * amdgpu_sriov_gpu_reset - reset the asic
+ * amdgpu_device_should_recover_gpu - check if we should try GPU recovery
  *
- * @adev: amdgpu device pointer
- * @voluntary: if this reset is requested by guest.
- *             (true means by guest and false means by HYPERVISOR )
+ * @adev: amdgpu_device pointer
  *
- * Attempt the reset the GPU if it has hung (all asics).
- * for SRIOV case.
- * Returns 0 for success or an error on failure.
+ * Check amdgpu_gpu_recovery and SRIOV status to see if we should try to recover
+ * a hung GPU.
  */
-int amdgpu_sriov_gpu_reset(struct amdgpu_device *adev, bool voluntary)
+bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev)
+{
+
+	if (amdgpu_gpu_recovery == 0)
+		goto disabled;
+
+	/* Skip soft reset check in fatal error mode */
+	if (!amdgpu_ras_is_poison_mode_supported(adev))
+		return true;
+
+	if (amdgpu_sriov_vf(adev))
+		return true;
+
+	if (amdgpu_gpu_recovery == -1) {
+		switch (adev->asic_type) {
+#ifdef CONFIG_DRM_AMDGPU_SI
+		case CHIP_VERDE:
+		case CHIP_TAHITI:
+		case CHIP_PITCAIRN:
+		case CHIP_OLAND:
+		case CHIP_HAINAN:
+#endif
+#ifdef CONFIG_DRM_AMDGPU_CIK
+		case CHIP_KAVERI:
+		case CHIP_KABINI:
+		case CHIP_MULLINS:
+#endif
+		case CHIP_CARRIZO:
+		case CHIP_STONEY:
+		case CHIP_CYAN_SKILLFISH:
+			goto disabled;
+		default:
+			break;
+		}
+	}
+
+	return true;
+
+disabled:
+		dev_info(adev->dev, "GPU recovery disabled.\n");
+		return false;
+}
+
+int amdgpu_device_mode1_reset(struct amdgpu_device *adev)
+{
+	u32 i;
+	int ret = 0;
+
+	if (adev->bios)
+		amdgpu_atombios_scratch_regs_engine_hung(adev, true);
+
+	dev_info(adev->dev, "GPU mode1 reset\n");
+
+	/* Cache the state before bus master disable. The saved config space
+	 * values are used in other cases like restore after mode-2 reset.
+	 */
+	amdgpu_device_cache_pci_state(adev->pdev);
+
+	/* disable BM */
+	pci_clear_master(adev->pdev);
+
+	if (amdgpu_dpm_is_mode1_reset_supported(adev)) {
+		dev_info(adev->dev, "GPU smu mode1 reset\n");
+		ret = amdgpu_dpm_mode1_reset(adev);
+	} else {
+		dev_info(adev->dev, "GPU psp mode1 reset\n");
+		ret = psp_gpu_reset(adev);
+	}
+
+	if (ret)
+		goto mode1_reset_failed;
+
+	amdgpu_device_load_pci_state(adev->pdev);
+	ret = amdgpu_psp_wait_for_bootloader(adev);
+	if (ret)
+		goto mode1_reset_failed;
+
+	/* wait for asic to come out of reset */
+	for (i = 0; i < adev->usec_timeout; i++) {
+		u32 memsize = adev->nbio.funcs->get_memsize(adev);
+
+		if (memsize != 0xffffffff)
+			break;
+		udelay(1);
+	}
+
+	if (i >= adev->usec_timeout) {
+		ret = -ETIMEDOUT;
+		goto mode1_reset_failed;
+	}
+
+	if (adev->bios)
+		amdgpu_atombios_scratch_regs_engine_hung(adev, false);
+
+	return 0;
+
+mode1_reset_failed:
+	dev_err(adev->dev, "GPU mode1 reset failed\n");
+	return ret;
+}
+
+int amdgpu_device_link_reset(struct amdgpu_device *adev)
+{
+	int ret = 0;
+
+	dev_info(adev->dev, "GPU link reset\n");
+
+	if (!amdgpu_reset_in_dpc(adev))
+		ret = amdgpu_dpm_link_reset(adev);
+
+	if (ret)
+		goto link_reset_failed;
+
+	ret = amdgpu_psp_wait_for_bootloader(adev);
+	if (ret)
+		goto link_reset_failed;
+
+	return 0;
+
+link_reset_failed:
+	dev_err(adev->dev, "GPU link reset failed\n");
+	return ret;
+}
+
+int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
+				 struct amdgpu_reset_context *reset_context)
 {
 	int i, r = 0;
-	int resched;
-	struct amdgpu_bo *bo, *tmp;
-	struct amdgpu_ring *ring;
-	struct dma_fence *fence = NULL, *next = NULL;
+	struct amdgpu_job *job = NULL;
+	struct amdgpu_device *tmp_adev = reset_context->reset_req_dev;
+	bool need_full_reset =
+		test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
+
+	if (reset_context->reset_req_dev == adev)
+		job = reset_context->job;
 
-	mutex_lock(&adev->virt.lock_reset);
-	atomic_inc(&adev->gpu_reset_counter);
-	adev->gfx.in_reset = true;
+	if (amdgpu_sriov_vf(adev))
+		amdgpu_virt_pre_reset(adev);
 
-	/* block TTM */
-	resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev);
+	amdgpu_fence_driver_isr_toggle(adev, true);
 
-	/* block scheduler */
+	/* block all schedulers and reset given job's ring */
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		ring = adev->rings[i];
+		struct amdgpu_ring *ring = adev->rings[i];
 
-		if (!ring || !ring->sched.thread)
+		if (!amdgpu_ring_sched_ready(ring))
 			continue;
 
-		kthread_park(ring->sched.thread);
-		amd_sched_hw_job_reset(&ring->sched);
+		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
+		amdgpu_fence_driver_force_completion(ring);
 	}
 
-	/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
-	amdgpu_fence_driver_force_completion(adev);
+	amdgpu_fence_driver_isr_toggle(adev, false);
+
+	if (job && job->vm)
+		drm_sched_increase_karma(&job->base);
 
-	/* request to take full control of GPU before re-initialization  */
-	if (voluntary)
-		amdgpu_virt_reset_gpu(adev);
+	r = amdgpu_reset_prepare_hwcontext(adev, reset_context);
+	/* If reset handler not implemented, continue; otherwise return */
+	if (r == -EOPNOTSUPP)
+		r = 0;
 	else
-		amdgpu_virt_request_full_gpu(adev, true);
+		return r;
 
+	/* Don't suspend on bare metal if we are not going to HW reset the ASIC */
+	if (!amdgpu_sriov_vf(adev)) {
 
-	/* Resume IP prior to SMC */
-	amdgpu_sriov_reinit_early(adev);
+		if (!need_full_reset)
+			need_full_reset = amdgpu_device_ip_need_full_reset(adev);
 
-	/* we need recover gart prior to run SMC/CP/SDMA resume */
-	amdgpu_ttm_recover_gart(adev);
+		if (!need_full_reset && amdgpu_gpu_recovery &&
+		    amdgpu_device_ip_check_soft_reset(adev)) {
+			amdgpu_device_ip_pre_soft_reset(adev);
+			r = amdgpu_device_ip_soft_reset(adev);
+			amdgpu_device_ip_post_soft_reset(adev);
+			if (r || amdgpu_device_ip_check_soft_reset(adev)) {
+				dev_info(adev->dev, "soft reset failed, will fallback to full reset!\n");
+				need_full_reset = true;
+			}
+		}
 
-	/* now we are okay to resume SMC/CP/SDMA */
-	amdgpu_sriov_reinit_late(adev);
+		if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags)) {
+			dev_info(tmp_adev->dev, "Dumping IP State\n");
+			/* Trigger ip dump before we reset the asic */
+			for (i = 0; i < tmp_adev->num_ip_blocks; i++)
+				if (tmp_adev->ip_blocks[i].version->funcs->dump_ip_state)
+					tmp_adev->ip_blocks[i].version->funcs
+						->dump_ip_state((void *)&tmp_adev->ip_blocks[i]);
+			dev_info(tmp_adev->dev, "Dumping IP State Completed\n");
+		}
 
-	amdgpu_irq_gpu_reset_resume_helper(adev);
+		if (need_full_reset)
+			r = amdgpu_device_ip_suspend(adev);
+		if (need_full_reset)
+			set_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
+		else
+			clear_bit(AMDGPU_NEED_FULL_RESET,
+				  &reset_context->flags);
+	}
 
-	if (amdgpu_ib_ring_tests(adev))
-		dev_err(adev->dev, "[GPU_RESET] ib ring test failed (%d).\n", r);
+	return r;
+}
 
-	/* release full control of GPU after ib test */
-	amdgpu_virt_release_full_gpu(adev, true);
+int amdgpu_device_reinit_after_reset(struct amdgpu_reset_context *reset_context)
+{
+	struct list_head *device_list_handle;
+	bool full_reset, vram_lost = false;
+	struct amdgpu_device *tmp_adev;
+	int r, init_level;
 
-	DRM_INFO("recover vram bo from shadow\n");
+	device_list_handle = reset_context->reset_device_list;
 
-	ring = adev->mman.buffer_funcs_ring;
-	mutex_lock(&adev->shadow_list_lock);
-	list_for_each_entry_safe(bo, tmp, &adev->shadow_list, shadow_list) {
-		next = NULL;
-		amdgpu_recover_vram_from_shadow(adev, ring, bo, &next);
-		if (fence) {
-			r = dma_fence_wait(fence, false);
+	if (!device_list_handle)
+		return -EINVAL;
+
+	full_reset = test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
+
+	/**
+	 * If it's reset on init, it's default init level, otherwise keep level
+	 * as recovery level.
+	 */
+	if (reset_context->method == AMD_RESET_METHOD_ON_INIT)
+			init_level = AMDGPU_INIT_LEVEL_DEFAULT;
+	else
+			init_level = AMDGPU_INIT_LEVEL_RESET_RECOVERY;
+
+	r = 0;
+	list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
+		amdgpu_set_init_level(tmp_adev, init_level);
+		if (full_reset) {
+			/* post card */
+			amdgpu_reset_set_dpc_status(tmp_adev, false);
+			amdgpu_ras_clear_err_state(tmp_adev);
+			r = amdgpu_device_asic_init(tmp_adev);
 			if (r) {
-				WARN(r, "recovery from shadow isn't completed\n");
-				break;
+				dev_warn(tmp_adev->dev, "asic atom init failed!");
+			} else {
+				dev_info(tmp_adev->dev, "GPU reset succeeded, trying to resume\n");
+
+				r = amdgpu_device_ip_resume_phase1(tmp_adev);
+				if (r)
+					goto out;
+
+				vram_lost = amdgpu_device_check_vram_lost(tmp_adev);
+
+				if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags))
+					amdgpu_coredump(tmp_adev, false, vram_lost, reset_context->job);
+
+				if (vram_lost) {
+					dev_info(
+						tmp_adev->dev,
+						"VRAM is lost due to GPU reset!\n");
+					amdgpu_inc_vram_lost(tmp_adev);
+				}
+
+				r = amdgpu_device_fw_loading(tmp_adev);
+				if (r)
+					return r;
+
+				r = amdgpu_xcp_restore_partition_mode(
+					tmp_adev->xcp_mgr);
+				if (r)
+					goto out;
+
+				r = amdgpu_device_ip_resume_phase2(tmp_adev);
+				if (r)
+					goto out;
+
+				if (tmp_adev->mman.buffer_funcs_ring->sched.ready)
+					amdgpu_ttm_set_buffer_funcs_status(tmp_adev, true);
+
+				r = amdgpu_device_ip_resume_phase3(tmp_adev);
+				if (r)
+					goto out;
+
+				if (vram_lost)
+					amdgpu_device_fill_reset_magic(tmp_adev);
+
+				/*
+				 * Add this ASIC as tracked as reset was already
+				 * complete successfully.
+				 */
+				amdgpu_register_gpu_instance(tmp_adev);
+
+				if (!reset_context->hive &&
+				    tmp_adev->gmc.xgmi.num_physical_nodes > 1)
+					amdgpu_xgmi_add_device(tmp_adev);
+
+				r = amdgpu_device_ip_late_init(tmp_adev);
+				if (r)
+					goto out;
+
+				r = amdgpu_userq_post_reset(tmp_adev, vram_lost);
+				if (r)
+					goto out;
+
+				drm_client_dev_resume(adev_to_drm(tmp_adev));
+
+				/*
+				 * The GPU enters bad state once faulty pages
+				 * by ECC has reached the threshold, and ras
+				 * recovery is scheduled next. So add one check
+				 * here to break recovery if it indeed exceeds
+				 * bad page threshold, and remind user to
+				 * retire this GPU or setting one bigger
+				 * bad_page_threshold value to fix this once
+				 * probing driver again.
+				 */
+				if (!amdgpu_ras_is_rma(tmp_adev)) {
+					/* must succeed. */
+					amdgpu_ras_resume(tmp_adev);
+				} else {
+					r = -EINVAL;
+					goto out;
+				}
+
+				/* Update PSP FW topology after reset */
+				if (reset_context->hive &&
+				    tmp_adev->gmc.xgmi.num_physical_nodes > 1)
+					r = amdgpu_xgmi_update_topology(
+						reset_context->hive, tmp_adev);
 			}
 		}
 
-		dma_fence_put(fence);
-		fence = next;
-	}
-	mutex_unlock(&adev->shadow_list_lock);
+out:
+		if (!r) {
+			/* IP init is complete now, set level as default */
+			amdgpu_set_init_level(tmp_adev,
+					      AMDGPU_INIT_LEVEL_DEFAULT);
+			amdgpu_irq_gpu_reset_resume_helper(tmp_adev);
+			r = amdgpu_ib_ring_tests(tmp_adev);
+			if (r) {
+				dev_err(tmp_adev->dev, "ib ring test failed (%d).\n", r);
+				r = -EAGAIN;
+				goto end;
+			}
+		}
 
-	if (fence) {
-		r = dma_fence_wait(fence, false);
 		if (r)
-			WARN(r, "recovery from shadow isn't completed\n");
+			tmp_adev->asic_reset_res = r;
 	}
-	dma_fence_put(fence);
 
-	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		if (!ring || !ring->sched.thread)
-			continue;
+end:
+	return r;
+}
+
+int amdgpu_do_asic_reset(struct list_head *device_list_handle,
+			 struct amdgpu_reset_context *reset_context)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+	bool need_full_reset, skip_hw_reset;
+	int r = 0;
+
+	/* Try reset handler method first */
+	tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device,
+				    reset_list);
+
+	reset_context->reset_device_list = device_list_handle;
+	r = amdgpu_reset_perform_reset(tmp_adev, reset_context);
+	/* If reset handler not implemented, continue; otherwise return */
+	if (r == -EOPNOTSUPP)
+		r = 0;
+	else
+		return r;
+
+	/* Reset handler not implemented, use the default method */
+	need_full_reset =
+		test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
+	skip_hw_reset = test_bit(AMDGPU_SKIP_HW_RESET, &reset_context->flags);
+
+	/*
+	 * ASIC reset has to be done on all XGMI hive nodes ASAP
+	 * to allow proper links negotiation in FW (within 1 sec)
+	 */
+	if (!skip_hw_reset && need_full_reset) {
+		list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
+			/* For XGMI run all resets in parallel to speed up the process */
+			if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
+				if (!queue_work(system_unbound_wq,
+						&tmp_adev->xgmi_reset_work))
+					r = -EALREADY;
+			} else
+				r = amdgpu_asic_reset(tmp_adev);
 
-		amd_sched_job_recovery(&ring->sched);
-		kthread_unpark(ring->sched.thread);
+			if (r) {
+				dev_err(tmp_adev->dev,
+					"ASIC reset failed with error, %d for drm dev, %s",
+					r, adev_to_drm(tmp_adev)->unique);
+				goto out;
+			}
+		}
+
+		/* For XGMI wait for all resets to complete before proceed */
+		if (!r) {
+			list_for_each_entry(tmp_adev, device_list_handle,
+					    reset_list) {
+				if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
+					flush_work(&tmp_adev->xgmi_reset_work);
+					r = tmp_adev->asic_reset_res;
+					if (r)
+						break;
+				}
+			}
+		}
 	}
 
-	drm_helper_resume_force_mode(adev->ddev);
-	ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched);
-	if (r) {
-		/* bad news, how to tell it to userspace ? */
-		dev_info(adev->dev, "GPU reset failed\n");
+	if (!r && amdgpu_ras_intr_triggered()) {
+		list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
+			amdgpu_ras_reset_error_count(tmp_adev,
+						     AMDGPU_RAS_BLOCK__MMHUB);
+		}
+
+		amdgpu_ras_intr_cleared();
 	}
 
-	adev->gfx.in_reset = false;
-	mutex_unlock(&adev->virt.lock_reset);
+	r = amdgpu_device_reinit_after_reset(reset_context);
+	if (r == -EAGAIN)
+		set_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
+	else
+		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags);
+
+out:
 	return r;
 }
 
-/**
- * amdgpu_gpu_reset - reset the asic
- *
- * @adev: amdgpu device pointer
- *
- * Attempt the reset the GPU if it has hung (all asics).
- * Returns 0 for success or an error on failure.
- */
-int amdgpu_gpu_reset(struct amdgpu_device *adev)
+static void amdgpu_device_set_mp1_state(struct amdgpu_device *adev)
 {
-	int i, r;
-	int resched;
-	bool need_full_reset;
 
-	if (amdgpu_sriov_vf(adev))
-		return amdgpu_sriov_gpu_reset(adev, true);
+	switch (amdgpu_asic_reset_method(adev)) {
+	case AMD_RESET_METHOD_MODE1:
+	case AMD_RESET_METHOD_LINK:
+		adev->mp1_state = PP_MP1_STATE_SHUTDOWN;
+		break;
+	case AMD_RESET_METHOD_MODE2:
+		adev->mp1_state = PP_MP1_STATE_RESET;
+		break;
+	default:
+		adev->mp1_state = PP_MP1_STATE_NONE;
+		break;
+	}
+}
 
-	if (!amdgpu_check_soft_reset(adev)) {
-		DRM_INFO("No hardware hang detected. Did some blocks stall?\n");
-		return 0;
+static void amdgpu_device_unset_mp1_state(struct amdgpu_device *adev)
+{
+	amdgpu_vf_error_trans_all(adev);
+	adev->mp1_state = PP_MP1_STATE_NONE;
+}
+
+static void amdgpu_device_resume_display_audio(struct amdgpu_device *adev)
+{
+	struct pci_dev *p = NULL;
+
+	p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus),
+			adev->pdev->bus->number, 1);
+	if (p) {
+		pm_runtime_enable(&(p->dev));
+		pm_runtime_resume(&(p->dev));
 	}
 
-	atomic_inc(&adev->gpu_reset_counter);
+	pci_dev_put(p);
+}
 
-	/* block TTM */
-	resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev);
+static int amdgpu_device_suspend_display_audio(struct amdgpu_device *adev)
+{
+	enum amd_reset_method reset_method;
+	struct pci_dev *p = NULL;
+	u64 expires;
 
-	/* block scheduler */
-	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-		struct amdgpu_ring *ring = adev->rings[i];
+	/*
+	 * For now, only BACO and mode1 reset are confirmed
+	 * to suffer the audio issue without proper suspended.
+	 */
+	reset_method = amdgpu_asic_reset_method(adev);
+	if ((reset_method != AMD_RESET_METHOD_BACO) &&
+	     (reset_method != AMD_RESET_METHOD_MODE1))
+		return -EINVAL;
 
-		if (!ring || !ring->sched.thread)
-			continue;
-		kthread_park(ring->sched.thread);
-		amd_sched_hw_job_reset(&ring->sched);
+	p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus),
+			adev->pdev->bus->number, 1);
+	if (!p)
+		return -ENODEV;
+
+	expires = pm_runtime_autosuspend_expiration(&(p->dev));
+	if (!expires)
+		/*
+		 * If we cannot get the audio device autosuspend delay,
+		 * a fixed 4S interval will be used. Considering 3S is
+		 * the audio controller default autosuspend delay setting.
+		 * 4S used here is guaranteed to cover that.
+		 */
+		expires = ktime_get_mono_fast_ns() + NSEC_PER_SEC * 4ULL;
+
+	while (!pm_runtime_status_suspended(&(p->dev))) {
+		if (!pm_runtime_suspend(&(p->dev)))
+			break;
+
+		if (expires < ktime_get_mono_fast_ns()) {
+			dev_warn(adev->dev, "failed to suspend display audio\n");
+			pci_dev_put(p);
+			/* TODO: abort the succeeding gpu reset? */
+			return -ETIMEDOUT;
+		}
+	}
+
+	pm_runtime_disable(&(p->dev));
+
+	pci_dev_put(p);
+	return 0;
+}
+
+static inline void amdgpu_device_stop_pending_resets(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+
+#if defined(CONFIG_DEBUG_FS)
+	if (!amdgpu_sriov_vf(adev))
+		cancel_work(&adev->reset_work);
+#endif
+	cancel_work(&adev->userq_reset_work);
+
+	if (adev->kfd.dev)
+		cancel_work(&adev->kfd.reset_work);
+
+	if (amdgpu_sriov_vf(adev))
+		cancel_work(&adev->virt.flr_work);
+
+	if (con && adev->ras_enabled)
+		cancel_work(&con->recovery_work);
+
+}
+
+static int amdgpu_device_health_check(struct list_head *device_list_handle)
+{
+	struct amdgpu_device *tmp_adev;
+	int ret = 0;
+
+	list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
+		ret |= amdgpu_device_bus_status_check(tmp_adev);
 	}
-	/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
-	amdgpu_fence_driver_force_completion(adev);
 
-	need_full_reset = amdgpu_need_full_reset(adev);
+	return ret;
+}
+
+static void amdgpu_device_recovery_prepare(struct amdgpu_device *adev,
+					  struct list_head *device_list,
+					  struct amdgpu_hive_info *hive)
+{
+	struct amdgpu_device *tmp_adev = NULL;
 
-	if (!need_full_reset) {
-		amdgpu_pre_soft_reset(adev);
-		r = amdgpu_soft_reset(adev);
-		amdgpu_post_soft_reset(adev);
-		if (r || amdgpu_check_soft_reset(adev)) {
-			DRM_INFO("soft reset failed, will fallback to full reset!\n");
-			need_full_reset = true;
+	/*
+	 * Build list of devices to reset.
+	 * In case we are in XGMI hive mode, resort the device list
+	 * to put adev in the 1st position.
+	 */
+	if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1) && hive) {
+		list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
+			list_add_tail(&tmp_adev->reset_list, device_list);
+			if (adev->shutdown)
+				tmp_adev->shutdown = true;
+			if (amdgpu_reset_in_dpc(adev))
+				tmp_adev->pcie_reset_ctx.in_link_reset = true;
 		}
+		if (!list_is_first(&adev->reset_list, device_list))
+			list_rotate_to_front(&adev->reset_list, device_list);
+	} else {
+		list_add_tail(&adev->reset_list, device_list);
 	}
+}
+
+static void amdgpu_device_recovery_get_reset_lock(struct amdgpu_device *adev,
+						  struct list_head *device_list)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+
+	if (list_empty(device_list))
+		return;
+	tmp_adev =
+		list_first_entry(device_list, struct amdgpu_device, reset_list);
+	amdgpu_device_lock_reset_domain(tmp_adev->reset_domain);
+}
+
+static void amdgpu_device_recovery_put_reset_lock(struct amdgpu_device *adev,
+						  struct list_head *device_list)
+{
+	struct amdgpu_device *tmp_adev = NULL;
 
-	if (need_full_reset) {
-		r = amdgpu_suspend(adev);
+	if (list_empty(device_list))
+		return;
+	tmp_adev =
+		list_first_entry(device_list, struct amdgpu_device, reset_list);
+	amdgpu_device_unlock_reset_domain(tmp_adev->reset_domain);
+}
+
+static void amdgpu_device_halt_activities(struct amdgpu_device *adev,
+					  struct amdgpu_job *job,
+					  struct amdgpu_reset_context *reset_context,
+					  struct list_head *device_list,
+					  struct amdgpu_hive_info *hive,
+					  bool need_emergency_restart)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+	int i;
+
+	/* block all schedulers and reset given job's ring */
+	list_for_each_entry(tmp_adev, device_list, reset_list) {
+		amdgpu_device_set_mp1_state(tmp_adev);
+
+		/*
+		 * Try to put the audio codec into suspend state
+		 * before gpu reset started.
+		 *
+		 * Due to the power domain of the graphics device
+		 * is shared with AZ power domain. Without this,
+		 * we may change the audio hardware from behind
+		 * the audio driver's back. That will trigger
+		 * some audio codec errors.
+		 */
+		if (!amdgpu_device_suspend_display_audio(tmp_adev))
+			tmp_adev->pcie_reset_ctx.audio_suspended = true;
+
+		amdgpu_ras_set_error_query_ready(tmp_adev, false);
+
+		cancel_delayed_work_sync(&tmp_adev->delayed_init_work);
+
+		amdgpu_amdkfd_pre_reset(tmp_adev, reset_context);
+
+		/*
+		 * Mark these ASICs to be reset as untracked first
+		 * And add them back after reset completed
+		 */
+		amdgpu_unregister_gpu_instance(tmp_adev);
+
+		drm_client_dev_suspend(adev_to_drm(tmp_adev));
+
+		/* disable ras on ALL IPs */
+		if (!need_emergency_restart && !amdgpu_reset_in_dpc(adev) &&
+		    amdgpu_device_ip_need_full_reset(tmp_adev))
+			amdgpu_ras_suspend(tmp_adev);
+
+		amdgpu_userq_pre_reset(tmp_adev);
+
+		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
+			struct amdgpu_ring *ring = tmp_adev->rings[i];
 
-retry:
-		/* Disable fb access */
-		if (adev->mode_info.num_crtc) {
-			struct amdgpu_mode_mc_save save;
-			amdgpu_display_stop_mc_access(adev, &save);
-			amdgpu_wait_for_idle(adev, AMD_IP_BLOCK_TYPE_GMC);
+			if (!amdgpu_ring_sched_ready(ring))
+				continue;
+
+			drm_sched_stop(&ring->sched, job ? &job->base : NULL);
+
+			if (need_emergency_restart)
+				amdgpu_job_stop_all_jobs_on_sched(&ring->sched);
 		}
-		if (adev->is_atom_fw)
-			amdgpu_atomfirmware_scratch_regs_save(adev);
-		else
-			amdgpu_atombios_scratch_regs_save(adev);
-		r = amdgpu_asic_reset(adev);
-		if (adev->is_atom_fw)
-			amdgpu_atomfirmware_scratch_regs_restore(adev);
-		else
-			amdgpu_atombios_scratch_regs_restore(adev);
-		/* post card */
-		amdgpu_atom_asic_init(adev->mode_info.atom_context);
+		atomic_inc(&tmp_adev->gpu_reset_counter);
+	}
+}
 
-		if (!r) {
-			dev_info(adev->dev, "GPU reset succeeded, trying to resume\n");
-			r = amdgpu_resume(adev);
+static int amdgpu_device_asic_reset(struct amdgpu_device *adev,
+			      struct list_head *device_list,
+			      struct amdgpu_reset_context *reset_context)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+	int retry_limit = AMDGPU_MAX_RETRY_LIMIT;
+	int r = 0;
+
+retry:	/* Rest of adevs pre asic reset from XGMI hive. */
+	list_for_each_entry(tmp_adev, device_list, reset_list) {
+		r = amdgpu_device_pre_asic_reset(tmp_adev, reset_context);
+		/*TODO Should we stop ?*/
+		if (r) {
+			dev_err(tmp_adev->dev, "GPU pre asic reset failed with err, %d for drm dev, %s ",
+				  r, adev_to_drm(tmp_adev)->unique);
+			tmp_adev->asic_reset_res = r;
 		}
 	}
-	if (!r) {
-		amdgpu_irq_gpu_reset_resume_helper(adev);
-		if (need_full_reset && amdgpu_need_backup(adev)) {
-			r = amdgpu_ttm_recover_gart(adev);
-			if (r)
-				DRM_ERROR("gart recovery failed!!!\n");
+
+	/* Actual ASIC resets if needed.*/
+	/* Host driver will handle XGMI hive reset for SRIOV */
+	if (amdgpu_sriov_vf(adev)) {
+
+		/* Bail out of reset early */
+		if (amdgpu_ras_is_rma(adev))
+			return -ENODEV;
+
+		if (amdgpu_ras_get_fed_status(adev) || amdgpu_virt_rcvd_ras_interrupt(adev)) {
+			dev_dbg(adev->dev, "Detected RAS error, wait for FLR completion\n");
+			amdgpu_ras_set_fed(adev, true);
+			set_bit(AMDGPU_HOST_FLR, &reset_context->flags);
 		}
-		r = amdgpu_ib_ring_tests(adev);
-		if (r) {
-			dev_err(adev->dev, "ib ring test failed (%d).\n", r);
-			r = amdgpu_suspend(adev);
-			need_full_reset = true;
+
+		r = amdgpu_device_reset_sriov(adev, reset_context);
+		if (AMDGPU_RETRY_SRIOV_RESET(r) && (retry_limit--) > 0) {
+			amdgpu_virt_release_full_gpu(adev, true);
 			goto retry;
 		}
-		/**
-		 * recovery vm page tables, since we cannot depend on VRAM is
-		 * consistent after gpu full reset.
+		if (r)
+			adev->asic_reset_res = r;
+	} else {
+		r = amdgpu_do_asic_reset(device_list, reset_context);
+		if (r && r == -EAGAIN)
+			goto retry;
+	}
+
+	list_for_each_entry(tmp_adev, device_list, reset_list) {
+		/*
+		 * Drop any pending non scheduler resets queued before reset is done.
+		 * Any reset scheduled after this point would be valid. Scheduler resets
+		 * were already dropped during drm_sched_stop and no new ones can come
+		 * in before drm_sched_start.
 		 */
-		if (need_full_reset && amdgpu_need_backup(adev)) {
-			struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
-			struct amdgpu_bo *bo, *tmp;
-			struct dma_fence *fence = NULL, *next = NULL;
-
-			DRM_INFO("recover vram bo from shadow\n");
-			mutex_lock(&adev->shadow_list_lock);
-			list_for_each_entry_safe(bo, tmp, &adev->shadow_list, shadow_list) {
-				next = NULL;
-				amdgpu_recover_vram_from_shadow(adev, ring, bo, &next);
-				if (fence) {
-					r = dma_fence_wait(fence, false);
-					if (r) {
-						WARN(r, "recovery from shadow isn't completed\n");
-						break;
-					}
-				}
+		amdgpu_device_stop_pending_resets(tmp_adev);
+	}
+
+	return r;
+}
+
+static int amdgpu_device_sched_resume(struct list_head *device_list,
+			      struct amdgpu_reset_context *reset_context,
+			      bool   job_signaled)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+	int i, r = 0;
+
+	/* Post ASIC reset for all devs .*/
+	list_for_each_entry(tmp_adev, device_list, reset_list) {
 
-				dma_fence_put(fence);
-				fence = next;
-			}
-			mutex_unlock(&adev->shadow_list_lock);
-			if (fence) {
-				r = dma_fence_wait(fence, false);
-				if (r)
-					WARN(r, "recovery from shadow isn't completed\n");
-			}
-			dma_fence_put(fence);
-		}
 		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-			struct amdgpu_ring *ring = adev->rings[i];
+			struct amdgpu_ring *ring = tmp_adev->rings[i];
 
-			if (!ring || !ring->sched.thread)
+			if (!amdgpu_ring_sched_ready(ring))
 				continue;
 
-			amd_sched_job_recovery(&ring->sched);
-			kthread_unpark(ring->sched.thread);
+			drm_sched_start(&ring->sched, 0);
 		}
-	} else {
-		dev_err(adev->dev, "asic resume failed (%d).\n", r);
-		for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
-			if (adev->rings[i] && adev->rings[i]->sched.thread) {
-				kthread_unpark(adev->rings[i]->sched.thread);
-			}
+
+		if (!drm_drv_uses_atomic_modeset(adev_to_drm(tmp_adev)) && !job_signaled)
+			drm_helper_resume_force_mode(adev_to_drm(tmp_adev));
+
+		if (tmp_adev->asic_reset_res) {
+			/* bad news, how to tell it to userspace ?
+			 * for ras error, we should report GPU bad status instead of
+			 * reset failure
+			 */
+			if (reset_context->src != AMDGPU_RESET_SRC_RAS ||
+			    !amdgpu_ras_eeprom_check_err_threshold(tmp_adev))
+				dev_info(
+					tmp_adev->dev,
+					"GPU reset(%d) failed with error %d \n",
+					atomic_read(
+						&tmp_adev->gpu_reset_counter),
+					tmp_adev->asic_reset_res);
+			amdgpu_vf_error_put(tmp_adev,
+					    AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0,
+					    tmp_adev->asic_reset_res);
+			if (!r)
+				r = tmp_adev->asic_reset_res;
+			tmp_adev->asic_reset_res = 0;
+		} else {
+			dev_info(tmp_adev->dev, "GPU reset(%d) succeeded!\n",
+				 atomic_read(&tmp_adev->gpu_reset_counter));
+			if (amdgpu_acpi_smart_shift_update(tmp_adev,
+							   AMDGPU_SS_DEV_D0))
+				dev_warn(tmp_adev->dev,
+					 "smart shift update failed\n");
 		}
 	}
 
-	drm_helper_resume_force_mode(adev->ddev);
+	return r;
+}
+
+static void amdgpu_device_gpu_resume(struct amdgpu_device *adev,
+			      struct list_head *device_list,
+			      bool   need_emergency_restart)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+
+	list_for_each_entry(tmp_adev, device_list, reset_list) {
+		/* unlock kfd: SRIOV would do it separately */
+		if (!need_emergency_restart && !amdgpu_sriov_vf(tmp_adev))
+			amdgpu_amdkfd_post_reset(tmp_adev);
+
+		/* kfd_post_reset will do nothing if kfd device is not initialized,
+		 * need to bring up kfd here if it's not be initialized before
+		 */
+		if (!adev->kfd.init_complete)
+			amdgpu_amdkfd_device_init(adev);
+
+		if (tmp_adev->pcie_reset_ctx.audio_suspended)
+			amdgpu_device_resume_display_audio(tmp_adev);
+
+		amdgpu_device_unset_mp1_state(tmp_adev);
+
+		amdgpu_ras_set_error_query_ready(tmp_adev, true);
 
-	ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched);
-	if (r) {
-		/* bad news, how to tell it to userspace ? */
-		dev_info(adev->dev, "GPU reset failed\n");
+	}
+}
+
+
+/**
+ * amdgpu_device_gpu_recover - reset the asic and recover scheduler
+ *
+ * @adev: amdgpu_device pointer
+ * @job: which job trigger hang
+ * @reset_context: amdgpu reset context pointer
+ *
+ * Attempt to reset the GPU if it has hung (all asics).
+ * Attempt to do soft-reset or full-reset and reinitialize Asic
+ * Returns 0 for success or an error on failure.
+ */
+
+int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
+			      struct amdgpu_job *job,
+			      struct amdgpu_reset_context *reset_context)
+{
+	struct list_head device_list;
+	bool job_signaled = false;
+	struct amdgpu_hive_info *hive = NULL;
+	int r = 0;
+	bool need_emergency_restart = false;
+
+	/*
+	 * If it reaches here because of hang/timeout and a RAS error is
+	 * detected at the same time, let RAS recovery take care of it.
+	 */
+	if (amdgpu_ras_is_err_state(adev, AMDGPU_RAS_BLOCK__ANY) &&
+	    !amdgpu_sriov_vf(adev) &&
+	    reset_context->src != AMDGPU_RESET_SRC_RAS) {
+		dev_dbg(adev->dev,
+			"Gpu recovery from source: %d yielding to RAS error recovery handling",
+			reset_context->src);
+		return 0;
+	}
+
+	/*
+	 * Special case: RAS triggered and full reset isn't supported
+	 */
+	need_emergency_restart = amdgpu_ras_need_emergency_restart(adev);
+
+	/*
+	 * Flush RAM to disk so that after reboot
+	 * the user can read log and see why the system rebooted.
+	 */
+	if (need_emergency_restart && amdgpu_ras_get_context(adev) &&
+		amdgpu_ras_get_context(adev)->reboot) {
+		dev_warn(adev->dev, "Emergency reboot.");
+
+		ksys_sync_helper();
+		emergency_restart();
+	}
+
+	dev_info(adev->dev, "GPU %s begin!. Source:  %d\n",
+		 need_emergency_restart ? "jobs stop" : "reset",
+		 reset_context->src);
+
+	if (!amdgpu_sriov_vf(adev))
+		hive = amdgpu_get_xgmi_hive(adev);
+	if (hive)
+		mutex_lock(&hive->hive_lock);
+
+	reset_context->job = job;
+	reset_context->hive = hive;
+	INIT_LIST_HEAD(&device_list);
+
+	amdgpu_device_recovery_prepare(adev, &device_list, hive);
+
+	if (!amdgpu_sriov_vf(adev)) {
+		r = amdgpu_device_health_check(&device_list);
+		if (r)
+			goto end_reset;
+	}
+
+	/* Cannot be called after locking reset domain */
+	amdgpu_ras_pre_reset(adev, &device_list);
+
+	/* We need to lock reset domain only once both for XGMI and single device */
+	amdgpu_device_recovery_get_reset_lock(adev, &device_list);
+
+	amdgpu_device_halt_activities(adev, job, reset_context, &device_list,
+				      hive, need_emergency_restart);
+	if (need_emergency_restart)
+		goto skip_sched_resume;
+	/*
+	 * Must check guilty signal here since after this point all old
+	 * HW fences are force signaled.
+	 *
+	 * job->base holds a reference to parent fence
+	 */
+	if (job && dma_fence_is_signaled(&job->hw_fence->base)) {
+		job_signaled = true;
+		dev_info(adev->dev, "Guilty job already signaled, skipping HW reset");
+		goto skip_hw_reset;
+	}
+
+	r = amdgpu_device_asic_reset(adev, &device_list, reset_context);
+	if (r)
+		goto reset_unlock;
+skip_hw_reset:
+	r = amdgpu_device_sched_resume(&device_list, reset_context, job_signaled);
+	if (r)
+		goto reset_unlock;
+skip_sched_resume:
+	amdgpu_device_gpu_resume(adev, &device_list, need_emergency_restart);
+reset_unlock:
+	amdgpu_device_recovery_put_reset_lock(adev, &device_list);
+	amdgpu_ras_post_reset(adev, &device_list);
+end_reset:
+	if (hive) {
+		mutex_unlock(&hive->hive_lock);
+		amdgpu_put_xgmi_hive(hive);
+	}
+
+	if (r)
+		dev_info(adev->dev, "GPU reset end with ret = %d\n", r);
+
+	atomic_set(&adev->reset_domain->reset_res, r);
+
+	if (!r) {
+		struct amdgpu_task_info *ti = NULL;
+
+		if (job)
+			ti = amdgpu_vm_get_task_info_pasid(adev, job->pasid);
+
+		drm_dev_wedged_event(adev_to_drm(adev), DRM_WEDGE_RECOVERY_NONE,
+				     ti ? &ti->task : NULL);
+
+		amdgpu_vm_put_task_info(ti);
 	}
 
 	return r;
 }
 
-void amdgpu_get_pcie_info(struct amdgpu_device *adev)
+/**
+ * amdgpu_device_partner_bandwidth - find the bandwidth of appropriate partner
+ *
+ * @adev: amdgpu_device pointer
+ * @speed: pointer to the speed of the link
+ * @width: pointer to the width of the link
+ *
+ * Evaluate the hierarchy to find the speed and bandwidth capabilities of the
+ * first physical partner to an AMD dGPU.
+ * This will exclude any virtual switches and links.
+ */
+static void amdgpu_device_partner_bandwidth(struct amdgpu_device *adev,
+					    enum pci_bus_speed *speed,
+					    enum pcie_link_width *width)
 {
-	u32 mask;
-	int ret;
+	struct pci_dev *parent = adev->pdev;
+
+	if (!speed || !width)
+		return;
+
+	*speed = PCI_SPEED_UNKNOWN;
+	*width = PCIE_LNK_WIDTH_UNKNOWN;
+
+	if (amdgpu_device_pcie_dynamic_switching_supported(adev)) {
+		while ((parent = pci_upstream_bridge(parent))) {
+			/* skip upstream/downstream switches internal to dGPU*/
+			if (parent->vendor == PCI_VENDOR_ID_ATI)
+				continue;
+			*speed = pcie_get_speed_cap(parent);
+			*width = pcie_get_width_cap(parent);
+			break;
+		}
+	} else {
+		/* use the current speeds rather than max if switching is not supported */
+		pcie_bandwidth_available(adev->pdev, NULL, speed, width);
+	}
+}
+
+/**
+ * amdgpu_device_gpu_bandwidth - find the bandwidth of the GPU
+ *
+ * @adev: amdgpu_device pointer
+ * @speed: pointer to the speed of the link
+ * @width: pointer to the width of the link
+ *
+ * Evaluate the hierarchy to find the speed and bandwidth capabilities of the
+ * AMD dGPU which may be a virtual upstream bridge.
+ */
+static void amdgpu_device_gpu_bandwidth(struct amdgpu_device *adev,
+					enum pci_bus_speed *speed,
+					enum pcie_link_width *width)
+{
+	struct pci_dev *parent = adev->pdev;
+
+	if (!speed || !width)
+		return;
+
+	parent = pci_upstream_bridge(parent);
+	if (parent && parent->vendor == PCI_VENDOR_ID_ATI) {
+		/* use the upstream/downstream switches internal to dGPU */
+		*speed = pcie_get_speed_cap(parent);
+		*width = pcie_get_width_cap(parent);
+		while ((parent = pci_upstream_bridge(parent))) {
+			if (parent->vendor == PCI_VENDOR_ID_ATI) {
+				/* use the upstream/downstream switches internal to dGPU */
+				*speed = pcie_get_speed_cap(parent);
+				*width = pcie_get_width_cap(parent);
+			}
+		}
+	} else {
+		/* use the device itself */
+		*speed = pcie_get_speed_cap(adev->pdev);
+		*width = pcie_get_width_cap(adev->pdev);
+	}
+}
+
+/**
+ * amdgpu_device_get_pcie_info - fence pcie info about the PCIE slot
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Fetches and stores in the driver the PCIE capabilities (gen speed
+ * and lanes) of the slot the device is in. Handles APUs and
+ * virtualized environments where PCIE config space may not be available.
+ */
+static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev)
+{
+	enum pci_bus_speed speed_cap, platform_speed_cap;
+	enum pcie_link_width platform_link_width, link_width;
 
 	if (amdgpu_pcie_gen_cap)
 		adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap;
@@ -2732,7 +6822,7 @@ void amdgpu_get_pcie_info(struct amdgpu_device *adev)
 		adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap;
 
 	/* covers APUs as well */
-	if (pci_is_root_bus(adev->pdev->bus)) {
+	if (pci_is_root_bus(adev->pdev->bus) && !amdgpu_passthrough(adev)) {
 		if (adev->pm.pcie_gen_mask == 0)
 			adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK;
 		if (adev->pm.pcie_mlw_mask == 0)
@@ -2740,773 +6830,1008 @@ void amdgpu_get_pcie_info(struct amdgpu_device *adev)
 		return;
 	}
 
+	if (adev->pm.pcie_gen_mask && adev->pm.pcie_mlw_mask)
+		return;
+
+	amdgpu_device_partner_bandwidth(adev, &platform_speed_cap,
+					&platform_link_width);
+	amdgpu_device_gpu_bandwidth(adev, &speed_cap, &link_width);
+
 	if (adev->pm.pcie_gen_mask == 0) {
-		ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask);
-		if (!ret) {
-			adev->pm.pcie_gen_mask = (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+		/* asic caps */
+		if (speed_cap == PCI_SPEED_UNKNOWN) {
+			adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
 						  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
 						  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3);
-
-			if (mask & DRM_PCIE_SPEED_25)
-				adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1;
-			if (mask & DRM_PCIE_SPEED_50)
-				adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2;
-			if (mask & DRM_PCIE_SPEED_80)
-				adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3;
 		} else {
-			adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK;
+			if (speed_cap == PCIE_SPEED_32_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN5);
+			else if (speed_cap == PCIE_SPEED_16_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4);
+			else if (speed_cap == PCIE_SPEED_8_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3);
+			else if (speed_cap == PCIE_SPEED_5_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2);
+			else
+				adev->pm.pcie_gen_mask |= CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1;
+		}
+		/* platform caps */
+		if (platform_speed_cap == PCI_SPEED_UNKNOWN) {
+			adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+						   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2);
+		} else {
+			if (platform_speed_cap == PCIE_SPEED_32_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN5);
+			else if (platform_speed_cap == PCIE_SPEED_16_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4);
+			else if (platform_speed_cap == PCIE_SPEED_8_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3);
+			else if (platform_speed_cap == PCIE_SPEED_5_0GT)
+				adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
+							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2);
+			else
+				adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1;
+
 		}
 	}
 	if (adev->pm.pcie_mlw_mask == 0) {
-		ret = drm_pcie_get_max_link_width(adev->ddev, &mask);
-		if (!ret) {
-			switch (mask) {
-			case 32:
-				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+		/* asic caps */
+		if (link_width == PCIE_LNK_WIDTH_UNKNOWN) {
+			adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_ASIC_PCIE_MLW_MASK;
+		} else {
+			switch (link_width) {
+			case PCIE_LNK_X32:
+				adev->pm.pcie_mlw_mask |= (CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X32 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X16 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X12 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X1);
 				break;
-			case 16:
-				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+			case PCIE_LNK_X16:
+				adev->pm.pcie_mlw_mask |= (CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X16 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X12 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X1);
 				break;
-			case 12:
-				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+			case PCIE_LNK_X12:
+				adev->pm.pcie_mlw_mask |= (CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X12 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X1);
 				break;
-			case 8:
-				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+			case PCIE_LNK_X8:
+				adev->pm.pcie_mlw_mask |= (CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X1);
 				break;
-			case 4:
-				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+			case PCIE_LNK_X4:
+				adev->pm.pcie_mlw_mask |= (CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X1);
 				break;
-			case 2:
-				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
-							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+			case PCIE_LNK_X2:
+				adev->pm.pcie_mlw_mask |= (CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X1);
 				break;
-			case 1:
-				adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1;
+			case PCIE_LNK_X1:
+				adev->pm.pcie_mlw_mask |= CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X1;
 				break;
 			default:
 				break;
 			}
+		}
+		/* platform caps */
+		if (platform_link_width == PCIE_LNK_WIDTH_UNKNOWN) {
+			adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_PCIE_MLW_MASK;
 		} else {
-			adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK;
+			switch (platform_link_width) {
+			case PCIE_LNK_X32:
+				adev->pm.pcie_mlw_mask |= (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+				break;
+			case PCIE_LNK_X16:
+				adev->pm.pcie_mlw_mask |= (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+				break;
+			case PCIE_LNK_X12:
+				adev->pm.pcie_mlw_mask |= (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+				break;
+			case PCIE_LNK_X8:
+				adev->pm.pcie_mlw_mask |= (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+				break;
+			case PCIE_LNK_X4:
+				adev->pm.pcie_mlw_mask |= (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+				break;
+			case PCIE_LNK_X2:
+				adev->pm.pcie_mlw_mask |= (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
+							   CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
+				break;
+			case PCIE_LNK_X1:
+				adev->pm.pcie_mlw_mask |= CAIL_PCIE_LINK_WIDTH_SUPPORT_X1;
+				break;
+			default:
+				break;
+			}
 		}
 	}
 }
 
-/*
- * Debugfs
+/**
+ * amdgpu_device_is_peer_accessible - Check peer access through PCIe BAR
+ *
+ * @adev: amdgpu_device pointer
+ * @peer_adev: amdgpu_device pointer for peer device trying to access @adev
+ *
+ * Return true if @peer_adev can access (DMA) @adev through the PCIe
+ * BAR, i.e. @adev is "large BAR" and the BAR matches the DMA mask of
+ * @peer_adev.
  */
-int amdgpu_debugfs_add_files(struct amdgpu_device *adev,
-			     const struct drm_info_list *files,
-			     unsigned nfiles)
+bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev,
+				      struct amdgpu_device *peer_adev)
 {
-	unsigned i;
-
-	for (i = 0; i < adev->debugfs_count; i++) {
-		if (adev->debugfs[i].files == files) {
-			/* Already registered */
-			return 0;
-		}
-	}
-
-	i = adev->debugfs_count + 1;
-	if (i > AMDGPU_DEBUGFS_MAX_COMPONENTS) {
-		DRM_ERROR("Reached maximum number of debugfs components.\n");
-		DRM_ERROR("Report so we increase "
-			  "AMDGPU_DEBUGFS_MAX_COMPONENTS.\n");
-		return -EINVAL;
-	}
-	adev->debugfs[adev->debugfs_count].files = files;
-	adev->debugfs[adev->debugfs_count].num_files = nfiles;
-	adev->debugfs_count = i;
-#if defined(CONFIG_DEBUG_FS)
-	drm_debugfs_create_files(files, nfiles,
-				 adev->ddev->primary->debugfs_root,
-				 adev->ddev->primary);
+#ifdef CONFIG_HSA_AMD_P2P
+	bool p2p_access =
+		!adev->gmc.xgmi.connected_to_cpu &&
+		!(pci_p2pdma_distance(adev->pdev, peer_adev->dev, false) < 0);
+	if (!p2p_access)
+		dev_info(adev->dev, "PCIe P2P access from peer device %s is not supported by the chipset\n",
+			pci_name(peer_adev->pdev));
+
+	bool is_large_bar = adev->gmc.visible_vram_size &&
+		adev->gmc.real_vram_size == adev->gmc.visible_vram_size;
+	bool p2p_addressable = amdgpu_device_check_iommu_remap(peer_adev);
+
+	if (!p2p_addressable) {
+		uint64_t address_mask = peer_adev->dev->dma_mask ?
+			~*peer_adev->dev->dma_mask : ~((1ULL << 32) - 1);
+		resource_size_t aper_limit =
+			adev->gmc.aper_base + adev->gmc.aper_size - 1;
+
+		p2p_addressable = !(adev->gmc.aper_base & address_mask ||
+				     aper_limit & address_mask);
+	}
+	return pcie_p2p && is_large_bar && p2p_access && p2p_addressable;
+#else
+	return false;
 #endif
-	return 0;
 }
 
-#if defined(CONFIG_DEBUG_FS)
-
-static ssize_t amdgpu_debugfs_regs_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+int amdgpu_device_baco_enter(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
-	int r;
-	bool pm_pg_lock, use_bank;
-	unsigned instance_bank, sh_bank, se_bank;
-
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
+	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
 
-	/* are we reading registers for which a PG lock is necessary? */
-	pm_pg_lock = (*pos >> 23) & 1;
-
-	if (*pos & (1ULL << 62)) {
-		se_bank = (*pos >> 24) & 0x3FF;
-		sh_bank = (*pos >> 34) & 0x3FF;
-		instance_bank = (*pos >> 44) & 0x3FF;
-
-		if (se_bank == 0x3FF)
-			se_bank = 0xFFFFFFFF;
-		if (sh_bank == 0x3FF)
-			sh_bank = 0xFFFFFFFF;
-		if (instance_bank == 0x3FF)
-			instance_bank = 0xFFFFFFFF;
-		use_bank = 1;
-	} else {
-		use_bank = 0;
-	}
+	if (!amdgpu_device_supports_baco(adev))
+		return -ENOTSUPP;
 
-	*pos &= (1UL << 22) - 1;
+	if (ras && adev->ras_enabled &&
+	    adev->nbio.funcs->enable_doorbell_interrupt)
+		adev->nbio.funcs->enable_doorbell_interrupt(adev, false);
 
-	if (use_bank) {
-		if ((sh_bank != 0xFFFFFFFF && sh_bank >= adev->gfx.config.max_sh_per_se) ||
-		    (se_bank != 0xFFFFFFFF && se_bank >= adev->gfx.config.max_shader_engines))
-			return -EINVAL;
-		mutex_lock(&adev->grbm_idx_mutex);
-		amdgpu_gfx_select_se_sh(adev, se_bank,
-					sh_bank, instance_bank);
-	}
+	return amdgpu_dpm_baco_enter(adev);
+}
 
-	if (pm_pg_lock)
-		mutex_lock(&adev->pm.mutex);
+int amdgpu_device_baco_exit(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+	int ret = 0;
 
-	while (size) {
-		uint32_t value;
+	if (!amdgpu_device_supports_baco(adev))
+		return -ENOTSUPP;
 
-		if (*pos > adev->rmmio_size)
-			goto end;
+	ret = amdgpu_dpm_baco_exit(adev);
+	if (ret)
+		return ret;
 
-		value = RREG32(*pos >> 2);
-		r = put_user(value, (uint32_t *)buf);
-		if (r) {
-			result = r;
-			goto end;
-		}
+	if (ras && adev->ras_enabled &&
+	    adev->nbio.funcs->enable_doorbell_interrupt)
+		adev->nbio.funcs->enable_doorbell_interrupt(adev, true);
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
-	}
+	if (amdgpu_passthrough(adev) && adev->nbio.funcs &&
+	    adev->nbio.funcs->clear_doorbell_interrupt)
+		adev->nbio.funcs->clear_doorbell_interrupt(adev);
 
-end:
-	if (use_bank) {
-		amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-		mutex_unlock(&adev->grbm_idx_mutex);
-	}
+	return 0;
+}
 
-	if (pm_pg_lock)
-		mutex_unlock(&adev->pm.mutex);
+/**
+ * amdgpu_pci_error_detected - Called when a PCI error is detected.
+ * @pdev: PCI device struct
+ * @state: PCI channel state
+ *
+ * Description: Called when a PCI error is detected.
+ *
+ * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT.
+ */
+pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
+{
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	struct amdgpu_hive_info *hive __free(xgmi_put_hive) =
+		amdgpu_get_xgmi_hive(adev);
+	struct amdgpu_reset_context reset_context;
+	struct list_head device_list;
+
+	dev_info(adev->dev, "PCI error: detected callback!!\n");
+
+	adev->pci_channel_state = state;
+
+	switch (state) {
+	case pci_channel_io_normal:
+		dev_info(adev->dev, "pci_channel_io_normal: state(%d)!!\n", state);
+		return PCI_ERS_RESULT_CAN_RECOVER;
+	case pci_channel_io_frozen:
+		/* Fatal error, prepare for slot reset */
+		dev_info(adev->dev, "pci_channel_io_frozen: state(%d)!!\n", state);
+		if (hive) {
+			/* Hive devices should be able to support FW based
+			 * link reset on other devices, if not return.
+			 */
+			if (!amdgpu_dpm_is_link_reset_supported(adev)) {
+				dev_warn(adev->dev,
+					 "No support for XGMI hive yet...\n");
+				return PCI_ERS_RESULT_DISCONNECT;
+			}
+			/* Set dpc status only if device is part of hive
+			 * Non-hive devices should be able to recover after
+			 * link reset.
+			 */
+			amdgpu_reset_set_dpc_status(adev, true);
 
-	return result;
+			mutex_lock(&hive->hive_lock);
+		}
+		memset(&reset_context, 0, sizeof(reset_context));
+		INIT_LIST_HEAD(&device_list);
+
+		amdgpu_device_recovery_prepare(adev, &device_list, hive);
+		amdgpu_device_recovery_get_reset_lock(adev, &device_list);
+		amdgpu_device_halt_activities(adev, NULL, &reset_context, &device_list,
+					      hive, false);
+		if (hive)
+			mutex_unlock(&hive->hive_lock);
+		return PCI_ERS_RESULT_NEED_RESET;
+	case pci_channel_io_perm_failure:
+		/* Permanent error, prepare for device removal */
+		dev_info(adev->dev, "pci_channel_io_perm_failure: state(%d)!!\n", state);
+		return PCI_ERS_RESULT_DISCONNECT;
+	}
+
+	return PCI_ERS_RESULT_NEED_RESET;
 }
 
-static ssize_t amdgpu_debugfs_regs_write(struct file *f, const char __user *buf,
-					 size_t size, loff_t *pos)
+/**
+ * amdgpu_pci_mmio_enabled - Enable MMIO and dump debug registers
+ * @pdev: pointer to PCI device
+ */
+pci_ers_result_t amdgpu_pci_mmio_enabled(struct pci_dev *pdev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
-	int r;
-	bool pm_pg_lock, use_bank;
-	unsigned instance_bank, sh_bank, se_bank;
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(dev);
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
+	dev_info(adev->dev, "PCI error: mmio enabled callback!!\n");
+
+	/* TODO - dump whatever for debugging purposes */
 
-	/* are we reading registers for which a PG lock is necessary? */
-	pm_pg_lock = (*pos >> 23) & 1;
-
-	if (*pos & (1ULL << 62)) {
-		se_bank = (*pos >> 24) & 0x3FF;
-		sh_bank = (*pos >> 34) & 0x3FF;
-		instance_bank = (*pos >> 44) & 0x3FF;
-
-		if (se_bank == 0x3FF)
-			se_bank = 0xFFFFFFFF;
-		if (sh_bank == 0x3FF)
-			sh_bank = 0xFFFFFFFF;
-		if (instance_bank == 0x3FF)
-			instance_bank = 0xFFFFFFFF;
-		use_bank = 1;
+	/* This called only if amdgpu_pci_error_detected returns
+	 * PCI_ERS_RESULT_CAN_RECOVER. Read/write to the device still
+	 * works, no need to reset slot.
+	 */
+
+	return PCI_ERS_RESULT_RECOVERED;
+}
+
+/**
+ * amdgpu_pci_slot_reset - Called when PCI slot has been reset.
+ * @pdev: PCI device struct
+ *
+ * Description: This routine is called by the pci error recovery
+ * code after the PCI slot has been reset, just before we
+ * should resume normal operations.
+ */
+pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev)
+{
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	struct amdgpu_reset_context reset_context;
+	struct amdgpu_device *tmp_adev;
+	struct amdgpu_hive_info *hive;
+	struct list_head device_list;
+	struct pci_dev *link_dev;
+	int r = 0, i, timeout;
+	u32 memsize;
+	u16 status;
+
+	dev_info(adev->dev, "PCI error: slot reset callback!!\n");
+
+	memset(&reset_context, 0, sizeof(reset_context));
+
+	if (adev->pcie_reset_ctx.swus)
+		link_dev = adev->pcie_reset_ctx.swus;
+	else
+		link_dev = adev->pdev;
+	/* wait for asic to come out of reset, timeout = 10s */
+	timeout = 10000;
+	do {
+		usleep_range(10000, 10500);
+		r = pci_read_config_word(link_dev, PCI_VENDOR_ID, &status);
+		timeout -= 10;
+	} while (timeout > 0 && (status != PCI_VENDOR_ID_ATI) &&
+		 (status != PCI_VENDOR_ID_AMD));
+
+	if ((status != PCI_VENDOR_ID_ATI) && (status != PCI_VENDOR_ID_AMD)) {
+		r = -ETIME;
+		goto out;
+	}
+
+	amdgpu_device_load_switch_state(adev);
+	/* Restore PCI confspace */
+	amdgpu_device_load_pci_state(pdev);
+
+	/* confirm  ASIC came out of reset */
+	for (i = 0; i < adev->usec_timeout; i++) {
+		memsize = amdgpu_asic_get_config_memsize(adev);
+
+		if (memsize != 0xffffffff)
+			break;
+		udelay(1);
+	}
+	if (memsize == 0xffffffff) {
+		r = -ETIME;
+		goto out;
+	}
+
+	reset_context.method = AMD_RESET_METHOD_NONE;
+	reset_context.reset_req_dev = adev;
+	set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
+	set_bit(AMDGPU_SKIP_COREDUMP, &reset_context.flags);
+	INIT_LIST_HEAD(&device_list);
+
+	hive = amdgpu_get_xgmi_hive(adev);
+	if (hive) {
+		mutex_lock(&hive->hive_lock);
+		reset_context.hive = hive;
+		list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
+			tmp_adev->pcie_reset_ctx.in_link_reset = true;
+			list_add_tail(&tmp_adev->reset_list, &device_list);
+		}
 	} else {
-		use_bank = 0;
+		set_bit(AMDGPU_SKIP_HW_RESET, &reset_context.flags);
+		list_add_tail(&adev->reset_list, &device_list);
 	}
 
-	*pos &= (1UL << 22) - 1;
-
-	if (use_bank) {
-		if ((sh_bank != 0xFFFFFFFF && sh_bank >= adev->gfx.config.max_sh_per_se) ||
-		    (se_bank != 0xFFFFFFFF && se_bank >= adev->gfx.config.max_shader_engines))
-			return -EINVAL;
-		mutex_lock(&adev->grbm_idx_mutex);
-		amdgpu_gfx_select_se_sh(adev, se_bank,
-					sh_bank, instance_bank);
+	r = amdgpu_device_asic_reset(adev, &device_list, &reset_context);
+out:
+	if (!r) {
+		if (amdgpu_device_cache_pci_state(adev->pdev))
+			pci_restore_state(adev->pdev);
+		dev_info(adev->dev, "PCIe error recovery succeeded\n");
+	} else {
+		dev_err(adev->dev, "PCIe error recovery failed, err:%d\n", r);
+		if (hive) {
+			list_for_each_entry(tmp_adev, &device_list, reset_list)
+				amdgpu_device_unset_mp1_state(tmp_adev);
+		}
+		amdgpu_device_recovery_put_reset_lock(adev, &device_list);
 	}
 
-	if (pm_pg_lock)
-		mutex_lock(&adev->pm.mutex);
+	if (hive) {
+		mutex_unlock(&hive->hive_lock);
+		amdgpu_put_xgmi_hive(hive);
+	}
 
-	while (size) {
-		uint32_t value;
+	return r ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
+}
 
-		if (*pos > adev->rmmio_size)
-			return result;
+/**
+ * amdgpu_pci_resume() - resume normal ops after PCI reset
+ * @pdev: pointer to PCI device
+ *
+ * Called when the error recovery driver tells us that its
+ * OK to resume normal operation.
+ */
+void amdgpu_pci_resume(struct pci_dev *pdev)
+{
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	struct list_head device_list;
+	struct amdgpu_hive_info *hive = NULL;
+	struct amdgpu_device *tmp_adev = NULL;
 
-		r = get_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
+	dev_info(adev->dev, "PCI error: resume callback!!\n");
 
-		WREG32(*pos >> 2, value);
+	/* Only continue execution for the case of pci_channel_io_frozen */
+	if (adev->pci_channel_state != pci_channel_io_frozen)
+		return;
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
-	}
+	INIT_LIST_HEAD(&device_list);
 
-	if (use_bank) {
-		amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
-		mutex_unlock(&adev->grbm_idx_mutex);
-	}
+	hive = amdgpu_get_xgmi_hive(adev);
+	if (hive) {
+		mutex_lock(&hive->hive_lock);
+		list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
+			tmp_adev->pcie_reset_ctx.in_link_reset = false;
+			list_add_tail(&tmp_adev->reset_list, &device_list);
+		}
+	} else
+		list_add_tail(&adev->reset_list, &device_list);
 
-	if (pm_pg_lock)
-		mutex_unlock(&adev->pm.mutex);
+	amdgpu_device_sched_resume(&device_list, NULL, NULL);
+	amdgpu_device_gpu_resume(adev, &device_list, false);
+	amdgpu_device_recovery_put_reset_lock(adev, &device_list);
 
-	return result;
+	if (hive) {
+		mutex_unlock(&hive->hive_lock);
+		amdgpu_put_xgmi_hive(hive);
+	}
 }
 
-static ssize_t amdgpu_debugfs_regs_pcie_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+static void amdgpu_device_cache_switch_state(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
+	struct pci_dev *swus, *swds;
 	int r;
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
-
-	while (size) {
-		uint32_t value;
+	swds = pci_upstream_bridge(adev->pdev);
+	if (!swds || swds->vendor != PCI_VENDOR_ID_ATI ||
+	    pci_pcie_type(swds) != PCI_EXP_TYPE_DOWNSTREAM)
+		return;
+	swus = pci_upstream_bridge(swds);
+	if (!swus ||
+	    (swus->vendor != PCI_VENDOR_ID_ATI &&
+	     swus->vendor != PCI_VENDOR_ID_AMD) ||
+	    pci_pcie_type(swus) != PCI_EXP_TYPE_UPSTREAM)
+		return;
 
-		value = RREG32_PCIE(*pos >> 2);
-		r = put_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
+	/* If already saved, return */
+	if (adev->pcie_reset_ctx.swus)
+		return;
+	/* Upstream bridge is ATI, assume it's SWUS/DS architecture */
+	r = pci_save_state(swds);
+	if (r)
+		return;
+	adev->pcie_reset_ctx.swds_pcistate = pci_store_saved_state(swds);
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
-	}
+	r = pci_save_state(swus);
+	if (r)
+		return;
+	adev->pcie_reset_ctx.swus_pcistate = pci_store_saved_state(swus);
 
-	return result;
+	adev->pcie_reset_ctx.swus = swus;
 }
 
-static ssize_t amdgpu_debugfs_regs_pcie_write(struct file *f, const char __user *buf,
-					 size_t size, loff_t *pos)
+static void amdgpu_device_load_switch_state(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
+	struct pci_dev *pdev;
 	int r;
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
-
-	while (size) {
-		uint32_t value;
-
-		r = get_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
-
-		WREG32_PCIE(*pos >> 2, value);
+	if (!adev->pcie_reset_ctx.swds_pcistate ||
+	    !adev->pcie_reset_ctx.swus_pcistate)
+		return;
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
+	pdev = adev->pcie_reset_ctx.swus;
+	r = pci_load_saved_state(pdev, adev->pcie_reset_ctx.swus_pcistate);
+	if (!r) {
+		pci_restore_state(pdev);
+	} else {
+		dev_warn(adev->dev, "Failed to load SWUS state, err:%d\n", r);
+		return;
 	}
 
-	return result;
+	pdev = pci_upstream_bridge(adev->pdev);
+	r = pci_load_saved_state(pdev, adev->pcie_reset_ctx.swds_pcistate);
+	if (!r)
+		pci_restore_state(pdev);
+	else
+		dev_warn(adev->dev, "Failed to load SWDS state, err:%d\n", r);
 }
 
-static ssize_t amdgpu_debugfs_regs_didt_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+bool amdgpu_device_cache_pci_state(struct pci_dev *pdev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	int r;
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
+	if (amdgpu_sriov_vf(adev))
+		return false;
 
-	while (size) {
-		uint32_t value;
+	r = pci_save_state(pdev);
+	if (!r) {
+		kfree(adev->pci_state);
 
-		value = RREG32_DIDT(*pos >> 2);
-		r = put_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
+		adev->pci_state = pci_store_saved_state(pdev);
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
+		if (!adev->pci_state) {
+			dev_err(adev->dev, "Failed to store PCI saved state");
+			return false;
+		}
+	} else {
+		dev_warn(adev->dev, "Failed to save PCI state, err:%d\n", r);
+		return false;
 	}
 
-	return result;
+	amdgpu_device_cache_switch_state(adev);
+
+	return true;
 }
 
-static ssize_t amdgpu_debugfs_regs_didt_write(struct file *f, const char __user *buf,
-					 size_t size, loff_t *pos)
+bool amdgpu_device_load_pci_state(struct pci_dev *pdev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	int r;
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
-
-	while (size) {
-		uint32_t value;
-
-		r = get_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
+	if (!adev->pci_state)
+		return false;
 
-		WREG32_DIDT(*pos >> 2, value);
+	r = pci_load_saved_state(pdev, adev->pci_state);
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
+	if (!r) {
+		pci_restore_state(pdev);
+	} else {
+		dev_warn(adev->dev, "Failed to load PCI state, err:%d\n", r);
+		return false;
 	}
 
-	return result;
+	return true;
 }
 
-static ssize_t amdgpu_debugfs_regs_smc_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+void amdgpu_device_flush_hdp(struct amdgpu_device *adev,
+		struct amdgpu_ring *ring)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
-	int r;
+#ifdef CONFIG_X86_64
+	if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev))
+		return;
+#endif
+	if (adev->gmc.xgmi.connected_to_cpu)
+		return;
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
+	if (ring && ring->funcs->emit_hdp_flush) {
+		amdgpu_ring_emit_hdp_flush(ring);
+		return;
+	}
 
-	while (size) {
-		uint32_t value;
+	if (!ring && amdgpu_sriov_runtime(adev)) {
+		if (!amdgpu_kiq_hdp_flush(adev))
+			return;
+	}
 
-		value = RREG32_SMC(*pos);
-		r = put_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
+	amdgpu_hdp_flush(adev, ring);
+}
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
-	}
+void amdgpu_device_invalidate_hdp(struct amdgpu_device *adev,
+		struct amdgpu_ring *ring)
+{
+#ifdef CONFIG_X86_64
+	if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev))
+		return;
+#endif
+	if (adev->gmc.xgmi.connected_to_cpu)
+		return;
 
-	return result;
+	amdgpu_hdp_invalidate(adev, ring);
 }
 
-static ssize_t amdgpu_debugfs_regs_smc_write(struct file *f, const char __user *buf,
-					 size_t size, loff_t *pos)
+int amdgpu_in_reset(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
-	int r;
+	return atomic_read(&adev->reset_domain->in_gpu_reset);
+}
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
+/**
+ * amdgpu_device_halt() - bring hardware to some kind of halt state
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Bring hardware to some kind of halt state so that no one can touch it
+ * any more. It will help to maintain error context when error occurred.
+ * Compare to a simple hang, the system will keep stable at least for SSH
+ * access. Then it should be trivial to inspect the hardware state and
+ * see what's going on. Implemented as following:
+ *
+ * 1. drm_dev_unplug() makes device inaccessible to user space(IOCTLs, etc),
+ *    clears all CPU mappings to device, disallows remappings through page faults
+ * 2. amdgpu_irq_disable_all() disables all interrupts
+ * 3. amdgpu_fence_driver_hw_fini() signals all HW fences
+ * 4. set adev->no_hw_access to avoid potential crashes after setp 5
+ * 5. amdgpu_device_unmap_mmio() clears all MMIO mappings
+ * 6. pci_disable_device() and pci_wait_for_pending_transaction()
+ *    flush any in flight DMA operations
+ */
+void amdgpu_device_halt(struct amdgpu_device *adev)
+{
+	struct pci_dev *pdev = adev->pdev;
+	struct drm_device *ddev = adev_to_drm(adev);
 
-	while (size) {
-		uint32_t value;
+	amdgpu_xcp_dev_unplug(adev);
+	drm_dev_unplug(ddev);
 
-		r = get_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
+	amdgpu_irq_disable_all(adev);
 
-		WREG32_SMC(*pos, value);
+	amdgpu_fence_driver_hw_fini(adev);
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
-	}
+	adev->no_hw_access = true;
+
+	amdgpu_device_unmap_mmio(adev);
 
-	return result;
+	pci_disable_device(pdev);
+	pci_wait_for_pending_transaction(pdev);
 }
 
-static ssize_t amdgpu_debugfs_gca_config_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+u32 amdgpu_device_pcie_port_rreg(struct amdgpu_device *adev,
+				u32 reg)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	ssize_t result = 0;
-	int r;
-	uint32_t *config, no_regs = 0;
+	unsigned long flags, address, data;
+	u32 r;
 
-	if (size & 0x3 || *pos & 0x3)
-		return -EINVAL;
+	address = adev->nbio.funcs->get_pcie_port_index_offset(adev);
+	data = adev->nbio.funcs->get_pcie_port_data_offset(adev);
 
-	config = kmalloc_array(256, sizeof(*config), GFP_KERNEL);
-	if (!config)
-		return -ENOMEM;
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	WREG32(address, reg * 4);
+	(void)RREG32(address);
+	r = RREG32(data);
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
+	return r;
+}
 
-	/* version, increment each time something is added */
-	config[no_regs++] = 3;
-	config[no_regs++] = adev->gfx.config.max_shader_engines;
-	config[no_regs++] = adev->gfx.config.max_tile_pipes;
-	config[no_regs++] = adev->gfx.config.max_cu_per_sh;
-	config[no_regs++] = adev->gfx.config.max_sh_per_se;
-	config[no_regs++] = adev->gfx.config.max_backends_per_se;
-	config[no_regs++] = adev->gfx.config.max_texture_channel_caches;
-	config[no_regs++] = adev->gfx.config.max_gprs;
-	config[no_regs++] = adev->gfx.config.max_gs_threads;
-	config[no_regs++] = adev->gfx.config.max_hw_contexts;
-	config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_frontend;
-	config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_backend;
-	config[no_regs++] = adev->gfx.config.sc_hiz_tile_fifo_size;
-	config[no_regs++] = adev->gfx.config.sc_earlyz_tile_fifo_size;
-	config[no_regs++] = adev->gfx.config.num_tile_pipes;
-	config[no_regs++] = adev->gfx.config.backend_enable_mask;
-	config[no_regs++] = adev->gfx.config.mem_max_burst_length_bytes;
-	config[no_regs++] = adev->gfx.config.mem_row_size_in_kb;
-	config[no_regs++] = adev->gfx.config.shader_engine_tile_size;
-	config[no_regs++] = adev->gfx.config.num_gpus;
-	config[no_regs++] = adev->gfx.config.multi_gpu_tile_size;
-	config[no_regs++] = adev->gfx.config.mc_arb_ramcfg;
-	config[no_regs++] = adev->gfx.config.gb_addr_config;
-	config[no_regs++] = adev->gfx.config.num_rbs;
-
-	/* rev==1 */
-	config[no_regs++] = adev->rev_id;
-	config[no_regs++] = adev->pg_flags;
-	config[no_regs++] = adev->cg_flags;
-
-	/* rev==2 */
-	config[no_regs++] = adev->family;
-	config[no_regs++] = adev->external_rev_id;
-
-	/* rev==3 */
-	config[no_regs++] = adev->pdev->device;
-	config[no_regs++] = adev->pdev->revision;
-	config[no_regs++] = adev->pdev->subsystem_device;
-	config[no_regs++] = adev->pdev->subsystem_vendor;
-
-	while (size && (*pos < no_regs * 4)) {
-		uint32_t value;
-
-		value = config[*pos >> 2];
-		r = put_user(value, (uint32_t *)buf);
-		if (r) {
-			kfree(config);
-			return r;
-		}
+void amdgpu_device_pcie_port_wreg(struct amdgpu_device *adev,
+				u32 reg, u32 v)
+{
+	unsigned long flags, address, data;
 
-		result += 4;
-		buf += 4;
-		*pos += 4;
-		size -= 4;
-	}
+	address = adev->nbio.funcs->get_pcie_port_index_offset(adev);
+	data = adev->nbio.funcs->get_pcie_port_data_offset(adev);
 
-	kfree(config);
-	return result;
+	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
+	WREG32(address, reg * 4);
+	(void)RREG32(address);
+	WREG32(data, v);
+	(void)RREG32(data);
+	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
 }
 
-static ssize_t amdgpu_debugfs_sensor_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+/**
+ * amdgpu_device_get_gang - return a reference to the current gang
+ * @adev: amdgpu_device pointer
+ *
+ * Returns: A new reference to the current gang leader.
+ */
+struct dma_fence *amdgpu_device_get_gang(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = file_inode(f)->i_private;
-	int idx, x, outsize, r, valuesize;
-	uint32_t values[16];
+	struct dma_fence *fence;
 
-	if (size & 3 || *pos & 0x3)
-		return -EINVAL;
-
-	if (amdgpu_dpm == 0)
-		return -EINVAL;
-
-	/* convert offset to sensor number */
-	idx = *pos >> 2;
+	rcu_read_lock();
+	fence = dma_fence_get_rcu_safe(&adev->gang_submit);
+	rcu_read_unlock();
+	return fence;
+}
 
-	valuesize = sizeof(values);
-	if (adev->powerplay.pp_funcs && adev->powerplay.pp_funcs->read_sensor)
-		r = adev->powerplay.pp_funcs->read_sensor(adev->powerplay.pp_handle, idx, &values[0], &valuesize);
-	else if (adev->pm.funcs && adev->pm.funcs->read_sensor)
-		r = adev->pm.funcs->read_sensor(adev, idx, &values[0],
-						&valuesize);
-	else
-		return -EINVAL;
+/**
+ * amdgpu_device_switch_gang - switch to a new gang
+ * @adev: amdgpu_device pointer
+ * @gang: the gang to switch to
+ *
+ * Try to switch to a new gang.
+ * Returns: NULL if we switched to the new gang or a reference to the current
+ * gang leader.
+ */
+struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev,
+					    struct dma_fence *gang)
+{
+	struct dma_fence *old = NULL;
 
-	if (size > valuesize)
-		return -EINVAL;
+	dma_fence_get(gang);
+	do {
+		dma_fence_put(old);
+		old = amdgpu_device_get_gang(adev);
+		if (old == gang)
+			break;
 
-	outsize = 0;
-	x = 0;
-	if (!r) {
-		while (size) {
-			r = put_user(values[x++], (int32_t *)buf);
-			buf += 4;
-			size -= 4;
-			outsize += 4;
+		if (!dma_fence_is_signaled(old)) {
+			dma_fence_put(gang);
+			return old;
 		}
-	}
 
-	return !r ? outsize : r;
+	} while (cmpxchg((struct dma_fence __force **)&adev->gang_submit,
+			 old, gang) != old);
+
+	/*
+	 * Drop it once for the exchanged reference in adev and once for the
+	 * thread local reference acquired in amdgpu_device_get_gang().
+	 */
+	dma_fence_put(old);
+	dma_fence_put(old);
+	return NULL;
 }
 
-static ssize_t amdgpu_debugfs_wave_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+/**
+ * amdgpu_device_enforce_isolation - enforce HW isolation
+ * @adev: the amdgpu device pointer
+ * @ring: the HW ring the job is supposed to run on
+ * @job: the job which is about to be pushed to the HW ring
+ *
+ * Makes sure that only one client at a time can use the GFX block.
+ * Returns: The dependency to wait on before the job can be pushed to the HW.
+ * The function is called multiple times until NULL is returned.
+ */
+struct dma_fence *amdgpu_device_enforce_isolation(struct amdgpu_device *adev,
+						  struct amdgpu_ring *ring,
+						  struct amdgpu_job *job)
 {
-	struct amdgpu_device *adev = f->f_inode->i_private;
-	int r, x;
-	ssize_t result=0;
-	uint32_t offset, se, sh, cu, wave, simd, data[32];
+	struct amdgpu_isolation *isolation = &adev->isolation[ring->xcp_id];
+	struct drm_sched_fence *f = job->base.s_fence;
+	struct dma_fence *dep;
+	void *owner;
+	int r;
 
-	if (size & 3 || *pos & 3)
-		return -EINVAL;
+	/*
+	 * For now enforce isolation only for the GFX block since we only need
+	 * the cleaner shader on those rings.
+	 */
+	if (ring->funcs->type != AMDGPU_RING_TYPE_GFX &&
+	    ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE)
+		return NULL;
 
-	/* decode offset */
-	offset = (*pos & 0x7F);
-	se = ((*pos >> 7) & 0xFF);
-	sh = ((*pos >> 15) & 0xFF);
-	cu = ((*pos >> 23) & 0xFF);
-	wave = ((*pos >> 31) & 0xFF);
-	simd = ((*pos >> 37) & 0xFF);
+	/*
+	 * All submissions where enforce isolation is false are handled as if
+	 * they come from a single client. Use ~0l as the owner to distinct it
+	 * from kernel submissions where the owner is NULL.
+	 */
+	owner = job->enforce_isolation ? f->owner : (void *)~0l;
 
-	/* switch to the specific se/sh/cu */
-	mutex_lock(&adev->grbm_idx_mutex);
-	amdgpu_gfx_select_se_sh(adev, se, sh, cu);
+	mutex_lock(&adev->enforce_isolation_mutex);
 
-	x = 0;
-	if (adev->gfx.funcs->read_wave_data)
-		adev->gfx.funcs->read_wave_data(adev, simd, wave, data, &x);
+	/*
+	 * The "spearhead" submission is the first one which changes the
+	 * ownership to its client. We always need to wait for it to be
+	 * pushed to the HW before proceeding with anything.
+	 */
+	if (&f->scheduled != isolation->spearhead &&
+	    !dma_fence_is_signaled(isolation->spearhead)) {
+		dep = isolation->spearhead;
+		goto out_grab_ref;
+	}
 
-	amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
-	mutex_unlock(&adev->grbm_idx_mutex);
+	if (isolation->owner != owner) {
 
-	if (!x)
-		return -EINVAL;
+		/*
+		 * Wait for any gang to be assembled before switching to a
+		 * different owner or otherwise we could deadlock the
+		 * submissions.
+		 */
+		if (!job->gang_submit) {
+			dep = amdgpu_device_get_gang(adev);
+			if (!dma_fence_is_signaled(dep))
+				goto out_return_dep;
+			dma_fence_put(dep);
+		}
 
-	while (size && (offset < x * 4)) {
-		uint32_t value;
+		dma_fence_put(isolation->spearhead);
+		isolation->spearhead = dma_fence_get(&f->scheduled);
+		amdgpu_sync_move(&isolation->active, &isolation->prev);
+		trace_amdgpu_isolation(isolation->owner, owner);
+		isolation->owner = owner;
+	}
 
-		value = data[offset >> 2];
-		r = put_user(value, (uint32_t *)buf);
-		if (r)
-			return r;
+	/*
+	 * Specifying the ring here helps to pipeline submissions even when
+	 * isolation is enabled. If that is not desired for testing NULL can be
+	 * used instead of the ring to enforce a CPU round trip while switching
+	 * between clients.
+	 */
+	dep = amdgpu_sync_peek_fence(&isolation->prev, ring);
+	r = amdgpu_sync_fence(&isolation->active, &f->finished, GFP_NOWAIT);
+	if (r)
+		dev_warn(adev->dev, "OOM tracking isolation\n");
+
+out_grab_ref:
+	dma_fence_get(dep);
+out_return_dep:
+	mutex_unlock(&adev->enforce_isolation_mutex);
+	return dep;
+}
 
-		result += 4;
-		buf += 4;
-		offset += 4;
-		size -= 4;
+bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev)
+{
+	switch (adev->asic_type) {
+#ifdef CONFIG_DRM_AMDGPU_SI
+	case CHIP_HAINAN:
+#endif
+	case CHIP_TOPAZ:
+		/* chips with no display hardware */
+		return false;
+#ifdef CONFIG_DRM_AMDGPU_SI
+	case CHIP_TAHITI:
+	case CHIP_PITCAIRN:
+	case CHIP_VERDE:
+	case CHIP_OLAND:
+#endif
+#ifdef CONFIG_DRM_AMDGPU_CIK
+	case CHIP_BONAIRE:
+	case CHIP_HAWAII:
+	case CHIP_KAVERI:
+	case CHIP_KABINI:
+	case CHIP_MULLINS:
+#endif
+	case CHIP_TONGA:
+	case CHIP_FIJI:
+	case CHIP_POLARIS10:
+	case CHIP_POLARIS11:
+	case CHIP_POLARIS12:
+	case CHIP_VEGAM:
+	case CHIP_CARRIZO:
+	case CHIP_STONEY:
+		/* chips with display hardware */
+		return true;
+	default:
+		/* IP discovery */
+		if (!amdgpu_ip_version(adev, DCE_HWIP, 0) ||
+		    (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK))
+			return false;
+		return true;
 	}
+}
 
-	return result;
+uint32_t amdgpu_device_wait_on_rreg(struct amdgpu_device *adev,
+		uint32_t inst, uint32_t reg_addr, char reg_name[],
+		uint32_t expected_value, uint32_t mask)
+{
+	uint32_t ret = 0;
+	uint32_t old_ = 0;
+	uint32_t tmp_ = RREG32(reg_addr);
+	uint32_t loop = adev->usec_timeout;
+
+	while ((tmp_ & (mask)) != (expected_value)) {
+		if (old_ != tmp_) {
+			loop = adev->usec_timeout;
+			old_ = tmp_;
+		} else
+			udelay(1);
+		tmp_ = RREG32(reg_addr);
+		loop--;
+		if (!loop) {
+			dev_warn(
+				adev->dev,
+				"Register(%d) [%s] failed to reach value 0x%08x != 0x%08xn",
+				inst, reg_name, (uint32_t)expected_value,
+				(uint32_t)(tmp_ & (mask)));
+			ret = -ETIMEDOUT;
+			break;
+		}
+	}
+	return ret;
 }
 
-static ssize_t amdgpu_debugfs_gpr_read(struct file *f, char __user *buf,
-					size_t size, loff_t *pos)
+ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring)
 {
-	struct amdgpu_device *adev = f->f_inode->i_private;
-	int r;
-	ssize_t result = 0;
-	uint32_t offset, se, sh, cu, wave, simd, thread, bank, *data;
+	ssize_t size = 0;
 
-	if (size & 3 || *pos & 3)
-		return -EINVAL;
+	if (!ring || !ring->adev)
+		return size;
 
-	/* decode offset */
-	offset = (*pos & 0xFFF);       /* in dwords */
-	se = ((*pos >> 12) & 0xFF);
-	sh = ((*pos >> 20) & 0xFF);
-	cu = ((*pos >> 28) & 0xFF);
-	wave = ((*pos >> 36) & 0xFF);
-	simd = ((*pos >> 44) & 0xFF);
-	thread = ((*pos >> 52) & 0xFF);
-	bank = ((*pos >> 60) & 1);
-
-	data = kmalloc_array(1024, sizeof(*data), GFP_KERNEL);
-	if (!data)
-		return -ENOMEM;
+	if (amdgpu_device_should_recover_gpu(ring->adev))
+		size |= AMDGPU_RESET_TYPE_FULL;
 
-	/* switch to the specific se/sh/cu */
-	mutex_lock(&adev->grbm_idx_mutex);
-	amdgpu_gfx_select_se_sh(adev, se, sh, cu);
+	if (unlikely(!ring->adev->debug_disable_soft_recovery) &&
+	    !amdgpu_sriov_vf(ring->adev) && ring->funcs->soft_recovery)
+		size |= AMDGPU_RESET_TYPE_SOFT_RESET;
 
-	if (bank == 0) {
-		if (adev->gfx.funcs->read_wave_vgprs)
-			adev->gfx.funcs->read_wave_vgprs(adev, simd, wave, thread, offset, size>>2, data);
-	} else {
-		if (adev->gfx.funcs->read_wave_sgprs)
-			adev->gfx.funcs->read_wave_sgprs(adev, simd, wave, offset, size>>2, data);
-	}
-
-	amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
-	mutex_unlock(&adev->grbm_idx_mutex);
+	return size;
+}
 
-	while (size) {
-		uint32_t value;
+ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset)
+{
+	ssize_t size = 0;
 
-		value = data[offset++];
-		r = put_user(value, (uint32_t *)buf);
-		if (r) {
-			result = r;
-			goto err;
-		}
+	if (supported_reset == 0) {
+		size += sysfs_emit_at(buf, size, "unsupported");
+		size += sysfs_emit_at(buf, size, "\n");
+		return size;
 
-		result += 4;
-		buf += 4;
-		size -= 4;
 	}
 
-err:
-	kfree(data);
-	return result;
-}
-
-static const struct file_operations amdgpu_debugfs_regs_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_regs_read,
-	.write = amdgpu_debugfs_regs_write,
-	.llseek = default_llseek
-};
-static const struct file_operations amdgpu_debugfs_regs_didt_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_regs_didt_read,
-	.write = amdgpu_debugfs_regs_didt_write,
-	.llseek = default_llseek
-};
-static const struct file_operations amdgpu_debugfs_regs_pcie_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_regs_pcie_read,
-	.write = amdgpu_debugfs_regs_pcie_write,
-	.llseek = default_llseek
-};
-static const struct file_operations amdgpu_debugfs_regs_smc_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_regs_smc_read,
-	.write = amdgpu_debugfs_regs_smc_write,
-	.llseek = default_llseek
-};
-
-static const struct file_operations amdgpu_debugfs_gca_config_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_gca_config_read,
-	.llseek = default_llseek
-};
+	if (supported_reset & AMDGPU_RESET_TYPE_SOFT_RESET)
+		size += sysfs_emit_at(buf, size, "soft ");
 
-static const struct file_operations amdgpu_debugfs_sensors_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_sensor_read,
-	.llseek = default_llseek
-};
+	if (supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE)
+		size += sysfs_emit_at(buf, size, "queue ");
 
-static const struct file_operations amdgpu_debugfs_wave_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_wave_read,
-	.llseek = default_llseek
-};
-static const struct file_operations amdgpu_debugfs_gpr_fops = {
-	.owner = THIS_MODULE,
-	.read = amdgpu_debugfs_gpr_read,
-	.llseek = default_llseek
-};
+	if (supported_reset & AMDGPU_RESET_TYPE_PER_PIPE)
+		size += sysfs_emit_at(buf, size, "pipe ");
 
-static const struct file_operations *debugfs_regs[] = {
-	&amdgpu_debugfs_regs_fops,
-	&amdgpu_debugfs_regs_didt_fops,
-	&amdgpu_debugfs_regs_pcie_fops,
-	&amdgpu_debugfs_regs_smc_fops,
-	&amdgpu_debugfs_gca_config_fops,
-	&amdgpu_debugfs_sensors_fops,
-	&amdgpu_debugfs_wave_fops,
-	&amdgpu_debugfs_gpr_fops,
-};
+	if (supported_reset & AMDGPU_RESET_TYPE_FULL)
+		size += sysfs_emit_at(buf, size, "full ");
 
-static const char *debugfs_regs_names[] = {
-	"amdgpu_regs",
-	"amdgpu_regs_didt",
-	"amdgpu_regs_pcie",
-	"amdgpu_regs_smc",
-	"amdgpu_gca_config",
-	"amdgpu_sensors",
-	"amdgpu_wave",
-	"amdgpu_gpr",
-};
+	size += sysfs_emit_at(buf, size, "\n");
+	return size;
+}
 
-static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
+void amdgpu_device_set_uid(struct amdgpu_uid *uid_info,
+			   enum amdgpu_uid_type type, uint8_t inst,
+			   uint64_t uid)
 {
-	struct drm_minor *minor = adev->ddev->primary;
-	struct dentry *ent, *root = minor->debugfs_root;
-	unsigned i, j;
+	if (!uid_info)
+		return;
 
-	for (i = 0; i < ARRAY_SIZE(debugfs_regs); i++) {
-		ent = debugfs_create_file(debugfs_regs_names[i],
-					  S_IFREG | S_IRUGO, root,
-					  adev, debugfs_regs[i]);
-		if (IS_ERR(ent)) {
-			for (j = 0; j < i; j++) {
-				debugfs_remove(adev->debugfs_regs[i]);
-				adev->debugfs_regs[i] = NULL;
-			}
-			return PTR_ERR(ent);
-		}
+	if (type >= AMDGPU_UID_TYPE_MAX) {
+		dev_err_once(uid_info->adev->dev, "Invalid UID type %d\n",
+			     type);
+		return;
+	}
 
-		if (!i)
-			i_size_write(ent->d_inode, adev->rmmio_size);
-		adev->debugfs_regs[i] = ent;
+	if (inst >= AMDGPU_UID_INST_MAX) {
+		dev_err_once(uid_info->adev->dev, "Invalid UID instance %d\n",
+			     inst);
+		return;
 	}
 
-	return 0;
+	if (uid_info->uid[type][inst] != 0) {
+		dev_warn_once(
+			uid_info->adev->dev,
+			"Overwriting existing UID %llu for type %d instance %d\n",
+			uid_info->uid[type][inst], type, inst);
+	}
+
+	uid_info->uid[type][inst] = uid;
 }
 
-static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev)
+u64 amdgpu_device_get_uid(struct amdgpu_uid *uid_info,
+			  enum amdgpu_uid_type type, uint8_t inst)
 {
-	unsigned i;
+	if (!uid_info)
+		return 0;
 
-	for (i = 0; i < ARRAY_SIZE(debugfs_regs); i++) {
-		if (adev->debugfs_regs[i]) {
-			debugfs_remove(adev->debugfs_regs[i]);
-			adev->debugfs_regs[i] = NULL;
-		}
+	if (type >= AMDGPU_UID_TYPE_MAX) {
+		dev_err_once(uid_info->adev->dev, "Invalid UID type %d\n",
+			     type);
+		return 0;
 	}
-}
 
-int amdgpu_debugfs_init(struct drm_minor *minor)
-{
-	return 0;
-}
-#else
-static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
-{
-	return 0;
+	if (inst >= AMDGPU_UID_INST_MAX) {
+		dev_err_once(uid_info->adev->dev, "Invalid UID instance %d\n",
+			     inst);
+		return 0;
+	}
+
+	return uid_info->uid[type][inst];
 }
-static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev) { }
-#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_df.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_df.h
new file mode 100644
index 000000000000..eb605e79ae0e
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_df.h
@@ -0,0 +1,64 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_DF_H__
+#define __AMDGPU_DF_H__
+
+struct amdgpu_df_hash_status {
+	bool hash_64k;
+	bool hash_2m;
+	bool hash_1g;
+};
+
+struct amdgpu_df_funcs {
+	void (*sw_init)(struct amdgpu_device *adev);
+	void (*sw_fini)(struct amdgpu_device *adev);
+	void (*hw_init)(struct amdgpu_device *adev);
+	void (*enable_broadcast_mode)(struct amdgpu_device *adev,
+				      bool enable);
+	u32 (*get_fb_channel_number)(struct amdgpu_device *adev);
+	u32 (*get_hbm_channel_number)(struct amdgpu_device *adev);
+	void (*update_medium_grain_clock_gating)(struct amdgpu_device *adev,
+						 bool enable);
+	void (*get_clockgating_state)(struct amdgpu_device *adev,
+				      u64 *flags);
+	void (*enable_ecc_force_par_wr_rmw)(struct amdgpu_device *adev,
+					    bool enable);
+	int (*pmc_start)(struct amdgpu_device *adev, uint64_t config,
+					 int counter_idx, int is_add);
+	int (*pmc_stop)(struct amdgpu_device *adev, uint64_t config,
+					 int counter_idx, int is_remove);
+	void (*pmc_get_count)(struct amdgpu_device *adev, uint64_t config,
+					 int counter_idx, uint64_t *count);
+	uint64_t (*get_fica)(struct amdgpu_device *adev, uint32_t ficaa_val);
+	void (*set_fica)(struct amdgpu_device *adev, uint32_t ficaa_val,
+			 uint32_t ficadl_val, uint32_t ficadh_val);
+	bool (*query_ras_poison_mode)(struct amdgpu_device *adev);
+};
+
+struct amdgpu_df {
+	struct amdgpu_df_hash_status	hash_status;
+	const struct amdgpu_df_funcs	*funcs;
+};
+
+#endif /* __AMDGPU_DF_H__ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c
new file mode 100644
index 000000000000..fa2a22dfa048
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c
@@ -0,0 +1,3241 @@
+/*
+ * Copyright 2018-2024 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/firmware.h>
+
+#include "amdgpu.h"
+#include "amdgpu_discovery.h"
+#include "soc15_hw_ip.h"
+#include "discovery.h"
+#include "amdgpu_ras.h"
+
+#include "soc15.h"
+#include "gfx_v9_0.h"
+#include "gfx_v9_4_3.h"
+#include "gmc_v9_0.h"
+#include "df_v1_7.h"
+#include "df_v3_6.h"
+#include "df_v4_3.h"
+#include "df_v4_6_2.h"
+#include "df_v4_15.h"
+#include "nbio_v6_1.h"
+#include "nbio_v7_0.h"
+#include "nbio_v7_4.h"
+#include "nbio_v7_9.h"
+#include "nbio_v7_11.h"
+#include "hdp_v4_0.h"
+#include "vega10_ih.h"
+#include "vega20_ih.h"
+#include "sdma_v4_0.h"
+#include "sdma_v4_4_2.h"
+#include "uvd_v7_0.h"
+#include "vce_v4_0.h"
+#include "vcn_v1_0.h"
+#include "vcn_v2_5.h"
+#include "jpeg_v2_5.h"
+#include "smuio_v9_0.h"
+#include "gmc_v10_0.h"
+#include "gmc_v11_0.h"
+#include "gmc_v12_0.h"
+#include "gfxhub_v2_0.h"
+#include "mmhub_v2_0.h"
+#include "nbio_v2_3.h"
+#include "nbio_v4_3.h"
+#include "nbio_v7_2.h"
+#include "nbio_v7_7.h"
+#include "nbif_v6_3_1.h"
+#include "hdp_v5_0.h"
+#include "hdp_v5_2.h"
+#include "hdp_v6_0.h"
+#include "hdp_v7_0.h"
+#include "nv.h"
+#include "soc21.h"
+#include "soc24.h"
+#include "navi10_ih.h"
+#include "ih_v6_0.h"
+#include "ih_v6_1.h"
+#include "ih_v7_0.h"
+#include "gfx_v10_0.h"
+#include "gfx_v11_0.h"
+#include "gfx_v12_0.h"
+#include "sdma_v5_0.h"
+#include "sdma_v5_2.h"
+#include "sdma_v6_0.h"
+#include "sdma_v7_0.h"
+#include "lsdma_v6_0.h"
+#include "lsdma_v7_0.h"
+#include "vcn_v2_0.h"
+#include "jpeg_v2_0.h"
+#include "vcn_v3_0.h"
+#include "jpeg_v3_0.h"
+#include "vcn_v4_0.h"
+#include "jpeg_v4_0.h"
+#include "vcn_v4_0_3.h"
+#include "jpeg_v4_0_3.h"
+#include "vcn_v4_0_5.h"
+#include "jpeg_v4_0_5.h"
+#include "amdgpu_vkms.h"
+#include "mes_v11_0.h"
+#include "mes_v12_0.h"
+#include "smuio_v11_0.h"
+#include "smuio_v11_0_6.h"
+#include "smuio_v13_0.h"
+#include "smuio_v13_0_3.h"
+#include "smuio_v13_0_6.h"
+#include "smuio_v14_0_2.h"
+#include "vcn_v5_0_0.h"
+#include "vcn_v5_0_1.h"
+#include "jpeg_v5_0_0.h"
+#include "jpeg_v5_0_1.h"
+#include "amdgpu_ras_mgr.h"
+
+#include "amdgpu_vpe.h"
+#if defined(CONFIG_DRM_AMD_ISP)
+#include "amdgpu_isp.h"
+#endif
+
+MODULE_FIRMWARE("amdgpu/ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/vega10_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/vega12_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/vega20_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/raven_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/raven2_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/picasso_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/arcturus_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/aldebaran_ip_discovery.bin");
+
+#define mmIP_DISCOVERY_VERSION  0x16A00
+#define mmRCC_CONFIG_MEMSIZE	0xde3
+#define mmMP0_SMN_C2PMSG_33	0x16061
+#define mmMM_INDEX		0x0
+#define mmMM_INDEX_HI		0x6
+#define mmMM_DATA		0x1
+
+static const char *hw_id_names[HW_ID_MAX] = {
+	[MP1_HWID]		= "MP1",
+	[MP2_HWID]		= "MP2",
+	[THM_HWID]		= "THM",
+	[SMUIO_HWID]		= "SMUIO",
+	[FUSE_HWID]		= "FUSE",
+	[CLKA_HWID]		= "CLKA",
+	[PWR_HWID]		= "PWR",
+	[GC_HWID]		= "GC",
+	[UVD_HWID]		= "UVD",
+	[AUDIO_AZ_HWID]		= "AUDIO_AZ",
+	[ACP_HWID]		= "ACP",
+	[DCI_HWID]		= "DCI",
+	[DMU_HWID]		= "DMU",
+	[DCO_HWID]		= "DCO",
+	[DIO_HWID]		= "DIO",
+	[XDMA_HWID]		= "XDMA",
+	[DCEAZ_HWID]		= "DCEAZ",
+	[DAZ_HWID]		= "DAZ",
+	[SDPMUX_HWID]		= "SDPMUX",
+	[NTB_HWID]		= "NTB",
+	[IOHC_HWID]		= "IOHC",
+	[L2IMU_HWID]		= "L2IMU",
+	[VCE_HWID]		= "VCE",
+	[MMHUB_HWID]		= "MMHUB",
+	[ATHUB_HWID]		= "ATHUB",
+	[DBGU_NBIO_HWID]	= "DBGU_NBIO",
+	[DFX_HWID]		= "DFX",
+	[DBGU0_HWID]		= "DBGU0",
+	[DBGU1_HWID]		= "DBGU1",
+	[OSSSYS_HWID]		= "OSSSYS",
+	[HDP_HWID]		= "HDP",
+	[SDMA0_HWID]		= "SDMA0",
+	[SDMA1_HWID]		= "SDMA1",
+	[SDMA2_HWID]		= "SDMA2",
+	[SDMA3_HWID]		= "SDMA3",
+	[LSDMA_HWID]		= "LSDMA",
+	[ISP_HWID]		= "ISP",
+	[DBGU_IO_HWID]		= "DBGU_IO",
+	[DF_HWID]		= "DF",
+	[CLKB_HWID]		= "CLKB",
+	[FCH_HWID]		= "FCH",
+	[DFX_DAP_HWID]		= "DFX_DAP",
+	[L1IMU_PCIE_HWID]	= "L1IMU_PCIE",
+	[L1IMU_NBIF_HWID]	= "L1IMU_NBIF",
+	[L1IMU_IOAGR_HWID]	= "L1IMU_IOAGR",
+	[L1IMU3_HWID]		= "L1IMU3",
+	[L1IMU4_HWID]		= "L1IMU4",
+	[L1IMU5_HWID]		= "L1IMU5",
+	[L1IMU6_HWID]		= "L1IMU6",
+	[L1IMU7_HWID]		= "L1IMU7",
+	[L1IMU8_HWID]		= "L1IMU8",
+	[L1IMU9_HWID]		= "L1IMU9",
+	[L1IMU10_HWID]		= "L1IMU10",
+	[L1IMU11_HWID]		= "L1IMU11",
+	[L1IMU12_HWID]		= "L1IMU12",
+	[L1IMU13_HWID]		= "L1IMU13",
+	[L1IMU14_HWID]		= "L1IMU14",
+	[L1IMU15_HWID]		= "L1IMU15",
+	[WAFLC_HWID]		= "WAFLC",
+	[FCH_USB_PD_HWID]	= "FCH_USB_PD",
+	[PCIE_HWID]		= "PCIE",
+	[PCS_HWID]		= "PCS",
+	[DDCL_HWID]		= "DDCL",
+	[SST_HWID]		= "SST",
+	[IOAGR_HWID]		= "IOAGR",
+	[NBIF_HWID]		= "NBIF",
+	[IOAPIC_HWID]		= "IOAPIC",
+	[SYSTEMHUB_HWID]	= "SYSTEMHUB",
+	[NTBCCP_HWID]		= "NTBCCP",
+	[UMC_HWID]		= "UMC",
+	[SATA_HWID]		= "SATA",
+	[USB_HWID]		= "USB",
+	[CCXSEC_HWID]		= "CCXSEC",
+	[XGMI_HWID]		= "XGMI",
+	[XGBE_HWID]		= "XGBE",
+	[MP0_HWID]		= "MP0",
+	[VPE_HWID]		= "VPE",
+};
+
+static int hw_id_map[MAX_HWIP] = {
+	[GC_HWIP]	= GC_HWID,
+	[HDP_HWIP]	= HDP_HWID,
+	[SDMA0_HWIP]	= SDMA0_HWID,
+	[SDMA1_HWIP]	= SDMA1_HWID,
+	[SDMA2_HWIP]    = SDMA2_HWID,
+	[SDMA3_HWIP]    = SDMA3_HWID,
+	[LSDMA_HWIP]    = LSDMA_HWID,
+	[MMHUB_HWIP]	= MMHUB_HWID,
+	[ATHUB_HWIP]	= ATHUB_HWID,
+	[NBIO_HWIP]	= NBIF_HWID,
+	[MP0_HWIP]	= MP0_HWID,
+	[MP1_HWIP]	= MP1_HWID,
+	[UVD_HWIP]	= UVD_HWID,
+	[VCE_HWIP]	= VCE_HWID,
+	[DF_HWIP]	= DF_HWID,
+	[DCE_HWIP]	= DMU_HWID,
+	[OSSSYS_HWIP]	= OSSSYS_HWID,
+	[SMUIO_HWIP]	= SMUIO_HWID,
+	[PWR_HWIP]	= PWR_HWID,
+	[NBIF_HWIP]	= NBIF_HWID,
+	[THM_HWIP]	= THM_HWID,
+	[CLK_HWIP]	= CLKA_HWID,
+	[UMC_HWIP]	= UMC_HWID,
+	[XGMI_HWIP]	= XGMI_HWID,
+	[DCI_HWIP]	= DCI_HWID,
+	[PCIE_HWIP]	= PCIE_HWID,
+	[VPE_HWIP]	= VPE_HWID,
+	[ISP_HWIP]	= ISP_HWID,
+};
+
+static int amdgpu_discovery_read_binary_from_sysmem(struct amdgpu_device *adev, uint8_t *binary)
+{
+	u64 tmr_offset, tmr_size, pos;
+	void *discv_regn;
+	int ret;
+
+	ret = amdgpu_acpi_get_tmr_info(adev, &tmr_offset, &tmr_size);
+	if (ret)
+		return ret;
+
+	pos = tmr_offset + tmr_size - DISCOVERY_TMR_OFFSET;
+
+	/* This region is read-only and reserved from system use */
+	discv_regn = memremap(pos, adev->discovery.size, MEMREMAP_WC);
+	if (discv_regn) {
+		memcpy(binary, discv_regn, adev->discovery.size);
+		memunmap(discv_regn);
+		return 0;
+	}
+
+	return -ENOENT;
+}
+
+#define IP_DISCOVERY_V2		2
+#define IP_DISCOVERY_V4		4
+
+static int amdgpu_discovery_read_binary_from_mem(struct amdgpu_device *adev,
+						 uint8_t *binary)
+{
+	bool sz_valid = true;
+	uint64_t vram_size;
+	int i, ret = 0;
+	u32 msg;
+
+	if (!amdgpu_sriov_vf(adev)) {
+		/* It can take up to two second for IFWI init to complete on some dGPUs,
+		 * but generally it should be in the 60-100ms range.  Normally this starts
+		 * as soon as the device gets power so by the time the OS loads this has long
+		 * completed.  However, when a card is hotplugged via e.g., USB4, we need to
+		 * wait for this to complete.  Once the C2PMSG is updated, we can
+		 * continue.
+		 */
+
+		for (i = 0; i < 2000; i++) {
+			msg = RREG32(mmMP0_SMN_C2PMSG_33);
+			if (msg & 0x80000000)
+				break;
+			msleep(1);
+		}
+	}
+
+	vram_size = RREG32(mmRCC_CONFIG_MEMSIZE);
+	if (!vram_size || vram_size == U32_MAX)
+		sz_valid = false;
+	else
+		vram_size <<= 20;
+
+	/*
+	 * If in VRAM, discovery TMR is marked for reservation. If it is in system mem,
+	 * then it is not required to be reserved.
+	 */
+	if (sz_valid) {
+		if (amdgpu_sriov_vf(adev) && adev->virt.is_dynamic_crit_regn_enabled) {
+			/* For SRIOV VFs with dynamic critical region enabled,
+			 * we will get the IPD binary via below call.
+			 * If dynamic critical is disabled, fall through to normal seq.
+			 */
+			if (amdgpu_virt_get_dynamic_data_info(adev,
+						AMD_SRIOV_MSG_IPD_TABLE_ID, binary,
+						&adev->discovery.size)) {
+				dev_err(adev->dev,
+						"failed to read discovery info from dynamic critical region.");
+				ret = -EINVAL;
+				goto exit;
+			}
+		} else {
+			uint64_t pos = vram_size - DISCOVERY_TMR_OFFSET;
+
+			amdgpu_device_vram_access(adev, pos, (uint32_t *)binary,
+					adev->discovery.size, false);
+			adev->discovery.reserve_tmr = true;
+		}
+	} else {
+		ret = amdgpu_discovery_read_binary_from_sysmem(adev, binary);
+	}
+
+	if (ret)
+		dev_err(adev->dev,
+			"failed to read discovery info from memory, vram size read: %llx",
+			vram_size);
+exit:
+	return ret;
+}
+
+static int amdgpu_discovery_read_binary_from_file(struct amdgpu_device *adev,
+							uint8_t *binary,
+							const char *fw_name)
+{
+	const struct firmware *fw;
+	int r;
+
+	r = firmware_request_nowarn(&fw, fw_name, adev->dev);
+	if (r) {
+		if (amdgpu_discovery == 2)
+			dev_err(adev->dev, "can't load firmware \"%s\"\n", fw_name);
+		else
+			drm_info(&adev->ddev, "Optional firmware \"%s\" was not found\n", fw_name);
+		return r;
+	}
+
+	memcpy((u8 *)binary, (u8 *)fw->data, fw->size);
+	release_firmware(fw);
+
+	return 0;
+}
+
+static uint16_t amdgpu_discovery_calculate_checksum(uint8_t *data, uint32_t size)
+{
+	uint16_t checksum = 0;
+	int i;
+
+	for (i = 0; i < size; i++)
+		checksum += data[i];
+
+	return checksum;
+}
+
+static inline bool amdgpu_discovery_verify_checksum(uint8_t *data, uint32_t size,
+						    uint16_t expected)
+{
+	return !!(amdgpu_discovery_calculate_checksum(data, size) == expected);
+}
+
+static inline bool amdgpu_discovery_verify_binary_signature(uint8_t *binary)
+{
+	struct binary_header *bhdr;
+	bhdr = (struct binary_header *)binary;
+
+	return (le32_to_cpu(bhdr->binary_signature) == BINARY_SIGNATURE);
+}
+
+static void amdgpu_discovery_harvest_config_quirk(struct amdgpu_device *adev)
+{
+	/*
+	 * So far, apply this quirk only on those Navy Flounder boards which
+	 * have a bad harvest table of VCN config.
+	 */
+	if ((amdgpu_ip_version(adev, UVD_HWIP, 1) == IP_VERSION(3, 0, 1)) &&
+	    (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(10, 3, 2))) {
+		switch (adev->pdev->revision) {
+		case 0xC1:
+		case 0xC2:
+		case 0xC3:
+		case 0xC5:
+		case 0xC7:
+		case 0xCF:
+		case 0xDF:
+			adev->vcn.harvest_config |= AMDGPU_VCN_HARVEST_VCN1;
+			adev->vcn.inst_mask &= ~AMDGPU_VCN_HARVEST_VCN1;
+			break;
+		default:
+			break;
+		}
+	}
+}
+
+static int amdgpu_discovery_verify_npsinfo(struct amdgpu_device *adev,
+					   struct binary_header *bhdr)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct table_info *info;
+	uint16_t checksum;
+	uint16_t offset;
+
+	info = &bhdr->table_list[NPS_INFO];
+	offset = le16_to_cpu(info->offset);
+	checksum = le16_to_cpu(info->checksum);
+
+	struct nps_info_header *nhdr =
+		(struct nps_info_header *)(discovery_bin + offset);
+
+	if (le32_to_cpu(nhdr->table_id) != NPS_INFO_TABLE_ID) {
+		dev_dbg(adev->dev, "invalid ip discovery nps info table id\n");
+		return -EINVAL;
+	}
+
+	if (!amdgpu_discovery_verify_checksum(discovery_bin + offset,
+					      le32_to_cpu(nhdr->size_bytes),
+					      checksum)) {
+		dev_dbg(adev->dev, "invalid nps info data table checksum\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static const char *amdgpu_discovery_get_fw_name(struct amdgpu_device *adev)
+{
+	if (amdgpu_discovery == 2) {
+		/* Assume there is valid discovery TMR in VRAM even if binary is sideloaded */
+		adev->discovery.reserve_tmr = true;
+		return "amdgpu/ip_discovery.bin";
+	}
+
+	switch (adev->asic_type) {
+	case CHIP_VEGA10:
+		return "amdgpu/vega10_ip_discovery.bin";
+	case CHIP_VEGA12:
+		return "amdgpu/vega12_ip_discovery.bin";
+	case CHIP_RAVEN:
+		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
+			return "amdgpu/raven2_ip_discovery.bin";
+		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
+			return "amdgpu/picasso_ip_discovery.bin";
+		else
+			return "amdgpu/raven_ip_discovery.bin";
+	case CHIP_VEGA20:
+		return "amdgpu/vega20_ip_discovery.bin";
+	case CHIP_ARCTURUS:
+		return "amdgpu/arcturus_ip_discovery.bin";
+	case CHIP_ALDEBARAN:
+		return "amdgpu/aldebaran_ip_discovery.bin";
+	default:
+		return NULL;
+	}
+}
+
+static int amdgpu_discovery_init(struct amdgpu_device *adev)
+{
+	struct table_info *info;
+	struct binary_header *bhdr;
+	uint8_t *discovery_bin;
+	const char *fw_name;
+	uint16_t offset;
+	uint16_t size;
+	uint16_t checksum;
+	int r;
+
+	adev->discovery.bin = kzalloc(DISCOVERY_TMR_SIZE, GFP_KERNEL);
+	if (!adev->discovery.bin)
+		return -ENOMEM;
+	adev->discovery.size = DISCOVERY_TMR_SIZE;
+	adev->discovery.debugfs_blob.data = adev->discovery.bin;
+	adev->discovery.debugfs_blob.size = adev->discovery.size;
+
+	discovery_bin = adev->discovery.bin;
+	/* Read from file if it is the preferred option */
+	fw_name = amdgpu_discovery_get_fw_name(adev);
+	if (fw_name != NULL) {
+		drm_dbg(&adev->ddev, "use ip discovery information from file");
+		r = amdgpu_discovery_read_binary_from_file(adev, discovery_bin,
+							   fw_name);
+		if (r)
+			goto out;
+	} else {
+		drm_dbg(&adev->ddev, "use ip discovery information from memory");
+		r = amdgpu_discovery_read_binary_from_mem(adev, discovery_bin);
+		if (r)
+			goto out;
+	}
+
+	/* check the ip discovery binary signature */
+	if (!amdgpu_discovery_verify_binary_signature(discovery_bin)) {
+		dev_err(adev->dev,
+			"get invalid ip discovery binary signature\n");
+		r = -EINVAL;
+		goto out;
+	}
+
+	bhdr = (struct binary_header *)discovery_bin;
+
+	offset = offsetof(struct binary_header, binary_checksum) +
+		sizeof(bhdr->binary_checksum);
+	size = le16_to_cpu(bhdr->binary_size) - offset;
+	checksum = le16_to_cpu(bhdr->binary_checksum);
+
+	if (!amdgpu_discovery_verify_checksum(discovery_bin + offset, size,
+					      checksum)) {
+		dev_err(adev->dev, "invalid ip discovery binary checksum\n");
+		r = -EINVAL;
+		goto out;
+	}
+
+	info = &bhdr->table_list[IP_DISCOVERY];
+	offset = le16_to_cpu(info->offset);
+	checksum = le16_to_cpu(info->checksum);
+
+	if (offset) {
+		struct ip_discovery_header *ihdr =
+			(struct ip_discovery_header *)(discovery_bin + offset);
+		if (le32_to_cpu(ihdr->signature) != DISCOVERY_TABLE_SIGNATURE) {
+			dev_err(adev->dev, "invalid ip discovery data table signature\n");
+			r = -EINVAL;
+			goto out;
+		}
+
+		if (!amdgpu_discovery_verify_checksum(discovery_bin + offset,
+						      le16_to_cpu(ihdr->size),
+						      checksum)) {
+			dev_err(adev->dev, "invalid ip discovery data table checksum\n");
+			r = -EINVAL;
+			goto out;
+		}
+	}
+
+	info = &bhdr->table_list[GC];
+	offset = le16_to_cpu(info->offset);
+	checksum = le16_to_cpu(info->checksum);
+
+	if (offset) {
+		struct gpu_info_header *ghdr =
+			(struct gpu_info_header *)(discovery_bin + offset);
+
+		if (le32_to_cpu(ghdr->table_id) != GC_TABLE_ID) {
+			dev_err(adev->dev, "invalid ip discovery gc table id\n");
+			r = -EINVAL;
+			goto out;
+		}
+
+		if (!amdgpu_discovery_verify_checksum(discovery_bin + offset,
+						      le32_to_cpu(ghdr->size),
+						      checksum)) {
+			dev_err(adev->dev, "invalid gc data table checksum\n");
+			r = -EINVAL;
+			goto out;
+		}
+	}
+
+	info = &bhdr->table_list[HARVEST_INFO];
+	offset = le16_to_cpu(info->offset);
+	checksum = le16_to_cpu(info->checksum);
+
+	if (offset) {
+		struct harvest_info_header *hhdr =
+			(struct harvest_info_header *)(discovery_bin + offset);
+
+		if (le32_to_cpu(hhdr->signature) != HARVEST_TABLE_SIGNATURE) {
+			dev_err(adev->dev, "invalid ip discovery harvest table signature\n");
+			r = -EINVAL;
+			goto out;
+		}
+
+		if (!amdgpu_discovery_verify_checksum(
+			    discovery_bin + offset,
+			    sizeof(struct harvest_table), checksum)) {
+			dev_err(adev->dev, "invalid harvest data table checksum\n");
+			r = -EINVAL;
+			goto out;
+		}
+	}
+
+	info = &bhdr->table_list[VCN_INFO];
+	offset = le16_to_cpu(info->offset);
+	checksum = le16_to_cpu(info->checksum);
+
+	if (offset) {
+		struct vcn_info_header *vhdr =
+			(struct vcn_info_header *)(discovery_bin + offset);
+
+		if (le32_to_cpu(vhdr->table_id) != VCN_INFO_TABLE_ID) {
+			dev_err(adev->dev, "invalid ip discovery vcn table id\n");
+			r = -EINVAL;
+			goto out;
+		}
+
+		if (!amdgpu_discovery_verify_checksum(
+			    discovery_bin + offset,
+			    le32_to_cpu(vhdr->size_bytes), checksum)) {
+			dev_err(adev->dev, "invalid vcn data table checksum\n");
+			r = -EINVAL;
+			goto out;
+		}
+	}
+
+	info = &bhdr->table_list[MALL_INFO];
+	offset = le16_to_cpu(info->offset);
+	checksum = le16_to_cpu(info->checksum);
+
+	if (0 && offset) {
+		struct mall_info_header *mhdr =
+			(struct mall_info_header *)(discovery_bin + offset);
+
+		if (le32_to_cpu(mhdr->table_id) != MALL_INFO_TABLE_ID) {
+			dev_err(adev->dev, "invalid ip discovery mall table id\n");
+			r = -EINVAL;
+			goto out;
+		}
+
+		if (!amdgpu_discovery_verify_checksum(
+			    discovery_bin + offset,
+			    le32_to_cpu(mhdr->size_bytes), checksum)) {
+			dev_err(adev->dev, "invalid mall data table checksum\n");
+			r = -EINVAL;
+			goto out;
+		}
+	}
+
+	return 0;
+
+out:
+	kfree(adev->discovery.bin);
+	adev->discovery.bin = NULL;
+	if ((amdgpu_discovery != 2) &&
+	    (RREG32(mmIP_DISCOVERY_VERSION) == 4))
+		amdgpu_ras_query_boot_status(adev, 4);
+	return r;
+}
+
+static void amdgpu_discovery_sysfs_fini(struct amdgpu_device *adev);
+
+void amdgpu_discovery_fini(struct amdgpu_device *adev)
+{
+	amdgpu_discovery_sysfs_fini(adev);
+	kfree(adev->discovery.bin);
+	adev->discovery.bin = NULL;
+}
+
+static int amdgpu_discovery_validate_ip(struct amdgpu_device *adev,
+					uint8_t instance, uint16_t hw_id)
+{
+	if (instance >= HWIP_MAX_INSTANCE) {
+		dev_err(adev->dev,
+			"Unexpected instance_number (%d) from ip discovery blob\n",
+			instance);
+		return -EINVAL;
+	}
+	if (hw_id >= HW_ID_MAX) {
+		dev_err(adev->dev,
+			"Unexpected hw_id (%d) from ip discovery blob\n",
+			hw_id);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static void amdgpu_discovery_read_harvest_bit_per_ip(struct amdgpu_device *adev,
+						uint32_t *vcn_harvest_count)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct binary_header *bhdr;
+	struct ip_discovery_header *ihdr;
+	struct die_header *dhdr;
+	struct ip *ip;
+	uint16_t die_offset, ip_offset, num_dies, num_ips;
+	uint16_t hw_id;
+	uint8_t inst;
+	int i, j;
+
+	bhdr = (struct binary_header *)discovery_bin;
+	ihdr = (struct ip_discovery_header
+			*)(discovery_bin +
+			   le16_to_cpu(bhdr->table_list[IP_DISCOVERY].offset));
+	num_dies = le16_to_cpu(ihdr->num_dies);
+
+	/* scan harvest bit of all IP data structures */
+	for (i = 0; i < num_dies; i++) {
+		die_offset = le16_to_cpu(ihdr->die_info[i].die_offset);
+		dhdr = (struct die_header *)(discovery_bin + die_offset);
+		num_ips = le16_to_cpu(dhdr->num_ips);
+		ip_offset = die_offset + sizeof(*dhdr);
+
+		for (j = 0; j < num_ips; j++) {
+			ip = (struct ip *)(discovery_bin + ip_offset);
+			inst = ip->number_instance;
+			hw_id = le16_to_cpu(ip->hw_id);
+			if (amdgpu_discovery_validate_ip(adev, inst, hw_id))
+				goto next_ip;
+
+			if (ip->harvest == 1) {
+				switch (hw_id) {
+				case VCN_HWID:
+					(*vcn_harvest_count)++;
+					if (inst == 0) {
+						adev->vcn.harvest_config |= AMDGPU_VCN_HARVEST_VCN0;
+						adev->vcn.inst_mask &=
+							~AMDGPU_VCN_HARVEST_VCN0;
+						adev->jpeg.inst_mask &=
+							~AMDGPU_VCN_HARVEST_VCN0;
+					} else {
+						adev->vcn.harvest_config |= AMDGPU_VCN_HARVEST_VCN1;
+						adev->vcn.inst_mask &=
+							~AMDGPU_VCN_HARVEST_VCN1;
+						adev->jpeg.inst_mask &=
+							~AMDGPU_VCN_HARVEST_VCN1;
+					}
+					break;
+				case DMU_HWID:
+					adev->harvest_ip_mask |= AMD_HARVEST_IP_DMU_MASK;
+					break;
+				default:
+					break;
+				}
+			}
+next_ip:
+			ip_offset += struct_size(ip, base_address,
+						 ip->num_base_address);
+		}
+	}
+}
+
+static void amdgpu_discovery_read_from_harvest_table(struct amdgpu_device *adev,
+						     uint32_t *vcn_harvest_count,
+						     uint32_t *umc_harvest_count)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct binary_header *bhdr;
+	struct harvest_table *harvest_info;
+	u16 offset;
+	int i;
+	uint32_t umc_harvest_config = 0;
+
+	bhdr = (struct binary_header *)discovery_bin;
+	offset = le16_to_cpu(bhdr->table_list[HARVEST_INFO].offset);
+
+	if (!offset) {
+		dev_err(adev->dev, "invalid harvest table offset\n");
+		return;
+	}
+
+	harvest_info = (struct harvest_table *)(discovery_bin + offset);
+
+	for (i = 0; i < 32; i++) {
+		if (le16_to_cpu(harvest_info->list[i].hw_id) == 0)
+			break;
+
+		switch (le16_to_cpu(harvest_info->list[i].hw_id)) {
+		case VCN_HWID:
+			(*vcn_harvest_count)++;
+			adev->vcn.harvest_config |=
+				(1 << harvest_info->list[i].number_instance);
+			adev->jpeg.harvest_config |=
+				(1 << harvest_info->list[i].number_instance);
+
+			adev->vcn.inst_mask &=
+				~(1U << harvest_info->list[i].number_instance);
+			adev->jpeg.inst_mask &=
+				~(1U << harvest_info->list[i].number_instance);
+			break;
+		case DMU_HWID:
+			adev->harvest_ip_mask |= AMD_HARVEST_IP_DMU_MASK;
+			break;
+		case UMC_HWID:
+			umc_harvest_config |=
+				1 << (le16_to_cpu(harvest_info->list[i].number_instance));
+			(*umc_harvest_count)++;
+			break;
+		case GC_HWID:
+			adev->gfx.xcc_mask &=
+				~(1U << harvest_info->list[i].number_instance);
+			break;
+		case SDMA0_HWID:
+			adev->sdma.sdma_mask &=
+				~(1U << harvest_info->list[i].number_instance);
+			break;
+#if defined(CONFIG_DRM_AMD_ISP)
+		case ISP_HWID:
+			adev->isp.harvest_config |=
+				~(1U << harvest_info->list[i].number_instance);
+			break;
+#endif
+		default:
+			break;
+		}
+	}
+
+	adev->umc.active_mask = ((1 << adev->umc.node_inst_num) - 1) &
+				~umc_harvest_config;
+}
+
+/* ================================================== */
+
+struct ip_hw_instance {
+	struct kobject kobj; /* ip_discovery/die/#die/#hw_id/#instance/<attrs...> */
+
+	int hw_id;
+	u8  num_instance;
+	u8  major, minor, revision;
+	u8  harvest;
+
+	int num_base_addresses;
+	u32 base_addr[] __counted_by(num_base_addresses);
+};
+
+struct ip_hw_id {
+	struct kset hw_id_kset;  /* ip_discovery/die/#die/#hw_id/, contains ip_hw_instance */
+	int hw_id;
+};
+
+struct ip_die_entry {
+	struct kset ip_kset;     /* ip_discovery/die/#die/, contains ip_hw_id  */
+	u16 num_ips;
+};
+
+/* -------------------------------------------------- */
+
+struct ip_hw_instance_attr {
+	struct attribute attr;
+	ssize_t (*show)(struct ip_hw_instance *ip_hw_instance, char *buf);
+};
+
+static ssize_t hw_id_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	return sysfs_emit(buf, "%d\n", ip_hw_instance->hw_id);
+}
+
+static ssize_t num_instance_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	return sysfs_emit(buf, "%d\n", ip_hw_instance->num_instance);
+}
+
+static ssize_t major_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	return sysfs_emit(buf, "%d\n", ip_hw_instance->major);
+}
+
+static ssize_t minor_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	return sysfs_emit(buf, "%d\n", ip_hw_instance->minor);
+}
+
+static ssize_t revision_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	return sysfs_emit(buf, "%d\n", ip_hw_instance->revision);
+}
+
+static ssize_t harvest_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	return sysfs_emit(buf, "0x%01X\n", ip_hw_instance->harvest);
+}
+
+static ssize_t num_base_addresses_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	return sysfs_emit(buf, "%d\n", ip_hw_instance->num_base_addresses);
+}
+
+static ssize_t base_addr_show(struct ip_hw_instance *ip_hw_instance, char *buf)
+{
+	ssize_t res, at;
+	int ii;
+
+	for (res = at = ii = 0; ii < ip_hw_instance->num_base_addresses; ii++) {
+		/* Here we satisfy the condition that, at + size <= PAGE_SIZE.
+		 */
+		if (at + 12 > PAGE_SIZE)
+			break;
+		res = sysfs_emit_at(buf, at, "0x%08X\n",
+				    ip_hw_instance->base_addr[ii]);
+		if (res <= 0)
+			break;
+		at += res;
+	}
+
+	return res < 0 ? res : at;
+}
+
+static struct ip_hw_instance_attr ip_hw_attr[] = {
+	__ATTR_RO(hw_id),
+	__ATTR_RO(num_instance),
+	__ATTR_RO(major),
+	__ATTR_RO(minor),
+	__ATTR_RO(revision),
+	__ATTR_RO(harvest),
+	__ATTR_RO(num_base_addresses),
+	__ATTR_RO(base_addr),
+};
+
+static struct attribute *ip_hw_instance_attrs[ARRAY_SIZE(ip_hw_attr) + 1];
+ATTRIBUTE_GROUPS(ip_hw_instance);
+
+#define to_ip_hw_instance(x) container_of(x, struct ip_hw_instance, kobj)
+#define to_ip_hw_instance_attr(x) container_of(x, struct ip_hw_instance_attr, attr)
+
+static ssize_t ip_hw_instance_attr_show(struct kobject *kobj,
+					struct attribute *attr,
+					char *buf)
+{
+	struct ip_hw_instance *ip_hw_instance = to_ip_hw_instance(kobj);
+	struct ip_hw_instance_attr *ip_hw_attr = to_ip_hw_instance_attr(attr);
+
+	if (!ip_hw_attr->show)
+		return -EIO;
+
+	return ip_hw_attr->show(ip_hw_instance, buf);
+}
+
+static const struct sysfs_ops ip_hw_instance_sysfs_ops = {
+	.show = ip_hw_instance_attr_show,
+};
+
+static void ip_hw_instance_release(struct kobject *kobj)
+{
+	struct ip_hw_instance *ip_hw_instance = to_ip_hw_instance(kobj);
+
+	kfree(ip_hw_instance);
+}
+
+static const struct kobj_type ip_hw_instance_ktype = {
+	.release = ip_hw_instance_release,
+	.sysfs_ops = &ip_hw_instance_sysfs_ops,
+	.default_groups = ip_hw_instance_groups,
+};
+
+/* -------------------------------------------------- */
+
+#define to_ip_hw_id(x)  container_of(to_kset(x), struct ip_hw_id, hw_id_kset)
+
+static void ip_hw_id_release(struct kobject *kobj)
+{
+	struct ip_hw_id *ip_hw_id = to_ip_hw_id(kobj);
+
+	if (!list_empty(&ip_hw_id->hw_id_kset.list))
+		DRM_ERROR("ip_hw_id->hw_id_kset is not empty");
+	kfree(ip_hw_id);
+}
+
+static const struct kobj_type ip_hw_id_ktype = {
+	.release = ip_hw_id_release,
+	.sysfs_ops = &kobj_sysfs_ops,
+};
+
+/* -------------------------------------------------- */
+
+static void die_kobj_release(struct kobject *kobj);
+static void ip_disc_release(struct kobject *kobj);
+
+struct ip_die_entry_attribute {
+	struct attribute attr;
+	ssize_t (*show)(struct ip_die_entry *ip_die_entry, char *buf);
+};
+
+#define to_ip_die_entry_attr(x)  container_of(x, struct ip_die_entry_attribute, attr)
+
+static ssize_t num_ips_show(struct ip_die_entry *ip_die_entry, char *buf)
+{
+	return sysfs_emit(buf, "%d\n", ip_die_entry->num_ips);
+}
+
+/* If there are more ip_die_entry attrs, other than the number of IPs,
+ * we can make this intro an array of attrs, and then initialize
+ * ip_die_entry_attrs in a loop.
+ */
+static struct ip_die_entry_attribute num_ips_attr =
+	__ATTR_RO(num_ips);
+
+static struct attribute *ip_die_entry_attrs[] = {
+	&num_ips_attr.attr,
+	NULL,
+};
+ATTRIBUTE_GROUPS(ip_die_entry); /* ip_die_entry_groups */
+
+#define to_ip_die_entry(x) container_of(to_kset(x), struct ip_die_entry, ip_kset)
+
+static ssize_t ip_die_entry_attr_show(struct kobject *kobj,
+				      struct attribute *attr,
+				      char *buf)
+{
+	struct ip_die_entry_attribute *ip_die_entry_attr = to_ip_die_entry_attr(attr);
+	struct ip_die_entry *ip_die_entry = to_ip_die_entry(kobj);
+
+	if (!ip_die_entry_attr->show)
+		return -EIO;
+
+	return ip_die_entry_attr->show(ip_die_entry, buf);
+}
+
+static void ip_die_entry_release(struct kobject *kobj)
+{
+	struct ip_die_entry *ip_die_entry = to_ip_die_entry(kobj);
+
+	if (!list_empty(&ip_die_entry->ip_kset.list))
+		DRM_ERROR("ip_die_entry->ip_kset is not empty");
+	kfree(ip_die_entry);
+}
+
+static const struct sysfs_ops ip_die_entry_sysfs_ops = {
+	.show = ip_die_entry_attr_show,
+};
+
+static const struct kobj_type ip_die_entry_ktype = {
+	.release = ip_die_entry_release,
+	.sysfs_ops = &ip_die_entry_sysfs_ops,
+	.default_groups = ip_die_entry_groups,
+};
+
+static const struct kobj_type die_kobj_ktype = {
+	.release = die_kobj_release,
+	.sysfs_ops = &kobj_sysfs_ops,
+};
+
+static const struct kobj_type ip_discovery_ktype = {
+	.release = ip_disc_release,
+	.sysfs_ops = &kobj_sysfs_ops,
+};
+
+struct ip_discovery_top {
+	struct kobject kobj;    /* ip_discovery/ */
+	struct kset die_kset;   /* ip_discovery/die/, contains ip_die_entry */
+	struct amdgpu_device *adev;
+};
+
+static void die_kobj_release(struct kobject *kobj)
+{
+	struct ip_discovery_top *ip_top = container_of(to_kset(kobj),
+						       struct ip_discovery_top,
+						       die_kset);
+	if (!list_empty(&ip_top->die_kset.list))
+		DRM_ERROR("ip_top->die_kset is not empty");
+}
+
+static void ip_disc_release(struct kobject *kobj)
+{
+	struct ip_discovery_top *ip_top = container_of(kobj, struct ip_discovery_top,
+						       kobj);
+	struct amdgpu_device *adev = ip_top->adev;
+
+	kfree(ip_top);
+	adev->discovery.ip_top = NULL;
+}
+
+static uint8_t amdgpu_discovery_get_harvest_info(struct amdgpu_device *adev,
+						 uint16_t hw_id, uint8_t inst)
+{
+	uint8_t harvest = 0;
+
+	/* Until a uniform way is figured, get mask based on hwid */
+	switch (hw_id) {
+	case VCN_HWID:
+		/* VCN vs UVD+VCE */
+		if (!amdgpu_ip_version(adev, VCE_HWIP, 0))
+			harvest = ((1 << inst) & adev->vcn.inst_mask) == 0;
+		break;
+	case DMU_HWID:
+		if (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK)
+			harvest = 0x1;
+		break;
+	case UMC_HWID:
+		/* TODO: It needs another parsing; for now, ignore.*/
+		break;
+	case GC_HWID:
+		harvest = ((1 << inst) & adev->gfx.xcc_mask) == 0;
+		break;
+	case SDMA0_HWID:
+		harvest = ((1 << inst) & adev->sdma.sdma_mask) == 0;
+		break;
+	default:
+		break;
+	}
+
+	return harvest;
+}
+
+static int amdgpu_discovery_sysfs_ips(struct amdgpu_device *adev,
+				      struct ip_die_entry *ip_die_entry,
+				      const size_t _ip_offset, const int num_ips,
+				      bool reg_base_64)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	int ii, jj, kk, res;
+	uint16_t hw_id;
+	uint8_t inst;
+
+	DRM_DEBUG("num_ips:%d", num_ips);
+
+	/* Find all IPs of a given HW ID, and add their instance to
+	 * #die/#hw_id/#instance/<attributes>
+	 */
+	for (ii = 0; ii < HW_ID_MAX; ii++) {
+		struct ip_hw_id *ip_hw_id = NULL;
+		size_t ip_offset = _ip_offset;
+
+		for (jj = 0; jj < num_ips; jj++) {
+			struct ip_v4 *ip;
+			struct ip_hw_instance *ip_hw_instance;
+
+			ip = (struct ip_v4 *)(discovery_bin + ip_offset);
+			inst = ip->instance_number;
+			hw_id = le16_to_cpu(ip->hw_id);
+			if (amdgpu_discovery_validate_ip(adev, inst, hw_id) ||
+			    hw_id != ii)
+				goto next_ip;
+
+			DRM_DEBUG("match:%d @ ip_offset:%zu", ii, ip_offset);
+
+			/* We have a hw_id match; register the hw
+			 * block if not yet registered.
+			 */
+			if (!ip_hw_id) {
+				ip_hw_id = kzalloc(sizeof(*ip_hw_id), GFP_KERNEL);
+				if (!ip_hw_id)
+					return -ENOMEM;
+				ip_hw_id->hw_id = ii;
+
+				kobject_set_name(&ip_hw_id->hw_id_kset.kobj, "%d", ii);
+				ip_hw_id->hw_id_kset.kobj.kset = &ip_die_entry->ip_kset;
+				ip_hw_id->hw_id_kset.kobj.ktype = &ip_hw_id_ktype;
+				res = kset_register(&ip_hw_id->hw_id_kset);
+				if (res) {
+					DRM_ERROR("Couldn't register ip_hw_id kset");
+					kfree(ip_hw_id);
+					return res;
+				}
+				if (hw_id_names[ii]) {
+					res = sysfs_create_link(&ip_die_entry->ip_kset.kobj,
+								&ip_hw_id->hw_id_kset.kobj,
+								hw_id_names[ii]);
+					if (res) {
+						DRM_ERROR("Couldn't create IP link %s in IP Die:%s\n",
+							  hw_id_names[ii],
+							  kobject_name(&ip_die_entry->ip_kset.kobj));
+					}
+				}
+			}
+
+			/* Now register its instance.
+			 */
+			ip_hw_instance = kzalloc(struct_size(ip_hw_instance,
+							     base_addr,
+							     ip->num_base_address),
+						 GFP_KERNEL);
+			if (!ip_hw_instance) {
+				DRM_ERROR("no memory for ip_hw_instance");
+				return -ENOMEM;
+			}
+			ip_hw_instance->hw_id = le16_to_cpu(ip->hw_id); /* == ii */
+			ip_hw_instance->num_instance = ip->instance_number;
+			ip_hw_instance->major = ip->major;
+			ip_hw_instance->minor = ip->minor;
+			ip_hw_instance->revision = ip->revision;
+			ip_hw_instance->harvest =
+				amdgpu_discovery_get_harvest_info(
+					adev, ip_hw_instance->hw_id,
+					ip_hw_instance->num_instance);
+			ip_hw_instance->num_base_addresses = ip->num_base_address;
+
+			for (kk = 0; kk < ip_hw_instance->num_base_addresses; kk++) {
+				if (reg_base_64)
+					ip_hw_instance->base_addr[kk] =
+						lower_32_bits(le64_to_cpu(ip->base_address_64[kk])) & 0x3FFFFFFF;
+				else
+					ip_hw_instance->base_addr[kk] = ip->base_address[kk];
+			}
+
+			kobject_init(&ip_hw_instance->kobj, &ip_hw_instance_ktype);
+			ip_hw_instance->kobj.kset = &ip_hw_id->hw_id_kset;
+			res = kobject_add(&ip_hw_instance->kobj, NULL,
+					  "%d", ip_hw_instance->num_instance);
+next_ip:
+			if (reg_base_64)
+				ip_offset += struct_size(ip, base_address_64,
+							 ip->num_base_address);
+			else
+				ip_offset += struct_size(ip, base_address,
+							 ip->num_base_address);
+		}
+	}
+
+	return 0;
+}
+
+static int amdgpu_discovery_sysfs_recurse(struct amdgpu_device *adev)
+{
+	struct ip_discovery_top *ip_top = adev->discovery.ip_top;
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct binary_header *bhdr;
+	struct ip_discovery_header *ihdr;
+	struct die_header *dhdr;
+	struct kset *die_kset = &ip_top->die_kset;
+	u16 num_dies, die_offset, num_ips;
+	size_t ip_offset;
+	int ii, res;
+
+	bhdr = (struct binary_header *)discovery_bin;
+	ihdr = (struct ip_discovery_header
+			*)(discovery_bin +
+			   le16_to_cpu(bhdr->table_list[IP_DISCOVERY].offset));
+	num_dies = le16_to_cpu(ihdr->num_dies);
+
+	DRM_DEBUG("number of dies: %d\n", num_dies);
+
+	for (ii = 0; ii < num_dies; ii++) {
+		struct ip_die_entry *ip_die_entry;
+
+		die_offset = le16_to_cpu(ihdr->die_info[ii].die_offset);
+		dhdr = (struct die_header *)(discovery_bin + die_offset);
+		num_ips = le16_to_cpu(dhdr->num_ips);
+		ip_offset = die_offset + sizeof(*dhdr);
+
+		/* Add the die to the kset.
+		 *
+		 * dhdr->die_id == ii, which was checked in
+		 * amdgpu_discovery_reg_base_init().
+		 */
+
+		ip_die_entry = kzalloc(sizeof(*ip_die_entry), GFP_KERNEL);
+		if (!ip_die_entry)
+			return -ENOMEM;
+
+		ip_die_entry->num_ips = num_ips;
+
+		kobject_set_name(&ip_die_entry->ip_kset.kobj, "%d", le16_to_cpu(dhdr->die_id));
+		ip_die_entry->ip_kset.kobj.kset = die_kset;
+		ip_die_entry->ip_kset.kobj.ktype = &ip_die_entry_ktype;
+		res = kset_register(&ip_die_entry->ip_kset);
+		if (res) {
+			DRM_ERROR("Couldn't register ip_die_entry kset");
+			kfree(ip_die_entry);
+			return res;
+		}
+
+		amdgpu_discovery_sysfs_ips(adev, ip_die_entry, ip_offset, num_ips, !!ihdr->base_addr_64_bit);
+	}
+
+	return 0;
+}
+
+static int amdgpu_discovery_sysfs_init(struct amdgpu_device *adev)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct ip_discovery_top *ip_top;
+	struct kset *die_kset;
+	int res, ii;
+
+	if (!discovery_bin)
+		return -EINVAL;
+
+	ip_top = kzalloc(sizeof(*ip_top), GFP_KERNEL);
+	if (!ip_top)
+		return -ENOMEM;
+
+	ip_top->adev = adev;
+	adev->discovery.ip_top = ip_top;
+	res = kobject_init_and_add(&ip_top->kobj, &ip_discovery_ktype,
+				   &adev->dev->kobj, "ip_discovery");
+	if (res) {
+		DRM_ERROR("Couldn't init and add ip_discovery/");
+		goto Err;
+	}
+
+	die_kset = &ip_top->die_kset;
+	kobject_set_name(&die_kset->kobj, "%s", "die");
+	die_kset->kobj.parent = &ip_top->kobj;
+	die_kset->kobj.ktype = &die_kobj_ktype;
+	res = kset_register(&ip_top->die_kset);
+	if (res) {
+		DRM_ERROR("Couldn't register die_kset");
+		goto Err;
+	}
+
+	for (ii = 0; ii < ARRAY_SIZE(ip_hw_attr); ii++)
+		ip_hw_instance_attrs[ii] = &ip_hw_attr[ii].attr;
+	ip_hw_instance_attrs[ii] = NULL;
+
+	res = amdgpu_discovery_sysfs_recurse(adev);
+
+	return res;
+Err:
+	kobject_put(&ip_top->kobj);
+	return res;
+}
+
+/* -------------------------------------------------- */
+
+#define list_to_kobj(el) container_of(el, struct kobject, entry)
+
+static void amdgpu_discovery_sysfs_ip_hw_free(struct ip_hw_id *ip_hw_id)
+{
+	struct list_head *el, *tmp;
+	struct kset *hw_id_kset;
+
+	hw_id_kset = &ip_hw_id->hw_id_kset;
+	spin_lock(&hw_id_kset->list_lock);
+	list_for_each_prev_safe(el, tmp, &hw_id_kset->list) {
+		list_del_init(el);
+		spin_unlock(&hw_id_kset->list_lock);
+		/* kobject is embedded in ip_hw_instance */
+		kobject_put(list_to_kobj(el));
+		spin_lock(&hw_id_kset->list_lock);
+	}
+	spin_unlock(&hw_id_kset->list_lock);
+	kobject_put(&ip_hw_id->hw_id_kset.kobj);
+}
+
+static void amdgpu_discovery_sysfs_die_free(struct ip_die_entry *ip_die_entry)
+{
+	struct list_head *el, *tmp;
+	struct kset *ip_kset;
+
+	ip_kset = &ip_die_entry->ip_kset;
+	spin_lock(&ip_kset->list_lock);
+	list_for_each_prev_safe(el, tmp, &ip_kset->list) {
+		list_del_init(el);
+		spin_unlock(&ip_kset->list_lock);
+		amdgpu_discovery_sysfs_ip_hw_free(to_ip_hw_id(list_to_kobj(el)));
+		spin_lock(&ip_kset->list_lock);
+	}
+	spin_unlock(&ip_kset->list_lock);
+	kobject_put(&ip_die_entry->ip_kset.kobj);
+}
+
+static void amdgpu_discovery_sysfs_fini(struct amdgpu_device *adev)
+{
+	struct ip_discovery_top *ip_top = adev->discovery.ip_top;
+	struct list_head *el, *tmp;
+	struct kset *die_kset;
+
+	die_kset = &ip_top->die_kset;
+	spin_lock(&die_kset->list_lock);
+	list_for_each_prev_safe(el, tmp, &die_kset->list) {
+		list_del_init(el);
+		spin_unlock(&die_kset->list_lock);
+		amdgpu_discovery_sysfs_die_free(to_ip_die_entry(list_to_kobj(el)));
+		spin_lock(&die_kset->list_lock);
+	}
+	spin_unlock(&die_kset->list_lock);
+	kobject_put(&ip_top->die_kset.kobj);
+	kobject_put(&ip_top->kobj);
+}
+
+/* ================================================== */
+
+static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev)
+{
+	uint8_t num_base_address, subrev, variant;
+	struct binary_header *bhdr;
+	struct ip_discovery_header *ihdr;
+	struct die_header *dhdr;
+	uint8_t *discovery_bin;
+	struct ip_v4 *ip;
+	uint16_t die_offset;
+	uint16_t ip_offset;
+	uint16_t num_dies;
+	uint32_t wafl_ver;
+	uint16_t num_ips;
+	uint16_t hw_id;
+	uint8_t inst;
+	int hw_ip;
+	int i, j, k;
+	int r;
+
+	r = amdgpu_discovery_init(adev);
+	if (r)
+		return r;
+	discovery_bin = adev->discovery.bin;
+	wafl_ver = 0;
+	adev->gfx.xcc_mask = 0;
+	adev->sdma.sdma_mask = 0;
+	adev->vcn.inst_mask = 0;
+	adev->jpeg.inst_mask = 0;
+	bhdr = (struct binary_header *)discovery_bin;
+	ihdr = (struct ip_discovery_header
+			*)(discovery_bin +
+			   le16_to_cpu(bhdr->table_list[IP_DISCOVERY].offset));
+	num_dies = le16_to_cpu(ihdr->num_dies);
+
+	DRM_DEBUG("number of dies: %d\n", num_dies);
+
+	for (i = 0; i < num_dies; i++) {
+		die_offset = le16_to_cpu(ihdr->die_info[i].die_offset);
+		dhdr = (struct die_header *)(discovery_bin + die_offset);
+		num_ips = le16_to_cpu(dhdr->num_ips);
+		ip_offset = die_offset + sizeof(*dhdr);
+
+		if (le16_to_cpu(dhdr->die_id) != i) {
+			DRM_ERROR("invalid die id %d, expected %d\n",
+					le16_to_cpu(dhdr->die_id), i);
+			return -EINVAL;
+		}
+
+		DRM_DEBUG("number of hardware IPs on die%d: %d\n",
+				le16_to_cpu(dhdr->die_id), num_ips);
+
+		for (j = 0; j < num_ips; j++) {
+			ip = (struct ip_v4 *)(discovery_bin + ip_offset);
+
+			inst = ip->instance_number;
+			hw_id = le16_to_cpu(ip->hw_id);
+			if (amdgpu_discovery_validate_ip(adev, inst, hw_id))
+				goto next_ip;
+
+			num_base_address = ip->num_base_address;
+
+			DRM_DEBUG("%s(%d) #%d v%d.%d.%d:\n",
+				  hw_id_names[le16_to_cpu(ip->hw_id)],
+				  le16_to_cpu(ip->hw_id),
+				  ip->instance_number,
+				  ip->major, ip->minor,
+				  ip->revision);
+
+			if (le16_to_cpu(ip->hw_id) == VCN_HWID) {
+				/* Bit [5:0]: original revision value
+				 * Bit [7:6]: en/decode capability:
+				 *     0b00 : VCN function normally
+				 *     0b10 : encode is disabled
+				 *     0b01 : decode is disabled
+				 */
+				if (adev->vcn.num_vcn_inst <
+				    AMDGPU_MAX_VCN_INSTANCES) {
+					adev->vcn.inst[adev->vcn.num_vcn_inst].vcn_config =
+						ip->revision & 0xc0;
+					adev->vcn.num_vcn_inst++;
+					adev->vcn.inst_mask |=
+						(1U << ip->instance_number);
+					adev->jpeg.inst_mask |=
+						(1U << ip->instance_number);
+				} else {
+					dev_err(adev->dev, "Too many VCN instances: %d vs %d\n",
+						adev->vcn.num_vcn_inst + 1,
+						AMDGPU_MAX_VCN_INSTANCES);
+				}
+				ip->revision &= ~0xc0;
+			}
+			if (le16_to_cpu(ip->hw_id) == SDMA0_HWID ||
+			    le16_to_cpu(ip->hw_id) == SDMA1_HWID ||
+			    le16_to_cpu(ip->hw_id) == SDMA2_HWID ||
+			    le16_to_cpu(ip->hw_id) == SDMA3_HWID) {
+				if (adev->sdma.num_instances <
+				    AMDGPU_MAX_SDMA_INSTANCES) {
+					adev->sdma.num_instances++;
+					adev->sdma.sdma_mask |=
+						(1U << ip->instance_number);
+				} else {
+					dev_err(adev->dev, "Too many SDMA instances: %d vs %d\n",
+						adev->sdma.num_instances + 1,
+						AMDGPU_MAX_SDMA_INSTANCES);
+				}
+			}
+
+			if (le16_to_cpu(ip->hw_id) == VPE_HWID) {
+				if (adev->vpe.num_instances < AMDGPU_MAX_VPE_INSTANCES)
+					adev->vpe.num_instances++;
+				else
+					dev_err(adev->dev, "Too many VPE instances: %d vs %d\n",
+						adev->vpe.num_instances + 1,
+						AMDGPU_MAX_VPE_INSTANCES);
+			}
+
+			if (le16_to_cpu(ip->hw_id) == UMC_HWID) {
+				adev->gmc.num_umc++;
+				adev->umc.node_inst_num++;
+			}
+
+			if (le16_to_cpu(ip->hw_id) == GC_HWID)
+				adev->gfx.xcc_mask |=
+					(1U << ip->instance_number);
+
+			if (!wafl_ver && le16_to_cpu(ip->hw_id) == WAFLC_HWID)
+				wafl_ver = IP_VERSION_FULL(ip->major, ip->minor,
+							   ip->revision, 0, 0);
+
+			for (k = 0; k < num_base_address; k++) {
+				/*
+				 * convert the endianness of base addresses in place,
+				 * so that we don't need to convert them when accessing adev->reg_offset.
+				 */
+				if (ihdr->base_addr_64_bit)
+					/* Truncate the 64bit base address from ip discovery
+					 * and only store lower 32bit ip base in reg_offset[].
+					 * Bits > 32 follows ASIC specific format, thus just
+					 * discard them and handle it within specific ASIC.
+					 * By this way reg_offset[] and related helpers can
+					 * stay unchanged.
+					 * The base address is in dwords, thus clear the
+					 * highest 2 bits to store.
+					 */
+					ip->base_address[k] =
+						lower_32_bits(le64_to_cpu(ip->base_address_64[k])) & 0x3FFFFFFF;
+				else
+					ip->base_address[k] = le32_to_cpu(ip->base_address[k]);
+				DRM_DEBUG("\t0x%08x\n", ip->base_address[k]);
+			}
+
+			for (hw_ip = 0; hw_ip < MAX_HWIP; hw_ip++) {
+				if (hw_id_map[hw_ip] == le16_to_cpu(ip->hw_id) &&
+				    hw_id_map[hw_ip] != 0) {
+					DRM_DEBUG("set register base offset for %s\n",
+							hw_id_names[le16_to_cpu(ip->hw_id)]);
+					adev->reg_offset[hw_ip][ip->instance_number] =
+						ip->base_address;
+					/* Instance support is somewhat inconsistent.
+					 * SDMA is a good example.  Sienna cichlid has 4 total
+					 * SDMA instances, each enumerated separately (HWIDs
+					 * 42, 43, 68, 69).  Arcturus has 8 total SDMA instances,
+					 * but they are enumerated as multiple instances of the
+					 * same HWIDs (4x HWID 42, 4x HWID 43).  UMC is another
+					 * example.  On most chips there are multiple instances
+					 * with the same HWID.
+					 */
+
+					if (ihdr->version < 3) {
+						subrev = 0;
+						variant = 0;
+					} else {
+						subrev = ip->sub_revision;
+						variant = ip->variant;
+					}
+
+					adev->ip_versions[hw_ip]
+							 [ip->instance_number] =
+						IP_VERSION_FULL(ip->major,
+								ip->minor,
+								ip->revision,
+								variant,
+								subrev);
+				}
+			}
+
+next_ip:
+			if (ihdr->base_addr_64_bit)
+				ip_offset += struct_size(ip, base_address_64, ip->num_base_address);
+			else
+				ip_offset += struct_size(ip, base_address, ip->num_base_address);
+		}
+	}
+
+	if (wafl_ver && !adev->ip_versions[XGMI_HWIP][0])
+		adev->ip_versions[XGMI_HWIP][0] = wafl_ver;
+
+	return 0;
+}
+
+static void amdgpu_discovery_harvest_ip(struct amdgpu_device *adev)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct ip_discovery_header *ihdr;
+	struct binary_header *bhdr;
+	int vcn_harvest_count = 0;
+	int umc_harvest_count = 0;
+	uint16_t offset, ihdr_ver;
+
+	bhdr = (struct binary_header *)discovery_bin;
+	offset = le16_to_cpu(bhdr->table_list[IP_DISCOVERY].offset);
+	ihdr = (struct ip_discovery_header *)(discovery_bin + offset);
+	ihdr_ver = le16_to_cpu(ihdr->version);
+	/*
+	 * Harvest table does not fit Navi1x and legacy GPUs,
+	 * so read harvest bit per IP data structure to set
+	 * harvest configuration.
+	 */
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(10, 2, 0) &&
+	    ihdr_ver <= 2) {
+		if ((adev->pdev->device == 0x731E &&
+			(adev->pdev->revision == 0xC6 ||
+			 adev->pdev->revision == 0xC7)) ||
+			(adev->pdev->device == 0x7340 &&
+			 adev->pdev->revision == 0xC9) ||
+			(adev->pdev->device == 0x7360 &&
+			 adev->pdev->revision == 0xC7))
+			amdgpu_discovery_read_harvest_bit_per_ip(adev,
+				&vcn_harvest_count);
+	} else {
+		amdgpu_discovery_read_from_harvest_table(adev,
+							 &vcn_harvest_count,
+							 &umc_harvest_count);
+	}
+
+	amdgpu_discovery_harvest_config_quirk(adev);
+
+	if (vcn_harvest_count == adev->vcn.num_vcn_inst) {
+		adev->harvest_ip_mask |= AMD_HARVEST_IP_VCN_MASK;
+		adev->harvest_ip_mask |= AMD_HARVEST_IP_JPEG_MASK;
+	}
+
+	if (umc_harvest_count < adev->gmc.num_umc) {
+		adev->gmc.num_umc -= umc_harvest_count;
+	}
+}
+
+union gc_info {
+	struct gc_info_v1_0 v1;
+	struct gc_info_v1_1 v1_1;
+	struct gc_info_v1_2 v1_2;
+	struct gc_info_v1_3 v1_3;
+	struct gc_info_v2_0 v2;
+	struct gc_info_v2_1 v2_1;
+};
+
+static int amdgpu_discovery_get_gfx_info(struct amdgpu_device *adev)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct binary_header *bhdr;
+	union gc_info *gc_info;
+	u16 offset;
+
+	if (!discovery_bin) {
+		DRM_ERROR("ip discovery uninitialized\n");
+		return -EINVAL;
+	}
+
+	bhdr = (struct binary_header *)discovery_bin;
+	offset = le16_to_cpu(bhdr->table_list[GC].offset);
+
+	if (!offset)
+		return 0;
+
+	gc_info = (union gc_info *)(discovery_bin + offset);
+
+	switch (le16_to_cpu(gc_info->v1.header.version_major)) {
+	case 1:
+		adev->gfx.config.max_shader_engines = le32_to_cpu(gc_info->v1.gc_num_se);
+		adev->gfx.config.max_cu_per_sh = 2 * (le32_to_cpu(gc_info->v1.gc_num_wgp0_per_sa) +
+						      le32_to_cpu(gc_info->v1.gc_num_wgp1_per_sa));
+		adev->gfx.config.max_sh_per_se = le32_to_cpu(gc_info->v1.gc_num_sa_per_se);
+		adev->gfx.config.max_backends_per_se = le32_to_cpu(gc_info->v1.gc_num_rb_per_se);
+		adev->gfx.config.max_texture_channel_caches = le32_to_cpu(gc_info->v1.gc_num_gl2c);
+		adev->gfx.config.max_gprs = le32_to_cpu(gc_info->v1.gc_num_gprs);
+		adev->gfx.config.max_gs_threads = le32_to_cpu(gc_info->v1.gc_num_max_gs_thds);
+		adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gc_info->v1.gc_gs_table_depth);
+		adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gc_info->v1.gc_gsprim_buff_depth);
+		adev->gfx.config.double_offchip_lds_buf = le32_to_cpu(gc_info->v1.gc_double_offchip_lds_buffer);
+		adev->gfx.cu_info.wave_front_size = le32_to_cpu(gc_info->v1.gc_wave_size);
+		adev->gfx.cu_info.max_waves_per_simd = le32_to_cpu(gc_info->v1.gc_max_waves_per_simd);
+		adev->gfx.cu_info.max_scratch_slots_per_cu = le32_to_cpu(gc_info->v1.gc_max_scratch_slots_per_cu);
+		adev->gfx.cu_info.lds_size = le32_to_cpu(gc_info->v1.gc_lds_size);
+		adev->gfx.config.num_sc_per_sh = le32_to_cpu(gc_info->v1.gc_num_sc_per_se) /
+			le32_to_cpu(gc_info->v1.gc_num_sa_per_se);
+		adev->gfx.config.num_packer_per_sc = le32_to_cpu(gc_info->v1.gc_num_packer_per_sc);
+		if (le16_to_cpu(gc_info->v1.header.version_minor) >= 1) {
+			adev->gfx.config.gc_num_tcp_per_sa = le32_to_cpu(gc_info->v1_1.gc_num_tcp_per_sa);
+			adev->gfx.config.gc_num_sdp_interface = le32_to_cpu(gc_info->v1_1.gc_num_sdp_interface);
+			adev->gfx.config.gc_num_tcps = le32_to_cpu(gc_info->v1_1.gc_num_tcps);
+		}
+		if (le16_to_cpu(gc_info->v1.header.version_minor) >= 2) {
+			adev->gfx.config.gc_num_tcp_per_wpg = le32_to_cpu(gc_info->v1_2.gc_num_tcp_per_wpg);
+			adev->gfx.config.gc_tcp_l1_size = le32_to_cpu(gc_info->v1_2.gc_tcp_l1_size);
+			adev->gfx.config.gc_num_sqc_per_wgp = le32_to_cpu(gc_info->v1_2.gc_num_sqc_per_wgp);
+			adev->gfx.config.gc_l1_instruction_cache_size_per_sqc = le32_to_cpu(gc_info->v1_2.gc_l1_instruction_cache_size_per_sqc);
+			adev->gfx.config.gc_l1_data_cache_size_per_sqc = le32_to_cpu(gc_info->v1_2.gc_l1_data_cache_size_per_sqc);
+			adev->gfx.config.gc_gl1c_per_sa = le32_to_cpu(gc_info->v1_2.gc_gl1c_per_sa);
+			adev->gfx.config.gc_gl1c_size_per_instance = le32_to_cpu(gc_info->v1_2.gc_gl1c_size_per_instance);
+			adev->gfx.config.gc_gl2c_per_gpu = le32_to_cpu(gc_info->v1_2.gc_gl2c_per_gpu);
+		}
+		if (le16_to_cpu(gc_info->v1.header.version_minor) >= 3) {
+			adev->gfx.config.gc_tcp_size_per_cu = le32_to_cpu(gc_info->v1_3.gc_tcp_size_per_cu);
+			adev->gfx.config.gc_tcp_cache_line_size = le32_to_cpu(gc_info->v1_3.gc_tcp_cache_line_size);
+			adev->gfx.config.gc_instruction_cache_size_per_sqc = le32_to_cpu(gc_info->v1_3.gc_instruction_cache_size_per_sqc);
+			adev->gfx.config.gc_instruction_cache_line_size = le32_to_cpu(gc_info->v1_3.gc_instruction_cache_line_size);
+			adev->gfx.config.gc_scalar_data_cache_size_per_sqc = le32_to_cpu(gc_info->v1_3.gc_scalar_data_cache_size_per_sqc);
+			adev->gfx.config.gc_scalar_data_cache_line_size = le32_to_cpu(gc_info->v1_3.gc_scalar_data_cache_line_size);
+			adev->gfx.config.gc_tcc_size = le32_to_cpu(gc_info->v1_3.gc_tcc_size);
+			adev->gfx.config.gc_tcc_cache_line_size = le32_to_cpu(gc_info->v1_3.gc_tcc_cache_line_size);
+		}
+		break;
+	case 2:
+		adev->gfx.config.max_shader_engines = le32_to_cpu(gc_info->v2.gc_num_se);
+		adev->gfx.config.max_cu_per_sh = le32_to_cpu(gc_info->v2.gc_num_cu_per_sh);
+		adev->gfx.config.max_sh_per_se = le32_to_cpu(gc_info->v2.gc_num_sh_per_se);
+		adev->gfx.config.max_backends_per_se = le32_to_cpu(gc_info->v2.gc_num_rb_per_se);
+		adev->gfx.config.max_texture_channel_caches = le32_to_cpu(gc_info->v2.gc_num_tccs);
+		adev->gfx.config.max_gprs = le32_to_cpu(gc_info->v2.gc_num_gprs);
+		adev->gfx.config.max_gs_threads = le32_to_cpu(gc_info->v2.gc_num_max_gs_thds);
+		adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gc_info->v2.gc_gs_table_depth);
+		adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gc_info->v2.gc_gsprim_buff_depth);
+		adev->gfx.config.double_offchip_lds_buf = le32_to_cpu(gc_info->v2.gc_double_offchip_lds_buffer);
+		adev->gfx.cu_info.wave_front_size = le32_to_cpu(gc_info->v2.gc_wave_size);
+		adev->gfx.cu_info.max_waves_per_simd = le32_to_cpu(gc_info->v2.gc_max_waves_per_simd);
+		adev->gfx.cu_info.max_scratch_slots_per_cu = le32_to_cpu(gc_info->v2.gc_max_scratch_slots_per_cu);
+		adev->gfx.cu_info.lds_size = le32_to_cpu(gc_info->v2.gc_lds_size);
+		adev->gfx.config.num_sc_per_sh = le32_to_cpu(gc_info->v2.gc_num_sc_per_se) /
+			le32_to_cpu(gc_info->v2.gc_num_sh_per_se);
+		adev->gfx.config.num_packer_per_sc = le32_to_cpu(gc_info->v2.gc_num_packer_per_sc);
+		if (le16_to_cpu(gc_info->v2.header.version_minor) == 1) {
+			adev->gfx.config.gc_num_tcp_per_sa = le32_to_cpu(gc_info->v2_1.gc_num_tcp_per_sh);
+			adev->gfx.config.gc_tcp_size_per_cu = le32_to_cpu(gc_info->v2_1.gc_tcp_size_per_cu);
+			adev->gfx.config.gc_num_sdp_interface = le32_to_cpu(gc_info->v2_1.gc_num_sdp_interface); /* per XCD */
+			adev->gfx.config.gc_num_cu_per_sqc = le32_to_cpu(gc_info->v2_1.gc_num_cu_per_sqc);
+			adev->gfx.config.gc_l1_instruction_cache_size_per_sqc = le32_to_cpu(gc_info->v2_1.gc_instruction_cache_size_per_sqc);
+			adev->gfx.config.gc_l1_data_cache_size_per_sqc = le32_to_cpu(gc_info->v2_1.gc_scalar_data_cache_size_per_sqc);
+			adev->gfx.config.gc_tcc_size = le32_to_cpu(gc_info->v2_1.gc_tcc_size); /* per XCD */
+		}
+		break;
+	default:
+		dev_err(adev->dev,
+			"Unhandled GC info table %d.%d\n",
+			le16_to_cpu(gc_info->v1.header.version_major),
+			le16_to_cpu(gc_info->v1.header.version_minor));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+union mall_info {
+	struct mall_info_v1_0 v1;
+	struct mall_info_v2_0 v2;
+};
+
+static int amdgpu_discovery_get_mall_info(struct amdgpu_device *adev)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct binary_header *bhdr;
+	union mall_info *mall_info;
+	u32 u, mall_size_per_umc, m_s_present, half_use;
+	u64 mall_size;
+	u16 offset;
+
+	if (!discovery_bin) {
+		DRM_ERROR("ip discovery uninitialized\n");
+		return -EINVAL;
+	}
+
+	bhdr = (struct binary_header *)discovery_bin;
+	offset = le16_to_cpu(bhdr->table_list[MALL_INFO].offset);
+
+	if (!offset)
+		return 0;
+
+	mall_info = (union mall_info *)(discovery_bin + offset);
+
+	switch (le16_to_cpu(mall_info->v1.header.version_major)) {
+	case 1:
+		mall_size = 0;
+		mall_size_per_umc = le32_to_cpu(mall_info->v1.mall_size_per_m);
+		m_s_present = le32_to_cpu(mall_info->v1.m_s_present);
+		half_use = le32_to_cpu(mall_info->v1.m_half_use);
+		for (u = 0; u < adev->gmc.num_umc; u++) {
+			if (m_s_present & (1 << u))
+				mall_size += mall_size_per_umc * 2;
+			else if (half_use & (1 << u))
+				mall_size += mall_size_per_umc / 2;
+			else
+				mall_size += mall_size_per_umc;
+		}
+		adev->gmc.mall_size = mall_size;
+		adev->gmc.m_half_use = half_use;
+		break;
+	case 2:
+		mall_size_per_umc = le32_to_cpu(mall_info->v2.mall_size_per_umc);
+		adev->gmc.mall_size = (uint64_t)mall_size_per_umc * adev->gmc.num_umc;
+		break;
+	default:
+		dev_err(adev->dev,
+			"Unhandled MALL info table %d.%d\n",
+			le16_to_cpu(mall_info->v1.header.version_major),
+			le16_to_cpu(mall_info->v1.header.version_minor));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+union vcn_info {
+	struct vcn_info_v1_0 v1;
+};
+
+static int amdgpu_discovery_get_vcn_info(struct amdgpu_device *adev)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct binary_header *bhdr;
+	union vcn_info *vcn_info;
+	u16 offset;
+	int v;
+
+	if (!discovery_bin) {
+		DRM_ERROR("ip discovery uninitialized\n");
+		return -EINVAL;
+	}
+
+	/* num_vcn_inst is currently limited to AMDGPU_MAX_VCN_INSTANCES
+	 * which is smaller than VCN_INFO_TABLE_MAX_NUM_INSTANCES
+	 * but that may change in the future with new GPUs so keep this
+	 * check for defensive purposes.
+	 */
+	if (adev->vcn.num_vcn_inst > VCN_INFO_TABLE_MAX_NUM_INSTANCES) {
+		dev_err(adev->dev, "invalid vcn instances\n");
+		return -EINVAL;
+	}
+
+	bhdr = (struct binary_header *)discovery_bin;
+	offset = le16_to_cpu(bhdr->table_list[VCN_INFO].offset);
+
+	if (!offset)
+		return 0;
+
+	vcn_info = (union vcn_info *)(discovery_bin + offset);
+
+	switch (le16_to_cpu(vcn_info->v1.header.version_major)) {
+	case 1:
+		/* num_vcn_inst is currently limited to AMDGPU_MAX_VCN_INSTANCES
+		 * so this won't overflow.
+		 */
+		for (v = 0; v < adev->vcn.num_vcn_inst; v++) {
+			adev->vcn.inst[v].vcn_codec_disable_mask =
+				le32_to_cpu(vcn_info->v1.instance_info[v].fuse_data.all_bits);
+		}
+		break;
+	default:
+		dev_err(adev->dev,
+			"Unhandled VCN info table %d.%d\n",
+			le16_to_cpu(vcn_info->v1.header.version_major),
+			le16_to_cpu(vcn_info->v1.header.version_minor));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+union nps_info {
+	struct nps_info_v1_0 v1;
+};
+
+static int amdgpu_discovery_refresh_nps_info(struct amdgpu_device *adev,
+					     union nps_info *nps_data)
+{
+	uint64_t vram_size, pos, offset;
+	struct nps_info_header *nhdr;
+	struct binary_header bhdr;
+	uint16_t checksum;
+
+	vram_size = (uint64_t)RREG32(mmRCC_CONFIG_MEMSIZE) << 20;
+	pos = vram_size - DISCOVERY_TMR_OFFSET;
+	amdgpu_device_vram_access(adev, pos, &bhdr, sizeof(bhdr), false);
+
+	offset = le16_to_cpu(bhdr.table_list[NPS_INFO].offset);
+	checksum = le16_to_cpu(bhdr.table_list[NPS_INFO].checksum);
+
+	amdgpu_device_vram_access(adev, (pos + offset), nps_data,
+				  sizeof(*nps_data), false);
+
+	nhdr = (struct nps_info_header *)(nps_data);
+	if (!amdgpu_discovery_verify_checksum((uint8_t *)nps_data,
+					      le32_to_cpu(nhdr->size_bytes),
+					      checksum)) {
+		dev_err(adev->dev, "nps data refresh, checksum mismatch\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+int amdgpu_discovery_get_nps_info(struct amdgpu_device *adev,
+				  uint32_t *nps_type,
+				  struct amdgpu_gmc_memrange **ranges,
+				  int *range_cnt, bool refresh)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct amdgpu_gmc_memrange *mem_ranges;
+	struct binary_header *bhdr;
+	union nps_info *nps_info;
+	union nps_info nps_data;
+	u16 offset;
+	int i, r;
+
+	if (!nps_type || !range_cnt || !ranges)
+		return -EINVAL;
+
+	if (refresh) {
+		r = amdgpu_discovery_refresh_nps_info(adev, &nps_data);
+		if (r)
+			return r;
+		nps_info = &nps_data;
+	} else {
+		if (!discovery_bin) {
+			dev_err(adev->dev,
+				"fetch mem range failed, ip discovery uninitialized\n");
+			return -EINVAL;
+		}
+
+		bhdr = (struct binary_header *)discovery_bin;
+		offset = le16_to_cpu(bhdr->table_list[NPS_INFO].offset);
+
+		if (!offset)
+			return -ENOENT;
+
+		/* If verification fails, return as if NPS table doesn't exist */
+		if (amdgpu_discovery_verify_npsinfo(adev, bhdr))
+			return -ENOENT;
+
+		nps_info = (union nps_info *)(discovery_bin + offset);
+	}
+
+	switch (le16_to_cpu(nps_info->v1.header.version_major)) {
+	case 1:
+		mem_ranges = kvcalloc(nps_info->v1.count,
+				      sizeof(*mem_ranges),
+				      GFP_KERNEL);
+		if (!mem_ranges)
+			return -ENOMEM;
+		*nps_type = nps_info->v1.nps_type;
+		*range_cnt = nps_info->v1.count;
+		for (i = 0; i < *range_cnt; i++) {
+			mem_ranges[i].base_address =
+				nps_info->v1.instance_info[i].base_address;
+			mem_ranges[i].limit_address =
+				nps_info->v1.instance_info[i].limit_address;
+			mem_ranges[i].nid_mask = -1;
+			mem_ranges[i].flags = 0;
+		}
+		*ranges = mem_ranges;
+		break;
+	default:
+		dev_err(adev->dev, "Unhandled NPS info table %d.%d\n",
+			le16_to_cpu(nps_info->v1.header.version_major),
+			le16_to_cpu(nps_info->v1.header.version_minor));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int amdgpu_discovery_set_common_ip_blocks(struct amdgpu_device *adev)
+{
+	/* what IP to use for this? */
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	case IP_VERSION(9, 0, 1):
+	case IP_VERSION(9, 1, 0):
+	case IP_VERSION(9, 2, 1):
+	case IP_VERSION(9, 2, 2):
+	case IP_VERSION(9, 3, 0):
+	case IP_VERSION(9, 4, 0):
+	case IP_VERSION(9, 4, 1):
+	case IP_VERSION(9, 4, 2):
+	case IP_VERSION(9, 4, 3):
+	case IP_VERSION(9, 4, 4):
+	case IP_VERSION(9, 5, 0):
+		amdgpu_device_ip_block_add(adev, &vega10_common_ip_block);
+		break;
+	case IP_VERSION(10, 1, 10):
+	case IP_VERSION(10, 1, 1):
+	case IP_VERSION(10, 1, 2):
+	case IP_VERSION(10, 1, 3):
+	case IP_VERSION(10, 1, 4):
+	case IP_VERSION(10, 3, 0):
+	case IP_VERSION(10, 3, 1):
+	case IP_VERSION(10, 3, 2):
+	case IP_VERSION(10, 3, 3):
+	case IP_VERSION(10, 3, 4):
+	case IP_VERSION(10, 3, 5):
+	case IP_VERSION(10, 3, 6):
+	case IP_VERSION(10, 3, 7):
+		amdgpu_device_ip_block_add(adev, &nv_common_ip_block);
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 1):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 5, 3):
+		amdgpu_device_ip_block_add(adev, &soc21_common_ip_block);
+		break;
+	case IP_VERSION(12, 0, 0):
+	case IP_VERSION(12, 0, 1):
+		amdgpu_device_ip_block_add(adev, &soc24_common_ip_block);
+		break;
+	default:
+		dev_err(adev->dev,
+			"Failed to add common ip block(GC_HWIP:0x%x)\n",
+			amdgpu_ip_version(adev, GC_HWIP, 0));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int amdgpu_discovery_set_gmc_ip_blocks(struct amdgpu_device *adev)
+{
+	/* use GC or MMHUB IP version */
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	case IP_VERSION(9, 0, 1):
+	case IP_VERSION(9, 1, 0):
+	case IP_VERSION(9, 2, 1):
+	case IP_VERSION(9, 2, 2):
+	case IP_VERSION(9, 3, 0):
+	case IP_VERSION(9, 4, 0):
+	case IP_VERSION(9, 4, 1):
+	case IP_VERSION(9, 4, 2):
+	case IP_VERSION(9, 4, 3):
+	case IP_VERSION(9, 4, 4):
+	case IP_VERSION(9, 5, 0):
+		amdgpu_device_ip_block_add(adev, &gmc_v9_0_ip_block);
+		break;
+	case IP_VERSION(10, 1, 10):
+	case IP_VERSION(10, 1, 1):
+	case IP_VERSION(10, 1, 2):
+	case IP_VERSION(10, 1, 3):
+	case IP_VERSION(10, 1, 4):
+	case IP_VERSION(10, 3, 0):
+	case IP_VERSION(10, 3, 1):
+	case IP_VERSION(10, 3, 2):
+	case IP_VERSION(10, 3, 3):
+	case IP_VERSION(10, 3, 4):
+	case IP_VERSION(10, 3, 5):
+	case IP_VERSION(10, 3, 6):
+	case IP_VERSION(10, 3, 7):
+		amdgpu_device_ip_block_add(adev, &gmc_v10_0_ip_block);
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 1):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 5, 3):
+		amdgpu_device_ip_block_add(adev, &gmc_v11_0_ip_block);
+		break;
+	case IP_VERSION(12, 0, 0):
+	case IP_VERSION(12, 0, 1):
+		amdgpu_device_ip_block_add(adev, &gmc_v12_0_ip_block);
+		break;
+	default:
+		dev_err(adev->dev, "Failed to add gmc ip block(GC_HWIP:0x%x)\n",
+			amdgpu_ip_version(adev, GC_HWIP, 0));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int amdgpu_discovery_set_ih_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, OSSSYS_HWIP, 0)) {
+	case IP_VERSION(4, 0, 0):
+	case IP_VERSION(4, 0, 1):
+	case IP_VERSION(4, 1, 0):
+	case IP_VERSION(4, 1, 1):
+	case IP_VERSION(4, 3, 0):
+		amdgpu_device_ip_block_add(adev, &vega10_ih_ip_block);
+		break;
+	case IP_VERSION(4, 2, 0):
+	case IP_VERSION(4, 2, 1):
+	case IP_VERSION(4, 4, 0):
+	case IP_VERSION(4, 4, 2):
+	case IP_VERSION(4, 4, 5):
+		amdgpu_device_ip_block_add(adev, &vega20_ih_ip_block);
+		break;
+	case IP_VERSION(5, 0, 0):
+	case IP_VERSION(5, 0, 1):
+	case IP_VERSION(5, 0, 2):
+	case IP_VERSION(5, 0, 3):
+	case IP_VERSION(5, 2, 0):
+	case IP_VERSION(5, 2, 1):
+		amdgpu_device_ip_block_add(adev, &navi10_ih_ip_block);
+		break;
+	case IP_VERSION(6, 0, 0):
+	case IP_VERSION(6, 0, 1):
+	case IP_VERSION(6, 0, 2):
+		amdgpu_device_ip_block_add(adev, &ih_v6_0_ip_block);
+		break;
+	case IP_VERSION(6, 1, 0):
+		amdgpu_device_ip_block_add(adev, &ih_v6_1_ip_block);
+		break;
+	case IP_VERSION(7, 0, 0):
+		amdgpu_device_ip_block_add(adev, &ih_v7_0_ip_block);
+		break;
+	default:
+		dev_err(adev->dev,
+			"Failed to add ih ip block(OSSSYS_HWIP:0x%x)\n",
+			amdgpu_ip_version(adev, OSSSYS_HWIP, 0));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int amdgpu_discovery_set_psp_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
+	case IP_VERSION(9, 0, 0):
+		amdgpu_device_ip_block_add(adev, &psp_v3_1_ip_block);
+		break;
+	case IP_VERSION(10, 0, 0):
+	case IP_VERSION(10, 0, 1):
+		amdgpu_device_ip_block_add(adev, &psp_v10_0_ip_block);
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 0, 5):
+	case IP_VERSION(11, 0, 9):
+	case IP_VERSION(11, 0, 7):
+	case IP_VERSION(11, 0, 11):
+	case IP_VERSION(11, 0, 12):
+	case IP_VERSION(11, 0, 13):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 2):
+		amdgpu_device_ip_block_add(adev, &psp_v11_0_ip_block);
+		break;
+	case IP_VERSION(11, 0, 8):
+		amdgpu_device_ip_block_add(adev, &psp_v11_0_8_ip_block);
+		break;
+	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(12, 0, 1):
+		amdgpu_device_ip_block_add(adev, &psp_v12_0_ip_block);
+		break;
+	case IP_VERSION(13, 0, 0):
+	case IP_VERSION(13, 0, 1):
+	case IP_VERSION(13, 0, 2):
+	case IP_VERSION(13, 0, 3):
+	case IP_VERSION(13, 0, 5):
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 7):
+	case IP_VERSION(13, 0, 8):
+	case IP_VERSION(13, 0, 10):
+	case IP_VERSION(13, 0, 11):
+	case IP_VERSION(13, 0, 12):
+	case IP_VERSION(13, 0, 14):
+	case IP_VERSION(14, 0, 0):
+	case IP_VERSION(14, 0, 1):
+	case IP_VERSION(14, 0, 4):
+		amdgpu_device_ip_block_add(adev, &psp_v13_0_ip_block);
+		break;
+	case IP_VERSION(13, 0, 4):
+		amdgpu_device_ip_block_add(adev, &psp_v13_0_4_ip_block);
+		break;
+	case IP_VERSION(14, 0, 2):
+	case IP_VERSION(14, 0, 3):
+	case IP_VERSION(14, 0, 5):
+		amdgpu_device_ip_block_add(adev, &psp_v14_0_ip_block);
+		break;
+	default:
+		dev_err(adev->dev,
+			"Failed to add psp ip block(MP0_HWIP:0x%x)\n",
+			amdgpu_ip_version(adev, MP0_HWIP, 0));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int amdgpu_discovery_set_smu_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
+	case IP_VERSION(9, 0, 0):
+	case IP_VERSION(10, 0, 0):
+	case IP_VERSION(10, 0, 1):
+	case IP_VERSION(11, 0, 2):
+		if (adev->asic_type == CHIP_ARCTURUS)
+			amdgpu_device_ip_block_add(adev, &smu_v11_0_ip_block);
+		else
+			amdgpu_device_ip_block_add(adev, &pp_smu_ip_block);
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 5):
+	case IP_VERSION(11, 0, 9):
+	case IP_VERSION(11, 0, 7):
+	case IP_VERSION(11, 0, 11):
+	case IP_VERSION(11, 0, 12):
+	case IP_VERSION(11, 0, 13):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 2):
+		amdgpu_device_ip_block_add(adev, &smu_v11_0_ip_block);
+		break;
+	case IP_VERSION(11, 0, 8):
+		if (adev->apu_flags & AMD_APU_IS_CYAN_SKILLFISH2)
+			amdgpu_device_ip_block_add(adev, &smu_v11_0_ip_block);
+		break;
+	case IP_VERSION(12, 0, 0):
+	case IP_VERSION(12, 0, 1):
+		amdgpu_device_ip_block_add(adev, &smu_v12_0_ip_block);
+		break;
+	case IP_VERSION(13, 0, 0):
+	case IP_VERSION(13, 0, 1):
+	case IP_VERSION(13, 0, 2):
+	case IP_VERSION(13, 0, 3):
+	case IP_VERSION(13, 0, 4):
+	case IP_VERSION(13, 0, 5):
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 7):
+	case IP_VERSION(13, 0, 8):
+	case IP_VERSION(13, 0, 10):
+	case IP_VERSION(13, 0, 11):
+	case IP_VERSION(13, 0, 14):
+	case IP_VERSION(13, 0, 12):
+		amdgpu_device_ip_block_add(adev, &smu_v13_0_ip_block);
+		break;
+	case IP_VERSION(14, 0, 0):
+	case IP_VERSION(14, 0, 1):
+	case IP_VERSION(14, 0, 2):
+	case IP_VERSION(14, 0, 3):
+	case IP_VERSION(14, 0, 4):
+	case IP_VERSION(14, 0, 5):
+		amdgpu_device_ip_block_add(adev, &smu_v14_0_ip_block);
+		break;
+	default:
+		dev_err(adev->dev,
+			"Failed to add smu ip block(MP1_HWIP:0x%x)\n",
+			amdgpu_ip_version(adev, MP1_HWIP, 0));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+#if defined(CONFIG_DRM_AMD_DC)
+static void amdgpu_discovery_set_sriov_display(struct amdgpu_device *adev)
+{
+	amdgpu_device_set_sriov_virtual_display(adev);
+	amdgpu_device_ip_block_add(adev, &amdgpu_vkms_ip_block);
+}
+#endif
+
+static int amdgpu_discovery_set_display_ip_blocks(struct amdgpu_device *adev)
+{
+	if (adev->enable_virtual_display) {
+		amdgpu_device_ip_block_add(adev, &amdgpu_vkms_ip_block);
+		return 0;
+	}
+
+	if (!amdgpu_device_has_dc_support(adev))
+		return 0;
+
+#if defined(CONFIG_DRM_AMD_DC)
+	if (amdgpu_ip_version(adev, DCE_HWIP, 0)) {
+		switch (amdgpu_ip_version(adev, DCE_HWIP, 0)) {
+		case IP_VERSION(1, 0, 0):
+		case IP_VERSION(1, 0, 1):
+		case IP_VERSION(2, 0, 2):
+		case IP_VERSION(2, 0, 0):
+		case IP_VERSION(2, 0, 3):
+		case IP_VERSION(2, 1, 0):
+		case IP_VERSION(3, 0, 0):
+		case IP_VERSION(3, 0, 2):
+		case IP_VERSION(3, 0, 3):
+		case IP_VERSION(3, 0, 1):
+		case IP_VERSION(3, 1, 2):
+		case IP_VERSION(3, 1, 3):
+		case IP_VERSION(3, 1, 4):
+		case IP_VERSION(3, 1, 5):
+		case IP_VERSION(3, 1, 6):
+		case IP_VERSION(3, 2, 0):
+		case IP_VERSION(3, 2, 1):
+		case IP_VERSION(3, 5, 0):
+		case IP_VERSION(3, 5, 1):
+		case IP_VERSION(3, 6, 0):
+		case IP_VERSION(4, 1, 0):
+			/* TODO: Fix IP version. DC code expects version 4.0.1 */
+			if (adev->ip_versions[DCE_HWIP][0] == IP_VERSION(4, 1, 0))
+				adev->ip_versions[DCE_HWIP][0] = IP_VERSION(4, 0, 1);
+
+			if (amdgpu_sriov_vf(adev))
+				amdgpu_discovery_set_sriov_display(adev);
+			else
+				amdgpu_device_ip_block_add(adev, &dm_ip_block);
+			break;
+		default:
+			dev_err(adev->dev,
+				"Failed to add dm ip block(DCE_HWIP:0x%x)\n",
+				amdgpu_ip_version(adev, DCE_HWIP, 0));
+			return -EINVAL;
+		}
+	} else if (amdgpu_ip_version(adev, DCI_HWIP, 0)) {
+		switch (amdgpu_ip_version(adev, DCI_HWIP, 0)) {
+		case IP_VERSION(12, 0, 0):
+		case IP_VERSION(12, 0, 1):
+		case IP_VERSION(12, 1, 0):
+			if (amdgpu_sriov_vf(adev))
+				amdgpu_discovery_set_sriov_display(adev);
+			else
+				amdgpu_device_ip_block_add(adev, &dm_ip_block);
+			break;
+		default:
+			dev_err(adev->dev,
+				"Failed to add dm ip block(DCI_HWIP:0x%x)\n",
+				amdgpu_ip_version(adev, DCI_HWIP, 0));
+			return -EINVAL;
+		}
+	}
+#endif
+	return 0;
+}
+
+static int amdgpu_discovery_set_gc_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	case IP_VERSION(9, 0, 1):
+	case IP_VERSION(9, 1, 0):
+	case IP_VERSION(9, 2, 1):
+	case IP_VERSION(9, 2, 2):
+	case IP_VERSION(9, 3, 0):
+	case IP_VERSION(9, 4, 0):
+	case IP_VERSION(9, 4, 1):
+	case IP_VERSION(9, 4, 2):
+		amdgpu_device_ip_block_add(adev, &gfx_v9_0_ip_block);
+		break;
+	case IP_VERSION(9, 4, 3):
+	case IP_VERSION(9, 4, 4):
+	case IP_VERSION(9, 5, 0):
+		amdgpu_device_ip_block_add(adev, &gfx_v9_4_3_ip_block);
+		break;
+	case IP_VERSION(10, 1, 10):
+	case IP_VERSION(10, 1, 2):
+	case IP_VERSION(10, 1, 1):
+	case IP_VERSION(10, 1, 3):
+	case IP_VERSION(10, 1, 4):
+	case IP_VERSION(10, 3, 0):
+	case IP_VERSION(10, 3, 2):
+	case IP_VERSION(10, 3, 1):
+	case IP_VERSION(10, 3, 4):
+	case IP_VERSION(10, 3, 5):
+	case IP_VERSION(10, 3, 6):
+	case IP_VERSION(10, 3, 3):
+	case IP_VERSION(10, 3, 7):
+		amdgpu_device_ip_block_add(adev, &gfx_v10_0_ip_block);
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 1):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 5, 3):
+		amdgpu_device_ip_block_add(adev, &gfx_v11_0_ip_block);
+		break;
+	case IP_VERSION(12, 0, 0):
+	case IP_VERSION(12, 0, 1):
+		amdgpu_device_ip_block_add(adev, &gfx_v12_0_ip_block);
+		break;
+	default:
+		dev_err(adev->dev, "Failed to add gfx ip block(GC_HWIP:0x%x)\n",
+			amdgpu_ip_version(adev, GC_HWIP, 0));
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int amdgpu_discovery_set_sdma_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, SDMA0_HWIP, 0)) {
+	case IP_VERSION(4, 0, 0):
+	case IP_VERSION(4, 0, 1):
+	case IP_VERSION(4, 1, 0):
+	case IP_VERSION(4, 1, 1):
+	case IP_VERSION(4, 1, 2):
+	case IP_VERSION(4, 2, 0):
+	case IP_VERSION(4, 2, 2):
+	case IP_VERSION(4, 4, 0):
+		amdgpu_device_ip_block_add(adev, &sdma_v4_0_ip_block);
+		break;
+	case IP_VERSION(4, 4, 2):
+	case IP_VERSION(4, 4, 5):
+	case IP_VERSION(4, 4, 4):
+		amdgpu_device_ip_block_add(adev, &sdma_v4_4_2_ip_block);
+		break;
+	case IP_VERSION(5, 0, 0):
+	case IP_VERSION(5, 0, 1):
+	case IP_VERSION(5, 0, 2):
+	case IP_VERSION(5, 0, 5):
+		amdgpu_device_ip_block_add(adev, &sdma_v5_0_ip_block);
+		break;
+	case IP_VERSION(5, 2, 0):
+	case IP_VERSION(5, 2, 2):
+	case IP_VERSION(5, 2, 4):
+	case IP_VERSION(5, 2, 5):
+	case IP_VERSION(5, 2, 6):
+	case IP_VERSION(5, 2, 3):
+	case IP_VERSION(5, 2, 1):
+	case IP_VERSION(5, 2, 7):
+		amdgpu_device_ip_block_add(adev, &sdma_v5_2_ip_block);
+		break;
+	case IP_VERSION(6, 0, 0):
+	case IP_VERSION(6, 0, 1):
+	case IP_VERSION(6, 0, 2):
+	case IP_VERSION(6, 0, 3):
+	case IP_VERSION(6, 1, 0):
+	case IP_VERSION(6, 1, 1):
+	case IP_VERSION(6, 1, 2):
+	case IP_VERSION(6, 1, 3):
+		amdgpu_device_ip_block_add(adev, &sdma_v6_0_ip_block);
+		break;
+	case IP_VERSION(7, 0, 0):
+	case IP_VERSION(7, 0, 1):
+		amdgpu_device_ip_block_add(adev, &sdma_v7_0_ip_block);
+		break;
+	default:
+		dev_err(adev->dev,
+			"Failed to add sdma ip block(SDMA0_HWIP:0x%x)\n",
+			amdgpu_ip_version(adev, SDMA0_HWIP, 0));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int amdgpu_discovery_set_ras_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 12):
+	case IP_VERSION(13, 0, 14):
+		amdgpu_device_ip_block_add(adev, &ras_v1_0_ip_block);
+		break;
+	default:
+		break;
+	}
+	return 0;
+}
+
+static int amdgpu_discovery_set_mm_ip_blocks(struct amdgpu_device *adev)
+{
+	if (amdgpu_ip_version(adev, VCE_HWIP, 0)) {
+		switch (amdgpu_ip_version(adev, UVD_HWIP, 0)) {
+		case IP_VERSION(7, 0, 0):
+		case IP_VERSION(7, 2, 0):
+			/* UVD is not supported on vega20 SR-IOV */
+			if (!(adev->asic_type == CHIP_VEGA20 && amdgpu_sriov_vf(adev)))
+				amdgpu_device_ip_block_add(adev, &uvd_v7_0_ip_block);
+			break;
+		default:
+			dev_err(adev->dev,
+				"Failed to add uvd v7 ip block(UVD_HWIP:0x%x)\n",
+				amdgpu_ip_version(adev, UVD_HWIP, 0));
+			return -EINVAL;
+		}
+		switch (amdgpu_ip_version(adev, VCE_HWIP, 0)) {
+		case IP_VERSION(4, 0, 0):
+		case IP_VERSION(4, 1, 0):
+			/* VCE is not supported on vega20 SR-IOV */
+			if (!(adev->asic_type == CHIP_VEGA20 && amdgpu_sriov_vf(adev)))
+				amdgpu_device_ip_block_add(adev, &vce_v4_0_ip_block);
+			break;
+		default:
+			dev_err(adev->dev,
+				"Failed to add VCE v4 ip block(VCE_HWIP:0x%x)\n",
+				amdgpu_ip_version(adev, VCE_HWIP, 0));
+			return -EINVAL;
+		}
+	} else {
+		switch (amdgpu_ip_version(adev, UVD_HWIP, 0)) {
+		case IP_VERSION(1, 0, 0):
+		case IP_VERSION(1, 0, 1):
+			amdgpu_device_ip_block_add(adev, &vcn_v1_0_ip_block);
+			break;
+		case IP_VERSION(2, 0, 0):
+		case IP_VERSION(2, 0, 2):
+		case IP_VERSION(2, 2, 0):
+			amdgpu_device_ip_block_add(adev, &vcn_v2_0_ip_block);
+			if (!amdgpu_sriov_vf(adev))
+				amdgpu_device_ip_block_add(adev, &jpeg_v2_0_ip_block);
+			break;
+		case IP_VERSION(2, 0, 3):
+			break;
+		case IP_VERSION(2, 5, 0):
+			amdgpu_device_ip_block_add(adev, &vcn_v2_5_ip_block);
+			amdgpu_device_ip_block_add(adev, &jpeg_v2_5_ip_block);
+			break;
+		case IP_VERSION(2, 6, 0):
+			amdgpu_device_ip_block_add(adev, &vcn_v2_6_ip_block);
+			amdgpu_device_ip_block_add(adev, &jpeg_v2_6_ip_block);
+			break;
+		case IP_VERSION(3, 0, 0):
+		case IP_VERSION(3, 0, 16):
+		case IP_VERSION(3, 1, 1):
+		case IP_VERSION(3, 1, 2):
+		case IP_VERSION(3, 0, 2):
+			amdgpu_device_ip_block_add(adev, &vcn_v3_0_ip_block);
+			if (!amdgpu_sriov_vf(adev))
+				amdgpu_device_ip_block_add(adev, &jpeg_v3_0_ip_block);
+			break;
+		case IP_VERSION(3, 0, 33):
+			amdgpu_device_ip_block_add(adev, &vcn_v3_0_ip_block);
+			break;
+		case IP_VERSION(4, 0, 0):
+		case IP_VERSION(4, 0, 2):
+		case IP_VERSION(4, 0, 4):
+			amdgpu_device_ip_block_add(adev, &vcn_v4_0_ip_block);
+			amdgpu_device_ip_block_add(adev, &jpeg_v4_0_ip_block);
+			break;
+		case IP_VERSION(4, 0, 3):
+			amdgpu_device_ip_block_add(adev, &vcn_v4_0_3_ip_block);
+			amdgpu_device_ip_block_add(adev, &jpeg_v4_0_3_ip_block);
+			break;
+		case IP_VERSION(4, 0, 5):
+		case IP_VERSION(4, 0, 6):
+			amdgpu_device_ip_block_add(adev, &vcn_v4_0_5_ip_block);
+			amdgpu_device_ip_block_add(adev, &jpeg_v4_0_5_ip_block);
+			break;
+		case IP_VERSION(5, 0, 0):
+			amdgpu_device_ip_block_add(adev, &vcn_v5_0_0_ip_block);
+			amdgpu_device_ip_block_add(adev, &jpeg_v5_0_0_ip_block);
+			break;
+		case IP_VERSION(5, 0, 1):
+			amdgpu_device_ip_block_add(adev, &vcn_v5_0_1_ip_block);
+			amdgpu_device_ip_block_add(adev, &jpeg_v5_0_1_ip_block);
+			break;
+		default:
+			dev_err(adev->dev,
+				"Failed to add vcn/jpeg ip block(UVD_HWIP:0x%x)\n",
+				amdgpu_ip_version(adev, UVD_HWIP, 0));
+			return -EINVAL;
+		}
+	}
+	return 0;
+}
+
+static int amdgpu_discovery_set_mes_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 1):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 5, 3):
+		amdgpu_device_ip_block_add(adev, &mes_v11_0_ip_block);
+		adev->enable_mes = true;
+		adev->enable_mes_kiq = true;
+		break;
+	case IP_VERSION(12, 0, 0):
+	case IP_VERSION(12, 0, 1):
+		amdgpu_device_ip_block_add(adev, &mes_v12_0_ip_block);
+		adev->enable_mes = true;
+		adev->enable_mes_kiq = true;
+		if (amdgpu_uni_mes)
+			adev->enable_uni_mes = true;
+		break;
+	default:
+		break;
+	}
+	return 0;
+}
+
+static void amdgpu_discovery_init_soc_config(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	case IP_VERSION(9, 4, 3):
+	case IP_VERSION(9, 4, 4):
+	case IP_VERSION(9, 5, 0):
+		aqua_vanjaram_init_soc_config(adev);
+		break;
+	default:
+		break;
+	}
+}
+
+static int amdgpu_discovery_set_vpe_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, VPE_HWIP, 0)) {
+	case IP_VERSION(6, 1, 0):
+	case IP_VERSION(6, 1, 1):
+	case IP_VERSION(6, 1, 3):
+		amdgpu_device_ip_block_add(adev, &vpe_v6_1_ip_block);
+		break;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+static int amdgpu_discovery_set_umsch_mm_ip_blocks(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
+	case IP_VERSION(4, 0, 5):
+	case IP_VERSION(4, 0, 6):
+		if (amdgpu_umsch_mm & 0x1) {
+			amdgpu_device_ip_block_add(adev, &umsch_mm_v4_0_ip_block);
+			adev->enable_umsch_mm = true;
+		}
+		break;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+static int amdgpu_discovery_set_isp_ip_blocks(struct amdgpu_device *adev)
+{
+#if defined(CONFIG_DRM_AMD_ISP)
+	switch (amdgpu_ip_version(adev, ISP_HWIP, 0)) {
+	case IP_VERSION(4, 1, 0):
+		amdgpu_device_ip_block_add(adev, &isp_v4_1_0_ip_block);
+		break;
+	case IP_VERSION(4, 1, 1):
+		amdgpu_device_ip_block_add(adev, &isp_v4_1_1_ip_block);
+		break;
+	default:
+		break;
+	}
+#endif
+
+	return 0;
+}
+
+int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
+{
+	int r;
+
+	switch (adev->asic_type) {
+	case CHIP_VEGA10:
+		/* This is not fatal.  We only need the discovery
+		 * binary for sysfs.  We don't need it for a
+		 * functional system.
+		 */
+		amdgpu_discovery_init(adev);
+		vega10_reg_base_init(adev);
+		adev->sdma.num_instances = 2;
+		adev->sdma.sdma_mask = 3;
+		adev->gmc.num_umc = 4;
+		adev->gfx.xcc_mask = 1;
+		adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 0, 0);
+		adev->ip_versions[HDP_HWIP][0] = IP_VERSION(4, 0, 0);
+		adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(4, 0, 0);
+		adev->ip_versions[SDMA1_HWIP][0] = IP_VERSION(4, 0, 0);
+		adev->ip_versions[DF_HWIP][0] = IP_VERSION(2, 1, 0);
+		adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(6, 1, 0);
+		adev->ip_versions[UMC_HWIP][0] = IP_VERSION(6, 0, 0);
+		adev->ip_versions[MP0_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[MP1_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[THM_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[GC_HWIP][0] = IP_VERSION(9, 0, 1);
+		adev->ip_versions[UVD_HWIP][0] = IP_VERSION(7, 0, 0);
+		adev->ip_versions[VCE_HWIP][0] = IP_VERSION(4, 0, 0);
+		adev->ip_versions[DCI_HWIP][0] = IP_VERSION(12, 0, 0);
+		break;
+	case CHIP_VEGA12:
+		/* This is not fatal.  We only need the discovery
+		 * binary for sysfs.  We don't need it for a
+		 * functional system.
+		 */
+		amdgpu_discovery_init(adev);
+		vega10_reg_base_init(adev);
+		adev->sdma.num_instances = 2;
+		adev->sdma.sdma_mask = 3;
+		adev->gmc.num_umc = 4;
+		adev->gfx.xcc_mask = 1;
+		adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 3, 0);
+		adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 3, 0);
+		adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 0, 1);
+		adev->ip_versions[HDP_HWIP][0] = IP_VERSION(4, 0, 1);
+		adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(4, 0, 1);
+		adev->ip_versions[SDMA1_HWIP][0] = IP_VERSION(4, 0, 1);
+		adev->ip_versions[DF_HWIP][0] = IP_VERSION(2, 5, 0);
+		adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(6, 2, 0);
+		adev->ip_versions[UMC_HWIP][0] = IP_VERSION(6, 1, 0);
+		adev->ip_versions[MP0_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[MP1_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[THM_HWIP][0] = IP_VERSION(9, 0, 0);
+		adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(9, 0, 1);
+		adev->ip_versions[GC_HWIP][0] = IP_VERSION(9, 2, 1);
+		adev->ip_versions[UVD_HWIP][0] = IP_VERSION(7, 0, 0);
+		adev->ip_versions[VCE_HWIP][0] = IP_VERSION(4, 0, 0);
+		adev->ip_versions[DCI_HWIP][0] = IP_VERSION(12, 0, 1);
+		break;
+	case CHIP_RAVEN:
+		/* This is not fatal.  We only need the discovery
+		 * binary for sysfs.  We don't need it for a
+		 * functional system.
+		 */
+		amdgpu_discovery_init(adev);
+		vega10_reg_base_init(adev);
+		adev->sdma.num_instances = 1;
+		adev->sdma.sdma_mask = 1;
+		adev->vcn.num_vcn_inst = 1;
+		adev->gmc.num_umc = 2;
+		adev->gfx.xcc_mask = 1;
+		if (adev->apu_flags & AMD_APU_IS_RAVEN2) {
+			adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 2, 0);
+			adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 2, 0);
+			adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 1, 1);
+			adev->ip_versions[HDP_HWIP][0] = IP_VERSION(4, 1, 1);
+			adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(4, 1, 1);
+			adev->ip_versions[DF_HWIP][0] = IP_VERSION(2, 1, 1);
+			adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(7, 0, 1);
+			adev->ip_versions[UMC_HWIP][0] = IP_VERSION(7, 5, 0);
+			adev->ip_versions[MP0_HWIP][0] = IP_VERSION(10, 0, 1);
+			adev->ip_versions[MP1_HWIP][0] = IP_VERSION(10, 0, 1);
+			adev->ip_versions[THM_HWIP][0] = IP_VERSION(10, 1, 0);
+			adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(10, 0, 1);
+			adev->ip_versions[GC_HWIP][0] = IP_VERSION(9, 2, 2);
+			adev->ip_versions[UVD_HWIP][0] = IP_VERSION(1, 0, 1);
+			adev->ip_versions[DCE_HWIP][0] = IP_VERSION(1, 0, 1);
+			adev->ip_versions[ISP_HWIP][0] = IP_VERSION(2, 0, 0);
+		} else {
+			adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 1, 0);
+			adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 1, 0);
+			adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 1, 0);
+			adev->ip_versions[HDP_HWIP][0] = IP_VERSION(4, 1, 0);
+			adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(4, 1, 0);
+			adev->ip_versions[DF_HWIP][0] = IP_VERSION(2, 1, 0);
+			adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(7, 0, 0);
+			adev->ip_versions[UMC_HWIP][0] = IP_VERSION(7, 0, 0);
+			adev->ip_versions[MP0_HWIP][0] = IP_VERSION(10, 0, 0);
+			adev->ip_versions[MP1_HWIP][0] = IP_VERSION(10, 0, 0);
+			adev->ip_versions[THM_HWIP][0] = IP_VERSION(10, 0, 0);
+			adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(10, 0, 0);
+			adev->ip_versions[GC_HWIP][0] = IP_VERSION(9, 1, 0);
+			adev->ip_versions[UVD_HWIP][0] = IP_VERSION(1, 0, 0);
+			adev->ip_versions[DCE_HWIP][0] = IP_VERSION(1, 0, 0);
+			adev->ip_versions[ISP_HWIP][0] = IP_VERSION(2, 0, 0);
+		}
+		break;
+	case CHIP_VEGA20:
+		/* This is not fatal.  We only need the discovery
+		 * binary for sysfs.  We don't need it for a
+		 * functional system.
+		 */
+		amdgpu_discovery_init(adev);
+		vega20_reg_base_init(adev);
+		adev->sdma.num_instances = 2;
+		adev->sdma.sdma_mask = 3;
+		adev->gmc.num_umc = 8;
+		adev->gfx.xcc_mask = 1;
+		adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 4, 0);
+		adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 4, 0);
+		adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 2, 0);
+		adev->ip_versions[HDP_HWIP][0] = IP_VERSION(4, 2, 0);
+		adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(4, 2, 0);
+		adev->ip_versions[SDMA1_HWIP][0] = IP_VERSION(4, 2, 0);
+		adev->ip_versions[DF_HWIP][0] = IP_VERSION(3, 6, 0);
+		adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(7, 4, 0);
+		adev->ip_versions[UMC_HWIP][0] = IP_VERSION(6, 1, 1);
+		adev->ip_versions[MP0_HWIP][0] = IP_VERSION(11, 0, 2);
+		adev->ip_versions[MP1_HWIP][0] = IP_VERSION(11, 0, 2);
+		adev->ip_versions[THM_HWIP][0] = IP_VERSION(11, 0, 2);
+		adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(11, 0, 2);
+		adev->ip_versions[GC_HWIP][0] = IP_VERSION(9, 4, 0);
+		adev->ip_versions[UVD_HWIP][0] = IP_VERSION(7, 2, 0);
+		adev->ip_versions[UVD_HWIP][1] = IP_VERSION(7, 2, 0);
+		adev->ip_versions[VCE_HWIP][0] = IP_VERSION(4, 1, 0);
+		adev->ip_versions[DCI_HWIP][0] = IP_VERSION(12, 1, 0);
+		break;
+	case CHIP_ARCTURUS:
+		/* This is not fatal.  We only need the discovery
+		 * binary for sysfs.  We don't need it for a
+		 * functional system.
+		 */
+		amdgpu_discovery_init(adev);
+		arct_reg_base_init(adev);
+		adev->sdma.num_instances = 8;
+		adev->sdma.sdma_mask = 0xff;
+		adev->vcn.num_vcn_inst = 2;
+		adev->gmc.num_umc = 8;
+		adev->gfx.xcc_mask = 1;
+		adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 4, 1);
+		adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 4, 1);
+		adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 2, 1);
+		adev->ip_versions[HDP_HWIP][0] = IP_VERSION(4, 2, 1);
+		adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[SDMA1_HWIP][0] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[SDMA1_HWIP][1] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[SDMA1_HWIP][2] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[SDMA1_HWIP][3] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[SDMA1_HWIP][4] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[SDMA1_HWIP][5] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[SDMA1_HWIP][6] = IP_VERSION(4, 2, 2);
+		adev->ip_versions[DF_HWIP][0] = IP_VERSION(3, 6, 1);
+		adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(7, 4, 1);
+		adev->ip_versions[UMC_HWIP][0] = IP_VERSION(6, 1, 2);
+		adev->ip_versions[MP0_HWIP][0] = IP_VERSION(11, 0, 4);
+		adev->ip_versions[MP1_HWIP][0] = IP_VERSION(11, 0, 2);
+		adev->ip_versions[THM_HWIP][0] = IP_VERSION(11, 0, 3);
+		adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(11, 0, 3);
+		adev->ip_versions[GC_HWIP][0] = IP_VERSION(9, 4, 1);
+		adev->ip_versions[UVD_HWIP][0] = IP_VERSION(2, 5, 0);
+		adev->ip_versions[UVD_HWIP][1] = IP_VERSION(2, 5, 0);
+		break;
+	case CHIP_ALDEBARAN:
+		/* This is not fatal.  We only need the discovery
+		 * binary for sysfs.  We don't need it for a
+		 * functional system.
+		 */
+		amdgpu_discovery_init(adev);
+		aldebaran_reg_base_init(adev);
+		adev->sdma.num_instances = 5;
+		adev->sdma.sdma_mask = 0x1f;
+		adev->vcn.num_vcn_inst = 2;
+		adev->gmc.num_umc = 4;
+		adev->gfx.xcc_mask = 1;
+		adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 4, 2);
+		adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 4, 2);
+		adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 4, 0);
+		adev->ip_versions[HDP_HWIP][0] = IP_VERSION(4, 4, 0);
+		adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(4, 4, 0);
+		adev->ip_versions[SDMA0_HWIP][1] = IP_VERSION(4, 4, 0);
+		adev->ip_versions[SDMA0_HWIP][2] = IP_VERSION(4, 4, 0);
+		adev->ip_versions[SDMA0_HWIP][3] = IP_VERSION(4, 4, 0);
+		adev->ip_versions[SDMA0_HWIP][4] = IP_VERSION(4, 4, 0);
+		adev->ip_versions[DF_HWIP][0] = IP_VERSION(3, 6, 2);
+		adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(7, 4, 4);
+		adev->ip_versions[UMC_HWIP][0] = IP_VERSION(6, 7, 0);
+		adev->ip_versions[MP0_HWIP][0] = IP_VERSION(13, 0, 2);
+		adev->ip_versions[MP1_HWIP][0] = IP_VERSION(13, 0, 2);
+		adev->ip_versions[THM_HWIP][0] = IP_VERSION(13, 0, 2);
+		adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(13, 0, 2);
+		adev->ip_versions[GC_HWIP][0] = IP_VERSION(9, 4, 2);
+		adev->ip_versions[UVD_HWIP][0] = IP_VERSION(2, 6, 0);
+		adev->ip_versions[UVD_HWIP][1] = IP_VERSION(2, 6, 0);
+		adev->ip_versions[XGMI_HWIP][0] = IP_VERSION(6, 1, 0);
+		break;
+	case CHIP_CYAN_SKILLFISH:
+		if (adev->apu_flags & AMD_APU_IS_CYAN_SKILLFISH2) {
+			r = amdgpu_discovery_reg_base_init(adev);
+			if (r)
+				return -EINVAL;
+
+			amdgpu_discovery_harvest_ip(adev);
+			amdgpu_discovery_get_gfx_info(adev);
+			amdgpu_discovery_get_mall_info(adev);
+			amdgpu_discovery_get_vcn_info(adev);
+		} else {
+			cyan_skillfish_reg_base_init(adev);
+			adev->sdma.num_instances = 2;
+			adev->sdma.sdma_mask = 3;
+			adev->gfx.xcc_mask = 1;
+			adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(2, 0, 3);
+			adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(2, 0, 3);
+			adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(5, 0, 1);
+			adev->ip_versions[HDP_HWIP][0] = IP_VERSION(5, 0, 1);
+			adev->ip_versions[SDMA0_HWIP][0] = IP_VERSION(5, 0, 1);
+			adev->ip_versions[SDMA1_HWIP][1] = IP_VERSION(5, 0, 1);
+			adev->ip_versions[DF_HWIP][0] = IP_VERSION(3, 5, 0);
+			adev->ip_versions[NBIO_HWIP][0] = IP_VERSION(2, 1, 1);
+			adev->ip_versions[UMC_HWIP][0] = IP_VERSION(8, 1, 1);
+			adev->ip_versions[MP0_HWIP][0] = IP_VERSION(11, 0, 8);
+			adev->ip_versions[MP1_HWIP][0] = IP_VERSION(11, 0, 8);
+			adev->ip_versions[THM_HWIP][0] = IP_VERSION(11, 0, 1);
+			adev->ip_versions[SMUIO_HWIP][0] = IP_VERSION(11, 0, 8);
+			adev->ip_versions[GC_HWIP][0] = IP_VERSION(10, 1, 3);
+			adev->ip_versions[UVD_HWIP][0] = IP_VERSION(2, 0, 3);
+		}
+		break;
+	default:
+		r = amdgpu_discovery_reg_base_init(adev);
+		if (r) {
+			drm_err(&adev->ddev, "discovery failed: %d\n", r);
+			return r;
+		}
+
+		amdgpu_discovery_harvest_ip(adev);
+		amdgpu_discovery_get_gfx_info(adev);
+		amdgpu_discovery_get_mall_info(adev);
+		amdgpu_discovery_get_vcn_info(adev);
+		break;
+	}
+
+	amdgpu_discovery_init_soc_config(adev);
+	amdgpu_discovery_sysfs_init(adev);
+
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	case IP_VERSION(9, 0, 1):
+	case IP_VERSION(9, 2, 1):
+	case IP_VERSION(9, 4, 0):
+	case IP_VERSION(9, 4, 1):
+	case IP_VERSION(9, 4, 2):
+	case IP_VERSION(9, 4, 3):
+	case IP_VERSION(9, 4, 4):
+	case IP_VERSION(9, 5, 0):
+		adev->family = AMDGPU_FAMILY_AI;
+		break;
+	case IP_VERSION(9, 1, 0):
+	case IP_VERSION(9, 2, 2):
+	case IP_VERSION(9, 3, 0):
+		adev->family = AMDGPU_FAMILY_RV;
+		break;
+	case IP_VERSION(10, 1, 10):
+	case IP_VERSION(10, 1, 1):
+	case IP_VERSION(10, 1, 2):
+	case IP_VERSION(10, 1, 3):
+	case IP_VERSION(10, 1, 4):
+	case IP_VERSION(10, 3, 0):
+	case IP_VERSION(10, 3, 2):
+	case IP_VERSION(10, 3, 4):
+	case IP_VERSION(10, 3, 5):
+		adev->family = AMDGPU_FAMILY_NV;
+		break;
+	case IP_VERSION(10, 3, 1):
+		adev->family = AMDGPU_FAMILY_VGH;
+		adev->apu_flags |= AMD_APU_IS_VANGOGH;
+		break;
+	case IP_VERSION(10, 3, 3):
+		adev->family = AMDGPU_FAMILY_YC;
+		break;
+	case IP_VERSION(10, 3, 6):
+		adev->family = AMDGPU_FAMILY_GC_10_3_6;
+		break;
+	case IP_VERSION(10, 3, 7):
+		adev->family = AMDGPU_FAMILY_GC_10_3_7;
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 3):
+		adev->family = AMDGPU_FAMILY_GC_11_0_0;
+		break;
+	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 4):
+		adev->family = AMDGPU_FAMILY_GC_11_0_1;
+		break;
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 1):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 5, 3):
+		adev->family = AMDGPU_FAMILY_GC_11_5_0;
+		break;
+	case IP_VERSION(12, 0, 0):
+	case IP_VERSION(12, 0, 1):
+		adev->family = AMDGPU_FAMILY_GC_12_0_0;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	case IP_VERSION(9, 1, 0):
+	case IP_VERSION(9, 2, 2):
+	case IP_VERSION(9, 3, 0):
+	case IP_VERSION(10, 1, 3):
+	case IP_VERSION(10, 1, 4):
+	case IP_VERSION(10, 3, 1):
+	case IP_VERSION(10, 3, 3):
+	case IP_VERSION(10, 3, 6):
+	case IP_VERSION(10, 3, 7):
+	case IP_VERSION(11, 0, 1):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 1):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 5, 3):
+		adev->flags |= AMD_IS_APU;
+		break;
+	default:
+		break;
+	}
+
+	/* set NBIO version */
+	switch (amdgpu_ip_version(adev, NBIO_HWIP, 0)) {
+	case IP_VERSION(6, 1, 0):
+	case IP_VERSION(6, 2, 0):
+		adev->nbio.funcs = &nbio_v6_1_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v6_1_hdp_flush_reg;
+		break;
+	case IP_VERSION(7, 0, 0):
+	case IP_VERSION(7, 0, 1):
+	case IP_VERSION(2, 5, 0):
+		adev->nbio.funcs = &nbio_v7_0_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v7_0_hdp_flush_reg;
+		break;
+	case IP_VERSION(7, 4, 0):
+	case IP_VERSION(7, 4, 1):
+	case IP_VERSION(7, 4, 4):
+		adev->nbio.funcs = &nbio_v7_4_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v7_4_hdp_flush_reg;
+		break;
+	case IP_VERSION(7, 9, 0):
+	case IP_VERSION(7, 9, 1):
+		adev->nbio.funcs = &nbio_v7_9_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v7_9_hdp_flush_reg;
+		break;
+	case IP_VERSION(7, 11, 0):
+	case IP_VERSION(7, 11, 1):
+	case IP_VERSION(7, 11, 2):
+	case IP_VERSION(7, 11, 3):
+		adev->nbio.funcs = &nbio_v7_11_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v7_11_hdp_flush_reg;
+		break;
+	case IP_VERSION(7, 2, 0):
+	case IP_VERSION(7, 2, 1):
+	case IP_VERSION(7, 3, 0):
+	case IP_VERSION(7, 5, 0):
+	case IP_VERSION(7, 5, 1):
+		adev->nbio.funcs = &nbio_v7_2_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v7_2_hdp_flush_reg;
+		break;
+	case IP_VERSION(2, 1, 1):
+	case IP_VERSION(2, 3, 0):
+	case IP_VERSION(2, 3, 1):
+	case IP_VERSION(2, 3, 2):
+	case IP_VERSION(3, 3, 0):
+	case IP_VERSION(3, 3, 1):
+	case IP_VERSION(3, 3, 2):
+	case IP_VERSION(3, 3, 3):
+		adev->nbio.funcs = &nbio_v2_3_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v2_3_hdp_flush_reg;
+		break;
+	case IP_VERSION(4, 3, 0):
+	case IP_VERSION(4, 3, 1):
+		if (amdgpu_sriov_vf(adev))
+			adev->nbio.funcs = &nbio_v4_3_sriov_funcs;
+		else
+			adev->nbio.funcs = &nbio_v4_3_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v4_3_hdp_flush_reg;
+		break;
+	case IP_VERSION(7, 7, 0):
+	case IP_VERSION(7, 7, 1):
+		adev->nbio.funcs = &nbio_v7_7_funcs;
+		adev->nbio.hdp_flush_reg = &nbio_v7_7_hdp_flush_reg;
+		break;
+	case IP_VERSION(6, 3, 1):
+		adev->nbio.funcs = &nbif_v6_3_1_funcs;
+		adev->nbio.hdp_flush_reg = &nbif_v6_3_1_hdp_flush_reg;
+		break;
+	default:
+		break;
+	}
+
+	switch (amdgpu_ip_version(adev, HDP_HWIP, 0)) {
+	case IP_VERSION(4, 0, 0):
+	case IP_VERSION(4, 0, 1):
+	case IP_VERSION(4, 1, 0):
+	case IP_VERSION(4, 1, 1):
+	case IP_VERSION(4, 1, 2):
+	case IP_VERSION(4, 2, 0):
+	case IP_VERSION(4, 2, 1):
+	case IP_VERSION(4, 4, 0):
+	case IP_VERSION(4, 4, 2):
+	case IP_VERSION(4, 4, 5):
+		adev->hdp.funcs = &hdp_v4_0_funcs;
+		break;
+	case IP_VERSION(5, 0, 0):
+	case IP_VERSION(5, 0, 1):
+	case IP_VERSION(5, 0, 2):
+	case IP_VERSION(5, 0, 3):
+	case IP_VERSION(5, 0, 4):
+	case IP_VERSION(5, 2, 0):
+		adev->hdp.funcs = &hdp_v5_0_funcs;
+		break;
+	case IP_VERSION(5, 2, 1):
+		adev->hdp.funcs = &hdp_v5_2_funcs;
+		break;
+	case IP_VERSION(6, 0, 0):
+	case IP_VERSION(6, 0, 1):
+	case IP_VERSION(6, 1, 0):
+		adev->hdp.funcs = &hdp_v6_0_funcs;
+		break;
+	case IP_VERSION(7, 0, 0):
+		adev->hdp.funcs = &hdp_v7_0_funcs;
+		break;
+	default:
+		break;
+	}
+
+	switch (amdgpu_ip_version(adev, DF_HWIP, 0)) {
+	case IP_VERSION(3, 6, 0):
+	case IP_VERSION(3, 6, 1):
+	case IP_VERSION(3, 6, 2):
+		adev->df.funcs = &df_v3_6_funcs;
+		break;
+	case IP_VERSION(2, 1, 0):
+	case IP_VERSION(2, 1, 1):
+	case IP_VERSION(2, 5, 0):
+	case IP_VERSION(3, 5, 1):
+	case IP_VERSION(3, 5, 2):
+		adev->df.funcs = &df_v1_7_funcs;
+		break;
+	case IP_VERSION(4, 3, 0):
+		adev->df.funcs = &df_v4_3_funcs;
+		break;
+	case IP_VERSION(4, 6, 2):
+		adev->df.funcs = &df_v4_6_2_funcs;
+		break;
+	case IP_VERSION(4, 15, 0):
+	case IP_VERSION(4, 15, 1):
+		adev->df.funcs = &df_v4_15_funcs;
+		break;
+	default:
+		break;
+	}
+
+	switch (amdgpu_ip_version(adev, SMUIO_HWIP, 0)) {
+	case IP_VERSION(9, 0, 0):
+	case IP_VERSION(9, 0, 1):
+	case IP_VERSION(10, 0, 0):
+	case IP_VERSION(10, 0, 1):
+	case IP_VERSION(10, 0, 2):
+		adev->smuio.funcs = &smuio_v9_0_funcs;
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 0, 7):
+	case IP_VERSION(11, 0, 8):
+		adev->smuio.funcs = &smuio_v11_0_funcs;
+		break;
+	case IP_VERSION(11, 0, 6):
+	case IP_VERSION(11, 0, 10):
+	case IP_VERSION(11, 0, 11):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(13, 0, 1):
+	case IP_VERSION(13, 0, 9):
+	case IP_VERSION(13, 0, 10):
+		adev->smuio.funcs = &smuio_v11_0_6_funcs;
+		break;
+	case IP_VERSION(13, 0, 2):
+		adev->smuio.funcs = &smuio_v13_0_funcs;
+		break;
+	case IP_VERSION(13, 0, 3):
+	case IP_VERSION(13, 0, 11):
+		adev->smuio.funcs = &smuio_v13_0_3_funcs;
+		if (adev->smuio.funcs->get_pkg_type(adev) == AMDGPU_PKG_TYPE_APU) {
+			adev->flags |= AMD_IS_APU;
+		}
+		break;
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 8):
+	case IP_VERSION(14, 0, 0):
+	case IP_VERSION(14, 0, 1):
+		adev->smuio.funcs = &smuio_v13_0_6_funcs;
+		break;
+	case IP_VERSION(14, 0, 2):
+		adev->smuio.funcs = &smuio_v14_0_2_funcs;
+		break;
+	default:
+		break;
+	}
+
+	switch (amdgpu_ip_version(adev, LSDMA_HWIP, 0)) {
+	case IP_VERSION(6, 0, 0):
+	case IP_VERSION(6, 0, 1):
+	case IP_VERSION(6, 0, 2):
+	case IP_VERSION(6, 0, 3):
+		adev->lsdma.funcs = &lsdma_v6_0_funcs;
+		break;
+	case IP_VERSION(7, 0, 0):
+	case IP_VERSION(7, 0, 1):
+		adev->lsdma.funcs = &lsdma_v7_0_funcs;
+		break;
+	default:
+		break;
+	}
+
+	r = amdgpu_discovery_set_common_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_gmc_ip_blocks(adev);
+	if (r)
+		return r;
+
+	/* For SR-IOV, PSP needs to be initialized before IH */
+	if (amdgpu_sriov_vf(adev)) {
+		r = amdgpu_discovery_set_psp_ip_blocks(adev);
+		if (r)
+			return r;
+		r = amdgpu_discovery_set_ih_ip_blocks(adev);
+		if (r)
+			return r;
+	} else {
+		r = amdgpu_discovery_set_ih_ip_blocks(adev);
+		if (r)
+			return r;
+
+		if (likely(adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) {
+			r = amdgpu_discovery_set_psp_ip_blocks(adev);
+			if (r)
+				return r;
+		}
+	}
+
+	if (likely(adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) {
+		r = amdgpu_discovery_set_smu_ip_blocks(adev);
+		if (r)
+			return r;
+	}
+
+	r = amdgpu_discovery_set_display_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_gc_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_sdma_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_ras_ip_blocks(adev);
+	if (r)
+		return r;
+
+	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT &&
+	     !amdgpu_sriov_vf(adev)) ||
+	    (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO && amdgpu_dpm == 1)) {
+		r = amdgpu_discovery_set_smu_ip_blocks(adev);
+		if (r)
+			return r;
+	}
+
+	r = amdgpu_discovery_set_mm_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_mes_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_vpe_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_umsch_mm_ip_blocks(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_discovery_set_isp_ip_blocks(adev);
+	if (r)
+		return r;
+	return 0;
+}
+
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.h
new file mode 100644
index 000000000000..4ce04486cc31
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.h
@@ -0,0 +1,50 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_DISCOVERY__
+#define __AMDGPU_DISCOVERY__
+
+#include <linux/debugfs.h>
+
+#define DISCOVERY_TMR_SIZE      (10 << 10)
+#define DISCOVERY_TMR_OFFSET    (64 << 10)
+
+struct ip_discovery_top;
+
+struct amdgpu_discovery_info {
+	struct debugfs_blob_wrapper debugfs_blob;
+	struct ip_discovery_top *ip_top;
+	uint32_t size;
+	uint8_t *bin;
+	bool reserve_tmr;
+};
+
+void amdgpu_discovery_fini(struct amdgpu_device *adev);
+int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev);
+
+int amdgpu_discovery_get_nps_info(struct amdgpu_device *adev,
+				  uint32_t *nps_type,
+				  struct amdgpu_gmc_memrange **ranges,
+				  int *range_cnt, bool refresh);
+
+#endif /* __AMDGPU_DISCOVERY__ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
index cdf2ab20166a..b5d34797d606 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
@@ -23,19 +23,73 @@
  * Authors: Dave Airlie
  *          Alex Deucher
  */
-#include <drm/drmP.h>
+
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 #include "amdgpu_i2c.h"
 #include "atom.h"
 #include "amdgpu_connectors.h"
+#include "amdgpu_display.h"
+#include "soc15_common.h"
+#include "gc/gc_11_0_0_offset.h"
+#include "gc/gc_11_0_0_sh_mask.h"
+#include "bif/bif_4_1_d.h"
 #include <asm/div64.h>
 
+#include <linux/pci.h>
 #include <linux/pm_runtime.h>
 #include <drm/drm_crtc_helper.h>
+#include <drm/drm_damage_helper.h>
+#include <drm/drm_drv.h>
 #include <drm/drm_edid.h>
+#include <drm/drm_fb_helper.h>
+#include <drm/drm_gem_framebuffer_helper.h>
+#include <drm/drm_fourcc.h>
+#include <drm/drm_modeset_helper.h>
+#include <drm/drm_vblank.h>
+
+/**
+ * amdgpu_display_hotplug_work_func - work handler for display hotplug event
+ *
+ * @work: work struct pointer
+ *
+ * This is the hotplug event work handler (all ASICs).
+ * The work gets scheduled from the IRQ handler if there
+ * was a hotplug interrupt.  It walks through the connector table
+ * and calls hotplug handler for each connector. After this, it sends
+ * a DRM hotplug event to alert userspace.
+ *
+ * This design approach is required in order to defer hotplug event handling
+ * from the IRQ handler to a work handler because hotplug handler has to use
+ * mutexes which cannot be locked in an IRQ handler (since &mutex_lock may
+ * sleep).
+ */
+void amdgpu_display_hotplug_work_func(struct work_struct *work)
+{
+	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
+						  hotplug_work.work);
+	struct drm_device *dev = adev_to_drm(adev);
+	struct drm_mode_config *mode_config = &dev->mode_config;
+	struct drm_connector *connector;
+	struct drm_connector_list_iter iter;
+
+	mutex_lock(&mode_config->mutex);
+	drm_connector_list_iter_begin(dev, &iter);
+	drm_for_each_connector_iter(connector, &iter)
+		amdgpu_connector_hotplug(connector);
+	drm_connector_list_iter_end(&iter);
+	mutex_unlock(&mode_config->mutex);
+	/* Just fire off a uevent and let userspace tell us what to do */
+	drm_helper_hpd_irq_event(dev);
+}
 
-static void amdgpu_flip_callback(struct dma_fence *f, struct dma_fence_cb *cb)
+static int amdgpu_display_framebuffer_init(struct drm_device *dev,
+					   struct amdgpu_framebuffer *rfb,
+					   const struct drm_mode_fb_cmd2 *mode_cmd,
+					   struct drm_gem_object *obj);
+
+static void amdgpu_display_flip_callback(struct dma_fence *f,
+					 struct dma_fence_cb *cb)
 {
 	struct amdgpu_flip_work *work =
 		container_of(cb, struct amdgpu_flip_work, cb);
@@ -44,24 +98,25 @@ static void amdgpu_flip_callback(struct dma_fence *f, struct dma_fence_cb *cb)
 	schedule_work(&work->flip_work.work);
 }
 
-static bool amdgpu_flip_handle_fence(struct amdgpu_flip_work *work,
-				     struct dma_fence **f)
+static bool amdgpu_display_flip_handle_fence(struct amdgpu_flip_work *work,
+					     struct dma_fence **f)
 {
-	struct dma_fence *fence= *f;
+	struct dma_fence *fence = *f;
 
 	if (fence == NULL)
 		return false;
 
 	*f = NULL;
 
-	if (!dma_fence_add_callback(fence, &work->cb, amdgpu_flip_callback))
+	if (!dma_fence_add_callback(fence, &work->cb,
+				    amdgpu_display_flip_callback))
 		return true;
 
 	dma_fence_put(fence);
 	return false;
 }
 
-static void amdgpu_flip_work_func(struct work_struct *__work)
+static void amdgpu_display_flip_work_func(struct work_struct *__work)
 {
 	struct delayed_work *delayed_work =
 		container_of(__work, struct delayed_work, work);
@@ -72,27 +127,24 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
 
 	struct drm_crtc *crtc = &amdgpu_crtc->base;
 	unsigned long flags;
-	unsigned i;
+	unsigned int i;
 	int vpos, hpos;
 
-	if (amdgpu_flip_handle_fence(work, &work->excl))
-		return;
-
 	for (i = 0; i < work->shared_count; ++i)
-		if (amdgpu_flip_handle_fence(work, &work->shared[i]))
+		if (amdgpu_display_flip_handle_fence(work, &work->shared[i]))
 			return;
 
 	/* Wait until we're out of the vertical blank period before the one
 	 * targeted by the flip
 	 */
 	if (amdgpu_crtc->enabled &&
-	    (amdgpu_get_crtc_scanoutpos(adev->ddev, work->crtc_id, 0,
-					&vpos, &hpos, NULL, NULL,
-					&crtc->hwmode)
+	    (amdgpu_display_get_crtc_scanoutpos(adev_to_drm(adev), work->crtc_id, 0,
+						&vpos, &hpos, NULL, NULL,
+						&crtc->hwmode)
 	     & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_IN_VBLANK)) ==
 	    (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_IN_VBLANK) &&
 	    (int)(work->target_vblank -
-		  amdgpu_get_vblank_counter_kms(adev->ddev, amdgpu_crtc->crtc_id)) > 0) {
+		  amdgpu_get_vblank_counter_kms(crtc)) > 0) {
 		schedule_delayed_work(&work->flip_work, usecs_to_jiffies(1000));
 		return;
 	}
@@ -108,15 +160,16 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
 	spin_unlock_irqrestore(&crtc->dev->event_lock, flags);
 
 
-	DRM_DEBUG_DRIVER("crtc:%d[%p], pflip_stat:AMDGPU_FLIP_SUBMITTED, work: %p,\n",
-					 amdgpu_crtc->crtc_id, amdgpu_crtc, work);
+	drm_dbg_vbl(adev_to_drm(adev),
+		    "crtc:%d[%p], pflip_stat:AMDGPU_FLIP_SUBMITTED, work: %p,\n",
+		    amdgpu_crtc->crtc_id, amdgpu_crtc, work);
 
 }
 
 /*
  * Handle unpin events outside the interrupt handler proper.
  */
-static void amdgpu_unpin_work_func(struct work_struct *__work)
+static void amdgpu_display_unpin_work_func(struct work_struct *__work)
 {
 	struct amdgpu_flip_work *work =
 		container_of(__work, struct amdgpu_flip_work, unpin_work);
@@ -125,10 +178,7 @@ static void amdgpu_unpin_work_func(struct work_struct *__work)
 	/* unpin of the old buffer */
 	r = amdgpu_bo_reserve(work->old_abo, true);
 	if (likely(r == 0)) {
-		r = amdgpu_bo_unpin(work->old_abo);
-		if (unlikely(r != 0)) {
-			DRM_ERROR("failed to unpin buffer after flip\n");
-		}
+		amdgpu_bo_unpin(work->old_abo);
 		amdgpu_bo_unreserve(work->old_abo);
 	} else
 		DRM_ERROR("failed to reserve buffer after flip\n");
@@ -138,31 +188,28 @@ static void amdgpu_unpin_work_func(struct work_struct *__work)
 	kfree(work);
 }
 
-int amdgpu_crtc_page_flip_target(struct drm_crtc *crtc,
-				 struct drm_framebuffer *fb,
-				 struct drm_pending_vblank_event *event,
-				 uint32_t page_flip_flags, uint32_t target,
-				 struct drm_modeset_acquire_ctx *ctx)
+int amdgpu_display_crtc_page_flip_target(struct drm_crtc *crtc,
+				struct drm_framebuffer *fb,
+				struct drm_pending_vblank_event *event,
+				uint32_t page_flip_flags, uint32_t target,
+				struct drm_modeset_acquire_ctx *ctx)
 {
 	struct drm_device *dev = crtc->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
-	struct amdgpu_framebuffer *old_amdgpu_fb;
-	struct amdgpu_framebuffer *new_amdgpu_fb;
 	struct drm_gem_object *obj;
 	struct amdgpu_flip_work *work;
 	struct amdgpu_bo *new_abo;
 	unsigned long flags;
 	u64 tiling_flags;
-	u64 base;
 	int i, r;
 
-	work = kzalloc(sizeof *work, GFP_KERNEL);
+	work = kzalloc(sizeof(*work), GFP_KERNEL);
 	if (work == NULL)
 		return -ENOMEM;
 
-	INIT_DELAYED_WORK(&work->flip_work, amdgpu_flip_work_func);
-	INIT_WORK(&work->unpin_work, amdgpu_unpin_work_func);
+	INIT_DELAYED_WORK(&work->flip_work, amdgpu_display_flip_work_func);
+	INIT_WORK(&work->unpin_work, amdgpu_display_unpin_work_func);
 
 	work->event = event;
 	work->adev = adev;
@@ -170,15 +217,13 @@ int amdgpu_crtc_page_flip_target(struct drm_crtc *crtc,
 	work->async = (page_flip_flags & DRM_MODE_PAGE_FLIP_ASYNC) != 0;
 
 	/* schedule unpin of the old buffer */
-	old_amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb);
-	obj = old_amdgpu_fb->obj;
+	obj = crtc->primary->fb->obj[0];
 
 	/* take a reference to the old object */
 	work->old_abo = gem_to_amdgpu_bo(obj);
 	amdgpu_bo_ref(work->old_abo);
 
-	new_amdgpu_fb = to_amdgpu_framebuffer(fb);
-	obj = new_amdgpu_fb->obj;
+	obj = fb->obj[0];
 	new_abo = gem_to_amdgpu_bo(obj);
 
 	/* pin the new buffer */
@@ -188,15 +233,25 @@ int amdgpu_crtc_page_flip_target(struct drm_crtc *crtc,
 		goto cleanup;
 	}
 
-	r = amdgpu_bo_pin(new_abo, AMDGPU_GEM_DOMAIN_VRAM, &base);
+	if (!adev->enable_virtual_display) {
+		new_abo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+		r = amdgpu_bo_pin(new_abo,
+				  amdgpu_display_supported_domains(adev, new_abo->flags));
+		if (unlikely(r != 0)) {
+			DRM_ERROR("failed to pin new abo buffer before flip\n");
+			goto unreserve;
+		}
+	}
+
+	r = amdgpu_ttm_alloc_gart(&new_abo->tbo);
 	if (unlikely(r != 0)) {
-		DRM_ERROR("failed to pin new abo buffer before flip\n");
-		goto unreserve;
+		DRM_ERROR("%p bind failed\n", new_abo);
+		goto unpin;
 	}
 
-	r = reservation_object_get_fences_rcu(new_abo->tbo.resv, &work->excl,
-					      &work->shared_count,
-					      &work->shared);
+	r = dma_resv_get_fences(new_abo->tbo.base.resv, DMA_RESV_USAGE_WRITE,
+				&work->shared_count,
+				&work->shared);
 	if (unlikely(r != 0)) {
 		DRM_ERROR("failed to get fences for buffer\n");
 		goto unpin;
@@ -205,9 +260,10 @@ int amdgpu_crtc_page_flip_target(struct drm_crtc *crtc,
 	amdgpu_bo_get_tiling_flags(new_abo, &tiling_flags);
 	amdgpu_bo_unreserve(new_abo);
 
-	work->base = base;
-	work->target_vblank = target - drm_crtc_vblank_count(crtc) +
-		amdgpu_get_vblank_counter_kms(dev, work->crtc_id);
+	if (!adev->enable_virtual_display)
+		work->base = amdgpu_bo_gpu_offset(new_abo);
+	work->target_vblank = target - (uint32_t)drm_crtc_vblank_count(crtc) +
+		amdgpu_get_vblank_counter_kms(crtc);
 
 	/* we borrow the event spin lock for protecting flip_wrok */
 	spin_lock_irqsave(&crtc->dev->event_lock, flags);
@@ -227,7 +283,7 @@ int amdgpu_crtc_page_flip_target(struct drm_crtc *crtc,
 	/* update crtc fb */
 	crtc->primary->fb = fb;
 	spin_unlock_irqrestore(&crtc->dev->event_lock, flags);
-	amdgpu_flip_work_func(&work->flip_work.work);
+	amdgpu_display_flip_work_func(&work->flip_work.work);
 	return 0;
 
 pflip_cleanup:
@@ -236,15 +292,14 @@ pflip_cleanup:
 		goto cleanup;
 	}
 unpin:
-	if (unlikely(amdgpu_bo_unpin(new_abo) != 0)) {
-		DRM_ERROR("failed to unpin new abo in error path\n");
-	}
+	if (!adev->enable_virtual_display)
+		amdgpu_bo_unpin(new_abo);
+
 unreserve:
 	amdgpu_bo_unreserve(new_abo);
 
 cleanup:
 	amdgpu_bo_unref(&work->old_abo);
-	dma_fence_put(work->excl);
 	for (i = 0; i < work->shared_count; ++i)
 		dma_fence_put(work->shared[i]);
 	kfree(work->shared);
@@ -253,8 +308,8 @@ cleanup:
 	return r;
 }
 
-int amdgpu_crtc_set_config(struct drm_mode_set *set,
-			   struct drm_modeset_acquire_ctx *ctx)
+int amdgpu_display_crtc_set_config(struct drm_mode_set *set,
+				   struct drm_modeset_acquire_ctx *ctx)
 {
 	struct drm_device *dev;
 	struct amdgpu_device *adev;
@@ -269,7 +324,7 @@ int amdgpu_crtc_set_config(struct drm_mode_set *set,
 
 	ret = pm_runtime_get_sync(dev->dev);
 	if (ret < 0)
-		return ret;
+		goto out;
 
 	ret = drm_crtc_helper_set_config(set, ctx);
 
@@ -277,22 +332,20 @@ int amdgpu_crtc_set_config(struct drm_mode_set *set,
 		if (crtc->enabled)
 			active = true;
 
-	pm_runtime_mark_last_busy(dev->dev);
-
-	adev = dev->dev_private;
+	adev = drm_to_adev(dev);
 	/* if we have active crtcs and we don't have a power ref,
-	   take the current one */
+	 * take the current one
+	 */
 	if (active && !adev->have_disp_power_ref) {
 		adev->have_disp_power_ref = true;
 		return ret;
 	}
-	/* if we have no active crtcs, then drop the power ref
-	   we got before */
-	if (!active && adev->have_disp_power_ref) {
-		pm_runtime_put_autosuspend(dev->dev);
+	/* if we have no active crtcs, then go to
+	 * drop the power ref we got before
+	 */
+	if (!active && adev->have_disp_power_ref)
 		adev->have_disp_power_ref = false;
-	}
-
+out:
 	/* drop the power reference we got coming in here */
 	pm_runtime_put_autosuspend(dev->dev);
 	return ret;
@@ -351,17 +404,19 @@ static const char *hpd_names[6] = {
 	"HPD6",
 };
 
-void amdgpu_print_display_setup(struct drm_device *dev)
+void amdgpu_display_print_display_setup(struct drm_device *dev)
 {
 	struct drm_connector *connector;
 	struct amdgpu_connector *amdgpu_connector;
 	struct drm_encoder *encoder;
 	struct amdgpu_encoder *amdgpu_encoder;
+	struct drm_connector_list_iter iter;
 	uint32_t devices;
 	int i = 0;
 
+	drm_connector_list_iter_begin(dev, &iter);
 	DRM_INFO("AMDGPU Display Connectors\n");
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
+	drm_for_each_connector_iter(connector, &iter) {
 		amdgpu_connector = to_amdgpu_connector(connector);
 		DRM_INFO("Connector %d:\n", i);
 		DRM_INFO("  %s\n", connector->name);
@@ -425,14 +480,11 @@ void amdgpu_print_display_setup(struct drm_device *dev)
 		}
 		i++;
 	}
+	drm_connector_list_iter_end(&iter);
 }
 
-/**
- * amdgpu_ddc_probe
- *
- */
-bool amdgpu_ddc_probe(struct amdgpu_connector *amdgpu_connector,
-		       bool use_aux)
+bool amdgpu_display_ddc_probe(struct amdgpu_connector *amdgpu_connector,
+			      bool use_aux)
 {
 	u8 out = 0x0;
 	u8 buf[8];
@@ -456,11 +508,10 @@ bool amdgpu_ddc_probe(struct amdgpu_connector *amdgpu_connector,
 	if (amdgpu_connector->router.ddc_valid)
 		amdgpu_i2c_router_select_ddc_port(amdgpu_connector);
 
-	if (use_aux) {
+	if (use_aux)
 		ret = i2c_transfer(&amdgpu_connector->ddc_bus->aux.ddc, msgs, 2);
-	} else {
+	else
 		ret = i2c_transfer(&amdgpu_connector->ddc_bus->adapter, msgs, 2);
-	}
 
 	if (ret != 2)
 		/* Couldn't find an accessible DDC on this connector */
@@ -469,172 +520,957 @@ bool amdgpu_ddc_probe(struct amdgpu_connector *amdgpu_connector,
 	 * EDID header starts with:
 	 * 0x00,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0x00.
 	 * Only the first 6 bytes must be valid as
-	 * drm_edid_block_valid() can fix the last 2 bytes */
+	 * drm_edid_block_valid() can fix the last 2 bytes
+	 */
 	if (drm_edid_header_is_valid(buf) < 6) {
 		/* Couldn't find an accessible EDID on this
-		 * connector */
+		 * connector
+		 */
 		return false;
 	}
 	return true;
 }
 
-static void amdgpu_user_framebuffer_destroy(struct drm_framebuffer *fb)
+static int amdgpu_dirtyfb(struct drm_framebuffer *fb, struct drm_file *file,
+			  unsigned int flags, unsigned int color,
+			  struct drm_clip_rect *clips, unsigned int num_clips)
 {
-	struct amdgpu_framebuffer *amdgpu_fb = to_amdgpu_framebuffer(fb);
 
-	drm_gem_object_unreference_unlocked(amdgpu_fb->obj);
-	drm_framebuffer_cleanup(fb);
-	kfree(amdgpu_fb);
+	if (file)
+		return -ENOSYS;
+
+	return drm_atomic_helper_dirtyfb(fb, file, flags, color, clips,
+					 num_clips);
 }
 
-static int amdgpu_user_framebuffer_create_handle(struct drm_framebuffer *fb,
-						  struct drm_file *file_priv,
-						  unsigned int *handle)
-{
-	struct amdgpu_framebuffer *amdgpu_fb = to_amdgpu_framebuffer(fb);
+static const struct drm_framebuffer_funcs amdgpu_fb_funcs = {
+	.destroy = drm_gem_fb_destroy,
+	.create_handle = drm_gem_fb_create_handle,
+};
 
-	return drm_gem_handle_create(file_priv, amdgpu_fb->obj, handle);
+static const struct drm_framebuffer_funcs amdgpu_fb_funcs_atomic = {
+	.destroy = drm_gem_fb_destroy,
+	.create_handle = drm_gem_fb_create_handle,
+	.dirty = amdgpu_dirtyfb
+};
+
+uint32_t amdgpu_display_supported_domains(struct amdgpu_device *adev,
+					  uint64_t bo_flags)
+{
+	uint32_t domain = AMDGPU_GEM_DOMAIN_VRAM;
+
+#if defined(CONFIG_DRM_AMD_DC)
+	/*
+	 * if amdgpu_bo_support_uswc returns false it means that USWC mappings
+	 * is not supported for this board. But this mapping is required
+	 * to avoid hang caused by placement of scanout BO in GTT on certain
+	 * APUs. So force the BO placement to VRAM in case this architecture
+	 * will not allow USWC mappings.
+	 * Also, don't allow GTT domain if the BO doesn't have USWC flag set.
+	 */
+	if ((bo_flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) &&
+	    amdgpu_bo_support_uswc(bo_flags) &&
+	    adev->dc_enabled &&
+	    adev->mode_info.gpu_vm_support)
+		domain |= AMDGPU_GEM_DOMAIN_GTT;
+#endif
+
+	return domain;
 }
 
-static const struct drm_framebuffer_funcs amdgpu_fb_funcs = {
-	.destroy = amdgpu_user_framebuffer_destroy,
-	.create_handle = amdgpu_user_framebuffer_create_handle,
+static const struct drm_format_info dcc_formats[] = {
+	{ .format = DRM_FORMAT_XRGB8888, .depth = 24, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	 { .format = DRM_FORMAT_XBGR8888, .depth = 24, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	{ .format = DRM_FORMAT_ARGB8888, .depth = 32, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	   .has_alpha = true, },
+	{ .format = DRM_FORMAT_ABGR8888, .depth = 32, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_BGRA8888, .depth = 32, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_XRGB2101010, .depth = 30, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	{ .format = DRM_FORMAT_XBGR2101010, .depth = 30, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	{ .format = DRM_FORMAT_ARGB2101010, .depth = 30, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_ABGR2101010, .depth = 30, .num_planes = 2,
+	  .cpp = { 4, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_RGB565, .depth = 16, .num_planes = 2,
+	  .cpp = { 2, 0, }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
 };
 
-int
-amdgpu_framebuffer_init(struct drm_device *dev,
-			struct amdgpu_framebuffer *rfb,
-			const struct drm_mode_fb_cmd2 *mode_cmd,
-			struct drm_gem_object *obj)
+static const struct drm_format_info dcc_retile_formats[] = {
+	{ .format = DRM_FORMAT_XRGB8888, .depth = 24, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	 { .format = DRM_FORMAT_XBGR8888, .depth = 24, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	{ .format = DRM_FORMAT_ARGB8888, .depth = 32, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	   .has_alpha = true, },
+	{ .format = DRM_FORMAT_ABGR8888, .depth = 32, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_BGRA8888, .depth = 32, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_XRGB2101010, .depth = 30, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	{ .format = DRM_FORMAT_XBGR2101010, .depth = 30, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+	{ .format = DRM_FORMAT_ARGB2101010, .depth = 30, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_ABGR2101010, .depth = 30, .num_planes = 3,
+	  .cpp = { 4, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1,
+	  .has_alpha = true, },
+	{ .format = DRM_FORMAT_RGB565, .depth = 16, .num_planes = 3,
+	  .cpp = { 2, 0, 0 }, .block_w = {1, 1, 1}, .block_h = {1, 1, 1}, .hsub = 1, .vsub = 1, },
+};
+
+static const struct drm_format_info *
+lookup_format_info(const struct drm_format_info formats[],
+		  int num_formats, u32 format)
+{
+	int i;
+
+	for (i = 0; i < num_formats; i++) {
+		if (formats[i].format == format)
+			return &formats[i];
+	}
+
+	return NULL;
+}
+
+const struct drm_format_info *
+amdgpu_lookup_format_info(u32 format, uint64_t modifier)
+{
+	if (!IS_AMD_FMT_MOD(modifier))
+		return NULL;
+
+	if (AMD_FMT_MOD_GET(TILE_VERSION, modifier) < AMD_FMT_MOD_TILE_VER_GFX9 ||
+	    AMD_FMT_MOD_GET(TILE_VERSION, modifier) >= AMD_FMT_MOD_TILE_VER_GFX12)
+		return NULL;
+
+	if (AMD_FMT_MOD_GET(DCC_RETILE, modifier))
+		return lookup_format_info(dcc_retile_formats,
+					  ARRAY_SIZE(dcc_retile_formats),
+					  format);
+
+	if (AMD_FMT_MOD_GET(DCC, modifier))
+		return lookup_format_info(dcc_formats, ARRAY_SIZE(dcc_formats),
+					  format);
+
+	/* returning NULL will cause the default format structs to be used. */
+	return NULL;
+}
+
+
+/*
+ * Tries to extract the renderable DCC offset from the opaque metadata attached
+ * to the buffer.
+ */
+static int
+extract_render_dcc_offset(struct amdgpu_device *adev,
+			  struct drm_gem_object *obj,
+			  uint64_t *offset)
+{
+	struct amdgpu_bo *rbo;
+	int r = 0;
+	uint32_t metadata[10]; /* Something that fits a descriptor + header. */
+	uint32_t size;
+
+	rbo = gem_to_amdgpu_bo(obj);
+	r = amdgpu_bo_reserve(rbo, false);
+
+	if (unlikely(r)) {
+		/* Don't show error message when returning -ERESTARTSYS */
+		if (r != -ERESTARTSYS)
+			DRM_ERROR("Unable to reserve buffer: %d\n", r);
+		return r;
+	}
+
+	r = amdgpu_bo_get_metadata(rbo, metadata, sizeof(metadata), &size, NULL);
+	amdgpu_bo_unreserve(rbo);
+
+	if (r)
+		return r;
+
+	/*
+	 * The first word is the metadata version, and we need space for at least
+	 * the version + pci vendor+device id + 8 words for a descriptor.
+	 */
+	if (size < 40  || metadata[0] != 1)
+		return -EINVAL;
+
+	if (adev->family >= AMDGPU_FAMILY_NV) {
+		/* resource word 6/7 META_DATA_ADDRESS{_LO} */
+		*offset = ((u64)metadata[9] << 16u) |
+			  ((metadata[8] & 0xFF000000u) >> 16);
+	} else {
+		/* resource word 5/7 META_DATA_ADDRESS */
+		*offset = ((u64)metadata[9] << 8u) |
+			  ((u64)(metadata[7] & 0x1FE0000u) << 23);
+	}
+
+	return 0;
+}
+
+static int convert_tiling_flags_to_modifier_gfx12(struct amdgpu_framebuffer *afb)
+{
+	u64 modifier = 0;
+	int swizzle_mode = AMDGPU_TILING_GET(afb->tiling_flags, GFX12_SWIZZLE_MODE);
+
+	if (!swizzle_mode) {
+		modifier = DRM_FORMAT_MOD_LINEAR;
+	} else {
+		int max_comp_block =
+			AMDGPU_TILING_GET(afb->tiling_flags, GFX12_DCC_MAX_COMPRESSED_BLOCK);
+
+		modifier =
+			AMD_FMT_MOD |
+			AMD_FMT_MOD_SET(TILE_VERSION, AMD_FMT_MOD_TILE_VER_GFX12) |
+			AMD_FMT_MOD_SET(TILE, swizzle_mode) |
+			AMD_FMT_MOD_SET(DCC, afb->gfx12_dcc) |
+			AMD_FMT_MOD_SET(DCC_MAX_COMPRESSED_BLOCK, max_comp_block);
+	}
+
+	afb->base.modifier = modifier;
+	afb->base.flags |= DRM_MODE_FB_MODIFIERS;
+	return 0;
+}
+
+static int convert_tiling_flags_to_modifier(struct amdgpu_framebuffer *afb)
+{
+	struct amdgpu_device *adev = drm_to_adev(afb->base.dev);
+	uint64_t modifier = 0;
+	int num_pipes = 0;
+	int num_pkrs = 0;
+
+	num_pkrs = adev->gfx.config.gb_addr_config_fields.num_pkrs;
+	num_pipes = adev->gfx.config.gb_addr_config_fields.num_pipes;
+
+	if (!afb->tiling_flags || !AMDGPU_TILING_GET(afb->tiling_flags, SWIZZLE_MODE)) {
+		modifier = DRM_FORMAT_MOD_LINEAR;
+	} else {
+		int swizzle = AMDGPU_TILING_GET(afb->tiling_flags, SWIZZLE_MODE);
+		bool has_xor = swizzle >= 16;
+		int block_size_bits;
+		int version;
+		int pipe_xor_bits = 0;
+		int bank_xor_bits = 0;
+		int packers = 0;
+		int rb = 0;
+		int pipes = ilog2(num_pipes);
+		uint32_t dcc_offset = AMDGPU_TILING_GET(afb->tiling_flags, DCC_OFFSET_256B);
+
+		switch (swizzle >> 2) {
+		case 0: /* 256B */
+			block_size_bits = 8;
+			break;
+		case 1: /* 4KiB */
+		case 5: /* 4KiB _X */
+			block_size_bits = 12;
+			break;
+		case 2: /* 64KiB */
+		case 4: /* 64 KiB _T */
+		case 6: /* 64 KiB _X */
+			block_size_bits = 16;
+			break;
+		case 7: /* 256 KiB */
+			block_size_bits = 18;
+			break;
+		default:
+			/* RESERVED or VAR */
+			return -EINVAL;
+		}
+
+		if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0))
+			version = AMD_FMT_MOD_TILE_VER_GFX11;
+		else if (amdgpu_ip_version(adev, GC_HWIP, 0) >=
+			 IP_VERSION(10, 3, 0))
+			version = AMD_FMT_MOD_TILE_VER_GFX10_RBPLUS;
+		else if (amdgpu_ip_version(adev, GC_HWIP, 0) >=
+			 IP_VERSION(10, 0, 0))
+			version = AMD_FMT_MOD_TILE_VER_GFX10;
+		else
+			version = AMD_FMT_MOD_TILE_VER_GFX9;
+
+		switch (swizzle & 3) {
+		case 0: /* Z microtiling */
+			return -EINVAL;
+		case 1: /* S microtiling */
+			if (amdgpu_ip_version(adev, GC_HWIP, 0) <
+			    IP_VERSION(11, 0, 0)) {
+				if (!has_xor)
+					version = AMD_FMT_MOD_TILE_VER_GFX9;
+			}
+			break;
+		case 2:
+			if (amdgpu_ip_version(adev, GC_HWIP, 0) <
+			    IP_VERSION(11, 0, 0)) {
+				if (!has_xor && afb->base.format->cpp[0] != 4)
+					version = AMD_FMT_MOD_TILE_VER_GFX9;
+			}
+			break;
+		case 3:
+			break;
+		}
+
+		if (has_xor) {
+			if (num_pipes == num_pkrs && num_pkrs == 0) {
+				DRM_ERROR("invalid number of pipes and packers\n");
+				return -EINVAL;
+			}
+
+			switch (version) {
+			case AMD_FMT_MOD_TILE_VER_GFX11:
+				pipe_xor_bits = min(block_size_bits - 8, pipes);
+				packers = ilog2(adev->gfx.config.gb_addr_config_fields.num_pkrs);
+				break;
+			case AMD_FMT_MOD_TILE_VER_GFX10_RBPLUS:
+				pipe_xor_bits = min(block_size_bits - 8, pipes);
+				packers = min(block_size_bits - 8 - pipe_xor_bits,
+					      ilog2(adev->gfx.config.gb_addr_config_fields.num_pkrs));
+				break;
+			case AMD_FMT_MOD_TILE_VER_GFX10:
+				pipe_xor_bits = min(block_size_bits - 8, pipes);
+				break;
+			case AMD_FMT_MOD_TILE_VER_GFX9:
+				rb = ilog2(adev->gfx.config.gb_addr_config_fields.num_se) +
+				     ilog2(adev->gfx.config.gb_addr_config_fields.num_rb_per_se);
+				pipe_xor_bits = min(block_size_bits - 8, pipes +
+						    ilog2(adev->gfx.config.gb_addr_config_fields.num_se));
+				bank_xor_bits = min(block_size_bits - 8 - pipe_xor_bits,
+						    ilog2(adev->gfx.config.gb_addr_config_fields.num_banks));
+				break;
+			}
+		}
+
+		modifier = AMD_FMT_MOD |
+			   AMD_FMT_MOD_SET(TILE, AMDGPU_TILING_GET(afb->tiling_flags, SWIZZLE_MODE)) |
+			   AMD_FMT_MOD_SET(TILE_VERSION, version) |
+			   AMD_FMT_MOD_SET(PIPE_XOR_BITS, pipe_xor_bits) |
+			   AMD_FMT_MOD_SET(BANK_XOR_BITS, bank_xor_bits) |
+			   AMD_FMT_MOD_SET(PACKERS, packers);
+
+		if (dcc_offset != 0) {
+			bool dcc_i64b = AMDGPU_TILING_GET(afb->tiling_flags, DCC_INDEPENDENT_64B) != 0;
+			bool dcc_i128b = version >= AMD_FMT_MOD_TILE_VER_GFX10_RBPLUS;
+			const struct drm_format_info *format_info;
+			u64 render_dcc_offset;
+
+			/* Enable constant encode on RAVEN2 and later. */
+			bool dcc_constant_encode =
+				(adev->asic_type > CHIP_RAVEN ||
+				 (adev->asic_type == CHIP_RAVEN &&
+				  adev->external_rev_id >= 0x81)) &&
+				amdgpu_ip_version(adev, GC_HWIP, 0) <
+					IP_VERSION(11, 0, 0);
+
+			int max_cblock_size = dcc_i64b ? AMD_FMT_MOD_DCC_BLOCK_64B :
+					      dcc_i128b ? AMD_FMT_MOD_DCC_BLOCK_128B :
+					      AMD_FMT_MOD_DCC_BLOCK_256B;
+
+			modifier |= AMD_FMT_MOD_SET(DCC, 1) |
+				    AMD_FMT_MOD_SET(DCC_CONSTANT_ENCODE, dcc_constant_encode) |
+				    AMD_FMT_MOD_SET(DCC_INDEPENDENT_64B, dcc_i64b) |
+				    AMD_FMT_MOD_SET(DCC_INDEPENDENT_128B, dcc_i128b) |
+				    AMD_FMT_MOD_SET(DCC_MAX_COMPRESSED_BLOCK, max_cblock_size);
+
+			afb->base.offsets[1] = dcc_offset * 256 + afb->base.offsets[0];
+			afb->base.pitches[1] =
+				AMDGPU_TILING_GET(afb->tiling_flags, DCC_PITCH_MAX) + 1;
+
+			/*
+			 * If the userspace driver uses retiling the tiling flags do not contain
+			 * info on the renderable DCC buffer. Luckily the opaque metadata contains
+			 * the info so we can try to extract it. The kernel does not use this info
+			 * but we should convert it to a modifier plane for getfb2, so the
+			 * userspace driver that gets it doesn't have to juggle around another DCC
+			 * plane internally.
+			 */
+			if (extract_render_dcc_offset(adev, afb->base.obj[0],
+						      &render_dcc_offset) == 0 &&
+			    render_dcc_offset != 0 &&
+			    render_dcc_offset != afb->base.offsets[1] &&
+			    render_dcc_offset < UINT_MAX) {
+				uint32_t dcc_block_bits;  /* of base surface data */
+
+				modifier |= AMD_FMT_MOD_SET(DCC_RETILE, 1);
+				afb->base.offsets[2] = render_dcc_offset;
+
+				if (adev->family >= AMDGPU_FAMILY_NV) {
+					int extra_pipe = 0;
+
+					if ((amdgpu_ip_version(adev, GC_HWIP,
+							       0) >=
+					     IP_VERSION(10, 3, 0)) &&
+					    pipes == packers && pipes > 1)
+						extra_pipe = 1;
+
+					dcc_block_bits = max(20, 16 + pipes + extra_pipe);
+				} else {
+					modifier |= AMD_FMT_MOD_SET(RB, rb) |
+						    AMD_FMT_MOD_SET(PIPE, pipes);
+					dcc_block_bits = max(20, 18 + rb);
+				}
+
+				dcc_block_bits -= ilog2(afb->base.format->cpp[0]);
+				afb->base.pitches[2] = ALIGN(afb->base.width,
+							     1u << ((dcc_block_bits + 1) / 2));
+			}
+			format_info = amdgpu_lookup_format_info(afb->base.format->format,
+								modifier);
+			if (!format_info)
+				return -EINVAL;
+
+			afb->base.format = format_info;
+		}
+	}
+
+	afb->base.modifier = modifier;
+	afb->base.flags |= DRM_MODE_FB_MODIFIERS;
+	return 0;
+}
+
+/* Mirrors the is_displayable check in radeonsi's gfx6_compute_surface */
+static int check_tiling_flags_gfx6(struct amdgpu_framebuffer *afb)
 {
+	u64 micro_tile_mode;
+
+	if (AMDGPU_TILING_GET(afb->tiling_flags, ARRAY_MODE) == 1) /* LINEAR_ALIGNED */
+		return 0;
+
+	micro_tile_mode = AMDGPU_TILING_GET(afb->tiling_flags, MICRO_TILE_MODE);
+	switch (micro_tile_mode) {
+	case 0: /* DISPLAY */
+	case 3: /* RENDER */
+		return 0;
+	default:
+		drm_dbg_kms(afb->base.dev,
+			    "Micro tile mode %llu not supported for scanout\n",
+			    micro_tile_mode);
+		return -EINVAL;
+	}
+}
+
+static void get_block_dimensions(unsigned int block_log2, unsigned int cpp,
+				 unsigned int *width, unsigned int *height)
+{
+	unsigned int cpp_log2 = ilog2(cpp);
+	unsigned int pixel_log2 = block_log2 - cpp_log2;
+	unsigned int width_log2 = (pixel_log2 + 1) / 2;
+	unsigned int height_log2 = pixel_log2 - width_log2;
+
+	*width = 1 << width_log2;
+	*height = 1 << height_log2;
+}
+
+static unsigned int get_dcc_block_size(uint64_t modifier, bool rb_aligned,
+				       bool pipe_aligned)
+{
+	unsigned int ver = AMD_FMT_MOD_GET(TILE_VERSION, modifier);
+
+	switch (ver) {
+	case AMD_FMT_MOD_TILE_VER_GFX9: {
+		/*
+		 * TODO: for pipe aligned we may need to check the alignment of the
+		 * total size of the surface, which may need to be bigger than the
+		 * natural alignment due to some HW workarounds
+		 */
+		return max(10 + (rb_aligned ? (int)AMD_FMT_MOD_GET(RB, modifier) : 0), 12);
+	}
+	case AMD_FMT_MOD_TILE_VER_GFX10:
+	case AMD_FMT_MOD_TILE_VER_GFX10_RBPLUS:
+	case AMD_FMT_MOD_TILE_VER_GFX11: {
+		int pipes_log2 = AMD_FMT_MOD_GET(PIPE_XOR_BITS, modifier);
+
+		if (ver >= AMD_FMT_MOD_TILE_VER_GFX10_RBPLUS && pipes_log2 > 1 &&
+		    AMD_FMT_MOD_GET(PACKERS, modifier) == pipes_log2)
+			++pipes_log2;
+
+		return max(8 + (pipe_aligned ? pipes_log2 : 0), 12);
+	}
+	default:
+		return 0;
+	}
+}
+
+static int amdgpu_display_verify_plane(struct amdgpu_framebuffer *rfb, int plane,
+				       const struct drm_format_info *format,
+				       unsigned int block_width, unsigned int block_height,
+				       unsigned int block_size_log2)
+{
+	unsigned int width = rfb->base.width /
+		((plane && plane < format->num_planes) ? format->hsub : 1);
+	unsigned int height = rfb->base.height /
+		((plane && plane < format->num_planes) ? format->vsub : 1);
+	unsigned int cpp = plane < format->num_planes ? format->cpp[plane] : 1;
+	unsigned int block_pitch = block_width * cpp;
+	unsigned int min_pitch = ALIGN(width * cpp, block_pitch);
+	unsigned int block_size = 1 << block_size_log2;
+	uint64_t size;
+
+	if (rfb->base.pitches[plane] % block_pitch) {
+		drm_dbg_kms(rfb->base.dev,
+			    "pitch %d for plane %d is not a multiple of block pitch %d\n",
+			    rfb->base.pitches[plane], plane, block_pitch);
+		return -EINVAL;
+	}
+	if (rfb->base.pitches[plane] < min_pitch) {
+		drm_dbg_kms(rfb->base.dev,
+			    "pitch %d for plane %d is less than minimum pitch %d\n",
+			    rfb->base.pitches[plane], plane, min_pitch);
+		return -EINVAL;
+	}
+
+	/* Force at least natural alignment. */
+	if (rfb->base.offsets[plane] % block_size) {
+		drm_dbg_kms(rfb->base.dev,
+			    "offset 0x%x for plane %d is not a multiple of block pitch 0x%x\n",
+			    rfb->base.offsets[plane], plane, block_size);
+		return -EINVAL;
+	}
+
+	size = rfb->base.offsets[plane] +
+		(uint64_t)rfb->base.pitches[plane] / block_pitch *
+		block_size * DIV_ROUND_UP(height, block_height);
+
+	if (rfb->base.obj[0]->size < size) {
+		drm_dbg_kms(rfb->base.dev,
+			    "BO size 0x%zx is less than 0x%llx required for plane %d\n",
+			    rfb->base.obj[0]->size, size, plane);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+
+static int amdgpu_display_verify_sizes(struct amdgpu_framebuffer *rfb)
+{
+	const struct drm_format_info *format_info = drm_format_info(rfb->base.format->format);
+	uint64_t modifier = rfb->base.modifier;
 	int ret;
-	rfb->obj = obj;
-	drm_helper_mode_fill_fb_struct(dev, &rfb->base, mode_cmd);
-	ret = drm_framebuffer_init(dev, &rfb->base, &amdgpu_fb_funcs);
-	if (ret) {
-		rfb->obj = NULL;
+	unsigned int i, block_width, block_height, block_size_log2;
+
+	if (rfb->base.dev->mode_config.fb_modifiers_not_supported)
+		return 0;
+
+	for (i = 0; i < format_info->num_planes; ++i) {
+		if (modifier == DRM_FORMAT_MOD_LINEAR) {
+			block_width = 256 / format_info->cpp[i];
+			block_height = 1;
+			block_size_log2 = 8;
+		} else if (AMD_FMT_MOD_GET(TILE_VERSION, modifier) >= AMD_FMT_MOD_TILE_VER_GFX12) {
+			int swizzle = AMD_FMT_MOD_GET(TILE, modifier);
+
+			switch (swizzle) {
+			case AMD_FMT_MOD_TILE_GFX12_256B_2D:
+				block_size_log2 = 8;
+				break;
+			case AMD_FMT_MOD_TILE_GFX12_4K_2D:
+				block_size_log2 = 12;
+				break;
+			case AMD_FMT_MOD_TILE_GFX12_64K_2D:
+				block_size_log2 = 16;
+				break;
+			case AMD_FMT_MOD_TILE_GFX12_256K_2D:
+				block_size_log2 = 18;
+				break;
+			default:
+				drm_dbg_kms(rfb->base.dev,
+					    "Gfx12 swizzle mode with unknown block size: %d\n", swizzle);
+				return -EINVAL;
+			}
+
+			get_block_dimensions(block_size_log2, format_info->cpp[i],
+					     &block_width, &block_height);
+		} else {
+			int swizzle = AMD_FMT_MOD_GET(TILE, modifier);
+
+			switch ((swizzle & ~3) + 1) {
+			case DC_SW_256B_S:
+				block_size_log2 = 8;
+				break;
+			case DC_SW_4KB_S:
+			case DC_SW_4KB_S_X:
+				block_size_log2 = 12;
+				break;
+			case DC_SW_64KB_S:
+			case DC_SW_64KB_S_T:
+			case DC_SW_64KB_S_X:
+				block_size_log2 = 16;
+				break;
+			case DC_SW_VAR_S_X:
+				block_size_log2 = 18;
+				break;
+			default:
+				drm_dbg_kms(rfb->base.dev,
+					    "Swizzle mode with unknown block size: %d\n", swizzle);
+				return -EINVAL;
+			}
+
+			get_block_dimensions(block_size_log2, format_info->cpp[i],
+					     &block_width, &block_height);
+		}
+
+		ret = amdgpu_display_verify_plane(rfb, i, format_info,
+						  block_width, block_height, block_size_log2);
+		if (ret)
+			return ret;
+	}
+
+	if (AMD_FMT_MOD_GET(TILE_VERSION, modifier) <= AMD_FMT_MOD_TILE_VER_GFX11 &&
+	    AMD_FMT_MOD_GET(DCC, modifier)) {
+		if (AMD_FMT_MOD_GET(DCC_RETILE, modifier)) {
+			block_size_log2 = get_dcc_block_size(modifier, false, false);
+			get_block_dimensions(block_size_log2 + 8, format_info->cpp[0],
+					     &block_width, &block_height);
+			ret = amdgpu_display_verify_plane(rfb, i, format_info,
+							  block_width, block_height,
+							  block_size_log2);
+			if (ret)
+				return ret;
+
+			++i;
+			block_size_log2 = get_dcc_block_size(modifier, true, true);
+		} else {
+			bool pipe_aligned = AMD_FMT_MOD_GET(DCC_PIPE_ALIGN, modifier);
+
+			block_size_log2 = get_dcc_block_size(modifier, true, pipe_aligned);
+		}
+		get_block_dimensions(block_size_log2 + 8, format_info->cpp[0],
+				     &block_width, &block_height);
+		ret = amdgpu_display_verify_plane(rfb, i, format_info,
+						  block_width, block_height, block_size_log2);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static int amdgpu_display_get_fb_info(const struct amdgpu_framebuffer *amdgpu_fb,
+				      uint64_t *tiling_flags, bool *tmz_surface,
+				      bool *gfx12_dcc)
+{
+	struct amdgpu_bo *rbo;
+	int r;
+
+	if (!amdgpu_fb) {
+		*tiling_flags = 0;
+		*tmz_surface = false;
+		*gfx12_dcc = false;
+		return 0;
+	}
+
+	rbo = gem_to_amdgpu_bo(amdgpu_fb->base.obj[0]);
+	r = amdgpu_bo_reserve(rbo, false);
+
+	if (unlikely(r)) {
+		/* Don't show error message when returning -ERESTARTSYS */
+		if (r != -ERESTARTSYS)
+			DRM_ERROR("Unable to reserve buffer: %d\n", r);
+		return r;
+	}
+
+	amdgpu_bo_get_tiling_flags(rbo, tiling_flags);
+	*tmz_surface = amdgpu_bo_encrypted(rbo);
+	*gfx12_dcc = rbo->flags & AMDGPU_GEM_CREATE_GFX12_DCC;
+
+	amdgpu_bo_unreserve(rbo);
+
+	return r;
+}
+
+static int amdgpu_display_gem_fb_verify_and_init(struct drm_device *dev,
+						 struct amdgpu_framebuffer *rfb,
+						 struct drm_file *file_priv,
+						 const struct drm_format_info *info,
+						 const struct drm_mode_fb_cmd2 *mode_cmd,
+						 struct drm_gem_object *obj)
+{
+	int ret;
+
+	rfb->base.obj[0] = obj;
+	drm_helper_mode_fill_fb_struct(dev, &rfb->base, info, mode_cmd);
+	/* Verify that the modifier is supported. */
+	if (!drm_any_plane_has_format(dev, mode_cmd->pixel_format,
+				      mode_cmd->modifier[0])) {
+		drm_dbg_kms(dev,
+			    "unsupported pixel format %p4cc / modifier 0x%llx\n",
+			    &mode_cmd->pixel_format, mode_cmd->modifier[0]);
+
+		ret = -EINVAL;
+		goto err;
+	}
+
+	ret = amdgpu_display_framebuffer_init(dev, rfb, mode_cmd, obj);
+	if (ret)
+		goto err;
+
+	if (drm_drv_uses_atomic_modeset(dev))
+		ret = drm_framebuffer_init(dev, &rfb->base,
+					   &amdgpu_fb_funcs_atomic);
+	else
+		ret = drm_framebuffer_init(dev, &rfb->base, &amdgpu_fb_funcs);
+
+	if (ret)
+		goto err;
+
+	return 0;
+err:
+	drm_dbg_kms(dev, "Failed to verify and init gem fb: %d\n", ret);
+	rfb->base.obj[0] = NULL;
+	return ret;
+}
+
+static int amdgpu_display_framebuffer_init(struct drm_device *dev,
+					   struct amdgpu_framebuffer *rfb,
+					   const struct drm_mode_fb_cmd2 *mode_cmd,
+					   struct drm_gem_object *obj)
+{
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	int ret, i;
+
+	/*
+	 * This needs to happen before modifier conversion as that might change
+	 * the number of planes.
+	 */
+	for (i = 1; i < rfb->base.format->num_planes; ++i) {
+		if (mode_cmd->handles[i] != mode_cmd->handles[0]) {
+			drm_dbg_kms(dev, "Plane 0 and %d have different BOs: %u vs. %u\n",
+				    i, mode_cmd->handles[0], mode_cmd->handles[i]);
+			ret = -EINVAL;
+			return ret;
+		}
+	}
+
+	ret = amdgpu_display_get_fb_info(rfb, &rfb->tiling_flags, &rfb->tmz_surface,
+					 &rfb->gfx12_dcc);
+	if (ret)
 		return ret;
+
+	if (dev->mode_config.fb_modifiers_not_supported && !adev->enable_virtual_display) {
+		drm_WARN_ONCE(dev, adev->family >= AMDGPU_FAMILY_AI,
+			      "GFX9+ requires FB check based on format modifier\n");
+		ret = check_tiling_flags_gfx6(rfb);
+		if (ret)
+			return ret;
 	}
+
+	if (!dev->mode_config.fb_modifiers_not_supported &&
+	    !(rfb->base.flags & DRM_MODE_FB_MODIFIERS)) {
+		if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(12, 0, 0))
+			ret = convert_tiling_flags_to_modifier_gfx12(rfb);
+		else
+			ret = convert_tiling_flags_to_modifier(rfb);
+
+		if (ret) {
+			drm_dbg_kms(dev, "Failed to convert tiling flags 0x%llX to a modifier",
+				    rfb->tiling_flags);
+			return ret;
+		}
+	}
+
+	ret = amdgpu_display_verify_sizes(rfb);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < rfb->base.format->num_planes; ++i) {
+		drm_gem_object_get(rfb->base.obj[0]);
+		rfb->base.obj[i] = rfb->base.obj[0];
+	}
+
 	return 0;
 }
 
-static struct drm_framebuffer *
-amdgpu_user_framebuffer_create(struct drm_device *dev,
-			       struct drm_file *file_priv,
-			       const struct drm_mode_fb_cmd2 *mode_cmd)
+struct drm_framebuffer *
+amdgpu_display_user_framebuffer_create(struct drm_device *dev,
+				       struct drm_file *file_priv,
+				       const struct drm_format_info *info,
+				       const struct drm_mode_fb_cmd2 *mode_cmd)
 {
-	struct drm_gem_object *obj;
 	struct amdgpu_framebuffer *amdgpu_fb;
+	struct drm_gem_object *obj;
+	struct amdgpu_bo *bo;
+	uint32_t domains;
 	int ret;
 
 	obj = drm_gem_object_lookup(file_priv, mode_cmd->handles[0]);
 	if (obj ==  NULL) {
-		dev_err(&dev->pdev->dev, "No GEM object associated to handle 0x%08X, "
-			"can't create framebuffer\n", mode_cmd->handles[0]);
+		drm_dbg_kms(dev,
+			    "No GEM object associated to handle 0x%08X, can't create framebuffer\n",
+			    mode_cmd->handles[0]);
+
 		return ERR_PTR(-ENOENT);
 	}
 
 	/* Handle is imported dma-buf, so cannot be migrated to VRAM for scanout */
-	if (obj->import_attach) {
-		DRM_DEBUG_KMS("Cannot create framebuffer from imported dma_buf\n");
+	bo = gem_to_amdgpu_bo(obj);
+	domains = amdgpu_display_supported_domains(drm_to_adev(dev), bo->flags);
+	if (drm_gem_is_imported(obj) && !(domains & AMDGPU_GEM_DOMAIN_GTT)) {
+		drm_dbg_kms(dev, "Cannot create framebuffer from imported dma_buf\n");
+		drm_gem_object_put(obj);
 		return ERR_PTR(-EINVAL);
 	}
 
 	amdgpu_fb = kzalloc(sizeof(*amdgpu_fb), GFP_KERNEL);
 	if (amdgpu_fb == NULL) {
-		drm_gem_object_unreference_unlocked(obj);
+		drm_gem_object_put(obj);
 		return ERR_PTR(-ENOMEM);
 	}
 
-	ret = amdgpu_framebuffer_init(dev, amdgpu_fb, mode_cmd, obj);
+	ret = amdgpu_display_gem_fb_verify_and_init(dev, amdgpu_fb, file_priv,
+						    info, mode_cmd, obj);
 	if (ret) {
 		kfree(amdgpu_fb);
-		drm_gem_object_unreference_unlocked(obj);
+		drm_gem_object_put(obj);
 		return ERR_PTR(ret);
 	}
 
+	drm_gem_object_put(obj);
 	return &amdgpu_fb->base;
 }
 
-static void amdgpu_output_poll_changed(struct drm_device *dev)
-{
-	struct amdgpu_device *adev = dev->dev_private;
-	amdgpu_fb_output_poll_changed(adev);
-}
-
 const struct drm_mode_config_funcs amdgpu_mode_funcs = {
-	.fb_create = amdgpu_user_framebuffer_create,
-	.output_poll_changed = amdgpu_output_poll_changed
+	.fb_create = amdgpu_display_user_framebuffer_create,
 };
 
-static const struct drm_prop_enum_list amdgpu_underscan_enum_list[] =
-{	{ UNDERSCAN_OFF, "off" },
+static const struct drm_prop_enum_list amdgpu_underscan_enum_list[] = {
+	{ UNDERSCAN_OFF, "off" },
 	{ UNDERSCAN_ON, "on" },
 	{ UNDERSCAN_AUTO, "auto" },
 };
 
-static const struct drm_prop_enum_list amdgpu_audio_enum_list[] =
-{	{ AMDGPU_AUDIO_DISABLE, "off" },
+static const struct drm_prop_enum_list amdgpu_audio_enum_list[] = {
+	{ AMDGPU_AUDIO_DISABLE, "off" },
 	{ AMDGPU_AUDIO_ENABLE, "on" },
 	{ AMDGPU_AUDIO_AUTO, "auto" },
 };
 
 /* XXX support different dither options? spatial, temporal, both, etc. */
-static const struct drm_prop_enum_list amdgpu_dither_enum_list[] =
-{	{ AMDGPU_FMT_DITHER_DISABLE, "off" },
+static const struct drm_prop_enum_list amdgpu_dither_enum_list[] = {
+	{ AMDGPU_FMT_DITHER_DISABLE, "off" },
 	{ AMDGPU_FMT_DITHER_ENABLE, "on" },
 };
 
-int amdgpu_modeset_create_props(struct amdgpu_device *adev)
+/**
+ * DOC: property for adaptive backlight modulation
+ *
+ * The 'adaptive backlight modulation' property is used for the compositor to
+ * directly control the adaptive backlight modulation power savings feature
+ * that is part of DCN hardware.
+ *
+ * The property will be attached specifically to eDP panels that support it.
+ *
+ * The property is by default set to 'sysfs' to allow the sysfs file 'panel_power_savings'
+ * to be able to control it.
+ * If set to 'off' the compositor will ensure it stays off.
+ * The other values 'min', 'bias min', 'bias max', and 'max' will control the
+ * intensity of the power savings.
+ *
+ * Modifying this value can have implications on color accuracy, so tread
+ * carefully.
+ */
+static int amdgpu_display_setup_abm_prop(struct amdgpu_device *adev)
+{
+	const struct drm_prop_enum_list props[] = {
+		{ ABM_SYSFS_CONTROL, "sysfs" },
+		{ ABM_LEVEL_OFF, "off" },
+		{ ABM_LEVEL_MIN, "min" },
+		{ ABM_LEVEL_BIAS_MIN, "bias min" },
+		{ ABM_LEVEL_BIAS_MAX, "bias max" },
+		{ ABM_LEVEL_MAX, "max" },
+	};
+	struct drm_property *prop;
+	int i;
+
+	if (!adev->dc_enabled)
+		return 0;
+
+	prop = drm_property_create(adev_to_drm(adev), DRM_MODE_PROP_ENUM,
+				"adaptive backlight modulation",
+				6);
+	if (!prop)
+		return -ENOMEM;
+
+	for (i = 0; i < ARRAY_SIZE(props); i++) {
+		int ret;
+
+		ret = drm_property_add_enum(prop, props[i].type,
+						props[i].name);
+
+		if (ret) {
+			drm_property_destroy(adev_to_drm(adev), prop);
+
+			return ret;
+		}
+	}
+
+	adev->mode_info.abm_level_property = prop;
+
+	return 0;
+}
+
+int amdgpu_display_modeset_create_props(struct amdgpu_device *adev)
 {
 	int sz;
 
 	adev->mode_info.coherent_mode_property =
-		drm_property_create_range(adev->ddev, 0 , "coherent", 0, 1);
+		drm_property_create_range(adev_to_drm(adev), 0, "coherent", 0, 1);
 	if (!adev->mode_info.coherent_mode_property)
 		return -ENOMEM;
 
 	adev->mode_info.load_detect_property =
-		drm_property_create_range(adev->ddev, 0, "load detection", 0, 1);
+		drm_property_create_range(adev_to_drm(adev), 0, "load detection", 0, 1);
 	if (!adev->mode_info.load_detect_property)
 		return -ENOMEM;
 
-	drm_mode_create_scaling_mode_property(adev->ddev);
+	drm_mode_create_scaling_mode_property(adev_to_drm(adev));
 
 	sz = ARRAY_SIZE(amdgpu_underscan_enum_list);
 	adev->mode_info.underscan_property =
-		drm_property_create_enum(adev->ddev, 0,
-				    "underscan",
-				    amdgpu_underscan_enum_list, sz);
+		drm_property_create_enum(adev_to_drm(adev), 0,
+					 "underscan",
+					 amdgpu_underscan_enum_list, sz);
 
 	adev->mode_info.underscan_hborder_property =
-		drm_property_create_range(adev->ddev, 0,
-					"underscan hborder", 0, 128);
+		drm_property_create_range(adev_to_drm(adev), 0,
+					  "underscan hborder", 0, 128);
 	if (!adev->mode_info.underscan_hborder_property)
 		return -ENOMEM;
 
 	adev->mode_info.underscan_vborder_property =
-		drm_property_create_range(adev->ddev, 0,
-					"underscan vborder", 0, 128);
+		drm_property_create_range(adev_to_drm(adev), 0,
+					  "underscan vborder", 0, 128);
 	if (!adev->mode_info.underscan_vborder_property)
 		return -ENOMEM;
 
 	sz = ARRAY_SIZE(amdgpu_audio_enum_list);
 	adev->mode_info.audio_property =
-		drm_property_create_enum(adev->ddev, 0,
+		drm_property_create_enum(adev_to_drm(adev), 0,
 					 "audio",
 					 amdgpu_audio_enum_list, sz);
 
 	sz = ARRAY_SIZE(amdgpu_dither_enum_list);
 	adev->mode_info.dither_property =
-		drm_property_create_enum(adev->ddev, 0,
+		drm_property_create_enum(adev_to_drm(adev), 0,
 					 "dither",
 					 amdgpu_dither_enum_list, sz);
 
-	return 0;
+	return amdgpu_display_setup_abm_prop(adev);
 }
 
-void amdgpu_update_display_priority(struct amdgpu_device *adev)
+void amdgpu_display_update_priority(struct amdgpu_device *adev)
 {
 	/* adjustment options for the display watermarks */
 	if ((amdgpu_disp_priority == 0) || (amdgpu_disp_priority > 2))
@@ -644,7 +1480,7 @@ void amdgpu_update_display_priority(struct amdgpu_device *adev)
 
 }
 
-static bool is_hdtv_mode(const struct drm_display_mode *mode)
+static bool amdgpu_display_is_hdtv_mode(const struct drm_display_mode *mode)
 {
 	/* try and guess if this is a tv or a monitor */
 	if ((mode->vdisplay == 480 && mode->hdisplay == 720) || /* 480p */
@@ -656,16 +1492,15 @@ static bool is_hdtv_mode(const struct drm_display_mode *mode)
 		return false;
 }
 
-bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
-				    const struct drm_display_mode *mode,
-				    struct drm_display_mode *adjusted_mode)
+bool amdgpu_display_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
+					const struct drm_display_mode *mode,
+					struct drm_display_mode *adjusted_mode)
 {
 	struct drm_device *dev = crtc->dev;
 	struct drm_encoder *encoder;
 	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
 	struct amdgpu_encoder *amdgpu_encoder;
 	struct drm_connector *connector;
-	struct amdgpu_connector *amdgpu_connector;
 	u32 src_v = 1, dst_v = 1;
 	u32 src_h = 1, dst_h = 1;
 
@@ -677,7 +1512,6 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
 			continue;
 		amdgpu_encoder = to_amdgpu_encoder(encoder);
 		connector = amdgpu_get_connector_for_encoder(encoder);
-		amdgpu_connector = to_amdgpu_connector(connector);
 
 		/* set scaling */
 		if (amdgpu_encoder->rmx_type == RMX_OFF)
@@ -700,8 +1534,8 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
 		if ((!(mode->flags & DRM_MODE_FLAG_INTERLACE)) &&
 		    ((amdgpu_encoder->underscan_type == UNDERSCAN_ON) ||
 		     ((amdgpu_encoder->underscan_type == UNDERSCAN_AUTO) &&
-		      drm_detect_hdmi_monitor(amdgpu_connector_edid(connector)) &&
-		      is_hdtv_mode(mode)))) {
+		      connector && connector->display_info.is_hdmi &&
+		      amdgpu_display_is_hdtv_mode(mode)))) {
 			if (amdgpu_encoder->underscan_hborder != 0)
 				amdgpu_crtc->h_border = amdgpu_encoder->underscan_hborder;
 			else
@@ -719,6 +1553,7 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
 	}
 	if (amdgpu_crtc->rmx_type != RMX_OFF) {
 		fixed20_12 a, b;
+
 		a.full = dfixed_const(src_v);
 		b.full = dfixed_const(dst_v);
 		amdgpu_crtc->vsc.full = dfixed_div(a, b);
@@ -738,7 +1573,7 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
  *
  * \param dev Device to query.
  * \param pipe Crtc to query.
- * \param flags Flags from caller (DRM_CALLED_FROM_VBLIRQ or 0).
+ * \param flags from caller (DRM_CALLED_FROM_VBLIRQ or 0).
  *              For driver internal use only also supports these flags:
  *
  *              USE_REAL_VBLANKSTART to use the real start of vblank instead
@@ -769,16 +1604,16 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
  * unknown small number of scanlines wrt. real scanout position.
  *
  */
-int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
-			       unsigned int flags, int *vpos, int *hpos,
-			       ktime_t *stime, ktime_t *etime,
-			       const struct drm_display_mode *mode)
+int amdgpu_display_get_crtc_scanoutpos(struct drm_device *dev,
+			unsigned int pipe, unsigned int flags, int *vpos,
+			int *hpos, ktime_t *stime, ktime_t *etime,
+			const struct drm_display_mode *mode)
 {
 	u32 vbl = 0, position = 0;
 	int vbl_start, vbl_end, vtotal, ret = 0;
 	bool in_vbl = true;
 
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 
 	/* preempt_disable_rt() should go right here in PREEMPT_RT patchset. */
 
@@ -805,8 +1640,7 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
 		ret |= DRM_SCANOUTPOS_ACCURATE;
 		vbl_start = vbl & 0x1fff;
 		vbl_end = (vbl >> 16) & 0x1fff;
-	}
-	else {
+	} else {
 		/* No: Fake something reasonable which gives at least ok results. */
 		vbl_start = mode->crtc_vdisplay;
 		vbl_end = 0;
@@ -814,8 +1648,8 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
 
 	/* Called from driver internal vblank counter query code? */
 	if (flags & GET_DISTANCE_TO_VBLANKSTART) {
-	    /* Caller wants distance from real vbl_start in *hpos */
-	    *hpos = *vpos - vbl_start;
+		/* Caller wants distance from real vbl_start in *hpos */
+		*hpos = *vpos - vbl_start;
 	}
 
 	/* Fudge vblank to start a few scanlines earlier to handle the
@@ -837,7 +1671,7 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
 
 	/* In vblank? */
 	if (in_vbl)
-	    ret |= DRM_SCANOUTPOS_IN_VBLANK;
+		ret |= DRM_SCANOUTPOS_IN_VBLANK;
 
 	/* Called from driver internal vblank counter query code? */
 	if (flags & GET_DISTANCE_TO_VBLANKSTART) {
@@ -855,7 +1689,12 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
 	/* Inside "upper part" of vblank area? Apply corrective offset if so: */
 	if (in_vbl && (*vpos >= vbl_start)) {
 		vtotal = mode->crtc_vtotal;
-		*vpos = *vpos - vtotal;
+
+		/* With variable refresh rate displays the vpos can exceed
+		 * the vtotal value. Clamp to 0 to return -vbl_end instead
+		 * of guessing the remaining number of lines until scanout.
+		 */
+		*vpos = (*vpos < vtotal) ? (*vpos - vtotal) : 0;
 	}
 
 	/* Correct for shifted end of vbl at vbl_end. */
@@ -864,7 +1703,7 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
 	return ret;
 }
 
-int amdgpu_crtc_idx_to_irq_type(struct amdgpu_device *adev, int crtc)
+int amdgpu_display_crtc_idx_to_irq_type(struct amdgpu_device *adev, int crtc)
 {
 	if (crtc < 0 || crtc >= adev->mode_info.num_crtc)
 		return AMDGPU_CRTC_IRQ_NONE;
@@ -886,3 +1725,204 @@ int amdgpu_crtc_idx_to_irq_type(struct amdgpu_device *adev, int crtc)
 		return AMDGPU_CRTC_IRQ_NONE;
 	}
 }
+
+bool amdgpu_crtc_get_scanout_position(struct drm_crtc *crtc,
+			bool in_vblank_irq, int *vpos,
+			int *hpos, ktime_t *stime, ktime_t *etime,
+			const struct drm_display_mode *mode)
+{
+	struct drm_device *dev = crtc->dev;
+	unsigned int pipe = crtc->index;
+
+	return amdgpu_display_get_crtc_scanoutpos(dev, pipe, 0, vpos, hpos,
+						  stime, etime, mode);
+}
+
+static bool
+amdgpu_display_robj_is_fb(struct amdgpu_device *adev, struct amdgpu_bo *robj)
+{
+	struct drm_device *dev = adev_to_drm(adev);
+	struct drm_fb_helper *fb_helper = dev->fb_helper;
+
+	if (!fb_helper || !fb_helper->buffer)
+		return false;
+
+	if (gem_to_amdgpu_bo(fb_helper->buffer->gem) != robj)
+		return false;
+
+	return true;
+}
+
+int amdgpu_display_suspend_helper(struct amdgpu_device *adev)
+{
+	struct drm_device *dev = adev_to_drm(adev);
+	struct drm_crtc *crtc;
+	struct drm_connector *connector;
+	struct drm_connector_list_iter iter;
+	int r;
+
+	drm_kms_helper_poll_disable(dev);
+
+	/* turn off display hw */
+	drm_modeset_lock_all(dev);
+	drm_connector_list_iter_begin(dev, &iter);
+	drm_for_each_connector_iter(connector, &iter)
+		drm_helper_connector_dpms(connector,
+					  DRM_MODE_DPMS_OFF);
+	drm_connector_list_iter_end(&iter);
+	drm_modeset_unlock_all(dev);
+	/* unpin the front buffers and cursors */
+	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
+		struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
+		struct drm_framebuffer *fb = crtc->primary->fb;
+		struct amdgpu_bo *robj;
+
+		if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) {
+			struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
+
+			r = amdgpu_bo_reserve(aobj, true);
+			if (r == 0) {
+				amdgpu_bo_unpin(aobj);
+				amdgpu_bo_unreserve(aobj);
+			}
+		}
+
+		if (!fb || !fb->obj[0])
+			continue;
+
+		robj = gem_to_amdgpu_bo(fb->obj[0]);
+		if (!amdgpu_display_robj_is_fb(adev, robj)) {
+			r = amdgpu_bo_reserve(robj, true);
+			if (r == 0) {
+				amdgpu_bo_unpin(robj);
+				amdgpu_bo_unreserve(robj);
+			}
+		}
+	}
+	return 0;
+}
+
+int amdgpu_display_resume_helper(struct amdgpu_device *adev)
+{
+	struct drm_device *dev = adev_to_drm(adev);
+	struct drm_connector *connector;
+	struct drm_connector_list_iter iter;
+	struct drm_crtc *crtc;
+	int r;
+
+	/* pin cursors */
+	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
+		struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
+
+		if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) {
+			struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
+
+			r = amdgpu_bo_reserve(aobj, true);
+			if (r == 0) {
+				aobj->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+				r = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM);
+				if (r != 0)
+					dev_err(adev->dev, "Failed to pin cursor BO (%d)\n", r);
+				amdgpu_crtc->cursor_addr = amdgpu_bo_gpu_offset(aobj);
+				amdgpu_bo_unreserve(aobj);
+			}
+		}
+	}
+
+	drm_helper_resume_force_mode(dev);
+
+	/* turn on display hw */
+	drm_modeset_lock_all(dev);
+
+	drm_connector_list_iter_begin(dev, &iter);
+	drm_for_each_connector_iter(connector, &iter)
+		drm_helper_connector_dpms(connector,
+					  DRM_MODE_DPMS_ON);
+	drm_connector_list_iter_end(&iter);
+
+	drm_modeset_unlock_all(dev);
+
+	drm_kms_helper_poll_enable(dev);
+
+	return 0;
+}
+
+/* panic_bo is set in amdgpu_dm_plane_get_scanout_buffer() and only used in amdgpu_dm_set_pixel()
+ * they are called from the panic handler, and protected by the drm_panic spinlock.
+ */
+static struct amdgpu_bo *panic_abo;
+
+/* Use the indirect MMIO to write each pixel to the GPU VRAM,
+ * This is a simplified version of amdgpu_device_mm_access()
+ */
+static void amdgpu_display_set_pixel(struct drm_scanout_buffer *sb,
+				     unsigned int x,
+				     unsigned int y,
+				     u32 color)
+{
+	struct amdgpu_res_cursor cursor;
+	unsigned long offset;
+	struct amdgpu_bo *abo = panic_abo;
+	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
+	uint32_t tmp;
+
+	offset = x * 4 + y * sb->pitch[0];
+	amdgpu_res_first(abo->tbo.resource, offset, 4, &cursor);
+
+	tmp = cursor.start >> 31;
+	WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t) cursor.start) | 0x80000000);
+	if (tmp != 0xffffffff)
+		WREG32_NO_KIQ(mmMM_INDEX_HI, tmp);
+	WREG32_NO_KIQ(mmMM_DATA, color);
+}
+
+int amdgpu_display_get_scanout_buffer(struct drm_plane *plane,
+				      struct drm_scanout_buffer *sb)
+{
+	struct amdgpu_bo *abo;
+	struct drm_framebuffer *fb = plane->state->fb;
+
+	if (!fb)
+		return -EINVAL;
+
+	DRM_DEBUG_KMS("Framebuffer %dx%d %p4cc\n", fb->width, fb->height, &fb->format->format);
+
+	abo = gem_to_amdgpu_bo(fb->obj[0]);
+	if (!abo)
+		return -EINVAL;
+
+	sb->width = fb->width;
+	sb->height = fb->height;
+	/* Use the generic linear format, because tiling will be disabled in panic_flush() */
+	sb->format = drm_format_info(fb->format->format);
+	if (!sb->format)
+		return -EINVAL;
+
+	sb->pitch[0] = fb->pitches[0];
+
+	if (abo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) {
+		if (abo->tbo.resource->mem_type != TTM_PL_VRAM) {
+			drm_warn(plane->dev, "amdgpu panic, framebuffer not in VRAM\n");
+			return -EINVAL;
+		}
+		/* Only handle 32bits format, to simplify mmio access */
+		if (fb->format->cpp[0] != 4) {
+			drm_warn(plane->dev, "amdgpu panic, pixel format is not 32bits\n");
+			return -EINVAL;
+		}
+		sb->set_pixel = amdgpu_display_set_pixel;
+		panic_abo = abo;
+		return 0;
+	}
+	if (!abo->kmap.virtual &&
+	    ttm_bo_kmap(&abo->tbo, 0, PFN_UP(abo->tbo.base.size), &abo->kmap)) {
+		drm_warn(plane->dev, "amdgpu bo map failed, panic won't be displayed\n");
+		return -ENOMEM;
+	}
+	if (abo->kmap.bo_kmap_type & TTM_BO_MAP_IOMEM_MASK)
+		iosys_map_set_vaddr_iomem(&sb->map[0], abo->kmap.virtual);
+	else
+		iosys_map_set_vaddr(&sb->map[0], abo->kmap.virtual);
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.h
new file mode 100644
index 000000000000..49a29bf47a37
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.h
@@ -0,0 +1,65 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_DISPLAY_H__
+#define __AMDGPU_DISPLAY_H__
+
+#include <drm/drm_panic.h>
+
+#define amdgpu_display_vblank_get_counter(adev, crtc) (adev)->mode_info.funcs->vblank_get_counter((adev), (crtc))
+#define amdgpu_display_backlight_set_level(adev, e, l) (adev)->mode_info.funcs->backlight_set_level((e), (l))
+#define amdgpu_display_backlight_get_level(adev, e) (adev)->mode_info.funcs->backlight_get_level((e))
+#define amdgpu_display_hpd_sense(adev, h) (adev)->mode_info.funcs->hpd_sense((adev), (h))
+#define amdgpu_display_hpd_set_polarity(adev, h) (adev)->mode_info.funcs->hpd_set_polarity((adev), (h))
+#define amdgpu_display_hpd_get_gpio_reg(adev) (adev)->mode_info.funcs->hpd_get_gpio_reg((adev))
+#define amdgpu_display_bandwidth_update(adev) (adev)->mode_info.funcs->bandwidth_update((adev))
+#define amdgpu_display_page_flip(adev, crtc, base, async) (adev)->mode_info.funcs->page_flip((adev), (crtc), (base), (async))
+#define amdgpu_display_page_flip_get_scanoutpos(adev, crtc, vbl, pos) (adev)->mode_info.funcs->page_flip_get_scanoutpos((adev), (crtc), (vbl), (pos))
+#define amdgpu_display_add_encoder(adev, e, s, c) (adev)->mode_info.funcs->add_encoder((adev), (e), (s), (c))
+#define amdgpu_display_add_connector(adev, ci, sd, ct, ib, coi, h, r) (adev)->mode_info.funcs->add_connector((adev), (ci), (sd), (ct), (ib), (coi), (h), (r))
+
+void amdgpu_display_hotplug_work_func(struct work_struct *work);
+void amdgpu_display_update_priority(struct amdgpu_device *adev);
+uint32_t amdgpu_display_supported_domains(struct amdgpu_device *adev,
+					  uint64_t bo_flags);
+struct drm_framebuffer *
+amdgpu_display_user_framebuffer_create(struct drm_device *dev,
+				       struct drm_file *file_priv,
+				       const struct drm_format_info *info,
+				       const struct drm_mode_fb_cmd2 *mode_cmd);
+const struct drm_format_info *
+amdgpu_lookup_format_info(u32 format, uint64_t modifier);
+
+int amdgpu_display_suspend_helper(struct amdgpu_device *adev);
+int amdgpu_display_resume_helper(struct amdgpu_device *adev);
+
+int amdgpu_display_get_scanout_buffer(struct drm_plane *plane,
+				      struct drm_scanout_buffer *sb);
+
+#define ABM_SYSFS_CONTROL	-1
+#define ABM_LEVEL_OFF		0
+#define ABM_LEVEL_MIN		1
+#define ABM_LEVEL_BIAS_MIN	2
+#define ABM_LEVEL_BIAS_MAX	3
+#define ABM_LEVEL_MAX		4
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
new file mode 100644
index 000000000000..e22cfa7c6d32
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
@@ -0,0 +1,605 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * based on nouveau_prime.c
+ *
+ * Authors: Alex Deucher
+ */
+
+/**
+ * DOC: PRIME Buffer Sharing
+ *
+ * The following callback implementations are used for :ref:`sharing GEM buffer
+ * objects between different devices via PRIME <prime_buffer_sharing>`.
+ */
+
+#include "amdgpu.h"
+#include "amdgpu_display.h"
+#include "amdgpu_gem.h"
+#include "amdgpu_dma_buf.h"
+#include "amdgpu_xgmi.h"
+#include "amdgpu_vm.h"
+#include <drm/amdgpu_drm.h>
+#include <drm/ttm/ttm_tt.h>
+#include <linux/dma-buf.h>
+#include <linux/dma-fence-array.h>
+#include <linux/pci-p2pdma.h>
+
+static const struct dma_buf_attach_ops amdgpu_dma_buf_attach_ops;
+
+/**
+ * dma_buf_attach_adev - Helper to get adev of an attachment
+ *
+ * @attach: attachment
+ *
+ * Returns:
+ * A struct amdgpu_device * if the attaching device is an amdgpu device or
+ * partition, NULL otherwise.
+ */
+static struct amdgpu_device *dma_buf_attach_adev(struct dma_buf_attachment *attach)
+{
+	if (attach->importer_ops == &amdgpu_dma_buf_attach_ops) {
+		struct drm_gem_object *obj = attach->importer_priv;
+		struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+
+		return amdgpu_ttm_adev(bo->tbo.bdev);
+	}
+
+	return NULL;
+}
+
+/**
+ * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation
+ *
+ * @dmabuf: DMA-buf where we attach to
+ * @attach: attachment to add
+ *
+ * Add the attachment as user to the exported DMA-buf.
+ */
+static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf,
+				 struct dma_buf_attachment *attach)
+{
+	struct amdgpu_device *attach_adev = dma_buf_attach_adev(attach);
+	struct drm_gem_object *obj = dmabuf->priv;
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	int r;
+
+	/*
+	 * Disable peer-to-peer access for DCC-enabled VRAM surfaces on GFX12+.
+	 * Such buffers cannot be safely accessed over P2P due to device-local
+	 * compression metadata. Fallback to system-memory path instead.
+	 * Device supports GFX12 (GC 12.x or newer)
+	 * BO was created with the AMDGPU_GEM_CREATE_GFX12_DCC flag
+	 *
+	 */
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(12, 0, 0) &&
+	    bo->flags & AMDGPU_GEM_CREATE_GFX12_DCC)
+		attach->peer2peer = false;
+
+	/*
+	 * Disable peer-to-peer access for DCC-enabled VRAM surfaces on GFX12+.
+	 * Such buffers cannot be safely accessed over P2P due to device-local
+	 * compression metadata. Fallback to system-memory path instead.
+	 * Device supports GFX12 (GC 12.x or newer)
+	 * BO was created with the AMDGPU_GEM_CREATE_GFX12_DCC flag
+	 *
+	 */
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(12, 0, 0) &&
+	    bo->flags & AMDGPU_GEM_CREATE_GFX12_DCC)
+		attach->peer2peer = false;
+
+	if (!amdgpu_dmabuf_is_xgmi_accessible(attach_adev, bo) &&
+	    pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0)
+		attach->peer2peer = false;
+
+	r = dma_resv_lock(bo->tbo.base.resv, NULL);
+	if (r)
+		return r;
+
+	amdgpu_vm_bo_update_shared(bo);
+
+	dma_resv_unlock(bo->tbo.base.resv);
+
+	return 0;
+}
+
+/**
+ * amdgpu_dma_buf_pin - &dma_buf_ops.pin implementation
+ *
+ * @attach: attachment to pin down
+ *
+ * Pin the BO which is backing the DMA-buf so that it can't move any more.
+ */
+static int amdgpu_dma_buf_pin(struct dma_buf_attachment *attach)
+{
+	struct dma_buf *dmabuf = attach->dmabuf;
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(dmabuf->priv);
+	u32 domains = bo->allowed_domains;
+
+	dma_resv_assert_held(dmabuf->resv);
+
+	/* Try pinning into VRAM to allow P2P with RDMA NICs without ODP
+	 * support if all attachments can do P2P. If any attachment can't do
+	 * P2P just pin into GTT instead.
+	 *
+	 * To avoid with conflicting pinnings between GPUs and RDMA when move
+	 * notifiers are disabled, only allow pinning in VRAM when move
+	 * notiers are enabled.
+	 */
+	if (!IS_ENABLED(CONFIG_DMABUF_MOVE_NOTIFY)) {
+		domains &= ~AMDGPU_GEM_DOMAIN_VRAM;
+	} else {
+		list_for_each_entry(attach, &dmabuf->attachments, node)
+			if (!attach->peer2peer)
+				domains &= ~AMDGPU_GEM_DOMAIN_VRAM;
+	}
+
+	if (domains & AMDGPU_GEM_DOMAIN_VRAM)
+		bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+
+	if (WARN_ON(!domains))
+		return -EINVAL;
+
+	return amdgpu_bo_pin(bo, domains);
+}
+
+/**
+ * amdgpu_dma_buf_unpin - &dma_buf_ops.unpin implementation
+ *
+ * @attach: attachment to unpin
+ *
+ * Unpin a previously pinned BO to make it movable again.
+ */
+static void amdgpu_dma_buf_unpin(struct dma_buf_attachment *attach)
+{
+	struct drm_gem_object *obj = attach->dmabuf->priv;
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+
+	amdgpu_bo_unpin(bo);
+}
+
+/**
+ * amdgpu_dma_buf_map - &dma_buf_ops.map_dma_buf implementation
+ * @attach: DMA-buf attachment
+ * @dir: DMA direction
+ *
+ * Makes sure that the shared DMA buffer can be accessed by the target device.
+ * For now, simply pins it to the GTT domain, where it should be accessible by
+ * all DMA devices.
+ *
+ * Returns:
+ * sg_table filled with the DMA addresses to use or ERR_PRT with negative error
+ * code.
+ */
+static struct sg_table *amdgpu_dma_buf_map(struct dma_buf_attachment *attach,
+					   enum dma_data_direction dir)
+{
+	struct dma_buf *dma_buf = attach->dmabuf;
+	struct drm_gem_object *obj = dma_buf->priv;
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct sg_table *sgt;
+	long r;
+
+	if (!bo->tbo.pin_count) {
+		/* move buffer into GTT or VRAM */
+		struct ttm_operation_ctx ctx = { false, false };
+		unsigned int domains = AMDGPU_GEM_DOMAIN_GTT;
+
+		if (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM &&
+		    attach->peer2peer) {
+			bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+			domains |= AMDGPU_GEM_DOMAIN_VRAM;
+		}
+		amdgpu_bo_placement_from_domain(bo, domains);
+		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+		if (r)
+			return ERR_PTR(r);
+	}
+
+	switch (bo->tbo.resource->mem_type) {
+	case TTM_PL_TT:
+		sgt = drm_prime_pages_to_sg(obj->dev,
+					    bo->tbo.ttm->pages,
+					    bo->tbo.ttm->num_pages);
+		if (IS_ERR(sgt))
+			return sgt;
+
+		if (dma_map_sgtable(attach->dev, sgt, dir,
+				    DMA_ATTR_SKIP_CPU_SYNC))
+			goto error_free;
+		break;
+
+	case TTM_PL_VRAM:
+		/* XGMI-accessible memory should never be DMA-mapped */
+		if (WARN_ON(amdgpu_dmabuf_is_xgmi_accessible(
+				dma_buf_attach_adev(attach), bo)))
+			return ERR_PTR(-EINVAL);
+
+		r = amdgpu_vram_mgr_alloc_sgt(adev, bo->tbo.resource, 0,
+					      bo->tbo.base.size, attach->dev,
+					      dir, &sgt);
+		if (r)
+			return ERR_PTR(r);
+		break;
+	default:
+		return ERR_PTR(-EINVAL);
+	}
+
+	return sgt;
+
+error_free:
+	sg_free_table(sgt);
+	kfree(sgt);
+	return ERR_PTR(-EBUSY);
+}
+
+/**
+ * amdgpu_dma_buf_unmap - &dma_buf_ops.unmap_dma_buf implementation
+ * @attach: DMA-buf attachment
+ * @sgt: sg_table to unmap
+ * @dir: DMA direction
+ *
+ * This is called when a shared DMA buffer no longer needs to be accessible by
+ * another device. For now, simply unpins the buffer from GTT.
+ */
+static void amdgpu_dma_buf_unmap(struct dma_buf_attachment *attach,
+				 struct sg_table *sgt,
+				 enum dma_data_direction dir)
+{
+	if (sg_page(sgt->sgl)) {
+		dma_unmap_sgtable(attach->dev, sgt, dir, 0);
+		sg_free_table(sgt);
+		kfree(sgt);
+	} else {
+		amdgpu_vram_mgr_free_sgt(attach->dev, dir, sgt);
+	}
+}
+
+/**
+ * amdgpu_dma_buf_begin_cpu_access - &dma_buf_ops.begin_cpu_access implementation
+ * @dma_buf: Shared DMA buffer
+ * @direction: Direction of DMA transfer
+ *
+ * This is called before CPU access to the shared DMA buffer's memory. If it's
+ * a read access, the buffer is moved to the GTT domain if possible, for optimal
+ * CPU read performance.
+ *
+ * Returns:
+ * 0 on success or a negative error code on failure.
+ */
+static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf,
+					   enum dma_data_direction direction)
+{
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(dma_buf->priv);
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct ttm_operation_ctx ctx = { true, false };
+	u32 domain = amdgpu_display_supported_domains(adev, bo->flags);
+	int ret;
+	bool reads = (direction == DMA_BIDIRECTIONAL ||
+		      direction == DMA_FROM_DEVICE);
+
+	if (!reads || !(domain & AMDGPU_GEM_DOMAIN_GTT))
+		return 0;
+
+	/* move to gtt */
+	ret = amdgpu_bo_reserve(bo, false);
+	if (unlikely(ret != 0))
+		return ret;
+
+	if (!bo->tbo.pin_count &&
+	    (bo->allowed_domains & AMDGPU_GEM_DOMAIN_GTT)) {
+		amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+		ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	}
+
+	amdgpu_bo_unreserve(bo);
+	return ret;
+}
+
+static int amdgpu_dma_buf_vmap(struct dma_buf *dma_buf, struct iosys_map *map)
+{
+	struct drm_gem_object *obj = dma_buf->priv;
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+	int ret;
+
+	/*
+	 * Pin to keep buffer in place while it's vmap'ed. The actual
+	 * domain is not that important as long as it's mapable. Using
+	 * GTT and VRAM should be compatible with most use cases.
+	 */
+	ret = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT | AMDGPU_GEM_DOMAIN_VRAM);
+	if (ret)
+		return ret;
+	ret = drm_gem_dmabuf_vmap(dma_buf, map);
+	if (ret)
+		amdgpu_bo_unpin(bo);
+
+	return ret;
+}
+
+static void amdgpu_dma_buf_vunmap(struct dma_buf *dma_buf, struct iosys_map *map)
+{
+	struct drm_gem_object *obj = dma_buf->priv;
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+
+	drm_gem_dmabuf_vunmap(dma_buf, map);
+	amdgpu_bo_unpin(bo);
+}
+
+const struct dma_buf_ops amdgpu_dmabuf_ops = {
+	.attach = amdgpu_dma_buf_attach,
+	.pin = amdgpu_dma_buf_pin,
+	.unpin = amdgpu_dma_buf_unpin,
+	.map_dma_buf = amdgpu_dma_buf_map,
+	.unmap_dma_buf = amdgpu_dma_buf_unmap,
+	.release = drm_gem_dmabuf_release,
+	.begin_cpu_access = amdgpu_dma_buf_begin_cpu_access,
+	.mmap = drm_gem_dmabuf_mmap,
+	.vmap = amdgpu_dma_buf_vmap,
+	.vunmap = amdgpu_dma_buf_vunmap,
+};
+
+/**
+ * amdgpu_gem_prime_export - &drm_driver.gem_prime_export implementation
+ * @gobj: GEM BO
+ * @flags: Flags such as DRM_CLOEXEC and DRM_RDWR.
+ *
+ * The main work is done by the &drm_gem_prime_export helper.
+ *
+ * Returns:
+ * Shared DMA buffer representing the GEM BO from the given device.
+ */
+struct dma_buf *amdgpu_gem_prime_export(struct drm_gem_object *gobj,
+					int flags)
+{
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
+	struct dma_buf *buf;
+	struct ttm_operation_ctx ctx = {
+		.interruptible = true,
+		.no_wait_gpu = true,
+		/* We opt to avoid OOM on system pages allocations */
+		.gfp_retry_mayfail = true,
+		.allow_res_evict = false,
+	};
+	int ret;
+
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) ||
+	    bo->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID)
+		return ERR_PTR(-EPERM);
+
+	ret = ttm_bo_setup_export(&bo->tbo, &ctx);
+	if (ret)
+		return ERR_PTR(ret);
+
+	buf = drm_gem_prime_export(gobj, flags);
+	if (!IS_ERR(buf))
+		buf->ops = &amdgpu_dmabuf_ops;
+
+	return buf;
+}
+
+/**
+ * amdgpu_dma_buf_create_obj - create BO for DMA-buf import
+ *
+ * @dev: DRM device
+ * @dma_buf: DMA-buf
+ *
+ * Creates an empty SG BO for DMA-buf import.
+ *
+ * Returns:
+ * A new GEM BO of the given DRM device, representing the memory
+ * described by the given DMA-buf attachment and scatter/gather table.
+ */
+static struct drm_gem_object *
+amdgpu_dma_buf_create_obj(struct drm_device *dev, struct dma_buf *dma_buf)
+{
+	struct dma_resv *resv = dma_buf->resv;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	struct drm_gem_object *gobj;
+	struct amdgpu_bo *bo;
+	uint64_t flags = 0;
+	int ret;
+
+	dma_resv_lock(resv, NULL);
+
+	if (dma_buf->ops == &amdgpu_dmabuf_ops) {
+		struct amdgpu_bo *other = gem_to_amdgpu_bo(dma_buf->priv);
+
+		flags |= other->flags & (AMDGPU_GEM_CREATE_CPU_GTT_USWC |
+					 AMDGPU_GEM_CREATE_COHERENT |
+					 AMDGPU_GEM_CREATE_EXT_COHERENT |
+					 AMDGPU_GEM_CREATE_UNCACHED);
+	}
+
+	ret = amdgpu_gem_object_create(adev, dma_buf->size, PAGE_SIZE,
+				       AMDGPU_GEM_DOMAIN_CPU, flags,
+				       ttm_bo_type_sg, resv, &gobj, 0);
+	if (ret)
+		goto error;
+
+	bo = gem_to_amdgpu_bo(gobj);
+	bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
+	bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
+
+	dma_resv_unlock(resv);
+	return gobj;
+
+error:
+	dma_resv_unlock(resv);
+	return ERR_PTR(ret);
+}
+
+/**
+ * amdgpu_dma_buf_move_notify - &attach.move_notify implementation
+ *
+ * @attach: the DMA-buf attachment
+ *
+ * Invalidate the DMA-buf attachment, making sure that the we re-create the
+ * mapping before the next use.
+ */
+static void
+amdgpu_dma_buf_move_notify(struct dma_buf_attachment *attach)
+{
+	struct drm_gem_object *obj = attach->importer_priv;
+	struct ww_acquire_ctx *ticket = dma_resv_locking_ctx(obj->resv);
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct ttm_operation_ctx ctx = { false, false };
+	struct ttm_placement placement = {};
+	struct amdgpu_vm_bo_base *bo_base;
+	int r;
+
+	/* FIXME: This should be after the "if", but needs a fix to make sure
+	 * DMABuf imports are initialized in the right VM list.
+	 */
+	amdgpu_vm_bo_invalidate(bo, false);
+	if (!bo->tbo.resource || bo->tbo.resource->mem_type == TTM_PL_SYSTEM)
+		return;
+
+	r = ttm_bo_validate(&bo->tbo, &placement, &ctx);
+	if (r) {
+		DRM_ERROR("Failed to invalidate DMA-buf import (%d))\n", r);
+		return;
+	}
+
+	for (bo_base = bo->vm_bo; bo_base; bo_base = bo_base->next) {
+		struct amdgpu_vm *vm = bo_base->vm;
+		struct dma_resv *resv = vm->root.bo->tbo.base.resv;
+
+		if (ticket) {
+			/* When we get an error here it means that somebody
+			 * else is holding the VM lock and updating page tables
+			 * So we can just continue here.
+			 */
+			r = dma_resv_lock(resv, ticket);
+			if (r)
+				continue;
+
+		} else {
+			/* TODO: This is more problematic and we actually need
+			 * to allow page tables updates without holding the
+			 * lock.
+			 */
+			if (!dma_resv_trylock(resv))
+				continue;
+		}
+
+		/* Reserve fences for two SDMA page table updates */
+		r = dma_resv_reserve_fences(resv, 2);
+		if (!r)
+			r = amdgpu_vm_clear_freed(adev, vm, NULL);
+		if (!r)
+			r = amdgpu_vm_handle_moved(adev, vm, ticket);
+
+		if (r && r != -EBUSY)
+			DRM_ERROR("Failed to invalidate VM page tables (%d))\n",
+				  r);
+
+		dma_resv_unlock(resv);
+	}
+}
+
+static const struct dma_buf_attach_ops amdgpu_dma_buf_attach_ops = {
+	.allow_peer2peer = true,
+	.move_notify = amdgpu_dma_buf_move_notify
+};
+
+/**
+ * amdgpu_gem_prime_import - &drm_driver.gem_prime_import implementation
+ * @dev: DRM device
+ * @dma_buf: Shared DMA buffer
+ *
+ * Import a dma_buf into a the driver and potentially create a new GEM object.
+ *
+ * Returns:
+ * GEM BO representing the shared DMA buffer for the given device.
+ */
+struct drm_gem_object *amdgpu_gem_prime_import(struct drm_device *dev,
+					       struct dma_buf *dma_buf)
+{
+	struct dma_buf_attachment *attach;
+	struct drm_gem_object *obj;
+
+	if (dma_buf->ops == &amdgpu_dmabuf_ops) {
+		obj = dma_buf->priv;
+		if (obj->dev == dev) {
+			/*
+			 * Importing dmabuf exported from out own gem increases
+			 * refcount on gem itself instead of f_count of dmabuf.
+			 */
+			drm_gem_object_get(obj);
+			return obj;
+		}
+	}
+
+	obj = amdgpu_dma_buf_create_obj(dev, dma_buf);
+	if (IS_ERR(obj))
+		return obj;
+
+	attach = dma_buf_dynamic_attach(dma_buf, dev->dev,
+					&amdgpu_dma_buf_attach_ops, obj);
+	if (IS_ERR(attach)) {
+		drm_gem_object_put(obj);
+		return ERR_CAST(attach);
+	}
+
+	get_dma_buf(dma_buf);
+	obj->import_attach = attach;
+	return obj;
+}
+
+/**
+ * amdgpu_dmabuf_is_xgmi_accessible - Check if xgmi available for P2P transfer
+ *
+ * @adev: amdgpu_device pointer of the importer
+ * @bo: amdgpu buffer object
+ *
+ * Returns:
+ * True if dmabuf accessible over xgmi, false otherwise.
+ */
+bool amdgpu_dmabuf_is_xgmi_accessible(struct amdgpu_device *adev,
+				      struct amdgpu_bo *bo)
+{
+	struct drm_gem_object *obj = &bo->tbo.base;
+	struct drm_gem_object *gobj;
+
+	if (!adev)
+		return false;
+
+	if (drm_gem_is_imported(obj)) {
+		struct dma_buf *dma_buf = obj->import_attach->dmabuf;
+
+		if (dma_buf->ops != &amdgpu_dmabuf_ops)
+			/* No XGMI with non AMD GPUs */
+			return false;
+
+		gobj = dma_buf->priv;
+		bo = gem_to_amdgpu_bo(gobj);
+	}
+
+	if (amdgpu_xgmi_same_hive(adev, amdgpu_ttm_adev(bo->tbo.bdev)) &&
+			(bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM))
+		return true;
+
+	return false;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
new file mode 100644
index 000000000000..3e93b9b407a9
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
@@ -0,0 +1,37 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_DMA_BUF_H__
+#define __AMDGPU_DMA_BUF_H__
+
+#include <drm/drm_gem.h>
+
+struct dma_buf *amdgpu_gem_prime_export(struct drm_gem_object *gobj,
+					int flags);
+struct drm_gem_object *amdgpu_gem_prime_import(struct drm_device *dev,
+					    struct dma_buf *dma_buf);
+bool amdgpu_dmabuf_is_xgmi_accessible(struct amdgpu_device *adev,
+				      struct amdgpu_bo *bo);
+
+extern const struct dma_buf_ops amdgpu_dmabuf_ops;
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h
new file mode 100644
index 000000000000..2675689ef70f
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h
@@ -0,0 +1,372 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef AMDGPU_DOORBELL_H
+#define AMDGPU_DOORBELL_H
+
+/*
+ * GPU doorbell structures, functions & helpers
+ */
+struct amdgpu_doorbell {
+	/* doorbell mmio */
+	resource_size_t		base;
+	resource_size_t		size;
+
+	/* Number of doorbells reserved for amdgpu kernel driver */
+	u32 num_kernel_doorbells;
+
+	/* Kernel doorbells */
+	struct amdgpu_bo *kernel_doorbells;
+
+	/* For CPU access of doorbells */
+	uint32_t *cpu_addr;
+};
+
+/* Reserved doorbells for amdgpu (including multimedia).
+ * KFD can use all the rest in the 2M doorbell bar.
+ * For asic before vega10, doorbell is 32-bit, so the
+ * index/offset is in dword. For vega10 and after, doorbell
+ * can be 64-bit, so the index defined is in qword.
+ */
+struct amdgpu_doorbell_index {
+	uint32_t kiq;
+	uint32_t mec_ring0;
+	uint32_t mec_ring1;
+	uint32_t mec_ring2;
+	uint32_t mec_ring3;
+	uint32_t mec_ring4;
+	uint32_t mec_ring5;
+	uint32_t mec_ring6;
+	uint32_t mec_ring7;
+	uint32_t userqueue_start;
+	uint32_t userqueue_end;
+	uint32_t gfx_ring0;
+	uint32_t gfx_ring1;
+	uint32_t gfx_userqueue_start;
+	uint32_t gfx_userqueue_end;
+	uint32_t sdma_engine[16];
+	uint32_t mes_ring0;
+	uint32_t mes_ring1;
+	uint32_t ih;
+	union {
+		struct {
+			uint32_t vcn_ring0_1;
+			uint32_t vcn_ring2_3;
+			uint32_t vcn_ring4_5;
+			uint32_t vcn_ring6_7;
+		} vcn;
+		struct {
+			uint32_t uvd_ring0_1;
+			uint32_t uvd_ring2_3;
+			uint32_t uvd_ring4_5;
+			uint32_t uvd_ring6_7;
+			uint32_t vce_ring0_1;
+			uint32_t vce_ring2_3;
+			uint32_t vce_ring4_5;
+			uint32_t vce_ring6_7;
+		} uvd_vce;
+	};
+	uint32_t vpe_ring;
+	uint32_t first_non_cp;
+	uint32_t last_non_cp;
+	uint32_t max_assignment;
+	/* Per engine SDMA doorbell size in dword */
+	uint32_t sdma_doorbell_range;
+	/* Per xcc doorbell size for KIQ/KCQ */
+	uint32_t xcc_doorbell_range;
+};
+
+enum AMDGPU_DOORBELL_ASSIGNMENT {
+	AMDGPU_DOORBELL_KIQ                     = 0x000,
+	AMDGPU_DOORBELL_HIQ                     = 0x001,
+	AMDGPU_DOORBELL_DIQ                     = 0x002,
+	AMDGPU_DOORBELL_MEC_RING0               = 0x010,
+	AMDGPU_DOORBELL_MEC_RING1               = 0x011,
+	AMDGPU_DOORBELL_MEC_RING2               = 0x012,
+	AMDGPU_DOORBELL_MEC_RING3               = 0x013,
+	AMDGPU_DOORBELL_MEC_RING4               = 0x014,
+	AMDGPU_DOORBELL_MEC_RING5               = 0x015,
+	AMDGPU_DOORBELL_MEC_RING6               = 0x016,
+	AMDGPU_DOORBELL_MEC_RING7               = 0x017,
+	AMDGPU_DOORBELL_GFX_RING0               = 0x020,
+	AMDGPU_DOORBELL_sDMA_ENGINE0            = 0x1E0,
+	AMDGPU_DOORBELL_sDMA_ENGINE1            = 0x1E1,
+	AMDGPU_DOORBELL_IH                      = 0x1E8,
+	AMDGPU_DOORBELL_MAX_ASSIGNMENT          = 0x3FF,
+	AMDGPU_DOORBELL_INVALID                 = 0xFFFF
+};
+
+enum AMDGPU_VEGA20_DOORBELL_ASSIGNMENT {
+
+	/* Compute + GFX: 0~255 */
+	AMDGPU_VEGA20_DOORBELL_KIQ                     = 0x000,
+	AMDGPU_VEGA20_DOORBELL_HIQ                     = 0x001,
+	AMDGPU_VEGA20_DOORBELL_DIQ                     = 0x002,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING0               = 0x003,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING1               = 0x004,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING2               = 0x005,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING3               = 0x006,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING4               = 0x007,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING5               = 0x008,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING6               = 0x009,
+	AMDGPU_VEGA20_DOORBELL_MEC_RING7               = 0x00A,
+	AMDGPU_VEGA20_DOORBELL_USERQUEUE_START	       = 0x00B,
+	AMDGPU_VEGA20_DOORBELL_USERQUEUE_END	       = 0x08A,
+	AMDGPU_VEGA20_DOORBELL_GFX_RING0               = 0x08B,
+	/* SDMA:256~335*/
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0            = 0x100,
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE1            = 0x10A,
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE2            = 0x114,
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE3            = 0x11E,
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE4            = 0x128,
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE5            = 0x132,
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE6            = 0x13C,
+	AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE7            = 0x146,
+	/* IH: 376~391 */
+	AMDGPU_VEGA20_DOORBELL_IH                      = 0x178,
+	/* MMSCH: 392~407
+	 * overlap the doorbell assignment with VCN as they are  mutually exclusive
+	 * VCN engine's doorbell is 32 bit and two VCN ring share one QWORD
+	 */
+	AMDGPU_VEGA20_DOORBELL64_VCN0_1                  = 0x188, /* VNC0 */
+	AMDGPU_VEGA20_DOORBELL64_VCN2_3                  = 0x189,
+	AMDGPU_VEGA20_DOORBELL64_VCN4_5                  = 0x18A,
+	AMDGPU_VEGA20_DOORBELL64_VCN6_7                  = 0x18B,
+
+	AMDGPU_VEGA20_DOORBELL64_VCN8_9                  = 0x18C, /* VNC1 */
+	AMDGPU_VEGA20_DOORBELL64_VCNa_b                  = 0x18D,
+	AMDGPU_VEGA20_DOORBELL64_VCNc_d                  = 0x18E,
+	AMDGPU_VEGA20_DOORBELL64_VCNe_f                  = 0x18F,
+
+	AMDGPU_VEGA20_DOORBELL64_UVD_RING0_1             = 0x188,
+	AMDGPU_VEGA20_DOORBELL64_UVD_RING2_3             = 0x189,
+	AMDGPU_VEGA20_DOORBELL64_UVD_RING4_5             = 0x18A,
+	AMDGPU_VEGA20_DOORBELL64_UVD_RING6_7             = 0x18B,
+
+	AMDGPU_VEGA20_DOORBELL64_VCE_RING0_1             = 0x18C,
+	AMDGPU_VEGA20_DOORBELL64_VCE_RING2_3             = 0x18D,
+	AMDGPU_VEGA20_DOORBELL64_VCE_RING4_5             = 0x18E,
+	AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7             = 0x18F,
+
+	AMDGPU_VEGA20_DOORBELL64_FIRST_NON_CP            = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0,
+	AMDGPU_VEGA20_DOORBELL64_LAST_NON_CP             = AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7,
+
+	/* kiq/kcq from second XCD. Max 8 XCDs */
+	AMDGPU_VEGA20_DOORBELL_XCC1_KIQ_START             = 0x190,
+	/* 8 compute rings per GC. Max to 0x1CE */
+	AMDGPU_VEGA20_DOORBELL_XCC1_MEC_RING0_START       = 0x197,
+
+	/* AID1 SDMA: 0x1D0 ~ 0x1F7 */
+	AMDGPU_VEGA20_DOORBELL_AID1_sDMA_START           = 0x1D0,
+
+	AMDGPU_VEGA20_DOORBELL_MAX_ASSIGNMENT            = 0x1F7,
+	AMDGPU_VEGA20_DOORBELL_INVALID                   = 0xFFFF
+};
+
+enum AMDGPU_NAVI10_DOORBELL_ASSIGNMENT {
+
+	/* Compute + GFX: 0~255 */
+	AMDGPU_NAVI10_DOORBELL_KIQ			= 0x000,
+	AMDGPU_NAVI10_DOORBELL_HIQ			= 0x001,
+	AMDGPU_NAVI10_DOORBELL_DIQ			= 0x002,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING0		= 0x003,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING1		= 0x004,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING2		= 0x005,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING3		= 0x006,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING4		= 0x007,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING5		= 0x008,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING6		= 0x009,
+	AMDGPU_NAVI10_DOORBELL_MEC_RING7		= 0x00A,
+	AMDGPU_NAVI10_DOORBELL_MES_RING0	        = 0x00B,
+	AMDGPU_NAVI10_DOORBELL_MES_RING1		= 0x00C,
+	AMDGPU_NAVI10_DOORBELL_USERQUEUE_START		= 0x00D,
+	AMDGPU_NAVI10_DOORBELL_USERQUEUE_END		= 0x08A,
+	AMDGPU_NAVI10_DOORBELL_GFX_RING0		= 0x08B,
+	AMDGPU_NAVI10_DOORBELL_GFX_RING1		= 0x08C,
+	AMDGPU_NAVI10_DOORBELL_GFX_USERQUEUE_START	= 0x08D,
+	AMDGPU_NAVI10_DOORBELL_GFX_USERQUEUE_END	= 0x0FF,
+
+	/* SDMA:256~335*/
+	AMDGPU_NAVI10_DOORBELL_sDMA_ENGINE0		= 0x100,
+	AMDGPU_NAVI10_DOORBELL_sDMA_ENGINE1		= 0x10A,
+	AMDGPU_NAVI10_DOORBELL_sDMA_ENGINE2		= 0x114,
+	AMDGPU_NAVI10_DOORBELL_sDMA_ENGINE3		= 0x11E,
+	/* IH: 376~391 */
+	AMDGPU_NAVI10_DOORBELL_IH			= 0x178,
+	/* MMSCH: 392~407
+	 * overlap the doorbell assignment with VCN as they are  mutually exclusive
+	 * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD
+	 */
+	AMDGPU_NAVI10_DOORBELL64_VCN0_1			= 0x188, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */
+	AMDGPU_NAVI10_DOORBELL64_VCN2_3			= 0x189,
+	AMDGPU_NAVI10_DOORBELL64_VCN4_5			= 0x18A,
+	AMDGPU_NAVI10_DOORBELL64_VCN6_7			= 0x18B,
+
+	AMDGPU_NAVI10_DOORBELL64_VCN8_9			= 0x18C,
+	AMDGPU_NAVI10_DOORBELL64_VCNa_b			= 0x18D,
+	AMDGPU_NAVI10_DOORBELL64_VCNc_d			= 0x18E,
+	AMDGPU_NAVI10_DOORBELL64_VCNe_f			= 0x18F,
+
+	AMDGPU_NAVI10_DOORBELL64_VPE			= 0x190,
+
+	AMDGPU_NAVI10_DOORBELL64_FIRST_NON_CP		= AMDGPU_NAVI10_DOORBELL_sDMA_ENGINE0,
+	AMDGPU_NAVI10_DOORBELL64_LAST_NON_CP		= AMDGPU_NAVI10_DOORBELL64_VPE,
+
+	AMDGPU_NAVI10_DOORBELL_MAX_ASSIGNMENT		= AMDGPU_NAVI10_DOORBELL64_VPE,
+	AMDGPU_NAVI10_DOORBELL_INVALID			= 0xFFFF
+};
+
+/*
+ * 64bit doorbell, offset are in QWORD, occupy 2KB doorbell space
+ */
+enum AMDGPU_DOORBELL64_ASSIGNMENT {
+	/*
+	 * All compute related doorbells: kiq, hiq, diq, traditional compute queue, user queue, should locate in
+	 * a continues range so that programming CP_MEC_DOORBELL_RANGE_LOWER/UPPER can cover this range.
+	 *  Compute related doorbells are allocated from 0x00 to 0x8a
+	 */
+
+
+	/* kernel scheduling */
+	AMDGPU_DOORBELL64_KIQ                     = 0x00,
+
+	/* HSA interface queue and debug queue */
+	AMDGPU_DOORBELL64_HIQ                     = 0x01,
+	AMDGPU_DOORBELL64_DIQ                     = 0x02,
+
+	/* Compute engines */
+	AMDGPU_DOORBELL64_MEC_RING0               = 0x03,
+	AMDGPU_DOORBELL64_MEC_RING1               = 0x04,
+	AMDGPU_DOORBELL64_MEC_RING2               = 0x05,
+	AMDGPU_DOORBELL64_MEC_RING3               = 0x06,
+	AMDGPU_DOORBELL64_MEC_RING4               = 0x07,
+	AMDGPU_DOORBELL64_MEC_RING5               = 0x08,
+	AMDGPU_DOORBELL64_MEC_RING6               = 0x09,
+	AMDGPU_DOORBELL64_MEC_RING7               = 0x0a,
+
+	/* User queue doorbell range (128 doorbells) */
+	AMDGPU_DOORBELL64_USERQUEUE_START         = 0x0b,
+	AMDGPU_DOORBELL64_USERQUEUE_END           = 0x8a,
+
+	/* Graphics engine */
+	AMDGPU_DOORBELL64_GFX_RING0               = 0x8b,
+
+	/*
+	 * Other graphics doorbells can be allocated here: from 0x8c to 0xdf
+	 * Graphics voltage island aperture 1
+	 * default non-graphics QWORD index is 0xe0 - 0xFF inclusive
+	 */
+
+	/* For vega10 sriov, the sdma doorbell must be fixed as follow
+	 * to keep the same setting with host driver, or it will
+	 * happen conflicts
+	 */
+	AMDGPU_DOORBELL64_sDMA_ENGINE0            = 0xF0,
+	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE0     = 0xF1,
+	AMDGPU_DOORBELL64_sDMA_ENGINE1            = 0xF2,
+	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE1     = 0xF3,
+
+	/* Interrupt handler */
+	AMDGPU_DOORBELL64_IH                      = 0xF4,  /* For legacy interrupt ring buffer */
+	AMDGPU_DOORBELL64_IH_RING1                = 0xF5,  /* For page migration request log */
+	AMDGPU_DOORBELL64_IH_RING2                = 0xF6,  /* For page migration translation/invalidation log */
+
+	/* VCN engine use 32 bits doorbell  */
+	AMDGPU_DOORBELL64_VCN0_1                  = 0xF8, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */
+	AMDGPU_DOORBELL64_VCN2_3                  = 0xF9,
+	AMDGPU_DOORBELL64_VCN4_5                  = 0xFA,
+	AMDGPU_DOORBELL64_VCN6_7                  = 0xFB,
+
+	/* overlap the doorbell assignment with VCN as they are  mutually exclusive
+	 * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD
+	 */
+	AMDGPU_DOORBELL64_UVD_RING0_1             = 0xF8,
+	AMDGPU_DOORBELL64_UVD_RING2_3             = 0xF9,
+	AMDGPU_DOORBELL64_UVD_RING4_5             = 0xFA,
+	AMDGPU_DOORBELL64_UVD_RING6_7             = 0xFB,
+
+	AMDGPU_DOORBELL64_VCE_RING0_1             = 0xFC,
+	AMDGPU_DOORBELL64_VCE_RING2_3             = 0xFD,
+	AMDGPU_DOORBELL64_VCE_RING4_5             = 0xFE,
+	AMDGPU_DOORBELL64_VCE_RING6_7             = 0xFF,
+
+	AMDGPU_DOORBELL64_FIRST_NON_CP            = AMDGPU_DOORBELL64_sDMA_ENGINE0,
+	AMDGPU_DOORBELL64_LAST_NON_CP             = AMDGPU_DOORBELL64_VCE_RING6_7,
+
+	AMDGPU_DOORBELL64_MAX_ASSIGNMENT          = 0xFF,
+	AMDGPU_DOORBELL64_INVALID                 = 0xFFFF
+};
+
+enum AMDGPU_DOORBELL_ASSIGNMENT_LAYOUT1 {
+
+	/* XCC0: 0x00 ~20, XCC1: 20 ~ 2F ... */
+
+	/* KIQ/HIQ/DIQ */
+	AMDGPU_DOORBELL_LAYOUT1_KIQ_START		= 0x000,
+	AMDGPU_DOORBELL_LAYOUT1_HIQ			= 0x001,
+	AMDGPU_DOORBELL_LAYOUT1_DIQ			= 0x002,
+	/* Compute: 0x08 ~ 0x20  */
+	AMDGPU_DOORBELL_LAYOUT1_MEC_RING_START		= 0x008,
+	AMDGPU_DOORBELL_LAYOUT1_MEC_RING_END		= 0x00F,
+	AMDGPU_DOORBELL_LAYOUT1_USERQUEUE_START		= 0x010,
+	AMDGPU_DOORBELL_LAYOUT1_USERQUEUE_END		= 0x01F,
+	AMDGPU_DOORBELL_LAYOUT1_XCC_RANGE		= 0x020,
+
+	/* SDMA: 0x100 ~ 0x19F */
+	AMDGPU_DOORBELL_LAYOUT1_sDMA_ENGINE_START	= 0x100,
+	AMDGPU_DOORBELL_LAYOUT1_sDMA_ENGINE_END		= 0x19F,
+	/* IH: 0x1A0 ~ 0x1AF */
+	AMDGPU_DOORBELL_LAYOUT1_IH                      = 0x1A0,
+	/* VCN: 0x1B0 ~ 0x1E8 */
+	AMDGPU_DOORBELL_LAYOUT1_VCN_START               = 0x1B0,
+	AMDGPU_DOORBELL_LAYOUT1_VCN_END                 = 0x1E8,
+
+	AMDGPU_DOORBELL_LAYOUT1_FIRST_NON_CP		= AMDGPU_DOORBELL_LAYOUT1_sDMA_ENGINE_START,
+	AMDGPU_DOORBELL_LAYOUT1_LAST_NON_CP		= AMDGPU_DOORBELL_LAYOUT1_VCN_END,
+
+	AMDGPU_DOORBELL_LAYOUT1_MAX_ASSIGNMENT          = 0x1E8,
+	AMDGPU_DOORBELL_LAYOUT1_INVALID                 = 0xFFFF
+};
+
+u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index);
+void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v);
+u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index);
+void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v);
+
+/*
+ * GPU doorbell aperture helpers function.
+ */
+int amdgpu_doorbell_init(struct amdgpu_device *adev);
+void amdgpu_doorbell_fini(struct amdgpu_device *adev);
+int amdgpu_doorbell_create_kernel_doorbells(struct amdgpu_device *adev);
+uint32_t amdgpu_doorbell_index_on_bar(struct amdgpu_device *adev,
+				      struct amdgpu_bo *db_bo,
+				      uint32_t doorbell_index,
+				      uint32_t db_size);
+
+#define RDOORBELL32(index) amdgpu_mm_rdoorbell(adev, (index))
+#define WDOORBELL32(index, v) amdgpu_mm_wdoorbell(adev, (index), (v))
+#define RDOORBELL64(index) amdgpu_mm_rdoorbell64(adev, (index))
+#define WDOORBELL64(index, v) amdgpu_mm_wdoorbell64(adev, (index), (v))
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c
new file mode 100644
index 000000000000..3040437d99c2
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell_mgr.c
@@ -0,0 +1,244 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "amdgpu.h"
+
+/**
+ * amdgpu_mm_rdoorbell - read a doorbell dword
+ *
+ * @adev: amdgpu_device pointer
+ * @index: doorbell index
+ *
+ * Returns the value in the doorbell aperture at the
+ * requested doorbell index (CIK).
+ */
+u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index)
+{
+	if (amdgpu_device_skip_hw_access(adev))
+		return 0;
+
+	if (index < adev->doorbell.num_kernel_doorbells)
+		return readl(adev->doorbell.cpu_addr + index);
+
+	dev_err(adev->dev, "reading beyond doorbell aperture: 0x%08x!\n",
+		index);
+	return 0;
+}
+
+/**
+ * amdgpu_mm_wdoorbell - write a doorbell dword
+ *
+ * @adev: amdgpu_device pointer
+ * @index: doorbell index
+ * @v: value to write
+ *
+ * Writes @v to the doorbell aperture at the
+ * requested doorbell index (CIK).
+ */
+void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v)
+{
+	if (amdgpu_device_skip_hw_access(adev))
+		return;
+
+	if (index < adev->doorbell.num_kernel_doorbells)
+		writel(v, adev->doorbell.cpu_addr + index);
+	else
+		dev_err(adev->dev,
+			"writing beyond doorbell aperture: 0x%08x!\n", index);
+}
+
+/**
+ * amdgpu_mm_rdoorbell64 - read a doorbell Qword
+ *
+ * @adev: amdgpu_device pointer
+ * @index: doorbell index
+ *
+ * Returns the value in the doorbell aperture at the
+ * requested doorbell index (VEGA10+).
+ */
+u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index)
+{
+	if (amdgpu_device_skip_hw_access(adev))
+		return 0;
+
+	if (index < adev->doorbell.num_kernel_doorbells)
+		return atomic64_read((atomic64_t *)(adev->doorbell.cpu_addr + index));
+
+	dev_err(adev->dev, "reading beyond doorbell aperture: 0x%08x!\n",
+		index);
+	return 0;
+}
+
+/**
+ * amdgpu_mm_wdoorbell64 - write a doorbell Qword
+ *
+ * @adev: amdgpu_device pointer
+ * @index: doorbell index
+ * @v: value to write
+ *
+ * Writes @v to the doorbell aperture at the
+ * requested doorbell index (VEGA10+).
+ */
+void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v)
+{
+	if (amdgpu_device_skip_hw_access(adev))
+		return;
+
+	if (index < adev->doorbell.num_kernel_doorbells)
+		atomic64_set((atomic64_t *)(adev->doorbell.cpu_addr + index), v);
+	else
+		dev_err(adev->dev,
+			"writing beyond doorbell aperture: 0x%08x!\n", index);
+}
+
+/**
+ * amdgpu_doorbell_index_on_bar - Find doorbell's absolute offset in BAR
+ *
+ * @adev: amdgpu_device pointer
+ * @db_bo: doorbell object's bo
+ * @doorbell_index: doorbell relative index in this doorbell object
+ * @db_size: doorbell size is in byte
+ *
+ * returns doorbell's absolute index in BAR
+ */
+uint32_t amdgpu_doorbell_index_on_bar(struct amdgpu_device *adev,
+				      struct amdgpu_bo *db_bo,
+				      uint32_t doorbell_index,
+				      uint32_t db_size)
+{
+	int db_bo_offset;
+
+	db_bo_offset = amdgpu_bo_gpu_offset_no_check(db_bo);
+
+	/* doorbell index is 32 bit but doorbell's size can be 32 bit
+	 * or 64 bit, so *db_size(in byte)/4 for alignment.
+	 */
+	return db_bo_offset / sizeof(u32) + doorbell_index *
+	       DIV_ROUND_UP(db_size, 4);
+}
+
+/**
+ * amdgpu_doorbell_create_kernel_doorbells - Create kernel doorbells for graphics
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Creates doorbells for graphics driver usages.
+ * returns 0 on success, error otherwise.
+ */
+int amdgpu_doorbell_create_kernel_doorbells(struct amdgpu_device *adev)
+{
+	int r;
+	int size;
+
+	/* SI HW does not have doorbells, skip allocation */
+	if (adev->doorbell.num_kernel_doorbells == 0)
+		return 0;
+
+	/* Reserve first num_kernel_doorbells (page-aligned) for kernel ops */
+	size = ALIGN(adev->doorbell.num_kernel_doorbells * sizeof(u32), PAGE_SIZE);
+
+	/* Allocate an extra page for MES kernel usages (ring test) */
+	adev->mes.db_start_dw_offset = size / sizeof(u32);
+	size += PAGE_SIZE;
+
+	r = amdgpu_bo_create_kernel(adev,
+				    size,
+				    PAGE_SIZE,
+				    AMDGPU_GEM_DOMAIN_DOORBELL,
+				    &adev->doorbell.kernel_doorbells,
+				    NULL,
+				    (void **)&adev->doorbell.cpu_addr);
+	if (r) {
+		dev_err(adev->dev,
+			"Failed to allocate kernel doorbells, err=%d\n", r);
+		return r;
+	}
+
+	adev->doorbell.num_kernel_doorbells = size / sizeof(u32);
+	return 0;
+}
+
+/*
+ * GPU doorbell aperture helpers function.
+ */
+/**
+ * amdgpu_doorbell_init - Init doorbell driver information.
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Init doorbell driver information (CIK)
+ * Returns 0 on success, error on failure.
+ */
+int amdgpu_doorbell_init(struct amdgpu_device *adev)
+{
+
+	/* No doorbell on SI hardware generation */
+	if (adev->asic_type < CHIP_BONAIRE) {
+		adev->doorbell.base = 0;
+		adev->doorbell.size = 0;
+		adev->doorbell.num_kernel_doorbells = 0;
+		return 0;
+	}
+
+	if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET)
+		return -EINVAL;
+
+	amdgpu_asic_init_doorbell_index(adev);
+
+	/* doorbell bar mapping */
+	adev->doorbell.base = pci_resource_start(adev->pdev, 2);
+	adev->doorbell.size = pci_resource_len(adev->pdev, 2);
+
+	adev->doorbell.num_kernel_doorbells =
+		min_t(u32, adev->doorbell.size / sizeof(u32),
+		      adev->doorbell_index.max_assignment + 1);
+	if (adev->doorbell.num_kernel_doorbells == 0)
+		return -EINVAL;
+
+	/*
+	 * For Vega, reserve and map two pages on doorbell BAR since SDMA
+	 * paging queue doorbell use the second page. The
+	 * AMDGPU_DOORBELL64_MAX_ASSIGNMENT definition assumes all the
+	 * doorbells are in the first page. So with paging queue enabled,
+	 * the max num_kernel_doorbells should + 1 page (0x400 in dword)
+	 */
+	if (adev->asic_type >= CHIP_VEGA10)
+		adev->doorbell.num_kernel_doorbells += 0x400;
+
+	return 0;
+}
+
+/**
+ * amdgpu_doorbell_fini - Tear down doorbell driver information.
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Tear down doorbell driver information (CIK)
+ */
+void amdgpu_doorbell_fini(struct amdgpu_device *adev)
+{
+	amdgpu_bo_free_kernel(&adev->doorbell.kernel_doorbells,
+			      NULL,
+			      (void **)&adev->doorbell.cpu_addr);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c
deleted file mode 100644
index 38e9b0d3659a..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c
+++ /dev/null
@@ -1,969 +0,0 @@
-/*
- * Copyright 2011 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors: Alex Deucher
- */
-
-#include "drmP.h"
-#include "amdgpu.h"
-#include "amdgpu_atombios.h"
-#include "amdgpu_i2c.h"
-#include "amdgpu_dpm.h"
-#include "atom.h"
-
-void amdgpu_dpm_print_class_info(u32 class, u32 class2)
-{
-	const char *s;
-
-	switch (class & ATOM_PPLIB_CLASSIFICATION_UI_MASK) {
-	case ATOM_PPLIB_CLASSIFICATION_UI_NONE:
-	default:
-		s = "none";
-		break;
-	case ATOM_PPLIB_CLASSIFICATION_UI_BATTERY:
-		s = "battery";
-		break;
-	case ATOM_PPLIB_CLASSIFICATION_UI_BALANCED:
-		s = "balanced";
-		break;
-	case ATOM_PPLIB_CLASSIFICATION_UI_PERFORMANCE:
-		s = "performance";
-		break;
-	}
-	printk("\tui class: %s\n", s);
-	printk("\tinternal class:");
-	if (((class & ~ATOM_PPLIB_CLASSIFICATION_UI_MASK) == 0) &&
-	    (class2 == 0))
-		pr_cont(" none");
-	else {
-		if (class & ATOM_PPLIB_CLASSIFICATION_BOOT)
-			pr_cont(" boot");
-		if (class & ATOM_PPLIB_CLASSIFICATION_THERMAL)
-			pr_cont(" thermal");
-		if (class & ATOM_PPLIB_CLASSIFICATION_LIMITEDPOWERSOURCE)
-			pr_cont(" limited_pwr");
-		if (class & ATOM_PPLIB_CLASSIFICATION_REST)
-			pr_cont(" rest");
-		if (class & ATOM_PPLIB_CLASSIFICATION_FORCED)
-			pr_cont(" forced");
-		if (class & ATOM_PPLIB_CLASSIFICATION_3DPERFORMANCE)
-			pr_cont(" 3d_perf");
-		if (class & ATOM_PPLIB_CLASSIFICATION_OVERDRIVETEMPLATE)
-			pr_cont(" ovrdrv");
-		if (class & ATOM_PPLIB_CLASSIFICATION_UVDSTATE)
-			pr_cont(" uvd");
-		if (class & ATOM_PPLIB_CLASSIFICATION_3DLOW)
-			pr_cont(" 3d_low");
-		if (class & ATOM_PPLIB_CLASSIFICATION_ACPI)
-			pr_cont(" acpi");
-		if (class & ATOM_PPLIB_CLASSIFICATION_HD2STATE)
-			pr_cont(" uvd_hd2");
-		if (class & ATOM_PPLIB_CLASSIFICATION_HDSTATE)
-			pr_cont(" uvd_hd");
-		if (class & ATOM_PPLIB_CLASSIFICATION_SDSTATE)
-			pr_cont(" uvd_sd");
-		if (class2 & ATOM_PPLIB_CLASSIFICATION2_LIMITEDPOWERSOURCE_2)
-			pr_cont(" limited_pwr2");
-		if (class2 & ATOM_PPLIB_CLASSIFICATION2_ULV)
-			pr_cont(" ulv");
-		if (class2 & ATOM_PPLIB_CLASSIFICATION2_MVC)
-			pr_cont(" uvd_mvc");
-	}
-	pr_cont("\n");
-}
-
-void amdgpu_dpm_print_cap_info(u32 caps)
-{
-	printk("\tcaps:");
-	if (caps & ATOM_PPLIB_SINGLE_DISPLAY_ONLY)
-		pr_cont(" single_disp");
-	if (caps & ATOM_PPLIB_SUPPORTS_VIDEO_PLAYBACK)
-		pr_cont(" video");
-	if (caps & ATOM_PPLIB_DISALLOW_ON_DC)
-		pr_cont(" no_dc");
-	pr_cont("\n");
-}
-
-void amdgpu_dpm_print_ps_status(struct amdgpu_device *adev,
-				struct amdgpu_ps *rps)
-{
-	printk("\tstatus:");
-	if (rps == adev->pm.dpm.current_ps)
-		pr_cont(" c");
-	if (rps == adev->pm.dpm.requested_ps)
-		pr_cont(" r");
-	if (rps == adev->pm.dpm.boot_ps)
-		pr_cont(" b");
-	pr_cont("\n");
-}
-
-
-u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev)
-{
-	struct drm_device *dev = adev->ddev;
-	struct drm_crtc *crtc;
-	struct amdgpu_crtc *amdgpu_crtc;
-	u32 vblank_in_pixels;
-	u32 vblank_time_us = 0xffffffff; /* if the displays are off, vblank time is max */
-
-	if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) {
-		list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
-			amdgpu_crtc = to_amdgpu_crtc(crtc);
-			if (crtc->enabled && amdgpu_crtc->enabled && amdgpu_crtc->hw_mode.clock) {
-				vblank_in_pixels =
-					amdgpu_crtc->hw_mode.crtc_htotal *
-					(amdgpu_crtc->hw_mode.crtc_vblank_end -
-					amdgpu_crtc->hw_mode.crtc_vdisplay +
-					(amdgpu_crtc->v_border * 2));
-
-				vblank_time_us = vblank_in_pixels * 1000 / amdgpu_crtc->hw_mode.clock;
-				break;
-			}
-		}
-	}
-
-	return vblank_time_us;
-}
-
-u32 amdgpu_dpm_get_vrefresh(struct amdgpu_device *adev)
-{
-	struct drm_device *dev = adev->ddev;
-	struct drm_crtc *crtc;
-	struct amdgpu_crtc *amdgpu_crtc;
-	u32 vrefresh = 0;
-
-	if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) {
-		list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
-			amdgpu_crtc = to_amdgpu_crtc(crtc);
-			if (crtc->enabled && amdgpu_crtc->enabled && amdgpu_crtc->hw_mode.clock) {
-				vrefresh = drm_mode_vrefresh(&amdgpu_crtc->hw_mode);
-				break;
-			}
-		}
-	}
-
-	return vrefresh;
-}
-
-void amdgpu_calculate_u_and_p(u32 i, u32 r_c, u32 p_b,
-			      u32 *p, u32 *u)
-{
-	u32 b_c = 0;
-	u32 i_c;
-	u32 tmp;
-
-	i_c = (i * r_c) / 100;
-	tmp = i_c >> p_b;
-
-	while (tmp) {
-		b_c++;
-		tmp >>= 1;
-	}
-
-	*u = (b_c + 1) / 2;
-	*p = i_c / (1 << (2 * (*u)));
-}
-
-int amdgpu_calculate_at(u32 t, u32 h, u32 fh, u32 fl, u32 *tl, u32 *th)
-{
-	u32 k, a, ah, al;
-	u32 t1;
-
-	if ((fl == 0) || (fh == 0) || (fl > fh))
-		return -EINVAL;
-
-	k = (100 * fh) / fl;
-	t1 = (t * (k - 100));
-	a = (1000 * (100 * h + t1)) / (10000 + (t1 / 100));
-	a = (a + 5) / 10;
-	ah = ((a * t) + 5000) / 10000;
-	al = a - ah;
-
-	*th = t - ah;
-	*tl = t + al;
-
-	return 0;
-}
-
-bool amdgpu_is_uvd_state(u32 class, u32 class2)
-{
-	if (class & ATOM_PPLIB_CLASSIFICATION_UVDSTATE)
-		return true;
-	if (class & ATOM_PPLIB_CLASSIFICATION_HD2STATE)
-		return true;
-	if (class & ATOM_PPLIB_CLASSIFICATION_HDSTATE)
-		return true;
-	if (class & ATOM_PPLIB_CLASSIFICATION_SDSTATE)
-		return true;
-	if (class2 & ATOM_PPLIB_CLASSIFICATION2_MVC)
-		return true;
-	return false;
-}
-
-bool amdgpu_is_internal_thermal_sensor(enum amdgpu_int_thermal_type sensor)
-{
-	switch (sensor) {
-	case THERMAL_TYPE_RV6XX:
-	case THERMAL_TYPE_RV770:
-	case THERMAL_TYPE_EVERGREEN:
-	case THERMAL_TYPE_SUMO:
-	case THERMAL_TYPE_NI:
-	case THERMAL_TYPE_SI:
-	case THERMAL_TYPE_CI:
-	case THERMAL_TYPE_KV:
-		return true;
-	case THERMAL_TYPE_ADT7473_WITH_INTERNAL:
-	case THERMAL_TYPE_EMC2103_WITH_INTERNAL:
-		return false; /* need special handling */
-	case THERMAL_TYPE_NONE:
-	case THERMAL_TYPE_EXTERNAL:
-	case THERMAL_TYPE_EXTERNAL_GPIO:
-	default:
-		return false;
-	}
-}
-
-union power_info {
-	struct _ATOM_POWERPLAY_INFO info;
-	struct _ATOM_POWERPLAY_INFO_V2 info_2;
-	struct _ATOM_POWERPLAY_INFO_V3 info_3;
-	struct _ATOM_PPLIB_POWERPLAYTABLE pplib;
-	struct _ATOM_PPLIB_POWERPLAYTABLE2 pplib2;
-	struct _ATOM_PPLIB_POWERPLAYTABLE3 pplib3;
-	struct _ATOM_PPLIB_POWERPLAYTABLE4 pplib4;
-	struct _ATOM_PPLIB_POWERPLAYTABLE5 pplib5;
-};
-
-union fan_info {
-	struct _ATOM_PPLIB_FANTABLE fan;
-	struct _ATOM_PPLIB_FANTABLE2 fan2;
-	struct _ATOM_PPLIB_FANTABLE3 fan3;
-};
-
-static int amdgpu_parse_clk_voltage_dep_table(struct amdgpu_clock_voltage_dependency_table *amdgpu_table,
-					      ATOM_PPLIB_Clock_Voltage_Dependency_Table *atom_table)
-{
-	u32 size = atom_table->ucNumEntries *
-		sizeof(struct amdgpu_clock_voltage_dependency_entry);
-	int i;
-	ATOM_PPLIB_Clock_Voltage_Dependency_Record *entry;
-
-	amdgpu_table->entries = kzalloc(size, GFP_KERNEL);
-	if (!amdgpu_table->entries)
-		return -ENOMEM;
-
-	entry = &atom_table->entries[0];
-	for (i = 0; i < atom_table->ucNumEntries; i++) {
-		amdgpu_table->entries[i].clk = le16_to_cpu(entry->usClockLow) |
-			(entry->ucClockHigh << 16);
-		amdgpu_table->entries[i].v = le16_to_cpu(entry->usVoltage);
-		entry = (ATOM_PPLIB_Clock_Voltage_Dependency_Record *)
-			((u8 *)entry + sizeof(ATOM_PPLIB_Clock_Voltage_Dependency_Record));
-	}
-	amdgpu_table->count = atom_table->ucNumEntries;
-
-	return 0;
-}
-
-int amdgpu_get_platform_caps(struct amdgpu_device *adev)
-{
-	struct amdgpu_mode_info *mode_info = &adev->mode_info;
-	union power_info *power_info;
-	int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo);
-	u16 data_offset;
-	u8 frev, crev;
-
-	if (!amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
-				   &frev, &crev, &data_offset))
-		return -EINVAL;
-	power_info = (union power_info *)(mode_info->atom_context->bios + data_offset);
-
-	adev->pm.dpm.platform_caps = le32_to_cpu(power_info->pplib.ulPlatformCaps);
-	adev->pm.dpm.backbias_response_time = le16_to_cpu(power_info->pplib.usBackbiasTime);
-	adev->pm.dpm.voltage_response_time = le16_to_cpu(power_info->pplib.usVoltageTime);
-
-	return 0;
-}
-
-/* sizeof(ATOM_PPLIB_EXTENDEDHEADER) */
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V2 12
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V3 14
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V4 16
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V5 18
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V6 20
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V7 22
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V8 24
-#define SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V9 26
-
-int amdgpu_parse_extended_power_table(struct amdgpu_device *adev)
-{
-	struct amdgpu_mode_info *mode_info = &adev->mode_info;
-	union power_info *power_info;
-	union fan_info *fan_info;
-	ATOM_PPLIB_Clock_Voltage_Dependency_Table *dep_table;
-	int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo);
-	u16 data_offset;
-	u8 frev, crev;
-	int ret, i;
-
-	if (!amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
-				   &frev, &crev, &data_offset))
-		return -EINVAL;
-	power_info = (union power_info *)(mode_info->atom_context->bios + data_offset);
-
-	/* fan table */
-	if (le16_to_cpu(power_info->pplib.usTableSize) >=
-	    sizeof(struct _ATOM_PPLIB_POWERPLAYTABLE3)) {
-		if (power_info->pplib3.usFanTableOffset) {
-			fan_info = (union fan_info *)(mode_info->atom_context->bios + data_offset +
-						      le16_to_cpu(power_info->pplib3.usFanTableOffset));
-			adev->pm.dpm.fan.t_hyst = fan_info->fan.ucTHyst;
-			adev->pm.dpm.fan.t_min = le16_to_cpu(fan_info->fan.usTMin);
-			adev->pm.dpm.fan.t_med = le16_to_cpu(fan_info->fan.usTMed);
-			adev->pm.dpm.fan.t_high = le16_to_cpu(fan_info->fan.usTHigh);
-			adev->pm.dpm.fan.pwm_min = le16_to_cpu(fan_info->fan.usPWMMin);
-			adev->pm.dpm.fan.pwm_med = le16_to_cpu(fan_info->fan.usPWMMed);
-			adev->pm.dpm.fan.pwm_high = le16_to_cpu(fan_info->fan.usPWMHigh);
-			if (fan_info->fan.ucFanTableFormat >= 2)
-				adev->pm.dpm.fan.t_max = le16_to_cpu(fan_info->fan2.usTMax);
-			else
-				adev->pm.dpm.fan.t_max = 10900;
-			adev->pm.dpm.fan.cycle_delay = 100000;
-			if (fan_info->fan.ucFanTableFormat >= 3) {
-				adev->pm.dpm.fan.control_mode = fan_info->fan3.ucFanControlMode;
-				adev->pm.dpm.fan.default_max_fan_pwm =
-					le16_to_cpu(fan_info->fan3.usFanPWMMax);
-				adev->pm.dpm.fan.default_fan_output_sensitivity = 4836;
-				adev->pm.dpm.fan.fan_output_sensitivity =
-					le16_to_cpu(fan_info->fan3.usFanOutputSensitivity);
-			}
-			adev->pm.dpm.fan.ucode_fan_control = true;
-		}
-	}
-
-	/* clock dependancy tables, shedding tables */
-	if (le16_to_cpu(power_info->pplib.usTableSize) >=
-	    sizeof(struct _ATOM_PPLIB_POWERPLAYTABLE4)) {
-		if (power_info->pplib4.usVddcDependencyOnSCLKOffset) {
-			dep_table = (ATOM_PPLIB_Clock_Voltage_Dependency_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(power_info->pplib4.usVddcDependencyOnSCLKOffset));
-			ret = amdgpu_parse_clk_voltage_dep_table(&adev->pm.dpm.dyn_state.vddc_dependency_on_sclk,
-								 dep_table);
-			if (ret) {
-				amdgpu_free_extended_power_table(adev);
-				return ret;
-			}
-		}
-		if (power_info->pplib4.usVddciDependencyOnMCLKOffset) {
-			dep_table = (ATOM_PPLIB_Clock_Voltage_Dependency_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(power_info->pplib4.usVddciDependencyOnMCLKOffset));
-			ret = amdgpu_parse_clk_voltage_dep_table(&adev->pm.dpm.dyn_state.vddci_dependency_on_mclk,
-								 dep_table);
-			if (ret) {
-				amdgpu_free_extended_power_table(adev);
-				return ret;
-			}
-		}
-		if (power_info->pplib4.usVddcDependencyOnMCLKOffset) {
-			dep_table = (ATOM_PPLIB_Clock_Voltage_Dependency_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(power_info->pplib4.usVddcDependencyOnMCLKOffset));
-			ret = amdgpu_parse_clk_voltage_dep_table(&adev->pm.dpm.dyn_state.vddc_dependency_on_mclk,
-								 dep_table);
-			if (ret) {
-				amdgpu_free_extended_power_table(adev);
-				return ret;
-			}
-		}
-		if (power_info->pplib4.usMvddDependencyOnMCLKOffset) {
-			dep_table = (ATOM_PPLIB_Clock_Voltage_Dependency_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(power_info->pplib4.usMvddDependencyOnMCLKOffset));
-			ret = amdgpu_parse_clk_voltage_dep_table(&adev->pm.dpm.dyn_state.mvdd_dependency_on_mclk,
-								 dep_table);
-			if (ret) {
-				amdgpu_free_extended_power_table(adev);
-				return ret;
-			}
-		}
-		if (power_info->pplib4.usMaxClockVoltageOnDCOffset) {
-			ATOM_PPLIB_Clock_Voltage_Limit_Table *clk_v =
-				(ATOM_PPLIB_Clock_Voltage_Limit_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(power_info->pplib4.usMaxClockVoltageOnDCOffset));
-			if (clk_v->ucNumEntries) {
-				adev->pm.dpm.dyn_state.max_clock_voltage_on_dc.sclk =
-					le16_to_cpu(clk_v->entries[0].usSclkLow) |
-					(clk_v->entries[0].ucSclkHigh << 16);
-				adev->pm.dpm.dyn_state.max_clock_voltage_on_dc.mclk =
-					le16_to_cpu(clk_v->entries[0].usMclkLow) |
-					(clk_v->entries[0].ucMclkHigh << 16);
-				adev->pm.dpm.dyn_state.max_clock_voltage_on_dc.vddc =
-					le16_to_cpu(clk_v->entries[0].usVddc);
-				adev->pm.dpm.dyn_state.max_clock_voltage_on_dc.vddci =
-					le16_to_cpu(clk_v->entries[0].usVddci);
-			}
-		}
-		if (power_info->pplib4.usVddcPhaseShedLimitsTableOffset) {
-			ATOM_PPLIB_PhaseSheddingLimits_Table *psl =
-				(ATOM_PPLIB_PhaseSheddingLimits_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(power_info->pplib4.usVddcPhaseShedLimitsTableOffset));
-			ATOM_PPLIB_PhaseSheddingLimits_Record *entry;
-
-			adev->pm.dpm.dyn_state.phase_shedding_limits_table.entries =
-				kzalloc(psl->ucNumEntries *
-					sizeof(struct amdgpu_phase_shedding_limits_entry),
-					GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.phase_shedding_limits_table.entries) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-
-			entry = &psl->entries[0];
-			for (i = 0; i < psl->ucNumEntries; i++) {
-				adev->pm.dpm.dyn_state.phase_shedding_limits_table.entries[i].sclk =
-					le16_to_cpu(entry->usSclkLow) | (entry->ucSclkHigh << 16);
-				adev->pm.dpm.dyn_state.phase_shedding_limits_table.entries[i].mclk =
-					le16_to_cpu(entry->usMclkLow) | (entry->ucMclkHigh << 16);
-				adev->pm.dpm.dyn_state.phase_shedding_limits_table.entries[i].voltage =
-					le16_to_cpu(entry->usVoltage);
-				entry = (ATOM_PPLIB_PhaseSheddingLimits_Record *)
-					((u8 *)entry + sizeof(ATOM_PPLIB_PhaseSheddingLimits_Record));
-			}
-			adev->pm.dpm.dyn_state.phase_shedding_limits_table.count =
-				psl->ucNumEntries;
-		}
-	}
-
-	/* cac data */
-	if (le16_to_cpu(power_info->pplib.usTableSize) >=
-	    sizeof(struct _ATOM_PPLIB_POWERPLAYTABLE5)) {
-		adev->pm.dpm.tdp_limit = le32_to_cpu(power_info->pplib5.ulTDPLimit);
-		adev->pm.dpm.near_tdp_limit = le32_to_cpu(power_info->pplib5.ulNearTDPLimit);
-		adev->pm.dpm.near_tdp_limit_adjusted = adev->pm.dpm.near_tdp_limit;
-		adev->pm.dpm.tdp_od_limit = le16_to_cpu(power_info->pplib5.usTDPODLimit);
-		if (adev->pm.dpm.tdp_od_limit)
-			adev->pm.dpm.power_control = true;
-		else
-			adev->pm.dpm.power_control = false;
-		adev->pm.dpm.tdp_adjustment = 0;
-		adev->pm.dpm.sq_ramping_threshold = le32_to_cpu(power_info->pplib5.ulSQRampingThreshold);
-		adev->pm.dpm.cac_leakage = le32_to_cpu(power_info->pplib5.ulCACLeakage);
-		adev->pm.dpm.load_line_slope = le16_to_cpu(power_info->pplib5.usLoadLineSlope);
-		if (power_info->pplib5.usCACLeakageTableOffset) {
-			ATOM_PPLIB_CAC_Leakage_Table *cac_table =
-				(ATOM_PPLIB_CAC_Leakage_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(power_info->pplib5.usCACLeakageTableOffset));
-			ATOM_PPLIB_CAC_Leakage_Record *entry;
-			u32 size = cac_table->ucNumEntries * sizeof(struct amdgpu_cac_leakage_table);
-			adev->pm.dpm.dyn_state.cac_leakage_table.entries = kzalloc(size, GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.cac_leakage_table.entries) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-			entry = &cac_table->entries[0];
-			for (i = 0; i < cac_table->ucNumEntries; i++) {
-				if (adev->pm.dpm.platform_caps & ATOM_PP_PLATFORM_CAP_EVV) {
-					adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc1 =
-						le16_to_cpu(entry->usVddc1);
-					adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc2 =
-						le16_to_cpu(entry->usVddc2);
-					adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc3 =
-						le16_to_cpu(entry->usVddc3);
-				} else {
-					adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].vddc =
-						le16_to_cpu(entry->usVddc);
-					adev->pm.dpm.dyn_state.cac_leakage_table.entries[i].leakage =
-						le32_to_cpu(entry->ulLeakageValue);
-				}
-				entry = (ATOM_PPLIB_CAC_Leakage_Record *)
-					((u8 *)entry + sizeof(ATOM_PPLIB_CAC_Leakage_Record));
-			}
-			adev->pm.dpm.dyn_state.cac_leakage_table.count = cac_table->ucNumEntries;
-		}
-	}
-
-	/* ext tables */
-	if (le16_to_cpu(power_info->pplib.usTableSize) >=
-	    sizeof(struct _ATOM_PPLIB_POWERPLAYTABLE3)) {
-		ATOM_PPLIB_EXTENDEDHEADER *ext_hdr = (ATOM_PPLIB_EXTENDEDHEADER *)
-			(mode_info->atom_context->bios + data_offset +
-			 le16_to_cpu(power_info->pplib3.usExtendendedHeaderOffset));
-		if ((le16_to_cpu(ext_hdr->usSize) >= SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V2) &&
-			ext_hdr->usVCETableOffset) {
-			VCEClockInfoArray *array = (VCEClockInfoArray *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usVCETableOffset) + 1);
-			ATOM_PPLIB_VCE_Clock_Voltage_Limit_Table *limits =
-				(ATOM_PPLIB_VCE_Clock_Voltage_Limit_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usVCETableOffset) + 1 +
-				 1 + array->ucNumEntries * sizeof(VCEClockInfo));
-			ATOM_PPLIB_VCE_State_Table *states =
-				(ATOM_PPLIB_VCE_State_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usVCETableOffset) + 1 +
-				 1 + (array->ucNumEntries * sizeof (VCEClockInfo)) +
-				 1 + (limits->numEntries * sizeof(ATOM_PPLIB_VCE_Clock_Voltage_Limit_Record)));
-			ATOM_PPLIB_VCE_Clock_Voltage_Limit_Record *entry;
-			ATOM_PPLIB_VCE_State_Record *state_entry;
-			VCEClockInfo *vce_clk;
-			u32 size = limits->numEntries *
-				sizeof(struct amdgpu_vce_clock_voltage_dependency_entry);
-			adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries =
-				kzalloc(size, GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-			adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.count =
-				limits->numEntries;
-			entry = &limits->entries[0];
-			state_entry = &states->entries[0];
-			for (i = 0; i < limits->numEntries; i++) {
-				vce_clk = (VCEClockInfo *)
-					((u8 *)&array->entries[0] +
-					 (entry->ucVCEClockInfoIndex * sizeof(VCEClockInfo)));
-				adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries[i].evclk =
-					le16_to_cpu(vce_clk->usEVClkLow) | (vce_clk->ucEVClkHigh << 16);
-				adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries[i].ecclk =
-					le16_to_cpu(vce_clk->usECClkLow) | (vce_clk->ucECClkHigh << 16);
-				adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table.entries[i].v =
-					le16_to_cpu(entry->usVoltage);
-				entry = (ATOM_PPLIB_VCE_Clock_Voltage_Limit_Record *)
-					((u8 *)entry + sizeof(ATOM_PPLIB_VCE_Clock_Voltage_Limit_Record));
-			}
-			adev->pm.dpm.num_of_vce_states =
-					states->numEntries > AMD_MAX_VCE_LEVELS ?
-					AMD_MAX_VCE_LEVELS : states->numEntries;
-			for (i = 0; i < adev->pm.dpm.num_of_vce_states; i++) {
-				vce_clk = (VCEClockInfo *)
-					((u8 *)&array->entries[0] +
-					 (state_entry->ucVCEClockInfoIndex * sizeof(VCEClockInfo)));
-				adev->pm.dpm.vce_states[i].evclk =
-					le16_to_cpu(vce_clk->usEVClkLow) | (vce_clk->ucEVClkHigh << 16);
-				adev->pm.dpm.vce_states[i].ecclk =
-					le16_to_cpu(vce_clk->usECClkLow) | (vce_clk->ucECClkHigh << 16);
-				adev->pm.dpm.vce_states[i].clk_idx =
-					state_entry->ucClockInfoIndex & 0x3f;
-				adev->pm.dpm.vce_states[i].pstate =
-					(state_entry->ucClockInfoIndex & 0xc0) >> 6;
-				state_entry = (ATOM_PPLIB_VCE_State_Record *)
-					((u8 *)state_entry + sizeof(ATOM_PPLIB_VCE_State_Record));
-			}
-		}
-		if ((le16_to_cpu(ext_hdr->usSize) >= SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V3) &&
-			ext_hdr->usUVDTableOffset) {
-			UVDClockInfoArray *array = (UVDClockInfoArray *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usUVDTableOffset) + 1);
-			ATOM_PPLIB_UVD_Clock_Voltage_Limit_Table *limits =
-				(ATOM_PPLIB_UVD_Clock_Voltage_Limit_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usUVDTableOffset) + 1 +
-				 1 + (array->ucNumEntries * sizeof (UVDClockInfo)));
-			ATOM_PPLIB_UVD_Clock_Voltage_Limit_Record *entry;
-			u32 size = limits->numEntries *
-				sizeof(struct amdgpu_uvd_clock_voltage_dependency_entry);
-			adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries =
-				kzalloc(size, GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-			adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.count =
-				limits->numEntries;
-			entry = &limits->entries[0];
-			for (i = 0; i < limits->numEntries; i++) {
-				UVDClockInfo *uvd_clk = (UVDClockInfo *)
-					((u8 *)&array->entries[0] +
-					 (entry->ucUVDClockInfoIndex * sizeof(UVDClockInfo)));
-				adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries[i].vclk =
-					le16_to_cpu(uvd_clk->usVClkLow) | (uvd_clk->ucVClkHigh << 16);
-				adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries[i].dclk =
-					le16_to_cpu(uvd_clk->usDClkLow) | (uvd_clk->ucDClkHigh << 16);
-				adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table.entries[i].v =
-					le16_to_cpu(entry->usVoltage);
-				entry = (ATOM_PPLIB_UVD_Clock_Voltage_Limit_Record *)
-					((u8 *)entry + sizeof(ATOM_PPLIB_UVD_Clock_Voltage_Limit_Record));
-			}
-		}
-		if ((le16_to_cpu(ext_hdr->usSize) >= SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V4) &&
-			ext_hdr->usSAMUTableOffset) {
-			ATOM_PPLIB_SAMClk_Voltage_Limit_Table *limits =
-				(ATOM_PPLIB_SAMClk_Voltage_Limit_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usSAMUTableOffset) + 1);
-			ATOM_PPLIB_SAMClk_Voltage_Limit_Record *entry;
-			u32 size = limits->numEntries *
-				sizeof(struct amdgpu_clock_voltage_dependency_entry);
-			adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.entries =
-				kzalloc(size, GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.entries) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-			adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.count =
-				limits->numEntries;
-			entry = &limits->entries[0];
-			for (i = 0; i < limits->numEntries; i++) {
-				adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.entries[i].clk =
-					le16_to_cpu(entry->usSAMClockLow) | (entry->ucSAMClockHigh << 16);
-				adev->pm.dpm.dyn_state.samu_clock_voltage_dependency_table.entries[i].v =
-					le16_to_cpu(entry->usVoltage);
-				entry = (ATOM_PPLIB_SAMClk_Voltage_Limit_Record *)
-					((u8 *)entry + sizeof(ATOM_PPLIB_SAMClk_Voltage_Limit_Record));
-			}
-		}
-		if ((le16_to_cpu(ext_hdr->usSize) >= SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V5) &&
-		    ext_hdr->usPPMTableOffset) {
-			ATOM_PPLIB_PPM_Table *ppm = (ATOM_PPLIB_PPM_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usPPMTableOffset));
-			adev->pm.dpm.dyn_state.ppm_table =
-				kzalloc(sizeof(struct amdgpu_ppm_table), GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.ppm_table) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-			adev->pm.dpm.dyn_state.ppm_table->ppm_design = ppm->ucPpmDesign;
-			adev->pm.dpm.dyn_state.ppm_table->cpu_core_number =
-				le16_to_cpu(ppm->usCpuCoreNumber);
-			adev->pm.dpm.dyn_state.ppm_table->platform_tdp =
-				le32_to_cpu(ppm->ulPlatformTDP);
-			adev->pm.dpm.dyn_state.ppm_table->small_ac_platform_tdp =
-				le32_to_cpu(ppm->ulSmallACPlatformTDP);
-			adev->pm.dpm.dyn_state.ppm_table->platform_tdc =
-				le32_to_cpu(ppm->ulPlatformTDC);
-			adev->pm.dpm.dyn_state.ppm_table->small_ac_platform_tdc =
-				le32_to_cpu(ppm->ulSmallACPlatformTDC);
-			adev->pm.dpm.dyn_state.ppm_table->apu_tdp =
-				le32_to_cpu(ppm->ulApuTDP);
-			adev->pm.dpm.dyn_state.ppm_table->dgpu_tdp =
-				le32_to_cpu(ppm->ulDGpuTDP);
-			adev->pm.dpm.dyn_state.ppm_table->dgpu_ulv_power =
-				le32_to_cpu(ppm->ulDGpuUlvPower);
-			adev->pm.dpm.dyn_state.ppm_table->tj_max =
-				le32_to_cpu(ppm->ulTjmax);
-		}
-		if ((le16_to_cpu(ext_hdr->usSize) >= SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V6) &&
-			ext_hdr->usACPTableOffset) {
-			ATOM_PPLIB_ACPClk_Voltage_Limit_Table *limits =
-				(ATOM_PPLIB_ACPClk_Voltage_Limit_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usACPTableOffset) + 1);
-			ATOM_PPLIB_ACPClk_Voltage_Limit_Record *entry;
-			u32 size = limits->numEntries *
-				sizeof(struct amdgpu_clock_voltage_dependency_entry);
-			adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.entries =
-				kzalloc(size, GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.entries) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-			adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.count =
-				limits->numEntries;
-			entry = &limits->entries[0];
-			for (i = 0; i < limits->numEntries; i++) {
-				adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.entries[i].clk =
-					le16_to_cpu(entry->usACPClockLow) | (entry->ucACPClockHigh << 16);
-				adev->pm.dpm.dyn_state.acp_clock_voltage_dependency_table.entries[i].v =
-					le16_to_cpu(entry->usVoltage);
-				entry = (ATOM_PPLIB_ACPClk_Voltage_Limit_Record *)
-					((u8 *)entry + sizeof(ATOM_PPLIB_ACPClk_Voltage_Limit_Record));
-			}
-		}
-		if ((le16_to_cpu(ext_hdr->usSize) >= SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V7) &&
-			ext_hdr->usPowerTuneTableOffset) {
-			u8 rev = *(u8 *)(mode_info->atom_context->bios + data_offset +
-					 le16_to_cpu(ext_hdr->usPowerTuneTableOffset));
-			ATOM_PowerTune_Table *pt;
-			adev->pm.dpm.dyn_state.cac_tdp_table =
-				kzalloc(sizeof(struct amdgpu_cac_tdp_table), GFP_KERNEL);
-			if (!adev->pm.dpm.dyn_state.cac_tdp_table) {
-				amdgpu_free_extended_power_table(adev);
-				return -ENOMEM;
-			}
-			if (rev > 0) {
-				ATOM_PPLIB_POWERTUNE_Table_V1 *ppt = (ATOM_PPLIB_POWERTUNE_Table_V1 *)
-					(mode_info->atom_context->bios + data_offset +
-					 le16_to_cpu(ext_hdr->usPowerTuneTableOffset));
-				adev->pm.dpm.dyn_state.cac_tdp_table->maximum_power_delivery_limit =
-					ppt->usMaximumPowerDeliveryLimit;
-				pt = &ppt->power_tune_table;
-			} else {
-				ATOM_PPLIB_POWERTUNE_Table *ppt = (ATOM_PPLIB_POWERTUNE_Table *)
-					(mode_info->atom_context->bios + data_offset +
-					 le16_to_cpu(ext_hdr->usPowerTuneTableOffset));
-				adev->pm.dpm.dyn_state.cac_tdp_table->maximum_power_delivery_limit = 255;
-				pt = &ppt->power_tune_table;
-			}
-			adev->pm.dpm.dyn_state.cac_tdp_table->tdp = le16_to_cpu(pt->usTDP);
-			adev->pm.dpm.dyn_state.cac_tdp_table->configurable_tdp =
-				le16_to_cpu(pt->usConfigurableTDP);
-			adev->pm.dpm.dyn_state.cac_tdp_table->tdc = le16_to_cpu(pt->usTDC);
-			adev->pm.dpm.dyn_state.cac_tdp_table->battery_power_limit =
-				le16_to_cpu(pt->usBatteryPowerLimit);
-			adev->pm.dpm.dyn_state.cac_tdp_table->small_power_limit =
-				le16_to_cpu(pt->usSmallPowerLimit);
-			adev->pm.dpm.dyn_state.cac_tdp_table->low_cac_leakage =
-				le16_to_cpu(pt->usLowCACLeakage);
-			adev->pm.dpm.dyn_state.cac_tdp_table->high_cac_leakage =
-				le16_to_cpu(pt->usHighCACLeakage);
-		}
-		if ((le16_to_cpu(ext_hdr->usSize) >= SIZE_OF_ATOM_PPLIB_EXTENDEDHEADER_V8) &&
-				ext_hdr->usSclkVddgfxTableOffset) {
-			dep_table = (ATOM_PPLIB_Clock_Voltage_Dependency_Table *)
-				(mode_info->atom_context->bios + data_offset +
-				 le16_to_cpu(ext_hdr->usSclkVddgfxTableOffset));
-			ret = amdgpu_parse_clk_voltage_dep_table(
-					&adev->pm.dpm.dyn_state.vddgfx_dependency_on_sclk,
-					dep_table);
-			if (ret) {
-				kfree(adev->pm.dpm.dyn_state.vddgfx_dependency_on_sclk.entries);
-				return ret;
-			}
-		}
-	}
-
-	return 0;
-}
-
-void amdgpu_free_extended_power_table(struct amdgpu_device *adev)
-{
-	struct amdgpu_dpm_dynamic_state *dyn_state = &adev->pm.dpm.dyn_state;
-
-	kfree(dyn_state->vddc_dependency_on_sclk.entries);
-	kfree(dyn_state->vddci_dependency_on_mclk.entries);
-	kfree(dyn_state->vddc_dependency_on_mclk.entries);
-	kfree(dyn_state->mvdd_dependency_on_mclk.entries);
-	kfree(dyn_state->cac_leakage_table.entries);
-	kfree(dyn_state->phase_shedding_limits_table.entries);
-	kfree(dyn_state->ppm_table);
-	kfree(dyn_state->cac_tdp_table);
-	kfree(dyn_state->vce_clock_voltage_dependency_table.entries);
-	kfree(dyn_state->uvd_clock_voltage_dependency_table.entries);
-	kfree(dyn_state->samu_clock_voltage_dependency_table.entries);
-	kfree(dyn_state->acp_clock_voltage_dependency_table.entries);
-	kfree(dyn_state->vddgfx_dependency_on_sclk.entries);
-}
-
-static const char *pp_lib_thermal_controller_names[] = {
-	"NONE",
-	"lm63",
-	"adm1032",
-	"adm1030",
-	"max6649",
-	"lm64",
-	"f75375",
-	"RV6xx",
-	"RV770",
-	"adt7473",
-	"NONE",
-	"External GPIO",
-	"Evergreen",
-	"emc2103",
-	"Sumo",
-	"Northern Islands",
-	"Southern Islands",
-	"lm96163",
-	"Sea Islands",
-	"Kaveri/Kabini",
-};
-
-void amdgpu_add_thermal_controller(struct amdgpu_device *adev)
-{
-	struct amdgpu_mode_info *mode_info = &adev->mode_info;
-	ATOM_PPLIB_POWERPLAYTABLE *power_table;
-	int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo);
-	ATOM_PPLIB_THERMALCONTROLLER *controller;
-	struct amdgpu_i2c_bus_rec i2c_bus;
-	u16 data_offset;
-	u8 frev, crev;
-
-	if (!amdgpu_atom_parse_data_header(mode_info->atom_context, index, NULL,
-				   &frev, &crev, &data_offset))
-		return;
-	power_table = (ATOM_PPLIB_POWERPLAYTABLE *)
-		(mode_info->atom_context->bios + data_offset);
-	controller = &power_table->sThermalController;
-
-	/* add the i2c bus for thermal/fan chip */
-	if (controller->ucType > 0) {
-		if (controller->ucFanParameters & ATOM_PP_FANPARAMETERS_NOFAN)
-			adev->pm.no_fan = true;
-		adev->pm.fan_pulses_per_revolution =
-			controller->ucFanParameters & ATOM_PP_FANPARAMETERS_TACHOMETER_PULSES_PER_REVOLUTION_MASK;
-		if (adev->pm.fan_pulses_per_revolution) {
-			adev->pm.fan_min_rpm = controller->ucFanMinRPM;
-			adev->pm.fan_max_rpm = controller->ucFanMaxRPM;
-		}
-		if (controller->ucType == ATOM_PP_THERMALCONTROLLER_RV6xx) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_RV6XX;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_RV770) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_RV770;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_EVERGREEN) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_EVERGREEN;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_SUMO) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_SUMO;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_NISLANDS) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_NI;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_SISLANDS) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_SI;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_CISLANDS) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_CI;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_KAVERI) {
-			DRM_INFO("Internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_KV;
-		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_EXTERNAL_GPIO) {
-			DRM_INFO("External GPIO thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_EXTERNAL_GPIO;
-		} else if (controller->ucType ==
-			   ATOM_PP_THERMALCONTROLLER_ADT7473_WITH_INTERNAL) {
-			DRM_INFO("ADT7473 with internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_ADT7473_WITH_INTERNAL;
-		} else if (controller->ucType ==
-			   ATOM_PP_THERMALCONTROLLER_EMC2103_WITH_INTERNAL) {
-			DRM_INFO("EMC2103 with internal thermal controller %s fan control\n",
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_EMC2103_WITH_INTERNAL;
-		} else if (controller->ucType < ARRAY_SIZE(pp_lib_thermal_controller_names)) {
-			DRM_INFO("Possible %s thermal controller at 0x%02x %s fan control\n",
-				 pp_lib_thermal_controller_names[controller->ucType],
-				 controller->ucI2cAddress >> 1,
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-			adev->pm.int_thermal_type = THERMAL_TYPE_EXTERNAL;
-			i2c_bus = amdgpu_atombios_lookup_i2c_gpio(adev, controller->ucI2cLine);
-			adev->pm.i2c_bus = amdgpu_i2c_lookup(adev, &i2c_bus);
-			if (adev->pm.i2c_bus) {
-				struct i2c_board_info info = { };
-				const char *name = pp_lib_thermal_controller_names[controller->ucType];
-				info.addr = controller->ucI2cAddress >> 1;
-				strlcpy(info.type, name, sizeof(info.type));
-				i2c_new_device(&adev->pm.i2c_bus->adapter, &info);
-			}
-		} else {
-			DRM_INFO("Unknown thermal controller type %d at 0x%02x %s fan control\n",
-				 controller->ucType,
-				 controller->ucI2cAddress >> 1,
-				 (controller->ucFanParameters &
-				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-		}
-	}
-}
-
-enum amdgpu_pcie_gen amdgpu_get_pcie_gen_support(struct amdgpu_device *adev,
-						 u32 sys_mask,
-						 enum amdgpu_pcie_gen asic_gen,
-						 enum amdgpu_pcie_gen default_gen)
-{
-	switch (asic_gen) {
-	case AMDGPU_PCIE_GEN1:
-		return AMDGPU_PCIE_GEN1;
-	case AMDGPU_PCIE_GEN2:
-		return AMDGPU_PCIE_GEN2;
-	case AMDGPU_PCIE_GEN3:
-		return AMDGPU_PCIE_GEN3;
-	default:
-		if ((sys_mask & DRM_PCIE_SPEED_80) && (default_gen == AMDGPU_PCIE_GEN3))
-			return AMDGPU_PCIE_GEN3;
-		else if ((sys_mask & DRM_PCIE_SPEED_50) && (default_gen == AMDGPU_PCIE_GEN2))
-			return AMDGPU_PCIE_GEN2;
-		else
-			return AMDGPU_PCIE_GEN1;
-	}
-	return AMDGPU_PCIE_GEN1;
-}
-
-u16 amdgpu_get_pcie_lane_support(struct amdgpu_device *adev,
-				 u16 asic_lanes,
-				 u16 default_lanes)
-{
-	switch (asic_lanes) {
-	case 0:
-	default:
-		return default_lanes;
-	case 1:
-		return 1;
-	case 2:
-		return 2;
-	case 4:
-		return 4;
-	case 8:
-		return 8;
-	case 12:
-		return 12;
-	case 16:
-		return 16;
-	}
-}
-
-u8 amdgpu_encode_pci_lane_width(u32 lanes)
-{
-	u8 encoded_lanes[] = { 0, 1, 2, 0, 3, 0, 0, 0, 4, 0, 0, 0, 5, 0, 0, 0, 6 };
-
-	if (lanes > 16)
-		return 0;
-
-	return encoded_lanes[lanes];
-}
-
-struct amd_vce_state*
-amdgpu_get_vce_clock_state(struct amdgpu_device *adev, unsigned idx)
-{
-	if (idx < adev->pm.dpm.num_of_vce_states)
-		return &adev->pm.dpm.vce_states[idx];
-
-	return NULL;
-}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h
deleted file mode 100644
index 8c96a4caa715..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h
+++ /dev/null
@@ -1,556 +0,0 @@
-/*
- * Copyright 2014 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- */
-#ifndef __AMDGPU_DPM_H__
-#define __AMDGPU_DPM_H__
-
-enum amdgpu_int_thermal_type {
-	THERMAL_TYPE_NONE,
-	THERMAL_TYPE_EXTERNAL,
-	THERMAL_TYPE_EXTERNAL_GPIO,
-	THERMAL_TYPE_RV6XX,
-	THERMAL_TYPE_RV770,
-	THERMAL_TYPE_ADT7473_WITH_INTERNAL,
-	THERMAL_TYPE_EVERGREEN,
-	THERMAL_TYPE_SUMO,
-	THERMAL_TYPE_NI,
-	THERMAL_TYPE_SI,
-	THERMAL_TYPE_EMC2103_WITH_INTERNAL,
-	THERMAL_TYPE_CI,
-	THERMAL_TYPE_KV,
-};
-
-enum amdgpu_dpm_auto_throttle_src {
-	AMDGPU_DPM_AUTO_THROTTLE_SRC_THERMAL,
-	AMDGPU_DPM_AUTO_THROTTLE_SRC_EXTERNAL
-};
-
-enum amdgpu_dpm_event_src {
-	AMDGPU_DPM_EVENT_SRC_ANALOG = 0,
-	AMDGPU_DPM_EVENT_SRC_EXTERNAL = 1,
-	AMDGPU_DPM_EVENT_SRC_DIGITAL = 2,
-	AMDGPU_DPM_EVENT_SRC_ANALOG_OR_EXTERNAL = 3,
-	AMDGPU_DPM_EVENT_SRC_DIGIAL_OR_EXTERNAL = 4
-};
-
-#define SCLK_DEEP_SLEEP_MASK 0x8
-
-struct amdgpu_ps {
-	u32 caps; /* vbios flags */
-	u32 class; /* vbios flags */
-	u32 class2; /* vbios flags */
-	/* UVD clocks */
-	u32 vclk;
-	u32 dclk;
-	/* VCE clocks */
-	u32 evclk;
-	u32 ecclk;
-	bool vce_active;
-	enum amd_vce_level vce_level;
-	/* asic priv */
-	void *ps_priv;
-};
-
-struct amdgpu_dpm_thermal {
-	/* thermal interrupt work */
-	struct work_struct work;
-	/* low temperature threshold */
-	int                min_temp;
-	/* high temperature threshold */
-	int                max_temp;
-	/* was last interrupt low to high or high to low */
-	bool               high_to_low;
-	/* interrupt source */
-	struct amdgpu_irq_src	irq;
-};
-
-enum amdgpu_clk_action
-{
-	AMDGPU_SCLK_UP = 1,
-	AMDGPU_SCLK_DOWN
-};
-
-struct amdgpu_blacklist_clocks
-{
-	u32 sclk;
-	u32 mclk;
-	enum amdgpu_clk_action action;
-};
-
-struct amdgpu_clock_and_voltage_limits {
-	u32 sclk;
-	u32 mclk;
-	u16 vddc;
-	u16 vddci;
-};
-
-struct amdgpu_clock_array {
-	u32 count;
-	u32 *values;
-};
-
-struct amdgpu_clock_voltage_dependency_entry {
-	u32 clk;
-	u16 v;
-};
-
-struct amdgpu_clock_voltage_dependency_table {
-	u32 count;
-	struct amdgpu_clock_voltage_dependency_entry *entries;
-};
-
-union amdgpu_cac_leakage_entry {
-	struct {
-		u16 vddc;
-		u32 leakage;
-	};
-	struct {
-		u16 vddc1;
-		u16 vddc2;
-		u16 vddc3;
-	};
-};
-
-struct amdgpu_cac_leakage_table {
-	u32 count;
-	union amdgpu_cac_leakage_entry *entries;
-};
-
-struct amdgpu_phase_shedding_limits_entry {
-	u16 voltage;
-	u32 sclk;
-	u32 mclk;
-};
-
-struct amdgpu_phase_shedding_limits_table {
-	u32 count;
-	struct amdgpu_phase_shedding_limits_entry *entries;
-};
-
-struct amdgpu_uvd_clock_voltage_dependency_entry {
-	u32 vclk;
-	u32 dclk;
-	u16 v;
-};
-
-struct amdgpu_uvd_clock_voltage_dependency_table {
-	u8 count;
-	struct amdgpu_uvd_clock_voltage_dependency_entry *entries;
-};
-
-struct amdgpu_vce_clock_voltage_dependency_entry {
-	u32 ecclk;
-	u32 evclk;
-	u16 v;
-};
-
-struct amdgpu_vce_clock_voltage_dependency_table {
-	u8 count;
-	struct amdgpu_vce_clock_voltage_dependency_entry *entries;
-};
-
-struct amdgpu_ppm_table {
-	u8 ppm_design;
-	u16 cpu_core_number;
-	u32 platform_tdp;
-	u32 small_ac_platform_tdp;
-	u32 platform_tdc;
-	u32 small_ac_platform_tdc;
-	u32 apu_tdp;
-	u32 dgpu_tdp;
-	u32 dgpu_ulv_power;
-	u32 tj_max;
-};
-
-struct amdgpu_cac_tdp_table {
-	u16 tdp;
-	u16 configurable_tdp;
-	u16 tdc;
-	u16 battery_power_limit;
-	u16 small_power_limit;
-	u16 low_cac_leakage;
-	u16 high_cac_leakage;
-	u16 maximum_power_delivery_limit;
-};
-
-struct amdgpu_dpm_dynamic_state {
-	struct amdgpu_clock_voltage_dependency_table vddc_dependency_on_sclk;
-	struct amdgpu_clock_voltage_dependency_table vddci_dependency_on_mclk;
-	struct amdgpu_clock_voltage_dependency_table vddc_dependency_on_mclk;
-	struct amdgpu_clock_voltage_dependency_table mvdd_dependency_on_mclk;
-	struct amdgpu_clock_voltage_dependency_table vddc_dependency_on_dispclk;
-	struct amdgpu_uvd_clock_voltage_dependency_table uvd_clock_voltage_dependency_table;
-	struct amdgpu_vce_clock_voltage_dependency_table vce_clock_voltage_dependency_table;
-	struct amdgpu_clock_voltage_dependency_table samu_clock_voltage_dependency_table;
-	struct amdgpu_clock_voltage_dependency_table acp_clock_voltage_dependency_table;
-	struct amdgpu_clock_voltage_dependency_table vddgfx_dependency_on_sclk;
-	struct amdgpu_clock_array valid_sclk_values;
-	struct amdgpu_clock_array valid_mclk_values;
-	struct amdgpu_clock_and_voltage_limits max_clock_voltage_on_dc;
-	struct amdgpu_clock_and_voltage_limits max_clock_voltage_on_ac;
-	u32 mclk_sclk_ratio;
-	u32 sclk_mclk_delta;
-	u16 vddc_vddci_delta;
-	u16 min_vddc_for_pcie_gen2;
-	struct amdgpu_cac_leakage_table cac_leakage_table;
-	struct amdgpu_phase_shedding_limits_table phase_shedding_limits_table;
-	struct amdgpu_ppm_table *ppm_table;
-	struct amdgpu_cac_tdp_table *cac_tdp_table;
-};
-
-struct amdgpu_dpm_fan {
-	u16 t_min;
-	u16 t_med;
-	u16 t_high;
-	u16 pwm_min;
-	u16 pwm_med;
-	u16 pwm_high;
-	u8 t_hyst;
-	u32 cycle_delay;
-	u16 t_max;
-	u8 control_mode;
-	u16 default_max_fan_pwm;
-	u16 default_fan_output_sensitivity;
-	u16 fan_output_sensitivity;
-	bool ucode_fan_control;
-};
-
-enum amdgpu_pcie_gen {
-	AMDGPU_PCIE_GEN1 = 0,
-	AMDGPU_PCIE_GEN2 = 1,
-	AMDGPU_PCIE_GEN3 = 2,
-	AMDGPU_PCIE_GEN_INVALID = 0xffff
-};
-
-struct amdgpu_dpm_funcs {
-	int (*get_temperature)(struct amdgpu_device *adev);
-	int (*pre_set_power_state)(struct amdgpu_device *adev);
-	int (*set_power_state)(struct amdgpu_device *adev);
-	void (*post_set_power_state)(struct amdgpu_device *adev);
-	void (*display_configuration_changed)(struct amdgpu_device *adev);
-	u32 (*get_sclk)(struct amdgpu_device *adev, bool low);
-	u32 (*get_mclk)(struct amdgpu_device *adev, bool low);
-	void (*print_power_state)(struct amdgpu_device *adev, struct amdgpu_ps *ps);
-	void (*debugfs_print_current_performance_level)(struct amdgpu_device *adev, struct seq_file *m);
-	int (*force_performance_level)(struct amdgpu_device *adev, enum amd_dpm_forced_level level);
-	bool (*vblank_too_short)(struct amdgpu_device *adev);
-	void (*powergate_uvd)(struct amdgpu_device *adev, bool gate);
-	void (*powergate_vce)(struct amdgpu_device *adev, bool gate);
-	void (*enable_bapm)(struct amdgpu_device *adev, bool enable);
-	void (*set_fan_control_mode)(struct amdgpu_device *adev, u32 mode);
-	u32 (*get_fan_control_mode)(struct amdgpu_device *adev);
-	int (*set_fan_speed_percent)(struct amdgpu_device *adev, u32 speed);
-	int (*get_fan_speed_percent)(struct amdgpu_device *adev, u32 *speed);
-	int (*force_clock_level)(struct amdgpu_device *adev, enum pp_clock_type type, uint32_t mask);
-	int (*print_clock_levels)(struct amdgpu_device *adev, enum pp_clock_type type, char *buf);
-	int (*get_sclk_od)(struct amdgpu_device *adev);
-	int (*set_sclk_od)(struct amdgpu_device *adev, uint32_t value);
-	int (*get_mclk_od)(struct amdgpu_device *adev);
-	int (*set_mclk_od)(struct amdgpu_device *adev, uint32_t value);
-	int (*check_state_equal)(struct amdgpu_device *adev,
-				struct amdgpu_ps *cps,
-				struct amdgpu_ps *rps,
-				bool *equal);
-	int (*read_sensor)(struct amdgpu_device *adev, int idx, void *value,
-			   int *size);
-
-	struct amd_vce_state* (*get_vce_clock_state)(struct amdgpu_device *adev, unsigned idx);
-	int (*reset_power_profile_state)(struct amdgpu_device *adev,
-			struct amd_pp_profile *request);
-	int (*get_power_profile_state)(struct amdgpu_device *adev,
-			struct amd_pp_profile *query);
-	int (*set_power_profile_state)(struct amdgpu_device *adev,
-			struct amd_pp_profile *request);
-	int (*switch_power_profile)(struct amdgpu_device *adev,
-			enum amd_pp_profile_type type);
-};
-
-#define amdgpu_dpm_pre_set_power_state(adev) (adev)->pm.funcs->pre_set_power_state((adev))
-#define amdgpu_dpm_set_power_state(adev) (adev)->pm.funcs->set_power_state((adev))
-#define amdgpu_dpm_post_set_power_state(adev) (adev)->pm.funcs->post_set_power_state((adev))
-#define amdgpu_dpm_display_configuration_changed(adev) (adev)->pm.funcs->display_configuration_changed((adev))
-#define amdgpu_dpm_print_power_state(adev, ps) (adev)->pm.funcs->print_power_state((adev), (ps))
-#define amdgpu_dpm_vblank_too_short(adev) (adev)->pm.funcs->vblank_too_short((adev))
-#define amdgpu_dpm_enable_bapm(adev, e) (adev)->pm.funcs->enable_bapm((adev), (e))
-
-#define amdgpu_dpm_read_sensor(adev, idx, value, size) \
-	((adev)->pp_enabled ? \
-		(adev)->powerplay.pp_funcs->read_sensor(adev->powerplay.pp_handle, (idx), (value), (size)) : \
-		(adev)->pm.funcs->read_sensor((adev), (idx), (value), (size)))
-
-#define amdgpu_dpm_get_temperature(adev) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->get_temperature((adev)->powerplay.pp_handle) : \
-	      (adev)->pm.funcs->get_temperature((adev)))
-
-#define amdgpu_dpm_set_fan_control_mode(adev, m) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->set_fan_control_mode((adev)->powerplay.pp_handle, (m)) : \
-	      (adev)->pm.funcs->set_fan_control_mode((adev), (m)))
-
-#define amdgpu_dpm_get_fan_control_mode(adev) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->get_fan_control_mode((adev)->powerplay.pp_handle) : \
-	      (adev)->pm.funcs->get_fan_control_mode((adev)))
-
-#define amdgpu_dpm_set_fan_speed_percent(adev, s) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->set_fan_speed_percent((adev)->powerplay.pp_handle, (s)) : \
-	      (adev)->pm.funcs->set_fan_speed_percent((adev), (s)))
-
-#define amdgpu_dpm_get_fan_speed_percent(adev, s) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->get_fan_speed_percent((adev)->powerplay.pp_handle, (s)) : \
-	      (adev)->pm.funcs->get_fan_speed_percent((adev), (s)))
-
-#define amdgpu_dpm_get_fan_speed_rpm(adev, s) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->get_fan_speed_rpm((adev)->powerplay.pp_handle, (s)) : \
-	      -EINVAL)
-
-#define amdgpu_dpm_get_sclk(adev, l) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->get_sclk((adev)->powerplay.pp_handle, (l)) : \
-		(adev)->pm.funcs->get_sclk((adev), (l)))
-
-#define amdgpu_dpm_get_mclk(adev, l)  \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->get_mclk((adev)->powerplay.pp_handle, (l)) : \
-	      (adev)->pm.funcs->get_mclk((adev), (l)))
-
-
-#define amdgpu_dpm_force_performance_level(adev, l) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->force_performance_level((adev)->powerplay.pp_handle, (l)) : \
-	      (adev)->pm.funcs->force_performance_level((adev), (l)))
-
-#define amdgpu_dpm_powergate_uvd(adev, g) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->powergate_uvd((adev)->powerplay.pp_handle, (g)) : \
-	      (adev)->pm.funcs->powergate_uvd((adev), (g)))
-
-#define amdgpu_dpm_powergate_vce(adev, g) \
-	((adev)->pp_enabled ?						\
-	      (adev)->powerplay.pp_funcs->powergate_vce((adev)->powerplay.pp_handle, (g)) : \
-	      (adev)->pm.funcs->powergate_vce((adev), (g)))
-
-#define amdgpu_dpm_get_current_power_state(adev) \
-	(adev)->powerplay.pp_funcs->get_current_power_state((adev)->powerplay.pp_handle)
-
-#define amdgpu_dpm_get_pp_num_states(adev, data) \
-	(adev)->powerplay.pp_funcs->get_pp_num_states((adev)->powerplay.pp_handle, data)
-
-#define amdgpu_dpm_get_pp_table(adev, table) \
-	(adev)->powerplay.pp_funcs->get_pp_table((adev)->powerplay.pp_handle, table)
-
-#define amdgpu_dpm_set_pp_table(adev, buf, size) \
-	(adev)->powerplay.pp_funcs->set_pp_table((adev)->powerplay.pp_handle, buf, size)
-
-#define amdgpu_dpm_print_clock_levels(adev, type, buf) \
-	(adev)->powerplay.pp_funcs->print_clock_levels((adev)->powerplay.pp_handle, type, buf)
-
-#define amdgpu_dpm_force_clock_level(adev, type, level) \
-		(adev)->powerplay.pp_funcs->force_clock_level((adev)->powerplay.pp_handle, type, level)
-
-#define amdgpu_dpm_get_sclk_od(adev) \
-	(adev)->powerplay.pp_funcs->get_sclk_od((adev)->powerplay.pp_handle)
-
-#define amdgpu_dpm_set_sclk_od(adev, value) \
-	(adev)->powerplay.pp_funcs->set_sclk_od((adev)->powerplay.pp_handle, value)
-
-#define amdgpu_dpm_get_mclk_od(adev) \
-	((adev)->powerplay.pp_funcs->get_mclk_od((adev)->powerplay.pp_handle))
-
-#define amdgpu_dpm_set_mclk_od(adev, value) \
-	((adev)->powerplay.pp_funcs->set_mclk_od((adev)->powerplay.pp_handle, value))
-
-#define amdgpu_dpm_dispatch_task(adev, event_id, input, output)		\
-	(adev)->powerplay.pp_funcs->dispatch_tasks((adev)->powerplay.pp_handle, (event_id), (input), (output))
-
-#define amgdpu_dpm_check_state_equal(adev, cps, rps, equal) (adev)->pm.funcs->check_state_equal((adev), (cps),(rps),(equal))
-
-#define amdgpu_dpm_get_vce_clock_state(adev, i)				\
-	((adev)->pp_enabled ?						\
-	 (adev)->powerplay.pp_funcs->get_vce_clock_state((adev)->powerplay.pp_handle, (i)) : \
-	 (adev)->pm.funcs->get_vce_clock_state((adev), (i)))
-
-#define amdgpu_dpm_get_performance_level(adev) \
-	((adev)->pp_enabled ?						\
-	(adev)->powerplay.pp_funcs->get_performance_level((adev)->powerplay.pp_handle) : \
-	(adev)->pm.dpm.forced_level)
-
-#define amdgpu_dpm_reset_power_profile_state(adev, request) \
-	((adev)->powerplay.pp_funcs->reset_power_profile_state(\
-			(adev)->powerplay.pp_handle, request))
-
-#define amdgpu_dpm_get_power_profile_state(adev, query) \
-	((adev)->powerplay.pp_funcs->get_power_profile_state(\
-			(adev)->powerplay.pp_handle, query))
-
-#define amdgpu_dpm_set_power_profile_state(adev, request) \
-	((adev)->powerplay.pp_funcs->set_power_profile_state(\
-			(adev)->powerplay.pp_handle, request))
-
-#define amdgpu_dpm_switch_power_profile(adev, type) \
-	((adev)->powerplay.pp_funcs->switch_power_profile(\
-			(adev)->powerplay.pp_handle, type))
-
-struct amdgpu_dpm {
-	struct amdgpu_ps        *ps;
-	/* number of valid power states */
-	int                     num_ps;
-	/* current power state that is active */
-	struct amdgpu_ps        *current_ps;
-	/* requested power state */
-	struct amdgpu_ps        *requested_ps;
-	/* boot up power state */
-	struct amdgpu_ps        *boot_ps;
-	/* default uvd power state */
-	struct amdgpu_ps        *uvd_ps;
-	/* vce requirements */
-	u32                  num_of_vce_states;
-	struct amd_vce_state vce_states[AMD_MAX_VCE_LEVELS];
-	enum amd_vce_level vce_level;
-	enum amd_pm_state_type state;
-	enum amd_pm_state_type user_state;
-	enum amd_pm_state_type last_state;
-	enum amd_pm_state_type last_user_state;
-	u32                     platform_caps;
-	u32                     voltage_response_time;
-	u32                     backbias_response_time;
-	void                    *priv;
-	u32			new_active_crtcs;
-	int			new_active_crtc_count;
-	u32			current_active_crtcs;
-	int			current_active_crtc_count;
-	struct amdgpu_dpm_dynamic_state dyn_state;
-	struct amdgpu_dpm_fan fan;
-	u32 tdp_limit;
-	u32 near_tdp_limit;
-	u32 near_tdp_limit_adjusted;
-	u32 sq_ramping_threshold;
-	u32 cac_leakage;
-	u16 tdp_od_limit;
-	u32 tdp_adjustment;
-	u16 load_line_slope;
-	bool power_control;
-	bool ac_power;
-	/* special states active */
-	bool                    thermal_active;
-	bool                    uvd_active;
-	bool                    vce_active;
-	/* thermal handling */
-	struct amdgpu_dpm_thermal thermal;
-	/* forced levels */
-	enum amd_dpm_forced_level forced_level;
-};
-
-struct amdgpu_pm {
-	struct mutex		mutex;
-	u32                     current_sclk;
-	u32                     current_mclk;
-	u32                     default_sclk;
-	u32                     default_mclk;
-	struct amdgpu_i2c_chan *i2c_bus;
-	/* internal thermal controller on rv6xx+ */
-	enum amdgpu_int_thermal_type int_thermal_type;
-	struct device	        *int_hwmon_dev;
-	/* fan control parameters */
-	bool                    no_fan;
-	u8                      fan_pulses_per_revolution;
-	u8                      fan_min_rpm;
-	u8                      fan_max_rpm;
-	/* dpm */
-	bool                    dpm_enabled;
-	bool                    sysfs_initialized;
-	struct amdgpu_dpm       dpm;
-	const struct firmware	*fw;	/* SMC firmware */
-	uint32_t                fw_version;
-	const struct amdgpu_dpm_funcs *funcs;
-	uint32_t                pcie_gen_mask;
-	uint32_t                pcie_mlw_mask;
-	struct amd_pp_display_configuration pm_display_cfg;/* set by DAL */
-};
-
-#define R600_SSTU_DFLT                               0
-#define R600_SST_DFLT                                0x00C8
-
-/* XXX are these ok? */
-#define R600_TEMP_RANGE_MIN (90 * 1000)
-#define R600_TEMP_RANGE_MAX (120 * 1000)
-
-#define FDO_PWM_MODE_STATIC  1
-#define FDO_PWM_MODE_STATIC_RPM 5
-
-enum amdgpu_td {
-	AMDGPU_TD_AUTO,
-	AMDGPU_TD_UP,
-	AMDGPU_TD_DOWN,
-};
-
-enum amdgpu_display_watermark {
-	AMDGPU_DISPLAY_WATERMARK_LOW = 0,
-	AMDGPU_DISPLAY_WATERMARK_HIGH = 1,
-};
-
-enum amdgpu_display_gap
-{
-    AMDGPU_PM_DISPLAY_GAP_VBLANK_OR_WM = 0,
-    AMDGPU_PM_DISPLAY_GAP_VBLANK       = 1,
-    AMDGPU_PM_DISPLAY_GAP_WATERMARK    = 2,
-    AMDGPU_PM_DISPLAY_GAP_IGNORE       = 3,
-};
-
-void amdgpu_dpm_print_class_info(u32 class, u32 class2);
-void amdgpu_dpm_print_cap_info(u32 caps);
-void amdgpu_dpm_print_ps_status(struct amdgpu_device *adev,
-				struct amdgpu_ps *rps);
-u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev);
-u32 amdgpu_dpm_get_vrefresh(struct amdgpu_device *adev);
-bool amdgpu_is_uvd_state(u32 class, u32 class2);
-void amdgpu_calculate_u_and_p(u32 i, u32 r_c, u32 p_b,
-			      u32 *p, u32 *u);
-int amdgpu_calculate_at(u32 t, u32 h, u32 fh, u32 fl, u32 *tl, u32 *th);
-
-bool amdgpu_is_internal_thermal_sensor(enum amdgpu_int_thermal_type sensor);
-
-int amdgpu_get_platform_caps(struct amdgpu_device *adev);
-
-int amdgpu_parse_extended_power_table(struct amdgpu_device *adev);
-void amdgpu_free_extended_power_table(struct amdgpu_device *adev);
-
-void amdgpu_add_thermal_controller(struct amdgpu_device *adev);
-
-enum amdgpu_pcie_gen amdgpu_get_pcie_gen_support(struct amdgpu_device *adev,
-						 u32 sys_mask,
-						 enum amdgpu_pcie_gen asic_gen,
-						 enum amdgpu_pcie_gen default_gen);
-
-u16 amdgpu_get_pcie_lane_support(struct amdgpu_device *adev,
-				 u16 asic_lanes,
-				 u16 default_lanes);
-u8 amdgpu_encode_pci_lane_width(u32 lanes);
-
-struct amd_vce_state*
-amdgpu_get_vce_clock_state(struct amdgpu_device *adev, unsigned idx);
-
-#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index f2d705e6a75a..2dfbddcef9ab 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -1,10 +1,3 @@
-/**
- * \file amdgpu_drv.c
- * AMD Amdgpu driver
- *
- * \author Gareth Hughes <gareth@valinux.com>
- */
-
 /*
  * Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
  * All Rights Reserved.
@@ -29,22 +22,38 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  */
 
-#include <drm/drmP.h>
 #include <drm/amdgpu_drm.h>
+#include <drm/clients/drm_client_setup.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_fbdev_ttm.h>
 #include <drm/drm_gem.h>
-#include "amdgpu_drv.h"
-
+#include <drm/drm_managed.h>
 #include <drm/drm_pciids.h>
-#include <linux/console.h>
+#include <drm/drm_probe_helper.h>
+#include <drm/drm_vblank.h>
+
+#include <linux/cc_platform.h>
+#include <linux/dynamic_debug.h>
 #include <linux/module.h>
+#include <linux/mmu_notifier.h>
 #include <linux/pm_runtime.h>
+#include <linux/suspend.h>
 #include <linux/vga_switcheroo.h>
-#include "drm_crtc_helper.h"
 
 #include "amdgpu.h"
-#include "amdgpu_irq.h"
-
 #include "amdgpu_amdkfd.h"
+#include "amdgpu_dma_buf.h"
+#include "amdgpu_drv.h"
+#include "amdgpu_fdinfo.h"
+#include "amdgpu_irq.h"
+#include "amdgpu_psp.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_reset.h"
+#include "amdgpu_sched.h"
+#include "amdgpu_xgmi.h"
+#include "amdgpu_userq.h"
+#include "amdgpu_userq_fence.h"
+#include "../amdxcp/amdgpu_xcp_drv.h"
 
 /*
  * KMS wrapper.
@@ -65,176 +74,1767 @@
  * - 3.13.0 - Add PRT support
  * - 3.14.0 - Fix race in amdgpu_ctx_get_fence() and note new functionality
  * - 3.15.0 - Export more gpu info for gfx9
+ * - 3.16.0 - Add reserved vmid support
+ * - 3.17.0 - Add AMDGPU_NUM_VRAM_CPU_PAGE_FAULTS.
+ * - 3.18.0 - Export gpu always on cu bitmap
+ * - 3.19.0 - Add support for UVD MJPEG decode
+ * - 3.20.0 - Add support for local BOs
+ * - 3.21.0 - Add DRM_AMDGPU_FENCE_TO_HANDLE ioctl
+ * - 3.22.0 - Add DRM_AMDGPU_SCHED ioctl
+ * - 3.23.0 - Add query for VRAM lost counter
+ * - 3.24.0 - Add high priority compute support for gfx9
+ * - 3.25.0 - Add support for sensor query info (stable pstate sclk/mclk).
+ * - 3.26.0 - GFX9: Process AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE.
+ * - 3.27.0 - Add new chunk to AMDGPU_CS to enable BO_LIST creation.
+ * - 3.28.0 - Add AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES
+ * - 3.29.0 - Add AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID
+ * - 3.30.0 - Add AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE.
+ * - 3.31.0 - Add support for per-flip tiling attribute changes with DC
+ * - 3.32.0 - Add syncobj timeline support to AMDGPU_CS.
+ * - 3.33.0 - Fixes for GDS ENOMEM failures in AMDGPU_CS.
+ * - 3.34.0 - Non-DC can flip correctly between buffers with different pitches
+ * - 3.35.0 - Add drm_amdgpu_info_device::tcc_disabled_mask
+ * - 3.36.0 - Allow reading more status registers on si/cik
+ * - 3.37.0 - L2 is invalidated before SDMA IBs, needed for correctness
+ * - 3.38.0 - Add AMDGPU_IB_FLAG_EMIT_MEM_SYNC
+ * - 3.39.0 - DMABUF implicit sync does a full pipeline sync
+ * - 3.40.0 - Add AMDGPU_IDS_FLAGS_TMZ
+ * - 3.41.0 - Add video codec query
+ * - 3.42.0 - Add 16bpc fixed point display support
+ * - 3.43.0 - Add device hot plug/unplug support
+ * - 3.44.0 - DCN3 supports DCC independent block settings: !64B && 128B, 64B && 128B
+ * - 3.45.0 - Add context ioctl stable pstate interface
+ * - 3.46.0 - To enable hot plug amdgpu tests in libdrm
+ * - 3.47.0 - Add AMDGPU_GEM_CREATE_DISCARDABLE and AMDGPU_VM_NOALLOC flags
+ * - 3.48.0 - Add IP discovery version info to HW INFO
+ * - 3.49.0 - Add gang submit into CS IOCTL
+ * - 3.50.0 - Update AMDGPU_INFO_DEV_INFO IOCTL for minimum engine and memory clock
+ *            Update AMDGPU_INFO_SENSOR IOCTL for PEAK_PSTATE engine and memory clock
+ *   3.51.0 - Return the PCIe gen and lanes from the INFO ioctl
+ *   3.52.0 - Add AMDGPU_IDS_FLAGS_CONFORMANT_TRUNC_COORD, add device_info fields:
+ *            tcp_cache_size, num_sqc_per_wgp, sqc_data_cache_size, sqc_inst_cache_size,
+ *            gl1c_cache_size, gl2c_cache_size, mall_size, enabled_rb_pipes_mask_hi
+ *   3.53.0 - Support for GFX11 CP GFX shadowing
+ *   3.54.0 - Add AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS support
+ * - 3.55.0 - Add AMDGPU_INFO_GPUVM_FAULT query
+ * - 3.56.0 - Update IB start address and size alignment for decode and encode
+ * - 3.57.0 - Compute tunneling on GFX10+
+ * - 3.58.0 - Add GFX12 DCC support
+ * - 3.59.0 - Cleared VRAM
+ * - 3.60.0 - Add AMDGPU_TILING_GFX12_DCC_WRITE_COMPRESS_DISABLE (Vulkan requirement)
+ * - 3.61.0 - Contains fix for RV/PCO compute queues
+ * - 3.62.0 - Add AMDGPU_IDS_FLAGS_MODE_PF, AMDGPU_IDS_FLAGS_MODE_VF & AMDGPU_IDS_FLAGS_MODE_PT
+ * - 3.63.0 - GFX12 display DCC supports 256B max compressed block size
+ * - 3.64.0 - Userq IP support query
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	15
+#define KMS_DRIVER_MINOR	64
 #define KMS_DRIVER_PATCHLEVEL	0
 
-int amdgpu_vram_limit = 0;
+/*
+ * amdgpu.debug module options. Are all disabled by default
+ */
+enum AMDGPU_DEBUG_MASK {
+	AMDGPU_DEBUG_VM = BIT(0),
+	AMDGPU_DEBUG_LARGEBAR = BIT(1),
+	AMDGPU_DEBUG_DISABLE_GPU_SOFT_RECOVERY = BIT(2),
+	AMDGPU_DEBUG_USE_VRAM_FW_BUF = BIT(3),
+	AMDGPU_DEBUG_ENABLE_RAS_ACA = BIT(4),
+	AMDGPU_DEBUG_ENABLE_EXP_RESETS = BIT(5),
+	AMDGPU_DEBUG_DISABLE_GPU_RING_RESET = BIT(6),
+	AMDGPU_DEBUG_SMU_POOL = BIT(7),
+	AMDGPU_DEBUG_VM_USERPTR = BIT(8),
+	AMDGPU_DEBUG_DISABLE_RAS_CE_LOG = BIT(9),
+	AMDGPU_DEBUG_ENABLE_CE_CS = BIT(10)
+};
+
+unsigned int amdgpu_vram_limit = UINT_MAX;
+int amdgpu_vis_vram_limit;
 int amdgpu_gart_size = -1; /* auto */
+int amdgpu_gtt_size = -1; /* auto */
 int amdgpu_moverate = -1; /* auto */
-int amdgpu_benchmarking = 0;
-int amdgpu_testing = 0;
 int amdgpu_audio = -1;
-int amdgpu_disp_priority = 0;
-int amdgpu_hw_i2c = 0;
+int amdgpu_disp_priority;
+int amdgpu_hw_i2c;
 int amdgpu_pcie_gen2 = -1;
 int amdgpu_msi = -1;
-int amdgpu_lockup_timeout = 0;
+char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH];
 int amdgpu_dpm = -1;
 int amdgpu_fw_load_type = -1;
 int amdgpu_aspm = -1;
 int amdgpu_runtime_pm = -1;
-unsigned amdgpu_ip_block_mask = 0xffffffff;
+uint amdgpu_ip_block_mask = 0xffffffff;
 int amdgpu_bapm = -1;
-int amdgpu_deep_color = 0;
+int amdgpu_deep_color;
 int amdgpu_vm_size = -1;
+int amdgpu_vm_fragment_size = -1;
 int amdgpu_vm_block_size = -1;
-int amdgpu_vm_fault_stop = 0;
-int amdgpu_vm_debug = 0;
-int amdgpu_vram_page_split = 1024;
-int amdgpu_exp_hw_support = 0;
+int amdgpu_vm_fault_stop;
+int amdgpu_vm_update_mode = -1;
+int amdgpu_exp_hw_support;
+int amdgpu_dc = -1;
 int amdgpu_sched_jobs = 32;
 int amdgpu_sched_hw_submission = 2;
-int amdgpu_no_evict = 0;
-int amdgpu_direct_gma_size = 0;
-unsigned amdgpu_pcie_gen_cap = 0;
-unsigned amdgpu_pcie_lane_cap = 0;
-unsigned amdgpu_cg_mask = 0xffffffff;
-unsigned amdgpu_pg_mask = 0xffffffff;
-char *amdgpu_disable_cu = NULL;
-char *amdgpu_virtual_display = NULL;
-unsigned amdgpu_pp_feature_mask = 0xffffffff;
-int amdgpu_ngg = 0;
-int amdgpu_prim_buf_per_se = 0;
-int amdgpu_pos_buf_per_se = 0;
-int amdgpu_cntl_sb_buf_per_se = 0;
-int amdgpu_param_buf_per_se = 0;
+uint amdgpu_pcie_gen_cap;
+uint amdgpu_pcie_lane_cap;
+u64 amdgpu_cg_mask = 0xffffffffffffffff;
+uint amdgpu_pg_mask = 0xffffffff;
+uint amdgpu_sdma_phase_quantum = 32;
+char *amdgpu_disable_cu;
+char *amdgpu_virtual_display;
+int amdgpu_enforce_isolation = -1;
+int amdgpu_modeset = -1;
+
+/* Specifies the default granularity for SVM, used in buffer
+ * migration and restoration of backing memory when handling
+ * recoverable page faults.
+ *
+ * The value is given as log(numPages(buffer)); for a 2 MiB
+ * buffer it computes to be 9
+ */
+uint amdgpu_svm_default_granularity = 9;
 
+/*
+ * OverDrive(bit 14) disabled by default
+ * GFX DCS(bit 19) disabled by default
+ */
+uint amdgpu_pp_feature_mask = 0xfff7bfff;
+uint amdgpu_force_long_training;
+int amdgpu_lbpw = -1;
+int amdgpu_compute_multipipe = -1;
+int amdgpu_gpu_recovery = -1; /* auto */
+int amdgpu_emu_mode;
+uint amdgpu_smu_memory_pool_size;
+int amdgpu_smu_pptable_id = -1;
+/*
+ * FBC (bit 0) disabled by default
+ * MULTI_MON_PP_MCLK_SWITCH (bit 1) enabled by default
+ *   - With this, for multiple monitors in sync(e.g. with the same model),
+ *     mclk switching will be allowed. And the mclk will be not foced to the
+ *     highest. That helps saving some idle power.
+ * DISABLE_FRACTIONAL_PWM (bit 2) disabled by default
+ * PSR (bit 3) disabled by default
+ * EDP NO POWER SEQUENCING (bit 4) disabled by default
+ */
+uint amdgpu_dc_feature_mask = 2;
+uint amdgpu_dc_debug_mask;
+uint amdgpu_dc_visual_confirm;
+int amdgpu_async_gfx_ring = 1;
+int amdgpu_mcbp = -1;
+int amdgpu_discovery = -1;
+int amdgpu_mes;
+int amdgpu_mes_log_enable = 0;
+int amdgpu_mes_kiq;
+int amdgpu_uni_mes = 1;
+int amdgpu_noretry = -1;
+int amdgpu_force_asic_type = -1;
+int amdgpu_tmz = -1; /* auto */
+uint amdgpu_freesync_vid_mode;
+int amdgpu_reset_method = -1; /* auto */
+int amdgpu_num_kcq = -1;
+int amdgpu_smartshift_bias;
+int amdgpu_use_xgmi_p2p = 1;
+int amdgpu_vcnfw_log;
+int amdgpu_sg_display = -1; /* auto */
+int amdgpu_user_partt_mode = AMDGPU_AUTO_COMPUTE_PARTITION_MODE;
+int amdgpu_umsch_mm;
+int amdgpu_seamless = -1; /* auto */
+uint amdgpu_debug_mask;
+int amdgpu_agp = -1; /* auto */
+int amdgpu_wbrf = -1;
+int amdgpu_damage_clips = -1; /* auto */
+int amdgpu_umsch_mm_fwlog;
+int amdgpu_rebar = -1; /* auto */
+int amdgpu_user_queue = -1;
+
+DECLARE_DYNDBG_CLASSMAP(drm_debug_classes, DD_CLASS_TYPE_DISJOINT_BITS, 0,
+			"DRM_UT_CORE",
+			"DRM_UT_DRIVER",
+			"DRM_UT_KMS",
+			"DRM_UT_PRIME",
+			"DRM_UT_ATOMIC",
+			"DRM_UT_VBL",
+			"DRM_UT_STATE",
+			"DRM_UT_LEASE",
+			"DRM_UT_DP",
+			"DRM_UT_DRMRES");
+
+struct amdgpu_mgpu_info mgpu_info = {
+	.mutex = __MUTEX_INITIALIZER(mgpu_info.mutex),
+};
+int amdgpu_ras_enable = -1;
+uint amdgpu_ras_mask = 0xffffffff;
+int amdgpu_bad_page_threshold = -1;
+struct amdgpu_watchdog_timer amdgpu_watchdog_timer = {
+	.timeout_fatal_disable = false,
+	.period = 0x0, /* default to 0x0 (timeout disable) */
+};
+
+/**
+ * DOC: vramlimit (int)
+ * Restrict the total amount of VRAM in MiB for testing.  The default is 0 (Use full VRAM).
+ */
 MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes");
 module_param_named(vramlimit, amdgpu_vram_limit, int, 0600);
 
-MODULE_PARM_DESC(gartsize, "Size of PCIE/IGP gart to setup in megabytes (32, 64, etc., -1 = auto)");
-module_param_named(gartsize, amdgpu_gart_size, int, 0600);
+/**
+ * DOC: vis_vramlimit (int)
+ * Restrict the amount of CPU visible VRAM in MiB for testing.  The default is 0 (Use full CPU visible VRAM).
+ */
+MODULE_PARM_DESC(vis_vramlimit, "Restrict visible VRAM for testing, in megabytes");
+module_param_named(vis_vramlimit, amdgpu_vis_vram_limit, int, 0444);
 
-MODULE_PARM_DESC(moverate, "Maximum buffer migration rate in MB/s. (32, 64, etc., -1=auto, 0=1=disabled)");
-module_param_named(moverate, amdgpu_moverate, int, 0600);
+/**
+ * DOC: gartsize (uint)
+ * Restrict the size of GART (for kernel use) in Mib (32, 64, etc.) for testing.
+ * The default is -1 (The size depends on asic).
+ */
+MODULE_PARM_DESC(gartsize, "Size of kernel GART to setup in megabytes (32, 64, etc., -1=auto)");
+module_param_named(gartsize, amdgpu_gart_size, uint, 0600);
 
-MODULE_PARM_DESC(benchmark, "Run benchmark");
-module_param_named(benchmark, amdgpu_benchmarking, int, 0444);
+/**
+ * DOC: gttsize (int)
+ * Restrict the size of GTT domain (for userspace use) in MiB for testing.
+ * The default is -1 (Use value specified by TTM).
+ * This parameter is deprecated and will be removed in the future.
+ */
+MODULE_PARM_DESC(gttsize, "Size of the GTT userspace domain in megabytes (-1 = auto)");
+module_param_named(gttsize, amdgpu_gtt_size, int, 0600);
 
-MODULE_PARM_DESC(test, "Run tests");
-module_param_named(test, amdgpu_testing, int, 0444);
+/**
+ * DOC: moverate (int)
+ * Set maximum buffer migration rate in MB/s. The default is -1 (8 MB/s).
+ */
+MODULE_PARM_DESC(moverate, "Maximum buffer migration rate in MB/s. (32, 64, etc., -1=auto, 0=1=disabled)");
+module_param_named(moverate, amdgpu_moverate, int, 0600);
 
-MODULE_PARM_DESC(audio, "Audio enable (-1 = auto, 0 = disable, 1 = enable)");
+/**
+ * DOC: audio (int)
+ * Set HDMI/DPAudio. Only affects non-DC display handling. The default is -1 (Enabled), set 0 to disabled it.
+ */
+MODULE_PARM_DESC(audio, "HDMI/DP Audio enable for non DC displays (-1 = auto, 0 = disable, 1 = enable)");
 module_param_named(audio, amdgpu_audio, int, 0444);
 
+/**
+ * DOC: disp_priority (int)
+ * Set display Priority (1 = normal, 2 = high). Only affects non-DC display handling. The default is 0 (auto).
+ */
 MODULE_PARM_DESC(disp_priority, "Display Priority (0 = auto, 1 = normal, 2 = high)");
 module_param_named(disp_priority, amdgpu_disp_priority, int, 0444);
 
+/**
+ * DOC: hw_i2c (int)
+ * To enable hw i2c engine. Only affects non-DC display handling. The default is 0 (Disabled).
+ */
 MODULE_PARM_DESC(hw_i2c, "hw i2c engine enable (0 = disable)");
 module_param_named(hw_i2c, amdgpu_hw_i2c, int, 0444);
 
+/**
+ * DOC: pcie_gen2 (int)
+ * To disable PCIE Gen2/3 mode (0 = disable, 1 = enable). The default is -1 (auto, enabled).
+ */
 MODULE_PARM_DESC(pcie_gen2, "PCIE Gen2 mode (-1 = auto, 0 = disable, 1 = enable)");
 module_param_named(pcie_gen2, amdgpu_pcie_gen2, int, 0444);
 
+/**
+ * DOC: msi (int)
+ * To disable Message Signaled Interrupts (MSI) functionality (1 = enable, 0 = disable). The default is -1 (auto, enabled).
+ */
 MODULE_PARM_DESC(msi, "MSI support (1 = enable, 0 = disable, -1 = auto)");
 module_param_named(msi, amdgpu_msi, int, 0444);
 
-MODULE_PARM_DESC(lockup_timeout, "GPU lockup timeout in ms (default 0 = disable)");
-module_param_named(lockup_timeout, amdgpu_lockup_timeout, int, 0444);
+/**
+ * DOC: svm_default_granularity (uint)
+ * Used in buffer migration and handling of recoverable page faults
+ */
+MODULE_PARM_DESC(svm_default_granularity, "SVM's default granularity in log(2^Pages), default 9 = 2^9 = 2 MiB");
+module_param_named(svm_default_granularity, amdgpu_svm_default_granularity, uint, 0644);
 
+/**
+ * DOC: lockup_timeout (string)
+ * Set GPU scheduler timeout value in ms.
+ *
+ * The format can be [single value] for setting all timeouts at once or
+ * [GFX,Compute,SDMA,Video] to set individual timeouts.
+ * Negative values mean infinity.
+ *
+ * By default(with no lockup_timeout settings), the timeout for all queues is 2000.
+ */
+MODULE_PARM_DESC(lockup_timeout,
+		 "GPU lockup timeout in ms (default: 2000. 0: keep default value. negative: infinity timeout), format: [single value for all] or [GFX,Compute,SDMA,Video].");
+module_param_string(lockup_timeout, amdgpu_lockup_timeout,
+		    sizeof(amdgpu_lockup_timeout), 0444);
+
+/**
+ * DOC: dpm (int)
+ * Override for dynamic power management setting
+ * (0 = disable, 1 = enable)
+ * The default is -1 (auto).
+ */
 MODULE_PARM_DESC(dpm, "DPM support (1 = enable, 0 = disable, -1 = auto)");
 module_param_named(dpm, amdgpu_dpm, int, 0444);
 
-MODULE_PARM_DESC(fw_load_type, "firmware loading type (0 = direct, 1 = SMU, 2 = PSP, -1 = auto)");
+/**
+ * DOC: fw_load_type (int)
+ * Set different firmware loading type for debugging, if supported.
+ * Set to 0 to force direct loading if supported by the ASIC.  Set
+ * to -1 to select the default loading mode for the ASIC, as defined
+ * by the driver.  The default is -1 (auto).
+ */
+MODULE_PARM_DESC(fw_load_type, "firmware loading type (3 = rlc backdoor autoload if supported, 2 = smu load if supported, 1 = psp load, 0 = force direct if supported, -1 = auto)");
 module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444);
 
+/**
+ * DOC: aspm (int)
+ * To disable ASPM (1 = enable, 0 = disable). The default is -1 (auto, enabled).
+ */
 MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)");
 module_param_named(aspm, amdgpu_aspm, int, 0444);
 
-MODULE_PARM_DESC(runpm, "PX runtime pm (1 = force enable, 0 = disable, -1 = PX only default)");
+/**
+ * DOC: runpm (int)
+ * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down
+ * the dGPUs when they are idle if supported. The default is -1 (auto enable).
+ * Setting the value to 0 disables this functionality.
+ * Setting the value to -2 is auto enabled with power down when displays are attached.
+ */
+MODULE_PARM_DESC(runpm, "PX runtime pm (2 = force enable with BAMACO, 1 = force enable with BACO, 0 = disable, -1 = auto, -2 = auto with displays)");
 module_param_named(runpm, amdgpu_runtime_pm, int, 0444);
 
+/**
+ * DOC: ip_block_mask (uint)
+ * Override what IP blocks are enabled on the GPU. Each GPU is a collection of IP blocks (gfx, display, video, etc.).
+ * Use this parameter to disable specific blocks. Note that the IP blocks do not have a fixed index. Some asics may not have
+ * some IPs or may include multiple instances of an IP so the ordering various from asic to asic. See the driver output in
+ * the kernel log for the list of IPs on the asic. The default is 0xffffffff (enable all blocks on a device).
+ */
 MODULE_PARM_DESC(ip_block_mask, "IP Block Mask (all blocks enabled (default))");
-module_param_named(ip_block_mask, amdgpu_ip_block_mask, uint, 0444);
+module_param_named_unsafe(ip_block_mask, amdgpu_ip_block_mask, uint, 0444);
 
+/**
+ * DOC: bapm (int)
+ * Bidirectional Application Power Management (BAPM) used to dynamically share TDP between CPU and GPU. Set value 0 to disable it.
+ * The default -1 (auto, enabled)
+ */
 MODULE_PARM_DESC(bapm, "BAPM support (1 = enable, 0 = disable, -1 = auto)");
 module_param_named(bapm, amdgpu_bapm, int, 0444);
 
+/**
+ * DOC: deep_color (int)
+ * Set 1 to enable Deep Color support. Only affects non-DC display handling. The default is 0 (disabled).
+ */
 MODULE_PARM_DESC(deep_color, "Deep Color support (1 = enable, 0 = disable (default))");
 module_param_named(deep_color, amdgpu_deep_color, int, 0444);
 
+/**
+ * DOC: vm_size (int)
+ * Override the size of the GPU's per client virtual address space in GiB.  The default is -1 (automatic for each asic).
+ */
 MODULE_PARM_DESC(vm_size, "VM address space size in gigabytes (default 64GB)");
 module_param_named(vm_size, amdgpu_vm_size, int, 0444);
 
+/**
+ * DOC: vm_fragment_size (int)
+ * Override VM fragment size in bits (4, 5, etc. 4 = 64K, 9 = 2M). The default is -1 (automatic for each asic).
+ */
+MODULE_PARM_DESC(vm_fragment_size, "VM fragment size in bits (4, 5, etc. 4 = 64K (default), Max 9 = 2M)");
+module_param_named(vm_fragment_size, amdgpu_vm_fragment_size, int, 0444);
+
+/**
+ * DOC: vm_block_size (int)
+ * Override VM page table size in bits (default depending on vm_size and hw setup). The default is -1 (automatic for each asic).
+ */
 MODULE_PARM_DESC(vm_block_size, "VM page table size in bits (default depending on vm_size)");
 module_param_named(vm_block_size, amdgpu_vm_block_size, int, 0444);
 
+/**
+ * DOC: vm_fault_stop (int)
+ * Stop on VM fault for debugging (0 = never, 1 = print first, 2 = always). The default is 0 (No stop).
+ */
 MODULE_PARM_DESC(vm_fault_stop, "Stop on VM fault (0 = never (default), 1 = print first, 2 = always)");
 module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444);
 
-MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)");
-module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
-
-MODULE_PARM_DESC(vram_page_split, "Number of pages after we split VRAM allocations (default 1024, -1 = disable)");
-module_param_named(vram_page_split, amdgpu_vram_page_split, int, 0444);
+/**
+ * DOC: vm_update_mode (int)
+ * Override VM update mode. VM updated by using CPU (0 = never, 1 = Graphics only, 2 = Compute only, 3 = Both). The default
+ * is -1 (Only in large BAR(LB) systems Compute VM tables will be updated by CPU, otherwise 0, never).
+ */
+MODULE_PARM_DESC(vm_update_mode, "VM update using CPU (0 = never (default except for large BAR(LB)), 1 = Graphics only, 2 = Compute only (default for LB), 3 = Both");
+module_param_named(vm_update_mode, amdgpu_vm_update_mode, int, 0444);
 
+/**
+ * DOC: exp_hw_support (int)
+ * Enable experimental hw support (1 = enable). The default is 0 (disabled).
+ */
 MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
-module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
+module_param_named_unsafe(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
+
+/**
+ * DOC: dc (int)
+ * Disable/Enable Display Core driver for debugging (1 = enable, 0 = disable). The default is -1 (automatic for each asic).
+ */
+MODULE_PARM_DESC(dc, "Display Core driver (1 = enable, 0 = disable, -1 = auto (default))");
+module_param_named(dc, amdgpu_dc, int, 0444);
 
+/**
+ * DOC: sched_jobs (int)
+ * Override the max number of jobs supported in the sw queue. The default is 32.
+ */
 MODULE_PARM_DESC(sched_jobs, "the max number of jobs supported in the sw queue (default 32)");
 module_param_named(sched_jobs, amdgpu_sched_jobs, int, 0444);
 
+/**
+ * DOC: sched_hw_submission (int)
+ * Override the max number of HW submissions. The default is 2.
+ */
 MODULE_PARM_DESC(sched_hw_submission, "the max number of HW submissions (default 2)");
 module_param_named(sched_hw_submission, amdgpu_sched_hw_submission, int, 0444);
 
+/**
+ * DOC: ppfeaturemask (hexint)
+ * Override power features enabled. See enum PP_FEATURE_MASK in drivers/gpu/drm/amd/include/amd_shared.h.
+ * The default is the current set of stable power features.
+ */
 MODULE_PARM_DESC(ppfeaturemask, "all power features enabled (default))");
-module_param_named(ppfeaturemask, amdgpu_pp_feature_mask, int, 0444);
-
-MODULE_PARM_DESC(no_evict, "Support pinning request from user space (1 = enable, 0 = disable (default))");
-module_param_named(no_evict, amdgpu_no_evict, int, 0444);
+module_param_named(ppfeaturemask, amdgpu_pp_feature_mask, hexint, 0444);
 
-MODULE_PARM_DESC(direct_gma_size, "Direct GMA size in megabytes (max 96MB)");
-module_param_named(direct_gma_size, amdgpu_direct_gma_size, int, 0444);
+/**
+ * DOC: forcelongtraining (uint)
+ * Force long memory training in resume.
+ * The default is zero, indicates short training in resume.
+ */
+MODULE_PARM_DESC(forcelongtraining, "force memory long training");
+module_param_named(forcelongtraining, amdgpu_force_long_training, uint, 0444);
 
+/**
+ * DOC: pcie_gen_cap (uint)
+ * Override PCIE gen speed capabilities. See the CAIL flags in drivers/gpu/drm/amd/include/amd_pcie.h.
+ * The default is 0 (automatic for each asic).
+ */
 MODULE_PARM_DESC(pcie_gen_cap, "PCIE Gen Caps (0: autodetect (default))");
 module_param_named(pcie_gen_cap, amdgpu_pcie_gen_cap, uint, 0444);
 
+/**
+ * DOC: pcie_lane_cap (uint)
+ * Override PCIE lanes capabilities. See the CAIL flags in drivers/gpu/drm/amd/include/amd_pcie.h.
+ * The default is 0 (automatic for each asic).
+ */
 MODULE_PARM_DESC(pcie_lane_cap, "PCIE Lane Caps (0: autodetect (default))");
 module_param_named(pcie_lane_cap, amdgpu_pcie_lane_cap, uint, 0444);
 
+/**
+ * DOC: cg_mask (ullong)
+ * Override Clockgating features enabled on GPU (0 = disable clock gating). See the AMD_CG_SUPPORT flags in
+ * drivers/gpu/drm/amd/include/amd_shared.h. The default is 0xffffffffffffffff (all enabled).
+ */
 MODULE_PARM_DESC(cg_mask, "Clockgating flags mask (0 = disable clock gating)");
-module_param_named(cg_mask, amdgpu_cg_mask, uint, 0444);
+module_param_named(cg_mask, amdgpu_cg_mask, ullong, 0444);
 
+/**
+ * DOC: pg_mask (uint)
+ * Override Powergating features enabled on GPU (0 = disable power gating). See the AMD_PG_SUPPORT flags in
+ * drivers/gpu/drm/amd/include/amd_shared.h. The default is 0xffffffff (all enabled).
+ */
 MODULE_PARM_DESC(pg_mask, "Powergating flags mask (0 = disable power gating)");
 module_param_named(pg_mask, amdgpu_pg_mask, uint, 0444);
 
+/**
+ * DOC: sdma_phase_quantum (uint)
+ * Override SDMA context switch phase quantum (x 1K GPU clock cycles, 0 = no change). The default is 32.
+ */
+MODULE_PARM_DESC(sdma_phase_quantum, "SDMA context switch phase quantum (x 1K GPU clock cycles, 0 = no change (default 32))");
+module_param_named(sdma_phase_quantum, amdgpu_sdma_phase_quantum, uint, 0444);
+
+/**
+ * DOC: disable_cu (charp)
+ * Set to disable CUs (It's set like se.sh.cu,...). The default is NULL.
+ */
 MODULE_PARM_DESC(disable_cu, "Disable CUs (se.sh.cu,...)");
 module_param_named(disable_cu, amdgpu_disable_cu, charp, 0444);
 
+/**
+ * DOC: virtual_display (charp)
+ * Set to enable virtual display feature. This feature provides a virtual display hardware on headless boards
+ * or in virtualized environments. It will be set like xxxx:xx:xx.x,x;xxxx:xx:xx.x,x. It's the pci address of
+ * the device, plus the number of crtcs to expose. E.g., 0000:26:00.0,4 would enable 4 virtual crtcs on the pci
+ * device at 26:00.0. The default is NULL.
+ */
 MODULE_PARM_DESC(virtual_display,
 		 "Enable virtual display feature (the virtual_display will be set like xxxx:xx:xx.x,x;xxxx:xx:xx.x,x)");
 module_param_named(virtual_display, amdgpu_virtual_display, charp, 0444);
 
-MODULE_PARM_DESC(ngg, "Next Generation Graphics (1 = enable, 0 = disable(default depending on gfx))");
-module_param_named(ngg, amdgpu_ngg, int, 0444);
+/**
+ * DOC: lbpw (int)
+ * Override Load Balancing Per Watt (LBPW) support (1 = enable, 0 = disable). The default is -1 (auto, enabled).
+ */
+MODULE_PARM_DESC(lbpw, "Load Balancing Per Watt (LBPW) support (1 = enable, 0 = disable, -1 = auto)");
+module_param_named(lbpw, amdgpu_lbpw, int, 0444);
+
+MODULE_PARM_DESC(compute_multipipe, "Force compute queues to be spread across pipes (1 = enable, 0 = disable, -1 = auto)");
+module_param_named(compute_multipipe, amdgpu_compute_multipipe, int, 0444);
+
+/**
+ * DOC: gpu_recovery (int)
+ * Set to enable GPU recovery mechanism (1 = enable, 0 = disable). The default is -1 (auto, disabled except SRIOV).
+ */
+MODULE_PARM_DESC(gpu_recovery, "Enable GPU recovery mechanism, (1 = enable, 0 = disable, -1 = auto)");
+module_param_named_unsafe(gpu_recovery, amdgpu_gpu_recovery, int, 0444);
+
+/**
+ * DOC: emu_mode (int)
+ * Set value 1 to enable emulation mode. This is only needed when running on an emulator. The default is 0 (disabled).
+ */
+MODULE_PARM_DESC(emu_mode, "Emulation mode, (1 = enable, 0 = disable)");
+module_param_named_unsafe(emu_mode, amdgpu_emu_mode, int, 0444);
+
+/**
+ * DOC: ras_enable (int)
+ * Enable RAS features on the GPU (0 = disable, 1 = enable, -1 = auto (default))
+ */
+MODULE_PARM_DESC(ras_enable, "Enable RAS features on the GPU (0 = disable, 1 = enable, -1 = auto (default))");
+module_param_named(ras_enable, amdgpu_ras_enable, int, 0444);
+
+/**
+ * DOC: ras_mask (uint)
+ * Mask of RAS features to enable (default 0xffffffff), only valid when ras_enable == 1
+ * See the flags in drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+ */
+MODULE_PARM_DESC(ras_mask, "Mask of RAS features to enable (default 0xffffffff), only valid when ras_enable == 1");
+module_param_named(ras_mask, amdgpu_ras_mask, uint, 0444);
+
+/**
+ * DOC: timeout_fatal_disable (bool)
+ * Disable Watchdog timeout fatal error event
+ */
+MODULE_PARM_DESC(timeout_fatal_disable, "disable watchdog timeout fatal error (false = default)");
+module_param_named(timeout_fatal_disable, amdgpu_watchdog_timer.timeout_fatal_disable, bool, 0644);
+
+/**
+ * DOC: timeout_period (uint)
+ * Modify the watchdog timeout max_cycles as (1 << period)
+ */
+MODULE_PARM_DESC(timeout_period, "watchdog timeout period (0 = timeout disabled, 1 ~ 0x23 = timeout maxcycles = (1 << period)");
+module_param_named(timeout_period, amdgpu_watchdog_timer.period, uint, 0644);
+
+/**
+ * DOC: si_support (int)
+ * 1 = enabled, 0 = disabled, -1 = default
+ *
+ * SI (Southern Islands) are first generation GCN GPUs, supported by both
+ * drivers: radeon (old) and amdgpu (new). This parameter controls whether
+ * amdgpu should support SI.
+ * By default, SI dedicated GPUs are supported by amdgpu.
+ * Only relevant when CONFIG_DRM_AMDGPU_SI is enabled to build SI support in amdgpu.
+ * See also radeon.si_support which should be disabled when amdgpu.si_support is
+ * enabled, and vice versa.
+ */
+int amdgpu_si_support = -1;
+#ifdef CONFIG_DRM_AMDGPU_SI
+MODULE_PARM_DESC(si_support, "SI support (1 = enabled, 0 = disabled, -1 = default)");
+module_param_named(si_support, amdgpu_si_support, int, 0444);
+#endif
+
+/**
+ * DOC: cik_support (int)
+ * 1 = enabled, 0 = disabled, -1 = default
+ *
+ * CIK (Sea Islands) are second generation GCN GPUs, supported by both
+ * drivers: radeon (old) and amdgpu (new). This parameter controls whether
+ * amdgpu should support CIK.
+ * By default:
+ * - CIK dedicated GPUs are supported by amdgpu.
+ * - CIK APUs are supported by radeon (except when radeon is not built).
+ * Only relevant when CONFIG_DRM_AMDGPU_CIK is enabled to build CIK support in amdgpu.
+ * See also radeon.cik_support which should be disabled when amdgpu.cik_support is
+ * enabled, and vice versa.
+ */
+int amdgpu_cik_support = -1;
+#ifdef CONFIG_DRM_AMDGPU_CIK
+MODULE_PARM_DESC(cik_support, "CIK support  (1 = enabled, 0 = disabled, -1 = default)");
+module_param_named(cik_support, amdgpu_cik_support, int, 0444);
+#endif
+
+/**
+ * DOC: smu_memory_pool_size (uint)
+ * It is used to reserve gtt for smu debug usage, setting value 0 to disable it. The actual size is value * 256MiB.
+ * E.g. 0x1 = 256Mbyte, 0x2 = 512Mbyte, 0x4 = 1 Gbyte, 0x8 = 2GByte. The default is 0 (disabled).
+ */
+MODULE_PARM_DESC(smu_memory_pool_size,
+	"reserve gtt for smu debug usage, 0 = disable,0x1 = 256Mbyte, 0x2 = 512Mbyte, 0x4 = 1 Gbyte, 0x8 = 2GByte");
+module_param_named(smu_memory_pool_size, amdgpu_smu_memory_pool_size, uint, 0444);
+
+/**
+ * DOC: async_gfx_ring (int)
+ * It is used to enable gfx rings that could be configured with different prioritites or equal priorities
+ */
+MODULE_PARM_DESC(async_gfx_ring,
+	"Asynchronous GFX rings that could be configured with either different priorities (HP3D ring and LP3D ring), or equal priorities (0 = disabled, 1 = enabled (default))");
+module_param_named(async_gfx_ring, amdgpu_async_gfx_ring, int, 0444);
+
+/**
+ * DOC: mcbp (int)
+ * It is used to enable mid command buffer preemption. (0 = disabled, 1 = enabled, -1 auto (default))
+ */
+MODULE_PARM_DESC(mcbp,
+	"Enable Mid-command buffer preemption (0 = disabled, 1 = enabled), -1 = auto (default)");
+module_param_named(mcbp, amdgpu_mcbp, int, 0444);
+
+/**
+ * DOC: discovery (int)
+ * Allow driver to discover hardware IP information from IP Discovery table at the top of VRAM.
+ * (-1 = auto (default), 0 = disabled, 1 = enabled, 2 = use ip_discovery table from file)
+ */
+MODULE_PARM_DESC(discovery,
+	"Allow driver to discover hardware IPs from IP Discovery table at the top of VRAM");
+module_param_named(discovery, amdgpu_discovery, int, 0444);
+
+/**
+ * DOC: mes (int)
+ * Enable Micro Engine Scheduler. This is a new hw scheduling engine for gfx, sdma, and compute.
+ * (0 = disabled (default), 1 = enabled)
+ */
+MODULE_PARM_DESC(mes,
+	"Enable Micro Engine Scheduler (0 = disabled (default), 1 = enabled)");
+module_param_named(mes, amdgpu_mes, int, 0444);
+
+/**
+ * DOC: mes_log_enable (int)
+ * Enable Micro Engine Scheduler log. This is used to enable/disable MES internal log.
+ * (0 = disabled (default), 1 = enabled)
+ */
+MODULE_PARM_DESC(mes_log_enable,
+	"Enable Micro Engine Scheduler log (0 = disabled (default), 1 = enabled)");
+module_param_named(mes_log_enable, amdgpu_mes_log_enable, int, 0444);
+
+/**
+ * DOC: mes_kiq (int)
+ * Enable Micro Engine Scheduler KIQ. This is a new engine pipe for kiq.
+ * (0 = disabled (default), 1 = enabled)
+ */
+MODULE_PARM_DESC(mes_kiq,
+	"Enable Micro Engine Scheduler KIQ (0 = disabled (default), 1 = enabled)");
+module_param_named(mes_kiq, amdgpu_mes_kiq, int, 0444);
+
+/**
+ * DOC: uni_mes (int)
+ * Enable Unified Micro Engine Scheduler. This is a new engine pipe for unified scheduler.
+ * (0 = disabled (default), 1 = enabled)
+ */
+MODULE_PARM_DESC(uni_mes,
+	"Enable Unified Micro Engine Scheduler (0 = disabled, 1 = enabled(default)");
+module_param_named(uni_mes, amdgpu_uni_mes, int, 0444);
+
+/**
+ * DOC: noretry (int)
+ * Disable XNACK retry in the SQ by default on GFXv9 hardware. On ASICs that
+ * do not support per-process XNACK this also disables retry page faults.
+ * (0 = retry enabled, 1 = retry disabled, -1 auto (default))
+ */
+MODULE_PARM_DESC(noretry,
+	"Disable retry faults (0 = retry enabled, 1 = retry disabled, -1 auto (default))");
+module_param_named(noretry, amdgpu_noretry, int, 0644);
+
+/**
+ * DOC: force_asic_type (int)
+ * A non negative value used to specify the asic type for all supported GPUs.
+ */
+MODULE_PARM_DESC(force_asic_type,
+	"A non negative value used to specify the asic type for all supported GPUs");
+module_param_named_unsafe(force_asic_type, amdgpu_force_asic_type, int, 0444);
+
+/**
+ * DOC: use_xgmi_p2p (int)
+ * Enables/disables XGMI P2P interface (0 = disable, 1 = enable).
+ */
+MODULE_PARM_DESC(use_xgmi_p2p,
+	"Enable XGMI P2P interface (0 = disable; 1 = enable (default))");
+module_param_named(use_xgmi_p2p, amdgpu_use_xgmi_p2p, int, 0444);
+
+
+#ifdef CONFIG_HSA_AMD
+/**
+ * DOC: sched_policy (int)
+ * Set scheduling policy. Default is HWS(hardware scheduling) with over-subscription.
+ * Setting 1 disables over-subscription. Setting 2 disables HWS and statically
+ * assigns queues to HQDs.
+ */
+int sched_policy = KFD_SCHED_POLICY_HWS;
+module_param_unsafe(sched_policy, int, 0444);
+MODULE_PARM_DESC(sched_policy,
+	"Scheduling policy (0 = HWS (Default), 1 = HWS without over-subscription, 2 = Non-HWS (Used for debugging only)");
+
+/**
+ * DOC: hws_max_conc_proc (int)
+ * Maximum number of processes that HWS can schedule concurrently. The maximum is the
+ * number of VMIDs assigned to the HWS, which is also the default.
+ */
+int hws_max_conc_proc = -1;
+module_param(hws_max_conc_proc, int, 0444);
+MODULE_PARM_DESC(hws_max_conc_proc,
+	"Max # processes HWS can execute concurrently when sched_policy=0 (0 = no concurrency, #VMIDs for KFD = Maximum(default))");
+
+/**
+ * DOC: cwsr_enable (int)
+ * CWSR(compute wave store and resume) allows the GPU to preempt shader execution in
+ * the middle of a compute wave. Default is 1 to enable this feature. Setting 0
+ * disables it.
+ */
+int cwsr_enable = 1;
+module_param(cwsr_enable, int, 0444);
+MODULE_PARM_DESC(cwsr_enable, "CWSR enable (0 = Off, 1 = On (Default))");
+
+/**
+ * DOC: max_num_of_queues_per_device (int)
+ * Maximum number of queues per device. Valid setting is between 1 and 4096. Default
+ * is 4096.
+ */
+int max_num_of_queues_per_device = KFD_MAX_NUM_OF_QUEUES_PER_DEVICE_DEFAULT;
+module_param(max_num_of_queues_per_device, int, 0444);
+MODULE_PARM_DESC(max_num_of_queues_per_device,
+	"Maximum number of supported queues per device (1 = Minimum, 4096 = default)");
+
+/**
+ * DOC: send_sigterm (int)
+ * Send sigterm to HSA process on unhandled exceptions. Default is not to send sigterm
+ * but just print errors on dmesg. Setting 1 enables sending sigterm.
+ */
+int send_sigterm;
+module_param(send_sigterm, int, 0444);
+MODULE_PARM_DESC(send_sigterm,
+	"Send sigterm to HSA process on unhandled exception (0 = disable, 1 = enable)");
+
+/**
+ * DOC: halt_if_hws_hang (int)
+ * Halt if HWS hang is detected. Default value, 0, disables the halt on hang.
+ * Setting 1 enables halt on hang.
+ */
+int halt_if_hws_hang;
+module_param_unsafe(halt_if_hws_hang, int, 0644);
+MODULE_PARM_DESC(halt_if_hws_hang, "Halt if HWS hang is detected (0 = off (default), 1 = on)");
+
+/**
+ * DOC: hws_gws_support(bool)
+ * Assume that HWS supports GWS barriers regardless of what firmware version
+ * check says. Default value: false (rely on MEC2 firmware version check).
+ */
+bool hws_gws_support;
+module_param_unsafe(hws_gws_support, bool, 0444);
+MODULE_PARM_DESC(hws_gws_support, "Assume MEC2 FW supports GWS barriers (false = rely on FW version check (Default), true = force supported)");
+
+/**
+ * DOC: queue_preemption_timeout_ms (int)
+ * queue preemption timeout in ms (1 = Minimum, 9000 = default)
+ */
+int queue_preemption_timeout_ms = 9000;
+module_param(queue_preemption_timeout_ms, int, 0644);
+MODULE_PARM_DESC(queue_preemption_timeout_ms, "queue preemption timeout in ms (1 = Minimum, 9000 = default)");
+
+/**
+ * DOC: debug_evictions(bool)
+ * Enable extra debug messages to help determine the cause of evictions
+ */
+bool debug_evictions;
+module_param(debug_evictions, bool, 0644);
+MODULE_PARM_DESC(debug_evictions, "enable eviction debug messages (false = default)");
+
+/**
+ * DOC: no_system_mem_limit(bool)
+ * Disable system memory limit, to support multiple process shared memory
+ */
+bool no_system_mem_limit;
+module_param(no_system_mem_limit, bool, 0644);
+MODULE_PARM_DESC(no_system_mem_limit, "disable system memory limit (false = default)");
+
+/**
+ * DOC: no_queue_eviction_on_vm_fault (int)
+ * If set, process queues will not be evicted on gpuvm fault. This is to keep the wavefront context for debugging (0 = queue eviction, 1 = no queue eviction). The default is 0 (queue eviction).
+ */
+int amdgpu_no_queue_eviction_on_vm_fault;
+MODULE_PARM_DESC(no_queue_eviction_on_vm_fault, "No queue eviction on VM fault (0 = queue eviction, 1 = no queue eviction)");
+module_param_named_unsafe(no_queue_eviction_on_vm_fault, amdgpu_no_queue_eviction_on_vm_fault, int, 0444);
+#endif
+
+/**
+ * DOC: mtype_local (int)
+ */
+int amdgpu_mtype_local;
+MODULE_PARM_DESC(mtype_local, "MTYPE for local memory (0 = MTYPE_RW (default), 1 = MTYPE_NC, 2 = MTYPE_CC)");
+module_param_named_unsafe(mtype_local, amdgpu_mtype_local, int, 0444);
+
+/**
+ * DOC: pcie_p2p (bool)
+ * Enable PCIe P2P (requires large-BAR). Default value: true (on)
+ */
+#ifdef CONFIG_HSA_AMD_P2P
+bool pcie_p2p = true;
+module_param(pcie_p2p, bool, 0444);
+MODULE_PARM_DESC(pcie_p2p, "Enable PCIe P2P (requires large-BAR). (N = off, Y = on(default))");
+#endif
+
+/**
+ * DOC: dcfeaturemask (uint)
+ * Override display features enabled. See enum DC_FEATURE_MASK in drivers/gpu/drm/amd/include/amd_shared.h.
+ * The default is the current set of stable display features.
+ */
+MODULE_PARM_DESC(dcfeaturemask, "all stable DC features enabled (default))");
+module_param_named(dcfeaturemask, amdgpu_dc_feature_mask, uint, 0444);
+
+/**
+ * DOC: dcdebugmask (uint)
+ * Display debug options. See enum DC_DEBUG_MASK in drivers/gpu/drm/amd/include/amd_shared.h.
+ */
+MODULE_PARM_DESC(dcdebugmask, "all debug options disabled (default))");
+module_param_named(dcdebugmask, amdgpu_dc_debug_mask, uint, 0444);
+
+MODULE_PARM_DESC(visualconfirm, "Visual confirm (0 = off (default), 1 = MPO, 5 = PSR)");
+module_param_named(visualconfirm, amdgpu_dc_visual_confirm, uint, 0444);
+
+/**
+ * DOC: abmlevel (uint)
+ * Override the default ABM (Adaptive Backlight Management) level used for DC
+ * enabled hardware. Requires DMCU to be supported and loaded.
+ * Valid levels are 0-4. A value of 0 indicates that ABM should be disabled by
+ * default. Values 1-4 control the maximum allowable brightness reduction via
+ * the ABM algorithm, with 1 being the least reduction and 4 being the most
+ * reduction.
+ *
+ * Defaults to -1, or auto. Userspace can only override this level after
+ * boot if it's set to auto.
+ */
+int amdgpu_dm_abm_level = -1;
+MODULE_PARM_DESC(abmlevel,
+		 "ABM level (0 = off, 1-4 = backlight reduction level, -1 auto (default))");
+module_param_named(abmlevel, amdgpu_dm_abm_level, int, 0444);
+
+int amdgpu_backlight = -1;
+MODULE_PARM_DESC(backlight, "Backlight control (0 = pwm, 1 = aux, -1 auto (default))");
+module_param_named(backlight, amdgpu_backlight, bint, 0444);
+
+/**
+ * DOC: damageclips (int)
+ * Enable or disable damage clips support. If damage clips support is disabled,
+ * we will force full frame updates, irrespective of what user space sends to
+ * us.
+ *
+ * Defaults to -1 (where it is enabled unless a PSR-SU display is detected).
+ */
+MODULE_PARM_DESC(damageclips,
+		 "Damage clips support (0 = disable, 1 = enable, -1 auto (default))");
+module_param_named(damageclips, amdgpu_damage_clips, int, 0444);
+
+/**
+ * DOC: tmz (int)
+ * Trusted Memory Zone (TMZ) is a method to protect data being written
+ * to or read from memory.
+ *
+ * The default value: 0 (off).  TODO: change to auto till it is completed.
+ */
+MODULE_PARM_DESC(tmz, "Enable TMZ feature (-1 = auto (default), 0 = off, 1 = on)");
+module_param_named(tmz, amdgpu_tmz, int, 0444);
+
+/**
+ * DOC: freesync_video (uint)
+ * Enable the optimization to adjust front porch timing to achieve seamless
+ * mode change experience when setting a freesync supported mode for which full
+ * modeset is not needed.
+ *
+ * The Display Core will add a set of modes derived from the base FreeSync
+ * video mode into the corresponding connector's mode list based on commonly
+ * used refresh rates and VRR range of the connected display, when users enable
+ * this feature. From the userspace perspective, they can see a seamless mode
+ * change experience when the change between different refresh rates under the
+ * same resolution. Additionally, userspace applications such as Video playback
+ * can read this modeset list and change the refresh rate based on the video
+ * frame rate. Finally, the userspace can also derive an appropriate mode for a
+ * particular refresh rate based on the FreeSync Mode and add it to the
+ * connector's mode list.
+ *
+ * Note: This is an experimental feature.
+ *
+ * The default value: 0 (off).
+ */
+MODULE_PARM_DESC(
+	freesync_video,
+	"Adds additional modes via VRR for refresh changes without a full modeset (0 = off (default), 1 = on)");
+module_param_named(freesync_video, amdgpu_freesync_vid_mode, uint, 0444);
+
+/**
+ * DOC: reset_method (int)
+ * GPU reset method (-1 = auto (default), 0 = legacy, 1 = mode0, 2 = mode1, 3 = mode2, 4 = baco)
+ */
+MODULE_PARM_DESC(reset_method, "GPU reset method (-1 = auto (default), 0 = legacy, 1 = mode0, 2 = mode1, 3 = mode2, 4 = baco/bamaco)");
+module_param_named_unsafe(reset_method, amdgpu_reset_method, int, 0644);
+
+/**
+ * DOC: bad_page_threshold (int) Bad page threshold is specifies the
+ * threshold value of faulty pages detected by RAS ECC, which may
+ * result in the GPU entering bad status when the number of total
+ * faulty pages by ECC exceeds the threshold value.
+ */
+MODULE_PARM_DESC(bad_page_threshold, "Bad page threshold(-1 = ignore threshold (default value), 0 = disable bad page retirement, -2 = threshold determined by a formula, 0 < threshold < max records, user-defined threshold)");
+module_param_named(bad_page_threshold, amdgpu_bad_page_threshold, int, 0444);
+
+MODULE_PARM_DESC(num_kcq, "number of kernel compute queue user want to setup (8 if set to greater than 8 or less than 0, only affect gfx 8+)");
+module_param_named(num_kcq, amdgpu_num_kcq, int, 0444);
+
+/**
+ * DOC: vcnfw_log (int)
+ * Enable vcnfw log output for debugging, the default is disabled.
+ */
+MODULE_PARM_DESC(vcnfw_log, "Enable vcnfw log(0 = disable (default value), 1 = enable)");
+module_param_named(vcnfw_log, amdgpu_vcnfw_log, int, 0444);
+
+/**
+ * DOC: sg_display (int)
+ * Disable S/G (scatter/gather) display (i.e., display from system memory).
+ * This option is only relevant on APUs.  Set this option to 0 to disable
+ * S/G display if you experience flickering or other issues under memory
+ * pressure and report the issue.
+ */
+MODULE_PARM_DESC(sg_display, "S/G Display (-1 = auto (default), 0 = disable)");
+module_param_named(sg_display, amdgpu_sg_display, int, 0444);
+
+/**
+ * DOC: umsch_mm (int)
+ * Enable Multi Media User Mode Scheduler. This is a HW scheduling engine for VCN and VPE.
+ * (0 = disabled (default), 1 = enabled)
+ */
+MODULE_PARM_DESC(umsch_mm,
+	"Enable Multi Media User Mode Scheduler (0 = disabled (default), 1 = enabled)");
+module_param_named(umsch_mm, amdgpu_umsch_mm, int, 0444);
+
+/**
+ * DOC: umsch_mm_fwlog (int)
+ * Enable umschfw log output for debugging, the default is disabled.
+ */
+MODULE_PARM_DESC(umsch_mm_fwlog, "Enable umschfw log(0 = disable (default value), 1 = enable)");
+module_param_named(umsch_mm_fwlog, amdgpu_umsch_mm_fwlog, int, 0444);
+
+/**
+ * DOC: smu_pptable_id (int)
+ * Used to override pptable id. id = 0 use VBIOS pptable.
+ * id > 0 use the soft pptable with specicfied id.
+ */
+MODULE_PARM_DESC(smu_pptable_id,
+	"specify pptable id to be used (-1 = auto(default) value, 0 = use pptable from vbios, > 0 = soft pptable id)");
+module_param_named(smu_pptable_id, amdgpu_smu_pptable_id, int, 0444);
+
+/**
+ * DOC: partition_mode (int)
+ * Used to override the default SPX mode.
+ */
+MODULE_PARM_DESC(
+	user_partt_mode,
+	"specify partition mode to be used (-2 = AMDGPU_AUTO_COMPUTE_PARTITION_MODE(default value) \
+						0 = AMDGPU_SPX_PARTITION_MODE, \
+						1 = AMDGPU_DPX_PARTITION_MODE, \
+						2 = AMDGPU_TPX_PARTITION_MODE, \
+						3 = AMDGPU_QPX_PARTITION_MODE, \
+						4 = AMDGPU_CPX_PARTITION_MODE)");
+module_param_named(user_partt_mode, amdgpu_user_partt_mode, uint, 0444);
+
 
-MODULE_PARM_DESC(prim_buf_per_se, "the size of Primitive Buffer per Shader Engine (default depending on gfx)");
-module_param_named(prim_buf_per_se, amdgpu_prim_buf_per_se, int, 0444);
+/**
+ * DOC: enforce_isolation (int)
+ * enforce process isolation between graphics and compute.
+ * (-1 = auto, 0 = disable, 1 = enable, 2 = enable legacy mode, 3 = enable without cleaner shader)
+ */
+module_param_named(enforce_isolation, amdgpu_enforce_isolation, int, 0444);
+MODULE_PARM_DESC(enforce_isolation,
+"enforce process isolation between graphics and compute. (-1 = auto, 0 = disable, 1 = enable, 2 = enable legacy mode, 3 = enable without cleaner shader)");
+
+/**
+ * DOC: modeset (int)
+ * Override nomodeset (1 = override, -1 = auto). The default is -1 (auto).
+ */
+MODULE_PARM_DESC(modeset, "Override nomodeset (1 = enable, -1 = auto)");
+module_param_named(modeset, amdgpu_modeset, int, 0444);
+
+/**
+ * DOC: seamless (int)
+ * Seamless boot will keep the image on the screen during the boot process.
+ */
+MODULE_PARM_DESC(seamless, "Seamless boot (-1 = auto (default), 0 = disable, 1 = enable)");
+module_param_named(seamless, amdgpu_seamless, int, 0444);
 
-MODULE_PARM_DESC(pos_buf_per_se, "the size of Position Buffer per Shader Engine (default depending on gfx)");
-module_param_named(pos_buf_per_se, amdgpu_pos_buf_per_se, int, 0444);
+/**
+ * DOC: debug_mask (uint)
+ * Debug options for amdgpu, work as a binary mask with the following options:
+ *
+ * - 0x1: Debug VM handling
+ * - 0x2: Enable simulating large-bar capability on non-large bar system. This
+ *   limits the VRAM size reported to ROCm applications to the visible
+ *   size, usually 256MB.
+ * - 0x4: Disable GPU soft recovery, always do a full reset
+ * - 0x8: Use VRAM for firmware loading
+ * - 0x10: Enable ACA based RAS logging
+ * - 0x20: Enable experimental resets
+ * - 0x40: Disable ring resets
+ * - 0x80: Use VRAM for SMU pool
+ */
+MODULE_PARM_DESC(debug_mask, "debug options for amdgpu, disabled by default");
+module_param_named_unsafe(debug_mask, amdgpu_debug_mask, uint, 0444);
+
+/**
+ * DOC: agp (int)
+ * Enable the AGP aperture.  This provides an aperture in the GPU's internal
+ * address space for direct access to system memory.  Note that these accesses
+ * are non-snooped, so they are only used for access to uncached memory.
+ */
+MODULE_PARM_DESC(agp, "AGP (-1 = auto (default), 0 = disable, 1 = enable)");
+module_param_named(agp, amdgpu_agp, int, 0444);
+
+/**
+ * DOC: wbrf (int)
+ * Enable Wifi RFI interference mitigation feature.
+ * Due to electrical and mechanical constraints there may be likely interference of
+ * relatively high-powered harmonics of the (G-)DDR memory clocks with local radio
+ * module frequency bands used by Wifi 6/6e/7. To mitigate the possible RFI interference,
+ * with this feature enabled, PMFW will use either “shadowed P-State” or “P-State” based
+ * on active list of frequencies in-use (to be avoided) as part of initial setting or
+ * P-state transition. However, there may be potential performance impact with this
+ * feature enabled.
+ * (0 = disabled, 1 = enabled, -1 = auto (default setting, will be enabled if supported))
+ */
+MODULE_PARM_DESC(wbrf,
+	"Enable Wifi RFI interference mitigation (0 = disabled, 1 = enabled, -1 = auto(default)");
+module_param_named(wbrf, amdgpu_wbrf, int, 0444);
 
-MODULE_PARM_DESC(cntl_sb_buf_per_se, "the size of Control Sideband per Shader Engine (default depending on gfx)");
-module_param_named(cntl_sb_buf_per_se, amdgpu_cntl_sb_buf_per_se, int, 0444);
+/**
+ * DOC: rebar (int)
+ * Allow BAR resizing.  Disable this to prevent the driver from attempting
+ * to resize the BAR if the GPU supports it and there is available MMIO space.
+ * Note that this just prevents the driver from resizing the BAR.  The BIOS
+ * may have already resized the BAR at boot time.
+ */
+MODULE_PARM_DESC(rebar, "Resizable BAR (-1 = auto (default), 0 = disable, 1 = enable)");
+module_param_named(rebar, amdgpu_rebar, int, 0444);
 
-MODULE_PARM_DESC(param_buf_per_se, "the size of Off-Chip Pramater Cache per Shader Engine (default depending on gfx)");
-module_param_named(param_buf_per_se, amdgpu_param_buf_per_se, int, 0444);
+/**
+ * DOC: user_queue (int)
+ * Enable user queues on systems that support user queues. Possible values:
+ *
+ * - -1 = auto (ASIC specific default)
+ * -  0 = user queues disabled
+ * -  1 = user queues enabled and kernel queues enabled (if supported)
+ * -  2 = user queues enabled and kernel queues disabled
+ */
+MODULE_PARM_DESC(user_queue, "Enable user queues (-1 = auto (default), 0 = disable, 1 = enable, 2 = enable UQs and disable KQs)");
+module_param_named(user_queue, amdgpu_user_queue, int, 0444);
 
+/* These devices are not supported by amdgpu.
+ * They are supported by the mach64, r128, radeon drivers
+ */
+static const u16 amdgpu_unsupported_pciidlist[] = {
+	/* mach64 */
+	0x4354,
+	0x4358,
+	0x4554,
+	0x4742,
+	0x4744,
+	0x4749,
+	0x474C,
+	0x474D,
+	0x474E,
+	0x474F,
+	0x4750,
+	0x4751,
+	0x4752,
+	0x4753,
+	0x4754,
+	0x4755,
+	0x4756,
+	0x4757,
+	0x4758,
+	0x4759,
+	0x475A,
+	0x4C42,
+	0x4C44,
+	0x4C47,
+	0x4C49,
+	0x4C4D,
+	0x4C4E,
+	0x4C50,
+	0x4C51,
+	0x4C52,
+	0x4C53,
+	0x5654,
+	0x5655,
+	0x5656,
+	/* r128 */
+	0x4c45,
+	0x4c46,
+	0x4d46,
+	0x4d4c,
+	0x5041,
+	0x5042,
+	0x5043,
+	0x5044,
+	0x5045,
+	0x5046,
+	0x5047,
+	0x5048,
+	0x5049,
+	0x504A,
+	0x504B,
+	0x504C,
+	0x504D,
+	0x504E,
+	0x504F,
+	0x5050,
+	0x5051,
+	0x5052,
+	0x5053,
+	0x5054,
+	0x5055,
+	0x5056,
+	0x5057,
+	0x5058,
+	0x5245,
+	0x5246,
+	0x5247,
+	0x524b,
+	0x524c,
+	0x534d,
+	0x5446,
+	0x544C,
+	0x5452,
+	/* radeon */
+	0x3150,
+	0x3151,
+	0x3152,
+	0x3154,
+	0x3155,
+	0x3E50,
+	0x3E54,
+	0x4136,
+	0x4137,
+	0x4144,
+	0x4145,
+	0x4146,
+	0x4147,
+	0x4148,
+	0x4149,
+	0x414A,
+	0x414B,
+	0x4150,
+	0x4151,
+	0x4152,
+	0x4153,
+	0x4154,
+	0x4155,
+	0x4156,
+	0x4237,
+	0x4242,
+	0x4336,
+	0x4337,
+	0x4437,
+	0x4966,
+	0x4967,
+	0x4A48,
+	0x4A49,
+	0x4A4A,
+	0x4A4B,
+	0x4A4C,
+	0x4A4D,
+	0x4A4E,
+	0x4A4F,
+	0x4A50,
+	0x4A54,
+	0x4B48,
+	0x4B49,
+	0x4B4A,
+	0x4B4B,
+	0x4B4C,
+	0x4C57,
+	0x4C58,
+	0x4C59,
+	0x4C5A,
+	0x4C64,
+	0x4C66,
+	0x4C67,
+	0x4E44,
+	0x4E45,
+	0x4E46,
+	0x4E47,
+	0x4E48,
+	0x4E49,
+	0x4E4A,
+	0x4E4B,
+	0x4E50,
+	0x4E51,
+	0x4E52,
+	0x4E53,
+	0x4E54,
+	0x4E56,
+	0x5144,
+	0x5145,
+	0x5146,
+	0x5147,
+	0x5148,
+	0x514C,
+	0x514D,
+	0x5157,
+	0x5158,
+	0x5159,
+	0x515A,
+	0x515E,
+	0x5460,
+	0x5462,
+	0x5464,
+	0x5548,
+	0x5549,
+	0x554A,
+	0x554B,
+	0x554C,
+	0x554D,
+	0x554E,
+	0x554F,
+	0x5550,
+	0x5551,
+	0x5552,
+	0x5554,
+	0x564A,
+	0x564B,
+	0x564F,
+	0x5652,
+	0x5653,
+	0x5657,
+	0x5834,
+	0x5835,
+	0x5954,
+	0x5955,
+	0x5974,
+	0x5975,
+	0x5960,
+	0x5961,
+	0x5962,
+	0x5964,
+	0x5965,
+	0x5969,
+	0x5a41,
+	0x5a42,
+	0x5a61,
+	0x5a62,
+	0x5b60,
+	0x5b62,
+	0x5b63,
+	0x5b64,
+	0x5b65,
+	0x5c61,
+	0x5c63,
+	0x5d48,
+	0x5d49,
+	0x5d4a,
+	0x5d4c,
+	0x5d4d,
+	0x5d4e,
+	0x5d4f,
+	0x5d50,
+	0x5d52,
+	0x5d57,
+	0x5e48,
+	0x5e4a,
+	0x5e4b,
+	0x5e4c,
+	0x5e4d,
+	0x5e4f,
+	0x6700,
+	0x6701,
+	0x6702,
+	0x6703,
+	0x6704,
+	0x6705,
+	0x6706,
+	0x6707,
+	0x6708,
+	0x6709,
+	0x6718,
+	0x6719,
+	0x671c,
+	0x671d,
+	0x671f,
+	0x6720,
+	0x6721,
+	0x6722,
+	0x6723,
+	0x6724,
+	0x6725,
+	0x6726,
+	0x6727,
+	0x6728,
+	0x6729,
+	0x6738,
+	0x6739,
+	0x673e,
+	0x6740,
+	0x6741,
+	0x6742,
+	0x6743,
+	0x6744,
+	0x6745,
+	0x6746,
+	0x6747,
+	0x6748,
+	0x6749,
+	0x674A,
+	0x6750,
+	0x6751,
+	0x6758,
+	0x6759,
+	0x675B,
+	0x675D,
+	0x675F,
+	0x6760,
+	0x6761,
+	0x6762,
+	0x6763,
+	0x6764,
+	0x6765,
+	0x6766,
+	0x6767,
+	0x6768,
+	0x6770,
+	0x6771,
+	0x6772,
+	0x6778,
+	0x6779,
+	0x677B,
+	0x6840,
+	0x6841,
+	0x6842,
+	0x6843,
+	0x6849,
+	0x684C,
+	0x6850,
+	0x6858,
+	0x6859,
+	0x6880,
+	0x6888,
+	0x6889,
+	0x688A,
+	0x688C,
+	0x688D,
+	0x6898,
+	0x6899,
+	0x689b,
+	0x689c,
+	0x689d,
+	0x689e,
+	0x68a0,
+	0x68a1,
+	0x68a8,
+	0x68a9,
+	0x68b0,
+	0x68b8,
+	0x68b9,
+	0x68ba,
+	0x68be,
+	0x68bf,
+	0x68c0,
+	0x68c1,
+	0x68c7,
+	0x68c8,
+	0x68c9,
+	0x68d8,
+	0x68d9,
+	0x68da,
+	0x68de,
+	0x68e0,
+	0x68e1,
+	0x68e4,
+	0x68e5,
+	0x68e8,
+	0x68e9,
+	0x68f1,
+	0x68f2,
+	0x68f8,
+	0x68f9,
+	0x68fa,
+	0x68fe,
+	0x7100,
+	0x7101,
+	0x7102,
+	0x7103,
+	0x7104,
+	0x7105,
+	0x7106,
+	0x7108,
+	0x7109,
+	0x710A,
+	0x710B,
+	0x710C,
+	0x710E,
+	0x710F,
+	0x7140,
+	0x7141,
+	0x7142,
+	0x7143,
+	0x7144,
+	0x7145,
+	0x7146,
+	0x7147,
+	0x7149,
+	0x714A,
+	0x714B,
+	0x714C,
+	0x714D,
+	0x714E,
+	0x714F,
+	0x7151,
+	0x7152,
+	0x7153,
+	0x715E,
+	0x715F,
+	0x7180,
+	0x7181,
+	0x7183,
+	0x7186,
+	0x7187,
+	0x7188,
+	0x718A,
+	0x718B,
+	0x718C,
+	0x718D,
+	0x718F,
+	0x7193,
+	0x7196,
+	0x719B,
+	0x719F,
+	0x71C0,
+	0x71C1,
+	0x71C2,
+	0x71C3,
+	0x71C4,
+	0x71C5,
+	0x71C6,
+	0x71C7,
+	0x71CD,
+	0x71CE,
+	0x71D2,
+	0x71D4,
+	0x71D5,
+	0x71D6,
+	0x71DA,
+	0x71DE,
+	0x7200,
+	0x7210,
+	0x7211,
+	0x7240,
+	0x7243,
+	0x7244,
+	0x7245,
+	0x7246,
+	0x7247,
+	0x7248,
+	0x7249,
+	0x724A,
+	0x724B,
+	0x724C,
+	0x724D,
+	0x724E,
+	0x724F,
+	0x7280,
+	0x7281,
+	0x7283,
+	0x7284,
+	0x7287,
+	0x7288,
+	0x7289,
+	0x728B,
+	0x728C,
+	0x7290,
+	0x7291,
+	0x7293,
+	0x7297,
+	0x7834,
+	0x7835,
+	0x791e,
+	0x791f,
+	0x793f,
+	0x7941,
+	0x7942,
+	0x796c,
+	0x796d,
+	0x796e,
+	0x796f,
+	0x9400,
+	0x9401,
+	0x9402,
+	0x9403,
+	0x9405,
+	0x940A,
+	0x940B,
+	0x940F,
+	0x94A0,
+	0x94A1,
+	0x94A3,
+	0x94B1,
+	0x94B3,
+	0x94B4,
+	0x94B5,
+	0x94B9,
+	0x9440,
+	0x9441,
+	0x9442,
+	0x9443,
+	0x9444,
+	0x9446,
+	0x944A,
+	0x944B,
+	0x944C,
+	0x944E,
+	0x9450,
+	0x9452,
+	0x9456,
+	0x945A,
+	0x945B,
+	0x945E,
+	0x9460,
+	0x9462,
+	0x946A,
+	0x946B,
+	0x947A,
+	0x947B,
+	0x9480,
+	0x9487,
+	0x9488,
+	0x9489,
+	0x948A,
+	0x948F,
+	0x9490,
+	0x9491,
+	0x9495,
+	0x9498,
+	0x949C,
+	0x949E,
+	0x949F,
+	0x94C0,
+	0x94C1,
+	0x94C3,
+	0x94C4,
+	0x94C5,
+	0x94C6,
+	0x94C7,
+	0x94C8,
+	0x94C9,
+	0x94CB,
+	0x94CC,
+	0x94CD,
+	0x9500,
+	0x9501,
+	0x9504,
+	0x9505,
+	0x9506,
+	0x9507,
+	0x9508,
+	0x9509,
+	0x950F,
+	0x9511,
+	0x9515,
+	0x9517,
+	0x9519,
+	0x9540,
+	0x9541,
+	0x9542,
+	0x954E,
+	0x954F,
+	0x9552,
+	0x9553,
+	0x9555,
+	0x9557,
+	0x955f,
+	0x9580,
+	0x9581,
+	0x9583,
+	0x9586,
+	0x9587,
+	0x9588,
+	0x9589,
+	0x958A,
+	0x958B,
+	0x958C,
+	0x958D,
+	0x958E,
+	0x958F,
+	0x9590,
+	0x9591,
+	0x9593,
+	0x9595,
+	0x9596,
+	0x9597,
+	0x9598,
+	0x9599,
+	0x959B,
+	0x95C0,
+	0x95C2,
+	0x95C4,
+	0x95C5,
+	0x95C6,
+	0x95C7,
+	0x95C9,
+	0x95CC,
+	0x95CD,
+	0x95CE,
+	0x95CF,
+	0x9610,
+	0x9611,
+	0x9612,
+	0x9613,
+	0x9614,
+	0x9615,
+	0x9616,
+	0x9640,
+	0x9641,
+	0x9642,
+	0x9643,
+	0x9644,
+	0x9645,
+	0x9647,
+	0x9648,
+	0x9649,
+	0x964a,
+	0x964b,
+	0x964c,
+	0x964e,
+	0x964f,
+	0x9710,
+	0x9711,
+	0x9712,
+	0x9713,
+	0x9714,
+	0x9715,
+	0x9802,
+	0x9803,
+	0x9804,
+	0x9805,
+	0x9806,
+	0x9807,
+	0x9808,
+	0x9809,
+	0x980A,
+	0x9900,
+	0x9901,
+	0x9903,
+	0x9904,
+	0x9905,
+	0x9906,
+	0x9907,
+	0x9908,
+	0x9909,
+	0x990A,
+	0x990B,
+	0x990C,
+	0x990D,
+	0x990E,
+	0x990F,
+	0x9910,
+	0x9913,
+	0x9917,
+	0x9918,
+	0x9919,
+	0x9990,
+	0x9991,
+	0x9992,
+	0x9993,
+	0x9994,
+	0x9995,
+	0x9996,
+	0x9997,
+	0x9998,
+	0x9999,
+	0x999A,
+	0x999B,
+	0x999C,
+	0x999D,
+	0x99A0,
+	0x99A2,
+	0x99A4,
+	/* radeon secondary ids */
+	0x3171,
+	0x3e70,
+	0x4164,
+	0x4165,
+	0x4166,
+	0x4168,
+	0x4170,
+	0x4171,
+	0x4172,
+	0x4173,
+	0x496e,
+	0x4a69,
+	0x4a6a,
+	0x4a6b,
+	0x4a70,
+	0x4a74,
+	0x4b69,
+	0x4b6b,
+	0x4b6c,
+	0x4c6e,
+	0x4e64,
+	0x4e65,
+	0x4e66,
+	0x4e67,
+	0x4e68,
+	0x4e69,
+	0x4e6a,
+	0x4e71,
+	0x4f73,
+	0x5569,
+	0x556b,
+	0x556d,
+	0x556f,
+	0x5571,
+	0x5854,
+	0x5874,
+	0x5940,
+	0x5941,
+	0x5b70,
+	0x5b72,
+	0x5b73,
+	0x5b74,
+	0x5b75,
+	0x5d44,
+	0x5d45,
+	0x5d6d,
+	0x5d6f,
+	0x5d72,
+	0x5d77,
+	0x5e6b,
+	0x5e6d,
+	0x7120,
+	0x7124,
+	0x7129,
+	0x712e,
+	0x712f,
+	0x7162,
+	0x7163,
+	0x7166,
+	0x7167,
+	0x7172,
+	0x7173,
+	0x71a0,
+	0x71a1,
+	0x71a3,
+	0x71a7,
+	0x71bb,
+	0x71e0,
+	0x71e1,
+	0x71e2,
+	0x71e6,
+	0x71e7,
+	0x71f2,
+	0x7269,
+	0x726b,
+	0x726e,
+	0x72a0,
+	0x72a8,
+	0x72b1,
+	0x72b3,
+	0x793f,
+};
 
 static const struct pci_device_id pciidlist[] = {
-#ifdef  CONFIG_DRM_AMDGPU_SI
 	{0x1002, 0x6780, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TAHITI},
 	{0x1002, 0x6784, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TAHITI},
 	{0x1002, 0x6788, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TAHITI},
@@ -307,8 +1907,6 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x6665, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_HAINAN|AMD_IS_MOBILITY},
 	{0x1002, 0x6667, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_HAINAN|AMD_IS_MOBILITY},
 	{0x1002, 0x666F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_HAINAN|AMD_IS_MOBILITY},
-#endif
-#ifdef CONFIG_DRM_AMDGPU_CIK
 	/* Kaveri */
 	{0x1002, 0x1304, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_MOBILITY|AMD_IS_APU},
 	{0x1002, 0x1305, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_KAVERI|AMD_IS_APU},
@@ -391,7 +1989,6 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x985D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
 	{0x1002, 0x985E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
 	{0x1002, 0x985F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU},
-#endif
 	/* topaz */
 	{0x1002, 0x6900, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ},
 	{0x1002, 0x6901, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ},
@@ -442,6 +2039,7 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x67CA, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS10},
 	{0x1002, 0x67CC, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS10},
 	{0x1002, 0x67CF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS10},
+	{0x1002, 0x6FDF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS10},
 	/* Polaris12 */
 	{0x1002, 0x6980, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
 	{0x1002, 0x6981, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
@@ -449,50 +2047,358 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x6986, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
 	{0x1002, 0x6987, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
 	{0x1002, 0x6995, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
+	{0x1002, 0x6997, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
 	{0x1002, 0x699F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
+	/* VEGAM */
+	{0x1002, 0x694C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM},
+	{0x1002, 0x694E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM},
+	{0x1002, 0x694F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGAM},
 	/* Vega 10 */
-	{0x1002, 0x6860, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x6861, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x6862, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x6863, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x6864, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x6867, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x6868, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x686c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x687f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10|AMD_EXP_HW_SUPPORT},
+	{0x1002, 0x6860, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x6861, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x6862, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x6863, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x6864, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x6867, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x6868, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x6869, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x686a, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x686b, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x686c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x686d, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x686e, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x686f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	{0x1002, 0x687f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
+	/* Vega 12 */
+	{0x1002, 0x69A0, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
+	{0x1002, 0x69A1, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
+	{0x1002, 0x69A2, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
+	{0x1002, 0x69A3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
+	{0x1002, 0x69AF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA12},
+	/* Vega 20 */
+	{0x1002, 0x66A0, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20},
+	{0x1002, 0x66A1, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20},
+	{0x1002, 0x66A2, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20},
+	{0x1002, 0x66A3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20},
+	{0x1002, 0x66A4, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20},
+	{0x1002, 0x66A7, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20},
+	{0x1002, 0x66AF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA20},
+	/* Raven */
+	{0x1002, 0x15dd, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RAVEN|AMD_IS_APU},
+	{0x1002, 0x15d8, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RAVEN|AMD_IS_APU},
+	/* Arcturus */
+	{0x1002, 0x738C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ARCTURUS},
+	{0x1002, 0x7388, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ARCTURUS},
+	{0x1002, 0x738E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ARCTURUS},
+	{0x1002, 0x7390, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ARCTURUS},
+	/* Navi10 */
+	{0x1002, 0x7310, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	{0x1002, 0x7312, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	{0x1002, 0x7318, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	{0x1002, 0x7319, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	{0x1002, 0x731A, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	{0x1002, 0x731B, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	{0x1002, 0x731E, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	{0x1002, 0x731F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI10},
+	/* Navi14 */
+	{0x1002, 0x7340, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14},
+	{0x1002, 0x7341, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14},
+	{0x1002, 0x7347, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14},
+	{0x1002, 0x734F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14},
+
+	/* Renoir */
+	{0x1002, 0x15E7, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RENOIR|AMD_IS_APU},
+	{0x1002, 0x1636, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RENOIR|AMD_IS_APU},
+	{0x1002, 0x1638, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RENOIR|AMD_IS_APU},
+	{0x1002, 0x164C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RENOIR|AMD_IS_APU},
+
+	/* Navi12 */
+	{0x1002, 0x7360, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI12},
+	{0x1002, 0x7362, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI12},
+
+	/* Sienna_Cichlid */
+	{0x1002, 0x73A0, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73A1, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73A2, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73A3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73A5, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73A8, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73A9, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73AB, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73AC, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73AD, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73AE, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73AF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+	{0x1002, 0x73BF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID},
+
+	/* Yellow Carp */
+	{0x1002, 0x164D, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_YELLOW_CARP|AMD_IS_APU},
+	{0x1002, 0x1681, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_YELLOW_CARP|AMD_IS_APU},
+
+	/* Navy_Flounder */
+	{0x1002, 0x73C0, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73C1, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73C3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73DA, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73DB, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73DC, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73DD, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73DE, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+	{0x1002, 0x73DF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVY_FLOUNDER},
+
+	/* DIMGREY_CAVEFISH */
+	{0x1002, 0x73E0, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73E1, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73E2, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73E3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73E8, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73E9, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73EA, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73EB, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73EC, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73ED, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73EF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+	{0x1002, 0x73FF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_DIMGREY_CAVEFISH},
+
+	/* Aldebaran */
+	{0x1002, 0x7408, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ALDEBARAN},
+	{0x1002, 0x740C, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ALDEBARAN},
+	{0x1002, 0x740F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ALDEBARAN},
+	{0x1002, 0x7410, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_ALDEBARAN},
+
+	/* CYAN_SKILLFISH */
+	{0x1002, 0x13DB, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CYAN_SKILLFISH|AMD_IS_APU},
+	{0x1002, 0x13F9, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CYAN_SKILLFISH|AMD_IS_APU},
+	{0x1002, 0x13FA, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CYAN_SKILLFISH|AMD_IS_APU},
+	{0x1002, 0x13FB, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CYAN_SKILLFISH|AMD_IS_APU},
+	{0x1002, 0x13FC, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CYAN_SKILLFISH|AMD_IS_APU},
+	{0x1002, 0x13FE, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CYAN_SKILLFISH|AMD_IS_APU},
+	{0x1002, 0x143F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CYAN_SKILLFISH|AMD_IS_APU},
+
+	/* BEIGE_GOBY */
+	{0x1002, 0x7420, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BEIGE_GOBY},
+	{0x1002, 0x7421, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BEIGE_GOBY},
+	{0x1002, 0x7422, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BEIGE_GOBY},
+	{0x1002, 0x7423, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BEIGE_GOBY},
+	{0x1002, 0x7424, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BEIGE_GOBY},
+	{0x1002, 0x743F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_BEIGE_GOBY},
+
+	{ PCI_DEVICE(0x1002, PCI_ANY_ID),
+	  .class = PCI_CLASS_DISPLAY_VGA << 8,
+	  .class_mask = 0xffffff,
+	  .driver_data = CHIP_IP_DISCOVERY },
+
+	{ PCI_DEVICE(0x1002, PCI_ANY_ID),
+	  .class = PCI_CLASS_DISPLAY_OTHER << 8,
+	  .class_mask = 0xffffff,
+	  .driver_data = CHIP_IP_DISCOVERY },
+
+	{ PCI_DEVICE(0x1002, PCI_ANY_ID),
+	  .class = PCI_CLASS_ACCELERATOR_PROCESSING << 8,
+	  .class_mask = 0xffffff,
+	  .driver_data = CHIP_IP_DISCOVERY },
+
 	{0, 0, 0}
 };
 
 MODULE_DEVICE_TABLE(pci, pciidlist);
 
-static struct drm_driver kms_driver;
+static const struct amdgpu_asic_type_quirk asic_type_quirks[] = {
+	/* differentiate between P10 and P11 asics with the same DID */
+	{0x67FF, 0xE3, CHIP_POLARIS10},
+	{0x67FF, 0xE7, CHIP_POLARIS10},
+	{0x67FF, 0xF3, CHIP_POLARIS10},
+	{0x67FF, 0xF7, CHIP_POLARIS10},
+};
+
+static const struct drm_driver amdgpu_kms_driver;
 
-static int amdgpu_kick_out_firmware_fb(struct pci_dev *pdev)
+static void amdgpu_get_secondary_funcs(struct amdgpu_device *adev)
 {
-	struct apertures_struct *ap;
-	bool primary = false;
+	struct pci_dev *p = NULL;
+	int i;
 
-	ap = alloc_apertures(1);
-	if (!ap)
-		return -ENOMEM;
+	/* 0 - GPU
+	 * 1 - audio
+	 * 2 - USB
+	 * 3 - UCSI
+	 */
+	for (i = 1; i < 4; i++) {
+		p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus),
+						adev->pdev->bus->number, i);
+		if (p) {
+			pm_runtime_get_sync(&p->dev);
+			pm_runtime_put_autosuspend(&p->dev);
+			pci_dev_put(p);
+		}
+	}
+}
 
-	ap->ranges[0].base = pci_resource_start(pdev, 0);
-	ap->ranges[0].size = pci_resource_len(pdev, 0);
+static void amdgpu_init_debug_options(struct amdgpu_device *adev)
+{
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_VM) {
+		pr_info("debug: VM handling debug enabled\n");
+		adev->debug_vm = true;
+	}
 
-#ifdef CONFIG_X86
-	primary = pdev->resource[PCI_ROM_RESOURCE].flags & IORESOURCE_ROM_SHADOW;
-#endif
-	drm_fb_helper_remove_conflicting_framebuffers(ap, "amdgpudrmfb", primary);
-	kfree(ap);
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_LARGEBAR) {
+		pr_info("debug: enabled simulating large-bar capability on non-large bar system\n");
+		adev->debug_largebar = true;
+	}
 
-	return 0;
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_DISABLE_GPU_SOFT_RECOVERY) {
+		pr_info("debug: soft reset for GPU recovery disabled\n");
+		adev->debug_disable_soft_recovery = true;
+	}
+
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_USE_VRAM_FW_BUF) {
+		pr_info("debug: place fw in vram for frontdoor loading\n");
+		adev->debug_use_vram_fw_buf = true;
+	}
+
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_ENABLE_RAS_ACA) {
+		pr_info("debug: enable RAS ACA\n");
+		adev->debug_enable_ras_aca = true;
+	}
+
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_ENABLE_EXP_RESETS) {
+		pr_info("debug: enable experimental reset features\n");
+		adev->debug_exp_resets = true;
+	}
+
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_DISABLE_GPU_RING_RESET) {
+		pr_info("debug: ring reset disabled\n");
+		adev->debug_disable_gpu_ring_reset = true;
+	}
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_SMU_POOL) {
+		pr_info("debug: use vram for smu pool\n");
+		adev->pm.smu_debug_mask |= SMU_DEBUG_POOL_USE_VRAM;
+	}
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_VM_USERPTR) {
+		pr_info("debug: VM mode debug for userptr is enabled\n");
+		adev->debug_vm_userptr = true;
+	}
+
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_DISABLE_RAS_CE_LOG) {
+		pr_info("debug: disable kernel logs of correctable errors\n");
+		adev->debug_disable_ce_logs = true;
+	}
+
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_ENABLE_CE_CS) {
+		pr_info("debug: allowing command submission to CE engine\n");
+		adev->debug_enable_ce_cs = true;
+	}
+}
+
+static unsigned long amdgpu_fix_asic_type(struct pci_dev *pdev, unsigned long flags)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(asic_type_quirks); i++) {
+		if (pdev->device == asic_type_quirks[i].device &&
+			pdev->revision == asic_type_quirks[i].revision) {
+				flags &= ~AMD_ASIC_MASK;
+				flags |= asic_type_quirks[i].type;
+				break;
+			}
+	}
+
+	return flags;
+}
+
+static bool amdgpu_support_enabled(struct device *dev,
+				   const enum amd_asic_type family)
+{
+	const char *gen;
+	const char *param;
+	int module_param = -1;
+	bool radeon_support_built = IS_ENABLED(CONFIG_DRM_RADEON);
+	bool amdgpu_support_built = false;
+	bool support_by_default = false;
+
+	switch (family) {
+	case CHIP_TAHITI:
+	case CHIP_PITCAIRN:
+	case CHIP_VERDE:
+	case CHIP_OLAND:
+	case CHIP_HAINAN:
+		gen = "SI";
+		param = "si_support";
+		module_param = amdgpu_si_support;
+		amdgpu_support_built = IS_ENABLED(CONFIG_DRM_AMDGPU_SI);
+		support_by_default = true;
+		break;
+
+	case CHIP_BONAIRE:
+	case CHIP_HAWAII:
+		support_by_default = true;
+		fallthrough;
+	case CHIP_KAVERI:
+	case CHIP_KABINI:
+	case CHIP_MULLINS:
+		gen = "CIK";
+		param = "cik_support";
+		module_param = amdgpu_cik_support;
+		amdgpu_support_built = IS_ENABLED(CONFIG_DRM_AMDGPU_CIK);
+		break;
+
+	default:
+		/* All other chips are supported by amdgpu only */
+		return true;
+	}
+
+	if (!amdgpu_support_built) {
+		dev_info(dev, "amdgpu built without %s support\n", gen);
+		return false;
+	}
+
+	if ((module_param == -1 && (support_by_default || !radeon_support_built)) ||
+	    module_param == 1) {
+		if (radeon_support_built)
+			dev_info(dev, "%s support provided by amdgpu.\n"
+				 "Use radeon.%s=1 amdgpu.%s=0 to override.\n",
+				 gen, param, param);
+
+		return true;
+	}
+
+	if (radeon_support_built)
+		dev_info(dev, "%s support provided by radeon.\n"
+			 "Use radeon.%s=0 amdgpu.%s=1 to override.\n",
+			 gen, param, param);
+	else if (module_param == 0)
+		dev_info(dev, "%s support disabled by module param\n", gen);
+
+	return false;
 }
 
 static int amdgpu_pci_probe(struct pci_dev *pdev,
 			    const struct pci_device_id *ent)
 {
+	struct drm_device *ddev;
+	struct amdgpu_device *adev;
 	unsigned long flags = ent->driver_data;
-	int ret;
+	int ret, retry = 0, i;
+	bool supports_atomic = false;
+
+	if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA ||
+	    (pdev->class >> 8) == PCI_CLASS_DISPLAY_OTHER) {
+		if (drm_firmware_drivers_only() && amdgpu_modeset == -1)
+			return -EINVAL;
+	}
+
+	/* skip devices which are owned by radeon */
+	for (i = 0; i < ARRAY_SIZE(amdgpu_unsupported_pciidlist); i++) {
+		if (amdgpu_unsupported_pciidlist[i] == pdev->device)
+			return -ENODEV;
+	}
+
+	if (amdgpu_aspm == -1 && !pcie_aspm_enabled(pdev))
+		amdgpu_aspm = 0;
+
+	if (amdgpu_virtual_display ||
+	    amdgpu_device_asic_has_dc_support(pdev, flags & AMD_ASIC_MASK))
+		supports_atomic = true;
 
 	if ((flags & AMD_EXP_HW_SUPPORT) && !amdgpu_exp_hw_support) {
 		DRM_INFO("This hardware requires experimental hardware support.\n"
@@ -500,123 +2406,480 @@ static int amdgpu_pci_probe(struct pci_dev *pdev,
 		return -ENODEV;
 	}
 
-	/*
-	 * Initialize amdkfd before starting radeon. If it was not loaded yet,
-	 * defer radeon probing
+	flags = amdgpu_fix_asic_type(pdev, flags);
+
+	/* Due to hardware bugs, S/G Display on raven requires a 1:1 IOMMU mapping,
+	 * however, SME requires an indirect IOMMU mapping because the encryption
+	 * bit is beyond the DMA mask of the chip.
 	 */
-	ret = amdgpu_amdkfd_init();
-	if (ret == -EPROBE_DEFER)
-		return ret;
+	if (cc_platform_has(CC_ATTR_MEM_ENCRYPT) &&
+	    ((flags & AMD_ASIC_MASK) == CHIP_RAVEN)) {
+		dev_info(&pdev->dev,
+			 "SME is not compatible with RAVEN\n");
+		return -ENOTSUPP;
+	}
+
+	if (!amdgpu_support_enabled(&pdev->dev, flags & AMD_ASIC_MASK))
+		return -ENODEV;
 
-	/* Get rid of things like offb */
-	ret = amdgpu_kick_out_firmware_fb(pdev);
+	adev = devm_drm_dev_alloc(&pdev->dev, &amdgpu_kms_driver, typeof(*adev), ddev);
+	if (IS_ERR(adev))
+		return PTR_ERR(adev);
+
+	adev->dev  = &pdev->dev;
+	adev->pdev = pdev;
+	ddev = adev_to_drm(adev);
+
+	if (!supports_atomic)
+		ddev->driver_features &= ~DRIVER_ATOMIC;
+
+	ret = pci_enable_device(pdev);
 	if (ret)
 		return ret;
 
-	return drm_get_pci_dev(pdev, ent, &kms_driver);
+	pci_set_drvdata(pdev, ddev);
+
+	amdgpu_init_debug_options(adev);
+
+	ret = amdgpu_driver_load_kms(adev, flags);
+	if (ret)
+		goto err_pci;
+
+retry_init:
+	ret = drm_dev_register(ddev, flags);
+	if (ret == -EAGAIN && ++retry <= 3) {
+		DRM_INFO("retry init %d\n", retry);
+		/* Don't request EX mode too frequently which is attacking */
+		msleep(5000);
+		goto retry_init;
+	} else if (ret) {
+		goto err_pci;
+	}
+
+	ret = amdgpu_xcp_dev_register(adev, ent);
+	if (ret)
+		goto err_pci;
+
+	ret = amdgpu_amdkfd_drm_client_create(adev);
+	if (ret)
+		goto err_pci;
+
+	/*
+	 * 1. don't init fbdev on hw without DCE
+	 * 2. don't init fbdev if there are no connectors
+	 */
+	if (adev->mode_info.mode_config_initialized &&
+	    !list_empty(&adev_to_drm(adev)->mode_config.connector_list)) {
+		const struct drm_format_info *format;
+
+		/* select 8 bpp console on low vram cards */
+		if (adev->gmc.real_vram_size <= (32*1024*1024))
+			format = drm_format_info(DRM_FORMAT_C8);
+		else
+			format = NULL;
+
+		drm_client_setup(adev_to_drm(adev), format);
+	}
+
+	ret = amdgpu_debugfs_init(adev);
+	if (ret)
+		DRM_ERROR("Creating debugfs files failed (%d).\n", ret);
+
+	if (adev->pm.rpm_mode != AMDGPU_RUNPM_NONE) {
+		/* only need to skip on ATPX */
+		if (amdgpu_device_supports_px(adev))
+			dev_pm_set_driver_flags(ddev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
+		/* we want direct complete for BOCO */
+		if (amdgpu_device_supports_boco(adev))
+			dev_pm_set_driver_flags(ddev->dev, DPM_FLAG_SMART_PREPARE |
+						DPM_FLAG_SMART_SUSPEND |
+						DPM_FLAG_MAY_SKIP_RESUME);
+		pm_runtime_use_autosuspend(ddev->dev);
+		pm_runtime_set_autosuspend_delay(ddev->dev, 5000);
+
+		pm_runtime_allow(ddev->dev);
+
+		pm_runtime_put_autosuspend(ddev->dev);
+
+		pci_wake_from_d3(pdev, TRUE);
+
+		/*
+		 * For runpm implemented via BACO, PMFW will handle the
+		 * timing for BACO in and out:
+		 *   - put ASIC into BACO state only when both video and
+		 *     audio functions are in D3 state.
+		 *   - pull ASIC out of BACO state when either video or
+		 *     audio function is in D0 state.
+		 * Also, at startup, PMFW assumes both functions are in
+		 * D0 state.
+		 *
+		 * So if snd driver was loaded prior to amdgpu driver
+		 * and audio function was put into D3 state, there will
+		 * be no PMFW-aware D-state transition(D0->D3) on runpm
+		 * suspend. Thus the BACO will be not correctly kicked in.
+		 *
+		 * Via amdgpu_get_secondary_funcs(), the audio dev is put
+		 * into D0 state. Then there will be a PMFW-aware D-state
+		 * transition(D0->D3) on runpm suspend.
+		 */
+		if (amdgpu_device_supports_baco(adev) &&
+		    !(adev->flags & AMD_IS_APU) &&
+		    adev->asic_type >= CHIP_NAVI10)
+			amdgpu_get_secondary_funcs(adev);
+	}
+
+	return 0;
+
+err_pci:
+	pci_disable_device(pdev);
+	return ret;
 }
 
 static void
 amdgpu_pci_remove(struct pci_dev *pdev)
 {
 	struct drm_device *dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(dev);
 
-	drm_put_dev(dev);
+	amdgpu_ras_eeprom_check_and_recover(adev);
+	amdgpu_xcp_dev_unplug(adev);
+	amdgpu_gmc_prepare_nps_mode_change(adev);
+	drm_dev_unplug(dev);
+
+	if (adev->pm.rpm_mode != AMDGPU_RUNPM_NONE) {
+		pm_runtime_get_sync(dev->dev);
+		pm_runtime_forbid(dev->dev);
+	}
+
+	amdgpu_driver_unload_kms(dev);
+
+	/*
+	 * Flush any in flight DMA operations from device.
+	 * Clear the Bus Master Enable bit and then wait on the PCIe Device
+	 * StatusTransactions Pending bit.
+	 */
+	pci_disable_device(pdev);
+	pci_wait_for_pending_transaction(pdev);
 }
 
 static void
 amdgpu_pci_shutdown(struct pci_dev *pdev)
 {
 	struct drm_device *dev = pci_get_drvdata(pdev);
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+
+	if (amdgpu_ras_intr_triggered())
+		return;
+
+	/* device maybe not resumed here, return immediately in this case */
+	if (adev->in_s4 && adev->in_suspend)
+		return;
 
 	/* if we are running in a VM, make sure the device
 	 * torn down properly on reboot/shutdown.
 	 * unfortunately we can't detect certain
 	 * hypervisors so just do this all the time.
 	 */
-	amdgpu_suspend(adev);
+	if (!amdgpu_passthrough(adev))
+		adev->mp1_state = PP_MP1_STATE_UNLOAD;
+	amdgpu_device_prepare(dev);
+	amdgpu_device_suspend(dev, true);
+	adev->mp1_state = PP_MP1_STATE_NONE;
+}
+
+static int amdgpu_pmops_prepare(struct device *dev)
+{
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+
+	/* device maybe not resumed here, return immediately in this case */
+	if (adev->in_s4 && adev->in_suspend)
+		return 0;
+
+	/* Return a positive number here so
+	 * DPM_FLAG_SMART_SUSPEND works properly
+	 */
+	if (amdgpu_device_supports_boco(adev) && pm_runtime_suspended(dev))
+		return 1;
+
+	/* if we will not support s3 or s2i for the device
+	 *  then skip suspend
+	 */
+	if (!amdgpu_acpi_is_s0ix_active(adev) &&
+	    !amdgpu_acpi_is_s3_active(adev))
+		return 1;
+
+	return amdgpu_device_prepare(drm_dev);
+}
+
+static void amdgpu_pmops_complete(struct device *dev)
+{
+	amdgpu_device_complete(dev_get_drvdata(dev));
 }
 
 static int amdgpu_pmops_suspend(struct device *dev)
 {
-	struct pci_dev *pdev = to_pci_dev(dev);
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+
+	if (amdgpu_acpi_is_s0ix_active(adev))
+		adev->in_s0ix = true;
+	else if (amdgpu_acpi_is_s3_active(adev))
+		adev->in_s3 = true;
+	if (!adev->in_s0ix && !adev->in_s3) {
+#if IS_ENABLED(CONFIG_SUSPEND)
+		/* don't allow going deep first time followed by s2idle the next time */
+		if (adev->last_suspend_state != PM_SUSPEND_ON &&
+		    adev->last_suspend_state != pm_suspend_target_state) {
+			drm_err_once(drm_dev, "Unsupported suspend state %d\n",
+				     pm_suspend_target_state);
+			return -EINVAL;
+		}
+#endif
+		return 0;
+	}
 
-	struct drm_device *drm_dev = pci_get_drvdata(pdev);
-	return amdgpu_device_suspend(drm_dev, true, true);
+#if IS_ENABLED(CONFIG_SUSPEND)
+	/* cache the state last used for suspend */
+	adev->last_suspend_state = pm_suspend_target_state;
+#endif
+
+	return amdgpu_device_suspend(drm_dev, true);
 }
 
-static int amdgpu_pmops_resume(struct device *dev)
+static int amdgpu_pmops_suspend_noirq(struct device *dev)
 {
-	struct pci_dev *pdev = to_pci_dev(dev);
-	struct drm_device *drm_dev = pci_get_drvdata(pdev);
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+	int r;
 
-	/* GPU comes up enabled by the bios on resume */
-	if (amdgpu_device_is_px(drm_dev)) {
-		pm_runtime_disable(dev);
-		pm_runtime_set_active(dev);
-		pm_runtime_enable(dev);
+	if (amdgpu_acpi_should_gpu_reset(adev)) {
+		amdgpu_device_lock_reset_domain(adev->reset_domain);
+		r = amdgpu_asic_reset(adev);
+		amdgpu_device_unlock_reset_domain(adev->reset_domain);
+		return r;
 	}
 
-	return amdgpu_device_resume(drm_dev, true, true);
+	return 0;
+}
+
+static int amdgpu_pmops_resume(struct device *dev)
+{
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+	int r;
+
+	if (!adev->in_s0ix && !adev->in_s3)
+		return 0;
+
+	/* Avoids registers access if device is physically gone */
+	if (!pci_device_is_present(adev->pdev))
+		adev->no_hw_access = true;
+
+	r = amdgpu_device_resume(drm_dev, true);
+	if (amdgpu_acpi_is_s0ix_active(adev))
+		adev->in_s0ix = false;
+	else
+		adev->in_s3 = false;
+	return r;
 }
 
 static int amdgpu_pmops_freeze(struct device *dev)
 {
-	struct pci_dev *pdev = to_pci_dev(dev);
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+	int r;
 
-	struct drm_device *drm_dev = pci_get_drvdata(pdev);
-	return amdgpu_device_suspend(drm_dev, false, true);
+	r = amdgpu_device_suspend(drm_dev, true);
+	if (r)
+		return r;
+
+	if (amdgpu_acpi_should_gpu_reset(adev))
+		return amdgpu_asic_reset(adev);
+	return 0;
 }
 
 static int amdgpu_pmops_thaw(struct device *dev)
 {
-	struct pci_dev *pdev = to_pci_dev(dev);
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
 
-	struct drm_device *drm_dev = pci_get_drvdata(pdev);
-	return amdgpu_device_resume(drm_dev, false, true);
+	/* do not resume device if it's normal hibernation */
+	if (!pm_hibernate_is_recovering() && !pm_hibernation_mode_is_suspend())
+		return 0;
+
+	return amdgpu_device_resume(drm_dev, true);
 }
 
 static int amdgpu_pmops_poweroff(struct device *dev)
 {
-	struct pci_dev *pdev = to_pci_dev(dev);
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
 
-	struct drm_device *drm_dev = pci_get_drvdata(pdev);
-	return amdgpu_device_suspend(drm_dev, true, true);
+	/* device maybe not resumed here, return immediately in this case */
+	if (adev->in_s4 && adev->in_suspend)
+		return 0;
+
+	return amdgpu_device_suspend(drm_dev, true);
 }
 
 static int amdgpu_pmops_restore(struct device *dev)
 {
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+
+	return amdgpu_device_resume(drm_dev, true);
+}
+
+static int amdgpu_runtime_idle_check_display(struct device *dev)
+{
 	struct pci_dev *pdev = to_pci_dev(dev);
+	struct drm_device *drm_dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+
+	if (adev->mode_info.num_crtc) {
+		struct drm_connector *list_connector;
+		struct drm_connector_list_iter iter;
+		int ret = 0;
+
+		if (amdgpu_runtime_pm != -2) {
+			/* XXX: Return busy if any displays are connected to avoid
+			 * possible display wakeups after runtime resume due to
+			 * hotplug events in case any displays were connected while
+			 * the GPU was in suspend.  Remove this once that is fixed.
+			 */
+			mutex_lock(&drm_dev->mode_config.mutex);
+			drm_connector_list_iter_begin(drm_dev, &iter);
+			drm_for_each_connector_iter(list_connector, &iter) {
+				if (list_connector->status == connector_status_connected) {
+					ret = -EBUSY;
+					break;
+				}
+			}
+			drm_connector_list_iter_end(&iter);
+			mutex_unlock(&drm_dev->mode_config.mutex);
+
+			if (ret)
+				return ret;
+		}
+
+		if (adev->dc_enabled) {
+			struct drm_crtc *crtc;
+
+			drm_for_each_crtc(crtc, drm_dev) {
+				drm_modeset_lock(&crtc->mutex, NULL);
+				if (crtc->state->active)
+					ret = -EBUSY;
+				drm_modeset_unlock(&crtc->mutex);
+				if (ret < 0)
+					break;
+			}
+		} else {
+			mutex_lock(&drm_dev->mode_config.mutex);
+			drm_modeset_lock(&drm_dev->mode_config.connection_mutex, NULL);
+
+			drm_connector_list_iter_begin(drm_dev, &iter);
+			drm_for_each_connector_iter(list_connector, &iter) {
+				if (list_connector->dpms ==  DRM_MODE_DPMS_ON) {
+					ret = -EBUSY;
+					break;
+				}
+			}
+
+			drm_connector_list_iter_end(&iter);
+
+			drm_modeset_unlock(&drm_dev->mode_config.connection_mutex);
+			mutex_unlock(&drm_dev->mode_config.mutex);
+		}
+		if (ret)
+			return ret;
+	}
 
+	return 0;
+}
+
+static int amdgpu_runtime_idle_check_userq(struct device *dev)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
 	struct drm_device *drm_dev = pci_get_drvdata(pdev);
-	return amdgpu_device_resume(drm_dev, false, true);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+
+	return xa_empty(&adev->userq_doorbell_xa) ? 0 : -EBUSY;
 }
 
 static int amdgpu_pmops_runtime_suspend(struct device *dev)
 {
 	struct pci_dev *pdev = to_pci_dev(dev);
 	struct drm_device *drm_dev = pci_get_drvdata(pdev);
-	int ret;
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+	int ret, i;
 
-	if (!amdgpu_device_is_px(drm_dev)) {
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_NONE) {
 		pm_runtime_forbid(dev);
 		return -EBUSY;
 	}
 
-	drm_dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
-	drm_kms_helper_poll_disable(drm_dev);
-	vga_switcheroo_set_dynamic_switch(pdev, VGA_SWITCHEROO_OFF);
+	ret = amdgpu_runtime_idle_check_display(dev);
+	if (ret)
+		return ret;
+	ret = amdgpu_runtime_idle_check_userq(dev);
+	if (ret)
+		return ret;
 
-	ret = amdgpu_device_suspend(drm_dev, false, false);
-	pci_save_state(pdev);
-	pci_disable_device(pdev);
-	pci_ignore_hotplug(pdev);
-	if (amdgpu_is_atpx_hybrid())
+	/* wait for all rings to drain before suspending */
+	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
+		struct amdgpu_ring *ring = adev->rings[i];
+
+		if (ring && ring->sched.ready) {
+			ret = amdgpu_fence_wait_empty(ring);
+			if (ret)
+				return -EBUSY;
+		}
+	}
+
+	adev->in_runpm = true;
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_PX)
+		drm_dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
+
+	/*
+	 * By setting mp1_state as PP_MP1_STATE_UNLOAD, MP1 will do some
+	 * proper cleanups and put itself into a state ready for PNP. That
+	 * can address some random resuming failure observed on BOCO capable
+	 * platforms.
+	 * TODO: this may be also needed for PX capable platform.
+	 */
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_BOCO)
+		adev->mp1_state = PP_MP1_STATE_UNLOAD;
+
+	ret = amdgpu_device_prepare(drm_dev);
+	if (ret)
+		return ret;
+	ret = amdgpu_device_suspend(drm_dev, false);
+	if (ret) {
+		adev->in_runpm = false;
+		if (adev->pm.rpm_mode == AMDGPU_RUNPM_BOCO)
+			adev->mp1_state = PP_MP1_STATE_NONE;
+		return ret;
+	}
+
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_BOCO)
+		adev->mp1_state = PP_MP1_STATE_NONE;
+
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_PX) {
+		/* Only need to handle PCI state in the driver for ATPX
+		 * PCI core handles it for _PR3.
+		 */
+		amdgpu_device_cache_pci_state(pdev);
+		pci_disable_device(pdev);
+		pci_ignore_hotplug(pdev);
 		pci_set_power_state(pdev, PCI_D3cold);
-	else if (!amdgpu_has_atpx_dgpu_power_cntl())
-		pci_set_power_state(pdev, PCI_D3hot);
-	drm_dev->switch_power_state = DRM_SWITCH_POWER_DYNAMIC_OFF;
+		drm_dev->switch_power_state = DRM_SWITCH_POWER_DYNAMIC_OFF;
+	} else if (adev->pm.rpm_mode == AMDGPU_RUNPM_BOCO) {
+		/* nothing to do */
+	} else if ((adev->pm.rpm_mode == AMDGPU_RUNPM_BACO) ||
+			(adev->pm.rpm_mode == AMDGPU_RUNPM_BAMACO)) {
+		amdgpu_device_baco_enter(adev);
+	}
+
+	dev_dbg(&pdev->dev, "asic/device is runtime suspended\n");
 
 	return 0;
 }
@@ -625,51 +2888,86 @@ static int amdgpu_pmops_runtime_resume(struct device *dev)
 {
 	struct pci_dev *pdev = to_pci_dev(dev);
 	struct drm_device *drm_dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
 	int ret;
 
-	if (!amdgpu_device_is_px(drm_dev))
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_NONE)
 		return -EINVAL;
 
-	drm_dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
+	/* Avoids registers access if device is physically gone */
+	if (!pci_device_is_present(adev->pdev))
+		adev->no_hw_access = true;
 
-	if (amdgpu_is_atpx_hybrid() ||
-	    !amdgpu_has_atpx_dgpu_power_cntl())
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_PX) {
+		drm_dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
+
+		/* Only need to handle PCI state in the driver for ATPX
+		 * PCI core handles it for _PR3.
+		 */
 		pci_set_power_state(pdev, PCI_D0);
-	pci_restore_state(pdev);
-	ret = pci_enable_device(pdev);
-	if (ret)
+		amdgpu_device_load_pci_state(pdev);
+		ret = pci_enable_device(pdev);
+		if (ret)
+			return ret;
+		pci_set_master(pdev);
+	} else if (adev->pm.rpm_mode == AMDGPU_RUNPM_BOCO) {
+		/* Only need to handle PCI state in the driver for ATPX
+		 * PCI core handles it for _PR3.
+		 */
+		pci_set_master(pdev);
+	} else if ((adev->pm.rpm_mode == AMDGPU_RUNPM_BACO) ||
+			(adev->pm.rpm_mode == AMDGPU_RUNPM_BAMACO)) {
+		amdgpu_device_baco_exit(adev);
+	}
+	ret = amdgpu_device_resume(drm_dev, false);
+	if (ret) {
+		if (adev->pm.rpm_mode == AMDGPU_RUNPM_PX)
+			pci_disable_device(pdev);
 		return ret;
-	pci_set_master(pdev);
+	}
 
-	ret = amdgpu_device_resume(drm_dev, false, false);
-	drm_kms_helper_poll_enable(drm_dev);
-	vga_switcheroo_set_dynamic_switch(pdev, VGA_SWITCHEROO_ON);
-	drm_dev->switch_power_state = DRM_SWITCH_POWER_ON;
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_PX)
+		drm_dev->switch_power_state = DRM_SWITCH_POWER_ON;
+	adev->in_runpm = false;
 	return 0;
 }
 
 static int amdgpu_pmops_runtime_idle(struct device *dev)
 {
-	struct pci_dev *pdev = to_pci_dev(dev);
-	struct drm_device *drm_dev = pci_get_drvdata(pdev);
-	struct drm_crtc *crtc;
+	struct drm_device *drm_dev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+	int ret;
 
-	if (!amdgpu_device_is_px(drm_dev)) {
+	if (adev->pm.rpm_mode == AMDGPU_RUNPM_NONE) {
 		pm_runtime_forbid(dev);
 		return -EBUSY;
 	}
 
-	list_for_each_entry(crtc, &drm_dev->mode_config.crtc_list, head) {
-		if (crtc->enabled) {
-			DRM_DEBUG_DRIVER("failing to power off - crtc active\n");
-			return -EBUSY;
-		}
-	}
+	ret = amdgpu_runtime_idle_check_display(dev);
+	if (ret)
+		goto done;
 
-	pm_runtime_mark_last_busy(dev);
+	ret = amdgpu_runtime_idle_check_userq(dev);
+done:
 	pm_runtime_autosuspend(dev);
-	/* we don't want the main rpm_idle to call suspend - we want to autosuspend */
-	return 1;
+	return ret;
+}
+
+static int amdgpu_drm_release(struct inode *inode, struct file *filp)
+{
+	struct drm_file *file_priv = filp->private_data;
+	struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
+	struct drm_device *dev = file_priv->minor->dev;
+	int idx;
+
+	if (fpriv && drm_dev_enter(dev, &idx)) {
+		fpriv->evf_mgr.fd_closing = true;
+		amdgpu_eviction_fence_destroy(&fpriv->evf_mgr);
+		amdgpu_userq_mgr_fini(&fpriv->userq_mgr);
+		drm_dev_exit(idx);
+	}
+
+	return drm_release(inode, filp);
 }
 
 long amdgpu_drm_ioctl(struct file *filp,
@@ -678,97 +2976,167 @@ long amdgpu_drm_ioctl(struct file *filp,
 	struct drm_file *file_priv = filp->private_data;
 	struct drm_device *dev;
 	long ret;
+
 	dev = file_priv->minor->dev;
 	ret = pm_runtime_get_sync(dev->dev);
 	if (ret < 0)
-		return ret;
+		goto out;
 
 	ret = drm_ioctl(filp, cmd, arg);
 
-	pm_runtime_mark_last_busy(dev->dev);
+out:
 	pm_runtime_put_autosuspend(dev->dev);
 	return ret;
 }
 
 static const struct dev_pm_ops amdgpu_pm_ops = {
-	.suspend = amdgpu_pmops_suspend,
-	.resume = amdgpu_pmops_resume,
-	.freeze = amdgpu_pmops_freeze,
-	.thaw = amdgpu_pmops_thaw,
-	.poweroff = amdgpu_pmops_poweroff,
-	.restore = amdgpu_pmops_restore,
+	.prepare = pm_sleep_ptr(amdgpu_pmops_prepare),
+	.complete = pm_sleep_ptr(amdgpu_pmops_complete),
+	.suspend = pm_sleep_ptr(amdgpu_pmops_suspend),
+	.suspend_noirq = pm_sleep_ptr(amdgpu_pmops_suspend_noirq),
+	.resume = pm_sleep_ptr(amdgpu_pmops_resume),
+	.freeze = pm_sleep_ptr(amdgpu_pmops_freeze),
+	.thaw = pm_sleep_ptr(amdgpu_pmops_thaw),
+	.poweroff = pm_sleep_ptr(amdgpu_pmops_poweroff),
+	.restore = pm_sleep_ptr(amdgpu_pmops_restore),
 	.runtime_suspend = amdgpu_pmops_runtime_suspend,
 	.runtime_resume = amdgpu_pmops_runtime_resume,
 	.runtime_idle = amdgpu_pmops_runtime_idle,
 };
 
+static int amdgpu_flush(struct file *f, fl_owner_t id)
+{
+	struct drm_file *file_priv = f->private_data;
+	struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
+	long timeout = MAX_WAIT_SCHED_ENTITY_Q_EMPTY;
+
+	timeout = amdgpu_ctx_mgr_entity_flush(&fpriv->ctx_mgr, timeout);
+	timeout = amdgpu_vm_wait_idle(&fpriv->vm, timeout);
+
+	return timeout >= 0 ? 0 : timeout;
+}
+
 static const struct file_operations amdgpu_driver_kms_fops = {
 	.owner = THIS_MODULE,
 	.open = drm_open,
-	.release = drm_release,
+	.flush = amdgpu_flush,
+	.release = amdgpu_drm_release,
 	.unlocked_ioctl = amdgpu_drm_ioctl,
-	.mmap = amdgpu_mmap,
+	.mmap = drm_gem_mmap,
 	.poll = drm_poll,
 	.read = drm_read,
 #ifdef CONFIG_COMPAT
 	.compat_ioctl = amdgpu_kms_compat_ioctl,
 #endif
+#ifdef CONFIG_PROC_FS
+	.show_fdinfo = drm_show_fdinfo,
+#endif
+	.fop_flags = FOP_UNSIGNED_OFFSET,
 };
 
-static struct drm_driver kms_driver = {
+int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv)
+{
+	struct drm_file *file;
+
+	if (!filp)
+		return -EINVAL;
+
+	if (filp->f_op != &amdgpu_driver_kms_fops)
+		return -EINVAL;
+
+	file = filp->private_data;
+	*fpriv = file->driver_priv;
+	return 0;
+}
+
+const struct drm_ioctl_desc amdgpu_ioctls_kms[] = {
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_CREATE, amdgpu_gem_create_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_CTX, amdgpu_ctx_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_VM, amdgpu_vm_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_SCHED, amdgpu_sched_ioctl, DRM_MASTER),
+	DRM_IOCTL_DEF_DRV(AMDGPU_BO_LIST, amdgpu_bo_list_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_FENCE_TO_HANDLE, amdgpu_cs_fence_to_handle_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	/* KMS */
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_MMAP, amdgpu_gem_mmap_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_WAIT_IDLE, amdgpu_gem_wait_idle_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_CS, amdgpu_cs_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_INFO, amdgpu_info_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_CS, amdgpu_cs_wait_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_FENCES, amdgpu_cs_wait_fences_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_METADATA, amdgpu_gem_metadata_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_USERQ, amdgpu_userq_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_USERQ_SIGNAL, amdgpu_userq_signal_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_USERQ_WAIT, amdgpu_userq_wait_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_LIST_HANDLES, amdgpu_gem_list_handles_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+};
+
+static const struct drm_driver amdgpu_kms_driver = {
 	.driver_features =
-	    DRIVER_USE_AGP |
-	    DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | DRIVER_GEM |
-	    DRIVER_PRIME | DRIVER_RENDER | DRIVER_MODESET,
-	.load = amdgpu_driver_load_kms,
+	    DRIVER_ATOMIC |
+	    DRIVER_GEM |
+	    DRIVER_RENDER | DRIVER_MODESET | DRIVER_SYNCOBJ |
+	    DRIVER_SYNCOBJ_TIMELINE,
 	.open = amdgpu_driver_open_kms,
 	.postclose = amdgpu_driver_postclose_kms,
-	.lastclose = amdgpu_driver_lastclose_kms,
-	.set_busid = drm_pci_set_busid,
-	.unload = amdgpu_driver_unload_kms,
-	.get_vblank_counter = amdgpu_get_vblank_counter_kms,
-	.enable_vblank = amdgpu_enable_vblank_kms,
-	.disable_vblank = amdgpu_disable_vblank_kms,
-	.get_vblank_timestamp = amdgpu_get_vblank_timestamp_kms,
-	.get_scanout_position = amdgpu_get_crtc_scanoutpos,
-#if defined(CONFIG_DEBUG_FS)
-	.debugfs_init = amdgpu_debugfs_init,
+	.ioctls = amdgpu_ioctls_kms,
+	.num_ioctls = ARRAY_SIZE(amdgpu_ioctls_kms),
+	.dumb_create = amdgpu_mode_dumb_create,
+	.dumb_map_offset = amdgpu_mode_dumb_mmap,
+	DRM_FBDEV_TTM_DRIVER_OPS,
+	.fops = &amdgpu_driver_kms_fops,
+	.release = &amdgpu_driver_release_kms,
+#ifdef CONFIG_PROC_FS
+	.show_fdinfo = amdgpu_show_fdinfo,
 #endif
-	.irq_preinstall = amdgpu_irq_preinstall,
-	.irq_postinstall = amdgpu_irq_postinstall,
-	.irq_uninstall = amdgpu_irq_uninstall,
-	.irq_handler = amdgpu_irq_handler,
+
+	.gem_prime_import = amdgpu_gem_prime_import,
+
+	.name = DRIVER_NAME,
+	.desc = DRIVER_DESC,
+	.major = KMS_DRIVER_MAJOR,
+	.minor = KMS_DRIVER_MINOR,
+	.patchlevel = KMS_DRIVER_PATCHLEVEL,
+};
+
+const struct drm_driver amdgpu_partition_driver = {
+	.driver_features =
+	    DRIVER_GEM | DRIVER_RENDER | DRIVER_SYNCOBJ |
+	    DRIVER_SYNCOBJ_TIMELINE,
+	.open = amdgpu_driver_open_kms,
+	.postclose = amdgpu_driver_postclose_kms,
 	.ioctls = amdgpu_ioctls_kms,
-	.gem_free_object_unlocked = amdgpu_gem_object_free,
-	.gem_open_object = amdgpu_gem_object_open,
-	.gem_close_object = amdgpu_gem_object_close,
+	.num_ioctls = ARRAY_SIZE(amdgpu_ioctls_kms),
 	.dumb_create = amdgpu_mode_dumb_create,
 	.dumb_map_offset = amdgpu_mode_dumb_mmap,
-	.dumb_destroy = drm_gem_dumb_destroy,
+	DRM_FBDEV_TTM_DRIVER_OPS,
 	.fops = &amdgpu_driver_kms_fops,
+	.release = &amdgpu_driver_release_kms,
 
-	.prime_handle_to_fd = drm_gem_prime_handle_to_fd,
-	.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
-	.gem_prime_export = amdgpu_gem_prime_export,
-	.gem_prime_import = drm_gem_prime_import,
-	.gem_prime_pin = amdgpu_gem_prime_pin,
-	.gem_prime_unpin = amdgpu_gem_prime_unpin,
-	.gem_prime_res_obj = amdgpu_gem_prime_res_obj,
-	.gem_prime_get_sg_table = amdgpu_gem_prime_get_sg_table,
-	.gem_prime_import_sg_table = amdgpu_gem_prime_import_sg_table,
-	.gem_prime_vmap = amdgpu_gem_prime_vmap,
-	.gem_prime_vunmap = amdgpu_gem_prime_vunmap,
+	.gem_prime_import = amdgpu_gem_prime_import,
 
 	.name = DRIVER_NAME,
 	.desc = DRIVER_DESC,
-	.date = DRIVER_DATE,
 	.major = KMS_DRIVER_MAJOR,
 	.minor = KMS_DRIVER_MINOR,
 	.patchlevel = KMS_DRIVER_PATCHLEVEL,
 };
 
-static struct drm_driver *driver;
-static struct pci_driver *pdriver;
+static struct pci_error_handlers amdgpu_pci_err_handler = {
+	.error_detected	= amdgpu_pci_error_detected,
+	.mmio_enabled	= amdgpu_pci_mmio_enabled,
+	.slot_reset	= amdgpu_pci_slot_reset,
+	.resume		= amdgpu_pci_resume,
+};
+
+static const struct attribute_group *amdgpu_sysfs_groups[] = {
+	&amdgpu_vram_mgr_attr_group,
+	&amdgpu_gtt_mgr_attr_group,
+	&amdgpu_flash_attr_group,
+	NULL,
+};
 
 static struct pci_driver amdgpu_kms_pci_driver = {
 	.name = DRIVER_NAME,
@@ -776,11 +3144,11 @@ static struct pci_driver amdgpu_kms_pci_driver = {
 	.probe = amdgpu_pci_probe,
 	.remove = amdgpu_pci_remove,
 	.shutdown = amdgpu_pci_shutdown,
-	.driver.pm = &amdgpu_pm_ops,
+	.driver.pm = pm_ptr(&amdgpu_pm_ops),
+	.err_handler = &amdgpu_pci_err_handler,
+	.dev_groups = amdgpu_sysfs_groups,
 };
 
-
-
 static int __init amdgpu_init(void)
 {
 	int r;
@@ -789,28 +3157,25 @@ static int __init amdgpu_init(void)
 	if (r)
 		goto error_sync;
 
-	r = amdgpu_fence_slab_init();
+	r = amdgpu_userq_fence_slab_init();
 	if (r)
 		goto error_fence;
 
-	r = amd_sched_fence_slab_init();
-	if (r)
-		goto error_sched;
-
-	if (vgacon_text_force()) {
-		DRM_ERROR("VGACON disables amdgpu kernel modesetting.\n");
-		return -EINVAL;
-	}
 	DRM_INFO("amdgpu kernel modesetting enabled.\n");
-	driver = &kms_driver;
-	pdriver = &amdgpu_kms_pci_driver;
-	driver->num_ioctls = amdgpu_max_kms_ioctl;
 	amdgpu_register_atpx_handler();
-	/* let modprobe override vga console setting */
-	return drm_pci_init(driver, pdriver);
+	amdgpu_acpi_detect();
 
-error_sched:
-	amdgpu_fence_slab_fini();
+	/* Ignore KFD init failures. Normal when CONFIG_HSA_AMD is not set. */
+	amdgpu_amdkfd_init();
+
+	if (amdgpu_pp_feature_mask & PP_OVERDRIVE_MASK) {
+		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK);
+		pr_crit("Overdrive is enabled, please disable it before "
+			"reporting any bugs unrelated to overdrive.\n");
+	}
+
+	/* let modprobe override vga console setting */
+	return pci_register_driver(&amdgpu_kms_pci_driver);
 
 error_fence:
 	amdgpu_sync_fini();
@@ -822,11 +3187,13 @@ error_sync:
 static void __exit amdgpu_exit(void)
 {
 	amdgpu_amdkfd_fini();
-	drm_pci_exit(driver, pdriver);
+	pci_unregister_driver(&amdgpu_kms_pci_driver);
 	amdgpu_unregister_atpx_handler();
+	amdgpu_acpi_release();
 	amdgpu_sync_fini();
-	amd_sched_fence_slab_fini();
-	amdgpu_fence_slab_fini();
+	amdgpu_userq_fence_slab_fini();
+	mmu_notifier_synchronize();
+	amdgpu_xcp_drv_release();
 }
 
 module_init(amdgpu_init);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.h
index e3a4f7048042..2d86cc6f7f4d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.h
@@ -40,9 +40,13 @@
 
 #define DRIVER_NAME		"amdgpu"
 #define DRIVER_DESC		"AMD GPU"
-#define DRIVER_DATE		"20150101"
+
+extern const struct drm_driver amdgpu_partition_driver;
 
 long amdgpu_drm_ioctl(struct file *filp,
 		      unsigned int cmd, unsigned long arg);
 
+long amdgpu_kms_compat_ioctl(struct file *filp,
+			     unsigned int cmd, unsigned long arg);
+
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eeprom.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_eeprom.c
new file mode 100644
index 000000000000..8cd69836dd99
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eeprom.c
@@ -0,0 +1,240 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "amdgpu_eeprom.h"
+#include "amdgpu.h"
+
+/* AT24CM02 and M24M02-R have a 256-byte write page size.
+ */
+#define EEPROM_PAGE_BITS   8
+#define EEPROM_PAGE_SIZE   (1U << EEPROM_PAGE_BITS)
+#define EEPROM_PAGE_MASK   (EEPROM_PAGE_SIZE - 1)
+
+#define EEPROM_OFFSET_SIZE 2
+
+/* EEPROM memory addresses are 19-bits long, which can
+ * be partitioned into 3, 8, 8 bits, for a total of 19.
+ * The upper 3 bits are sent as part of the 7-bit
+ * "Device Type Identifier"--an I2C concept, which for EEPROM devices
+ * is hard-coded as 1010b, indicating that it is an EEPROM
+ * device--this is the wire format, followed by the upper
+ * 3 bits of the 19-bit address, followed by the direction,
+ * followed by two bytes holding the rest of the 16-bits of
+ * the EEPROM memory address. The format on the wire for EEPROM
+ * devices is: 1010XYZD, A15:A8, A7:A0,
+ * Where D is the direction and sequenced out by the hardware.
+ * Bits XYZ are memory address bits 18, 17 and 16.
+ * These bits are compared to how pins 1-3 of the part are connected,
+ * depending on the size of the part, more on that later.
+ *
+ * Note that of this wire format, a client is in control
+ * of, and needs to specify only XYZ, A15:A8, A7:0, bits,
+ * which is exactly the EEPROM memory address, or offset,
+ * in order to address up to 8 EEPROM devices on the I2C bus.
+ *
+ * For instance, a 2-Mbit I2C EEPROM part, addresses all its bytes,
+ * using an 18-bit address, bit 17 to 0 and thus would use all but one bit of
+ * the 19 bits previously mentioned. The designer would then not connect
+ * pins 1 and 2, and pin 3 usually named "A_2" or "E2", would be connected to
+ * either Vcc or GND. This would allow for up to two 2-Mbit parts on
+ * the same bus, where one would be addressable with bit 18 as 1, and
+ * the other with bit 18 of the address as 0.
+ *
+ * For a 2-Mbit part, bit 18 is usually known as the "Chip Enable" or
+ * "Hardware Address Bit". This bit is compared to the load on pin 3
+ * of the device, described above, and if there is a match, then this
+ * device responds to the command. This way, you can connect two
+ * 2-Mbit EEPROM devices on the same bus, but see one contiguous
+ * memory from 0 to 7FFFFh, where address 0 to 3FFFF is in the device
+ * whose pin 3 is connected to GND, and address 40000 to 7FFFFh is in
+ * the 2nd device, whose pin 3 is connected to Vcc.
+ *
+ * This addressing you encode in the 32-bit "eeprom_addr" below,
+ * namely the 19-bits "XYZ,A15:A0", as a single 19-bit address. For
+ * instance, eeprom_addr = 0x6DA01, is 110_1101_1010_0000_0001, where
+ * XYZ=110b, and A15:A0=DA01h. The XYZ bits become part of the device
+ * address, and the rest of the address bits are sent as the memory
+ * address bytes.
+ *
+ * That is, for an I2C EEPROM driver everything is controlled by
+ * the "eeprom_addr".
+ *
+ * See also top of amdgpu_ras_eeprom.c.
+ *
+ * P.S. If you need to write, lock and read the Identification Page,
+ * (M24M02-DR device only, which we do not use), change the "7" to
+ * "0xF" in the macro below, and let the client set bit 20 to 1 in
+ * "eeprom_addr", and set A10 to 0 to write into it, and A10 and A1 to
+ * 1 to lock it permanently.
+ */
+#define MAKE_I2C_ADDR(_aa) ((0xA << 3) | (((_aa) >> 16) & 0xF))
+
+static int __amdgpu_eeprom_xfer(struct i2c_adapter *i2c_adap, u32 eeprom_addr,
+				u8 *eeprom_buf, u32 buf_size, bool read)
+{
+	u8 eeprom_offset_buf[EEPROM_OFFSET_SIZE];
+	struct i2c_msg msgs[] = {
+		{
+			.flags = 0,
+			.len = EEPROM_OFFSET_SIZE,
+			.buf = eeprom_offset_buf,
+		},
+		{
+			.flags = read ? I2C_M_RD : 0,
+		},
+	};
+	const u8 *p = eeprom_buf;
+	int r;
+	u16 len;
+
+	for (r = 0; buf_size > 0;
+	      buf_size -= len, eeprom_addr += len, eeprom_buf += len) {
+		/* Set the EEPROM address we want to write to/read from.
+		 */
+		msgs[0].addr = MAKE_I2C_ADDR(eeprom_addr);
+		msgs[1].addr = msgs[0].addr;
+		msgs[0].buf[0] = (eeprom_addr >> 8) & 0xff;
+		msgs[0].buf[1] = eeprom_addr & 0xff;
+
+		if (!read) {
+			/* Write the maximum amount of data, without
+			 * crossing the device's page boundary, as per
+			 * its spec. Partial page writes are allowed,
+			 * starting at any location within the page,
+			 * so long as the page boundary isn't crossed
+			 * over (actually the page pointer rolls
+			 * over).
+			 *
+			 * As per the AT24CM02 EEPROM spec, after
+			 * writing into a page, the I2C driver should
+			 * terminate the transfer, i.e. in
+			 * "i2c_transfer()" below, with a STOP
+			 * condition, so that the self-timed write
+			 * cycle begins. This is implied for the
+			 * "i2c_transfer()" abstraction.
+			 */
+			len = min(EEPROM_PAGE_SIZE - (eeprom_addr & EEPROM_PAGE_MASK),
+					buf_size);
+		} else {
+			/* Reading from the EEPROM has no limitation
+			 * on the number of bytes read from the EEPROM
+			 * device--they are simply sequenced out.
+			 * Keep in mind that i2c_msg.len is u16 type.
+			 */
+			len = min(U16_MAX, buf_size);
+		}
+		msgs[1].len = len;
+		msgs[1].buf = eeprom_buf;
+
+		/* This constitutes a START-STOP transaction.
+		 */
+		r = i2c_transfer(i2c_adap, msgs, ARRAY_SIZE(msgs));
+		if (r != ARRAY_SIZE(msgs))
+			break;
+
+		if (!read) {
+			/* According to EEPROM specs the length of the
+			 * self-writing cycle, tWR (tW), is 10 ms.
+			 *
+			 * TODO: Use polling on ACK, aka Acknowledge
+			 * Polling, to minimize waiting for the
+			 * internal write cycle to complete, as it is
+			 * usually smaller than tWR (tW).
+			 */
+			msleep(10);
+		}
+	}
+
+	return r < 0 ? r : eeprom_buf - p;
+}
+
+/**
+ * amdgpu_eeprom_xfer -- Read/write from/to an I2C EEPROM device
+ * @i2c_adap: pointer to the I2C adapter to use
+ * @eeprom_addr: EEPROM address from which to read/write
+ * @eeprom_buf: pointer to data buffer to read into/write from
+ * @buf_size: the size of @eeprom_buf
+ * @read: True if reading from the EEPROM, false if writing
+ *
+ * Returns the number of bytes read/written; -errno on error.
+ */
+static int amdgpu_eeprom_xfer(struct i2c_adapter *i2c_adap, u32 eeprom_addr,
+			      u8 *eeprom_buf, u32 buf_size, bool read)
+{
+	const struct i2c_adapter_quirks *quirks = i2c_adap->quirks;
+	u16 limit;
+	u16 ps; /* Partial size */
+	int res = 0, r;
+
+	if (!quirks)
+		limit = 0;
+	else if (read)
+		limit = quirks->max_read_len;
+	else
+		limit = quirks->max_write_len;
+
+	if (limit == 0) {
+		return __amdgpu_eeprom_xfer(i2c_adap, eeprom_addr,
+					    eeprom_buf, buf_size, read);
+	} else if (limit <= EEPROM_OFFSET_SIZE) {
+		dev_err_ratelimited(&i2c_adap->dev,
+				    "maddr:0x%04X size:0x%02X:quirk max_%s_len must be > %d",
+				    eeprom_addr, buf_size,
+				    str_read_write(read), EEPROM_OFFSET_SIZE);
+		return -EINVAL;
+	}
+
+	/* The "limit" includes all data bytes sent/received,
+	 * which would include the EEPROM_OFFSET_SIZE bytes.
+	 * Account for them here.
+	 */
+	limit -= EEPROM_OFFSET_SIZE;
+	for ( ; buf_size > 0;
+	      buf_size -= ps, eeprom_addr += ps, eeprom_buf += ps) {
+		ps = min(limit, buf_size);
+
+		r = __amdgpu_eeprom_xfer(i2c_adap, eeprom_addr,
+					 eeprom_buf, ps, read);
+		if (r < 0)
+			return r;
+		res += r;
+	}
+
+	return res;
+}
+
+int amdgpu_eeprom_read(struct i2c_adapter *i2c_adap,
+		       u32 eeprom_addr, u8 *eeprom_buf,
+		       u32 bytes)
+{
+	return amdgpu_eeprom_xfer(i2c_adap, eeprom_addr, eeprom_buf, bytes,
+				  true);
+}
+
+int amdgpu_eeprom_write(struct i2c_adapter *i2c_adap,
+			u32 eeprom_addr, u8 *eeprom_buf,
+			u32 bytes)
+{
+	return amdgpu_eeprom_xfer(i2c_adap, eeprom_addr, eeprom_buf, bytes,
+				  false);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eeprom.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_eeprom.h
new file mode 100644
index 000000000000..8083b8253ef4
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eeprom.h
@@ -0,0 +1,37 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef _AMDGPU_EEPROM_H
+#define _AMDGPU_EEPROM_H
+
+#include <linux/i2c.h>
+
+int amdgpu_eeprom_read(struct i2c_adapter *i2c_adap,
+		       u32 eeprom_addr, u8 *eeprom_buf,
+		       u32 bytes);
+
+int amdgpu_eeprom_write(struct i2c_adapter *i2c_adap,
+			u32 eeprom_addr, u8 *eeprom_buf,
+			u32 bytes);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_encoders.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_encoders.c
index 94138abe093b..3aaeed2d3562 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_encoders.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_encoders.c
@@ -23,30 +23,32 @@
  * Authors: Dave Airlie
  *          Alex Deucher
  */
-#include <drm/drmP.h>
-#include <drm/drm_crtc_helper.h>
+
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 #include "amdgpu_connectors.h"
+#include "amdgpu_display.h"
 #include "atom.h"
 #include "atombios_encoders.h"
 
 void
 amdgpu_link_encoder_connector(struct drm_device *dev)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct drm_connector *connector;
+	struct drm_connector_list_iter iter;
 	struct amdgpu_connector *amdgpu_connector;
 	struct drm_encoder *encoder;
 	struct amdgpu_encoder *amdgpu_encoder;
 
+	drm_connector_list_iter_begin(dev, &iter);
 	/* walk the list and link encoders to connectors */
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
+	drm_for_each_connector_iter(connector, &iter) {
 		amdgpu_connector = to_amdgpu_connector(connector);
 		list_for_each_entry(encoder, &dev->mode_config.encoder_list, head) {
 			amdgpu_encoder = to_amdgpu_encoder(encoder);
 			if (amdgpu_encoder->devices & amdgpu_connector->devices) {
-				drm_mode_connector_attach_encoder(connector, encoder);
+				drm_connector_attach_encoder(connector, encoder);
 				if (amdgpu_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT)) {
 					amdgpu_atombios_encoder_init_backlight(amdgpu_encoder, connector);
 					adev->mode_info.bl_encoder = amdgpu_encoder;
@@ -54,6 +56,7 @@ amdgpu_link_encoder_connector(struct drm_device *dev)
 			}
 		}
 	}
+	drm_connector_list_iter_end(&iter);
 }
 
 void amdgpu_encoder_set_active_device(struct drm_encoder *encoder)
@@ -61,16 +64,20 @@ void amdgpu_encoder_set_active_device(struct drm_encoder *encoder)
 	struct drm_device *dev = encoder->dev;
 	struct amdgpu_encoder *amdgpu_encoder = to_amdgpu_encoder(encoder);
 	struct drm_connector *connector;
+	struct drm_connector_list_iter iter;
 
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
+	drm_connector_list_iter_begin(dev, &iter);
+	drm_for_each_connector_iter(connector, &iter) {
 		if (connector->encoder == encoder) {
 			struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector);
+
 			amdgpu_encoder->active_device = amdgpu_encoder->devices & amdgpu_connector->devices;
 			DRM_DEBUG_KMS("setting active device to %08x from %08x %08x for encoder %d\n",
 				  amdgpu_encoder->active_device, amdgpu_encoder->devices,
 				  amdgpu_connector->devices, encoder->encoder_type);
 		}
 	}
+	drm_connector_list_iter_end(&iter);
 }
 
 struct drm_connector *
@@ -78,15 +85,20 @@ amdgpu_get_connector_for_encoder(struct drm_encoder *encoder)
 {
 	struct drm_device *dev = encoder->dev;
 	struct amdgpu_encoder *amdgpu_encoder = to_amdgpu_encoder(encoder);
-	struct drm_connector *connector;
+	struct drm_connector *connector, *found = NULL;
+	struct drm_connector_list_iter iter;
 	struct amdgpu_connector *amdgpu_connector;
 
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
+	drm_connector_list_iter_begin(dev, &iter);
+	drm_for_each_connector_iter(connector, &iter) {
 		amdgpu_connector = to_amdgpu_connector(connector);
-		if (amdgpu_encoder->active_device & amdgpu_connector->devices)
-			return connector;
+		if (amdgpu_encoder->active_device & amdgpu_connector->devices) {
+			found = connector;
+			break;
+		}
 	}
-	return NULL;
+	drm_connector_list_iter_end(&iter);
+	return found;
 }
 
 struct drm_connector *
@@ -94,15 +106,20 @@ amdgpu_get_connector_for_encoder_init(struct drm_encoder *encoder)
 {
 	struct drm_device *dev = encoder->dev;
 	struct amdgpu_encoder *amdgpu_encoder = to_amdgpu_encoder(encoder);
-	struct drm_connector *connector;
+	struct drm_connector *connector, *found = NULL;
+	struct drm_connector_list_iter iter;
 	struct amdgpu_connector *amdgpu_connector;
 
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
+	drm_connector_list_iter_begin(dev, &iter);
+	drm_for_each_connector_iter(connector, &iter) {
 		amdgpu_connector = to_amdgpu_connector(connector);
-		if (amdgpu_encoder->devices & amdgpu_connector->devices)
-			return connector;
+		if (amdgpu_encoder->devices & amdgpu_connector->devices) {
+			found = connector;
+			break;
+		}
 	}
-	return NULL;
+	drm_connector_list_iter_end(&iter);
+	return found;
 }
 
 struct drm_encoder *amdgpu_get_external_encoder(struct drm_encoder *encoder)
@@ -149,12 +166,12 @@ void amdgpu_panel_mode_fixup(struct drm_encoder *encoder,
 {
 	struct amdgpu_encoder *amdgpu_encoder = to_amdgpu_encoder(encoder);
 	struct drm_display_mode *native_mode = &amdgpu_encoder->native_mode;
-	unsigned hblank = native_mode->htotal - native_mode->hdisplay;
-	unsigned vblank = native_mode->vtotal - native_mode->vdisplay;
-	unsigned hover = native_mode->hsync_start - native_mode->hdisplay;
-	unsigned vover = native_mode->vsync_start - native_mode->vdisplay;
-	unsigned hsync_width = native_mode->hsync_end - native_mode->hsync_start;
-	unsigned vsync_width = native_mode->vsync_end - native_mode->vsync_start;
+	unsigned int hblank = native_mode->htotal - native_mode->hdisplay;
+	unsigned int vblank = native_mode->vtotal - native_mode->vdisplay;
+	unsigned int hover = native_mode->hsync_start - native_mode->hdisplay;
+	unsigned int vover = native_mode->vsync_start - native_mode->vdisplay;
+	unsigned int hsync_width = native_mode->hsync_end - native_mode->hsync_start;
+	unsigned int vsync_width = native_mode->vsync_end - native_mode->vsync_start;
 
 	adjusted_mode->clock = native_mode->clock;
 	adjusted_mode->flags = native_mode->flags;
@@ -205,7 +222,7 @@ bool amdgpu_dig_monitor_is_duallink(struct drm_encoder *encoder,
 	case DRM_MODE_CONNECTOR_HDMIB:
 		if (amdgpu_connector->use_digital) {
 			/* HDMI 1.3 supports up to 340 Mhz over single link */
-			if (drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
+			if (connector->display_info.is_hdmi) {
 				if (pixel_clock > 340000)
 					return true;
 				else
@@ -227,7 +244,7 @@ bool amdgpu_dig_monitor_is_duallink(struct drm_encoder *encoder,
 			return false;
 		else {
 			/* HDMI 1.3 supports up to 340 Mhz over single link */
-			if (drm_detect_hdmi_monitor(amdgpu_connector_edid(connector))) {
+			if (connector->display_info.is_hdmi) {
 				if (pixel_clock > 340000)
 					return true;
 				else
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
new file mode 100644
index 000000000000..23d7d0b0d625
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2024 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include <linux/sched.h>
+#include <drm/drm_exec.h>
+#include "amdgpu.h"
+
+#define work_to_evf_mgr(w, name) container_of(w, struct amdgpu_eviction_fence_mgr, name)
+#define evf_mgr_to_fpriv(e) container_of(e, struct amdgpu_fpriv, evf_mgr)
+
+static const char *
+amdgpu_eviction_fence_get_driver_name(struct dma_fence *fence)
+{
+	return "amdgpu_eviction_fence";
+}
+
+static const char *
+amdgpu_eviction_fence_get_timeline_name(struct dma_fence *f)
+{
+	struct amdgpu_eviction_fence *ef;
+
+	ef = container_of(f, struct amdgpu_eviction_fence, base);
+	return ef->timeline_name;
+}
+
+int
+amdgpu_eviction_fence_replace_fence(struct amdgpu_eviction_fence_mgr *evf_mgr,
+				    struct drm_exec *exec)
+{
+	struct amdgpu_eviction_fence *old_ef, *new_ef;
+	struct drm_gem_object *obj;
+	unsigned long index;
+	int ret;
+
+	if (evf_mgr->ev_fence &&
+	    !dma_fence_is_signaled(&evf_mgr->ev_fence->base))
+		return 0;
+	/*
+	 * Steps to replace eviction fence:
+	 * * lock all objects in exec (caller)
+	 * * create a new eviction fence
+	 * * update new eviction fence in evf_mgr
+	 * * attach the new eviction fence to BOs
+	 * * release the old fence
+	 * * unlock the objects (caller)
+	 */
+	new_ef = amdgpu_eviction_fence_create(evf_mgr);
+	if (!new_ef) {
+		DRM_ERROR("Failed to create new eviction fence\n");
+		return -ENOMEM;
+	}
+
+	/* Update the eviction fence now */
+	spin_lock(&evf_mgr->ev_fence_lock);
+	old_ef = evf_mgr->ev_fence;
+	evf_mgr->ev_fence = new_ef;
+	spin_unlock(&evf_mgr->ev_fence_lock);
+
+	/* Attach the new fence */
+	drm_exec_for_each_locked_object(exec, index, obj) {
+		struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+
+		if (!bo)
+			continue;
+		ret = amdgpu_eviction_fence_attach(evf_mgr, bo);
+		if (ret) {
+			DRM_ERROR("Failed to attch new eviction fence\n");
+			goto free_err;
+		}
+	}
+
+	/* Free old fence */
+	if (old_ef)
+		dma_fence_put(&old_ef->base);
+	return 0;
+
+free_err:
+	kfree(new_ef);
+	return ret;
+}
+
+static void
+amdgpu_eviction_fence_suspend_worker(struct work_struct *work)
+{
+	struct amdgpu_eviction_fence_mgr *evf_mgr = work_to_evf_mgr(work, suspend_work.work);
+	struct amdgpu_fpriv *fpriv = evf_mgr_to_fpriv(evf_mgr);
+	struct amdgpu_userq_mgr *uq_mgr = &fpriv->userq_mgr;
+	struct amdgpu_eviction_fence *ev_fence;
+
+	mutex_lock(&uq_mgr->userq_mutex);
+	spin_lock(&evf_mgr->ev_fence_lock);
+	ev_fence = evf_mgr->ev_fence;
+	if (ev_fence)
+		dma_fence_get(&ev_fence->base);
+	else
+		goto unlock;
+	spin_unlock(&evf_mgr->ev_fence_lock);
+
+	amdgpu_userq_evict(uq_mgr, ev_fence);
+
+	mutex_unlock(&uq_mgr->userq_mutex);
+	dma_fence_put(&ev_fence->base);
+	return;
+
+unlock:
+	spin_unlock(&evf_mgr->ev_fence_lock);
+	mutex_unlock(&uq_mgr->userq_mutex);
+}
+
+static bool amdgpu_eviction_fence_enable_signaling(struct dma_fence *f)
+{
+	struct amdgpu_eviction_fence_mgr *evf_mgr;
+	struct amdgpu_eviction_fence *ev_fence;
+
+	if (!f)
+		return true;
+
+	ev_fence = to_ev_fence(f);
+	evf_mgr = ev_fence->evf_mgr;
+
+	schedule_delayed_work(&evf_mgr->suspend_work, 0);
+	return true;
+}
+
+static const struct dma_fence_ops amdgpu_eviction_fence_ops = {
+	.get_driver_name = amdgpu_eviction_fence_get_driver_name,
+	.get_timeline_name = amdgpu_eviction_fence_get_timeline_name,
+	.enable_signaling = amdgpu_eviction_fence_enable_signaling,
+};
+
+void amdgpu_eviction_fence_signal(struct amdgpu_eviction_fence_mgr *evf_mgr,
+				  struct amdgpu_eviction_fence *ev_fence)
+{
+	spin_lock(&evf_mgr->ev_fence_lock);
+	dma_fence_signal(&ev_fence->base);
+	spin_unlock(&evf_mgr->ev_fence_lock);
+}
+
+struct amdgpu_eviction_fence *
+amdgpu_eviction_fence_create(struct amdgpu_eviction_fence_mgr *evf_mgr)
+{
+	struct amdgpu_eviction_fence *ev_fence;
+
+	ev_fence = kzalloc(sizeof(*ev_fence), GFP_KERNEL);
+	if (!ev_fence)
+		return NULL;
+
+	ev_fence->evf_mgr = evf_mgr;
+	get_task_comm(ev_fence->timeline_name, current);
+	spin_lock_init(&ev_fence->lock);
+	dma_fence_init64(&ev_fence->base, &amdgpu_eviction_fence_ops,
+			 &ev_fence->lock, evf_mgr->ev_fence_ctx,
+			 atomic_inc_return(&evf_mgr->ev_fence_seq));
+	return ev_fence;
+}
+
+void amdgpu_eviction_fence_destroy(struct amdgpu_eviction_fence_mgr *evf_mgr)
+{
+	struct amdgpu_eviction_fence *ev_fence;
+
+	/* Wait for any pending work to execute */
+	flush_delayed_work(&evf_mgr->suspend_work);
+
+	spin_lock(&evf_mgr->ev_fence_lock);
+	ev_fence = evf_mgr->ev_fence;
+	spin_unlock(&evf_mgr->ev_fence_lock);
+
+	if (!ev_fence)
+		return;
+
+	dma_fence_wait(&ev_fence->base, false);
+
+	/* Last unref of ev_fence */
+	dma_fence_put(&ev_fence->base);
+}
+
+int amdgpu_eviction_fence_attach(struct amdgpu_eviction_fence_mgr *evf_mgr,
+				 struct amdgpu_bo *bo)
+{
+	struct amdgpu_eviction_fence *ev_fence;
+	struct dma_resv *resv = bo->tbo.base.resv;
+	int ret;
+
+	if (!resv)
+		return 0;
+
+	ret = dma_resv_reserve_fences(resv, 1);
+	if (ret) {
+		DRM_DEBUG_DRIVER("Failed to resv fence space\n");
+		return ret;
+	}
+
+	spin_lock(&evf_mgr->ev_fence_lock);
+	ev_fence = evf_mgr->ev_fence;
+	if (ev_fence)
+		dma_resv_add_fence(resv, &ev_fence->base, DMA_RESV_USAGE_BOOKKEEP);
+	spin_unlock(&evf_mgr->ev_fence_lock);
+
+	return 0;
+}
+
+void amdgpu_eviction_fence_detach(struct amdgpu_eviction_fence_mgr *evf_mgr,
+				  struct amdgpu_bo *bo)
+{
+	struct dma_fence *stub = dma_fence_get_stub();
+
+	dma_resv_replace_fences(bo->tbo.base.resv, evf_mgr->ev_fence_ctx,
+				stub, DMA_RESV_USAGE_BOOKKEEP);
+	dma_fence_put(stub);
+}
+
+int amdgpu_eviction_fence_init(struct amdgpu_eviction_fence_mgr *evf_mgr)
+{
+	/* This needs to be done one time per open */
+	atomic_set(&evf_mgr->ev_fence_seq, 0);
+	evf_mgr->ev_fence_ctx = dma_fence_context_alloc(1);
+	spin_lock_init(&evf_mgr->ev_fence_lock);
+
+	INIT_DELAYED_WORK(&evf_mgr->suspend_work, amdgpu_eviction_fence_suspend_worker);
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
new file mode 100644
index 000000000000..fcd867b7147d
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
@@ -0,0 +1,69 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef AMDGPU_EV_FENCE_H_
+#define AMDGPU_EV_FENCE_H_
+
+struct amdgpu_eviction_fence {
+	struct dma_fence base;
+	spinlock_t	 lock;
+	char		 timeline_name[TASK_COMM_LEN];
+	struct amdgpu_eviction_fence_mgr *evf_mgr;
+};
+
+struct amdgpu_eviction_fence_mgr {
+	u64			ev_fence_ctx;
+	atomic_t		ev_fence_seq;
+	spinlock_t		ev_fence_lock;
+	struct amdgpu_eviction_fence *ev_fence;
+	struct delayed_work	suspend_work;
+	uint8_t fd_closing;
+};
+
+/* Eviction fence helper functions */
+struct amdgpu_eviction_fence *
+amdgpu_eviction_fence_create(struct amdgpu_eviction_fence_mgr *evf_mgr);
+
+void
+amdgpu_eviction_fence_destroy(struct amdgpu_eviction_fence_mgr *evf_mgr);
+
+int
+amdgpu_eviction_fence_attach(struct amdgpu_eviction_fence_mgr *evf_mgr,
+			     struct amdgpu_bo *bo);
+
+void
+amdgpu_eviction_fence_detach(struct amdgpu_eviction_fence_mgr *evf_mgr,
+			     struct amdgpu_bo *bo);
+
+int
+amdgpu_eviction_fence_init(struct amdgpu_eviction_fence_mgr *evf_mgr);
+
+void
+amdgpu_eviction_fence_signal(struct amdgpu_eviction_fence_mgr *evf_mgr,
+			     struct amdgpu_eviction_fence *ev_fence);
+
+int
+amdgpu_eviction_fence_replace_fence(struct amdgpu_eviction_fence_mgr *evf_mgr,
+				    struct drm_exec *exec);
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c
deleted file mode 100644
index 236d9950221b..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c
+++ /dev/null
@@ -1,440 +0,0 @@
-/*
- * Copyright © 2007 David Airlie
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *     David Airlie
- */
-#include <linux/module.h>
-#include <linux/slab.h>
-#include <linux/pm_runtime.h>
-
-#include <drm/drmP.h>
-#include <drm/drm_crtc.h>
-#include <drm/drm_crtc_helper.h>
-#include <drm/amdgpu_drm.h>
-#include "amdgpu.h"
-#include "cikd.h"
-
-#include <drm/drm_fb_helper.h>
-
-#include <linux/vga_switcheroo.h>
-
-/* object hierarchy -
-   this contains a helper + a amdgpu fb
-   the helper contains a pointer to amdgpu framebuffer baseclass.
-*/
-struct amdgpu_fbdev {
-	struct drm_fb_helper helper;
-	struct amdgpu_framebuffer rfb;
-	struct amdgpu_device *adev;
-};
-
-static int
-amdgpufb_open(struct fb_info *info, int user)
-{
-	struct amdgpu_fbdev *rfbdev = info->par;
-	struct amdgpu_device *adev = rfbdev->adev;
-	int ret = pm_runtime_get_sync(adev->ddev->dev);
-	if (ret < 0 && ret != -EACCES) {
-		pm_runtime_mark_last_busy(adev->ddev->dev);
-		pm_runtime_put_autosuspend(adev->ddev->dev);
-		return ret;
-	}
-	return 0;
-}
-
-static int
-amdgpufb_release(struct fb_info *info, int user)
-{
-	struct amdgpu_fbdev *rfbdev = info->par;
-	struct amdgpu_device *adev = rfbdev->adev;
-
-	pm_runtime_mark_last_busy(adev->ddev->dev);
-	pm_runtime_put_autosuspend(adev->ddev->dev);
-	return 0;
-}
-
-static struct fb_ops amdgpufb_ops = {
-	.owner = THIS_MODULE,
-	DRM_FB_HELPER_DEFAULT_OPS,
-	.fb_open = amdgpufb_open,
-	.fb_release = amdgpufb_release,
-	.fb_fillrect = drm_fb_helper_cfb_fillrect,
-	.fb_copyarea = drm_fb_helper_cfb_copyarea,
-	.fb_imageblit = drm_fb_helper_cfb_imageblit,
-};
-
-
-int amdgpu_align_pitch(struct amdgpu_device *adev, int width, int cpp, bool tiled)
-{
-	int aligned = width;
-	int pitch_mask = 0;
-
-	switch (cpp) {
-	case 1:
-		pitch_mask = 255;
-		break;
-	case 2:
-		pitch_mask = 127;
-		break;
-	case 3:
-	case 4:
-		pitch_mask = 63;
-		break;
-	}
-
-	aligned += pitch_mask;
-	aligned &= ~pitch_mask;
-	return aligned * cpp;
-}
-
-static void amdgpufb_destroy_pinned_object(struct drm_gem_object *gobj)
-{
-	struct amdgpu_bo *abo = gem_to_amdgpu_bo(gobj);
-	int ret;
-
-	ret = amdgpu_bo_reserve(abo, true);
-	if (likely(ret == 0)) {
-		amdgpu_bo_kunmap(abo);
-		amdgpu_bo_unpin(abo);
-		amdgpu_bo_unreserve(abo);
-	}
-	drm_gem_object_unreference_unlocked(gobj);
-}
-
-static int amdgpufb_create_pinned_object(struct amdgpu_fbdev *rfbdev,
-					 struct drm_mode_fb_cmd2 *mode_cmd,
-					 struct drm_gem_object **gobj_p)
-{
-	struct amdgpu_device *adev = rfbdev->adev;
-	struct drm_gem_object *gobj = NULL;
-	struct amdgpu_bo *abo = NULL;
-	bool fb_tiled = false; /* useful for testing */
-	u32 tiling_flags = 0;
-	int ret;
-	int aligned_size, size;
-	int height = mode_cmd->height;
-	u32 cpp;
-
-	cpp = drm_format_plane_cpp(mode_cmd->pixel_format, 0);
-
-	/* need to align pitch with crtc limits */
-	mode_cmd->pitches[0] = amdgpu_align_pitch(adev, mode_cmd->width, cpp,
-						  fb_tiled);
-
-	height = ALIGN(mode_cmd->height, 8);
-	size = mode_cmd->pitches[0] * height;
-	aligned_size = ALIGN(size, PAGE_SIZE);
-	ret = amdgpu_gem_object_create(adev, aligned_size, 0,
-				       AMDGPU_GEM_DOMAIN_VRAM,
-				       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-				       AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
-				       AMDGPU_GEM_CREATE_VRAM_CLEARED,
-				       true, &gobj);
-	if (ret) {
-		pr_err("failed to allocate framebuffer (%d)\n", aligned_size);
-		return -ENOMEM;
-	}
-	abo = gem_to_amdgpu_bo(gobj);
-
-	if (fb_tiled)
-		tiling_flags = AMDGPU_TILING_SET(ARRAY_MODE, GRPH_ARRAY_2D_TILED_THIN1);
-
-	ret = amdgpu_bo_reserve(abo, false);
-	if (unlikely(ret != 0))
-		goto out_unref;
-
-	if (tiling_flags) {
-		ret = amdgpu_bo_set_tiling_flags(abo,
-						 tiling_flags);
-		if (ret)
-			dev_err(adev->dev, "FB failed to set tiling flags\n");
-	}
-
-
-	ret = amdgpu_bo_pin(abo, AMDGPU_GEM_DOMAIN_VRAM, NULL);
-	if (ret) {
-		amdgpu_bo_unreserve(abo);
-		goto out_unref;
-	}
-	ret = amdgpu_bo_kmap(abo, NULL);
-	amdgpu_bo_unreserve(abo);
-	if (ret) {
-		goto out_unref;
-	}
-
-	*gobj_p = gobj;
-	return 0;
-out_unref:
-	amdgpufb_destroy_pinned_object(gobj);
-	*gobj_p = NULL;
-	return ret;
-}
-
-static int amdgpufb_create(struct drm_fb_helper *helper,
-			   struct drm_fb_helper_surface_size *sizes)
-{
-	struct amdgpu_fbdev *rfbdev = (struct amdgpu_fbdev *)helper;
-	struct amdgpu_device *adev = rfbdev->adev;
-	struct fb_info *info;
-	struct drm_framebuffer *fb = NULL;
-	struct drm_mode_fb_cmd2 mode_cmd;
-	struct drm_gem_object *gobj = NULL;
-	struct amdgpu_bo *abo = NULL;
-	int ret;
-	unsigned long tmp;
-
-	mode_cmd.width = sizes->surface_width;
-	mode_cmd.height = sizes->surface_height;
-
-	if (sizes->surface_bpp == 24)
-		sizes->surface_bpp = 32;
-
-	mode_cmd.pixel_format = drm_mode_legacy_fb_format(sizes->surface_bpp,
-							  sizes->surface_depth);
-
-	ret = amdgpufb_create_pinned_object(rfbdev, &mode_cmd, &gobj);
-	if (ret) {
-		DRM_ERROR("failed to create fbcon object %d\n", ret);
-		return ret;
-	}
-
-	abo = gem_to_amdgpu_bo(gobj);
-
-	/* okay we have an object now allocate the framebuffer */
-	info = drm_fb_helper_alloc_fbi(helper);
-	if (IS_ERR(info)) {
-		ret = PTR_ERR(info);
-		goto out;
-	}
-
-	info->par = rfbdev;
-	info->skip_vt_switch = true;
-
-	ret = amdgpu_framebuffer_init(adev->ddev, &rfbdev->rfb, &mode_cmd, gobj);
-	if (ret) {
-		DRM_ERROR("failed to initialize framebuffer %d\n", ret);
-		goto out;
-	}
-
-	fb = &rfbdev->rfb.base;
-
-	/* setup helper */
-	rfbdev->helper.fb = fb;
-
-	strcpy(info->fix.id, "amdgpudrmfb");
-
-	drm_fb_helper_fill_fix(info, fb->pitches[0], fb->format->depth);
-
-	info->flags = FBINFO_DEFAULT | FBINFO_CAN_FORCE_OUTPUT;
-	info->fbops = &amdgpufb_ops;
-
-	tmp = amdgpu_bo_gpu_offset(abo) - adev->mc.vram_start;
-	info->fix.smem_start = adev->mc.aper_base + tmp;
-	info->fix.smem_len = amdgpu_bo_size(abo);
-	info->screen_base = abo->kptr;
-	info->screen_size = amdgpu_bo_size(abo);
-
-	drm_fb_helper_fill_var(info, &rfbdev->helper, sizes->fb_width, sizes->fb_height);
-
-	/* setup aperture base/size for vesafb takeover */
-	info->apertures->ranges[0].base = adev->ddev->mode_config.fb_base;
-	info->apertures->ranges[0].size = adev->mc.aper_size;
-
-	/* Use default scratch pixmap (info->pixmap.flags = FB_PIXMAP_SYSTEM) */
-
-	if (info->screen_base == NULL) {
-		ret = -ENOSPC;
-		goto out;
-	}
-
-	DRM_INFO("fb mappable at 0x%lX\n",  info->fix.smem_start);
-	DRM_INFO("vram apper at 0x%lX\n",  (unsigned long)adev->mc.aper_base);
-	DRM_INFO("size %lu\n", (unsigned long)amdgpu_bo_size(abo));
-	DRM_INFO("fb depth is %d\n", fb->format->depth);
-	DRM_INFO("   pitch is %d\n", fb->pitches[0]);
-
-	vga_switcheroo_client_fb_set(adev->ddev->pdev, info);
-	return 0;
-
-out:
-	if (abo) {
-
-	}
-	if (fb && ret) {
-		drm_gem_object_unreference_unlocked(gobj);
-		drm_framebuffer_unregister_private(fb);
-		drm_framebuffer_cleanup(fb);
-		kfree(fb);
-	}
-	return ret;
-}
-
-void amdgpu_fb_output_poll_changed(struct amdgpu_device *adev)
-{
-	if (adev->mode_info.rfbdev)
-		drm_fb_helper_hotplug_event(&adev->mode_info.rfbdev->helper);
-}
-
-static int amdgpu_fbdev_destroy(struct drm_device *dev, struct amdgpu_fbdev *rfbdev)
-{
-	struct amdgpu_framebuffer *rfb = &rfbdev->rfb;
-
-	drm_fb_helper_unregister_fbi(&rfbdev->helper);
-
-	if (rfb->obj) {
-		amdgpufb_destroy_pinned_object(rfb->obj);
-		rfb->obj = NULL;
-	}
-	drm_fb_helper_fini(&rfbdev->helper);
-	drm_framebuffer_unregister_private(&rfb->base);
-	drm_framebuffer_cleanup(&rfb->base);
-
-	return 0;
-}
-
-/** Sets the color ramps on behalf of fbcon */
-static void amdgpu_crtc_fb_gamma_set(struct drm_crtc *crtc, u16 red, u16 green,
-				      u16 blue, int regno)
-{
-	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
-
-	amdgpu_crtc->lut_r[regno] = red >> 6;
-	amdgpu_crtc->lut_g[regno] = green >> 6;
-	amdgpu_crtc->lut_b[regno] = blue >> 6;
-}
-
-/** Gets the color ramps on behalf of fbcon */
-static void amdgpu_crtc_fb_gamma_get(struct drm_crtc *crtc, u16 *red, u16 *green,
-				      u16 *blue, int regno)
-{
-	struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
-
-	*red = amdgpu_crtc->lut_r[regno] << 6;
-	*green = amdgpu_crtc->lut_g[regno] << 6;
-	*blue = amdgpu_crtc->lut_b[regno] << 6;
-}
-
-static const struct drm_fb_helper_funcs amdgpu_fb_helper_funcs = {
-	.gamma_set = amdgpu_crtc_fb_gamma_set,
-	.gamma_get = amdgpu_crtc_fb_gamma_get,
-	.fb_probe = amdgpufb_create,
-};
-
-int amdgpu_fbdev_init(struct amdgpu_device *adev)
-{
-	struct amdgpu_fbdev *rfbdev;
-	int bpp_sel = 32;
-	int ret;
-
-	/* don't init fbdev on hw without DCE */
-	if (!adev->mode_info.mode_config_initialized)
-		return 0;
-
-	/* don't init fbdev if there are no connectors */
-	if (list_empty(&adev->ddev->mode_config.connector_list))
-		return 0;
-
-	/* select 8 bpp console on low vram cards */
-	if (adev->mc.real_vram_size <= (32*1024*1024))
-		bpp_sel = 8;
-
-	rfbdev = kzalloc(sizeof(struct amdgpu_fbdev), GFP_KERNEL);
-	if (!rfbdev)
-		return -ENOMEM;
-
-	rfbdev->adev = adev;
-	adev->mode_info.rfbdev = rfbdev;
-
-	drm_fb_helper_prepare(adev->ddev, &rfbdev->helper,
-			&amdgpu_fb_helper_funcs);
-
-	ret = drm_fb_helper_init(adev->ddev, &rfbdev->helper,
-				 AMDGPUFB_CONN_LIMIT);
-	if (ret) {
-		kfree(rfbdev);
-		return ret;
-	}
-
-	drm_fb_helper_single_add_all_connectors(&rfbdev->helper);
-
-	/* disable all the possible outputs/crtcs before entering KMS mode */
-	drm_helper_disable_unused_functions(adev->ddev);
-
-	drm_fb_helper_initial_config(&rfbdev->helper, bpp_sel);
-	return 0;
-}
-
-void amdgpu_fbdev_fini(struct amdgpu_device *adev)
-{
-	if (!adev->mode_info.rfbdev)
-		return;
-
-	amdgpu_fbdev_destroy(adev->ddev, adev->mode_info.rfbdev);
-	kfree(adev->mode_info.rfbdev);
-	adev->mode_info.rfbdev = NULL;
-}
-
-void amdgpu_fbdev_set_suspend(struct amdgpu_device *adev, int state)
-{
-	if (adev->mode_info.rfbdev)
-		drm_fb_helper_set_suspend(&adev->mode_info.rfbdev->helper,
-			state);
-}
-
-int amdgpu_fbdev_total_size(struct amdgpu_device *adev)
-{
-	struct amdgpu_bo *robj;
-	int size = 0;
-
-	if (!adev->mode_info.rfbdev)
-		return 0;
-
-	robj = gem_to_amdgpu_bo(adev->mode_info.rfbdev->rfb.obj);
-	size += amdgpu_bo_size(robj);
-	return size;
-}
-
-bool amdgpu_fbdev_robj_is_fb(struct amdgpu_device *adev, struct amdgpu_bo *robj)
-{
-	if (!adev->mode_info.rfbdev)
-		return false;
-	if (robj == gem_to_amdgpu_bo(adev->mode_info.rfbdev->rfb.obj))
-		return true;
-	return false;
-}
-
-void amdgpu_fbdev_restore_mode(struct amdgpu_device *adev)
-{
-	struct amdgpu_fbdev *afbdev = adev->mode_info.rfbdev;
-	struct drm_fb_helper *fb_helper;
-	int ret;
-
-	if (!afbdev)
-		return;
-
-	fb_helper = &afbdev->helper;
-
-	ret = drm_fb_helper_restore_fbdev_mode_unlocked(fb_helper);
-	if (ret)
-		DRM_DEBUG("failed to restore crtc mode\n");
-}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
new file mode 100644
index 000000000000..b349bb3676d5
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c
@@ -0,0 +1,124 @@
+// SPDX-License-Identifier: MIT
+/* Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: David Nieto
+ *          Roy Sun
+ */
+
+#include <linux/debugfs.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/uaccess.h>
+#include <linux/reboot.h>
+#include <linux/syscalls.h>
+
+#include <drm/amdgpu_drm.h>
+#include <drm/drm_debugfs.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_file.h>
+
+#include "amdgpu.h"
+#include "amdgpu_vm.h"
+#include "amdgpu_gem.h"
+#include "amdgpu_ctx.h"
+#include "amdgpu_fdinfo.h"
+
+
+static const char *amdgpu_ip_name[AMDGPU_HW_IP_NUM] = {
+	[AMDGPU_HW_IP_GFX]	=	"gfx",
+	[AMDGPU_HW_IP_COMPUTE]	=	"compute",
+	[AMDGPU_HW_IP_DMA]	=	"dma",
+	[AMDGPU_HW_IP_UVD]	=	"dec",
+	[AMDGPU_HW_IP_VCE]	=	"enc",
+	[AMDGPU_HW_IP_UVD_ENC]	=	"enc_1",
+	[AMDGPU_HW_IP_VCN_DEC]	=	"dec",
+	[AMDGPU_HW_IP_VCN_ENC]	=	"enc",
+	[AMDGPU_HW_IP_VCN_JPEG]	=	"jpeg",
+	[AMDGPU_HW_IP_VPE]	=	"vpe",
+};
+
+void amdgpu_show_fdinfo(struct drm_printer *p, struct drm_file *file)
+{
+	struct amdgpu_fpriv *fpriv = file->driver_priv;
+	struct amdgpu_vm *vm = &fpriv->vm;
+
+	struct amdgpu_mem_stats stats[__AMDGPU_PL_NUM];
+	ktime_t usage[AMDGPU_HW_IP_NUM];
+	const char *pl_name[] = {
+		[TTM_PL_VRAM] = "vram",
+		[TTM_PL_TT] = "gtt",
+		[TTM_PL_SYSTEM] = "cpu",
+		[AMDGPU_PL_GDS] = "gds",
+		[AMDGPU_PL_GWS] = "gws",
+		[AMDGPU_PL_OA] = "oa",
+		[AMDGPU_PL_DOORBELL] = "doorbell",
+		[AMDGPU_PL_MMIO_REMAP] = "mmioremap",
+	};
+	unsigned int hw_ip, i;
+
+	amdgpu_vm_get_memory(vm, stats);
+	amdgpu_ctx_mgr_usage(&fpriv->ctx_mgr, usage);
+
+	/*
+	 * ******************************************************************
+	 * For text output format description please see drm-usage-stats.rst!
+	 * ******************************************************************
+	 */
+
+	drm_printf(p, "pasid:\t%u\n", fpriv->vm.pasid);
+
+	for (i = 0; i < ARRAY_SIZE(pl_name); i++) {
+		if (!pl_name[i])
+			continue;
+
+		drm_print_memory_stats(p,
+				       &stats[i].drm,
+				       DRM_GEM_OBJECT_RESIDENT |
+				       DRM_GEM_OBJECT_PURGEABLE,
+				       pl_name[i]);
+	}
+
+	/* Legacy amdgpu keys, alias to drm-resident-memory-: */
+	drm_printf(p, "drm-memory-vram:\t%llu KiB\n",
+		   stats[TTM_PL_VRAM].drm.resident/1024UL);
+	drm_printf(p, "drm-memory-gtt: \t%llu KiB\n",
+		   stats[TTM_PL_TT].drm.resident/1024UL);
+	drm_printf(p, "drm-memory-cpu: \t%llu KiB\n",
+		   stats[TTM_PL_SYSTEM].drm.resident/1024UL);
+
+	/* Amdgpu specific memory accounting keys: */
+	drm_printf(p, "amd-evicted-vram:\t%llu KiB\n",
+		   stats[TTM_PL_VRAM].evicted/1024UL);
+	drm_printf(p, "amd-requested-vram:\t%llu KiB\n",
+		   (stats[TTM_PL_VRAM].drm.shared +
+		    stats[TTM_PL_VRAM].drm.private) / 1024UL);
+	drm_printf(p, "amd-requested-gtt:\t%llu KiB\n",
+		   (stats[TTM_PL_TT].drm.shared +
+		    stats[TTM_PL_TT].drm.private) / 1024UL);
+
+	for (hw_ip = 0; hw_ip < AMDGPU_HW_IP_NUM; ++hw_ip) {
+		if (!usage[hw_ip])
+			continue;
+
+		drm_printf(p, "drm-engine-%s:\t%lld ns\n", amdgpu_ip_name[hw_ip],
+			   ktime_to_ns(usage[hw_ip]));
+	}
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h
new file mode 100644
index 000000000000..0398f5a159ef
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.h
@@ -0,0 +1,42 @@
+/* SPDX-License-Identifier: MIT
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: David Nieto
+ *          Roy Sun
+ */
+#ifndef __AMDGPU_SMI_H__
+#define __AMDGPU_SMI_H__
+
+#include <linux/idr.h>
+#include <linux/kfifo.h>
+#include <linux/rbtree.h>
+#include <drm/gpu_scheduler.h>
+#include <drm/drm_file.h>
+#include <linux/sched/mm.h>
+
+#include "amdgpu_sync.h"
+#include "amdgpu_ring.h"
+#include "amdgpu_ids.h"
+
+uint32_t amdgpu_get_ip_count(struct amdgpu_device *adev, int id);
+void amdgpu_show_fdinfo(struct drm_printer *p, struct drm_file *file);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 7b60fb79c3a6..c7843e336310 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -34,44 +34,14 @@
 #include <linux/kref.h>
 #include <linux/slab.h>
 #include <linux/firmware.h>
-#include <drm/drmP.h>
+#include <linux/pm_runtime.h>
+
+#include <drm/drm_drv.h>
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
+#include "amdgpu_reset.h"
 
 /*
- * Fences
- * Fences mark an event in the GPUs pipeline and are used
- * for GPU/CPU synchronization.  When the fence is written,
- * it is expected that all buffers associated with that fence
- * are no longer in use by the associated ring on the GPU and
- * that the the relevant GPU caches have been flushed.
- */
-
-struct amdgpu_fence {
-	struct dma_fence base;
-
-	/* RB, DMA, etc. */
-	struct amdgpu_ring		*ring;
-};
-
-static struct kmem_cache *amdgpu_fence_slab;
-
-int amdgpu_fence_slab_init(void)
-{
-	amdgpu_fence_slab = kmem_cache_create(
-		"amdgpu_fence", sizeof(struct amdgpu_fence), 0,
-		SLAB_HWCACHE_ALIGN, NULL);
-	if (!amdgpu_fence_slab)
-		return -ENOMEM;
-	return 0;
-}
-
-void amdgpu_fence_slab_fini(void)
-{
-	rcu_barrier();
-	kmem_cache_destroy(amdgpu_fence_slab);
-}
-/*
  * Cast helper
  */
 static const struct dma_fence_ops amdgpu_fence_ops;
@@ -79,10 +49,7 @@ static inline struct amdgpu_fence *to_amdgpu_fence(struct dma_fence *f)
 {
 	struct amdgpu_fence *__f = container_of(f, struct amdgpu_fence, base);
 
-	if (__f->base.ops == &amdgpu_fence_ops)
-		return __f;
-
-	return NULL;
+	return __f;
 }
 
 /**
@@ -126,44 +93,90 @@ static u32 amdgpu_fence_read(struct amdgpu_ring *ring)
  * amdgpu_fence_emit - emit a fence on the requested ring
  *
  * @ring: ring the fence is associated with
- * @f: resulting fence object
+ * @af: amdgpu fence input
+ * @flags: flags to pass into the subordinate .emit_fence() call
  *
  * Emits a fence command on the requested ring (all asics).
  * Returns 0 on success, -ENOMEM on failure.
  */
-int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f)
+int amdgpu_fence_emit(struct amdgpu_ring *ring, struct amdgpu_fence *af,
+		      unsigned int flags)
 {
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_fence *fence;
-	struct dma_fence *old, **ptr;
+	struct dma_fence *fence;
+	struct dma_fence __rcu **ptr;
 	uint32_t seq;
+	int r;
 
-	fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL);
-	if (fence == NULL)
-		return -ENOMEM;
+	fence = &af->base;
+	af->ring = ring;
 
 	seq = ++ring->fence_drv.sync_seq;
-	fence->ring = ring;
-	dma_fence_init(&fence->base, &amdgpu_fence_ops,
+	dma_fence_init(fence, &amdgpu_fence_ops,
 		       &ring->fence_drv.lock,
-		       adev->fence_context + ring->idx,
-		       seq);
-	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
-			       seq, AMDGPU_FENCE_FLAG_INT);
+		       adev->fence_context + ring->idx, seq);
 
+	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
+			       seq, flags | AMDGPU_FENCE_FLAG_INT);
+	amdgpu_fence_save_wptr(af);
+	pm_runtime_get_noresume(adev_to_drm(adev)->dev);
 	ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask];
+	if (unlikely(rcu_dereference_protected(*ptr, 1))) {
+		struct dma_fence *old;
+
+		rcu_read_lock();
+		old = dma_fence_get_rcu_safe(ptr);
+		rcu_read_unlock();
+
+		if (old) {
+			r = dma_fence_wait(old, false);
+			dma_fence_put(old);
+			if (r)
+				return r;
+		}
+	}
+
+	to_amdgpu_fence(fence)->start_timestamp = ktime_get();
+
 	/* This function can't be called concurrently anyway, otherwise
 	 * emitting the fence would mess up the hardware ring buffer.
 	 */
-	old = rcu_dereference_protected(*ptr, 1);
-	if (old && !dma_fence_is_signaled(old)) {
-		DRM_INFO("rcu slot is busy\n");
-		dma_fence_wait(old, false);
-	}
+	rcu_assign_pointer(*ptr, dma_fence_get(fence));
 
-	rcu_assign_pointer(*ptr, dma_fence_get(&fence->base));
+	return 0;
+}
 
-	*f = &fence->base;
+/**
+ * amdgpu_fence_emit_polling - emit a fence on the requeste ring
+ *
+ * @ring: ring the fence is associated with
+ * @s: resulting sequence number
+ * @timeout: the timeout for waiting in usecs
+ *
+ * Emits a fence command on the requested ring (all asics).
+ * Used For polling fence.
+ * Returns 0 on success, -ENOMEM on failure.
+ */
+int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s,
+			      uint32_t timeout)
+{
+	uint32_t seq;
+	signed long r;
+
+	if (!s)
+		return -EINVAL;
+
+	seq = ++ring->fence_drv.sync_seq;
+	r = amdgpu_fence_wait_polling(ring,
+				      seq - ring->fence_drv.num_fences_mask,
+				      timeout);
+	if (r < 1)
+		return -ETIMEDOUT;
+
+	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
+			       seq, 0);
+
+	*s = seq;
 
 	return 0;
 }
@@ -189,12 +202,14 @@ static void amdgpu_fence_schedule_fallback(struct amdgpu_ring *ring)
  * Checks the current fence value and calculates the last
  * signalled fence value. Wakes the fence queue if the
  * sequence number has increased.
+ *
+ * Returns true if fence was processed
  */
-void amdgpu_fence_process(struct amdgpu_ring *ring)
+bool amdgpu_fence_process(struct amdgpu_ring *ring)
 {
 	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	struct amdgpu_device *adev = ring->adev;
 	uint32_t seq, last_seq;
-	int r;
 
 	do {
 		last_seq = atomic_read(&ring->fence_drv.last_seq);
@@ -202,17 +217,19 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
 
 	} while (atomic_cmpxchg(&drv->last_seq, last_seq, seq) != last_seq);
 
-	if (seq != ring->fence_drv.sync_seq)
+	if (timer_delete(&ring->fence_drv.fallback_timer) &&
+	    seq != ring->fence_drv.sync_seq)
 		amdgpu_fence_schedule_fallback(ring);
 
 	if (unlikely(seq == last_seq))
-		return;
+		return false;
 
 	last_seq &= drv->num_fences_mask;
 	seq &= drv->num_fences_mask;
 
 	do {
 		struct dma_fence *fence, **ptr;
+		struct amdgpu_fence *am_fence;
 
 		++last_seq;
 		last_seq &= drv->num_fences_mask;
@@ -225,34 +242,41 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
 		if (!fence)
 			continue;
 
-		r = dma_fence_signal(fence);
-		if (!r)
-			DMA_FENCE_TRACE(fence, "signaled from irq context\n");
-		else
-			BUG();
-
+		/* Save the wptr in the fence driver so we know what the last processed
+		 * wptr was.  This is required for re-emitting the ring state for
+		 * queues that are reset but are not guilty and thus have no guilty fence.
+		 */
+		am_fence = container_of(fence, struct amdgpu_fence, base);
+		drv->signalled_wptr = am_fence->wptr;
+		dma_fence_signal(fence);
 		dma_fence_put(fence);
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
 	} while (last_seq != seq);
+
+	return true;
 }
 
 /**
  * amdgpu_fence_fallback - fallback for hardware interrupts
  *
- * @work: delayed work item
+ * @t: timer context used to obtain the pointer to ring structure
  *
  * Checks for fence activity.
  */
-static void amdgpu_fence_fallback(unsigned long arg)
+static void amdgpu_fence_fallback(struct timer_list *t)
 {
-	struct amdgpu_ring *ring = (void *)arg;
+	struct amdgpu_ring *ring = timer_container_of(ring, t,
+						      fence_drv.fallback_timer);
 
-	amdgpu_fence_process(ring);
+	if (amdgpu_fence_process(ring))
+		dev_warn(ring->adev->dev,
+			 "Fence fallback timer expired on ring %s\n",
+			 ring->name);
 }
 
 /**
  * amdgpu_fence_wait_empty - wait for all fences to signal
  *
- * @adev: amdgpu device pointer
  * @ring: ring index the fence is associated with
  *
  * Wait for all fences on the requested ring to signal (all asics).
@@ -260,7 +284,7 @@ static void amdgpu_fence_fallback(unsigned long arg)
  */
 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
 {
-	uint64_t seq = ACCESS_ONCE(ring->fence_drv.sync_seq);
+	uint64_t seq = READ_ONCE(ring->fence_drv.sync_seq);
 	struct dma_fence *fence, **ptr;
 	int r;
 
@@ -282,6 +306,27 @@ int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
 }
 
 /**
+ * amdgpu_fence_wait_polling - busy wait for givn sequence number
+ *
+ * @ring: ring index the fence is associated with
+ * @wait_seq: sequence number to wait
+ * @timeout: the timeout for waiting in usecs
+ *
+ * Wait for all fences on the requested ring to signal (all asics).
+ * Returns left time if no timeout, 0 or minus if timeout.
+ */
+signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring,
+				      uint32_t wait_seq,
+				      signed long timeout)
+{
+
+	while ((int32_t)(wait_seq - amdgpu_fence_read(ring)) > 0 && timeout > 0) {
+		udelay(2);
+		timeout -= 2;
+	}
+	return timeout > 0 ? timeout : 0;
+}
+/**
  * amdgpu_fence_count_emitted - get the count of emitted fences
  *
  * @ring: ring the fence is associated with
@@ -290,21 +335,71 @@ int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
  * Returns the number of emitted fences on the ring.  Used by the
  * dynpm code to ring track activity.
  */
-unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring)
+unsigned int amdgpu_fence_count_emitted(struct amdgpu_ring *ring)
 {
 	uint64_t emitted;
 
 	/* We are not protected by ring lock when reading the last sequence
 	 * but it's ok to report slightly wrong fence count here.
 	 */
-	amdgpu_fence_process(ring);
 	emitted = 0x100000000ull;
 	emitted -= atomic_read(&ring->fence_drv.last_seq);
-	emitted += ACCESS_ONCE(ring->fence_drv.sync_seq);
+	emitted += READ_ONCE(ring->fence_drv.sync_seq);
 	return lower_32_bits(emitted);
 }
 
 /**
+ * amdgpu_fence_last_unsignaled_time_us - the time fence emitted until now
+ * @ring: ring the fence is associated with
+ *
+ * Find the earliest fence unsignaled until now, calculate the time delta
+ * between the time fence emitted and now.
+ */
+u64 amdgpu_fence_last_unsignaled_time_us(struct amdgpu_ring *ring)
+{
+	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	struct dma_fence *fence;
+	uint32_t last_seq, sync_seq;
+
+	last_seq = atomic_read(&ring->fence_drv.last_seq);
+	sync_seq = READ_ONCE(ring->fence_drv.sync_seq);
+	if (last_seq == sync_seq)
+		return 0;
+
+	++last_seq;
+	last_seq &= drv->num_fences_mask;
+	fence = drv->fences[last_seq];
+	if (!fence)
+		return 0;
+
+	return ktime_us_delta(ktime_get(),
+		to_amdgpu_fence(fence)->start_timestamp);
+}
+
+/**
+ * amdgpu_fence_update_start_timestamp - update the timestamp of the fence
+ * @ring: ring the fence is associated with
+ * @seq: the fence seq number to update.
+ * @timestamp: the start timestamp to update.
+ *
+ * The function called at the time the fence and related ib is about to
+ * resubmit to gpu in MCBP scenario. Thus we do not consider race condition
+ * with amdgpu_fence_process to modify the same fence.
+ */
+void amdgpu_fence_update_start_timestamp(struct amdgpu_ring *ring, uint32_t seq, ktime_t timestamp)
+{
+	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	struct dma_fence *fence;
+
+	seq &= drv->num_fences_mask;
+	fence = drv->fences[seq];
+	if (!fence)
+		return;
+
+	to_amdgpu_fence(fence)->start_timestamp = timestamp;
+}
+
+/**
  * amdgpu_fence_driver_start_ring - make the fence driver
  * ready for use on the requested ring.
  *
@@ -319,30 +414,28 @@ unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring)
  */
 int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
 				   struct amdgpu_irq_src *irq_src,
-				   unsigned irq_type)
+				   unsigned int irq_type)
 {
 	struct amdgpu_device *adev = ring->adev;
 	uint64_t index;
 
-	if (ring != &adev->uvd.ring) {
-		ring->fence_drv.cpu_addr = &adev->wb.wb[ring->fence_offs];
-		ring->fence_drv.gpu_addr = adev->wb.gpu_addr + (ring->fence_offs * 4);
+	if (ring->funcs->type != AMDGPU_RING_TYPE_UVD) {
+		ring->fence_drv.cpu_addr = ring->fence_cpu_addr;
+		ring->fence_drv.gpu_addr = ring->fence_gpu_addr;
 	} else {
 		/* put fence directly behind firmware */
 		index = ALIGN(adev->uvd.fw->size, 8);
-		ring->fence_drv.cpu_addr = adev->uvd.cpu_addr + index;
-		ring->fence_drv.gpu_addr = adev->uvd.gpu_addr + index;
+		ring->fence_drv.cpu_addr = adev->uvd.inst[ring->me].cpu_addr + index;
+		ring->fence_drv.gpu_addr = adev->uvd.inst[ring->me].gpu_addr + index;
 	}
 	amdgpu_fence_write(ring, atomic_read(&ring->fence_drv.last_seq));
-	amdgpu_irq_get(adev, irq_src, irq_type);
 
 	ring->fence_drv.irq_src = irq_src;
 	ring->fence_drv.irq_type = irq_type;
 	ring->fence_drv.initialized = true;
 
-	dev_info(adev->dev, "fence driver on ring %d use gpu addr 0x%016llx, "
-		 "cpu addr 0x%p\n", ring->idx,
-		 ring->fence_drv.gpu_addr, ring->fence_drv.cpu_addr);
+	DRM_DEV_DEBUG(adev->dev, "fence driver on ring %s use gpu addr 0x%016llx\n",
+		      ring->name, ring->fence_drv.gpu_addr);
 	return 0;
 }
 
@@ -351,19 +444,18 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
  * for the requested ring.
  *
  * @ring: ring to init the fence driver on
- * @num_hw_submission: number of entries on the hardware queue
  *
  * Init the fence driver for the requested ring (all asics).
  * Helper function for amdgpu_fence_driver_init().
  */
-int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
-				  unsigned num_hw_submission)
+int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
 {
-	long timeout;
-	int r;
+	struct amdgpu_device *adev = ring->adev;
 
-	/* Check that num_hw_submission is a power of two */
-	if ((num_hw_submission & (num_hw_submission - 1)) != 0)
+	if (!adev)
+		return -EINVAL;
+
+	if (!is_power_of_2(ring->num_hw_submission))
 		return -EINVAL;
 
 	ring->fence_drv.cpu_addr = NULL;
@@ -372,44 +464,21 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 	atomic_set(&ring->fence_drv.last_seq, 0);
 	ring->fence_drv.initialized = false;
 
-	setup_timer(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback,
-		    (unsigned long)ring);
+	timer_setup(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback, 0);
 
-	ring->fence_drv.num_fences_mask = num_hw_submission * 2 - 1;
+	ring->fence_drv.num_fences_mask = ring->num_hw_submission * 2 - 1;
 	spin_lock_init(&ring->fence_drv.lock);
-	ring->fence_drv.fences = kcalloc(num_hw_submission * 2, sizeof(void *),
+	ring->fence_drv.fences = kcalloc(ring->num_hw_submission * 2, sizeof(void *),
 					 GFP_KERNEL);
+
 	if (!ring->fence_drv.fences)
 		return -ENOMEM;
 
-	/* No need to setup the GPU scheduler for KIQ ring */
-	if (ring->funcs->type != AMDGPU_RING_TYPE_KIQ) {
-		timeout = msecs_to_jiffies(amdgpu_lockup_timeout);
-		if (timeout == 0) {
-			/*
-			 * FIXME:
-			 * Delayed workqueue cannot use it directly,
-			 * so the scheduler will not use delayed workqueue if
-			 * MAX_SCHEDULE_TIMEOUT is set.
-			 * Currently keep it simple and silly.
-			 */
-			timeout = MAX_SCHEDULE_TIMEOUT;
-		}
-		r = amd_sched_init(&ring->sched, &amdgpu_sched_ops,
-				   num_hw_submission,
-				   timeout, ring->name);
-		if (r) {
-			DRM_ERROR("Failed to create scheduler on ring %s.\n",
-				  ring->name);
-			return r;
-		}
-	}
-
 	return 0;
 }
 
 /**
- * amdgpu_fence_driver_init - init the fence driver
+ * amdgpu_fence_driver_sw_init - init the fence driver
  * for all possible rings.
  *
  * @adev: amdgpu device pointer
@@ -420,125 +489,308 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
  * amdgpu_fence_driver_start_ring().
  * Returns 0 for success.
  */
-int amdgpu_fence_driver_init(struct amdgpu_device *adev)
+int amdgpu_fence_driver_sw_init(struct amdgpu_device *adev)
 {
-	if (amdgpu_debugfs_fence_init(adev))
-		dev_err(adev->dev, "fence debugfs file creation failed\n");
-
 	return 0;
 }
 
 /**
- * amdgpu_fence_driver_fini - tear down the fence driver
+ * amdgpu_fence_need_ring_interrupt_restore - helper function to check whether
+ * fence driver interrupts need to be restored.
+ *
+ * @ring: ring that to be checked
+ *
+ * Interrupts for rings that belong to GFX IP don't need to be restored
+ * when the target power state is s0ix.
+ *
+ * Return true if need to restore interrupts, false otherwise.
+ */
+static bool amdgpu_fence_need_ring_interrupt_restore(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	bool is_gfx_power_domain = false;
+
+	switch (ring->funcs->type) {
+	case AMDGPU_RING_TYPE_SDMA:
+	/* SDMA 5.x+ is part of GFX power domain so it's covered by GFXOFF */
+		if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >=
+		    IP_VERSION(5, 0, 0))
+			is_gfx_power_domain = true;
+		break;
+	case AMDGPU_RING_TYPE_GFX:
+	case AMDGPU_RING_TYPE_COMPUTE:
+	case AMDGPU_RING_TYPE_KIQ:
+	case AMDGPU_RING_TYPE_MES:
+		is_gfx_power_domain = true;
+		break;
+	default:
+		break;
+	}
+
+	return !(adev->in_s0ix && is_gfx_power_domain);
+}
+
+/**
+ * amdgpu_fence_driver_hw_fini - tear down the fence driver
  * for all possible rings.
  *
  * @adev: amdgpu device pointer
  *
  * Tear down the fence driver for all possible rings (all asics).
  */
-void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
+void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
 {
-	unsigned i, j;
-	int r;
+	int i, r;
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
 		struct amdgpu_ring *ring = adev->rings[i];
 
 		if (!ring || !ring->fence_drv.initialized)
 			continue;
-		r = amdgpu_fence_wait_empty(ring);
-		if (r) {
-			/* no need to trigger GPU reset as we are unloading */
-			amdgpu_fence_driver_force_completion(adev);
-		}
-		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
-			       ring->fence_drv.irq_type);
-		amd_sched_fini(&ring->sched);
-		del_timer_sync(&ring->fence_drv.fallback_timer);
-		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
-			dma_fence_put(ring->fence_drv.fences[j]);
-		kfree(ring->fence_drv.fences);
-		ring->fence_drv.fences = NULL;
-		ring->fence_drv.initialized = false;
+
+		/* You can't wait for HW to signal if it's gone */
+		if (!drm_dev_is_unplugged(adev_to_drm(adev)))
+			r = amdgpu_fence_wait_empty(ring);
+		else
+			r = -ENODEV;
+		/* no need to trigger GPU reset as we are unloading */
+		if (r)
+			amdgpu_fence_driver_force_completion(ring);
+
+		if (!drm_dev_is_unplugged(adev_to_drm(adev)) &&
+		    ring->fence_drv.irq_src &&
+		    amdgpu_fence_need_ring_interrupt_restore(ring))
+			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
+				       ring->fence_drv.irq_type);
+
+		timer_delete_sync(&ring->fence_drv.fallback_timer);
 	}
 }
 
-/**
- * amdgpu_fence_driver_suspend - suspend the fence driver
- * for all possible rings.
- *
- * @adev: amdgpu device pointer
- *
- * Suspend the fence driver for all possible rings (all asics).
- */
-void amdgpu_fence_driver_suspend(struct amdgpu_device *adev)
+/* Will either stop and flush handlers for amdgpu interrupt or reanble it */
+void amdgpu_fence_driver_isr_toggle(struct amdgpu_device *adev, bool stop)
 {
-	int i, r;
+	int i;
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
 		struct amdgpu_ring *ring = adev->rings[i];
+
+		if (!ring || !ring->fence_drv.initialized || !ring->fence_drv.irq_src)
+			continue;
+
+		if (stop)
+			disable_irq(adev->irq.irq);
+		else
+			enable_irq(adev->irq.irq);
+	}
+}
+
+void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev)
+{
+	unsigned int i, j;
+
+	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
+		struct amdgpu_ring *ring = adev->rings[i];
+
 		if (!ring || !ring->fence_drv.initialized)
 			continue;
 
-		/* wait for gpu to finish processing current batch */
-		r = amdgpu_fence_wait_empty(ring);
-		if (r) {
-			/* delay GPU reset to resume */
-			amdgpu_fence_driver_force_completion(adev);
-		}
+		/*
+		 * Notice we check for sched.ops since there's some
+		 * override on the meaning of sched.ready by amdgpu.
+		 * The natural check would be sched.ready, which is
+		 * set as drm_sched_init() finishes...
+		 */
+		if (ring->sched.ops)
+			drm_sched_fini(&ring->sched);
 
-		/* disable the interrupt */
-		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
-			       ring->fence_drv.irq_type);
+		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
+			dma_fence_put(ring->fence_drv.fences[j]);
+		kfree(ring->fence_drv.fences);
+		ring->fence_drv.fences = NULL;
+		ring->fence_drv.initialized = false;
 	}
 }
 
 /**
- * amdgpu_fence_driver_resume - resume the fence driver
+ * amdgpu_fence_driver_hw_init - enable the fence driver
  * for all possible rings.
  *
  * @adev: amdgpu device pointer
  *
- * Resume the fence driver for all possible rings (all asics).
+ * Enable the fence driver for all possible rings (all asics).
  * Not all asics have all rings, so each asic will only
  * start the fence driver on the rings it has using
  * amdgpu_fence_driver_start_ring().
  * Returns 0 for success.
  */
-void amdgpu_fence_driver_resume(struct amdgpu_device *adev)
+void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev)
 {
 	int i;
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
 		struct amdgpu_ring *ring = adev->rings[i];
+
 		if (!ring || !ring->fence_drv.initialized)
 			continue;
 
 		/* enable the interrupt */
-		amdgpu_irq_get(adev, ring->fence_drv.irq_src,
-			       ring->fence_drv.irq_type);
+		if (ring->fence_drv.irq_src &&
+		    amdgpu_fence_need_ring_interrupt_restore(ring))
+			amdgpu_irq_get(adev, ring->fence_drv.irq_src,
+				       ring->fence_drv.irq_type);
 	}
 }
 
 /**
- * amdgpu_fence_driver_force_completion - force all fence waiter to complete
+ * amdgpu_fence_driver_set_error - set error code on fences
+ * @ring: the ring which contains the fences
+ * @error: the error code to set
  *
- * @adev: amdgpu device pointer
- *
- * In case of GPU reset failure make sure no process keep waiting on fence
- * that will never complete.
+ * Set an error code to all the fences pending on the ring.
  */
-void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev)
+void amdgpu_fence_driver_set_error(struct amdgpu_ring *ring, int error)
 {
-	int i;
+	struct amdgpu_fence_driver *drv = &ring->fence_drv;
+	unsigned long flags;
 
-	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		if (!ring || !ring->fence_drv.initialized)
-			continue;
+	spin_lock_irqsave(&drv->lock, flags);
+	for (unsigned int i = 0; i <= drv->num_fences_mask; ++i) {
+		struct dma_fence *fence;
 
-		amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
+		fence = rcu_dereference_protected(drv->fences[i],
+						  lockdep_is_held(&drv->lock));
+		if (fence && !dma_fence_is_signaled_locked(fence))
+			dma_fence_set_error(fence, error);
 	}
+	spin_unlock_irqrestore(&drv->lock, flags);
+}
+
+/**
+ * amdgpu_fence_driver_force_completion - force signal latest fence of ring
+ *
+ * @ring: fence of the ring to signal
+ *
+ */
+void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring)
+{
+	amdgpu_fence_driver_set_error(ring, -ECANCELED);
+	amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
+	amdgpu_fence_process(ring);
+}
+
+
+/*
+ * Kernel queue reset handling
+ *
+ * The driver can reset individual queues for most engines, but those queues
+ * may contain work from multiple contexts.  Resetting the queue will reset
+ * lose all of that state.  In order to minimize the collateral damage, the
+ * driver will save the ring contents which are not associated with the guilty
+ * context prior to resetting the queue.  After resetting the queue the queue
+ * contents from the other contexts is re-emitted to the rings so that it can
+ * be processed by the engine.  To handle this, we save the queue's write
+ * pointer (wptr) in the fences associated with each context.  If we get a
+ * queue timeout, we can then use the wptrs from the fences to determine
+ * which data needs to be saved out of the queue's ring buffer.
+ */
+
+/**
+ * amdgpu_fence_driver_guilty_force_completion - force signal of specified sequence
+ *
+ * @af: fence of the ring to signal
+ *
+ */
+void amdgpu_fence_driver_guilty_force_completion(struct amdgpu_fence *af)
+{
+	struct dma_fence *unprocessed;
+	struct dma_fence __rcu **ptr;
+	struct amdgpu_fence *fence;
+	struct amdgpu_ring *ring = af->ring;
+	unsigned long flags;
+	u32 seq, last_seq;
+
+	last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask;
+	seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask;
+
+	/* mark all fences from the guilty context with an error */
+	spin_lock_irqsave(&ring->fence_drv.lock, flags);
+	do {
+		last_seq++;
+		last_seq &= ring->fence_drv.num_fences_mask;
+
+		ptr = &ring->fence_drv.fences[last_seq];
+		rcu_read_lock();
+		unprocessed = rcu_dereference(*ptr);
+
+		if (unprocessed && !dma_fence_is_signaled_locked(unprocessed)) {
+			fence = container_of(unprocessed, struct amdgpu_fence, base);
+
+			if (fence == af)
+				dma_fence_set_error(&fence->base, -ETIME);
+			else if (fence->context == af->context)
+				dma_fence_set_error(&fence->base, -ECANCELED);
+		}
+		rcu_read_unlock();
+	} while (last_seq != seq);
+	spin_unlock_irqrestore(&ring->fence_drv.lock, flags);
+	/* signal the guilty fence */
+	amdgpu_fence_write(ring, (u32)af->base.seqno);
+	amdgpu_fence_process(ring);
+}
+
+void amdgpu_fence_save_wptr(struct amdgpu_fence *af)
+{
+	af->wptr = af->ring->wptr;
+}
+
+static void amdgpu_ring_backup_unprocessed_command(struct amdgpu_ring *ring,
+						   u64 start_wptr, u32 end_wptr)
+{
+	unsigned int first_idx = start_wptr & ring->buf_mask;
+	unsigned int last_idx = end_wptr & ring->buf_mask;
+	unsigned int i;
+
+	/* Backup the contents of the ring buffer. */
+	for (i = first_idx; i != last_idx; ++i, i &= ring->buf_mask)
+		ring->ring_backup[ring->ring_backup_entries_to_copy++] = ring->ring[i];
+}
+
+void amdgpu_ring_backup_unprocessed_commands(struct amdgpu_ring *ring,
+					     struct amdgpu_fence *guilty_fence)
+{
+	struct dma_fence *unprocessed;
+	struct dma_fence __rcu **ptr;
+	struct amdgpu_fence *fence;
+	u64 wptr;
+	u32 seq, last_seq;
+
+	last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask;
+	seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask;
+	wptr = ring->fence_drv.signalled_wptr;
+	ring->ring_backup_entries_to_copy = 0;
+
+	do {
+		last_seq++;
+		last_seq &= ring->fence_drv.num_fences_mask;
+
+		ptr = &ring->fence_drv.fences[last_seq];
+		rcu_read_lock();
+		unprocessed = rcu_dereference(*ptr);
+
+		if (unprocessed && !dma_fence_is_signaled(unprocessed)) {
+			fence = container_of(unprocessed, struct amdgpu_fence, base);
+
+			/* save everything if the ring is not guilty, otherwise
+			 * just save the content from other contexts.
+			 */
+			if (!guilty_fence || (fence->context != guilty_fence->context))
+				amdgpu_ring_backup_unprocessed_command(ring, wptr,
+								       fence->wptr);
+			wptr = fence->wptr;
+		}
+		rcu_read_unlock();
+	} while (last_seq != seq);
 }
 
 /*
@@ -552,13 +804,12 @@ static const char *amdgpu_fence_get_driver_name(struct dma_fence *fence)
 
 static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
 {
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	return (const char *)fence->ring->name;
+	return (const char *)to_amdgpu_fence(f)->ring->name;
 }
 
 /**
  * amdgpu_fence_enable_signaling - enable signalling on fence
- * @fence: fence
+ * @f: fence
  *
  * This function is called with fence_queue lock held, and adds a callback
  * to fence_queue that checks if this fence is signaled, and if so it
@@ -566,13 +817,8 @@ static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
  */
 static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
 {
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	struct amdgpu_ring *ring = fence->ring;
-
-	if (!timer_pending(&ring->fence_drv.fallback_timer))
-		amdgpu_fence_schedule_fallback(ring);
-
-	DMA_FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx);
+	if (!timer_pending(&to_amdgpu_fence(f)->ring->fence_drv.fallback_timer))
+		amdgpu_fence_schedule_fallback(to_amdgpu_fence(f)->ring);
 
 	return true;
 }
@@ -587,14 +833,15 @@ static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
 static void amdgpu_fence_free(struct rcu_head *rcu)
 {
 	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);
-	struct amdgpu_fence *fence = to_amdgpu_fence(f);
-	kmem_cache_free(amdgpu_fence_slab, fence);
+
+	/* free fence_slab if it's separated fence*/
+	kfree(to_amdgpu_fence(f));
 }
 
 /**
  * amdgpu_fence_release - callback that fence can be freed
  *
- * @fence: fence
+ * @f: fence
  *
  * This function is called when the reference count becomes zero.
  * It just RCU schedules freeing up the fence.
@@ -608,7 +855,6 @@ static const struct dma_fence_ops amdgpu_fence_ops = {
 	.get_driver_name = amdgpu_fence_get_driver_name,
 	.get_timeline_name = amdgpu_fence_get_timeline_name,
 	.enable_signaling = amdgpu_fence_enable_signaling,
-	.wait = dma_fence_default_wait,
 	.release = amdgpu_fence_release,
 };
 
@@ -616,58 +862,115 @@ static const struct dma_fence_ops amdgpu_fence_ops = {
  * Fence debugfs
  */
 #if defined(CONFIG_DEBUG_FS)
-static int amdgpu_debugfs_fence_info(struct seq_file *m, void *data)
+static int amdgpu_debugfs_fence_info_show(struct seq_file *m, void *unused)
 {
-	struct drm_info_node *node = (struct drm_info_node *)m->private;
-	struct drm_device *dev = node->minor->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = m->private;
 	int i;
 
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
 		struct amdgpu_ring *ring = adev->rings[i];
+
 		if (!ring || !ring->fence_drv.initialized)
 			continue;
 
 		amdgpu_fence_process(ring);
 
 		seq_printf(m, "--- ring %d (%s) ---\n", i, ring->name);
-		seq_printf(m, "Last signaled fence 0x%08x\n",
+		seq_printf(m, "Last signaled fence          0x%08x\n",
 			   atomic_read(&ring->fence_drv.last_seq));
-		seq_printf(m, "Last emitted        0x%08x\n",
+		seq_printf(m, "Last emitted                 0x%08x\n",
 			   ring->fence_drv.sync_seq);
+
+		if (ring->funcs->type == AMDGPU_RING_TYPE_GFX ||
+		    ring->funcs->type == AMDGPU_RING_TYPE_SDMA) {
+			seq_printf(m, "Last signaled trailing fence 0x%08x\n",
+				   le32_to_cpu(*ring->trail_fence_cpu_addr));
+			seq_printf(m, "Last emitted                 0x%08x\n",
+				   ring->trail_seq);
+		}
+
+		if (ring->funcs->type != AMDGPU_RING_TYPE_GFX)
+			continue;
+
+		/* set in CP_VMID_PREEMPT and preemption occurred */
+		seq_printf(m, "Last preempted               0x%08x\n",
+			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 2)));
+		/* set in CP_VMID_RESET and reset occurred */
+		seq_printf(m, "Last reset                   0x%08x\n",
+			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 4)));
+		/* Both preemption and reset occurred */
+		seq_printf(m, "Last both                    0x%08x\n",
+			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 6)));
 	}
 	return 0;
 }
 
-/**
- * amdgpu_debugfs_gpu_reset - manually trigger a gpu reset
+/*
+ * amdgpu_debugfs_gpu_recover - manually trigger a gpu reset & recover
  *
  * Manually trigger a gpu reset at the next fence wait.
  */
-static int amdgpu_debugfs_gpu_reset(struct seq_file *m, void *data)
+static int gpu_recover_get(void *data, u64 *val)
 {
-	struct drm_info_node *node = (struct drm_info_node *) m->private;
-	struct drm_device *dev = node->minor->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	struct drm_device *dev = adev_to_drm(adev);
+	int r;
+
+	r = pm_runtime_get_sync(dev->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(dev->dev);
+		return 0;
+	}
+
+	if (amdgpu_reset_domain_schedule(adev->reset_domain, &adev->reset_work))
+		flush_work(&adev->reset_work);
 
-	seq_printf(m, "gpu reset\n");
-	amdgpu_gpu_reset(adev);
+	*val = atomic_read(&adev->reset_domain->reset_res);
+
+	pm_runtime_put_autosuspend(dev->dev);
 
 	return 0;
 }
 
-static const struct drm_info_list amdgpu_debugfs_fence_list[] = {
-	{"amdgpu_fence_info", &amdgpu_debugfs_fence_info, 0, NULL},
-	{"amdgpu_gpu_reset", &amdgpu_debugfs_gpu_reset, 0, NULL}
-};
+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_fence_info);
+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_gpu_recover_fops, gpu_recover_get, NULL,
+			 "%lld\n");
+
+static void amdgpu_debugfs_reset_work(struct work_struct *work)
+{
+	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
+						  reset_work);
+
+	struct amdgpu_reset_context reset_context;
+
+	memset(&reset_context, 0, sizeof(reset_context));
+
+	reset_context.method = AMD_RESET_METHOD_NONE;
+	reset_context.reset_req_dev = adev;
+	reset_context.src = AMDGPU_RESET_SRC_USER;
+	set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
+	set_bit(AMDGPU_SKIP_COREDUMP, &reset_context.flags);
+
+	amdgpu_device_gpu_recover(adev, NULL, &reset_context);
+}
+
 #endif
 
-int amdgpu_debugfs_fence_init(struct amdgpu_device *adev)
+void amdgpu_debugfs_fence_init(struct amdgpu_device *adev)
 {
 #if defined(CONFIG_DEBUG_FS)
-	return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_fence_list, 2);
-#else
-	return 0;
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+
+	debugfs_create_file("amdgpu_fence_info", 0444, root, adev,
+			    &amdgpu_debugfs_fence_info_fops);
+
+	if (!amdgpu_sriov_vf(adev)) {
+
+		INIT_WORK(&adev->reset_work, amdgpu_debugfs_reset_work);
+		debugfs_create_file("amdgpu_gpu_recover", 0444, root, adev,
+				    &amdgpu_debugfs_gpu_recover_fops);
+	}
 #endif
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
new file mode 100644
index 000000000000..b0082aa7f3c6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
@@ -0,0 +1,408 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include <linux/pci.h>
+
+#include "amdgpu.h"
+#include "amdgpu_i2c.h"
+#include "smu_v11_0_i2c.h"
+#include "atom.h"
+#include "amdgpu_fru_eeprom.h"
+#include "amdgpu_eeprom.h"
+
+#define FRU_EEPROM_MADDR_6      0x60000
+#define FRU_EEPROM_MADDR_8      0x80000
+#define FRU_EEPROM_MADDR_INV    0xFFFFF
+
+static bool is_fru_eeprom_supported(struct amdgpu_device *adev, u32 *fru_addr)
+{
+	/* Only server cards have the FRU EEPROM
+	 * TODO: See if we can figure this out dynamically instead of
+	 * having to parse VBIOS versions.
+	 */
+	struct atom_context *atom_ctx = adev->mode_info.atom_context;
+
+	/* The i2c access is blocked on VF
+	 * TODO: Need other way to get the info
+	 * Also, FRU not valid for APU devices.
+	 */
+	if (amdgpu_sriov_vf(adev) || (adev->flags & AMD_IS_APU))
+		return false;
+
+	/* The default I2C EEPROM address of the FRU.
+	 */
+	if (fru_addr)
+		*fru_addr = FRU_EEPROM_MADDR_8;
+
+	/* VBIOS is of the format ###-DXXXYYYY-##. For SKU identification,
+	 * we can use just the "DXXX" portion. If there were more models, we
+	 * could convert the 3 characters to a hex integer and use a switch
+	 * for ease/speed/readability. For now, 2 string comparisons are
+	 * reasonable and not too expensive
+	 */
+	switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
+	case IP_VERSION(11, 0, 2):
+		switch (adev->asic_type) {
+		case CHIP_VEGA20:
+			/* D161 and D163 are the VG20 server SKUs */
+			if (atom_ctx && (strnstr(atom_ctx->vbios_pn, "D161",
+						 sizeof(atom_ctx->vbios_pn)) ||
+					 strnstr(atom_ctx->vbios_pn, "D163",
+						 sizeof(atom_ctx->vbios_pn)))) {
+				if (fru_addr)
+					*fru_addr = FRU_EEPROM_MADDR_6;
+				return true;
+			} else {
+				return false;
+			}
+		case CHIP_ARCTURUS:
+		default:
+			return false;
+		}
+	case IP_VERSION(11, 0, 7):
+		if (atom_ctx && strnstr(atom_ctx->vbios_pn, "D603",
+					sizeof(atom_ctx->vbios_pn))) {
+			if (strnstr(atom_ctx->vbios_pn, "D603GLXE",
+				    sizeof(atom_ctx->vbios_pn))) {
+				return false;
+			}
+
+			if (fru_addr)
+				*fru_addr = FRU_EEPROM_MADDR_6;
+			return true;
+
+		} else {
+			return false;
+		}
+	case IP_VERSION(13, 0, 2):
+		/* All Aldebaran SKUs have an FRU */
+		if (atom_ctx && !strnstr(atom_ctx->vbios_pn, "D673",
+					 sizeof(atom_ctx->vbios_pn)))
+			if (fru_addr)
+				*fru_addr = FRU_EEPROM_MADDR_6;
+		return true;
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 14):
+			if (fru_addr)
+				*fru_addr = FRU_EEPROM_MADDR_8;
+			return true;
+	case IP_VERSION(13, 0, 12):
+			if (fru_addr)
+				*fru_addr = FRU_EEPROM_MADDR_INV;
+			return true;
+	default:
+		return false;
+	}
+}
+
+int amdgpu_fru_get_product_info(struct amdgpu_device *adev)
+{
+	struct amdgpu_fru_info *fru_info;
+	unsigned char buf[8], *pia;
+	u32 addr, fru_addr;
+	int size, len;
+	u8 csum;
+
+	if (!is_fru_eeprom_supported(adev, &fru_addr))
+		return 0;
+
+	/* FRU data avaialble, but no direct EEPROM access */
+	if (fru_addr == FRU_EEPROM_MADDR_INV)
+		return 0;
+
+	if (!adev->fru_info) {
+		adev->fru_info = kzalloc(sizeof(*adev->fru_info), GFP_KERNEL);
+		if (!adev->fru_info)
+			return -ENOMEM;
+	}
+
+	fru_info = adev->fru_info;
+	/* For Arcturus-and-later, default value of serial_number is unique_id
+	 * so convert it to a 16-digit HEX string for convenience and
+	 * backwards-compatibility.
+	 */
+	sprintf(fru_info->serial, "%llx", adev->unique_id);
+
+	/* If algo exists, it means that the i2c_adapter's initialized */
+	if (!adev->pm.fru_eeprom_i2c_bus || !adev->pm.fru_eeprom_i2c_bus->algo) {
+		dev_warn(adev->dev,
+			 "Cannot access FRU, EEPROM accessor not initialized");
+		return -ENODEV;
+	}
+
+	/* Read the IPMI Common header */
+	len = amdgpu_eeprom_read(adev->pm.fru_eeprom_i2c_bus, fru_addr, buf,
+				 sizeof(buf));
+	if (len != 8) {
+		dev_err(adev->dev, "Couldn't read the IPMI Common Header: %d",
+			len);
+		return len < 0 ? len : -EIO;
+	}
+
+	if (buf[0] != 1) {
+		dev_err(adev->dev, "Bad IPMI Common Header version: 0x%02x",
+			buf[0]);
+		return -EIO;
+	}
+
+	for (csum = 0; len > 0; len--)
+		csum += buf[len - 1];
+	if (csum) {
+		dev_err(adev->dev, "Bad IPMI Common Header checksum: 0x%02x",
+			csum);
+		return -EIO;
+	}
+
+	/* Get the offset to the Product Info Area (PIA). */
+	addr = buf[4] * 8;
+	if (!addr)
+		return 0;
+
+	/* Get the absolute address to the PIA. */
+	addr += fru_addr;
+
+	/* Read the header of the PIA. */
+	len = amdgpu_eeprom_read(adev->pm.fru_eeprom_i2c_bus, addr, buf, 3);
+	if (len != 3) {
+		dev_err(adev->dev,
+			"Couldn't read the Product Info Area header: %d", len);
+		return len < 0 ? len : -EIO;
+	}
+
+	if (buf[0] != 1) {
+		dev_err(adev->dev, "Bad IPMI Product Info Area version: 0x%02x",
+			buf[0]);
+		return -EIO;
+	}
+
+	size = buf[1] * 8;
+	pia = kzalloc(size, GFP_KERNEL);
+	if (!pia)
+		return -ENOMEM;
+
+	/* Read the whole PIA. */
+	len = amdgpu_eeprom_read(adev->pm.fru_eeprom_i2c_bus, addr, pia, size);
+	if (len != size) {
+		kfree(pia);
+		dev_err(adev->dev, "Couldn't read the Product Info Area: %d",
+			len);
+		return len < 0 ? len : -EIO;
+	}
+
+	for (csum = 0; size > 0; size--)
+		csum += pia[size - 1];
+	if (csum) {
+		dev_err(adev->dev, "Bad Product Info Area checksum: 0x%02x",
+			csum);
+		kfree(pia);
+		return -EIO;
+	}
+
+	/* Now extract useful information from the PIA.
+	 *
+	 * Read Manufacturer Name field whose length is [3].
+	 */
+	addr = 3;
+	if (addr + 1 >= len)
+		goto Out;
+	memcpy(fru_info->manufacturer_name, pia + addr + 1,
+	       min_t(size_t, sizeof(fru_info->manufacturer_name),
+		     pia[addr] & 0x3F));
+	fru_info->manufacturer_name[sizeof(fru_info->manufacturer_name) - 1] =
+		'\0';
+
+	/* Read Product Name field. */
+	addr += 1 + (pia[addr] & 0x3F);
+	if (addr + 1 >= len)
+		goto Out;
+	memcpy(fru_info->product_name, pia + addr + 1,
+	       min_t(size_t, sizeof(fru_info->product_name), pia[addr] & 0x3F));
+	fru_info->product_name[sizeof(fru_info->product_name) - 1] = '\0';
+
+	/* Go to the Product Part/Model Number field. */
+	addr += 1 + (pia[addr] & 0x3F);
+	if (addr + 1 >= len)
+		goto Out;
+	memcpy(fru_info->product_number, pia + addr + 1,
+	       min_t(size_t, sizeof(fru_info->product_number),
+		     pia[addr] & 0x3F));
+	fru_info->product_number[sizeof(fru_info->product_number) - 1] = '\0';
+
+	/* Go to the Product Version field. */
+	addr += 1 + (pia[addr] & 0x3F);
+
+	/* Go to the Product Serial Number field. */
+	addr += 1 + (pia[addr] & 0x3F);
+	if (addr + 1 >= len)
+		goto Out;
+	memcpy(fru_info->serial, pia + addr + 1,
+	       min_t(size_t, sizeof(fru_info->serial), pia[addr] & 0x3F));
+	fru_info->serial[sizeof(fru_info->serial) - 1] = '\0';
+
+	/* Asset Tag field */
+	addr += 1 + (pia[addr] & 0x3F);
+
+	/* FRU File Id field. This could be 'null'. */
+	addr += 1 + (pia[addr] & 0x3F);
+	if ((addr + 1 >= len) || !(pia[addr] & 0x3F))
+		goto Out;
+	memcpy(fru_info->fru_id, pia + addr + 1,
+	       min_t(size_t, sizeof(fru_info->fru_id), pia[addr] & 0x3F));
+	fru_info->fru_id[sizeof(fru_info->fru_id) - 1] = '\0';
+
+Out:
+	kfree(pia);
+	return 0;
+}
+
+/**
+ * DOC: product_name
+ *
+ * The amdgpu driver provides a sysfs API for reporting the product name
+ * for the device
+ * The file product_name is used for this and returns the product name
+ * as returned from the FRU.
+ * NOTE: This is only available for certain server cards
+ */
+
+static ssize_t amdgpu_fru_product_name_show(struct device *dev,
+					    struct device_attribute *attr,
+					    char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	return sysfs_emit(buf, "%s\n", adev->fru_info->product_name);
+}
+
+static DEVICE_ATTR(product_name, 0444, amdgpu_fru_product_name_show, NULL);
+
+/**
+ * DOC: product_number
+ *
+ * The amdgpu driver provides a sysfs API for reporting the part number
+ * for the device
+ * The file product_number is used for this and returns the part number
+ * as returned from the FRU.
+ * NOTE: This is only available for certain server cards
+ */
+
+static ssize_t amdgpu_fru_product_number_show(struct device *dev,
+					      struct device_attribute *attr,
+					      char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	return sysfs_emit(buf, "%s\n", adev->fru_info->product_number);
+}
+
+static DEVICE_ATTR(product_number, 0444, amdgpu_fru_product_number_show, NULL);
+
+/**
+ * DOC: serial_number
+ *
+ * The amdgpu driver provides a sysfs API for reporting the serial number
+ * for the device
+ * The file serial_number is used for this and returns the serial number
+ * as returned from the FRU.
+ * NOTE: This is only available for certain server cards
+ */
+
+static ssize_t amdgpu_fru_serial_number_show(struct device *dev,
+					     struct device_attribute *attr,
+					     char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	return sysfs_emit(buf, "%s\n", adev->fru_info->serial);
+}
+
+static DEVICE_ATTR(serial_number, 0444, amdgpu_fru_serial_number_show, NULL);
+
+/**
+ * DOC: fru_id
+ *
+ * The amdgpu driver provides a sysfs API for reporting FRU File Id
+ * for the device.
+ * The file fru_id is used for this and returns the File Id value
+ * as returned from the FRU.
+ * NOTE: This is only available for certain server cards
+ */
+
+static ssize_t amdgpu_fru_id_show(struct device *dev,
+				  struct device_attribute *attr, char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	return sysfs_emit(buf, "%s\n", adev->fru_info->fru_id);
+}
+
+static DEVICE_ATTR(fru_id, 0444, amdgpu_fru_id_show, NULL);
+
+/**
+ * DOC: manufacturer
+ *
+ * The amdgpu driver provides a sysfs API for reporting manufacturer name from
+ * FRU information.
+ * The file manufacturer returns the value as returned from the FRU.
+ * NOTE: This is only available for certain server cards
+ */
+
+static ssize_t amdgpu_fru_manufacturer_name_show(struct device *dev,
+						 struct device_attribute *attr,
+						 char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	return sysfs_emit(buf, "%s\n", adev->fru_info->manufacturer_name);
+}
+
+static DEVICE_ATTR(manufacturer, 0444, amdgpu_fru_manufacturer_name_show, NULL);
+
+static const struct attribute *amdgpu_fru_attributes[] = {
+	&dev_attr_product_name.attr,
+	&dev_attr_product_number.attr,
+	&dev_attr_serial_number.attr,
+	&dev_attr_fru_id.attr,
+	&dev_attr_manufacturer.attr,
+	NULL
+};
+
+int amdgpu_fru_sysfs_init(struct amdgpu_device *adev)
+{
+	if (!is_fru_eeprom_supported(adev, NULL) || !adev->fru_info)
+		return 0;
+
+	return sysfs_create_files(&adev->dev->kobj, amdgpu_fru_attributes);
+}
+
+void amdgpu_fru_sysfs_fini(struct amdgpu_device *adev)
+{
+	if (!adev->fru_info)
+		return;
+
+	sysfs_remove_files(&adev->dev->kobj, amdgpu_fru_attributes);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.h
new file mode 100644
index 000000000000..98f3196599ef
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.h
@@ -0,0 +1,42 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_FRU_EEPROM_H__
+#define __AMDGPU_FRU_EEPROM_H__
+
+#define AMDGPU_PRODUCT_NAME_LEN 64
+
+/* FRU product information */
+struct amdgpu_fru_info {
+	char				product_number[20];
+	char				product_name[AMDGPU_PRODUCT_NAME_LEN];
+	char				serial[20];
+	char				manufacturer_name[32];
+	char				fru_id[50];
+};
+
+int amdgpu_fru_get_product_info(struct amdgpu_device *adev);
+int amdgpu_fru_sysfs_init(struct amdgpu_device *adev);
+void amdgpu_fru_sysfs_fini(struct amdgpu_device *adev);
+
+#endif  // __AMDGPU_FRU_EEPROM_H__
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fw_attestation.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fw_attestation.c
new file mode 100644
index 000000000000..328a1b963548
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fw_attestation.c
@@ -0,0 +1,145 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/debugfs.h>
+#include <linux/firmware.h>
+#include <linux/dma-mapping.h>
+
+#include "amdgpu.h"
+#include "amdgpu_fw_attestation.h"
+#include "amdgpu_psp.h"
+#include "amdgpu_ucode.h"
+#include "soc15_common.h"
+
+#define FW_ATTESTATION_DB_COOKIE        0x143b6a37
+#define FW_ATTESTATION_RECORD_VALID	1
+#define FW_ATTESTATION_MAX_SIZE		4096
+
+struct FW_ATT_DB_HEADER {
+	uint32_t AttDbVersion;           /* version of the fwar feature */
+	uint32_t AttDbCookie;            /* cookie as an extra check for corrupt data */
+};
+
+struct FW_ATT_RECORD {
+	uint16_t AttFwIdV1;              /* Legacy FW Type field */
+	uint16_t AttFwIdV2;              /* V2 FW ID field */
+	uint32_t AttFWVersion;           /* FW Version */
+	uint16_t AttFWActiveFunctionID;  /* The VF ID (only in VF Attestation Table) */
+	uint8_t  AttSource;              /* FW source indicator */
+	uint8_t  RecordValid;            /* Indicates whether the record is a valid entry */
+	uint32_t AttFwTaId;              /* Ta ID (only in TA Attestation Table) */
+};
+
+static ssize_t amdgpu_fw_attestation_debugfs_read(struct file *f,
+						  char __user *buf,
+						  size_t size,
+						  loff_t *pos)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
+	uint64_t records_addr = 0;
+	uint64_t vram_pos = 0;
+	struct FW_ATT_DB_HEADER fw_att_hdr = {0};
+	struct FW_ATT_RECORD fw_att_record = {0};
+
+	if (size < sizeof(struct FW_ATT_RECORD)) {
+		DRM_WARN("FW attestation input buffer not enough memory");
+		return -EINVAL;
+	}
+
+	if ((*pos + sizeof(struct FW_ATT_DB_HEADER)) >= FW_ATTESTATION_MAX_SIZE) {
+		DRM_WARN("FW attestation out of bounds");
+		return 0;
+	}
+
+	if (psp_get_fw_attestation_records_addr(&adev->psp, &records_addr)) {
+		DRM_WARN("Failed to get FW attestation record address");
+		return -EINVAL;
+	}
+
+	vram_pos =  records_addr - adev->gmc.vram_start;
+
+	if (*pos == 0) {
+		amdgpu_device_vram_access(adev,
+					  vram_pos,
+					  (uint32_t *)&fw_att_hdr,
+					  sizeof(struct FW_ATT_DB_HEADER),
+					  false);
+
+		if (fw_att_hdr.AttDbCookie != FW_ATTESTATION_DB_COOKIE) {
+			DRM_WARN("Invalid FW attestation cookie");
+			return -EINVAL;
+		}
+
+		DRM_INFO("FW attestation version = 0x%X", fw_att_hdr.AttDbVersion);
+	}
+
+	amdgpu_device_vram_access(adev,
+				  vram_pos + sizeof(struct FW_ATT_DB_HEADER) + *pos,
+				  (uint32_t *)&fw_att_record,
+				  sizeof(struct FW_ATT_RECORD),
+				  false);
+
+	if (fw_att_record.RecordValid != FW_ATTESTATION_RECORD_VALID)
+		return 0;
+
+	if (copy_to_user(buf, (void *)&fw_att_record, sizeof(struct FW_ATT_RECORD)))
+		return -EINVAL;
+
+	*pos += sizeof(struct FW_ATT_RECORD);
+
+	return sizeof(struct FW_ATT_RECORD);
+}
+
+static const struct file_operations amdgpu_fw_attestation_debugfs_ops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_fw_attestation_debugfs_read,
+	.write = NULL,
+	.llseek = default_llseek
+};
+
+static int amdgpu_is_fw_attestation_supported(struct amdgpu_device *adev)
+{
+	if (adev->flags & AMD_IS_APU)
+		return 0;
+
+	if (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(14, 0, 2) ||
+	    amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(14, 0, 3))
+		return 0;
+
+	if (adev->asic_type >= CHIP_SIENNA_CICHLID)
+		return 1;
+
+	return 0;
+}
+
+void amdgpu_fw_attestation_debugfs_init(struct amdgpu_device *adev)
+{
+	if (!amdgpu_is_fw_attestation_supported(adev))
+		return;
+
+	debugfs_create_file("amdgpu_fw_attestation",
+			    0400,
+			    adev_to_drm(adev)->primary->debugfs_root,
+			    adev,
+			    &amdgpu_fw_attestation_debugfs_ops);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fw_attestation.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_fw_attestation.h
new file mode 100644
index 000000000000..90af4fe58c99
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fw_attestation.h
@@ -0,0 +1,30 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ *
+ */
+#ifndef _AMDGPU_FW_ATTESTATION_H
+#define _AMDGPU_FW_ATTESTATION_H
+
+#include "amdgpu.h"
+
+void amdgpu_fw_attestation_debugfs_init(struct amdgpu_device *adev);
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index 902e6015abca..d2237ce9da70 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -25,12 +25,18 @@
  *          Alex Deucher
  *          Jerome Glisse
  */
-#include <drm/drmP.h>
+
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
 #include <drm/amdgpu_drm.h>
 #ifdef CONFIG_X86
 #include <asm/set_memory.h>
 #endif
 #include "amdgpu.h"
+#include "amdgpu_reset.h"
+#include <drm/drm_drv.h>
+#include <drm/ttm/ttm_tt.h>
 
 /*
  * GART
@@ -55,144 +61,213 @@
 /*
  * Common GART table functions.
  */
+
 /**
- * amdgpu_gart_table_ram_alloc - allocate system ram for gart page table
+ * amdgpu_gart_dummy_page_init - init dummy page used by the driver
  *
  * @adev: amdgpu_device pointer
  *
- * Allocate system memory for GART page table
- * (r1xx-r3xx, non-pcie r4xx, rs400).  These asics require the
- * gart table to be in system memory.
- * Returns 0 for success, -ENOMEM for failure.
+ * Allocate the dummy page used by the driver (all asics).
+ * This dummy page is used by the driver as a filler for gart entries
+ * when pages are taken out of the GART
+ * Returns 0 on sucess, -ENOMEM on failure.
  */
-int amdgpu_gart_table_ram_alloc(struct amdgpu_device *adev)
+static int amdgpu_gart_dummy_page_init(struct amdgpu_device *adev)
 {
-	void *ptr;
+	struct page *dummy_page = ttm_glob.dummy_read_page;
 
-	ptr = pci_alloc_consistent(adev->pdev, adev->gart.table_size,
-				   &adev->gart.table_addr);
-	if (ptr == NULL) {
+	if (adev->dummy_page_addr)
+		return 0;
+	adev->dummy_page_addr = dma_map_page_attrs(&adev->pdev->dev, dummy_page, 0,
+							PAGE_SIZE, DMA_BIDIRECTIONAL,
+							DMA_ATTR_SKIP_CPU_SYNC);
+	if (dma_mapping_error(&adev->pdev->dev, adev->dummy_page_addr)) {
+		dev_err(&adev->pdev->dev, "Failed to DMA MAP the dummy page\n");
+		adev->dummy_page_addr = 0;
 		return -ENOMEM;
 	}
-#ifdef CONFIG_X86
-	if (0) {
-		set_memory_uc((unsigned long)ptr,
-			      adev->gart.table_size >> PAGE_SHIFT);
-	}
-#endif
-	adev->gart.ptr = ptr;
-	memset((void *)adev->gart.ptr, 0, adev->gart.table_size);
 	return 0;
 }
 
 /**
- * amdgpu_gart_table_ram_free - free system ram for gart page table
+ * amdgpu_gart_dummy_page_fini - free dummy page used by the driver
  *
  * @adev: amdgpu_device pointer
  *
- * Free system memory for GART page table
- * (r1xx-r3xx, non-pcie r4xx, rs400).  These asics require the
- * gart table to be in system memory.
+ * Frees the dummy page used by the driver (all asics).
  */
-void amdgpu_gart_table_ram_free(struct amdgpu_device *adev)
+void amdgpu_gart_dummy_page_fini(struct amdgpu_device *adev)
 {
-	if (adev->gart.ptr == NULL) {
+	if (!adev->dummy_page_addr)
 		return;
-	}
-#ifdef CONFIG_X86
-	if (0) {
-		set_memory_wb((unsigned long)adev->gart.ptr,
-			      adev->gart.table_size >> PAGE_SHIFT);
-	}
-#endif
-	pci_free_consistent(adev->pdev, adev->gart.table_size,
-			    (void *)adev->gart.ptr,
-			    adev->gart.table_addr);
-	adev->gart.ptr = NULL;
-	adev->gart.table_addr = 0;
+	dma_unmap_page_attrs(&adev->pdev->dev, adev->dummy_page_addr, PAGE_SIZE,
+				DMA_BIDIRECTIONAL,
+				DMA_ATTR_SKIP_CPU_SYNC);
+	adev->dummy_page_addr = 0;
 }
 
 /**
- * amdgpu_gart_table_vram_alloc - allocate vram for gart page table
+ * amdgpu_gart_table_ram_alloc - allocate system ram for gart page table
  *
  * @adev: amdgpu_device pointer
  *
- * Allocate video memory for GART page table
- * (pcie r4xx, r5xx+).  These asics require the
- * gart table to be in video memory.
+ * Allocate system memory for GART page table for ASICs that don't have
+ * dedicated VRAM.
  * Returns 0 for success, error for failure.
  */
-int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
+int amdgpu_gart_table_ram_alloc(struct amdgpu_device *adev)
 {
-	int r;
+	unsigned int order = get_order(adev->gart.table_size);
+	gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO;
+	struct amdgpu_bo *bo = NULL;
+	struct sg_table *sg = NULL;
+	struct amdgpu_bo_param bp;
+	dma_addr_t dma_addr;
+	struct page *p;
+	unsigned long x;
+	int ret;
 
-	if (adev->gart.robj == NULL) {
-		r = amdgpu_bo_create(adev, adev->gart.table_size,
-				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
-				     NULL, NULL, &adev->gart.robj);
-		if (r) {
-			return r;
-		}
+	if (adev->gart.bo != NULL)
+		return 0;
+
+	p = alloc_pages(gfp_flags, order);
+	if (!p)
+		return -ENOMEM;
+
+	/* assign pages to this device */
+	for (x = 0; x < (1UL << order); x++)
+		p[x].mapping = adev->mman.bdev.dev_mapping;
+
+	/* If the hardware does not support UTCL2 snooping of the CPU caches
+	 * then set_memory_wc() could be used as a workaround to mark the pages
+	 * as write combine memory.
+	 */
+	dma_addr = dma_map_page(&adev->pdev->dev, p, 0, adev->gart.table_size,
+				DMA_BIDIRECTIONAL);
+	if (dma_mapping_error(&adev->pdev->dev, dma_addr)) {
+		dev_err(&adev->pdev->dev, "Failed to DMA MAP the GART BO page\n");
+		__free_pages(p, order);
+		p = NULL;
+		return -EFAULT;
+	}
+
+	dev_info(adev->dev, "%s dma_addr:%pad\n", __func__, &dma_addr);
+	/* Create SG table */
+	sg = kmalloc(sizeof(*sg), GFP_KERNEL);
+	if (!sg) {
+		ret = -ENOMEM;
+		goto error;
+	}
+	ret = sg_alloc_table(sg, 1, GFP_KERNEL);
+	if (ret)
+		goto error;
+
+	sg_dma_address(sg->sgl) = dma_addr;
+	sg->sgl->length = adev->gart.table_size;
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+	sg->sgl->dma_length = adev->gart.table_size;
+#endif
+	/* Create SG BO */
+	memset(&bp, 0, sizeof(bp));
+	bp.size = adev->gart.table_size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_CPU;
+	bp.type = ttm_bo_type_sg;
+	bp.resv = NULL;
+	bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+	bp.flags = 0;
+	ret = amdgpu_bo_create(adev, &bp, &bo);
+	if (ret)
+		goto error;
+
+	bo->tbo.sg = sg;
+	bo->tbo.ttm->sg = sg;
+	bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
+	bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
+
+	ret = amdgpu_bo_reserve(bo, true);
+	if (ret) {
+		dev_err(adev->dev, "(%d) failed to reserve bo for GART system bo\n", ret);
+		goto error;
 	}
+
+	ret = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT);
+	WARN(ret, "Pinning the GART table failed");
+	if (ret)
+		goto error_resv;
+
+	adev->gart.bo = bo;
+	adev->gart.ptr = page_to_virt(p);
+	/* Make GART table accessible in VMID0 */
+	ret = amdgpu_ttm_alloc_gart(&adev->gart.bo->tbo);
+	if (ret)
+		amdgpu_gart_table_ram_free(adev);
+	amdgpu_bo_unreserve(bo);
+
 	return 0;
+
+error_resv:
+	amdgpu_bo_unreserve(bo);
+error:
+	amdgpu_bo_unref(&bo);
+	if (sg) {
+		sg_free_table(sg);
+		kfree(sg);
+	}
+	__free_pages(p, order);
+	return ret;
 }
 
 /**
- * amdgpu_gart_table_vram_pin - pin gart page table in vram
+ * amdgpu_gart_table_ram_free - free gart page table system ram
  *
  * @adev: amdgpu_device pointer
  *
- * Pin the GART page table in vram so it will not be moved
- * by the memory manager (pcie r4xx, r5xx+).  These asics require the
- * gart table to be in video memory.
- * Returns 0 for success, error for failure.
+ * Free the system memory used for the GART page tableon ASICs that don't
+ * have dedicated VRAM.
  */
-int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev)
+void amdgpu_gart_table_ram_free(struct amdgpu_device *adev)
 {
-	uint64_t gpu_addr;
-	int r;
+	unsigned int order = get_order(adev->gart.table_size);
+	struct sg_table *sg = adev->gart.bo->tbo.sg;
+	struct page *p;
+	unsigned long x;
+	int ret;
 
-	r = amdgpu_bo_reserve(adev->gart.robj, false);
-	if (unlikely(r != 0))
-		return r;
-	r = amdgpu_bo_pin(adev->gart.robj,
-				AMDGPU_GEM_DOMAIN_VRAM, &gpu_addr);
-	if (r) {
-		amdgpu_bo_unreserve(adev->gart.robj);
-		return r;
+	ret = amdgpu_bo_reserve(adev->gart.bo, false);
+	if (!ret) {
+		amdgpu_bo_unpin(adev->gart.bo);
+		amdgpu_bo_unreserve(adev->gart.bo);
 	}
-	r = amdgpu_bo_kmap(adev->gart.robj, &adev->gart.ptr);
-	if (r)
-		amdgpu_bo_unpin(adev->gart.robj);
-	amdgpu_bo_unreserve(adev->gart.robj);
-	adev->gart.table_addr = gpu_addr;
-	return r;
+	amdgpu_bo_unref(&adev->gart.bo);
+	sg_free_table(sg);
+	kfree(sg);
+	p = virt_to_page(adev->gart.ptr);
+	for (x = 0; x < (1UL << order); x++)
+		p[x].mapping = NULL;
+	__free_pages(p, order);
+
+	adev->gart.ptr = NULL;
 }
 
 /**
- * amdgpu_gart_table_vram_unpin - unpin gart page table in vram
+ * amdgpu_gart_table_vram_alloc - allocate vram for gart page table
  *
  * @adev: amdgpu_device pointer
  *
- * Unpin the GART page table in vram (pcie r4xx, r5xx+).
- * These asics require the gart table to be in video memory.
+ * Allocate video memory for GART page table
+ * (pcie r4xx, r5xx+).  These asics require the
+ * gart table to be in video memory.
+ * Returns 0 for success, error for failure.
  */
-void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev)
+int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
 {
-	int r;
+	if (adev->gart.bo != NULL)
+		return 0;
 
-	if (adev->gart.robj == NULL) {
-		return;
-	}
-	r = amdgpu_bo_reserve(adev->gart.robj, true);
-	if (likely(r == 0)) {
-		amdgpu_bo_kunmap(adev->gart.robj);
-		amdgpu_bo_unpin(adev->gart.robj);
-		amdgpu_bo_unreserve(adev->gart.robj);
-		adev->gart.ptr = NULL;
-	}
+	return amdgpu_bo_create_kernel(adev,  adev->gart.table_size, PAGE_SIZE,
+				       AMDGPU_GEM_DOMAIN_VRAM, &adev->gart.bo,
+				       NULL, (void *)&adev->gart.ptr);
 }
 
 /**
@@ -206,10 +281,7 @@ void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev)
  */
 void amdgpu_gart_table_vram_free(struct amdgpu_device *adev)
 {
-	if (adev->gart.robj == NULL) {
-		return;
-	}
-	amdgpu_bo_unref(&adev->gart.robj);
+	amdgpu_bo_free_kernel(&adev->gart.bo, NULL, (void *)&adev->gart.ptr);
 }
 
 /*
@@ -224,87 +296,158 @@ void amdgpu_gart_table_vram_free(struct amdgpu_device *adev)
  *
  * Unbinds the requested pages from the gart page table and
  * replaces them with the dummy page (all asics).
+ * Returns 0 for success, -EINVAL for failure.
  */
 void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset,
 			int pages)
 {
 	unsigned t;
-	unsigned p;
 	int i, j;
 	u64 page_base;
 	/* Starting from VEGA10, system bit must be 0 to mean invalid. */
 	uint64_t flags = 0;
+	int idx;
 
-	if (!adev->gart.ready) {
-		WARN(1, "trying to unbind memory from uninitialized GART !\n");
+	if (!adev->gart.ptr)
+		return;
+
+	if (!drm_dev_enter(adev_to_drm(adev), &idx))
 		return;
-	}
 
 	t = offset / AMDGPU_GPU_PAGE_SIZE;
-	p = t / (PAGE_SIZE / AMDGPU_GPU_PAGE_SIZE);
-	for (i = 0; i < pages; i++, p++) {
-#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
-		adev->gart.pages[p] = NULL;
-#endif
-		page_base = adev->dummy_page.addr;
+	for (i = 0; i < pages; i++) {
+		page_base = adev->dummy_page_addr;
 		if (!adev->gart.ptr)
 			continue;
 
-		for (j = 0; j < (PAGE_SIZE / AMDGPU_GPU_PAGE_SIZE); j++, t++) {
-			amdgpu_gart_set_pte_pde(adev, adev->gart.ptr,
-						t, page_base, flags);
+		for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) {
+			amdgpu_gmc_set_pte_pde(adev, adev->gart.ptr,
+					       t, page_base, flags);
 			page_base += AMDGPU_GPU_PAGE_SIZE;
 		}
 	}
-	mb();
-	amdgpu_gart_flush_gpu_tlb(adev, 0);
+	amdgpu_gart_invalidate_tlb(adev);
+
+	drm_dev_exit(idx);
 }
 
 /**
- * amdgpu_gart_bind - bind pages into the gart page table
+ * amdgpu_gart_map - map dma_addresses into GART entries
  *
  * @adev: amdgpu_device pointer
  * @offset: offset into the GPU's gart aperture
  * @pages: number of pages to bind
- * @pagelist: pages to bind
  * @dma_addr: DMA addresses of pages
+ * @flags: page table entry flags
+ * @dst: CPU address of the gart table
  *
- * Binds the requested pages to the gart page table
- * (all asics).
+ * Map the dma_addresses into GART entries (all asics).
  * Returns 0 for success, -EINVAL for failure.
  */
-int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
-		     int pages, struct page **pagelist, dma_addr_t *dma_addr,
-		     uint64_t flags)
+void amdgpu_gart_map(struct amdgpu_device *adev, uint64_t offset,
+		    int pages, dma_addr_t *dma_addr, uint64_t flags,
+		    void *dst)
 {
-	unsigned t;
-	unsigned p;
 	uint64_t page_base;
-	int i, j;
+	unsigned i, j, t;
+	int idx;
 
-	if (!adev->gart.ready) {
-		WARN(1, "trying to bind memory to uninitialized GART !\n");
-		return -EINVAL;
-	}
+	if (!drm_dev_enter(adev_to_drm(adev), &idx))
+		return;
 
 	t = offset / AMDGPU_GPU_PAGE_SIZE;
-	p = t / (PAGE_SIZE / AMDGPU_GPU_PAGE_SIZE);
 
-	for (i = 0; i < pages; i++, p++) {
-#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
-		adev->gart.pages[p] = pagelist[i];
-#endif
-		if (adev->gart.ptr) {
-			page_base = dma_addr[i];
-			for (j = 0; j < (PAGE_SIZE / AMDGPU_GPU_PAGE_SIZE); j++, t++) {
-				amdgpu_gart_set_pte_pde(adev, adev->gart.ptr, t, page_base, flags);
-				page_base += AMDGPU_GPU_PAGE_SIZE;
-			}
+	for (i = 0; i < pages; i++) {
+		page_base = dma_addr[i];
+		for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) {
+			amdgpu_gmc_set_pte_pde(adev, dst, t, page_base, flags);
+			page_base += AMDGPU_GPU_PAGE_SIZE;
 		}
 	}
+	drm_dev_exit(idx);
+}
+
+/**
+ * amdgpu_gart_map_vram_range - map VRAM pages into the GART page table
+ *
+ * @adev: amdgpu_device pointer
+ * @pa: physical address of the first page to be mapped
+ * @start_page: first page to map in the GART aperture
+ * @num_pages: number of pages to be mapped
+ * @flags: page table entry flags
+ * @dst: CPU address of the GART table
+ *
+ * Binds a BO that is allocated in VRAM to the GART page table
+ * (all ASICs).
+ *
+ * Useful when a kernel BO is located in VRAM but
+ * needs to be accessed from the GART address space.
+ */
+void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
+				uint64_t start_page, uint64_t num_pages,
+				uint64_t flags, void *dst)
+{
+	u32 i, idx;
+
+	/* The SYSTEM flag indicates the pages aren't in VRAM. */
+	WARN_ON_ONCE(flags & AMDGPU_PTE_SYSTEM);
+
+	if (!drm_dev_enter(adev_to_drm(adev), &idx))
+		return;
+
+	for (i = 0; i < num_pages; ++i) {
+		amdgpu_gmc_set_pte_pde(adev, adev->gart.ptr,
+			start_page + i, pa + AMDGPU_GPU_PAGE_SIZE * i, flags);
+	}
+
+	drm_dev_exit(idx);
+}
+
+/**
+ * amdgpu_gart_bind - bind pages into the gart page table
+ *
+ * @adev: amdgpu_device pointer
+ * @offset: offset into the GPU's gart aperture
+ * @pages: number of pages to bind
+ * @dma_addr: DMA addresses of pages
+ * @flags: page table entry flags
+ *
+ * Binds the requested pages to the gart page table
+ * (all asics).
+ * Returns 0 for success, -EINVAL for failure.
+ */
+void amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
+		     int pages, dma_addr_t *dma_addr,
+		     uint64_t flags)
+{
+	if (!adev->gart.ptr)
+		return;
+
+	amdgpu_gart_map(adev, offset, pages, dma_addr, flags, adev->gart.ptr);
+}
+
+/**
+ * amdgpu_gart_invalidate_tlb - invalidate gart TLB
+ *
+ * @adev: amdgpu device driver pointer
+ *
+ * Invalidate gart TLB which can be use as a way to flush gart changes
+ *
+ */
+void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev)
+{
+	int i;
+
+	if (!adev->gart.ptr)
+		return;
+
 	mb();
-	amdgpu_gart_flush_gpu_tlb(adev, 0);
-	return 0;
+	if (down_read_trylock(&adev->reset_domain->sem)) {
+		amdgpu_device_flush_hdp(adev, NULL);
+		up_read(&adev->reset_domain->sem);
+	}
+	for_each_set_bit(i, adev->vmhubs_mask, AMDGPU_MAX_VMHUBS)
+		amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
 }
 
 /**
@@ -319,7 +462,7 @@ int amdgpu_gart_init(struct amdgpu_device *adev)
 {
 	int r;
 
-	if (adev->dummy_page.page)
+	if (adev->dummy_page_addr)
 		return 0;
 
 	/* We need PAGE_SIZE >= AMDGPU_GPU_PAGE_SIZE */
@@ -327,44 +470,14 @@ int amdgpu_gart_init(struct amdgpu_device *adev)
 		DRM_ERROR("Page size is smaller than GPU page size!\n");
 		return -EINVAL;
 	}
-	r = amdgpu_dummy_page_init(adev);
+	r = amdgpu_gart_dummy_page_init(adev);
 	if (r)
 		return r;
 	/* Compute table size */
-	adev->gart.num_cpu_pages = adev->mc.gtt_size / PAGE_SIZE;
-	adev->gart.num_gpu_pages = adev->mc.gtt_size / AMDGPU_GPU_PAGE_SIZE;
+	adev->gart.num_cpu_pages = adev->gmc.gart_size / PAGE_SIZE;
+	adev->gart.num_gpu_pages = adev->gmc.gart_size / AMDGPU_GPU_PAGE_SIZE;
 	DRM_INFO("GART: num cpu pages %u, num gpu pages %u\n",
 		 adev->gart.num_cpu_pages, adev->gart.num_gpu_pages);
 
-#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
-	/* Allocate pages table */
-	adev->gart.pages = vzalloc(sizeof(void *) * adev->gart.num_cpu_pages);
-	if (adev->gart.pages == NULL) {
-		amdgpu_gart_fini(adev);
-		return -ENOMEM;
-	}
-#endif
-
 	return 0;
 }
-
-/**
- * amdgpu_gart_fini - tear down the driver info for managing the gart
- *
- * @adev: amdgpu_device pointer
- *
- * Tear down the gart driver info and free the dummy page (all asics).
- */
-void amdgpu_gart_fini(struct amdgpu_device *adev)
-{
-	if (adev->gart.ready) {
-		/* unbind pages */
-		amdgpu_gart_unbind(adev, 0, adev->gart.num_cpu_pages);
-	}
-	adev->gart.ready = false;
-#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
-	vfree(adev->gart.pages);
-	adev->gart.pages = NULL;
-#endif
-	amdgpu_dummy_page_fini(adev);
-}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
new file mode 100644
index 000000000000..d3118275ddae
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
@@ -0,0 +1,71 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_GART_H__
+#define __AMDGPU_GART_H__
+
+#include <linux/types.h>
+
+/*
+ * GART structures, functions & helpers
+ */
+struct amdgpu_device;
+struct amdgpu_bo;
+
+#define AMDGPU_GPU_PAGE_SIZE 4096
+#define AMDGPU_GPU_PAGE_MASK (AMDGPU_GPU_PAGE_SIZE - 1)
+#define AMDGPU_GPU_PAGE_SHIFT 12
+#define AMDGPU_GPU_PAGE_ALIGN(a) (((a) + AMDGPU_GPU_PAGE_MASK) & ~AMDGPU_GPU_PAGE_MASK)
+
+#define AMDGPU_GPU_PAGES_IN_CPU_PAGE (PAGE_SIZE / AMDGPU_GPU_PAGE_SIZE)
+
+struct amdgpu_gart {
+	struct amdgpu_bo		*bo;
+	/* CPU kmapped address of gart table */
+	void				*ptr;
+	unsigned			num_gpu_pages;
+	unsigned			num_cpu_pages;
+	unsigned			table_size;
+
+	/* Asic default pte flags */
+	uint64_t			gart_pte_flags;
+};
+
+int amdgpu_gart_table_ram_alloc(struct amdgpu_device *adev);
+void amdgpu_gart_table_ram_free(struct amdgpu_device *adev);
+int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev);
+void amdgpu_gart_table_vram_free(struct amdgpu_device *adev);
+int amdgpu_gart_init(struct amdgpu_device *adev);
+void amdgpu_gart_dummy_page_fini(struct amdgpu_device *adev);
+void amdgpu_gart_unbind(struct amdgpu_device *adev, uint64_t offset,
+			int pages);
+void amdgpu_gart_map(struct amdgpu_device *adev, uint64_t offset,
+		     int pages, dma_addr_t *dma_addr, uint64_t flags,
+		     void *dst);
+void amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
+		      int pages, dma_addr_t *dma_addr, uint64_t flags);
+void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
+				uint64_t start_page, uint64_t num_pages,
+				uint64_t flags, void *dst);
+void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev);
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h
index e73728d90388..f6ac1e9548f2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h
@@ -24,34 +24,14 @@
 #ifndef __AMDGPU_GDS_H__
 #define __AMDGPU_GDS_H__
 
-/* Because TTM request that alloacted buffer should be PAGE_SIZE aligned,
- * we should report GDS/GWS/OA size as PAGE_SIZE aligned
- * */
-#define AMDGPU_GDS_SHIFT	2
-#define AMDGPU_GWS_SHIFT	PAGE_SHIFT
-#define AMDGPU_OA_SHIFT		PAGE_SHIFT
-
 struct amdgpu_ring;
 struct amdgpu_bo;
 
-struct amdgpu_gds_asic_info {
-	uint32_t	total_size;
-	uint32_t	gfx_partition_size;
-	uint32_t	cs_partition_size;
-};
-
 struct amdgpu_gds {
-	struct amdgpu_gds_asic_info	mem;
-	struct amdgpu_gds_asic_info	gws;
-	struct amdgpu_gds_asic_info	oa;
-	/* At present, GDS, GWS and OA resources for gfx (graphics)
-	 * is always pre-allocated and available for graphics operation.
-	 * Such resource is shared between all gfx clients.
-	 * TODO: move this operation to user space
-	 * */
-	struct amdgpu_bo*		gds_gfx_bo;
-	struct amdgpu_bo*		gws_gfx_bo;
-	struct amdgpu_bo*		oa_gfx_bo;
+	uint32_t gds_size;
+	uint32_t gws_size;
+	uint32_t oa_size;
+	uint32_t gds_compute_max_wave_id;
 };
 
 struct amdgpu_gds_reg_offset {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 94cb91cf93eb..3e38c5db2987 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -26,71 +26,219 @@
  *          Jerome Glisse
  */
 #include <linux/ktime.h>
+#include <linux/module.h>
 #include <linux/pagemap.h>
-#include <drm/drmP.h>
+#include <linux/pci.h>
+#include <linux/dma-buf.h>
+
 #include <drm/amdgpu_drm.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_exec.h>
+#include <drm/drm_gem_ttm_helper.h>
+#include <drm/ttm/ttm_tt.h>
+#include <drm/drm_syncobj.h>
+
 #include "amdgpu.h"
+#include "amdgpu_display.h"
+#include "amdgpu_dma_buf.h"
+#include "amdgpu_hmm.h"
+#include "amdgpu_xgmi.h"
+#include "amdgpu_vm.h"
+
+static int
+amdgpu_gem_add_input_fence(struct drm_file *filp,
+			   uint64_t syncobj_handles_array,
+			   uint32_t num_syncobj_handles)
+{
+	struct dma_fence *fence;
+	uint32_t *syncobj_handles;
+	int ret, i;
+
+	if (!num_syncobj_handles)
+		return 0;
+
+	syncobj_handles = memdup_user(u64_to_user_ptr(syncobj_handles_array),
+				      size_mul(sizeof(uint32_t), num_syncobj_handles));
+	if (IS_ERR(syncobj_handles))
+		return PTR_ERR(syncobj_handles);
+
+	for (i = 0; i < num_syncobj_handles; i++) {
+
+		if (!syncobj_handles[i]) {
+			ret = -EINVAL;
+			goto free_memdup;
+		}
+
+		ret = drm_syncobj_find_fence(filp, syncobj_handles[i], 0, 0, &fence);
+		if (ret)
+			goto free_memdup;
 
-void amdgpu_gem_object_free(struct drm_gem_object *gobj)
+		dma_fence_wait(fence, false);
+
+		/* TODO: optimize async handling */
+		dma_fence_put(fence);
+	}
+
+free_memdup:
+	kfree(syncobj_handles);
+	return ret;
+}
+
+static int
+amdgpu_gem_update_timeline_node(struct drm_file *filp,
+				uint32_t syncobj_handle,
+				uint64_t point,
+				struct drm_syncobj **syncobj,
+				struct dma_fence_chain **chain)
 {
-	struct amdgpu_bo *robj = gem_to_amdgpu_bo(gobj);
+	if (!syncobj_handle)
+		return 0;
+
+	/* Find the sync object */
+	*syncobj = drm_syncobj_find(filp, syncobj_handle);
+	if (!*syncobj)
+		return -ENOENT;
+
+	if (!point)
+		return 0;
 
-	if (robj) {
-		if (robj->gem_base.import_attach)
-			drm_prime_gem_destroy(&robj->gem_base, robj->tbo.sg);
-		amdgpu_mn_unregister(robj);
-		amdgpu_bo_unref(&robj);
+	/* Allocate the chain node */
+	*chain = dma_fence_chain_alloc();
+	if (!*chain) {
+		drm_syncobj_put(*syncobj);
+		return -ENOMEM;
 	}
+
+	return 0;
 }
 
-int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
-				int alignment, u32 initial_domain,
-				u64 flags, bool kernel,
-				struct drm_gem_object **obj)
+static void
+amdgpu_gem_update_bo_mapping(struct drm_file *filp,
+			     struct amdgpu_bo_va *bo_va,
+			     uint32_t operation,
+			     uint64_t point,
+			     struct dma_fence *fence,
+			     struct drm_syncobj *syncobj,
+			     struct dma_fence_chain *chain)
 {
-	struct amdgpu_bo *robj;
-	unsigned long max_size;
-	int r;
+	struct amdgpu_bo *bo = bo_va ? bo_va->base.bo : NULL;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
+	struct amdgpu_vm *vm = &fpriv->vm;
+	struct dma_fence *last_update;
 
-	*obj = NULL;
-	/* At least align on page size */
-	if (alignment < PAGE_SIZE) {
-		alignment = PAGE_SIZE;
+	if (!syncobj)
+		return;
+
+	/* Find the last update fence */
+	switch (operation) {
+	case AMDGPU_VA_OP_MAP:
+	case AMDGPU_VA_OP_REPLACE:
+		if (bo && (bo->tbo.base.resv == vm->root.bo->tbo.base.resv))
+			last_update = vm->last_update;
+		else
+			last_update = bo_va->last_pt_update;
+		break;
+	case AMDGPU_VA_OP_UNMAP:
+	case AMDGPU_VA_OP_CLEAR:
+		last_update = fence;
+		break;
+	default:
+		return;
 	}
 
-	if (!(initial_domain & (AMDGPU_GEM_DOMAIN_GDS | AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA))) {
-		/* Maximum bo size is the unpinned gtt size since we use the gtt to
-		 * handle vram to system pool migrations.
-		 */
-		max_size = adev->mc.gtt_size - adev->gart_pin_size;
-		if (size > max_size) {
-			DRM_DEBUG("Allocation size %ldMb bigger than %ldMb limit\n",
-				  size >> 20, max_size >> 20);
-			return -ENOMEM;
+	/* Add fence to timeline */
+	if (!point)
+		drm_syncobj_replace_fence(syncobj, last_update);
+	else
+		drm_syncobj_add_point(syncobj, chain, last_update, point);
+}
+
+static vm_fault_t amdgpu_gem_fault(struct vm_fault *vmf)
+{
+	struct ttm_buffer_object *bo = vmf->vma->vm_private_data;
+	struct drm_device *ddev = bo->base.dev;
+	vm_fault_t ret;
+	int idx;
+
+	ret = ttm_bo_vm_reserve(bo, vmf);
+	if (ret)
+		return ret;
+
+	if (drm_dev_enter(ddev, &idx)) {
+		ret = amdgpu_bo_fault_reserve_notify(bo);
+		if (ret) {
+			drm_dev_exit(idx);
+			goto unlock;
 		}
+
+		ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot,
+					       TTM_BO_VM_NUM_PREFAULT);
+
+		drm_dev_exit(idx);
+	} else {
+		ret = ttm_bo_vm_dummy_page(vmf, vmf->vma->vm_page_prot);
 	}
-retry:
-	r = amdgpu_bo_create(adev, size, alignment, kernel, initial_domain,
-			     flags, NULL, NULL, &robj);
-	if (r) {
-		if (r != -ERESTARTSYS) {
-			if (initial_domain == AMDGPU_GEM_DOMAIN_VRAM) {
-				initial_domain |= AMDGPU_GEM_DOMAIN_GTT;
-				goto retry;
-			}
-			DRM_ERROR("Failed to allocate GEM object (%ld, %d, %u, %d)\n",
-				  size, initial_domain, alignment, r);
-		}
+	if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
+		return ret;
+
+unlock:
+	dma_resv_unlock(bo->base.resv);
+	return ret;
+}
+
+static const struct vm_operations_struct amdgpu_gem_vm_ops = {
+	.fault = amdgpu_gem_fault,
+	.open = ttm_bo_vm_open,
+	.close = ttm_bo_vm_close,
+	.access = ttm_bo_vm_access
+};
+
+static void amdgpu_gem_object_free(struct drm_gem_object *gobj)
+{
+	struct amdgpu_bo *aobj = gem_to_amdgpu_bo(gobj);
+
+	amdgpu_hmm_unregister(aobj);
+	ttm_bo_fini(&aobj->tbo);
+}
+
+int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
+			     int alignment, u32 initial_domain,
+			     u64 flags, enum ttm_bo_type type,
+			     struct dma_resv *resv,
+			     struct drm_gem_object **obj, int8_t xcp_id_plus1)
+{
+	struct amdgpu_bo *bo;
+	struct amdgpu_bo_user *ubo;
+	struct amdgpu_bo_param bp;
+	int r;
+
+	memset(&bp, 0, sizeof(bp));
+	*obj = NULL;
+	flags |= AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE;
+
+	bp.size = size;
+	bp.byte_align = alignment;
+	bp.type = type;
+	bp.resv = resv;
+	bp.preferred_domain = initial_domain;
+	bp.flags = flags;
+	bp.domain = initial_domain;
+	bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+	bp.xcp_id_plus1 = xcp_id_plus1;
+
+	r = amdgpu_bo_create_user(adev, &bp, &ubo);
+	if (r)
 		return r;
-	}
-	*obj = &robj->gem_base;
+
+	bo = &ubo->bo;
+	*obj = &bo->tbo.base;
 
 	return 0;
 }
 
 void amdgpu_gem_force_release(struct amdgpu_device *adev)
 {
-	struct drm_device *ddev = adev->ddev;
+	struct drm_device *ddev = adev_to_drm(adev);
 	struct drm_file *file;
 
 	mutex_lock(&ddev->filelist_mutex);
@@ -103,7 +251,7 @@ void amdgpu_gem_force_release(struct amdgpu_device *adev)
 		spin_lock(&file->table_lock);
 		idr_for_each_entry(&file->object_idr, gobj, handle) {
 			WARN_ONCE(1, "And also active allocations!\n");
-			drm_gem_object_unreference_unlocked(gobj);
+			drm_gem_object_put(gobj);
 		}
 		idr_destroy(&file->object_idr);
 		spin_unlock(&file->table_lock);
@@ -116,203 +264,280 @@ void amdgpu_gem_force_release(struct amdgpu_device *adev)
  * Call from drm_gem_handle_create which appear in both new and open ioctl
  * case.
  */
-int amdgpu_gem_object_open(struct drm_gem_object *obj,
-			   struct drm_file *file_priv)
+static int amdgpu_gem_object_open(struct drm_gem_object *obj,
+				  struct drm_file *file_priv)
 {
 	struct amdgpu_bo *abo = gem_to_amdgpu_bo(obj);
 	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
 	struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_va *bo_va;
+	struct mm_struct *mm;
 	int r;
+
+	mm = amdgpu_ttm_tt_get_usermm(abo->tbo.ttm);
+	if (mm && mm != current->mm)
+		return -EPERM;
+
+	if (abo->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID &&
+	    !amdgpu_vm_is_bo_always_valid(vm, abo))
+		return -EPERM;
+
 	r = amdgpu_bo_reserve(abo, false);
 	if (r)
 		return r;
 
+	amdgpu_vm_bo_update_shared(abo);
 	bo_va = amdgpu_vm_bo_find(vm, abo);
-	if (!bo_va) {
+	if (!bo_va)
 		bo_va = amdgpu_vm_bo_add(adev, vm, abo);
-	} else {
+	else
 		++bo_va->ref_count;
-	}
-	amdgpu_bo_unreserve(abo);
-	return 0;
-}
-
-static int amdgpu_gem_vm_check(void *param, struct amdgpu_bo *bo)
-{
-	/* if anything is swapped out don't swap it in here,
-	   just abort and wait for the next CS */
-	if (!amdgpu_bo_gpu_accessible(bo))
-		return -ERESTARTSYS;
-
-	if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
-		return -ERESTARTSYS;
 
-	return 0;
-}
+	/* attach gfx eviction fence */
+	r = amdgpu_eviction_fence_attach(&fpriv->evf_mgr, abo);
+	if (r) {
+		DRM_DEBUG_DRIVER("Failed to attach eviction fence to BO\n");
+		amdgpu_bo_unreserve(abo);
+		return r;
+	}
 
-static bool amdgpu_gem_vm_ready(struct amdgpu_device *adev,
-				struct amdgpu_vm *vm,
-				struct list_head *list)
-{
-	struct ttm_validate_buffer *entry;
+	amdgpu_bo_unreserve(abo);
 
-	list_for_each_entry(entry, list, head) {
-		struct amdgpu_bo *bo =
-			container_of(entry->bo, struct amdgpu_bo, tbo);
-		if (amdgpu_gem_vm_check(NULL, bo))
-			return false;
+	/* Validate and add eviction fence to DMABuf imports with dynamic
+	 * attachment in compute VMs. Re-validation will be done by
+	 * amdgpu_vm_validate. Fences are on the reservation shared with the
+	 * export, which is currently required to be validated and fenced
+	 * already by amdgpu_amdkfd_gpuvm_restore_process_bos.
+	 *
+	 * Nested locking below for the case that a GEM object is opened in
+	 * kfd_mem_export_dmabuf. Since the lock below is only taken for imports,
+	 * but not for export, this is a different lock class that cannot lead to
+	 * circular lock dependencies.
+	 */
+	if (!vm->is_compute_context || !vm->process_info)
+		return 0;
+	if (!drm_gem_is_imported(obj) ||
+	    !dma_buf_is_dynamic(obj->import_attach->dmabuf))
+		return 0;
+	mutex_lock_nested(&vm->process_info->lock, 1);
+	if (!WARN_ON(!vm->process_info->eviction_fence)) {
+		r = amdgpu_amdkfd_bo_validate_and_fence(abo, AMDGPU_GEM_DOMAIN_GTT,
+							&vm->process_info->eviction_fence->base);
+		if (r) {
+			struct amdgpu_task_info *ti = amdgpu_vm_get_task_info_vm(vm);
+
+			dev_warn(adev->dev, "validate_and_fence failed: %d\n", r);
+			if (ti) {
+				dev_warn(adev->dev, "pid %d\n", ti->task.pid);
+				amdgpu_vm_put_task_info(ti);
+			}
+		}
 	}
+	mutex_unlock(&vm->process_info->lock);
 
-	return !amdgpu_vm_validate_pt_bos(adev, vm, amdgpu_gem_vm_check, NULL);
+	return r;
 }
 
-void amdgpu_gem_object_close(struct drm_gem_object *obj,
-			     struct drm_file *file_priv)
+static void amdgpu_gem_object_close(struct drm_gem_object *obj,
+				    struct drm_file *file_priv)
 {
 	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
 	struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
 	struct amdgpu_vm *vm = &fpriv->vm;
 
-	struct amdgpu_bo_list_entry vm_pd;
-	struct list_head list;
-	struct ttm_validate_buffer tv;
-	struct ww_acquire_ctx ticket;
+	struct dma_fence *fence = NULL;
 	struct amdgpu_bo_va *bo_va;
-	int r;
-
-	INIT_LIST_HEAD(&list);
-
-	tv.bo = &bo->tbo;
-	tv.shared = true;
-	list_add(&tv.head, &list);
+	struct drm_exec exec;
+	long r;
+
+	drm_exec_init(&exec, DRM_EXEC_IGNORE_DUPLICATES, 0);
+	drm_exec_until_all_locked(&exec) {
+		r = drm_exec_prepare_obj(&exec, &bo->tbo.base, 1);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(r))
+			goto out_unlock;
+
+		r = amdgpu_vm_lock_pd(vm, &exec, 0);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(r))
+			goto out_unlock;
+	}
 
-	amdgpu_vm_get_pd_bo(vm, &list, &vm_pd);
+	if (!amdgpu_vm_is_bo_always_valid(vm, bo))
+		amdgpu_eviction_fence_detach(&fpriv->evf_mgr, bo);
 
-	r = ttm_eu_reserve_buffers(&ticket, &list, false, NULL);
-	if (r) {
-		dev_err(adev->dev, "leaking bo va because "
-			"we fail to reserve bo (%d)\n", r);
-		return;
-	}
 	bo_va = amdgpu_vm_bo_find(vm, bo);
-	if (bo_va && --bo_va->ref_count == 0) {
-		amdgpu_vm_bo_rmv(adev, bo_va);
+	if (!bo_va || --bo_va->ref_count)
+		goto out_unlock;
 
-		if (amdgpu_gem_vm_ready(adev, vm, &list)) {
-			struct dma_fence *fence = NULL;
+	amdgpu_vm_bo_del(adev, bo_va);
+	amdgpu_vm_bo_update_shared(bo);
+	if (!amdgpu_vm_ready(vm))
+		goto out_unlock;
 
-			r = amdgpu_vm_clear_freed(adev, vm, &fence);
-			if (unlikely(r)) {
-				dev_err(adev->dev, "failed to clear page "
-					"tables on GEM object close (%d)\n", r);
-			}
+	r = amdgpu_vm_clear_freed(adev, vm, &fence);
+	if (unlikely(r < 0))
+		dev_err(adev->dev, "failed to clear page "
+			"tables on GEM object close (%ld)\n", r);
+	if (r || !fence)
+		goto out_unlock;
 
-			if (fence) {
-				amdgpu_bo_fence(bo, fence, true);
-				dma_fence_put(fence);
-			}
-		}
-	}
-	ttm_eu_backoff_reservation(&ticket, &list);
+	amdgpu_bo_fence(bo, fence, true);
+	dma_fence_put(fence);
+
+out_unlock:
+	if (r)
+		dev_err(adev->dev, "leaking bo va (%ld)\n", r);
+	drm_exec_fini(&exec);
 }
 
-static int amdgpu_gem_handle_lockup(struct amdgpu_device *adev, int r)
+static int amdgpu_gem_object_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
 {
-	if (r == -EDEADLK) {
-		r = amdgpu_gpu_reset(adev);
-		if (!r)
-			r = -EAGAIN;
-	}
-	return r;
+	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
+
+	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm))
+		return -EPERM;
+	if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
+		return -EPERM;
+
+	/* Workaround for Thunk bug creating PROT_NONE,MAP_PRIVATE mappings
+	 * for debugger access to invisible VRAM. Should have used MAP_SHARED
+	 * instead. Clearing VM_MAYWRITE prevents the mapping from ever
+	 * becoming writable and makes is_cow_mapping(vm_flags) false.
+	 */
+	if (is_cow_mapping(vma->vm_flags) &&
+	    !(vma->vm_flags & VM_ACCESS_FLAGS))
+		vm_flags_clear(vma, VM_MAYWRITE);
+
+	return drm_gem_ttm_mmap(obj, vma);
 }
 
+const struct drm_gem_object_funcs amdgpu_gem_object_funcs = {
+	.free = amdgpu_gem_object_free,
+	.open = amdgpu_gem_object_open,
+	.close = amdgpu_gem_object_close,
+	.export = amdgpu_gem_prime_export,
+	.vmap = drm_gem_ttm_vmap,
+	.vunmap = drm_gem_ttm_vunmap,
+	.mmap = amdgpu_gem_object_mmap,
+	.vm_ops = &amdgpu_gem_vm_ops,
+};
+
 /*
  * GEM ioctls.
  */
 int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
 			    struct drm_file *filp)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
+	struct amdgpu_vm *vm = &fpriv->vm;
 	union drm_amdgpu_gem_create *args = data;
+	uint64_t flags = args->in.domain_flags;
 	uint64_t size = args->in.bo_size;
+	struct dma_resv *resv = NULL;
 	struct drm_gem_object *gobj;
-	uint32_t handle;
-	bool kernel = false;
+	uint32_t handle, initial_domain;
 	int r;
 
 	/* reject invalid gem flags */
-	if (args->in.domain_flags & ~(AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-				      AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
-				      AMDGPU_GEM_CREATE_CPU_GTT_USWC |
-				      AMDGPU_GEM_CREATE_VRAM_CLEARED|
-				      AMDGPU_GEM_CREATE_SHADOW |
-				      AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)) {
-		r = -EINVAL;
-		goto error_unlock;
-	}
+	if (flags & ~AMDGPU_GEM_CREATE_SETTABLE_MASK)
+		return -EINVAL;
+
 	/* reject invalid gem domains */
-	if (args->in.domains & ~(AMDGPU_GEM_DOMAIN_CPU |
-				 AMDGPU_GEM_DOMAIN_GTT |
-				 AMDGPU_GEM_DOMAIN_VRAM |
-				 AMDGPU_GEM_DOMAIN_GDS |
-				 AMDGPU_GEM_DOMAIN_GWS |
-				 AMDGPU_GEM_DOMAIN_OA)) {
-		r = -EINVAL;
-		goto error_unlock;
+	if (args->in.domains & ~AMDGPU_GEM_DOMAIN_MASK)
+		return -EINVAL;
+
+	if (!amdgpu_is_tmz(adev) && (flags & AMDGPU_GEM_CREATE_ENCRYPTED)) {
+		DRM_NOTE_ONCE("Cannot allocate secure buffer since TMZ is disabled\n");
+		return -EINVAL;
 	}
 
+	/* always clear VRAM */
+	flags |= AMDGPU_GEM_CREATE_VRAM_CLEARED;
+
+	if (args->in.domains & AMDGPU_GEM_DOMAIN_MMIO_REMAP)
+		return -EINVAL;
+
 	/* create a gem object to contain this object in */
 	if (args->in.domains & (AMDGPU_GEM_DOMAIN_GDS |
 	    AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA)) {
-		kernel = true;
-		if (args->in.domains == AMDGPU_GEM_DOMAIN_GDS)
-			size = size << AMDGPU_GDS_SHIFT;
-		else if (args->in.domains == AMDGPU_GEM_DOMAIN_GWS)
-			size = size << AMDGPU_GWS_SHIFT;
-		else if (args->in.domains == AMDGPU_GEM_DOMAIN_OA)
-			size = size << AMDGPU_OA_SHIFT;
-		else {
-			r = -EINVAL;
-			goto error_unlock;
+		if (flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID) {
+			/* if gds bo is created from user space, it must be
+			 * passed to bo list
+			 */
+			DRM_ERROR("GDS bo cannot be per-vm-bo\n");
+			return -EINVAL;
 		}
+		flags |= AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
+	}
+
+	if (flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID) {
+		r = amdgpu_bo_reserve(vm->root.bo, false);
+		if (r)
+			return r;
+
+		resv = vm->root.bo->tbo.base.resv;
 	}
-	size = roundup(size, PAGE_SIZE);
 
+	initial_domain = (u32)(0xffffffff & args->in.domains);
+retry:
 	r = amdgpu_gem_object_create(adev, size, args->in.alignment,
-				     (u32)(0xffffffff & args->in.domains),
-				     args->in.domain_flags,
-				     kernel, &gobj);
+				     initial_domain,
+				     flags, ttm_bo_type_device, resv, &gobj, fpriv->xcp_id + 1);
+	if (r && r != -ERESTARTSYS) {
+		if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) {
+			flags &= ~AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+			goto retry;
+		}
+
+		if (initial_domain == AMDGPU_GEM_DOMAIN_VRAM) {
+			initial_domain |= AMDGPU_GEM_DOMAIN_GTT;
+			goto retry;
+		}
+		DRM_DEBUG("Failed to allocate GEM object (%llu, %d, %llu, %d)\n",
+				size, initial_domain, args->in.alignment, r);
+	}
+
+	if (flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID) {
+		if (!r) {
+			struct amdgpu_bo *abo = gem_to_amdgpu_bo(gobj);
+
+			abo->parent = amdgpu_bo_ref(vm->root.bo);
+		}
+		amdgpu_bo_unreserve(vm->root.bo);
+	}
 	if (r)
-		goto error_unlock;
+		return r;
 
 	r = drm_gem_handle_create(filp, gobj, &handle);
 	/* drop reference from allocate - handle holds it now */
-	drm_gem_object_unreference_unlocked(gobj);
+	drm_gem_object_put(gobj);
 	if (r)
-		goto error_unlock;
+		return r;
 
 	memset(args, 0, sizeof(*args));
 	args->out.handle = handle;
 	return 0;
-
-error_unlock:
-	r = amdgpu_gem_handle_lockup(adev, r);
-	return r;
 }
 
 int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
 			     struct drm_file *filp)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct ttm_operation_ctx ctx = { true, false };
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct drm_amdgpu_gem_userptr *args = data;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	struct drm_gem_object *gobj;
+	struct amdgpu_hmm_range *range;
 	struct amdgpu_bo *bo;
 	uint32_t handle;
 	int r;
 
+	args->addr = untagged_addr(args->addr);
+
 	if (offset_in_page(args->addr | args->size))
 		return -EINVAL;
 
@@ -330,66 +555,55 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
 	}
 
 	/* create a gem object to contain this object in */
-	r = amdgpu_gem_object_create(adev, args->size, 0,
-				     AMDGPU_GEM_DOMAIN_CPU, 0,
-				     0, &gobj);
+	r = amdgpu_gem_object_create(adev, args->size, 0, AMDGPU_GEM_DOMAIN_CPU,
+				     0, ttm_bo_type_device, NULL, &gobj, fpriv->xcp_id + 1);
 	if (r)
-		goto handle_lockup;
+		return r;
 
 	bo = gem_to_amdgpu_bo(gobj);
-	bo->prefered_domains = AMDGPU_GEM_DOMAIN_GTT;
+	bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
 	bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
-	r = amdgpu_ttm_tt_set_userptr(bo->tbo.ttm, args->addr, args->flags);
+	r = amdgpu_ttm_tt_set_userptr(&bo->tbo, args->addr, args->flags);
 	if (r)
 		goto release_object;
 
-	if (args->flags & AMDGPU_GEM_USERPTR_REGISTER) {
-		r = amdgpu_mn_register(bo, args->addr);
-		if (r)
-			goto release_object;
-	}
+	r = amdgpu_hmm_register(bo, args->addr);
+	if (r)
+		goto release_object;
 
 	if (args->flags & AMDGPU_GEM_USERPTR_VALIDATE) {
-		down_read(&current->mm->mmap_sem);
-
-		r = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm,
-						 bo->tbo.ttm->pages);
-		if (r)
-			goto unlock_mmap_sem;
-
+		range = amdgpu_hmm_range_alloc(NULL);
+		if (unlikely(!range))
+			return -ENOMEM;
+		r = amdgpu_ttm_tt_get_user_pages(bo, range);
+		if (r) {
+			amdgpu_hmm_range_free(range);
+			goto release_object;
+		}
 		r = amdgpu_bo_reserve(bo, true);
 		if (r)
-			goto free_pages;
+			goto user_pages_done;
 
-		amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
-		r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
+		amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, range);
+
+		amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
 		amdgpu_bo_unreserve(bo);
 		if (r)
-			goto free_pages;
-
-		up_read(&current->mm->mmap_sem);
+			goto user_pages_done;
 	}
 
 	r = drm_gem_handle_create(filp, gobj, &handle);
-	/* drop reference from allocate - handle holds it now */
-	drm_gem_object_unreference_unlocked(gobj);
 	if (r)
-		goto handle_lockup;
+		goto user_pages_done;
 
 	args->handle = handle;
-	return 0;
-
-free_pages:
-	release_pages(bo->tbo.ttm->pages, bo->tbo.ttm->num_pages, false);
-
-unlock_mmap_sem:
-	up_read(&current->mm->mmap_sem);
 
+user_pages_done:
+	if (args->flags & AMDGPU_GEM_USERPTR_VALIDATE)
+		amdgpu_hmm_range_free(range);
 release_object:
-	drm_gem_object_unreference_unlocked(gobj);
-
-handle_lockup:
-	r = amdgpu_gem_handle_lockup(adev, r);
+	drm_gem_object_put(gobj);
 
 	return r;
 }
@@ -402,17 +616,17 @@ int amdgpu_mode_dumb_mmap(struct drm_file *filp,
 	struct amdgpu_bo *robj;
 
 	gobj = drm_gem_object_lookup(filp, handle);
-	if (gobj == NULL) {
+	if (!gobj)
 		return -ENOENT;
-	}
+
 	robj = gem_to_amdgpu_bo(gobj);
 	if (amdgpu_ttm_tt_get_usermm(robj->tbo.ttm) ||
 	    (robj->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) {
-		drm_gem_object_unreference_unlocked(gobj);
+		drm_gem_object_put(gobj);
 		return -EPERM;
 	}
 	*offset_p = amdgpu_bo_mmap_offset(robj);
-	drm_gem_object_unreference_unlocked(gobj);
+	drm_gem_object_put(gobj);
 	return 0;
 }
 
@@ -421,6 +635,7 @@ int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
 {
 	union drm_amdgpu_gem_mmap *args = data;
 	uint32_t handle = args->in.handle;
+
 	memset(args, 0, sizeof(*args));
 	return amdgpu_mode_dumb_mmap(filp, dev, handle, &args->out.addr_ptr);
 }
@@ -447,7 +662,7 @@ unsigned long amdgpu_gem_timeout(uint64_t timeout_ns)
 
 	timeout_jiffies = nsecs_to_jiffies(ktime_to_ns(timeout));
 	/*  clamp timeout to avoid unsigned-> signed overflow */
-	if (timeout_jiffies > MAX_SCHEDULE_TIMEOUT )
+	if (timeout_jiffies > MAX_SCHEDULE_TIMEOUT)
 		return MAX_SCHEDULE_TIMEOUT - 1;
 
 	return timeout_jiffies;
@@ -456,7 +671,6 @@ unsigned long amdgpu_gem_timeout(uint64_t timeout_ns)
 int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
 			      struct drm_file *filp)
 {
-	struct amdgpu_device *adev = dev->dev_private;
 	union drm_amdgpu_gem_wait_idle *args = data;
 	struct drm_gem_object *gobj;
 	struct amdgpu_bo *robj;
@@ -466,12 +680,12 @@ int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
 	long ret;
 
 	gobj = drm_gem_object_lookup(filp, handle);
-	if (gobj == NULL) {
+	if (!gobj)
 		return -ENOENT;
-	}
+
 	robj = gem_to_amdgpu_bo(gobj);
-	ret = reservation_object_wait_timeout_rcu(robj->tbo.resv, true, true,
-						  timeout);
+	ret = dma_resv_wait_timeout(robj->tbo.base.resv, DMA_RESV_USAGE_READ,
+				    true, timeout);
 
 	/* ret == 0 means not signaled,
 	 * ret > 0 means signaled
@@ -483,8 +697,7 @@ int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
 	} else
 		r = ret;
 
-	drm_gem_object_unreference_unlocked(gobj);
-	r = amdgpu_gem_handle_lockup(adev, r);
+	drm_gem_object_put(gobj);
 	return r;
 }
 
@@ -496,7 +709,7 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
 	struct amdgpu_bo *robj;
 	int r = -1;
 
-	DRM_DEBUG("%d \n", args->handle);
+	DRM_DEBUG("%d\n", args->handle);
 	gobj = drm_gem_object_lookup(filp, args->handle);
 	if (gobj == NULL)
 		return -ENOENT;
@@ -527,7 +740,7 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
 unreserve:
 	amdgpu_bo_unreserve(robj);
 out:
-	drm_gem_object_unreference_unlocked(gobj);
+	drm_gem_object_put(gobj);
 	return r;
 }
 
@@ -537,38 +750,44 @@ out:
  * @adev: amdgpu_device pointer
  * @vm: vm to update
  * @bo_va: bo_va to update
- * @list: validation list
  * @operation: map, unmap or clear
  *
  * Update the bo_va directly after setting its address. Errors are not
  * vital here, so they are not reported back to userspace.
+ *
+ * Returns resulting fence if freed BO(s) got cleared from the PT.
+ * otherwise stub fence in case of error.
  */
-static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
-				    struct amdgpu_vm *vm,
-				    struct amdgpu_bo_va *bo_va,
-				    struct list_head *list,
-				    uint32_t operation)
+static struct dma_fence *
+amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
+			struct amdgpu_vm *vm,
+			struct amdgpu_bo_va *bo_va,
+			uint32_t operation)
 {
-	int r = -ERESTARTSYS;
+	struct dma_fence *fence = dma_fence_get_stub();
+	int r;
 
-	if (!amdgpu_gem_vm_ready(adev, vm, list))
-		goto error;
+	if (!amdgpu_vm_ready(vm))
+		return fence;
 
-	r = amdgpu_vm_update_directories(adev, vm);
-	if (r)
-		goto error;
-
-	r = amdgpu_vm_clear_freed(adev, vm, NULL);
+	r = amdgpu_vm_clear_freed(adev, vm, &fence);
 	if (r)
 		goto error;
 
 	if (operation == AMDGPU_VA_OP_MAP ||
-	    operation == AMDGPU_VA_OP_REPLACE)
+	    operation == AMDGPU_VA_OP_REPLACE) {
 		r = amdgpu_vm_bo_update(adev, bo_va, false);
+		if (r)
+			goto error;
+	}
+
+	r = amdgpu_vm_update_pdes(adev, vm, false);
 
 error:
 	if (r && r != -ERESTARTSYS)
 		DRM_ERROR("Couldn't update BO_VA (%d)\n", r);
+
+	return fence;
 }
 
 int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
@@ -576,36 +795,53 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 {
 	const uint32_t valid_flags = AMDGPU_VM_DELAY_UPDATE |
 		AMDGPU_VM_PAGE_READABLE | AMDGPU_VM_PAGE_WRITEABLE |
-		AMDGPU_VM_PAGE_EXECUTABLE | AMDGPU_VM_MTYPE_MASK;
+		AMDGPU_VM_PAGE_EXECUTABLE | AMDGPU_VM_MTYPE_MASK |
+		AMDGPU_VM_PAGE_NOALLOC;
 	const uint32_t prt_flags = AMDGPU_VM_DELAY_UPDATE |
 		AMDGPU_VM_PAGE_PRT;
 
 	struct drm_amdgpu_gem_va *args = data;
 	struct drm_gem_object *gobj;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	struct amdgpu_bo *abo;
 	struct amdgpu_bo_va *bo_va;
-	struct amdgpu_bo_list_entry vm_pd;
-	struct ttm_validate_buffer tv;
-	struct ww_acquire_ctx ticket;
-	struct list_head list;
-	uint64_t va_flags;
+	struct drm_syncobj *timeline_syncobj = NULL;
+	struct dma_fence_chain *timeline_chain = NULL;
+	struct dma_fence *fence;
+	struct drm_exec exec;
+	uint64_t vm_size;
 	int r = 0;
 
-	if (!adev->vm_manager.enabled)
-		return -ENOTTY;
+	if (args->va_address < AMDGPU_VA_RESERVED_BOTTOM) {
+		dev_dbg(dev->dev,
+			"va_address 0x%llx is in reserved area 0x%llx\n",
+			args->va_address, AMDGPU_VA_RESERVED_BOTTOM);
+		return -EINVAL;
+	}
 
-	if (args->va_address < AMDGPU_VA_RESERVED_SIZE) {
-		dev_err(&dev->pdev->dev,
-			"va_address 0x%lX is in reserved area 0x%X\n",
-			(unsigned long)args->va_address,
-			AMDGPU_VA_RESERVED_SIZE);
+	if (args->va_address >= AMDGPU_GMC_HOLE_START &&
+	    args->va_address < AMDGPU_GMC_HOLE_END) {
+		dev_dbg(dev->dev,
+			"va_address 0x%llx is in VA hole 0x%llx-0x%llx\n",
+			args->va_address, AMDGPU_GMC_HOLE_START,
+			AMDGPU_GMC_HOLE_END);
+		return -EINVAL;
+	}
+
+	args->va_address &= AMDGPU_GMC_HOLE_MASK;
+
+	vm_size = adev->vm_manager.max_pfn * AMDGPU_GPU_PAGE_SIZE;
+	vm_size -= AMDGPU_VA_RESERVED_TOP;
+	if (args->va_address + args->map_size > vm_size) {
+		dev_dbg(dev->dev,
+			"va_address 0x%llx is in top reserved area 0x%llx\n",
+			args->va_address + args->map_size, vm_size);
 		return -EINVAL;
 	}
 
 	if ((args->flags & ~valid_flags) && (args->flags & ~prt_flags)) {
-		dev_err(&dev->pdev->dev, "invalid flags combination 0x%08X\n",
+		dev_dbg(dev->dev, "invalid flags combination 0x%08X\n",
 			args->flags);
 		return -EINVAL;
 	}
@@ -617,37 +853,49 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 	case AMDGPU_VA_OP_REPLACE:
 		break;
 	default:
-		dev_err(&dev->pdev->dev, "unsupported operation %d\n",
+		dev_dbg(dev->dev, "unsupported operation %d\n",
 			args->operation);
 		return -EINVAL;
 	}
 
-	INIT_LIST_HEAD(&list);
 	if ((args->operation != AMDGPU_VA_OP_CLEAR) &&
 	    !(args->flags & AMDGPU_VM_PAGE_PRT)) {
 		gobj = drm_gem_object_lookup(filp, args->handle);
 		if (gobj == NULL)
 			return -ENOENT;
 		abo = gem_to_amdgpu_bo(gobj);
-		tv.bo = &abo->tbo;
-		tv.shared = false;
-		list_add(&tv.head, &list);
 	} else {
 		gobj = NULL;
 		abo = NULL;
 	}
 
-	amdgpu_vm_get_pd_bo(&fpriv->vm, &list, &vm_pd);
-
-	r = ttm_eu_reserve_buffers(&ticket, &list, true, NULL);
+	r = amdgpu_gem_add_input_fence(filp,
+				       args->input_fence_syncobj_handles,
+				       args->num_syncobj_handles);
 	if (r)
-		goto error_unref;
+		goto error_put_gobj;
+
+	drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT |
+		      DRM_EXEC_IGNORE_DUPLICATES, 0);
+	drm_exec_until_all_locked(&exec) {
+		if (gobj) {
+			r = drm_exec_lock_obj(&exec, gobj);
+			drm_exec_retry_on_contention(&exec);
+			if (unlikely(r))
+				goto error;
+		}
+
+		r = amdgpu_vm_lock_pd(&fpriv->vm, &exec, 2);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(r))
+			goto error;
+	}
 
 	if (abo) {
 		bo_va = amdgpu_vm_bo_find(&fpriv->vm, abo);
 		if (!bo_va) {
 			r = -ENOENT;
-			goto error_backoff;
+			goto error;
 		}
 	} else if (args->operation != AMDGPU_VA_OP_CLEAR) {
 		bo_va = fpriv->prt_va;
@@ -655,17 +903,19 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 		bo_va = NULL;
 	}
 
+	r = amdgpu_gem_update_timeline_node(filp,
+					    args->vm_timeline_syncobj_out,
+					    args->vm_timeline_point,
+					    &timeline_syncobj,
+					    &timeline_chain);
+	if (r)
+		goto error;
+
 	switch (args->operation) {
 	case AMDGPU_VA_OP_MAP:
-		r = amdgpu_vm_alloc_pts(adev, bo_va->vm, args->va_address,
-					args->map_size);
-		if (r)
-			goto error_backoff;
-
-		va_flags = amdgpu_vm_get_pte_flags(adev, args->flags);
 		r = amdgpu_vm_bo_map(adev, bo_va, args->va_address,
 				     args->offset_in_bo, args->map_size,
-				     va_flags);
+				     args->flags);
 		break;
 	case AMDGPU_VA_OP_UNMAP:
 		r = amdgpu_vm_bo_unmap(adev, bo_va, args->va_address);
@@ -677,28 +927,32 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
 						args->map_size);
 		break;
 	case AMDGPU_VA_OP_REPLACE:
-		r = amdgpu_vm_alloc_pts(adev, bo_va->vm, args->va_address,
-					args->map_size);
-		if (r)
-			goto error_backoff;
-
-		va_flags = amdgpu_vm_get_pte_flags(adev, args->flags);
 		r = amdgpu_vm_bo_replace_map(adev, bo_va, args->va_address,
 					     args->offset_in_bo, args->map_size,
-					     va_flags);
+					     args->flags);
 		break;
 	default:
 		break;
 	}
-	if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE) && !amdgpu_vm_debug)
-		amdgpu_gem_va_update_vm(adev, &fpriv->vm, bo_va, &list,
-					args->operation);
+	if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE) && !adev->debug_vm) {
+		fence = amdgpu_gem_va_update_vm(adev, &fpriv->vm, bo_va,
+						args->operation);
+
+		if (timeline_syncobj)
+			amdgpu_gem_update_bo_mapping(filp, bo_va,
+					     args->operation,
+					     args->vm_timeline_point,
+					     fence, timeline_syncobj,
+					     timeline_chain);
+		else
+			dma_fence_put(fence);
 
-error_backoff:
-	ttm_eu_backoff_reservation(&ticket, &list);
+	}
 
-error_unref:
-	drm_gem_object_unreference_unlocked(gobj);
+error:
+	drm_exec_fini(&exec);
+error_put_gobj:
+	drm_gem_object_put(gobj);
 	return r;
 }
 
@@ -707,135 +961,294 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
 {
 	struct drm_amdgpu_gem_op *args = data;
 	struct drm_gem_object *gobj;
+	struct amdgpu_vm_bo_base *base;
 	struct amdgpu_bo *robj;
+	struct drm_exec exec;
+	struct amdgpu_fpriv *fpriv = filp->driver_priv;
 	int r;
 
+	if (args->padding)
+		return -EINVAL;
+
 	gobj = drm_gem_object_lookup(filp, args->handle);
-	if (gobj == NULL) {
+	if (!gobj)
 		return -ENOENT;
-	}
+
 	robj = gem_to_amdgpu_bo(gobj);
 
-	r = amdgpu_bo_reserve(robj, false);
-	if (unlikely(r))
-		goto out;
+	drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT |
+			  DRM_EXEC_IGNORE_DUPLICATES, 0);
+	drm_exec_until_all_locked(&exec) {
+		r = drm_exec_lock_obj(&exec, gobj);
+		drm_exec_retry_on_contention(&exec);
+		if (r)
+			goto out_exec;
+
+		if (args->op == AMDGPU_GEM_OP_GET_MAPPING_INFO) {
+			r = amdgpu_vm_lock_pd(&fpriv->vm, &exec, 0);
+			drm_exec_retry_on_contention(&exec);
+			if (r)
+				goto out_exec;
+		}
+	}
 
 	switch (args->op) {
 	case AMDGPU_GEM_OP_GET_GEM_CREATE_INFO: {
 		struct drm_amdgpu_gem_create_in info;
-		void __user *out = (void __user *)(uintptr_t)args->value;
+		void __user *out = u64_to_user_ptr(args->value);
 
-		info.bo_size = robj->gem_base.size;
-		info.alignment = robj->tbo.mem.page_alignment << PAGE_SHIFT;
-		info.domains = robj->prefered_domains;
+		info.bo_size = robj->tbo.base.size;
+		info.alignment = robj->tbo.page_alignment << PAGE_SHIFT;
+		info.domains = robj->preferred_domains;
 		info.domain_flags = robj->flags;
-		amdgpu_bo_unreserve(robj);
+		drm_exec_fini(&exec);
 		if (copy_to_user(out, &info, sizeof(info)))
 			r = -EFAULT;
 		break;
 	}
 	case AMDGPU_GEM_OP_SET_PLACEMENT:
-		if (robj->prime_shared_count && (args->value & AMDGPU_GEM_DOMAIN_VRAM)) {
+		if (drm_gem_is_imported(&robj->tbo.base) &&
+		    args->value & AMDGPU_GEM_DOMAIN_VRAM) {
 			r = -EINVAL;
-			amdgpu_bo_unreserve(robj);
-			break;
+			goto out_exec;
 		}
 		if (amdgpu_ttm_tt_get_usermm(robj->tbo.ttm)) {
 			r = -EPERM;
-			amdgpu_bo_unreserve(robj);
-			break;
+			goto out_exec;
 		}
-		robj->prefered_domains = args->value & (AMDGPU_GEM_DOMAIN_VRAM |
+		for (base = robj->vm_bo; base; base = base->next)
+			if (amdgpu_xgmi_same_hive(amdgpu_ttm_adev(robj->tbo.bdev),
+				amdgpu_ttm_adev(base->vm->root.bo->tbo.bdev))) {
+				r = -EINVAL;
+				goto out_exec;
+			}
+
+
+		robj->preferred_domains = args->value & (AMDGPU_GEM_DOMAIN_VRAM |
 							AMDGPU_GEM_DOMAIN_GTT |
 							AMDGPU_GEM_DOMAIN_CPU);
-		robj->allowed_domains = robj->prefered_domains;
+		robj->allowed_domains = robj->preferred_domains;
 		if (robj->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
 			robj->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
 
-		amdgpu_bo_unreserve(robj);
+		if (robj->flags & AMDGPU_GEM_CREATE_VM_ALWAYS_VALID)
+			amdgpu_vm_bo_invalidate(robj, true);
+		drm_exec_fini(&exec);
 		break;
+	case AMDGPU_GEM_OP_GET_MAPPING_INFO: {
+		struct amdgpu_bo_va *bo_va = amdgpu_vm_bo_find(&fpriv->vm, robj);
+		struct drm_amdgpu_gem_vm_entry *vm_entries;
+		struct amdgpu_bo_va_mapping *mapping;
+		int num_mappings = 0;
+		/*
+		 * num_entries is set as an input to the size of the user-allocated array of
+		 * drm_amdgpu_gem_vm_entry stored at args->value.
+		 * num_entries is sent back as output as the number of mappings the bo has.
+		 * If that number is larger than the size of the array, the ioctl must
+		 * be retried.
+		 */
+		vm_entries = kvcalloc(args->num_entries, sizeof(*vm_entries), GFP_KERNEL);
+		if (!vm_entries)
+			return -ENOMEM;
+
+		amdgpu_vm_bo_va_for_each_valid_mapping(bo_va, mapping) {
+			if (num_mappings < args->num_entries) {
+				vm_entries[num_mappings].addr = mapping->start * AMDGPU_GPU_PAGE_SIZE;
+				vm_entries[num_mappings].size = (mapping->last - mapping->start + 1) * AMDGPU_GPU_PAGE_SIZE;
+				vm_entries[num_mappings].offset = mapping->offset;
+				vm_entries[num_mappings].flags = mapping->flags;
+			}
+			num_mappings += 1;
+		}
+
+		amdgpu_vm_bo_va_for_each_invalid_mapping(bo_va, mapping) {
+			if (num_mappings < args->num_entries) {
+				vm_entries[num_mappings].addr = mapping->start * AMDGPU_GPU_PAGE_SIZE;
+				vm_entries[num_mappings].size = (mapping->last - mapping->start + 1) * AMDGPU_GPU_PAGE_SIZE;
+				vm_entries[num_mappings].offset = mapping->offset;
+				vm_entries[num_mappings].flags = mapping->flags;
+			}
+			num_mappings += 1;
+		}
+
+		drm_exec_fini(&exec);
+
+		if (num_mappings > 0 && num_mappings <= args->num_entries)
+			if (copy_to_user(u64_to_user_ptr(args->value), vm_entries, num_mappings * sizeof(*vm_entries)))
+				r = -EFAULT;
+
+		args->num_entries = num_mappings;
+
+		kvfree(vm_entries);
+		break;
+	}
 	default:
-		amdgpu_bo_unreserve(robj);
+		drm_exec_fini(&exec);
 		r = -EINVAL;
 	}
 
-out:
-	drm_gem_object_unreference_unlocked(gobj);
+	drm_gem_object_put(gobj);
+	return r;
+out_exec:
+	drm_exec_fini(&exec);
+	drm_gem_object_put(gobj);
 	return r;
 }
 
+/**
+ * amdgpu_gem_list_handles_ioctl - get information about a process' buffer objects
+ *
+ * @dev: drm device pointer
+ * @data: drm_amdgpu_gem_list_handles
+ * @filp: drm file pointer
+ *
+ * num_entries is set as an input to the size of the entries array.
+ * num_entries is sent back as output as the number of bos in the process.
+ * If that number is larger than the size of the array, the ioctl must
+ * be retried.
+ *
+ * Returns:
+ * 0 for success, -errno for errors.
+ */
+int amdgpu_gem_list_handles_ioctl(struct drm_device *dev, void *data,
+				  struct drm_file *filp)
+{
+	struct drm_amdgpu_gem_list_handles *args = data;
+	struct drm_amdgpu_gem_list_handles_entry *bo_entries;
+	struct drm_gem_object *gobj;
+	int id, ret = 0;
+	int bo_index = 0;
+	int num_bos = 0;
+
+	spin_lock(&filp->table_lock);
+	idr_for_each_entry(&filp->object_idr, gobj, id)
+		num_bos += 1;
+	spin_unlock(&filp->table_lock);
+
+	if (args->num_entries < num_bos) {
+		args->num_entries = num_bos;
+		return 0;
+	}
+
+	if (num_bos == 0) {
+		args->num_entries = 0;
+		return 0;
+	}
+
+	bo_entries = kvcalloc(num_bos, sizeof(*bo_entries), GFP_KERNEL);
+	if (!bo_entries)
+		return -ENOMEM;
+
+	spin_lock(&filp->table_lock);
+	idr_for_each_entry(&filp->object_idr, gobj, id) {
+		struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
+		struct drm_amdgpu_gem_list_handles_entry *bo_entry;
+
+		if (bo_index >= num_bos) {
+			ret = -EAGAIN;
+			break;
+		}
+
+		bo_entry = &bo_entries[bo_index];
+
+		bo_entry->size = amdgpu_bo_size(bo);
+		bo_entry->alloc_flags = bo->flags & AMDGPU_GEM_CREATE_SETTABLE_MASK;
+		bo_entry->preferred_domains = bo->preferred_domains;
+		bo_entry->gem_handle = id;
+		bo_entry->alignment = bo->tbo.page_alignment;
+
+		if (bo->tbo.base.import_attach)
+			bo_entry->flags |= AMDGPU_GEM_LIST_HANDLES_FLAG_IS_IMPORT;
+
+		bo_index += 1;
+	}
+	spin_unlock(&filp->table_lock);
+
+	args->num_entries = bo_index;
+
+	if (!ret)
+		if (copy_to_user(u64_to_user_ptr(args->entries), bo_entries, num_bos * sizeof(*bo_entries)))
+			ret = -EFAULT;
+
+	kvfree(bo_entries);
+
+	return ret;
+}
+
+static int amdgpu_gem_align_pitch(struct amdgpu_device *adev,
+				  int width,
+				  int cpp,
+				  bool tiled)
+{
+	int aligned = width;
+	int pitch_mask = 0;
+
+	switch (cpp) {
+	case 1:
+		pitch_mask = 255;
+		break;
+	case 2:
+		pitch_mask = 127;
+		break;
+	case 3:
+	case 4:
+		pitch_mask = 63;
+		break;
+	}
+
+	aligned += pitch_mask;
+	aligned &= ~pitch_mask;
+	return aligned * cpp;
+}
+
 int amdgpu_mode_dumb_create(struct drm_file *file_priv,
 			    struct drm_device *dev,
 			    struct drm_mode_create_dumb *args)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
 	struct drm_gem_object *gobj;
 	uint32_t handle;
+	u64 flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+		    AMDGPU_GEM_CREATE_CPU_GTT_USWC |
+		    AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+	u32 domain;
 	int r;
 
-	args->pitch = amdgpu_align_pitch(adev, args->width,
-					 DIV_ROUND_UP(args->bpp, 8), 0);
+	/*
+	 * The buffer returned from this function should be cleared, but
+	 * it can only be done if the ring is enabled or we'll fail to
+	 * create the buffer.
+	 */
+	if (adev->mman.buffer_funcs_enabled)
+		flags |= AMDGPU_GEM_CREATE_VRAM_CLEARED;
+
+	args->pitch = amdgpu_gem_align_pitch(adev, args->width,
+					     DIV_ROUND_UP(args->bpp, 8), 0);
 	args->size = (u64)args->pitch * args->height;
 	args->size = ALIGN(args->size, PAGE_SIZE);
-
-	r = amdgpu_gem_object_create(adev, args->size, 0,
-				     AMDGPU_GEM_DOMAIN_VRAM,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
-				     ttm_bo_type_device,
-				     &gobj);
+	domain = amdgpu_bo_get_preferred_domain(adev,
+				amdgpu_display_supported_domains(adev, flags));
+	r = amdgpu_gem_object_create(adev, args->size, 0, domain, flags,
+				     ttm_bo_type_device, NULL, &gobj, fpriv->xcp_id + 1);
 	if (r)
 		return -ENOMEM;
 
 	r = drm_gem_handle_create(file_priv, gobj, &handle);
 	/* drop reference from allocate - handle holds it now */
-	drm_gem_object_unreference_unlocked(gobj);
-	if (r) {
+	drm_gem_object_put(gobj);
+	if (r)
 		return r;
-	}
+
 	args->handle = handle;
 	return 0;
 }
 
 #if defined(CONFIG_DEBUG_FS)
-static int amdgpu_debugfs_gem_bo_info(int id, void *ptr, void *data)
+static int amdgpu_debugfs_gem_info_show(struct seq_file *m, void *unused)
 {
-	struct drm_gem_object *gobj = ptr;
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
-	struct seq_file *m = data;
-
-	unsigned domain;
-	const char *placement;
-	unsigned pin_count;
-
-	domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-	switch (domain) {
-	case AMDGPU_GEM_DOMAIN_VRAM:
-		placement = "VRAM";
-		break;
-	case AMDGPU_GEM_DOMAIN_GTT:
-		placement = " GTT";
-		break;
-	case AMDGPU_GEM_DOMAIN_CPU:
-	default:
-		placement = " CPU";
-		break;
-	}
-	seq_printf(m, "\t0x%08x: %12ld byte %s @ 0x%010Lx",
-		   id, amdgpu_bo_size(bo), placement,
-		   amdgpu_bo_gpu_offset(bo));
-
-	pin_count = ACCESS_ONCE(bo->pin_count);
-	if (pin_count)
-		seq_printf(m, " pin count %d", pin_count);
-	seq_printf(m, "\n");
-
-	return 0;
-}
-
-static int amdgpu_debugfs_gem_info(struct seq_file *m, void *data)
-{
-	struct drm_info_node *node = (struct drm_info_node *)m->private;
-	struct drm_device *dev = node->minor->dev;
+	struct amdgpu_device *adev = m->private;
+	struct drm_device *dev = adev_to_drm(adev);
 	struct drm_file *file;
 	int r;
 
@@ -845,6 +1258,9 @@ static int amdgpu_debugfs_gem_info(struct seq_file *m, void *data)
 
 	list_for_each_entry(file, &dev->filelist, lhead) {
 		struct task_struct *task;
+		struct drm_gem_object *gobj;
+		struct pid *pid;
+		int id;
 
 		/*
 		 * Although we have a valid reference on file->pid, that does
@@ -853,13 +1269,18 @@ static int amdgpu_debugfs_gem_info(struct seq_file *m, void *data)
 		 * Therefore, we need to protect this ->comm access using RCU.
 		 */
 		rcu_read_lock();
-		task = pid_task(file->pid, PIDTYPE_PID);
-		seq_printf(m, "pid %8d command %s:\n", pid_nr(file->pid),
+		pid = rcu_dereference(file->pid);
+		task = pid_task(pid, PIDTYPE_TGID);
+		seq_printf(m, "pid %8d command %s:\n", pid_nr(pid),
 			   task ? task->comm : "<unknown>");
 		rcu_read_unlock();
 
 		spin_lock(&file->table_lock);
-		idr_for_each(&file->object_idr, amdgpu_debugfs_gem_bo_info, m);
+		idr_for_each_entry(&file->object_idr, gobj, id) {
+			struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
+
+			amdgpu_bo_print_info(id, bo, m);
+		}
 		spin_unlock(&file->table_lock);
 	}
 
@@ -867,15 +1288,17 @@ static int amdgpu_debugfs_gem_info(struct seq_file *m, void *data)
 	return 0;
 }
 
-static const struct drm_info_list amdgpu_debugfs_gem_list[] = {
-	{"amdgpu_gem_info", &amdgpu_debugfs_gem_info, 0, NULL},
-};
+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_gem_info);
+
 #endif
 
-int amdgpu_gem_debugfs_init(struct amdgpu_device *adev)
+void amdgpu_debugfs_gem_init(struct amdgpu_device *adev)
 {
 #if defined(CONFIG_DEBUG_FS)
-	return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_gem_list, 1);
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+
+	debugfs_create_file("amdgpu_gem_info", 0444, root, adev,
+			    &amdgpu_debugfs_gem_info_fops);
 #endif
-	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h
new file mode 100644
index 000000000000..b558336bc4c6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h
@@ -0,0 +1,90 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_GEM_H__
+#define __AMDGPU_GEM_H__
+
+#include <drm/amdgpu_drm.h>
+#include <drm/drm_gem.h>
+
+/*
+ * GEM.
+ */
+
+#define AMDGPU_GEM_DOMAIN_MAX		0x3
+#define gem_to_amdgpu_bo(gobj) container_of((gobj), struct amdgpu_bo, tbo.base)
+
+extern const struct drm_gem_object_funcs amdgpu_gem_object_funcs;
+
+unsigned long amdgpu_gem_timeout(uint64_t timeout_ns);
+
+/*
+ * GEM objects.
+ */
+void amdgpu_gem_force_release(struct amdgpu_device *adev);
+int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
+			     int alignment, u32 initial_domain,
+			     u64 flags, enum ttm_bo_type type,
+			     struct dma_resv *resv,
+			     struct drm_gem_object **obj, int8_t xcp_id_plus1);
+int amdgpu_mode_dumb_create(struct drm_file *file_priv,
+			    struct drm_device *dev,
+			    struct drm_mode_create_dumb *args);
+int amdgpu_mode_dumb_mmap(struct drm_file *filp,
+			  struct drm_device *dev,
+			  uint32_t handle, uint64_t *offset_p);
+
+int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
+			    struct drm_file *filp);
+int amdgpu_gem_info_ioctl(struct drm_device *dev, void *data,
+			  struct drm_file *filp);
+int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *filp);
+int amdgpu_gem_mmap_ioctl(struct drm_device *dev, void *data,
+			  struct drm_file *filp);
+int amdgpu_gem_wait_idle_ioctl(struct drm_device *dev, void *data,
+			      struct drm_file *filp);
+int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
+			  struct drm_file *filp);
+int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *filp);
+int amdgpu_gem_list_handles_ioctl(struct drm_device *dev, void *data,
+				  struct drm_file *filp);
+
+int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
+				struct drm_file *filp);
+
+#define AMDGPU_GEM_CREATE_SETTABLE_MASK	(AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | \
+	AMDGPU_GEM_CREATE_NO_CPU_ACCESS | \
+	AMDGPU_GEM_CREATE_CPU_GTT_USWC | \
+	AMDGPU_GEM_CREATE_VRAM_CLEARED | \
+	AMDGPU_GEM_CREATE_VM_ALWAYS_VALID | \
+	AMDGPU_GEM_CREATE_EXPLICIT_SYNC | \
+	AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE | \
+	AMDGPU_GEM_CREATE_ENCRYPTED | \
+	AMDGPU_GEM_CREATE_GFX12_DCC | \
+	AMDGPU_GEM_CREATE_DISCARDABLE | \
+	AMDGPU_GEM_CREATE_COHERENT | \
+	AMDGPU_GEM_CREATE_UNCACHED | \
+	AMDGPU_GEM_CREATE_EXT_COHERENT)
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 19943356cca7..8b118c53f351 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -22,47 +22,79 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  *
  */
-#include <drm/drmP.h>
+
+#include <linux/firmware.h>
+#include <linux/pm_runtime.h>
+
 #include "amdgpu.h"
 #include "amdgpu_gfx.h"
+#include "amdgpu_rlc.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_reset.h"
+#include "amdgpu_xcp.h"
+#include "amdgpu_xgmi.h"
+#include "amdgpu_mes.h"
+#include "nvd.h"
+
+/* delay 0.1 second to enable gfx off feature */
+#define GFX_OFF_DELAY_ENABLE         msecs_to_jiffies(100)
+
+#define GFX_OFF_NO_DELAY 0
 
 /*
- * GPU scratch registers helpers function.
+ * GPU GFX IP block helpers function.
  */
-/**
- * amdgpu_gfx_scratch_get - Allocate a scratch register
- *
- * @adev: amdgpu_device pointer
- * @reg: scratch register mmio offset
- *
- * Allocate a CP scratch register for use by the driver (all asics).
- * Returns 0 on success or -EINVAL on failure.
- */
-int amdgpu_gfx_scratch_get(struct amdgpu_device *adev, uint32_t *reg)
+
+int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
+				int pipe, int queue)
 {
-	int i;
+	int bit = 0;
 
-	i = ffs(adev->gfx.scratch.free_mask);
-	if (i != 0 && i <= adev->gfx.scratch.num_reg) {
-		i--;
-		adev->gfx.scratch.free_mask &= ~(1u << i);
-		*reg = adev->gfx.scratch.reg_base + i;
-		return 0;
-	}
-	return -EINVAL;
+	bit += mec * adev->gfx.mec.num_pipe_per_mec
+		* adev->gfx.mec.num_queue_per_pipe;
+	bit += pipe * adev->gfx.mec.num_queue_per_pipe;
+	bit += queue;
+
+	return bit;
 }
 
-/**
- * amdgpu_gfx_scratch_free - Free a scratch register
- *
- * @adev: amdgpu_device pointer
- * @reg: scratch register mmio offset
- *
- * Free a CP scratch register allocated for use by the driver (all asics)
- */
-void amdgpu_gfx_scratch_free(struct amdgpu_device *adev, uint32_t reg)
+void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
+				 int *mec, int *pipe, int *queue)
+{
+	*queue = bit % adev->gfx.mec.num_queue_per_pipe;
+	*pipe = (bit / adev->gfx.mec.num_queue_per_pipe)
+		% adev->gfx.mec.num_pipe_per_mec;
+	*mec = (bit / adev->gfx.mec.num_queue_per_pipe)
+	       / adev->gfx.mec.num_pipe_per_mec;
+
+}
+
+bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev,
+				     int xcc_id, int mec, int pipe, int queue)
 {
-	adev->gfx.scratch.free_mask |= 1u << (reg - adev->gfx.scratch.reg_base);
+	return test_bit(amdgpu_gfx_mec_queue_to_bit(adev, mec, pipe, queue),
+			adev->gfx.mec_bitmap[xcc_id].queue_bitmap);
+}
+
+static int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev,
+				      int me, int pipe, int queue)
+{
+	int num_queue_per_pipe = 1; /* we only enable 1 KGQ per pipe */
+	int bit = 0;
+
+	bit += me * adev->gfx.me.num_pipe_per_me
+		* num_queue_per_pipe;
+	bit += pipe * num_queue_per_pipe;
+	bit += queue;
+
+	return bit;
+}
+
+bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev,
+				    int me, int pipe, int queue)
+{
+	return test_bit(amdgpu_gfx_me_queue_to_bit(adev, me, pipe, queue),
+			adev->gfx.me.queue_bitmap);
 }
 
 /**
@@ -75,9 +107,9 @@ void amdgpu_gfx_scratch_free(struct amdgpu_device *adev, uint32_t reg)
  * The bitmask of CUs to be disabled in the shader array determined by se and
  * sh is stored in mask[se * max_sh + sh].
  */
-void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se, unsigned max_sh)
+void amdgpu_gfx_parse_disable_cu(unsigned int *mask, unsigned int max_se, unsigned int max_sh)
 {
-	unsigned se, sh, cu;
+	unsigned int se, sh, cu;
 	const char *p;
 
 	memset(mask, 0, sizeof(*mask) * max_se * max_sh);
@@ -89,6 +121,7 @@ void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se, unsigned max_s
 	for (;;) {
 		char *next;
 		int ret = sscanf(p, "%u.%u.%u", &se, &sh, &cu);
+
 		if (ret < 3) {
 			DRM_ERROR("amdgpu: could not parse disable_cu\n");
 			return;
@@ -108,3 +141,2417 @@ void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se, unsigned max_s
 		p = next + 1;
 	}
 }
+
+static bool amdgpu_gfx_is_graphics_multipipe_capable(struct amdgpu_device *adev)
+{
+	return amdgpu_async_gfx_ring && adev->gfx.me.num_pipe_per_me > 1;
+}
+
+static bool amdgpu_gfx_is_compute_multipipe_capable(struct amdgpu_device *adev)
+{
+	if (amdgpu_compute_multipipe != -1) {
+		dev_info(adev->dev, "amdgpu: forcing compute pipe policy %d\n",
+			 amdgpu_compute_multipipe);
+		return amdgpu_compute_multipipe == 1;
+	}
+
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) > IP_VERSION(9, 0, 0))
+		return true;
+
+	/* FIXME: spreading the queues across pipes causes perf regressions
+	 * on POLARIS11 compute workloads */
+	if (adev->asic_type == CHIP_POLARIS11)
+		return false;
+
+	return adev->gfx.mec.num_mec > 1;
+}
+
+bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev,
+						struct amdgpu_ring *ring)
+{
+	int queue = ring->queue;
+	int pipe = ring->pipe;
+
+	/* Policy: use pipe1 queue0 as high priority graphics queue if we
+	 * have more than one gfx pipe.
+	 */
+	if (amdgpu_gfx_is_graphics_multipipe_capable(adev) &&
+	    adev->gfx.num_gfx_rings > 1 && pipe == 1 && queue == 0) {
+		int me = ring->me;
+		int bit;
+
+		bit = amdgpu_gfx_me_queue_to_bit(adev, me, pipe, queue);
+		if (ring == &adev->gfx.gfx_ring[bit])
+			return true;
+	}
+
+	return false;
+}
+
+bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
+					       struct amdgpu_ring *ring)
+{
+	/* Policy: use 1st queue as high priority compute queue if we
+	 * have more than one compute queue.
+	 */
+	if (adev->gfx.num_compute_rings > 1 &&
+	    ring == &adev->gfx.compute_ring[0])
+		return true;
+
+	return false;
+}
+
+void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev)
+{
+	int i, j, queue, pipe;
+	bool multipipe_policy = amdgpu_gfx_is_compute_multipipe_capable(adev);
+	int max_queues_per_mec = min(adev->gfx.mec.num_pipe_per_mec *
+				     adev->gfx.mec.num_queue_per_pipe,
+				     adev->gfx.num_compute_rings);
+	int num_xcc = adev->gfx.xcc_mask ? NUM_XCC(adev->gfx.xcc_mask) : 1;
+
+	if (multipipe_policy) {
+		/* policy: make queues evenly cross all pipes on MEC1 only
+		 * for multiple xcc, just use the original policy for simplicity */
+		for (j = 0; j < num_xcc; j++) {
+			for (i = 0; i < max_queues_per_mec; i++) {
+				pipe = i % adev->gfx.mec.num_pipe_per_mec;
+				queue = (i / adev->gfx.mec.num_pipe_per_mec) %
+					 adev->gfx.mec.num_queue_per_pipe;
+
+				set_bit(pipe * adev->gfx.mec.num_queue_per_pipe + queue,
+					adev->gfx.mec_bitmap[j].queue_bitmap);
+			}
+		}
+	} else {
+		/* policy: amdgpu owns all queues in the given pipe */
+		for (j = 0; j < num_xcc; j++) {
+			for (i = 0; i < max_queues_per_mec; ++i)
+				set_bit(i, adev->gfx.mec_bitmap[j].queue_bitmap);
+		}
+	}
+
+	for (j = 0; j < num_xcc; j++) {
+		dev_dbg(adev->dev, "mec queue bitmap weight=%d\n",
+			bitmap_weight(adev->gfx.mec_bitmap[j].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES));
+	}
+}
+
+void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev)
+{
+	int i, queue, pipe;
+	bool multipipe_policy = amdgpu_gfx_is_graphics_multipipe_capable(adev);
+	int num_queue_per_pipe = 1; /* we only enable 1 KGQ per pipe */
+	int max_queues_per_me = adev->gfx.me.num_pipe_per_me * num_queue_per_pipe;
+
+	if (multipipe_policy) {
+		/* policy: amdgpu owns the first queue per pipe at this stage
+		 * will extend to mulitple queues per pipe later */
+		for (i = 0; i < max_queues_per_me; i++) {
+			pipe = i % adev->gfx.me.num_pipe_per_me;
+			queue = (i / adev->gfx.me.num_pipe_per_me) %
+				num_queue_per_pipe;
+
+			set_bit(pipe * num_queue_per_pipe + queue,
+				adev->gfx.me.queue_bitmap);
+		}
+	} else {
+		for (i = 0; i < max_queues_per_me; ++i)
+			set_bit(i, adev->gfx.me.queue_bitmap);
+	}
+
+	/* update the number of active graphics rings */
+	if (adev->gfx.num_gfx_rings)
+		adev->gfx.num_gfx_rings =
+			bitmap_weight(adev->gfx.me.queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
+}
+
+static int amdgpu_gfx_kiq_acquire(struct amdgpu_device *adev,
+				  struct amdgpu_ring *ring, int xcc_id)
+{
+	int queue_bit;
+	int mec, pipe, queue;
+
+	queue_bit = adev->gfx.mec.num_mec
+		    * adev->gfx.mec.num_pipe_per_mec
+		    * adev->gfx.mec.num_queue_per_pipe;
+
+	while (--queue_bit >= 0) {
+		if (test_bit(queue_bit, adev->gfx.mec_bitmap[xcc_id].queue_bitmap))
+			continue;
+
+		amdgpu_queue_mask_bit_to_mec_queue(adev, queue_bit, &mec, &pipe, &queue);
+
+		/*
+		 * 1. Using pipes 2/3 from MEC 2 seems cause problems.
+		 * 2. It must use queue id 0, because CGPG_IDLE/SAVE/LOAD/RUN
+		 * only can be issued on queue 0.
+		 */
+		if ((mec == 1 && pipe > 1) || queue != 0)
+			continue;
+
+		ring->me = mec + 1;
+		ring->pipe = pipe;
+		ring->queue = queue;
+
+		return 0;
+	}
+
+	dev_err(adev->dev, "Failed to find a queue for KIQ\n");
+	return -EINVAL;
+}
+
+int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_irq_src *irq = &kiq->irq;
+	struct amdgpu_ring *ring = &kiq->ring;
+	int r = 0;
+
+	spin_lock_init(&kiq->ring_lock);
+
+	ring->adev = NULL;
+	ring->ring_obj = NULL;
+	ring->use_doorbell = true;
+	ring->xcc_id = xcc_id;
+	ring->vm_hub = AMDGPU_GFXHUB(xcc_id);
+	ring->doorbell_index =
+		(adev->doorbell_index.kiq +
+		 xcc_id * adev->doorbell_index.xcc_doorbell_range)
+		<< 1;
+
+	r = amdgpu_gfx_kiq_acquire(adev, ring, xcc_id);
+	if (r)
+		return r;
+
+	ring->eop_gpu_addr = kiq->eop_gpu_addr;
+	ring->no_scheduler = true;
+	snprintf(ring->name, sizeof(ring->name), "kiq_%hhu.%hhu.%hhu.%hhu",
+		 (unsigned char)xcc_id, (unsigned char)ring->me,
+		 (unsigned char)ring->pipe, (unsigned char)ring->queue);
+	r = amdgpu_ring_init(adev, ring, 1024, irq, AMDGPU_CP_KIQ_IRQ_DRIVER0,
+			     AMDGPU_RING_PRIO_DEFAULT, NULL);
+	if (r)
+		dev_warn(adev->dev, "(%d) failed to init kiq ring\n", r);
+
+	return r;
+}
+
+void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_fini(ring);
+}
+
+void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+
+	amdgpu_bo_free_kernel(&kiq->eop_obj, &kiq->eop_gpu_addr, NULL);
+}
+
+int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
+			unsigned int hpd_size, int xcc_id)
+{
+	int r;
+	u32 *hpd;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+
+	r = amdgpu_bo_create_kernel(adev, hpd_size, PAGE_SIZE,
+				    AMDGPU_GEM_DOMAIN_GTT, &kiq->eop_obj,
+				    &kiq->eop_gpu_addr, (void **)&hpd);
+	if (r) {
+		dev_warn(adev->dev, "failed to create KIQ bo (%d).\n", r);
+		return r;
+	}
+
+	memset(hpd, 0, hpd_size);
+
+	r = amdgpu_bo_reserve(kiq->eop_obj, true);
+	if (unlikely(r != 0))
+		dev_warn(adev->dev, "(%d) reserve kiq eop bo failed\n", r);
+	amdgpu_bo_kunmap(kiq->eop_obj);
+	amdgpu_bo_unreserve(kiq->eop_obj);
+
+	return 0;
+}
+
+/* create MQD for each compute/gfx queue */
+int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
+			   unsigned int mqd_size, int xcc_id)
+{
+	int r, i, j;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *ring = &kiq->ring;
+	u32 domain = AMDGPU_GEM_DOMAIN_GTT;
+
+#if !defined(CONFIG_ARM) && !defined(CONFIG_ARM64)
+	/* Only enable on gfx10 and 11 for now to avoid changing behavior on older chips */
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(10, 0, 0))
+		domain |= AMDGPU_GEM_DOMAIN_VRAM;
+#endif
+
+	/* create MQD for KIQ */
+	if (!adev->enable_mes_kiq && !ring->mqd_obj) {
+		/* originaly the KIQ MQD is put in GTT domain, but for SRIOV VRAM domain is a must
+		 * otherwise hypervisor trigger SAVE_VF fail after driver unloaded which mean MQD
+		 * deallocated and gart_unbind, to strict diverage we decide to use VRAM domain for
+		 * KIQ MQD no matter SRIOV or Bare-metal
+		 */
+		r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE,
+					    AMDGPU_GEM_DOMAIN_VRAM |
+					    AMDGPU_GEM_DOMAIN_GTT,
+					    &ring->mqd_obj,
+					    &ring->mqd_gpu_addr,
+					    &ring->mqd_ptr);
+		if (r) {
+			dev_warn(adev->dev, "failed to create ring mqd ob (%d)", r);
+			return r;
+		}
+
+		/* prepare MQD backup */
+		kiq->mqd_backup = kzalloc(mqd_size, GFP_KERNEL);
+		if (!kiq->mqd_backup) {
+			dev_warn(adev->dev,
+				 "no memory to create MQD backup for ring %s\n", ring->name);
+			return -ENOMEM;
+		}
+	}
+
+	if (adev->asic_type >= CHIP_NAVI10 && amdgpu_async_gfx_ring) {
+		/* create MQD for each KGQ */
+		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+			ring = &adev->gfx.gfx_ring[i];
+			if (!ring->mqd_obj) {
+				r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE,
+							    domain, &ring->mqd_obj,
+							    &ring->mqd_gpu_addr, &ring->mqd_ptr);
+				if (r) {
+					dev_warn(adev->dev, "failed to create ring mqd bo (%d)", r);
+					return r;
+				}
+
+				ring->mqd_size = mqd_size;
+				/* prepare MQD backup */
+				adev->gfx.me.mqd_backup[i] = kzalloc(mqd_size, GFP_KERNEL);
+				if (!adev->gfx.me.mqd_backup[i]) {
+					dev_warn(adev->dev, "no memory to create MQD backup for ring %s\n", ring->name);
+					return -ENOMEM;
+				}
+			}
+		}
+	}
+
+	/* create MQD for each KCQ */
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		j = i + xcc_id * adev->gfx.num_compute_rings;
+		ring = &adev->gfx.compute_ring[j];
+		if (!ring->mqd_obj) {
+			r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE,
+						    domain, &ring->mqd_obj,
+						    &ring->mqd_gpu_addr, &ring->mqd_ptr);
+			if (r) {
+				dev_warn(adev->dev, "failed to create ring mqd bo (%d)", r);
+				return r;
+			}
+
+			ring->mqd_size = mqd_size;
+			/* prepare MQD backup */
+			adev->gfx.mec.mqd_backup[j] = kzalloc(mqd_size, GFP_KERNEL);
+			if (!adev->gfx.mec.mqd_backup[j]) {
+				dev_warn(adev->dev, "no memory to create MQD backup for ring %s\n", ring->name);
+				return -ENOMEM;
+			}
+		}
+	}
+
+	return 0;
+}
+
+void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_ring *ring = NULL;
+	int i, j;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+
+	if (adev->asic_type >= CHIP_NAVI10 && amdgpu_async_gfx_ring) {
+		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+			ring = &adev->gfx.gfx_ring[i];
+			kfree(adev->gfx.me.mqd_backup[i]);
+			amdgpu_bo_free_kernel(&ring->mqd_obj,
+					      &ring->mqd_gpu_addr,
+					      &ring->mqd_ptr);
+		}
+	}
+
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		j = i + xcc_id * adev->gfx.num_compute_rings;
+		ring = &adev->gfx.compute_ring[j];
+		kfree(adev->gfx.mec.mqd_backup[j]);
+		amdgpu_bo_free_kernel(&ring->mqd_obj,
+				      &ring->mqd_gpu_addr,
+				      &ring->mqd_ptr);
+	}
+
+	ring = &kiq->ring;
+	kfree(kiq->mqd_backup);
+	amdgpu_bo_free_kernel(&ring->mqd_obj,
+			      &ring->mqd_gpu_addr,
+			      &ring->mqd_ptr);
+}
+
+int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *kiq_ring = &kiq->ring;
+	int i, r = 0;
+	int j;
+
+	if (adev->enable_mes) {
+		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+			j = i + xcc_id * adev->gfx.num_compute_rings;
+			amdgpu_mes_unmap_legacy_queue(adev,
+						   &adev->gfx.compute_ring[j],
+						   RESET_QUEUES, 0, 0);
+		}
+		return 0;
+	}
+
+	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
+		return -EINVAL;
+
+	if (!kiq_ring->sched.ready || amdgpu_in_reset(adev))
+		return 0;
+
+	spin_lock(&kiq->ring_lock);
+	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size *
+					adev->gfx.num_compute_rings)) {
+		spin_unlock(&kiq->ring_lock);
+		return -ENOMEM;
+	}
+
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		j = i + xcc_id * adev->gfx.num_compute_rings;
+		kiq->pmf->kiq_unmap_queues(kiq_ring,
+					   &adev->gfx.compute_ring[j],
+					   RESET_QUEUES, 0, 0);
+	}
+	/* Submit unmap queue packet */
+	amdgpu_ring_commit(kiq_ring);
+	/*
+	 * Ring test will do a basic scratch register change check. Just run
+	 * this to ensure that unmap queues that is submitted before got
+	 * processed successfully before returning.
+	 */
+	r = amdgpu_ring_test_helper(kiq_ring);
+
+	spin_unlock(&kiq->ring_lock);
+
+	return r;
+}
+
+int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *kiq_ring = &kiq->ring;
+	int i, r = 0;
+	int j;
+
+	if (adev->enable_mes) {
+		if (amdgpu_gfx_is_master_xcc(adev, xcc_id)) {
+			for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+				j = i + xcc_id * adev->gfx.num_gfx_rings;
+				amdgpu_mes_unmap_legacy_queue(adev,
+						      &adev->gfx.gfx_ring[j],
+						      PREEMPT_QUEUES, 0, 0);
+			}
+		}
+		return 0;
+	}
+
+	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
+		return -EINVAL;
+
+	if (!adev->gfx.kiq[0].ring.sched.ready || amdgpu_in_reset(adev))
+		return 0;
+
+	if (amdgpu_gfx_is_master_xcc(adev, xcc_id)) {
+		spin_lock(&kiq->ring_lock);
+		if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size *
+						adev->gfx.num_gfx_rings)) {
+			spin_unlock(&kiq->ring_lock);
+			return -ENOMEM;
+		}
+
+		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+			j = i + xcc_id * adev->gfx.num_gfx_rings;
+			kiq->pmf->kiq_unmap_queues(kiq_ring,
+						   &adev->gfx.gfx_ring[j],
+						   PREEMPT_QUEUES, 0, 0);
+		}
+		/* Submit unmap queue packet */
+		amdgpu_ring_commit(kiq_ring);
+
+		/*
+		 * Ring test will do a basic scratch register change check.
+		 * Just run this to ensure that unmap queues that is submitted
+		 * before got processed successfully before returning.
+		 */
+		r = amdgpu_ring_test_helper(kiq_ring);
+		spin_unlock(&kiq->ring_lock);
+	}
+
+	return r;
+}
+
+int amdgpu_queue_mask_bit_to_set_resource_bit(struct amdgpu_device *adev,
+					int queue_bit)
+{
+	int mec, pipe, queue;
+	int set_resource_bit = 0;
+
+	amdgpu_queue_mask_bit_to_mec_queue(adev, queue_bit, &mec, &pipe, &queue);
+
+	set_resource_bit = mec * 4 * 8 + pipe * 8 + queue;
+
+	return set_resource_bit;
+}
+
+static int amdgpu_gfx_mes_enable_kcq(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *kiq_ring = &kiq->ring;
+	uint64_t queue_mask = ~0ULL;
+	int r, i, j;
+
+	amdgpu_device_flush_hdp(adev, NULL);
+
+	if (!adev->enable_uni_mes) {
+		spin_lock(&kiq->ring_lock);
+		r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->set_resources_size);
+		if (r) {
+			dev_err(adev->dev, "Failed to lock KIQ (%d).\n", r);
+			spin_unlock(&kiq->ring_lock);
+			return r;
+		}
+
+		kiq->pmf->kiq_set_resources(kiq_ring, queue_mask);
+		r = amdgpu_ring_test_helper(kiq_ring);
+		spin_unlock(&kiq->ring_lock);
+		if (r)
+			dev_err(adev->dev, "KIQ failed to set resources\n");
+	}
+
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		j = i + xcc_id * adev->gfx.num_compute_rings;
+		r = amdgpu_mes_map_legacy_queue(adev,
+						&adev->gfx.compute_ring[j]);
+		if (r) {
+			dev_err(adev->dev, "failed to map compute queue\n");
+			return r;
+		}
+	}
+
+	return 0;
+}
+
+int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *kiq_ring = &kiq->ring;
+	uint64_t queue_mask = 0;
+	int r, i, j;
+
+	if (adev->mes.enable_legacy_queue_map)
+		return amdgpu_gfx_mes_enable_kcq(adev, xcc_id);
+
+	if (!kiq->pmf || !kiq->pmf->kiq_map_queues || !kiq->pmf->kiq_set_resources)
+		return -EINVAL;
+
+	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
+		if (!test_bit(i, adev->gfx.mec_bitmap[xcc_id].queue_bitmap))
+			continue;
+
+		/* This situation may be hit in the future if a new HW
+		 * generation exposes more than 64 queues. If so, the
+		 * definition of queue_mask needs updating */
+		if (WARN_ON(i > (sizeof(queue_mask)*8))) {
+			dev_err(adev->dev, "Invalid KCQ enabled: %d\n", i);
+			break;
+		}
+
+		queue_mask |= (1ull << amdgpu_queue_mask_bit_to_set_resource_bit(adev, i));
+	}
+
+	amdgpu_device_flush_hdp(adev, NULL);
+
+	dev_info(adev->dev, "kiq ring mec %d pipe %d q %d\n", kiq_ring->me,
+		 kiq_ring->pipe, kiq_ring->queue);
+
+	spin_lock(&kiq->ring_lock);
+	r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size *
+					adev->gfx.num_compute_rings +
+					kiq->pmf->set_resources_size);
+	if (r) {
+		dev_err(adev->dev, "Failed to lock KIQ (%d).\n", r);
+		spin_unlock(&kiq->ring_lock);
+		return r;
+	}
+
+	kiq->pmf->kiq_set_resources(kiq_ring, queue_mask);
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		j = i + xcc_id * adev->gfx.num_compute_rings;
+		kiq->pmf->kiq_map_queues(kiq_ring,
+					 &adev->gfx.compute_ring[j]);
+	}
+	/* Submit map queue packet */
+	amdgpu_ring_commit(kiq_ring);
+	/*
+	 * Ring test will do a basic scratch register change check. Just run
+	 * this to ensure that map queues that is submitted before got
+	 * processed successfully before returning.
+	 */
+	r = amdgpu_ring_test_helper(kiq_ring);
+	spin_unlock(&kiq->ring_lock);
+	if (r)
+		dev_err(adev->dev, "KCQ enable failed\n");
+
+	return r;
+}
+
+int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *kiq_ring = &kiq->ring;
+	int r, i, j;
+
+	if (!kiq->pmf || !kiq->pmf->kiq_map_queues)
+		return -EINVAL;
+
+	amdgpu_device_flush_hdp(adev, NULL);
+
+	if (adev->mes.enable_legacy_queue_map) {
+		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+			j = i + xcc_id * adev->gfx.num_gfx_rings;
+			r = amdgpu_mes_map_legacy_queue(adev,
+							&adev->gfx.gfx_ring[j]);
+			if (r) {
+				dev_err(adev->dev, "failed to map gfx queue\n");
+				return r;
+			}
+		}
+
+		return 0;
+	}
+
+	spin_lock(&kiq->ring_lock);
+	/* No need to map kcq on the slave */
+	if (amdgpu_gfx_is_master_xcc(adev, xcc_id)) {
+		r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size *
+						adev->gfx.num_gfx_rings);
+		if (r) {
+			dev_err(adev->dev, "Failed to lock KIQ (%d).\n", r);
+			spin_unlock(&kiq->ring_lock);
+			return r;
+		}
+
+		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+			j = i + xcc_id * adev->gfx.num_gfx_rings;
+			kiq->pmf->kiq_map_queues(kiq_ring,
+						 &adev->gfx.gfx_ring[j]);
+		}
+	}
+	/* Submit map queue packet */
+	amdgpu_ring_commit(kiq_ring);
+	/*
+	 * Ring test will do a basic scratch register change check. Just run
+	 * this to ensure that map queues that is submitted before got
+	 * processed successfully before returning.
+	 */
+	r = amdgpu_ring_test_helper(kiq_ring);
+	spin_unlock(&kiq->ring_lock);
+	if (r)
+		dev_err(adev->dev, "KGQ enable failed\n");
+
+	return r;
+}
+
+static void amdgpu_gfx_do_off_ctrl(struct amdgpu_device *adev, bool enable,
+				   bool no_delay)
+{
+	unsigned long delay = GFX_OFF_DELAY_ENABLE;
+
+	if (!(adev->pm.pp_feature & PP_GFXOFF_MASK))
+		return;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	if (enable) {
+		/* If the count is already 0, it means there's an imbalance bug somewhere.
+		 * Note that the bug may be in a different caller than the one which triggers the
+		 * WARN_ON_ONCE.
+		 */
+		if (WARN_ON_ONCE(adev->gfx.gfx_off_req_count == 0))
+			goto unlock;
+
+		adev->gfx.gfx_off_req_count--;
+
+		if (adev->gfx.gfx_off_req_count == 0 &&
+		    !adev->gfx.gfx_off_state) {
+			/* If going to s2idle, no need to wait */
+			if (no_delay) {
+				if (!amdgpu_dpm_set_powergating_by_smu(adev,
+						AMD_IP_BLOCK_TYPE_GFX, true, 0))
+					adev->gfx.gfx_off_state = true;
+			} else {
+				schedule_delayed_work(&adev->gfx.gfx_off_delay_work,
+					      delay);
+			}
+		}
+	} else {
+		if (adev->gfx.gfx_off_req_count == 0) {
+			cancel_delayed_work_sync(&adev->gfx.gfx_off_delay_work);
+
+			if (adev->gfx.gfx_off_state &&
+			    !amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, false, 0)) {
+				adev->gfx.gfx_off_state = false;
+
+				if (adev->gfx.funcs->init_spm_golden) {
+					dev_dbg(adev->dev,
+						"GFXOFF is disabled, re-init SPM golden settings\n");
+					amdgpu_gfx_init_spm_golden(adev);
+				}
+			}
+		}
+
+		adev->gfx.gfx_off_req_count++;
+	}
+
+unlock:
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+}
+
+/* amdgpu_gfx_off_ctrl - Handle gfx off feature enable/disable
+ *
+ * @adev: amdgpu_device pointer
+ * @bool enable true: enable gfx off feature, false: disable gfx off feature
+ *
+ * 1. gfx off feature will be enabled by gfx ip after gfx cg pg enabled.
+ * 2. other client can send request to disable gfx off feature, the request should be honored.
+ * 3. other client can cancel their request of disable gfx off feature
+ * 4. other client should not send request to enable gfx off feature before disable gfx off feature.
+ *
+ * gfx off allow will be delayed by GFX_OFF_DELAY_ENABLE ms.
+ */
+void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable)
+{
+	/* If going to s2idle, no need to wait */
+	bool no_delay = adev->in_s0ix ? true : false;
+
+	amdgpu_gfx_do_off_ctrl(adev, enable, no_delay);
+}
+
+/* amdgpu_gfx_off_ctrl_immediate - Handle gfx off feature enable/disable
+ *
+ * @adev: amdgpu_device pointer
+ * @bool enable true: enable gfx off feature, false: disable gfx off feature
+ *
+ * 1. gfx off feature will be enabled by gfx ip after gfx cg pg enabled.
+ * 2. other client can send request to disable gfx off feature, the request should be honored.
+ * 3. other client can cancel their request of disable gfx off feature
+ * 4. other client should not send request to enable gfx off feature before disable gfx off feature.
+ *
+ * gfx off allow will be issued immediately.
+ */
+void amdgpu_gfx_off_ctrl_immediate(struct amdgpu_device *adev, bool enable)
+{
+	amdgpu_gfx_do_off_ctrl(adev, enable, true);
+}
+
+int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value)
+{
+	int r = 0;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	r = amdgpu_dpm_set_residency_gfxoff(adev, value);
+
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+
+	return r;
+}
+
+int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *value)
+{
+	int r = 0;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	r = amdgpu_dpm_get_residency_gfxoff(adev, value);
+
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+
+	return r;
+}
+
+int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value)
+{
+	int r = 0;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	r = amdgpu_dpm_get_entrycount_gfxoff(adev, value);
+
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+
+	return r;
+}
+
+int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value)
+{
+
+	int r = 0;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	r = amdgpu_dpm_get_status_gfxoff(adev, value);
+
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+
+	return r;
+}
+
+int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
+{
+	int r;
+
+	if (amdgpu_ras_is_supported(adev, ras_block->block)) {
+		if (!amdgpu_persistent_edc_harvesting_supported(adev)) {
+			r = amdgpu_ras_reset_error_status(adev, AMDGPU_RAS_BLOCK__GFX);
+			if (r)
+				return r;
+		}
+
+		r = amdgpu_ras_block_late_init(adev, ras_block);
+		if (r)
+			return r;
+
+		if (amdgpu_sriov_vf(adev))
+			return r;
+
+		if (adev->gfx.cp_ecc_error_irq.funcs) {
+			r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
+			if (r)
+				goto late_fini;
+		}
+	} else {
+		amdgpu_ras_feature_enable_on_boot(adev, ras_block, 0);
+	}
+
+	return 0;
+late_fini:
+	amdgpu_ras_block_late_fini(adev, ras_block);
+	return r;
+}
+
+int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err = 0;
+	struct amdgpu_gfx_ras *ras = NULL;
+
+	/* adev->gfx.ras is NULL, which means gfx does not
+	 * support ras function, then do nothing here.
+	 */
+	if (!adev->gfx.ras)
+		return 0;
+
+	ras = adev->gfx.ras;
+
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register gfx ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "gfx");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__GFX;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+	adev->gfx.ras_if = &ras->ras_block.ras_comm;
+
+	/* If not define special ras_late_init function, use gfx default ras_late_init */
+	if (!ras->ras_block.ras_late_init)
+		ras->ras_block.ras_late_init = amdgpu_gfx_ras_late_init;
+
+	/* If not defined special ras_cb function, use default ras_cb */
+	if (!ras->ras_block.ras_cb)
+		ras->ras_block.ras_cb = amdgpu_gfx_process_ras_data_cb;
+
+	return 0;
+}
+
+int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev,
+						struct amdgpu_iv_entry *entry)
+{
+	if (adev->gfx.ras && adev->gfx.ras->poison_consumption_handler)
+		return adev->gfx.ras->poison_consumption_handler(adev, entry);
+
+	return 0;
+}
+
+int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
+		void *err_data,
+		struct amdgpu_iv_entry *entry)
+{
+	/* TODO ue will trigger an interrupt.
+	 *
+	 * When “Full RAS” is enabled, the per-IP interrupt sources should
+	 * be disabled and the driver should only look for the aggregated
+	 * interrupt via sync flood
+	 */
+	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) {
+		kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
+		if (adev->gfx.ras && adev->gfx.ras->ras_block.hw_ops &&
+		    adev->gfx.ras->ras_block.hw_ops->query_ras_error_count)
+			adev->gfx.ras->ras_block.hw_ops->query_ras_error_count(adev, err_data);
+		amdgpu_ras_reset_gpu(adev);
+	}
+	return AMDGPU_RAS_SUCCESS;
+}
+
+int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
+				  struct amdgpu_irq_src *source,
+				  struct amdgpu_iv_entry *entry)
+{
+	struct ras_common_if *ras_if = adev->gfx.ras_if;
+	struct ras_dispatch_if ih_data = {
+		.entry = entry,
+	};
+
+	if (!ras_if)
+		return 0;
+
+	ih_data.head = *ras_if;
+
+	dev_err(adev->dev, "CP ECC ERROR IRQ\n");
+	amdgpu_ras_interrupt_dispatch(adev, &ih_data);
+	return 0;
+}
+
+void amdgpu_gfx_ras_error_func(struct amdgpu_device *adev,
+		void *ras_error_status,
+		void (*func)(struct amdgpu_device *adev, void *ras_error_status,
+				int xcc_id))
+{
+	int i;
+	int num_xcc = adev->gfx.xcc_mask ? NUM_XCC(adev->gfx.xcc_mask) : 1;
+	uint32_t xcc_mask = GENMASK(num_xcc - 1, 0);
+	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
+
+	if (err_data) {
+		err_data->ue_count = 0;
+		err_data->ce_count = 0;
+	}
+
+	for_each_inst(i, xcc_mask)
+		func(adev, ras_error_status, i);
+}
+
+uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg, uint32_t xcc_id)
+{
+	signed long r, cnt = 0;
+	unsigned long flags;
+	uint32_t seq, reg_val_offs = 0, value = 0;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *ring = &kiq->ring;
+
+	if (amdgpu_device_skip_hw_access(adev))
+		return 0;
+
+	if (adev->mes.ring[0].sched.ready)
+		return amdgpu_mes_rreg(adev, reg);
+
+	BUG_ON(!ring->funcs->emit_rreg);
+
+	spin_lock_irqsave(&kiq->ring_lock, flags);
+	if (amdgpu_device_wb_get(adev, &reg_val_offs)) {
+		pr_err("critical bug! too many kiq readers\n");
+		goto failed_unlock;
+	}
+	r = amdgpu_ring_alloc(ring, 32);
+	if (r)
+		goto failed_unlock;
+
+	amdgpu_ring_emit_rreg(ring, reg, reg_val_offs);
+	r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
+	if (r)
+		goto failed_undo;
+
+	amdgpu_ring_commit(ring);
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+
+	r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+
+	/* don't wait anymore for gpu reset case because this way may
+	 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
+	 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
+	 * never return if we keep waiting in virt_kiq_rreg, which cause
+	 * gpu_recover() hang there.
+	 *
+	 * also don't wait anymore for IRQ context
+	 * */
+	if (r < 1 && (amdgpu_in_reset(adev) || in_interrupt()))
+		goto failed_kiq_read;
+
+	might_sleep();
+	while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
+		if (amdgpu_in_reset(adev))
+			goto failed_kiq_read;
+
+		msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
+		r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+	}
+
+	if (cnt > MAX_KIQ_REG_TRY)
+		goto failed_kiq_read;
+
+	mb();
+	value = adev->wb.wb[reg_val_offs];
+	amdgpu_device_wb_free(adev, reg_val_offs);
+	return value;
+
+failed_undo:
+	amdgpu_ring_undo(ring);
+failed_unlock:
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+failed_kiq_read:
+	if (reg_val_offs)
+		amdgpu_device_wb_free(adev, reg_val_offs);
+	dev_err(adev->dev, "failed to read reg:%x\n", reg);
+	return ~0;
+}
+
+void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, uint32_t xcc_id)
+{
+	signed long r, cnt = 0;
+	unsigned long flags;
+	uint32_t seq;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
+	struct amdgpu_ring *ring = &kiq->ring;
+
+	BUG_ON(!ring->funcs->emit_wreg);
+
+	if (amdgpu_device_skip_hw_access(adev))
+		return;
+
+	if (adev->mes.ring[0].sched.ready) {
+		amdgpu_mes_wreg(adev, reg, v);
+		return;
+	}
+
+	spin_lock_irqsave(&kiq->ring_lock, flags);
+	r = amdgpu_ring_alloc(ring, 32);
+	if (r)
+		goto failed_unlock;
+
+	amdgpu_ring_emit_wreg(ring, reg, v);
+	r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
+	if (r)
+		goto failed_undo;
+
+	amdgpu_ring_commit(ring);
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+
+	r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+
+	/* don't wait anymore for gpu reset case because this way may
+	 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
+	 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
+	 * never return if we keep waiting in virt_kiq_rreg, which cause
+	 * gpu_recover() hang there.
+	 *
+	 * also don't wait anymore for IRQ context
+	 * */
+	if (r < 1 && (amdgpu_in_reset(adev) || in_interrupt()))
+		goto failed_kiq_write;
+
+	might_sleep();
+	while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
+		if (amdgpu_in_reset(adev))
+			goto failed_kiq_write;
+
+		msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
+		r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+	}
+
+	if (cnt > MAX_KIQ_REG_TRY)
+		goto failed_kiq_write;
+
+	return;
+
+failed_undo:
+	amdgpu_ring_undo(ring);
+failed_unlock:
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+failed_kiq_write:
+	dev_err(adev->dev, "failed to write reg:%x\n", reg);
+}
+
+int amdgpu_kiq_hdp_flush(struct amdgpu_device *adev)
+{
+	signed long r, cnt = 0;
+	unsigned long flags;
+	uint32_t seq;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
+	struct amdgpu_ring *ring = &kiq->ring;
+
+	if (amdgpu_device_skip_hw_access(adev))
+		return 0;
+
+	if (adev->enable_mes_kiq && adev->mes.ring[0].sched.ready)
+		return amdgpu_mes_hdp_flush(adev);
+
+	if (!ring->funcs->emit_hdp_flush) {
+		return -EOPNOTSUPP;
+	}
+
+	spin_lock_irqsave(&kiq->ring_lock, flags);
+	r = amdgpu_ring_alloc(ring, 32);
+	if (r)
+		goto failed_unlock;
+
+	amdgpu_ring_emit_hdp_flush(ring);
+	r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
+	if (r)
+		goto failed_undo;
+
+	amdgpu_ring_commit(ring);
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+
+	r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+
+	/* don't wait anymore for gpu reset case because this way may
+	 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
+	 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
+	 * never return if we keep waiting in virt_kiq_rreg, which cause
+	 * gpu_recover() hang there.
+	 *
+	 * also don't wait anymore for IRQ context
+	 * */
+	if (r < 1 && (amdgpu_in_reset(adev) || in_interrupt()))
+		goto failed_kiq_hdp_flush;
+
+	might_sleep();
+	while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
+		if (amdgpu_in_reset(adev))
+			goto failed_kiq_hdp_flush;
+
+		msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
+		r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+	}
+
+	if (cnt > MAX_KIQ_REG_TRY) {
+		dev_err(adev->dev, "failed to flush HDP via KIQ timeout\n");
+		return -ETIMEDOUT;
+	}
+
+	return 0;
+
+failed_undo:
+	amdgpu_ring_undo(ring);
+failed_unlock:
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+failed_kiq_hdp_flush:
+	dev_err(adev->dev, "failed to flush HDP via KIQ\n");
+	return r < 0 ? r : -EIO;
+}
+
+int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev)
+{
+	if (amdgpu_num_kcq == -1) {
+		return 8;
+	} else if (amdgpu_num_kcq > 8 || amdgpu_num_kcq < 0) {
+		dev_warn(adev->dev, "set kernel compute queue number to 8 due to invalid parameter provided by user\n");
+		return 8;
+	}
+	return amdgpu_num_kcq;
+}
+
+void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev,
+				  uint32_t ucode_id)
+{
+	const struct gfx_firmware_header_v1_0 *cp_hdr;
+	const struct gfx_firmware_header_v2_0 *cp_hdr_v2_0;
+	struct amdgpu_firmware_info *info = NULL;
+	const struct firmware *ucode_fw;
+	unsigned int fw_size;
+
+	switch (ucode_id) {
+	case AMDGPU_UCODE_ID_CP_PFP:
+		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
+			adev->gfx.pfp_fw->data;
+		adev->gfx.pfp_fw_version =
+			le32_to_cpu(cp_hdr->header.ucode_version);
+		adev->gfx.pfp_feature_version =
+			le32_to_cpu(cp_hdr->ucode_feature_version);
+		ucode_fw = adev->gfx.pfp_fw;
+		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_PFP:
+		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)
+			adev->gfx.pfp_fw->data;
+		adev->gfx.pfp_fw_version =
+			le32_to_cpu(cp_hdr_v2_0->header.ucode_version);
+		adev->gfx.pfp_feature_version =
+			le32_to_cpu(cp_hdr_v2_0->ucode_feature_version);
+		ucode_fw = adev->gfx.pfp_fw;
+		fw_size = le32_to_cpu(cp_hdr_v2_0->ucode_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_PFP_P0_STACK:
+	case AMDGPU_UCODE_ID_CP_RS64_PFP_P1_STACK:
+		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)
+			adev->gfx.pfp_fw->data;
+		ucode_fw = adev->gfx.pfp_fw;
+		fw_size = le32_to_cpu(cp_hdr_v2_0->data_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_ME:
+		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
+			adev->gfx.me_fw->data;
+		adev->gfx.me_fw_version =
+			le32_to_cpu(cp_hdr->header.ucode_version);
+		adev->gfx.me_feature_version =
+			le32_to_cpu(cp_hdr->ucode_feature_version);
+		ucode_fw = adev->gfx.me_fw;
+		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_ME:
+		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)
+			adev->gfx.me_fw->data;
+		adev->gfx.me_fw_version =
+			le32_to_cpu(cp_hdr_v2_0->header.ucode_version);
+		adev->gfx.me_feature_version =
+			le32_to_cpu(cp_hdr_v2_0->ucode_feature_version);
+		ucode_fw = adev->gfx.me_fw;
+		fw_size = le32_to_cpu(cp_hdr_v2_0->ucode_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_ME_P0_STACK:
+	case AMDGPU_UCODE_ID_CP_RS64_ME_P1_STACK:
+		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)
+			adev->gfx.me_fw->data;
+		ucode_fw = adev->gfx.me_fw;
+		fw_size = le32_to_cpu(cp_hdr_v2_0->data_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_CE:
+		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
+			adev->gfx.ce_fw->data;
+		adev->gfx.ce_fw_version =
+			le32_to_cpu(cp_hdr->header.ucode_version);
+		adev->gfx.ce_feature_version =
+			le32_to_cpu(cp_hdr->ucode_feature_version);
+		ucode_fw = adev->gfx.ce_fw;
+		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC1:
+		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
+			adev->gfx.mec_fw->data;
+		adev->gfx.mec_fw_version =
+			le32_to_cpu(cp_hdr->header.ucode_version);
+		adev->gfx.mec_feature_version =
+			le32_to_cpu(cp_hdr->ucode_feature_version);
+		ucode_fw = adev->gfx.mec_fw;
+		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes) -
+			  le32_to_cpu(cp_hdr->jt_size) * 4;
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC1_JT:
+		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
+			adev->gfx.mec_fw->data;
+		ucode_fw = adev->gfx.mec_fw;
+		fw_size = le32_to_cpu(cp_hdr->jt_size) * 4;
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC2:
+		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
+			adev->gfx.mec2_fw->data;
+		adev->gfx.mec2_fw_version =
+			le32_to_cpu(cp_hdr->header.ucode_version);
+		adev->gfx.mec2_feature_version =
+			le32_to_cpu(cp_hdr->ucode_feature_version);
+		ucode_fw = adev->gfx.mec2_fw;
+		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes) -
+			  le32_to_cpu(cp_hdr->jt_size) * 4;
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC2_JT:
+		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
+			adev->gfx.mec2_fw->data;
+		ucode_fw = adev->gfx.mec2_fw;
+		fw_size = le32_to_cpu(cp_hdr->jt_size) * 4;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_MEC:
+		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)
+			adev->gfx.mec_fw->data;
+		adev->gfx.mec_fw_version =
+			le32_to_cpu(cp_hdr_v2_0->header.ucode_version);
+		adev->gfx.mec_feature_version =
+			le32_to_cpu(cp_hdr_v2_0->ucode_feature_version);
+		ucode_fw = adev->gfx.mec_fw;
+		fw_size = le32_to_cpu(cp_hdr_v2_0->ucode_size_bytes);
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P0_STACK:
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P1_STACK:
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P2_STACK:
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P3_STACK:
+		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)
+			adev->gfx.mec_fw->data;
+		ucode_fw = adev->gfx.mec_fw;
+		fw_size = le32_to_cpu(cp_hdr_v2_0->data_size_bytes);
+		break;
+	default:
+		dev_err(adev->dev, "Invalid ucode id %u\n", ucode_id);
+		return;
+	}
+
+	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+		info = &adev->firmware.ucode[ucode_id];
+		info->ucode_id = ucode_id;
+		info->fw = ucode_fw;
+		adev->firmware.fw_size += ALIGN(fw_size, PAGE_SIZE);
+	}
+}
+
+bool amdgpu_gfx_is_master_xcc(struct amdgpu_device *adev, int xcc_id)
+{
+	return !(xcc_id % (adev->gfx.num_xcc_per_xcp ?
+			adev->gfx.num_xcc_per_xcp : 1));
+}
+
+static ssize_t amdgpu_gfx_get_current_compute_partition(struct device *dev,
+						struct device_attribute *addr,
+						char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	int mode;
+
+	/* Only minimal precaution taken to reject requests while in reset.*/
+	if (amdgpu_in_reset(adev))
+		return -EPERM;
+
+	mode = amdgpu_xcp_query_partition_mode(adev->xcp_mgr,
+					       AMDGPU_XCP_FL_NONE);
+
+	return sysfs_emit(buf, "%s\n", amdgpu_gfx_compute_mode_desc(mode));
+}
+
+static ssize_t amdgpu_gfx_set_compute_partition(struct device *dev,
+						struct device_attribute *addr,
+						const char *buf, size_t count)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	enum amdgpu_gfx_partition mode;
+	int ret = 0, num_xcc;
+
+	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
+	if (num_xcc % 2 != 0)
+		return -EINVAL;
+
+	if (!strncasecmp("SPX", buf, strlen("SPX"))) {
+		mode = AMDGPU_SPX_PARTITION_MODE;
+	} else if (!strncasecmp("DPX", buf, strlen("DPX"))) {
+		/*
+		 * DPX mode needs AIDs to be in multiple of 2.
+		 * Each AID connects 2 XCCs.
+		 */
+		if (num_xcc%4)
+			return -EINVAL;
+		mode = AMDGPU_DPX_PARTITION_MODE;
+	} else if (!strncasecmp("TPX", buf, strlen("TPX"))) {
+		if (num_xcc != 6)
+			return -EINVAL;
+		mode = AMDGPU_TPX_PARTITION_MODE;
+	} else if (!strncasecmp("QPX", buf, strlen("QPX"))) {
+		if (num_xcc != 8)
+			return -EINVAL;
+		mode = AMDGPU_QPX_PARTITION_MODE;
+	} else if (!strncasecmp("CPX", buf, strlen("CPX"))) {
+		mode = AMDGPU_CPX_PARTITION_MODE;
+	} else {
+		return -EINVAL;
+	}
+
+	/* Don't allow a switch while under reset */
+	if (!down_read_trylock(&adev->reset_domain->sem))
+		return -EPERM;
+
+	ret = amdgpu_xcp_switch_partition_mode(adev->xcp_mgr, mode);
+
+	up_read(&adev->reset_domain->sem);
+
+	if (ret)
+		return ret;
+
+	return count;
+}
+
+static const char *xcp_desc[] = {
+	[AMDGPU_SPX_PARTITION_MODE] = "SPX",
+	[AMDGPU_DPX_PARTITION_MODE] = "DPX",
+	[AMDGPU_TPX_PARTITION_MODE] = "TPX",
+	[AMDGPU_QPX_PARTITION_MODE] = "QPX",
+	[AMDGPU_CPX_PARTITION_MODE] = "CPX",
+};
+
+static ssize_t amdgpu_gfx_get_available_compute_partition(struct device *dev,
+						struct device_attribute *addr,
+						char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct amdgpu_xcp_mgr *xcp_mgr = adev->xcp_mgr;
+	int size = 0, mode;
+	char *sep = "";
+
+	if (!xcp_mgr || !xcp_mgr->avail_xcp_modes)
+		return sysfs_emit(buf, "Not supported\n");
+
+	for_each_inst(mode, xcp_mgr->avail_xcp_modes) {
+		size += sysfs_emit_at(buf, size, "%s%s", sep, xcp_desc[mode]);
+		sep = ", ";
+	}
+
+	size += sysfs_emit_at(buf, size, "\n");
+
+	return size;
+}
+
+static int amdgpu_gfx_run_cleaner_shader_job(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	struct drm_gpu_scheduler *sched = &ring->sched;
+	struct drm_sched_entity entity;
+	static atomic_t counter;
+	struct dma_fence *f;
+	struct amdgpu_job *job;
+	struct amdgpu_ib *ib;
+	void *owner;
+	int i, r;
+
+	/* Initialize the scheduler entity */
+	r = drm_sched_entity_init(&entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
+	if (r) {
+		dev_err(adev->dev, "Failed setting up GFX kernel entity.\n");
+		goto err;
+	}
+
+	/*
+	 * Use some unique dummy value as the owner to make sure we execute
+	 * the cleaner shader on each submission. The value just need to change
+	 * for each submission and is otherwise meaningless.
+	 */
+	owner = (void *)(unsigned long)atomic_inc_return(&counter);
+
+	r = amdgpu_job_alloc_with_ib(ring->adev, &entity, owner,
+				     64, 0, &job,
+				     AMDGPU_KERNEL_JOB_ID_CLEANER_SHADER);
+	if (r)
+		goto err;
+
+	job->enforce_isolation = true;
+	/* always run the cleaner shader */
+	job->run_cleaner_shader = true;
+
+	ib = &job->ibs[0];
+	for (i = 0; i <= ring->funcs->align_mask; ++i)
+		ib->ptr[i] = ring->funcs->nop;
+	ib->length_dw = ring->funcs->align_mask + 1;
+
+	f = amdgpu_job_submit(job);
+
+	r = dma_fence_wait(f, false);
+	if (r)
+		goto err;
+
+	dma_fence_put(f);
+
+	/* Clean up the scheduler entity */
+	drm_sched_entity_destroy(&entity);
+	return 0;
+
+err:
+	return r;
+}
+
+static int amdgpu_gfx_run_cleaner_shader(struct amdgpu_device *adev, int xcp_id)
+{
+	int num_xcc = NUM_XCC(adev->gfx.xcc_mask);
+	struct amdgpu_ring *ring;
+	int num_xcc_to_clear;
+	int i, r, xcc_id;
+
+	if (adev->gfx.num_xcc_per_xcp)
+		num_xcc_to_clear = adev->gfx.num_xcc_per_xcp;
+	else
+		num_xcc_to_clear = 1;
+
+	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
+		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+			ring = &adev->gfx.compute_ring[i + xcc_id * adev->gfx.num_compute_rings];
+			if ((ring->xcp_id == xcp_id) && ring->sched.ready) {
+				r = amdgpu_gfx_run_cleaner_shader_job(ring);
+				if (r)
+					return r;
+				num_xcc_to_clear--;
+				break;
+			}
+		}
+	}
+
+	if (num_xcc_to_clear)
+		return -ENOENT;
+
+	return 0;
+}
+
+/**
+ * amdgpu_gfx_set_run_cleaner_shader - Execute the AMDGPU GFX Cleaner Shader
+ * @dev: The device structure
+ * @attr: The device attribute structure
+ * @buf: The buffer containing the input data
+ * @count: The size of the input data
+ *
+ * Provides the sysfs interface to manually run a cleaner shader, which is
+ * used to clear the GPU state between different tasks. Writing a value to the
+ * 'run_cleaner_shader' sysfs file triggers the cleaner shader execution.
+ * The value written corresponds to the partition index on multi-partition
+ * devices. On single-partition devices, the value should be '0'.
+ *
+ * The cleaner shader clears the Local Data Store (LDS) and General Purpose
+ * Registers (GPRs) to ensure data isolation between GPU workloads.
+ *
+ * Return: The number of bytes written to the sysfs file.
+ */
+static ssize_t amdgpu_gfx_set_run_cleaner_shader(struct device *dev,
+						 struct device_attribute *attr,
+						 const char *buf,
+						 size_t count)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	int ret;
+	long value;
+
+	if (amdgpu_in_reset(adev))
+		return -EPERM;
+	if (adev->in_suspend && !adev->in_runpm)
+		return -EPERM;
+
+	if (adev->gfx.disable_kq)
+		return -EPERM;
+
+	ret = kstrtol(buf, 0, &value);
+
+	if (ret)
+		return -EINVAL;
+
+	if (value < 0)
+		return -EINVAL;
+
+	if (adev->xcp_mgr) {
+		if (value >= adev->xcp_mgr->num_xcps)
+			return -EINVAL;
+	} else {
+		if (value > 1)
+			return -EINVAL;
+	}
+
+	ret = pm_runtime_get_sync(ddev->dev);
+	if (ret < 0) {
+		pm_runtime_put_autosuspend(ddev->dev);
+		return ret;
+	}
+
+	ret = amdgpu_gfx_run_cleaner_shader(adev, value);
+
+	pm_runtime_put_autosuspend(ddev->dev);
+
+	if (ret)
+		return ret;
+
+	return count;
+}
+
+/**
+ * amdgpu_gfx_get_enforce_isolation - Query AMDGPU GFX Enforce Isolation Settings
+ * @dev: The device structure
+ * @attr: The device attribute structure
+ * @buf: The buffer to store the output data
+ *
+ * Provides the sysfs read interface to get the current settings of the 'enforce_isolation'
+ * feature for each GPU partition. Reading from the 'enforce_isolation'
+ * sysfs file returns the isolation settings for all partitions, where '0'
+ * indicates disabled, '1' indicates enabled, and '2' indicates enabled in legacy mode,
+ * and '3' indicates enabled without cleaner shader.
+ *
+ * Return: The number of bytes read from the sysfs file.
+ */
+static ssize_t amdgpu_gfx_get_enforce_isolation(struct device *dev,
+						struct device_attribute *attr,
+						char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	int i;
+	ssize_t size = 0;
+
+	if (adev->xcp_mgr) {
+		for (i = 0; i < adev->xcp_mgr->num_xcps; i++) {
+			size += sysfs_emit_at(buf, size, "%u", adev->enforce_isolation[i]);
+			if (i < (adev->xcp_mgr->num_xcps - 1))
+				size += sysfs_emit_at(buf, size, " ");
+		}
+		buf[size++] = '\n';
+	} else {
+		size = sysfs_emit_at(buf, 0, "%u\n", adev->enforce_isolation[0]);
+	}
+
+	return size;
+}
+
+/**
+ * amdgpu_gfx_set_enforce_isolation - Control AMDGPU GFX Enforce Isolation
+ * @dev: The device structure
+ * @attr: The device attribute structure
+ * @buf: The buffer containing the input data
+ * @count: The size of the input data
+ *
+ * This function allows control over the 'enforce_isolation' feature, which
+ * serializes access to the graphics engine. Writing '0' to disable, '1' to
+ * enable isolation with cleaner shader, '2' to enable legacy isolation without
+ * cleaner shader, or '3' to enable process isolation without submitting the
+ * cleaner shader to the 'enforce_isolation' sysfs file sets the isolation mode
+ * for each partition. The input should specify the setting for all
+ * partitions.
+ *
+ * Return: The number of bytes written to the sysfs file.
+ */
+static ssize_t amdgpu_gfx_set_enforce_isolation(struct device *dev,
+						struct device_attribute *attr,
+						const char *buf, size_t count)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	long partition_values[MAX_XCP] = {0};
+	int ret, i, num_partitions;
+	const char *input_buf = buf;
+
+	for (i = 0; i < (adev->xcp_mgr ? adev->xcp_mgr->num_xcps : 1); i++) {
+		ret = sscanf(input_buf, "%ld", &partition_values[i]);
+		if (ret <= 0)
+			break;
+
+		/* Move the pointer to the next value in the string */
+		input_buf = strchr(input_buf, ' ');
+		if (input_buf) {
+			input_buf++;
+		} else {
+			i++;
+			break;
+		}
+	}
+	num_partitions = i;
+
+	if (adev->xcp_mgr && num_partitions != adev->xcp_mgr->num_xcps)
+		return -EINVAL;
+
+	if (!adev->xcp_mgr && num_partitions != 1)
+		return -EINVAL;
+
+	for (i = 0; i < num_partitions; i++) {
+		if (partition_values[i] != 0 &&
+		    partition_values[i] != 1 &&
+		    partition_values[i] != 2 &&
+		    partition_values[i] != 3)
+			return -EINVAL;
+	}
+
+	mutex_lock(&adev->enforce_isolation_mutex);
+	for (i = 0; i < num_partitions; i++) {
+		switch (partition_values[i]) {
+		case 0:
+		default:
+			adev->enforce_isolation[i] = AMDGPU_ENFORCE_ISOLATION_DISABLE;
+			break;
+		case 1:
+			adev->enforce_isolation[i] =
+				AMDGPU_ENFORCE_ISOLATION_ENABLE;
+			break;
+		case 2:
+			adev->enforce_isolation[i] =
+				AMDGPU_ENFORCE_ISOLATION_ENABLE_LEGACY;
+			break;
+		case 3:
+			adev->enforce_isolation[i] =
+				AMDGPU_ENFORCE_ISOLATION_NO_CLEANER_SHADER;
+			break;
+		}
+	}
+	mutex_unlock(&adev->enforce_isolation_mutex);
+
+	amdgpu_mes_update_enforce_isolation(adev);
+
+	return count;
+}
+
+static ssize_t amdgpu_gfx_get_gfx_reset_mask(struct device *dev,
+						struct device_attribute *attr,
+						char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	if (!adev)
+		return -ENODEV;
+
+	return amdgpu_show_reset_mask(buf, adev->gfx.gfx_supported_reset);
+}
+
+static ssize_t amdgpu_gfx_get_compute_reset_mask(struct device *dev,
+						struct device_attribute *attr,
+						char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	if (!adev)
+		return -ENODEV;
+
+	return amdgpu_show_reset_mask(buf, adev->gfx.compute_supported_reset);
+}
+
+static DEVICE_ATTR(run_cleaner_shader, 0200,
+		   NULL, amdgpu_gfx_set_run_cleaner_shader);
+
+static DEVICE_ATTR(enforce_isolation, 0644,
+		   amdgpu_gfx_get_enforce_isolation,
+		   amdgpu_gfx_set_enforce_isolation);
+
+static DEVICE_ATTR(current_compute_partition, 0644,
+		   amdgpu_gfx_get_current_compute_partition,
+		   amdgpu_gfx_set_compute_partition);
+
+static DEVICE_ATTR(available_compute_partition, 0444,
+		   amdgpu_gfx_get_available_compute_partition, NULL);
+static DEVICE_ATTR(gfx_reset_mask, 0444,
+		   amdgpu_gfx_get_gfx_reset_mask, NULL);
+
+static DEVICE_ATTR(compute_reset_mask, 0444,
+		   amdgpu_gfx_get_compute_reset_mask, NULL);
+
+static int amdgpu_gfx_sysfs_xcp_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_xcp_mgr *xcp_mgr = adev->xcp_mgr;
+	bool xcp_switch_supported;
+	int r;
+
+	if (!xcp_mgr)
+		return 0;
+
+	xcp_switch_supported =
+		(xcp_mgr->funcs && xcp_mgr->funcs->switch_partition_mode);
+
+	if (!xcp_switch_supported)
+		dev_attr_current_compute_partition.attr.mode &=
+			~(S_IWUSR | S_IWGRP | S_IWOTH);
+
+	r = device_create_file(adev->dev, &dev_attr_current_compute_partition);
+	if (r)
+		return r;
+
+	if (xcp_switch_supported)
+		r = device_create_file(adev->dev,
+				       &dev_attr_available_compute_partition);
+
+	return r;
+}
+
+static void amdgpu_gfx_sysfs_xcp_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_xcp_mgr *xcp_mgr = adev->xcp_mgr;
+	bool xcp_switch_supported;
+
+	if (!xcp_mgr)
+		return;
+
+	xcp_switch_supported =
+		(xcp_mgr->funcs && xcp_mgr->funcs->switch_partition_mode);
+	device_remove_file(adev->dev, &dev_attr_current_compute_partition);
+
+	if (xcp_switch_supported)
+		device_remove_file(adev->dev,
+				   &dev_attr_available_compute_partition);
+}
+
+static int amdgpu_gfx_sysfs_isolation_shader_init(struct amdgpu_device *adev)
+{
+	int r;
+
+	r = device_create_file(adev->dev, &dev_attr_enforce_isolation);
+	if (r)
+		return r;
+	if (adev->gfx.enable_cleaner_shader)
+		r = device_create_file(adev->dev, &dev_attr_run_cleaner_shader);
+
+	return r;
+}
+
+static void amdgpu_gfx_sysfs_isolation_shader_fini(struct amdgpu_device *adev)
+{
+	device_remove_file(adev->dev, &dev_attr_enforce_isolation);
+	if (adev->gfx.enable_cleaner_shader)
+		device_remove_file(adev->dev, &dev_attr_run_cleaner_shader);
+}
+
+static int amdgpu_gfx_sysfs_reset_mask_init(struct amdgpu_device *adev)
+{
+	int r = 0;
+
+	if (!amdgpu_gpu_recovery)
+		return r;
+
+	if (adev->gfx.num_gfx_rings) {
+		r = device_create_file(adev->dev, &dev_attr_gfx_reset_mask);
+		if (r)
+			return r;
+	}
+
+	if (adev->gfx.num_compute_rings) {
+		r = device_create_file(adev->dev, &dev_attr_compute_reset_mask);
+		if (r)
+			return r;
+	}
+
+	return r;
+}
+
+static void amdgpu_gfx_sysfs_reset_mask_fini(struct amdgpu_device *adev)
+{
+	if (!amdgpu_gpu_recovery)
+		return;
+
+	if (adev->gfx.num_gfx_rings)
+		device_remove_file(adev->dev, &dev_attr_gfx_reset_mask);
+
+	if (adev->gfx.num_compute_rings)
+		device_remove_file(adev->dev, &dev_attr_compute_reset_mask);
+}
+
+int amdgpu_gfx_sysfs_init(struct amdgpu_device *adev)
+{
+	int r;
+
+	r = amdgpu_gfx_sysfs_xcp_init(adev);
+	if (r) {
+		dev_err(adev->dev, "failed to create xcp sysfs files");
+		return r;
+	}
+
+	r = amdgpu_gfx_sysfs_isolation_shader_init(adev);
+	if (r)
+		dev_err(adev->dev, "failed to create isolation sysfs files");
+
+	r = amdgpu_gfx_sysfs_reset_mask_init(adev);
+	if (r)
+		dev_err(adev->dev, "failed to create reset mask sysfs files");
+
+	return r;
+}
+
+void amdgpu_gfx_sysfs_fini(struct amdgpu_device *adev)
+{
+	if (adev->dev->kobj.sd) {
+		amdgpu_gfx_sysfs_xcp_fini(adev);
+		amdgpu_gfx_sysfs_isolation_shader_fini(adev);
+		amdgpu_gfx_sysfs_reset_mask_fini(adev);
+	}
+}
+
+int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev,
+				      unsigned int cleaner_shader_size)
+{
+	if (!adev->gfx.enable_cleaner_shader)
+		return -EOPNOTSUPP;
+
+	return amdgpu_bo_create_kernel(adev, cleaner_shader_size, PAGE_SIZE,
+				       AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT,
+				       &adev->gfx.cleaner_shader_obj,
+				       &adev->gfx.cleaner_shader_gpu_addr,
+				       (void **)&adev->gfx.cleaner_shader_cpu_ptr);
+}
+
+void amdgpu_gfx_cleaner_shader_sw_fini(struct amdgpu_device *adev)
+{
+	if (!adev->gfx.enable_cleaner_shader)
+		return;
+
+	amdgpu_bo_free_kernel(&adev->gfx.cleaner_shader_obj,
+			      &adev->gfx.cleaner_shader_gpu_addr,
+			      (void **)&adev->gfx.cleaner_shader_cpu_ptr);
+}
+
+void amdgpu_gfx_cleaner_shader_init(struct amdgpu_device *adev,
+				    unsigned int cleaner_shader_size,
+				    const void *cleaner_shader_ptr)
+{
+	if (!adev->gfx.enable_cleaner_shader)
+		return;
+
+	if (adev->gfx.cleaner_shader_cpu_ptr && cleaner_shader_ptr)
+		memcpy_toio(adev->gfx.cleaner_shader_cpu_ptr, cleaner_shader_ptr,
+			    cleaner_shader_size);
+}
+
+/**
+ * amdgpu_gfx_kfd_sch_ctrl - Control the KFD scheduler from the KGD (Graphics Driver)
+ * @adev: amdgpu_device pointer
+ * @idx: Index of the scheduler to control
+ * @enable: Whether to enable or disable the KFD scheduler
+ *
+ * This function is used to control the KFD (Kernel Fusion Driver) scheduler
+ * from the KGD. It is part of the cleaner shader feature. This function plays
+ * a key role in enforcing process isolation on the GPU.
+ *
+ * The function uses a reference count mechanism (kfd_sch_req_count) to keep
+ * track of the number of requests to enable the KFD scheduler. When a request
+ * to enable the KFD scheduler is made, the reference count is decremented.
+ * When the reference count reaches zero, a delayed work is scheduled to
+ * enforce isolation after a delay of GFX_SLICE_PERIOD.
+ *
+ * When a request to disable the KFD scheduler is made, the function first
+ * checks if the reference count is zero. If it is, it cancels the delayed work
+ * for enforcing isolation and checks if the KFD scheduler is active. If the
+ * KFD scheduler is active, it sends a request to stop the KFD scheduler and
+ * sets the KFD scheduler state to inactive. Then, it increments the reference
+ * count.
+ *
+ * The function is synchronized using the kfd_sch_mutex to ensure that the KFD
+ * scheduler state and reference count are updated atomically.
+ *
+ * Note: If the reference count is already zero when a request to enable the
+ * KFD scheduler is made, it means there's an imbalance bug somewhere. The
+ * function triggers a warning in this case.
+ */
+static void amdgpu_gfx_kfd_sch_ctrl(struct amdgpu_device *adev, u32 idx,
+				    bool enable)
+{
+	mutex_lock(&adev->gfx.userq_sch_mutex);
+
+	if (enable) {
+		/* If the count is already 0, it means there's an imbalance bug somewhere.
+		 * Note that the bug may be in a different caller than the one which triggers the
+		 * WARN_ON_ONCE.
+		 */
+		if (WARN_ON_ONCE(adev->gfx.userq_sch_req_count[idx] == 0)) {
+			dev_err(adev->dev, "Attempted to enable KFD scheduler when reference count is already zero\n");
+			goto unlock;
+		}
+
+		adev->gfx.userq_sch_req_count[idx]--;
+
+		if (adev->gfx.userq_sch_req_count[idx] == 0 &&
+		    adev->gfx.userq_sch_inactive[idx]) {
+			schedule_delayed_work(&adev->gfx.enforce_isolation[idx].work,
+					      msecs_to_jiffies(adev->gfx.enforce_isolation_time[idx]));
+		}
+	} else {
+		if (adev->gfx.userq_sch_req_count[idx] == 0) {
+			cancel_delayed_work_sync(&adev->gfx.enforce_isolation[idx].work);
+			if (!adev->gfx.userq_sch_inactive[idx]) {
+				amdgpu_userq_stop_sched_for_enforce_isolation(adev, idx);
+				if (adev->kfd.init_complete)
+					amdgpu_amdkfd_stop_sched(adev, idx);
+				adev->gfx.userq_sch_inactive[idx] = true;
+			}
+		}
+
+		adev->gfx.userq_sch_req_count[idx]++;
+	}
+
+unlock:
+	mutex_unlock(&adev->gfx.userq_sch_mutex);
+}
+
+/**
+ * amdgpu_gfx_enforce_isolation_handler - work handler for enforcing shader isolation
+ *
+ * @work: work_struct.
+ *
+ * This function is the work handler for enforcing shader isolation on AMD GPUs.
+ * It counts the number of emitted fences for each GFX and compute ring. If there
+ * are any fences, it schedules the `enforce_isolation_work` to be run after a
+ * delay of `GFX_SLICE_PERIOD`. If there are no fences, it signals the Kernel Fusion
+ * Driver (KFD) to resume the runqueue. The function is synchronized using the
+ * `enforce_isolation_mutex`.
+ */
+void amdgpu_gfx_enforce_isolation_handler(struct work_struct *work)
+{
+	struct amdgpu_isolation_work *isolation_work =
+		container_of(work, struct amdgpu_isolation_work, work.work);
+	struct amdgpu_device *adev = isolation_work->adev;
+	u32 i, idx, fences = 0;
+
+	if (isolation_work->xcp_id == AMDGPU_XCP_NO_PARTITION)
+		idx = 0;
+	else
+		idx = isolation_work->xcp_id;
+
+	if (idx >= MAX_XCP)
+		return;
+
+	mutex_lock(&adev->enforce_isolation_mutex);
+	for (i = 0; i < AMDGPU_MAX_GFX_RINGS; ++i) {
+		if (isolation_work->xcp_id == adev->gfx.gfx_ring[i].xcp_id)
+			fences += amdgpu_fence_count_emitted(&adev->gfx.gfx_ring[i]);
+	}
+	for (i = 0; i < (AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES); ++i) {
+		if (isolation_work->xcp_id == adev->gfx.compute_ring[i].xcp_id)
+			fences += amdgpu_fence_count_emitted(&adev->gfx.compute_ring[i]);
+	}
+	if (fences) {
+		/* we've already had our timeslice, so let's wrap this up */
+		schedule_delayed_work(&adev->gfx.enforce_isolation[idx].work,
+				      msecs_to_jiffies(1));
+	} else {
+		/* Tell KFD to resume the runqueue */
+		WARN_ON_ONCE(!adev->gfx.userq_sch_inactive[idx]);
+		WARN_ON_ONCE(adev->gfx.userq_sch_req_count[idx]);
+
+		amdgpu_userq_start_sched_for_enforce_isolation(adev, idx);
+		if (adev->kfd.init_complete)
+			amdgpu_amdkfd_start_sched(adev, idx);
+		adev->gfx.userq_sch_inactive[idx] = false;
+	}
+	mutex_unlock(&adev->enforce_isolation_mutex);
+}
+
+/**
+ * amdgpu_gfx_enforce_isolation_wait_for_kfd - Manage KFD wait period for process isolation
+ * @adev: amdgpu_device pointer
+ * @idx: Index of the GPU partition
+ *
+ * When kernel submissions come in, the jobs are given a time slice and once
+ * that time slice is up, if there are KFD user queues active, kernel
+ * submissions are blocked until KFD has had its time slice. Once the KFD time
+ * slice is up, KFD user queues are preempted and kernel submissions are
+ * unblocked and allowed to run again.
+ */
+static void
+amdgpu_gfx_enforce_isolation_wait_for_kfd(struct amdgpu_device *adev,
+					  u32 idx)
+{
+	unsigned long cjiffies;
+	bool wait = false;
+
+	mutex_lock(&adev->enforce_isolation_mutex);
+	if (adev->enforce_isolation[idx] == AMDGPU_ENFORCE_ISOLATION_ENABLE) {
+		/* set the initial values if nothing is set */
+		if (!adev->gfx.enforce_isolation_jiffies[idx]) {
+			adev->gfx.enforce_isolation_jiffies[idx] = jiffies;
+			adev->gfx.enforce_isolation_time[idx] =	GFX_SLICE_PERIOD_MS;
+		}
+		/* Make sure KFD gets a chance to run */
+		if (amdgpu_amdkfd_compute_active(adev, idx)) {
+			cjiffies = jiffies;
+			if (time_after(cjiffies, adev->gfx.enforce_isolation_jiffies[idx])) {
+				cjiffies -= adev->gfx.enforce_isolation_jiffies[idx];
+				if ((jiffies_to_msecs(cjiffies) >= GFX_SLICE_PERIOD_MS)) {
+					/* if our time is up, let KGD work drain before scheduling more */
+					wait = true;
+					/* reset the timer period */
+					adev->gfx.enforce_isolation_time[idx] =	GFX_SLICE_PERIOD_MS;
+				} else {
+					/* set the timer period to what's left in our time slice */
+					adev->gfx.enforce_isolation_time[idx] =
+						GFX_SLICE_PERIOD_MS - jiffies_to_msecs(cjiffies);
+				}
+			} else {
+				/* if jiffies wrap around we will just wait a little longer */
+				adev->gfx.enforce_isolation_jiffies[idx] = jiffies;
+			}
+		} else {
+			/* if there is no KFD work, then set the full slice period */
+			adev->gfx.enforce_isolation_jiffies[idx] = jiffies;
+			adev->gfx.enforce_isolation_time[idx] = GFX_SLICE_PERIOD_MS;
+		}
+	}
+	mutex_unlock(&adev->enforce_isolation_mutex);
+
+	if (wait)
+		msleep(GFX_SLICE_PERIOD_MS);
+}
+
+/**
+ * amdgpu_gfx_enforce_isolation_ring_begin_use - Begin use of a ring with enforced isolation
+ * @ring: Pointer to the amdgpu_ring structure
+ *
+ * Ring begin_use helper implementation for gfx which serializes access to the
+ * gfx IP between kernel submission IOCTLs and KFD user queues when isolation
+ * enforcement is enabled. The kernel submission IOCTLs and KFD user queues
+ * each get a time slice when both are active.
+ */
+void amdgpu_gfx_enforce_isolation_ring_begin_use(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	u32 idx;
+	bool sched_work = false;
+
+	if (!adev->gfx.enable_cleaner_shader)
+		return;
+
+	if (ring->xcp_id == AMDGPU_XCP_NO_PARTITION)
+		idx = 0;
+	else
+		idx = ring->xcp_id;
+
+	if (idx >= MAX_XCP)
+		return;
+
+	/* Don't submit more work until KFD has had some time */
+	amdgpu_gfx_enforce_isolation_wait_for_kfd(adev, idx);
+
+	mutex_lock(&adev->enforce_isolation_mutex);
+	if (adev->enforce_isolation[idx] == AMDGPU_ENFORCE_ISOLATION_ENABLE) {
+		if (adev->kfd.init_complete)
+			sched_work = true;
+	}
+	mutex_unlock(&adev->enforce_isolation_mutex);
+
+	if (sched_work)
+		amdgpu_gfx_kfd_sch_ctrl(adev, idx, false);
+}
+
+/**
+ * amdgpu_gfx_enforce_isolation_ring_end_use - End use of a ring with enforced isolation
+ * @ring: Pointer to the amdgpu_ring structure
+ *
+ * Ring end_use helper implementation for gfx which serializes access to the
+ * gfx IP between kernel submission IOCTLs and KFD user queues when isolation
+ * enforcement is enabled. The kernel submission IOCTLs and KFD user queues
+ * each get a time slice when both are active.
+ */
+void amdgpu_gfx_enforce_isolation_ring_end_use(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	u32 idx;
+	bool sched_work = false;
+
+	if (!adev->gfx.enable_cleaner_shader)
+		return;
+
+	if (ring->xcp_id == AMDGPU_XCP_NO_PARTITION)
+		idx = 0;
+	else
+		idx = ring->xcp_id;
+
+	if (idx >= MAX_XCP)
+		return;
+
+	mutex_lock(&adev->enforce_isolation_mutex);
+	if (adev->enforce_isolation[idx] == AMDGPU_ENFORCE_ISOLATION_ENABLE) {
+		if (adev->kfd.init_complete)
+			sched_work = true;
+	}
+	mutex_unlock(&adev->enforce_isolation_mutex);
+
+	if (sched_work)
+		amdgpu_gfx_kfd_sch_ctrl(adev, idx, true);
+}
+
+void amdgpu_gfx_profile_idle_work_handler(struct work_struct *work)
+{
+	struct amdgpu_device *adev =
+		container_of(work, struct amdgpu_device, gfx.idle_work.work);
+	enum PP_SMC_POWER_PROFILE profile;
+	u32 i, fences = 0;
+	int r;
+
+	if (adev->gfx.num_gfx_rings)
+		profile = PP_SMC_POWER_PROFILE_FULLSCREEN3D;
+	else
+		profile = PP_SMC_POWER_PROFILE_COMPUTE;
+
+	for (i = 0; i < AMDGPU_MAX_GFX_RINGS; ++i)
+		fences += amdgpu_fence_count_emitted(&adev->gfx.gfx_ring[i]);
+	for (i = 0; i < (AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES); ++i)
+		fences += amdgpu_fence_count_emitted(&adev->gfx.compute_ring[i]);
+	if (!fences && !atomic_read(&adev->gfx.total_submission_cnt)) {
+		mutex_lock(&adev->gfx.workload_profile_mutex);
+		if (adev->gfx.workload_profile_active) {
+			r = amdgpu_dpm_switch_power_profile(adev, profile, false);
+			if (r)
+				dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
+					 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
+					 "fullscreen 3D" : "compute");
+			adev->gfx.workload_profile_active = false;
+		}
+		mutex_unlock(&adev->gfx.workload_profile_mutex);
+	} else {
+		schedule_delayed_work(&adev->gfx.idle_work, GFX_PROFILE_IDLE_TIMEOUT);
+	}
+}
+
+void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	enum PP_SMC_POWER_PROFILE profile;
+	int r;
+
+	if (amdgpu_dpm_is_overdrive_enabled(adev))
+		return;
+
+	if (adev->gfx.num_gfx_rings)
+		profile = PP_SMC_POWER_PROFILE_FULLSCREEN3D;
+	else
+		profile = PP_SMC_POWER_PROFILE_COMPUTE;
+
+	atomic_inc(&adev->gfx.total_submission_cnt);
+
+	cancel_delayed_work_sync(&adev->gfx.idle_work);
+
+	/* We can safely return early here because we've cancelled the
+	 * the delayed work so there is no one else to set it to false
+	 * and we don't care if someone else sets it to true.
+	 */
+	if (adev->gfx.workload_profile_active)
+		return;
+
+	mutex_lock(&adev->gfx.workload_profile_mutex);
+	if (!adev->gfx.workload_profile_active) {
+		r = amdgpu_dpm_switch_power_profile(adev, profile, true);
+		if (r)
+			dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
+				 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
+				 "fullscreen 3D" : "compute");
+		adev->gfx.workload_profile_active = true;
+	}
+	mutex_unlock(&adev->gfx.workload_profile_mutex);
+}
+
+void amdgpu_gfx_profile_ring_end_use(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+
+	if (amdgpu_dpm_is_overdrive_enabled(adev))
+		return;
+
+	atomic_dec(&ring->adev->gfx.total_submission_cnt);
+
+	schedule_delayed_work(&ring->adev->gfx.idle_work, GFX_PROFILE_IDLE_TIMEOUT);
+}
+
+/**
+ * amdgpu_gfx_csb_preamble_start - Set CSB preamble start
+ *
+ * @buffer: This is an output variable that gets the PACKET3 preamble setup.
+ *
+ * Return:
+ * return the latest index.
+ */
+u32 amdgpu_gfx_csb_preamble_start(u32 *buffer)
+{
+	u32 count = 0;
+
+	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
+	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
+
+	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
+	buffer[count++] = cpu_to_le32(0x80000000);
+	buffer[count++] = cpu_to_le32(0x80000000);
+
+	return count;
+}
+
+/**
+ * amdgpu_gfx_csb_data_parser - Parser CS data
+ *
+ * @adev: amdgpu_device pointer used to get the CS data and other gfx info.
+ * @buffer: This is an output variable that gets the PACKET3 preamble end.
+ * @count: Index to start set the preemble end.
+ *
+ * Return:
+ * return the latest index.
+ */
+u32 amdgpu_gfx_csb_data_parser(struct amdgpu_device *adev, u32 *buffer, u32 count)
+{
+	const struct cs_section_def *sect = NULL;
+	const struct cs_extent_def *ext = NULL;
+	u32 i;
+
+	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
+		for (ext = sect->section; ext->extent != NULL; ++ext) {
+			if (sect->id == SECT_CONTEXT) {
+				buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
+				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
+
+				for (i = 0; i < ext->reg_count; i++)
+					buffer[count++] = cpu_to_le32(ext->extent[i]);
+			}
+		}
+	}
+
+	return count;
+}
+
+/**
+ * amdgpu_gfx_csb_preamble_end - Set CSB preamble end
+ *
+ * @buffer: This is an output variable that gets the PACKET3 preamble end.
+ * @count: Index to start set the preemble end.
+ */
+void amdgpu_gfx_csb_preamble_end(u32 *buffer, u32 count)
+{
+	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
+	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
+
+	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
+	buffer[count++] = cpu_to_le32(0);
+}
+
+/*
+ * debugfs for to enable/disable gfx job submission to specific core.
+ */
+#if defined(CONFIG_DEBUG_FS)
+static int amdgpu_debugfs_gfx_sched_mask_set(void *data, u64 val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	u32 i;
+	u64 mask = 0;
+	struct amdgpu_ring *ring;
+
+	if (!adev)
+		return -ENODEV;
+
+	mask = (1ULL << adev->gfx.num_gfx_rings) - 1;
+	if ((val & mask) == 0)
+		return -EINVAL;
+
+	for (i = 0; i < adev->gfx.num_gfx_rings; ++i) {
+		ring = &adev->gfx.gfx_ring[i];
+		if (val & (1 << i))
+			ring->sched.ready = true;
+		else
+			ring->sched.ready = false;
+	}
+	/* publish sched.ready flag update effective immediately across smp */
+	smp_rmb();
+	return 0;
+}
+
+static int amdgpu_debugfs_gfx_sched_mask_get(void *data, u64 *val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	u32 i;
+	u64 mask = 0;
+	struct amdgpu_ring *ring;
+
+	if (!adev)
+		return -ENODEV;
+	for (i = 0; i < adev->gfx.num_gfx_rings; ++i) {
+		ring = &adev->gfx.gfx_ring[i];
+		if (ring->sched.ready)
+			mask |= 1ULL << i;
+	}
+
+	*val = mask;
+	return 0;
+}
+
+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_gfx_sched_mask_fops,
+			 amdgpu_debugfs_gfx_sched_mask_get,
+			 amdgpu_debugfs_gfx_sched_mask_set, "%llx\n");
+
+#endif
+
+void amdgpu_debugfs_gfx_sched_mask_init(struct amdgpu_device *adev)
+{
+#if defined(CONFIG_DEBUG_FS)
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+	char name[32];
+
+	if (!(adev->gfx.num_gfx_rings > 1))
+		return;
+	sprintf(name, "amdgpu_gfx_sched_mask");
+	debugfs_create_file(name, 0600, root, adev,
+			    &amdgpu_debugfs_gfx_sched_mask_fops);
+#endif
+}
+
+/*
+ * debugfs for to enable/disable compute job submission to specific core.
+ */
+#if defined(CONFIG_DEBUG_FS)
+static int amdgpu_debugfs_compute_sched_mask_set(void *data, u64 val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	u32 i;
+	u64 mask = 0;
+	struct amdgpu_ring *ring;
+
+	if (!adev)
+		return -ENODEV;
+
+	mask = (1ULL << adev->gfx.num_compute_rings) - 1;
+	if ((val & mask) == 0)
+		return -EINVAL;
+
+	for (i = 0; i < adev->gfx.num_compute_rings; ++i) {
+		ring = &adev->gfx.compute_ring[i];
+		if (val & (1 << i))
+			ring->sched.ready = true;
+		else
+			ring->sched.ready = false;
+	}
+
+	/* publish sched.ready flag update effective immediately across smp */
+	smp_rmb();
+	return 0;
+}
+
+static int amdgpu_debugfs_compute_sched_mask_get(void *data, u64 *val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	u32 i;
+	u64 mask = 0;
+	struct amdgpu_ring *ring;
+
+	if (!adev)
+		return -ENODEV;
+	for (i = 0; i < adev->gfx.num_compute_rings; ++i) {
+		ring = &adev->gfx.compute_ring[i];
+		if (ring->sched.ready)
+			mask |= 1ULL << i;
+	}
+
+	*val = mask;
+	return 0;
+}
+
+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_compute_sched_mask_fops,
+			 amdgpu_debugfs_compute_sched_mask_get,
+			 amdgpu_debugfs_compute_sched_mask_set, "%llx\n");
+
+#endif
+
+void amdgpu_debugfs_compute_sched_mask_init(struct amdgpu_device *adev)
+{
+#if defined(CONFIG_DEBUG_FS)
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+	char name[32];
+
+	if (!(adev->gfx.num_compute_rings > 1))
+		return;
+	sprintf(name, "amdgpu_compute_sched_mask");
+	debugfs_create_file(name, 0600, root, adev,
+			    &amdgpu_debugfs_compute_sched_mask_fops);
+#endif
+}
+
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index e02044086445..efd61a1ccc66 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -24,10 +24,648 @@
 #ifndef __AMDGPU_GFX_H__
 #define __AMDGPU_GFX_H__
 
-int amdgpu_gfx_scratch_get(struct amdgpu_device *adev, uint32_t *reg);
-void amdgpu_gfx_scratch_free(struct amdgpu_device *adev, uint32_t reg);
+/*
+ * GFX stuff
+ */
+#include "clearstate_defs.h"
+#include "amdgpu_ring.h"
+#include "amdgpu_rlc.h"
+#include "amdgpu_imu.h"
+#include "soc15.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_ring_mux.h"
+#include "amdgpu_xcp.h"
+
+/* GFX current status */
+#define AMDGPU_GFX_NORMAL_MODE			0x00000000L
+#define AMDGPU_GFX_SAFE_MODE			0x00000001L
+#define AMDGPU_GFX_PG_DISABLED_MODE		0x00000002L
+#define AMDGPU_GFX_CG_DISABLED_MODE		0x00000004L
+#define AMDGPU_GFX_LBPW_DISABLED_MODE		0x00000008L
+
+#define AMDGPU_MAX_GC_INSTANCES		8
+#define AMDGPU_MAX_QUEUES		128
+
+#define AMDGPU_MAX_GFX_QUEUES AMDGPU_MAX_QUEUES
+#define AMDGPU_MAX_COMPUTE_QUEUES AMDGPU_MAX_QUEUES
+
+enum amdgpu_gfx_pipe_priority {
+	AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1,
+	AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2
+};
+
+#define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM  0
+#define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM  15
+
+/* 1 second timeout */
+#define GFX_PROFILE_IDLE_TIMEOUT	msecs_to_jiffies(1000)
+
+enum amdgpu_gfx_partition {
+	AMDGPU_SPX_PARTITION_MODE = 0,
+	AMDGPU_DPX_PARTITION_MODE = 1,
+	AMDGPU_TPX_PARTITION_MODE = 2,
+	AMDGPU_QPX_PARTITION_MODE = 3,
+	AMDGPU_CPX_PARTITION_MODE = 4,
+	AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE = -1,
+	/* Automatically choose the right mode */
+	AMDGPU_AUTO_COMPUTE_PARTITION_MODE = -2,
+};
+
+#define NUM_XCC(x) hweight16(x)
+
+enum amdgpu_gfx_ras_mem_id_type {
+	AMDGPU_GFX_CP_MEM = 0,
+	AMDGPU_GFX_GCEA_MEM,
+	AMDGPU_GFX_GC_CANE_MEM,
+	AMDGPU_GFX_GCUTCL2_MEM,
+	AMDGPU_GFX_GDS_MEM,
+	AMDGPU_GFX_LDS_MEM,
+	AMDGPU_GFX_RLC_MEM,
+	AMDGPU_GFX_SP_MEM,
+	AMDGPU_GFX_SPI_MEM,
+	AMDGPU_GFX_SQC_MEM,
+	AMDGPU_GFX_SQ_MEM,
+	AMDGPU_GFX_TA_MEM,
+	AMDGPU_GFX_TCC_MEM,
+	AMDGPU_GFX_TCA_MEM,
+	AMDGPU_GFX_TCI_MEM,
+	AMDGPU_GFX_TCP_MEM,
+	AMDGPU_GFX_TD_MEM,
+	AMDGPU_GFX_TCX_MEM,
+	AMDGPU_GFX_ATC_L2_MEM,
+	AMDGPU_GFX_UTCL2_MEM,
+	AMDGPU_GFX_VML2_MEM,
+	AMDGPU_GFX_VML2_WALKER_MEM,
+	AMDGPU_GFX_MEM_TYPE_NUM
+};
+
+struct amdgpu_mec {
+	struct amdgpu_bo	*hpd_eop_obj;
+	u64			hpd_eop_gpu_addr;
+	struct amdgpu_bo	*mec_fw_obj;
+	u64			mec_fw_gpu_addr;
+	struct amdgpu_bo	*mec_fw_data_obj;
+	u64			mec_fw_data_gpu_addr;
+
+	u32 num_mec;
+	u32 num_pipe_per_mec;
+	u32 num_queue_per_pipe;
+	void			*mqd_backup[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES];
+};
+
+struct amdgpu_mec_bitmap {
+	/* These are the resources for which amdgpu takes ownership */
+	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
+};
+
+enum amdgpu_unmap_queues_action {
+	PREEMPT_QUEUES = 0,
+	RESET_QUEUES,
+	DISABLE_PROCESS_QUEUES,
+	PREEMPT_QUEUES_NO_UNMAP,
+};
+
+struct kiq_pm4_funcs {
+	/* Support ASIC-specific kiq pm4 packets*/
+	void (*kiq_set_resources)(struct amdgpu_ring *kiq_ring,
+					uint64_t queue_mask);
+	void (*kiq_map_queues)(struct amdgpu_ring *kiq_ring,
+					struct amdgpu_ring *ring);
+	void (*kiq_unmap_queues)(struct amdgpu_ring *kiq_ring,
+				 struct amdgpu_ring *ring,
+				 enum amdgpu_unmap_queues_action action,
+				 u64 gpu_addr, u64 seq);
+	void (*kiq_query_status)(struct amdgpu_ring *kiq_ring,
+					struct amdgpu_ring *ring,
+					u64 addr,
+					u64 seq);
+	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
+				uint16_t pasid, uint32_t flush_type,
+				bool all_hub);
+	void (*kiq_reset_hw_queue)(struct amdgpu_ring *kiq_ring,
+				   uint32_t queue_type, uint32_t me_id,
+				   uint32_t pipe_id, uint32_t queue_id,
+				   uint32_t xcc_id, uint32_t vmid);
+	/* Packet sizes */
+	int set_resources_size;
+	int map_queues_size;
+	int unmap_queues_size;
+	int query_status_size;
+	int invalidate_tlbs_size;
+};
+
+struct amdgpu_kiq {
+	u64			eop_gpu_addr;
+	struct amdgpu_bo	*eop_obj;
+	spinlock_t              ring_lock;
+	struct amdgpu_ring	ring;
+	struct amdgpu_irq_src	irq;
+	const struct kiq_pm4_funcs *pmf;
+	void			*mqd_backup;
+};
+
+/*
+ * GFX configurations
+ */
+#define AMDGPU_GFX_MAX_SE 4
+#define AMDGPU_GFX_MAX_SH_PER_SE 2
+
+/**
+ * amdgpu_rb_config - Configure a single Render Backend (RB)
+ *
+ * Bad RBs are fused off and there is a harvest register the driver reads to
+ * determine which RB(s) are fused off so that the driver can configure the
+ * hardware state so that nothing gets sent to them. There are also user
+ * harvest registers that the driver can program to disable additional RBs,
+ * etc., for testing purposes.
+ */
+struct amdgpu_rb_config {
+	/**
+	 * @rb_backend_disable:
+	 *
+	 * The value captured from register RB_BACKEND_DISABLE indicates if the
+	 * RB backend is disabled or not.
+	 */
+	uint32_t rb_backend_disable;
+
+	/**
+	 * @user_rb_backend_disable:
+	 *
+	 * The value captured from register USER_RB_BACKEND_DISABLE indicates
+	 * if the User RB backend is disabled or not.
+	 */
+	uint32_t user_rb_backend_disable;
+
+	/**
+	 * @raster_config:
+	 *
+	 * To set up all of the states, it is necessary to have two registers
+	 * to keep all of the states. This field holds the first register.
+	 */
+	uint32_t raster_config;
+
+	/**
+	 * @raster_config_1:
+	 *
+	 * To set up all of the states, it is necessary to have two registers
+	 * to keep all of the states. This field holds the second register.
+	 */
+	uint32_t raster_config_1;
+};
+
+struct gb_addr_config {
+	uint16_t pipe_interleave_size;
+	uint8_t num_pipes;
+	uint8_t max_compress_frags;
+	uint8_t num_banks;
+	uint8_t num_se;
+	uint8_t num_rb_per_se;
+	uint8_t num_pkrs;
+};
+
+struct amdgpu_gfx_config {
+	unsigned max_shader_engines;
+	unsigned max_tile_pipes;
+	unsigned max_cu_per_sh;
+	unsigned max_sh_per_se;
+	unsigned max_backends_per_se;
+	unsigned max_texture_channel_caches;
+	unsigned max_gprs;
+	unsigned max_gs_threads;
+	unsigned max_hw_contexts;
+	unsigned sc_prim_fifo_size_frontend;
+	unsigned sc_prim_fifo_size_backend;
+	unsigned sc_hiz_tile_fifo_size;
+	unsigned sc_earlyz_tile_fifo_size;
+
+	unsigned num_tile_pipes;
+	unsigned backend_enable_mask;
+	unsigned mem_max_burst_length_bytes;
+	unsigned mem_row_size_in_kb;
+	unsigned shader_engine_tile_size;
+	unsigned num_gpus;
+	unsigned multi_gpu_tile_size;
+	unsigned mc_arb_ramcfg;
+	unsigned num_banks;
+	unsigned num_ranks;
+	unsigned gb_addr_config;
+	unsigned num_rbs;
+	unsigned gs_vgt_table_depth;
+	unsigned gs_prim_buffer_depth;
+
+	uint32_t tile_mode_array[32];
+	uint32_t macrotile_mode_array[16];
+
+	struct gb_addr_config gb_addr_config_fields;
+
+	/**
+	 * @rb_config:
+	 *
+	 * Matrix that keeps all the Render Backend (color and depth buffer
+	 * handling) configuration on the 3D engine.
+	 */
+	struct amdgpu_rb_config rb_config[AMDGPU_GFX_MAX_SE][AMDGPU_GFX_MAX_SH_PER_SE];
+
+	/* gfx configure feature */
+	uint32_t double_offchip_lds_buf;
+	/* cached value of DB_DEBUG2 */
+	uint32_t db_debug2;
+	/* gfx10 specific config */
+	uint32_t num_sc_per_sh;
+	uint32_t num_packer_per_sc;
+	uint32_t pa_sc_tile_steering_override;
+	/* Whether texture coordinate truncation is conformant. */
+	bool ta_cntl2_truncate_coord_mode;
+	uint64_t tcc_disabled_mask;
+	uint32_t gc_num_tcp_per_sa;
+	uint32_t gc_num_sdp_interface;
+	uint32_t gc_num_tcps;
+	uint32_t gc_num_tcp_per_wpg;
+	uint32_t gc_tcp_l1_size;
+	uint32_t gc_num_sqc_per_wgp;
+	uint32_t gc_l1_instruction_cache_size_per_sqc;
+	uint32_t gc_l1_data_cache_size_per_sqc;
+	uint32_t gc_gl1c_per_sa;
+	uint32_t gc_gl1c_size_per_instance;
+	uint32_t gc_gl2c_per_gpu;
+	uint32_t gc_tcp_size_per_cu;
+	uint32_t gc_num_cu_per_sqc;
+	uint32_t gc_tcc_size;
+	uint32_t gc_tcp_cache_line_size;
+	uint32_t gc_instruction_cache_size_per_sqc;
+	uint32_t gc_instruction_cache_line_size;
+	uint32_t gc_scalar_data_cache_size_per_sqc;
+	uint32_t gc_scalar_data_cache_line_size;
+	uint32_t gc_tcc_cache_line_size;
+};
+
+struct amdgpu_cu_info {
+	uint32_t simd_per_cu;
+	uint32_t max_waves_per_simd;
+	uint32_t wave_front_size;
+	uint32_t max_scratch_slots_per_cu;
+	uint32_t lds_size;
+
+	/* total active CU number */
+	uint32_t number;
+	uint32_t ao_cu_mask;
+	uint32_t ao_cu_bitmap[4][4];
+	uint32_t bitmap[AMDGPU_MAX_GC_INSTANCES][4][4];
+};
+
+struct amdgpu_gfx_ras {
+	struct amdgpu_ras_block_object  ras_block;
+	void (*enable_watchdog_timer)(struct amdgpu_device *adev);
+	int (*rlc_gc_fed_irq)(struct amdgpu_device *adev,
+				struct amdgpu_irq_src *source,
+				struct amdgpu_iv_entry *entry);
+	int (*poison_consumption_handler)(struct amdgpu_device *adev,
+						struct amdgpu_iv_entry *entry);
+};
+
+struct amdgpu_gfx_shadow_info {
+	u32 shadow_size;
+	u32 shadow_alignment;
+	u32 csa_size;
+	u32 csa_alignment;
+};
+
+struct amdgpu_gfx_funcs {
+	/* get the gpu clock counter */
+	uint64_t (*get_gpu_clock_counter)(struct amdgpu_device *adev);
+	void (*select_se_sh)(struct amdgpu_device *adev, u32 se_num,
+			     u32 sh_num, u32 instance, int xcc_id);
+	void (*read_wave_data)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
+			       uint32_t wave, uint32_t *dst, int *no_fields);
+	void (*read_wave_vgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
+				uint32_t wave, uint32_t thread, uint32_t start,
+				uint32_t size, uint32_t *dst);
+	void (*read_wave_sgprs)(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
+				uint32_t wave, uint32_t start, uint32_t size,
+				uint32_t *dst);
+	void (*select_me_pipe_q)(struct amdgpu_device *adev, u32 me, u32 pipe,
+				 u32 queue, u32 vmid, u32 xcc_id);
+	void (*init_spm_golden)(struct amdgpu_device *adev);
+	void (*update_perfmon_mgcg)(struct amdgpu_device *adev, bool enable);
+	int (*get_gfx_shadow_info)(struct amdgpu_device *adev,
+				   struct amdgpu_gfx_shadow_info *shadow_info,
+				   bool skip_check);
+	enum amdgpu_gfx_partition
+			(*query_partition_mode)(struct amdgpu_device *adev);
+	int (*switch_partition_mode)(struct amdgpu_device *adev,
+				     int num_xccs_per_xcp);
+	int (*ih_node_to_logical_xcc)(struct amdgpu_device *adev, int ih_node);
+	int (*get_xccs_per_xcp)(struct amdgpu_device *adev);
+};
+
+struct sq_work {
+	struct work_struct	work;
+	unsigned ih_data;
+};
+
+struct amdgpu_pfp {
+	struct amdgpu_bo		*pfp_fw_obj;
+	uint64_t			pfp_fw_gpu_addr;
+	uint32_t			*pfp_fw_ptr;
+
+	struct amdgpu_bo		*pfp_fw_data_obj;
+	uint64_t			pfp_fw_data_gpu_addr;
+	uint32_t			*pfp_fw_data_ptr;
+};
+
+struct amdgpu_ce {
+	struct amdgpu_bo		*ce_fw_obj;
+	uint64_t			ce_fw_gpu_addr;
+	uint32_t			*ce_fw_ptr;
+};
+
+struct amdgpu_me {
+	struct amdgpu_bo		*me_fw_obj;
+	uint64_t			me_fw_gpu_addr;
+	uint32_t			*me_fw_ptr;
+
+	struct amdgpu_bo		*me_fw_data_obj;
+	uint64_t			me_fw_data_gpu_addr;
+	uint32_t			*me_fw_data_ptr;
+
+	uint32_t			num_me;
+	uint32_t			num_pipe_per_me;
+	uint32_t			num_queue_per_pipe;
+	void				*mqd_backup[AMDGPU_MAX_GFX_RINGS];
+
+	/* These are the resources for which amdgpu takes ownership */
+	DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
+};
+
+struct amdgpu_isolation_work {
+	struct amdgpu_device		*adev;
+	u32				xcp_id;
+	struct delayed_work		work;
+};
+
+struct amdgpu_gfx {
+	struct mutex			gpu_clock_mutex;
+	struct amdgpu_gfx_config	config;
+	struct amdgpu_rlc		rlc;
+	struct amdgpu_pfp		pfp;
+	struct amdgpu_ce		ce;
+	struct amdgpu_me		me;
+	struct amdgpu_mec		mec;
+	struct amdgpu_mec_bitmap	mec_bitmap[AMDGPU_MAX_GC_INSTANCES];
+	struct amdgpu_kiq		kiq[AMDGPU_MAX_GC_INSTANCES];
+	struct amdgpu_imu		imu;
+	bool				rs64_enable; /* firmware format */
+	const struct firmware		*me_fw;	/* ME firmware */
+	uint32_t			me_fw_version;
+	const struct firmware		*pfp_fw; /* PFP firmware */
+	uint32_t			pfp_fw_version;
+	const struct firmware		*ce_fw;	/* CE firmware */
+	uint32_t			ce_fw_version;
+	const struct firmware		*rlc_fw; /* RLC firmware */
+	uint32_t			rlc_fw_version;
+	const struct firmware		*mec_fw; /* MEC firmware */
+	uint32_t			mec_fw_version;
+	const struct firmware		*mec2_fw; /* MEC2 firmware */
+	uint32_t			mec2_fw_version;
+	const struct firmware		*imu_fw; /* IMU firmware */
+	uint32_t			imu_fw_version;
+	uint32_t			me_feature_version;
+	uint32_t			ce_feature_version;
+	uint32_t			pfp_feature_version;
+	uint32_t			rlc_feature_version;
+	uint32_t			rlc_srlc_fw_version;
+	uint32_t			rlc_srlc_feature_version;
+	uint32_t			rlc_srlg_fw_version;
+	uint32_t			rlc_srlg_feature_version;
+	uint32_t			rlc_srls_fw_version;
+	uint32_t			rlc_srls_feature_version;
+	uint32_t			rlcp_ucode_version;
+	uint32_t			rlcp_ucode_feature_version;
+	uint32_t			rlcv_ucode_version;
+	uint32_t			rlcv_ucode_feature_version;
+	uint32_t			mec_feature_version;
+	uint32_t			mec2_feature_version;
+	bool				mec_fw_write_wait;
+	bool				me_fw_write_wait;
+	bool				cp_fw_write_wait;
+	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
+	unsigned			num_gfx_rings;
+	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES];
+	unsigned			num_compute_rings;
+	struct amdgpu_irq_src		eop_irq;
+	struct amdgpu_irq_src		priv_reg_irq;
+	struct amdgpu_irq_src		priv_inst_irq;
+	struct amdgpu_irq_src		bad_op_irq;
+	struct amdgpu_irq_src		cp_ecc_error_irq;
+	struct amdgpu_irq_src		sq_irq;
+	struct amdgpu_irq_src		rlc_gc_fed_irq;
+	struct sq_work			sq_work;
+
+	/* gfx status */
+	uint32_t			gfx_current_status;
+	/* ce ram size*/
+	unsigned			ce_ram_size;
+	struct amdgpu_cu_info		cu_info;
+	const struct amdgpu_gfx_funcs	*funcs;
+
+	/* reset mask */
+	uint32_t                        grbm_soft_reset;
+	uint32_t                        srbm_soft_reset;
+	uint32_t 			gfx_supported_reset;
+	uint32_t 			compute_supported_reset;
+
+	/* gfx off */
+	bool                            gfx_off_state;      /* true: enabled, false: disabled */
+	struct mutex                    gfx_off_mutex;      /* mutex to change gfxoff state */
+	uint32_t                        gfx_off_req_count;  /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
+	struct delayed_work             gfx_off_delay_work; /* async work to set gfx block off */
+	uint32_t                        gfx_off_residency;  /* last logged residency */
+	uint64_t                        gfx_off_entrycount; /* count of times GPU has get into GFXOFF state */
+
+	/* pipe reservation */
+	struct mutex			pipe_reserve_mutex;
+	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
+
+	/*ras */
+	struct ras_common_if		*ras_if;
+	struct amdgpu_gfx_ras		*ras;
+
+	bool				is_poweron;
+
+	struct amdgpu_ring		sw_gfx_ring[AMDGPU_MAX_SW_GFX_RINGS];
+	struct amdgpu_ring_mux          muxer;
+
+	bool				cp_gfx_shadow; /* for gfx11 */
+
+	uint16_t 			xcc_mask;
+	uint32_t			num_xcc_per_xcp;
+	struct mutex			partition_mutex;
+	bool				mcbp; /* mid command buffer preemption */
+
+	/* IP reg dump */
+	uint32_t			*ip_dump_core;
+	uint32_t			*ip_dump_compute_queues;
+	uint32_t			*ip_dump_gfx_queues;
+
+	struct mutex			reset_sem_mutex;
+
+	/* cleaner shader */
+	struct amdgpu_bo		*cleaner_shader_obj;
+	unsigned int                    cleaner_shader_size;
+	u64				cleaner_shader_gpu_addr;
+	void				*cleaner_shader_cpu_ptr;
+	const void			*cleaner_shader_ptr;
+	bool				enable_cleaner_shader;
+	struct amdgpu_isolation_work	enforce_isolation[MAX_XCP];
+	/* Mutex for synchronizing KFD scheduler operations */
+	struct mutex                    userq_sch_mutex;
+	u64				userq_sch_req_count[MAX_XCP];
+	bool				userq_sch_inactive[MAX_XCP];
+	unsigned long			enforce_isolation_jiffies[MAX_XCP];
+	unsigned long			enforce_isolation_time[MAX_XCP];
+
+	atomic_t			total_submission_cnt;
+	struct delayed_work		idle_work;
+	bool				workload_profile_active;
+	struct mutex                    workload_profile_mutex;
+
+	bool				disable_kq;
+	bool				disable_uq;
+};
+
+struct amdgpu_gfx_ras_reg_entry {
+	struct amdgpu_ras_err_status_reg_entry reg_entry;
+	enum amdgpu_gfx_ras_mem_id_type mem_id_type;
+	uint32_t se_num;
+};
+
+struct amdgpu_gfx_ras_mem_id_entry {
+	const struct amdgpu_ras_memory_id_entry *mem_id_ent;
+	uint32_t size;
+};
+
+#define AMDGPU_GFX_MEMID_ENT(x) {(x), ARRAY_SIZE(x)},
+
+#define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
+#define amdgpu_gfx_select_se_sh(adev, se, sh, instance, xcc_id) ((adev)->gfx.funcs->select_se_sh((adev), (se), (sh), (instance), (xcc_id)))
+#define amdgpu_gfx_select_me_pipe_q(adev, me, pipe, q, vmid, xcc_id) ((adev)->gfx.funcs->select_me_pipe_q((adev), (me), (pipe), (q), (vmid), (xcc_id)))
+#define amdgpu_gfx_init_spm_golden(adev) (adev)->gfx.funcs->init_spm_golden((adev))
+#define amdgpu_gfx_get_gfx_shadow_info(adev, si) ((adev)->gfx.funcs->get_gfx_shadow_info((adev), (si), false))
+
+/**
+ * amdgpu_gfx_create_bitmask - create a bitmask
+ *
+ * @bit_width: length of the mask
+ *
+ * create a variable length bit mask.
+ * Returns the bitmask.
+ */
+static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width)
+{
+	return (u32)((1ULL << bit_width) - 1);
+}
 
 void amdgpu_gfx_parse_disable_cu(unsigned *mask, unsigned max_se,
-		unsigned max_sh);
+				 unsigned max_sh);
+
+int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, int xcc_id);
+
+void amdgpu_gfx_kiq_free_ring(struct amdgpu_ring *ring);
+
+void amdgpu_gfx_kiq_fini(struct amdgpu_device *adev, int xcc_id);
+int amdgpu_gfx_kiq_init(struct amdgpu_device *adev,
+			unsigned hpd_size, int xcc_id);
+
+int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
+			   unsigned mqd_size, int xcc_id);
+void amdgpu_gfx_mqd_sw_fini(struct amdgpu_device *adev, int xcc_id);
+int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id);
+int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev, int xcc_id);
+int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id);
+int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id);
+
+void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev);
+void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev);
+
+int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
+				int pipe, int queue);
+void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
+				 int *mec, int *pipe, int *queue);
+bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int xcc_id,
+				     int mec, int pipe, int queue);
+bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
+					       struct amdgpu_ring *ring);
+bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev,
+						struct amdgpu_ring *ring);
+bool amdgpu_gfx_is_me_queue_enabled(struct amdgpu_device *adev, int me,
+				    int pipe, int queue);
+void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable);
+void amdgpu_gfx_off_ctrl_immediate(struct amdgpu_device *adev, bool enable);
+int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value);
+int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block);
+void amdgpu_gfx_ras_fini(struct amdgpu_device *adev);
+int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value);
+int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *residency);
+int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value);
+int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
+		void *err_data,
+		struct amdgpu_iv_entry *entry);
+int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
+				  struct amdgpu_irq_src *source,
+				  struct amdgpu_iv_entry *entry);
+uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg, uint32_t xcc_id);
+void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, uint32_t xcc_id);
+int amdgpu_kiq_hdp_flush(struct amdgpu_device *adev);
+int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev);
+void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id);
+
+int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev);
+int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev,
+						struct amdgpu_iv_entry *entry);
+
+bool amdgpu_gfx_is_master_xcc(struct amdgpu_device *adev, int xcc_id);
+int amdgpu_gfx_sysfs_init(struct amdgpu_device *adev);
+void amdgpu_gfx_sysfs_fini(struct amdgpu_device *adev);
+void amdgpu_gfx_ras_error_func(struct amdgpu_device *adev,
+		void *ras_error_status,
+		void (*func)(struct amdgpu_device *adev, void *ras_error_status,
+				int xcc_id));
+int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev,
+				      unsigned int cleaner_shader_size);
+void amdgpu_gfx_cleaner_shader_sw_fini(struct amdgpu_device *adev);
+void amdgpu_gfx_cleaner_shader_init(struct amdgpu_device *adev,
+				    unsigned int cleaner_shader_size,
+				    const void *cleaner_shader_ptr);
+void amdgpu_gfx_enforce_isolation_handler(struct work_struct *work);
+void amdgpu_gfx_enforce_isolation_ring_begin_use(struct amdgpu_ring *ring);
+void amdgpu_gfx_enforce_isolation_ring_end_use(struct amdgpu_ring *ring);
+
+void amdgpu_gfx_profile_idle_work_handler(struct work_struct *work);
+void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring);
+void amdgpu_gfx_profile_ring_end_use(struct amdgpu_ring *ring);
+u32 amdgpu_gfx_csb_preamble_start(u32 *buffer);
+u32 amdgpu_gfx_csb_data_parser(struct amdgpu_device *adev, u32 *buffer, u32 count);
+void amdgpu_gfx_csb_preamble_end(u32 *buffer, u32 count);
+
+void amdgpu_debugfs_gfx_sched_mask_init(struct amdgpu_device *adev);
+void amdgpu_debugfs_compute_sched_mask_init(struct amdgpu_device *adev);
+
+static inline const char *amdgpu_gfx_compute_mode_desc(int mode)
+{
+	switch (mode) {
+	case AMDGPU_SPX_PARTITION_MODE:
+		return "SPX";
+	case AMDGPU_DPX_PARTITION_MODE:
+		return "DPX";
+	case AMDGPU_TPX_PARTITION_MODE:
+		return "TPX";
+	case AMDGPU_QPX_PARTITION_MODE:
+		return "QPX";
+	case AMDGPU_CPX_PARTITION_MODE:
+		return "CPX";
+	default:
+		return "UNKNOWN";
+	}
+}
 
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h
new file mode 100644
index 000000000000..c7b44aeb671b
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_GFXHUB_H__
+#define __AMDGPU_GFXHUB_H__
+
+struct amdgpu_gfxhub_funcs {
+	u64 (*get_fb_location)(struct amdgpu_device *adev);
+	u64 (*get_mc_fb_offset)(struct amdgpu_device *adev);
+	void (*setup_vm_pt_regs)(struct amdgpu_device *adev, uint32_t vmid,
+			uint64_t page_table_base);
+	int (*gart_enable)(struct amdgpu_device *adev);
+
+	void (*gart_disable)(struct amdgpu_device *adev);
+	void (*set_fault_enable_default)(struct amdgpu_device *adev, bool value);
+	void (*init)(struct amdgpu_device *adev);
+	int (*get_xgmi_info)(struct amdgpu_device *adev);
+	void (*utcl2_harvest)(struct amdgpu_device *adev);
+	void (*mode2_save_regs)(struct amdgpu_device *adev);
+	void (*mode2_restore_regs)(struct amdgpu_device *adev);
+	void (*halt)(struct amdgpu_device *adev);
+};
+
+struct amdgpu_gfxhub {
+	const struct amdgpu_gfxhub_funcs *funcs;
+};
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
new file mode 100644
index 000000000000..869bceb0fe2c
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
@@ -0,0 +1,1682 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+
+#include <linux/io-64-nonatomic-lo-hi.h>
+#ifdef CONFIG_X86
+#include <asm/hypervisor.h>
+#endif
+
+#include "amdgpu.h"
+#include "amdgpu_gmc.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_reset.h"
+#include "amdgpu_xgmi.h"
+
+#include <drm/drm_drv.h>
+#include <drm/ttm/ttm_tt.h>
+
+static const u64 four_gb = 0x100000000ULL;
+
+bool amdgpu_gmc_is_pdb0_enabled(struct amdgpu_device *adev)
+{
+	return adev->gmc.xgmi.connected_to_cpu || amdgpu_virt_xgmi_migrate_enabled(adev);
+}
+
+/**
+ * amdgpu_gmc_pdb0_alloc - allocate vram for pdb0
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Allocate video memory for pdb0 and map it for CPU access
+ * Returns 0 for success, error for failure.
+ */
+int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev)
+{
+	int r;
+	struct amdgpu_bo_param bp;
+	u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
+	uint32_t pde0_page_shift = adev->gmc.vmid0_page_table_block_size + 21;
+	uint32_t npdes = (vram_size + (1ULL << pde0_page_shift) - 1) >> pde0_page_shift;
+
+	memset(&bp, 0, sizeof(bp));
+	bp.size = PAGE_ALIGN((npdes + 1) * 8);
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+	bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+		AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+	bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+
+	r = amdgpu_bo_create(adev, &bp, &adev->gmc.pdb0_bo);
+	if (r)
+		return r;
+
+	r = amdgpu_bo_reserve(adev->gmc.pdb0_bo, false);
+	if (unlikely(r != 0))
+		goto bo_reserve_failure;
+
+	r = amdgpu_bo_pin(adev->gmc.pdb0_bo, AMDGPU_GEM_DOMAIN_VRAM);
+	if (r)
+		goto bo_pin_failure;
+	r = amdgpu_bo_kmap(adev->gmc.pdb0_bo, &adev->gmc.ptr_pdb0);
+	if (r)
+		goto bo_kmap_failure;
+
+	amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
+	return 0;
+
+bo_kmap_failure:
+	amdgpu_bo_unpin(adev->gmc.pdb0_bo);
+bo_pin_failure:
+	amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
+bo_reserve_failure:
+	amdgpu_bo_unref(&adev->gmc.pdb0_bo);
+	return r;
+}
+
+/**
+ * amdgpu_gmc_get_pde_for_bo - get the PDE for a BO
+ *
+ * @bo: the BO to get the PDE for
+ * @level: the level in the PD hirarchy
+ * @addr: resulting addr
+ * @flags: resulting flags
+ *
+ * Get the address and flags to be used for a PDE (Page Directory Entry).
+ */
+void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
+			       uint64_t *addr, uint64_t *flags)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+
+	switch (bo->tbo.resource->mem_type) {
+	case TTM_PL_TT:
+		*addr = bo->tbo.ttm->dma_address[0];
+		break;
+	case TTM_PL_VRAM:
+		*addr = amdgpu_bo_gpu_offset(bo);
+		break;
+	default:
+		*addr = 0;
+		break;
+	}
+	*flags = amdgpu_ttm_tt_pde_flags(bo->tbo.ttm, bo->tbo.resource);
+	amdgpu_gmc_get_vm_pde(adev, level, addr, flags);
+}
+
+/*
+ * amdgpu_gmc_pd_addr - return the address of the root directory
+ */
+uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	uint64_t pd_addr;
+
+	/* TODO: move that into ASIC specific code */
+	if (adev->asic_type >= CHIP_VEGA10) {
+		uint64_t flags = AMDGPU_PTE_VALID;
+
+		amdgpu_gmc_get_pde_for_bo(bo, -1, &pd_addr, &flags);
+		pd_addr |= flags;
+	} else {
+		pd_addr = amdgpu_bo_gpu_offset(bo);
+	}
+	return pd_addr;
+}
+
+/**
+ * amdgpu_gmc_set_pte_pde - update the page tables using CPU
+ *
+ * @adev: amdgpu_device pointer
+ * @cpu_pt_addr: cpu address of the page table
+ * @gpu_page_idx: entry in the page table to update
+ * @addr: dst addr to write into pte/pde
+ * @flags: access flags
+ *
+ * Update the page tables using CPU.
+ */
+int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
+				uint32_t gpu_page_idx, uint64_t addr,
+				uint64_t flags)
+{
+	void __iomem *ptr = (void *)cpu_pt_addr;
+	uint64_t value;
+
+	/*
+	 * The following is for PTE only. GART does not have PDEs.
+	*/
+	value = addr & 0x0000FFFFFFFFF000ULL;
+	value |= flags;
+	writeq(value, ptr + (gpu_page_idx * 8));
+
+	return 0;
+}
+
+/**
+ * amdgpu_gmc_agp_addr - return the address in the AGP address space
+ *
+ * @bo: TTM BO which needs the address, must be in GTT domain
+ *
+ * Tries to figure out how to access the BO through the AGP aperture. Returns
+ * AMDGPU_BO_INVALID_OFFSET if that is not possible.
+ */
+uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
+
+	if (!bo->ttm)
+		return AMDGPU_BO_INVALID_OFFSET;
+
+	if (bo->ttm->num_pages != 1 || bo->ttm->caching == ttm_cached)
+		return AMDGPU_BO_INVALID_OFFSET;
+
+	if (bo->ttm->dma_address[0] + PAGE_SIZE >= adev->gmc.agp_size)
+		return AMDGPU_BO_INVALID_OFFSET;
+
+	return adev->gmc.agp_start + bo->ttm->dma_address[0];
+}
+
+/**
+ * amdgpu_gmc_vram_location - try to find VRAM location
+ *
+ * @adev: amdgpu device structure holding all necessary information
+ * @mc: memory controller structure holding memory information
+ * @base: base address at which to put VRAM
+ *
+ * Function will try to place VRAM at base address provided
+ * as parameter.
+ */
+void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
+			      u64 base)
+{
+	uint64_t vis_limit = (uint64_t)amdgpu_vis_vram_limit << 20;
+	uint64_t limit = (uint64_t)amdgpu_vram_limit << 20;
+
+	mc->vram_start = base;
+	mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
+	if (limit < mc->real_vram_size)
+		mc->real_vram_size = limit;
+
+	if (vis_limit && vis_limit < mc->visible_vram_size)
+		mc->visible_vram_size = vis_limit;
+
+	if (mc->real_vram_size < mc->visible_vram_size)
+		mc->visible_vram_size = mc->real_vram_size;
+
+	if (mc->xgmi.num_physical_nodes == 0) {
+		mc->fb_start = mc->vram_start;
+		mc->fb_end = mc->vram_end;
+	}
+	dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
+			mc->mc_vram_size >> 20, mc->vram_start,
+			mc->vram_end, mc->real_vram_size >> 20);
+}
+
+/** amdgpu_gmc_sysvm_location - place vram and gart in sysvm aperture
+ *
+ * @adev: amdgpu device structure holding all necessary information
+ * @mc: memory controller structure holding memory information
+ *
+ * This function is only used if use GART for FB translation. In such
+ * case, we use sysvm aperture (vmid0 page tables) for both vram
+ * and gart (aka system memory) access.
+ *
+ * GPUVM (and our organization of vmid0 page tables) require sysvm
+ * aperture to be placed at a location aligned with 8 times of native
+ * page size. For example, if vm_context0_cntl.page_table_block_size
+ * is 12, then native page size is 8G (2M*2^12), sysvm should start
+ * with a 64G aligned address. For simplicity, we just put sysvm at
+ * address 0. So vram start at address 0 and gart is right after vram.
+ */
+void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc)
+{
+	u64 hive_vram_start = 0;
+	u64 hive_vram_end = mc->xgmi.node_segment_size * mc->xgmi.num_physical_nodes - 1;
+	mc->vram_start = mc->xgmi.node_segment_size * mc->xgmi.physical_node_id;
+	mc->vram_end = mc->vram_start + mc->xgmi.node_segment_size - 1;
+	/* node_segment_size may not 4GB aligned on SRIOV, align up is needed. */
+	mc->gart_start = ALIGN(hive_vram_end + 1, four_gb);
+	mc->gart_end = mc->gart_start + mc->gart_size - 1;
+	if (amdgpu_virt_xgmi_migrate_enabled(adev)) {
+		/* set mc->vram_start to 0 to switch the returned GPU address of
+		 * amdgpu_bo_create_reserved() from FB aperture to GART aperture.
+		 */
+		mc->vram_start = 0;
+		mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
+		mc->visible_vram_size = min(mc->visible_vram_size, mc->real_vram_size);
+	} else {
+		mc->fb_start = hive_vram_start;
+		mc->fb_end = hive_vram_end;
+	}
+	dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
+			mc->mc_vram_size >> 20, mc->vram_start,
+			mc->vram_end, mc->real_vram_size >> 20);
+	dev_info(adev->dev, "GART: %lluM 0x%016llX - 0x%016llX\n",
+			mc->gart_size >> 20, mc->gart_start, mc->gart_end);
+}
+
+/**
+ * amdgpu_gmc_gart_location - try to find GART location
+ *
+ * @adev: amdgpu device structure holding all necessary information
+ * @mc: memory controller structure holding memory information
+ * @gart_placement: GART placement policy with respect to VRAM
+ *
+ * Function will try to place GART before or after VRAM.
+ * If GART size is bigger than space left then we ajust GART size.
+ * Thus function will never fails.
+ */
+void amdgpu_gmc_gart_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
+			      enum amdgpu_gart_placement gart_placement)
+{
+	u64 size_af, size_bf;
+	/*To avoid the hole, limit the max mc address to AMDGPU_GMC_HOLE_START*/
+	u64 max_mc_address = min(adev->gmc.mc_mask, AMDGPU_GMC_HOLE_START - 1);
+
+	/* VCE doesn't like it when BOs cross a 4GB segment, so align
+	 * the GART base on a 4GB boundary as well.
+	 */
+	size_bf = mc->fb_start;
+	size_af = max_mc_address + 1 - ALIGN(mc->fb_end + 1, four_gb);
+
+	if (mc->gart_size > max(size_bf, size_af)) {
+		dev_warn(adev->dev, "limiting GART\n");
+		mc->gart_size = max(size_bf, size_af);
+	}
+
+	switch (gart_placement) {
+	case AMDGPU_GART_PLACEMENT_HIGH:
+		mc->gart_start = max_mc_address - mc->gart_size + 1;
+		break;
+	case AMDGPU_GART_PLACEMENT_LOW:
+		mc->gart_start = 0;
+		break;
+	case AMDGPU_GART_PLACEMENT_BEST_FIT:
+	default:
+		if ((size_bf >= mc->gart_size && size_bf < size_af) ||
+		    (size_af < mc->gart_size))
+			mc->gart_start = 0;
+		else
+			mc->gart_start = max_mc_address - mc->gart_size + 1;
+		break;
+	}
+
+	mc->gart_start &= ~(four_gb - 1);
+	mc->gart_end = mc->gart_start + mc->gart_size - 1;
+	dev_info(adev->dev, "GART: %lluM 0x%016llX - 0x%016llX\n",
+			mc->gart_size >> 20, mc->gart_start, mc->gart_end);
+}
+
+/**
+ * amdgpu_gmc_agp_location - try to find AGP location
+ * @adev: amdgpu device structure holding all necessary information
+ * @mc: memory controller structure holding memory information
+ *
+ * Function will place try to find a place for the AGP BAR in the MC address
+ * space.
+ *
+ * AGP BAR will be assigned the largest available hole in the address space.
+ * Should be called after VRAM and GART locations are setup.
+ */
+void amdgpu_gmc_agp_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc)
+{
+	const uint64_t sixteen_gb = 1ULL << 34;
+	const uint64_t sixteen_gb_mask = ~(sixteen_gb - 1);
+	u64 size_af, size_bf;
+
+	if (mc->fb_start > mc->gart_start) {
+		size_bf = (mc->fb_start & sixteen_gb_mask) -
+			ALIGN(mc->gart_end + 1, sixteen_gb);
+		size_af = mc->mc_mask + 1 - ALIGN(mc->fb_end + 1, sixteen_gb);
+	} else {
+		size_bf = mc->fb_start & sixteen_gb_mask;
+		size_af = (mc->gart_start & sixteen_gb_mask) -
+			ALIGN(mc->fb_end + 1, sixteen_gb);
+	}
+
+	if (size_bf > size_af) {
+		mc->agp_start = (mc->fb_start - size_bf) & sixteen_gb_mask;
+		mc->agp_size = size_bf;
+	} else {
+		mc->agp_start = ALIGN(mc->fb_end + 1, sixteen_gb);
+		mc->agp_size = size_af;
+	}
+
+	mc->agp_end = mc->agp_start + mc->agp_size - 1;
+	dev_info(adev->dev, "AGP: %lluM 0x%016llX - 0x%016llX\n",
+			mc->agp_size >> 20, mc->agp_start, mc->agp_end);
+}
+
+/**
+ * amdgpu_gmc_set_agp_default - Set the default AGP aperture value.
+ * @adev: amdgpu device structure holding all necessary information
+ * @mc: memory controller structure holding memory information
+ *
+ * To disable the AGP aperture, you need to set the start to a larger
+ * value than the end.  This function sets the default value which
+ * can then be overridden using amdgpu_gmc_agp_location() if you want
+ * to enable the AGP aperture on a specific chip.
+ *
+ */
+void amdgpu_gmc_set_agp_default(struct amdgpu_device *adev,
+				struct amdgpu_gmc *mc)
+{
+	mc->agp_start = 0xffffffffffff;
+	mc->agp_end = 0;
+	mc->agp_size = 0;
+}
+
+/**
+ * amdgpu_gmc_fault_key - get hask key from vm fault address and pasid
+ *
+ * @addr: 48 bit physical address, page aligned (36 significant bits)
+ * @pasid: 16 bit process address space identifier
+ */
+static inline uint64_t amdgpu_gmc_fault_key(uint64_t addr, uint16_t pasid)
+{
+	return addr << 4 | pasid;
+}
+
+/**
+ * amdgpu_gmc_filter_faults - filter VM faults
+ *
+ * @adev: amdgpu device structure
+ * @ih: interrupt ring that the fault received from
+ * @addr: address of the VM fault
+ * @pasid: PASID of the process causing the fault
+ * @timestamp: timestamp of the fault
+ *
+ * Returns:
+ * True if the fault was filtered and should not be processed further.
+ * False if the fault is a new one and needs to be handled.
+ */
+bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev,
+			      struct amdgpu_ih_ring *ih, uint64_t addr,
+			      uint16_t pasid, uint64_t timestamp)
+{
+	struct amdgpu_gmc *gmc = &adev->gmc;
+	uint64_t stamp, key = amdgpu_gmc_fault_key(addr, pasid);
+	struct amdgpu_gmc_fault *fault;
+	uint32_t hash;
+
+	/* Stale retry fault if timestamp goes backward */
+	if (amdgpu_ih_ts_after(timestamp, ih->processed_timestamp))
+		return true;
+
+	/* If we don't have space left in the ring buffer return immediately */
+	stamp = max(timestamp, AMDGPU_GMC_FAULT_TIMEOUT + 1) -
+		AMDGPU_GMC_FAULT_TIMEOUT;
+	if (gmc->fault_ring[gmc->last_fault].timestamp >= stamp)
+		return true;
+
+	/* Try to find the fault in the hash */
+	hash = hash_64(key, AMDGPU_GMC_FAULT_HASH_ORDER);
+	fault = &gmc->fault_ring[gmc->fault_hash[hash].idx];
+	while (fault->timestamp >= stamp) {
+		uint64_t tmp;
+
+		if (atomic64_read(&fault->key) == key) {
+			/*
+			 * if we get a fault which is already present in
+			 * the fault_ring and the timestamp of
+			 * the fault is after the expired timestamp,
+			 * then this is a new fault that needs to be added
+			 * into the fault ring.
+			 */
+			if (fault->timestamp_expiry != 0 &&
+			    amdgpu_ih_ts_after(fault->timestamp_expiry,
+					       timestamp))
+				break;
+			else
+				return true;
+		}
+
+		tmp = fault->timestamp;
+		fault = &gmc->fault_ring[fault->next];
+
+		/* Check if the entry was reused */
+		if (fault->timestamp >= tmp)
+			break;
+	}
+
+	/* Add the fault to the ring */
+	fault = &gmc->fault_ring[gmc->last_fault];
+	atomic64_set(&fault->key, key);
+	fault->timestamp = timestamp;
+
+	/* And update the hash */
+	fault->next = gmc->fault_hash[hash].idx;
+	gmc->fault_hash[hash].idx = gmc->last_fault++;
+	return false;
+}
+
+/**
+ * amdgpu_gmc_filter_faults_remove - remove address from VM faults filter
+ *
+ * @adev: amdgpu device structure
+ * @addr: address of the VM fault
+ * @pasid: PASID of the process causing the fault
+ *
+ * Remove the address from fault filter, then future vm fault on this address
+ * will pass to retry fault handler to recover.
+ */
+void amdgpu_gmc_filter_faults_remove(struct amdgpu_device *adev, uint64_t addr,
+				     uint16_t pasid)
+{
+	struct amdgpu_gmc *gmc = &adev->gmc;
+	uint64_t key = amdgpu_gmc_fault_key(addr, pasid);
+	struct amdgpu_ih_ring *ih;
+	struct amdgpu_gmc_fault *fault;
+	uint32_t last_wptr;
+	uint64_t last_ts;
+	uint32_t hash;
+	uint64_t tmp;
+
+	if (adev->irq.retry_cam_enabled)
+		return;
+
+	ih = &adev->irq.ih1;
+	/* Get the WPTR of the last entry in IH ring */
+	last_wptr = amdgpu_ih_get_wptr(adev, ih);
+	/* Order wptr with ring data. */
+	rmb();
+	/* Get the timetamp of the last entry in IH ring */
+	last_ts = amdgpu_ih_decode_iv_ts(adev, ih, last_wptr, -1);
+
+	hash = hash_64(key, AMDGPU_GMC_FAULT_HASH_ORDER);
+	fault = &gmc->fault_ring[gmc->fault_hash[hash].idx];
+	do {
+		if (atomic64_read(&fault->key) == key) {
+			/*
+			 * Update the timestamp when this fault
+			 * expired.
+			 */
+			fault->timestamp_expiry = last_ts;
+			break;
+		}
+
+		tmp = fault->timestamp;
+		fault = &gmc->fault_ring[fault->next];
+	} while (fault->timestamp < tmp);
+}
+
+int amdgpu_gmc_ras_sw_init(struct amdgpu_device *adev)
+{
+	int r;
+
+	/* umc ras block */
+	r = amdgpu_umc_ras_sw_init(adev);
+	if (r)
+		return r;
+
+	/* mmhub ras block */
+	r = amdgpu_mmhub_ras_sw_init(adev);
+	if (r)
+		return r;
+
+	/* hdp ras block */
+	r = amdgpu_hdp_ras_sw_init(adev);
+	if (r)
+		return r;
+
+	/* mca.x ras block */
+	r = amdgpu_mca_mp0_ras_sw_init(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_mca_mp1_ras_sw_init(adev);
+	if (r)
+		return r;
+
+	r = amdgpu_mca_mpio_ras_sw_init(adev);
+	if (r)
+		return r;
+
+	/* xgmi ras block */
+	r = amdgpu_xgmi_ras_sw_init(adev);
+	if (r)
+		return r;
+
+	return 0;
+}
+
+int amdgpu_gmc_ras_late_init(struct amdgpu_device *adev)
+{
+	return 0;
+}
+
+void amdgpu_gmc_ras_fini(struct amdgpu_device *adev)
+{
+
+}
+
+	/*
+	 * The latest engine allocation on gfx9/10 is:
+	 * Engine 2, 3: firmware
+	 * Engine 0, 1, 4~16: amdgpu ring,
+	 *                    subject to change when ring number changes
+	 * Engine 17: Gart flushes
+	 */
+#define AMDGPU_VMHUB_INV_ENG_BITMAP		0x1FFF3
+
+int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev)
+{
+	struct amdgpu_ring *ring;
+	unsigned vm_inv_engs[AMDGPU_MAX_VMHUBS] = {0};
+	unsigned i;
+	unsigned vmhub, inv_eng;
+	struct amdgpu_ring *shared_ring;
+
+	/* init the vm inv eng for all vmhubs */
+	for_each_set_bit(i, adev->vmhubs_mask, AMDGPU_MAX_VMHUBS) {
+		vm_inv_engs[i] = AMDGPU_VMHUB_INV_ENG_BITMAP;
+		/* reserve engine 5 for firmware */
+		if (adev->enable_mes)
+			vm_inv_engs[i] &= ~(1 << 5);
+		/* reserve engine 6 for uni mes */
+		if (adev->enable_uni_mes)
+			vm_inv_engs[i] &= ~(1 << 6);
+		/* reserve mmhub engine 3 for firmware */
+		if (adev->enable_umsch_mm)
+			vm_inv_engs[i] &= ~(1 << 3);
+	}
+
+	for (i = 0; i < adev->num_rings; ++i) {
+		ring = adev->rings[i];
+		vmhub = ring->vm_hub;
+
+		if (ring == &adev->mes.ring[0] ||
+		    ring == &adev->mes.ring[1] ||
+		    ring == &adev->umsch_mm.ring ||
+		    ring == &adev->cper.ring_buf)
+			continue;
+
+		/* Skip if the ring is a shared ring */
+		if (amdgpu_sdma_is_shared_inv_eng(adev, ring))
+			continue;
+
+		inv_eng = ffs(vm_inv_engs[vmhub]);
+		if (!inv_eng) {
+			dev_err(adev->dev, "no VM inv eng for ring %s\n",
+				ring->name);
+			return -EINVAL;
+		}
+
+		ring->vm_inv_eng = inv_eng - 1;
+		vm_inv_engs[vmhub] &= ~(1 << ring->vm_inv_eng);
+
+		dev_info(adev->dev, "ring %s uses VM inv eng %u on hub %u\n",
+			 ring->name, ring->vm_inv_eng, ring->vm_hub);
+		/* SDMA has a special packet which allows it to use the same
+		 * invalidation engine for all the rings in one instance.
+		 * Therefore, we do not allocate a separate VM invalidation engine
+		 * for SDMA page rings. Instead, they share the VM invalidation
+		 * engine with the SDMA gfx ring. This change ensures efficient
+		 * resource management and avoids the issue of insufficient VM
+		 * invalidation engines.
+		 */
+		shared_ring = amdgpu_sdma_get_shared_ring(adev, ring);
+		if (shared_ring) {
+			shared_ring->vm_inv_eng = ring->vm_inv_eng;
+			dev_info(adev->dev, "ring %s shares VM invalidation engine %u with ring %s on hub %u\n",
+					ring->name, ring->vm_inv_eng, shared_ring->name, ring->vm_hub);
+			continue;
+		}
+	}
+
+	return 0;
+}
+
+void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
+			      uint32_t vmhub, uint32_t flush_type)
+{
+	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+	struct amdgpu_vmhub *hub = &adev->vmhub[vmhub];
+	struct dma_fence *fence;
+	struct amdgpu_job *job;
+	int r;
+
+	if (!hub->sdma_invalidation_workaround || vmid ||
+	    !adev->mman.buffer_funcs_enabled || !adev->ib_pool_ready ||
+	    !ring->sched.ready) {
+		/*
+		 * A GPU reset should flush all TLBs anyway, so no need to do
+		 * this while one is ongoing.
+		 */
+		if (!down_read_trylock(&adev->reset_domain->sem))
+			return;
+
+		if (adev->gmc.flush_tlb_needs_extra_type_2)
+			adev->gmc.gmc_funcs->flush_gpu_tlb(adev, vmid,
+							   vmhub, 2);
+
+		if (adev->gmc.flush_tlb_needs_extra_type_0 && flush_type == 2)
+			adev->gmc.gmc_funcs->flush_gpu_tlb(adev, vmid,
+							   vmhub, 0);
+
+		adev->gmc.gmc_funcs->flush_gpu_tlb(adev, vmid, vmhub,
+						   flush_type);
+		up_read(&adev->reset_domain->sem);
+		return;
+	}
+
+	/* The SDMA on Navi 1x has a bug which can theoretically result in memory
+	 * corruption if an invalidation happens at the same time as an VA
+	 * translation. Avoid this by doing the invalidation from the SDMA
+	 * itself at least for GART.
+	 */
+	mutex_lock(&adev->mman.gtt_window_lock);
+	r = amdgpu_job_alloc_with_ib(ring->adev, &adev->mman.high_pr,
+				     AMDGPU_FENCE_OWNER_UNDEFINED,
+				     16 * 4, AMDGPU_IB_POOL_IMMEDIATE,
+				     &job, AMDGPU_KERNEL_JOB_ID_FLUSH_GPU_TLB);
+	if (r)
+		goto error_alloc;
+
+	job->vm_pd_addr = amdgpu_gmc_pd_addr(adev->gart.bo);
+	job->vm_needs_flush = true;
+	job->ibs->ptr[job->ibs->length_dw++] = ring->funcs->nop;
+	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
+	fence = amdgpu_job_submit(job);
+	mutex_unlock(&adev->mman.gtt_window_lock);
+
+	dma_fence_wait(fence, false);
+	dma_fence_put(fence);
+
+	return;
+
+error_alloc:
+	mutex_unlock(&adev->mman.gtt_window_lock);
+	dev_err(adev->dev, "Error flushing GPU TLB using the SDMA (%d)!\n", r);
+}
+
+int amdgpu_gmc_flush_gpu_tlb_pasid(struct amdgpu_device *adev, uint16_t pasid,
+				   uint32_t flush_type, bool all_hub,
+				   uint32_t inst)
+{
+	struct amdgpu_ring *ring = &adev->gfx.kiq[inst].ring;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[inst];
+	unsigned int ndw;
+	int r, cnt = 0;
+	uint32_t seq;
+
+	/*
+	 * A GPU reset should flush all TLBs anyway, so no need to do
+	 * this while one is ongoing.
+	 */
+	if (!down_read_trylock(&adev->reset_domain->sem))
+		return 0;
+
+	if (!adev->gmc.flush_pasid_uses_kiq || !ring->sched.ready) {
+		if (adev->gmc.flush_tlb_needs_extra_type_2)
+			adev->gmc.gmc_funcs->flush_gpu_tlb_pasid(adev, pasid,
+								 2, all_hub,
+								 inst);
+
+		if (adev->gmc.flush_tlb_needs_extra_type_0 && flush_type == 2)
+			adev->gmc.gmc_funcs->flush_gpu_tlb_pasid(adev, pasid,
+								 0, all_hub,
+								 inst);
+
+		adev->gmc.gmc_funcs->flush_gpu_tlb_pasid(adev, pasid,
+							 flush_type, all_hub,
+							 inst);
+		r = 0;
+	} else {
+		/* 2 dwords flush + 8 dwords fence */
+		ndw = kiq->pmf->invalidate_tlbs_size + 8;
+
+		if (adev->gmc.flush_tlb_needs_extra_type_2)
+			ndw += kiq->pmf->invalidate_tlbs_size;
+
+		if (adev->gmc.flush_tlb_needs_extra_type_0)
+			ndw += kiq->pmf->invalidate_tlbs_size;
+
+		spin_lock(&adev->gfx.kiq[inst].ring_lock);
+		r = amdgpu_ring_alloc(ring, ndw);
+		if (r) {
+			spin_unlock(&adev->gfx.kiq[inst].ring_lock);
+			goto error_unlock_reset;
+		}
+		if (adev->gmc.flush_tlb_needs_extra_type_2)
+			kiq->pmf->kiq_invalidate_tlbs(ring, pasid, 2, all_hub);
+
+		if (flush_type == 2 && adev->gmc.flush_tlb_needs_extra_type_0)
+			kiq->pmf->kiq_invalidate_tlbs(ring, pasid, 0, all_hub);
+
+		kiq->pmf->kiq_invalidate_tlbs(ring, pasid, flush_type, all_hub);
+		r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
+		if (r) {
+			amdgpu_ring_undo(ring);
+			spin_unlock(&adev->gfx.kiq[inst].ring_lock);
+			goto error_unlock_reset;
+		}
+
+		amdgpu_ring_commit(ring);
+		spin_unlock(&adev->gfx.kiq[inst].ring_lock);
+
+		r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+
+		might_sleep();
+		while (r < 1 && cnt++ < MAX_KIQ_REG_TRY &&
+		       !amdgpu_reset_pending(adev->reset_domain)) {
+			msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
+			r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+		}
+
+		if (cnt > MAX_KIQ_REG_TRY) {
+			dev_err(adev->dev, "timeout waiting for kiq fence\n");
+			r = -ETIME;
+		} else
+			r = 0;
+	}
+
+error_unlock_reset:
+	up_read(&adev->reset_domain->sem);
+	return r;
+}
+
+void amdgpu_gmc_fw_reg_write_reg_wait(struct amdgpu_device *adev,
+				      uint32_t reg0, uint32_t reg1,
+				      uint32_t ref, uint32_t mask,
+				      uint32_t xcc_inst)
+{
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_inst];
+	struct amdgpu_ring *ring = &kiq->ring;
+	signed long r, cnt = 0;
+	unsigned long flags;
+	uint32_t seq;
+
+	if (adev->mes.ring[0].sched.ready) {
+		amdgpu_mes_reg_write_reg_wait(adev, reg0, reg1,
+					      ref, mask);
+		return;
+	}
+
+	spin_lock_irqsave(&kiq->ring_lock, flags);
+	amdgpu_ring_alloc(ring, 32);
+	amdgpu_ring_emit_reg_write_reg_wait(ring, reg0, reg1,
+					    ref, mask);
+	r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
+	if (r)
+		goto failed_undo;
+
+	amdgpu_ring_commit(ring);
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+
+	r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+
+	/* don't wait anymore for IRQ context */
+	if (r < 1 && in_interrupt())
+		goto failed_kiq;
+
+	might_sleep();
+	while (r < 1 && cnt++ < MAX_KIQ_REG_TRY &&
+	       !amdgpu_reset_pending(adev->reset_domain)) {
+
+		msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
+		r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
+	}
+
+	if (cnt > MAX_KIQ_REG_TRY)
+		goto failed_kiq;
+
+	return;
+
+failed_undo:
+	amdgpu_ring_undo(ring);
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+failed_kiq:
+	dev_err(adev->dev, "failed to write reg %x wait reg %x\n", reg0, reg1);
+}
+
+/**
+ * amdgpu_gmc_tmz_set -- check and set if a device supports TMZ
+ * @adev: amdgpu_device pointer
+ *
+ * Check and set if an the device @adev supports Trusted Memory
+ * Zones (TMZ).
+ */
+void amdgpu_gmc_tmz_set(struct amdgpu_device *adev)
+{
+	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	/* RAVEN */
+	case IP_VERSION(9, 2, 2):
+	case IP_VERSION(9, 1, 0):
+	/* RENOIR looks like RAVEN */
+	case IP_VERSION(9, 3, 0):
+	/* GC 10.3.7 */
+	case IP_VERSION(10, 3, 7):
+	/* GC 11.0.1 */
+	case IP_VERSION(11, 0, 1):
+		if (amdgpu_tmz == 0) {
+			adev->gmc.tmz_enabled = false;
+			dev_info(adev->dev,
+				 "Trusted Memory Zone (TMZ) feature disabled (cmd line)\n");
+		} else {
+			adev->gmc.tmz_enabled = true;
+			dev_info(adev->dev,
+				 "Trusted Memory Zone (TMZ) feature enabled\n");
+		}
+		break;
+	case IP_VERSION(10, 1, 10):
+	case IP_VERSION(10, 1, 1):
+	case IP_VERSION(10, 1, 2):
+	case IP_VERSION(10, 1, 3):
+	case IP_VERSION(10, 3, 0):
+	case IP_VERSION(10, 3, 2):
+	case IP_VERSION(10, 3, 4):
+	case IP_VERSION(10, 3, 5):
+	case IP_VERSION(10, 3, 6):
+	/* VANGOGH */
+	case IP_VERSION(10, 3, 1):
+	/* YELLOW_CARP*/
+	case IP_VERSION(10, 3, 3):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 1):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 5, 3):
+		/* Don't enable it by default yet.
+		 */
+		if (amdgpu_tmz < 1) {
+			adev->gmc.tmz_enabled = false;
+			dev_info(adev->dev,
+				 "Trusted Memory Zone (TMZ) feature disabled as experimental (default)\n");
+		} else {
+			adev->gmc.tmz_enabled = true;
+			dev_info(adev->dev,
+				 "Trusted Memory Zone (TMZ) feature enabled as experimental (cmd line)\n");
+		}
+		break;
+	default:
+		adev->gmc.tmz_enabled = false;
+		dev_info(adev->dev,
+			 "Trusted Memory Zone (TMZ) feature not supported\n");
+		break;
+	}
+}
+
+/**
+ * amdgpu_gmc_noretry_set -- set per asic noretry defaults
+ * @adev: amdgpu_device pointer
+ *
+ * Set a per asic default for the no-retry parameter.
+ *
+ */
+void amdgpu_gmc_noretry_set(struct amdgpu_device *adev)
+{
+	struct amdgpu_gmc *gmc = &adev->gmc;
+	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
+	bool noretry_default = (gc_ver == IP_VERSION(9, 0, 1) ||
+				gc_ver == IP_VERSION(9, 4, 0) ||
+				gc_ver == IP_VERSION(9, 4, 1) ||
+				gc_ver == IP_VERSION(9, 4, 2) ||
+				gc_ver == IP_VERSION(9, 4, 3) ||
+				gc_ver == IP_VERSION(9, 4, 4) ||
+				gc_ver == IP_VERSION(9, 5, 0) ||
+				gc_ver >= IP_VERSION(10, 3, 0));
+
+	if (!amdgpu_sriov_xnack_support(adev))
+		gmc->noretry = 1;
+	else
+		gmc->noretry = (amdgpu_noretry == -1) ? noretry_default : amdgpu_noretry;
+}
+
+void amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type,
+				   bool enable)
+{
+	struct amdgpu_vmhub *hub;
+	u32 tmp, reg, i;
+
+	hub = &adev->vmhub[hub_type];
+	for (i = 0; i < 16; i++) {
+		reg = hub->vm_context0_cntl + hub->ctx_distance * i;
+
+		tmp = (hub_type == AMDGPU_GFXHUB(0)) ?
+			RREG32_SOC15_IP(GC, reg) :
+			RREG32_SOC15_IP(MMHUB, reg);
+
+		if (enable)
+			tmp |= hub->vm_cntx_cntl_vm_fault;
+		else
+			tmp &= ~hub->vm_cntx_cntl_vm_fault;
+
+		(hub_type == AMDGPU_GFXHUB(0)) ?
+			WREG32_SOC15_IP(GC, reg, tmp) :
+			WREG32_SOC15_IP(MMHUB, reg, tmp);
+	}
+}
+
+void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev)
+{
+	unsigned size;
+
+	/*
+	 * Some ASICs need to reserve a region of video memory to avoid access
+	 * from driver
+	 */
+	adev->mman.stolen_reserved_offset = 0;
+	adev->mman.stolen_reserved_size = 0;
+
+	/*
+	 * TODO:
+	 * Currently there is a bug where some memory client outside
+	 * of the driver writes to first 8M of VRAM on S3 resume,
+	 * this overrides GART which by default gets placed in first 8M and
+	 * causes VM_FAULTS once GTT is accessed.
+	 * Keep the stolen memory reservation until the while this is not solved.
+	 */
+	switch (adev->asic_type) {
+	case CHIP_VEGA10:
+		adev->mman.keep_stolen_vga_memory = true;
+		/*
+		 * VEGA10 SRIOV VF with MS_HYPERV host needs some firmware reserved area.
+		 */
+#ifdef CONFIG_X86
+		if (amdgpu_sriov_vf(adev) && hypervisor_is_type(X86_HYPER_MS_HYPERV)) {
+			adev->mman.stolen_reserved_offset = 0x500000;
+			adev->mman.stolen_reserved_size = 0x200000;
+		}
+#endif
+		break;
+	case CHIP_RAVEN:
+	case CHIP_RENOIR:
+		adev->mman.keep_stolen_vga_memory = true;
+		break;
+	default:
+		adev->mman.keep_stolen_vga_memory = false;
+		break;
+	}
+
+	if (amdgpu_sriov_vf(adev) ||
+	    !amdgpu_device_has_display_hardware(adev)) {
+		size = 0;
+	} else {
+		size = amdgpu_gmc_get_vbios_fb_size(adev);
+
+		if (adev->mman.keep_stolen_vga_memory)
+			size = max(size, (unsigned)AMDGPU_VBIOS_VGA_ALLOCATION);
+	}
+
+	/* set to 0 if the pre-OS buffer uses up most of vram */
+	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
+		size = 0;
+
+	if (size > AMDGPU_VBIOS_VGA_ALLOCATION) {
+		adev->mman.stolen_vga_size = AMDGPU_VBIOS_VGA_ALLOCATION;
+		adev->mman.stolen_extended_size = size - adev->mman.stolen_vga_size;
+	} else {
+		adev->mman.stolen_vga_size = size;
+		adev->mman.stolen_extended_size = 0;
+	}
+}
+
+/**
+ * amdgpu_gmc_init_pdb0 - initialize PDB0
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * This function is only used when GART page table is used
+ * for FB address translatioin. In such a case, we construct
+ * a 2-level system VM page table: PDB0->PTB, to cover both
+ * VRAM of the hive and system memory.
+ *
+ * PDB0 is static, initialized once on driver initialization.
+ * The first n entries of PDB0 are used as PTE by setting
+ * P bit to 1, pointing to VRAM. The n+1'th entry points
+ * to a big PTB covering system memory.
+ *
+ */
+void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev)
+{
+	int i;
+	uint64_t flags = adev->gart.gart_pte_flags; //TODO it is UC. explore NC/RW?
+	/* Each PDE0 (used as PTE) covers (2^vmid0_page_table_block_size)*2M
+	 */
+	u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
+	u64 pde0_page_size = (1ULL<<adev->gmc.vmid0_page_table_block_size)<<21;
+	u64 vram_addr, vram_end;
+	u64 gart_ptb_gpu_pa = amdgpu_gmc_vram_pa(adev, adev->gart.bo);
+	int idx;
+
+	if (!drm_dev_enter(adev_to_drm(adev), &idx))
+		return;
+
+	flags |= AMDGPU_PTE_VALID | AMDGPU_PTE_READABLE;
+	flags |= AMDGPU_PTE_WRITEABLE;
+	flags |= AMDGPU_PTE_SNOOPED;
+	flags |= AMDGPU_PTE_FRAG((adev->gmc.vmid0_page_table_block_size + 9*1));
+	flags |= AMDGPU_PDE_PTE_FLAG(adev);
+
+	vram_addr = adev->vm_manager.vram_base_offset;
+	if (!amdgpu_virt_xgmi_migrate_enabled(adev))
+		vram_addr -= adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
+	vram_end = vram_addr + vram_size;
+
+	/* The first n PDE0 entries are used as PTE,
+	 * pointing to vram
+	 */
+	for (i = 0; vram_addr < vram_end; i++, vram_addr += pde0_page_size)
+		amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, vram_addr, flags);
+
+	/* The n+1'th PDE0 entry points to a huge
+	 * PTB who has more than 512 entries each
+	 * pointing to a 4K system page
+	 */
+	flags = AMDGPU_PTE_VALID;
+	flags |= AMDGPU_PTE_SNOOPED | AMDGPU_PDE_BFS_FLAG(adev, 0);
+	/* Requires gart_ptb_gpu_pa to be 4K aligned */
+	amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, gart_ptb_gpu_pa, flags);
+	drm_dev_exit(idx);
+}
+
+/**
+ * amdgpu_gmc_vram_mc2pa - calculate vram buffer's physical address from MC
+ * address
+ *
+ * @adev: amdgpu_device pointer
+ * @mc_addr: MC address of buffer
+ */
+uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr)
+{
+	return mc_addr - adev->gmc.vram_start + adev->vm_manager.vram_base_offset;
+}
+
+/**
+ * amdgpu_gmc_vram_pa - calculate vram buffer object's physical address from
+ * GPU's view
+ *
+ * @adev: amdgpu_device pointer
+ * @bo: amdgpu buffer object
+ */
+uint64_t amdgpu_gmc_vram_pa(struct amdgpu_device *adev, struct amdgpu_bo *bo)
+{
+	return amdgpu_gmc_vram_mc2pa(adev, amdgpu_bo_gpu_offset(bo));
+}
+
+int amdgpu_gmc_vram_checking(struct amdgpu_device *adev)
+{
+	struct amdgpu_bo *vram_bo = NULL;
+	uint64_t vram_gpu = 0;
+	void *vram_ptr = NULL;
+
+	int ret, size = 0x100000;
+	uint8_t cptr[10];
+
+	ret = amdgpu_bo_create_kernel(adev, size, PAGE_SIZE,
+				AMDGPU_GEM_DOMAIN_VRAM,
+				&vram_bo,
+				&vram_gpu,
+				&vram_ptr);
+	if (ret)
+		return ret;
+
+	memset(vram_ptr, 0x86, size);
+	memset(cptr, 0x86, 10);
+
+	/**
+	 * Check the start, the mid, and the end of the memory if the content of
+	 * each byte is the pattern "0x86". If yes, we suppose the vram bo is
+	 * workable.
+	 *
+	 * Note: If check the each byte of whole 1M bo, it will cost too many
+	 * seconds, so here, we just pick up three parts for emulation.
+	 */
+	ret = memcmp(vram_ptr, cptr, 10);
+	if (ret) {
+		ret = -EIO;
+		goto release_buffer;
+	}
+
+	ret = memcmp(vram_ptr + (size / 2), cptr, 10);
+	if (ret) {
+		ret = -EIO;
+		goto release_buffer;
+	}
+
+	ret = memcmp(vram_ptr + size - 10, cptr, 10);
+	if (ret) {
+		ret = -EIO;
+		goto release_buffer;
+	}
+
+release_buffer:
+	amdgpu_bo_free_kernel(&vram_bo, &vram_gpu,
+			&vram_ptr);
+
+	return ret;
+}
+
+static const char *nps_desc[] = {
+	[AMDGPU_NPS1_PARTITION_MODE] = "NPS1",
+	[AMDGPU_NPS2_PARTITION_MODE] = "NPS2",
+	[AMDGPU_NPS3_PARTITION_MODE] = "NPS3",
+	[AMDGPU_NPS4_PARTITION_MODE] = "NPS4",
+	[AMDGPU_NPS6_PARTITION_MODE] = "NPS6",
+	[AMDGPU_NPS8_PARTITION_MODE] = "NPS8",
+};
+
+static ssize_t available_memory_partition_show(struct device *dev,
+					       struct device_attribute *addr,
+					       char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	int size = 0, mode;
+	char *sep = "";
+
+	for_each_inst(mode, adev->gmc.supported_nps_modes) {
+		size += sysfs_emit_at(buf, size, "%s%s", sep, nps_desc[mode]);
+		sep = ", ";
+	}
+	size += sysfs_emit_at(buf, size, "\n");
+
+	return size;
+}
+
+static ssize_t current_memory_partition_store(struct device *dev,
+					      struct device_attribute *attr,
+					      const char *buf, size_t count)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	enum amdgpu_memory_partition mode;
+	struct amdgpu_hive_info *hive;
+	int i;
+
+	mode = UNKNOWN_MEMORY_PARTITION_MODE;
+	for_each_inst(i, adev->gmc.supported_nps_modes) {
+		if (!strncasecmp(nps_desc[i], buf, strlen(nps_desc[i]))) {
+			mode = i;
+			break;
+		}
+	}
+
+	if (mode == UNKNOWN_MEMORY_PARTITION_MODE)
+		return -EINVAL;
+
+	if (mode == adev->gmc.gmc_funcs->query_mem_partition_mode(adev)) {
+		dev_info(
+			adev->dev,
+			"requested NPS mode is same as current NPS mode, skipping\n");
+		return count;
+	}
+
+	/* If device is part of hive, all devices in the hive should request the
+	 * same mode. Hence store the requested mode in hive.
+	 */
+	hive = amdgpu_get_xgmi_hive(adev);
+	if (hive) {
+		atomic_set(&hive->requested_nps_mode, mode);
+		amdgpu_put_xgmi_hive(hive);
+	} else {
+		adev->gmc.requested_nps_mode = mode;
+	}
+
+	dev_info(
+		adev->dev,
+		"NPS mode change requested, please remove and reload the driver\n");
+
+	return count;
+}
+
+static ssize_t current_memory_partition_show(
+	struct device *dev, struct device_attribute *addr, char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	enum amdgpu_memory_partition mode;
+
+	/* Only minimal precaution taken to reject requests while in reset */
+	if (amdgpu_in_reset(adev))
+		return -EPERM;
+
+	mode = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
+	if ((mode >= ARRAY_SIZE(nps_desc)) ||
+	    (BIT(mode) & AMDGPU_ALL_NPS_MASK) != BIT(mode))
+		return sysfs_emit(buf, "UNKNOWN\n");
+
+	return sysfs_emit(buf, "%s\n", nps_desc[mode]);
+}
+
+static DEVICE_ATTR_RW(current_memory_partition);
+static DEVICE_ATTR_RO(available_memory_partition);
+
+int amdgpu_gmc_sysfs_init(struct amdgpu_device *adev)
+{
+	bool nps_switch_support;
+	int r = 0;
+
+	if (!adev->gmc.gmc_funcs->query_mem_partition_mode)
+		return 0;
+
+	nps_switch_support = (hweight32(adev->gmc.supported_nps_modes &
+					AMDGPU_ALL_NPS_MASK) > 1);
+	if (!nps_switch_support)
+		dev_attr_current_memory_partition.attr.mode &=
+			~(S_IWUSR | S_IWGRP | S_IWOTH);
+	else
+		r = device_create_file(adev->dev,
+				       &dev_attr_available_memory_partition);
+
+	if (r)
+		return r;
+
+	return device_create_file(adev->dev,
+				  &dev_attr_current_memory_partition);
+}
+
+void amdgpu_gmc_sysfs_fini(struct amdgpu_device *adev)
+{
+	if (!adev->gmc.gmc_funcs->query_mem_partition_mode)
+		return;
+
+	device_remove_file(adev->dev, &dev_attr_current_memory_partition);
+	device_remove_file(adev->dev, &dev_attr_available_memory_partition);
+}
+
+int amdgpu_gmc_get_nps_memranges(struct amdgpu_device *adev,
+				 struct amdgpu_mem_partition_info *mem_ranges,
+				 uint8_t *exp_ranges)
+{
+	struct amdgpu_gmc_memrange *ranges;
+	int range_cnt, ret, i, j;
+	uint32_t nps_type;
+	bool refresh;
+
+	if (!mem_ranges || !exp_ranges)
+		return -EINVAL;
+
+	refresh = (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) &&
+		  (adev->gmc.reset_flags & AMDGPU_GMC_INIT_RESET_NPS);
+	ret = amdgpu_discovery_get_nps_info(adev, &nps_type, &ranges,
+					    &range_cnt, refresh);
+
+	if (ret)
+		return ret;
+
+	/* TODO: For now, expect ranges and partition count to be the same.
+	 * Adjust if there are holes expected in any NPS domain.
+	 */
+	if (*exp_ranges && (range_cnt != *exp_ranges)) {
+		dev_warn(
+			adev->dev,
+			"NPS config mismatch - expected ranges: %d discovery - nps mode: %d, nps ranges: %d",
+			*exp_ranges, nps_type, range_cnt);
+		ret = -EINVAL;
+		goto err;
+	}
+
+	for (i = 0; i < range_cnt; ++i) {
+		if (ranges[i].base_address >= ranges[i].limit_address) {
+			dev_warn(
+				adev->dev,
+				"Invalid NPS range - nps mode: %d, range[%d]: base: %llx limit: %llx",
+				nps_type, i, ranges[i].base_address,
+				ranges[i].limit_address);
+			ret = -EINVAL;
+			goto err;
+		}
+
+		/* Check for overlaps, not expecting any now */
+		for (j = i - 1; j >= 0; j--) {
+			if (max(ranges[j].base_address,
+				ranges[i].base_address) <=
+			    min(ranges[j].limit_address,
+				ranges[i].limit_address)) {
+				dev_warn(
+					adev->dev,
+					"overlapping ranges detected [ %llx - %llx ] | [%llx - %llx]",
+					ranges[j].base_address,
+					ranges[j].limit_address,
+					ranges[i].base_address,
+					ranges[i].limit_address);
+				ret = -EINVAL;
+				goto err;
+			}
+		}
+
+		mem_ranges[i].range.fpfn =
+			(ranges[i].base_address -
+			 adev->vm_manager.vram_base_offset) >>
+			AMDGPU_GPU_PAGE_SHIFT;
+		mem_ranges[i].range.lpfn =
+			(ranges[i].limit_address -
+			 adev->vm_manager.vram_base_offset) >>
+			AMDGPU_GPU_PAGE_SHIFT;
+		mem_ranges[i].size =
+			ranges[i].limit_address - ranges[i].base_address + 1;
+	}
+
+	if (!*exp_ranges)
+		*exp_ranges = range_cnt;
+err:
+	kfree(ranges);
+
+	return ret;
+}
+
+int amdgpu_gmc_request_memory_partition(struct amdgpu_device *adev,
+					int nps_mode)
+{
+	/* Not supported on VF devices and APUs */
+	if (amdgpu_sriov_vf(adev) || (adev->flags & AMD_IS_APU))
+		return -EOPNOTSUPP;
+
+	if (!adev->psp.funcs) {
+		dev_err(adev->dev,
+			"PSP interface not available for nps mode change request");
+		return -EINVAL;
+	}
+
+	return psp_memory_partition(&adev->psp, nps_mode);
+}
+
+static inline bool amdgpu_gmc_need_nps_switch_req(struct amdgpu_device *adev,
+						  int req_nps_mode,
+						  int cur_nps_mode)
+{
+	return (((BIT(req_nps_mode) & adev->gmc.supported_nps_modes) ==
+			BIT(req_nps_mode)) &&
+		req_nps_mode != cur_nps_mode);
+}
+
+void amdgpu_gmc_prepare_nps_mode_change(struct amdgpu_device *adev)
+{
+	int req_nps_mode, cur_nps_mode, r;
+	struct amdgpu_hive_info *hive;
+
+	if (amdgpu_sriov_vf(adev) || !adev->gmc.supported_nps_modes ||
+	    !adev->gmc.gmc_funcs->request_mem_partition_mode)
+		return;
+
+	cur_nps_mode = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
+	hive = amdgpu_get_xgmi_hive(adev);
+	if (hive) {
+		req_nps_mode = atomic_read(&hive->requested_nps_mode);
+		if (!amdgpu_gmc_need_nps_switch_req(adev, req_nps_mode,
+						    cur_nps_mode)) {
+			amdgpu_put_xgmi_hive(hive);
+			return;
+		}
+		r = amdgpu_xgmi_request_nps_change(adev, hive, req_nps_mode);
+		amdgpu_put_xgmi_hive(hive);
+		goto out;
+	}
+
+	req_nps_mode = adev->gmc.requested_nps_mode;
+	if (!amdgpu_gmc_need_nps_switch_req(adev, req_nps_mode, cur_nps_mode))
+		return;
+
+	/* even if this fails, we should let driver unload w/o blocking */
+	r = adev->gmc.gmc_funcs->request_mem_partition_mode(adev, req_nps_mode);
+out:
+	if (r)
+		dev_err(adev->dev, "NPS mode change request failed\n");
+	else
+		dev_info(
+			adev->dev,
+			"NPS mode change request done, reload driver to complete the change\n");
+}
+
+bool amdgpu_gmc_need_reset_on_init(struct amdgpu_device *adev)
+{
+	if (adev->gmc.gmc_funcs->need_reset_on_init)
+		return adev->gmc.gmc_funcs->need_reset_on_init(adev);
+
+	return false;
+}
+
+enum amdgpu_memory_partition
+amdgpu_gmc_get_vf_memory_partition(struct amdgpu_device *adev)
+{
+	switch (adev->gmc.num_mem_partitions) {
+	case 0:
+		return UNKNOWN_MEMORY_PARTITION_MODE;
+	case 1:
+		return AMDGPU_NPS1_PARTITION_MODE;
+	case 2:
+		return AMDGPU_NPS2_PARTITION_MODE;
+	case 4:
+		return AMDGPU_NPS4_PARTITION_MODE;
+	case 8:
+		return AMDGPU_NPS8_PARTITION_MODE;
+	default:
+		return AMDGPU_NPS1_PARTITION_MODE;
+	}
+}
+
+enum amdgpu_memory_partition
+amdgpu_gmc_get_memory_partition(struct amdgpu_device *adev, u32 *supp_modes)
+{
+	enum amdgpu_memory_partition mode = UNKNOWN_MEMORY_PARTITION_MODE;
+
+	if (adev->nbio.funcs &&
+	    adev->nbio.funcs->get_memory_partition_mode)
+		mode = adev->nbio.funcs->get_memory_partition_mode(adev,
+								   supp_modes);
+	else
+		dev_warn(adev->dev, "memory partition mode query is not supported\n");
+
+	return mode;
+}
+
+enum amdgpu_memory_partition
+amdgpu_gmc_query_memory_partition(struct amdgpu_device *adev)
+{
+	if (amdgpu_sriov_vf(adev))
+		return amdgpu_gmc_get_vf_memory_partition(adev);
+	else
+		return amdgpu_gmc_get_memory_partition(adev, NULL);
+}
+
+static bool amdgpu_gmc_validate_partition_info(struct amdgpu_device *adev)
+{
+	enum amdgpu_memory_partition mode;
+	u32 supp_modes;
+	bool valid;
+
+	mode = amdgpu_gmc_get_memory_partition(adev, &supp_modes);
+
+	/* Mode detected by hardware not present in supported modes */
+	if ((mode != UNKNOWN_MEMORY_PARTITION_MODE) &&
+	    !(BIT(mode - 1) & supp_modes))
+		return false;
+
+	switch (mode) {
+	case UNKNOWN_MEMORY_PARTITION_MODE:
+	case AMDGPU_NPS1_PARTITION_MODE:
+		valid = (adev->gmc.num_mem_partitions == 1);
+		break;
+	case AMDGPU_NPS2_PARTITION_MODE:
+		valid = (adev->gmc.num_mem_partitions == 2);
+		break;
+	case AMDGPU_NPS4_PARTITION_MODE:
+		valid = (adev->gmc.num_mem_partitions == 3 ||
+			 adev->gmc.num_mem_partitions == 4);
+		break;
+	case AMDGPU_NPS8_PARTITION_MODE:
+		valid = (adev->gmc.num_mem_partitions == 8);
+		break;
+	default:
+		valid = false;
+	}
+
+	return valid;
+}
+
+static bool amdgpu_gmc_is_node_present(int *node_ids, int num_ids, int nid)
+{
+	int i;
+
+	/* Check if node with id 'nid' is present in 'node_ids' array */
+	for (i = 0; i < num_ids; ++i)
+		if (node_ids[i] == nid)
+			return true;
+
+	return false;
+}
+
+static void
+amdgpu_gmc_init_acpi_mem_ranges(struct amdgpu_device *adev,
+				struct amdgpu_mem_partition_info *mem_ranges)
+{
+	struct amdgpu_numa_info numa_info;
+	int node_ids[AMDGPU_MAX_MEM_RANGES];
+	int num_ranges = 0, ret;
+	int num_xcc, xcc_id;
+	uint32_t xcc_mask;
+
+	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
+	xcc_mask = (1U << num_xcc) - 1;
+
+	for_each_inst(xcc_id, xcc_mask)	{
+		ret = amdgpu_acpi_get_mem_info(adev, xcc_id, &numa_info);
+		if (ret)
+			continue;
+
+		if (numa_info.nid == NUMA_NO_NODE) {
+			mem_ranges[0].size = numa_info.size;
+			mem_ranges[0].numa.node = numa_info.nid;
+			num_ranges = 1;
+			break;
+		}
+
+		if (amdgpu_gmc_is_node_present(node_ids, num_ranges,
+					     numa_info.nid))
+			continue;
+
+		node_ids[num_ranges] = numa_info.nid;
+		mem_ranges[num_ranges].numa.node = numa_info.nid;
+		mem_ranges[num_ranges].size = numa_info.size;
+		++num_ranges;
+	}
+
+	adev->gmc.num_mem_partitions = num_ranges;
+}
+
+void amdgpu_gmc_init_sw_mem_ranges(struct amdgpu_device *adev,
+				   struct amdgpu_mem_partition_info *mem_ranges)
+{
+	enum amdgpu_memory_partition mode;
+	u32 start_addr = 0, size;
+	int i, r, l;
+
+	mode = amdgpu_gmc_query_memory_partition(adev);
+
+	switch (mode) {
+	case UNKNOWN_MEMORY_PARTITION_MODE:
+		adev->gmc.num_mem_partitions = 0;
+		break;
+	case AMDGPU_NPS1_PARTITION_MODE:
+		adev->gmc.num_mem_partitions = 1;
+		break;
+	case AMDGPU_NPS2_PARTITION_MODE:
+		adev->gmc.num_mem_partitions = 2;
+		break;
+	case AMDGPU_NPS4_PARTITION_MODE:
+		if (adev->flags & AMD_IS_APU)
+			adev->gmc.num_mem_partitions = 3;
+		else
+			adev->gmc.num_mem_partitions = 4;
+		break;
+	case AMDGPU_NPS8_PARTITION_MODE:
+		adev->gmc.num_mem_partitions = 8;
+		break;
+	default:
+		adev->gmc.num_mem_partitions = 1;
+		break;
+	}
+
+	/* Use NPS range info, if populated */
+	r = amdgpu_gmc_get_nps_memranges(adev, mem_ranges,
+					 &adev->gmc.num_mem_partitions);
+	if (!r) {
+		l = 0;
+		for (i = 1; i < adev->gmc.num_mem_partitions; ++i) {
+			if (mem_ranges[i].range.lpfn >
+			    mem_ranges[i - 1].range.lpfn)
+				l = i;
+		}
+
+	} else {
+		if (!adev->gmc.num_mem_partitions) {
+			dev_warn(adev->dev,
+				 "Not able to detect NPS mode, fall back to NPS1\n");
+			adev->gmc.num_mem_partitions = 1;
+		}
+		/* Fallback to sw based calculation */
+		size = (adev->gmc.real_vram_size + SZ_16M) >> AMDGPU_GPU_PAGE_SHIFT;
+		size /= adev->gmc.num_mem_partitions;
+
+		for (i = 0; i < adev->gmc.num_mem_partitions; ++i) {
+			mem_ranges[i].range.fpfn = start_addr;
+			mem_ranges[i].size =
+				((u64)size << AMDGPU_GPU_PAGE_SHIFT);
+			mem_ranges[i].range.lpfn = start_addr + size - 1;
+			start_addr += size;
+		}
+
+		l = adev->gmc.num_mem_partitions - 1;
+	}
+
+	/* Adjust the last one */
+	mem_ranges[l].range.lpfn =
+		(adev->gmc.real_vram_size >> AMDGPU_GPU_PAGE_SHIFT) - 1;
+	mem_ranges[l].size =
+		adev->gmc.real_vram_size -
+		((u64)mem_ranges[l].range.fpfn << AMDGPU_GPU_PAGE_SHIFT);
+}
+
+int amdgpu_gmc_init_mem_ranges(struct amdgpu_device *adev)
+{
+	bool valid;
+
+	adev->gmc.mem_partitions = kcalloc(AMDGPU_MAX_MEM_RANGES,
+					   sizeof(struct amdgpu_mem_partition_info),
+					   GFP_KERNEL);
+	if (!adev->gmc.mem_partitions)
+		return -ENOMEM;
+
+	if (adev->gmc.is_app_apu)
+		amdgpu_gmc_init_acpi_mem_ranges(adev, adev->gmc.mem_partitions);
+	else
+		amdgpu_gmc_init_sw_mem_ranges(adev, adev->gmc.mem_partitions);
+
+	if (amdgpu_sriov_vf(adev))
+		valid = true;
+	else
+		valid = amdgpu_gmc_validate_partition_info(adev);
+	if (!valid) {
+		/* TODO: handle invalid case */
+		dev_warn(adev->dev,
+			 "Mem ranges not matching with hardware config\n");
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
new file mode 100644
index 000000000000..727342689d4b
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
@@ -0,0 +1,476 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+#ifndef __AMDGPU_GMC_H__
+#define __AMDGPU_GMC_H__
+
+#include <linux/types.h>
+
+#include "amdgpu_irq.h"
+#include "amdgpu_xgmi.h"
+#include "amdgpu_ras.h"
+
+/* VA hole for 48bit addresses on Vega10 */
+#define AMDGPU_GMC_HOLE_START	0x0000800000000000ULL
+#define AMDGPU_GMC_HOLE_END	0xffff800000000000ULL
+
+/*
+ * Hardware is programmed as if the hole doesn't exists with start and end
+ * address values.
+ *
+ * This mask is used to remove the upper 16bits of the VA and so come up with
+ * the linear addr value.
+ */
+#define AMDGPU_GMC_HOLE_MASK	0x0000ffffffffffffULL
+
+/*
+ * Ring size as power of two for the log of recent faults.
+ */
+#define AMDGPU_GMC_FAULT_RING_ORDER	8
+#define AMDGPU_GMC_FAULT_RING_SIZE	(1 << AMDGPU_GMC_FAULT_RING_ORDER)
+
+/*
+ * Hash size as power of two for the log of recent faults
+ */
+#define AMDGPU_GMC_FAULT_HASH_ORDER	8
+#define AMDGPU_GMC_FAULT_HASH_SIZE	(1 << AMDGPU_GMC_FAULT_HASH_ORDER)
+
+/*
+ * Number of IH timestamp ticks until a fault is considered handled
+ */
+#define AMDGPU_GMC_FAULT_TIMEOUT	5000ULL
+
+/* XNACK flags */
+#define AMDGPU_GMC_XNACK_FLAG_CHAIN BIT(0)
+
+struct firmware;
+
+enum amdgpu_memory_partition {
+	UNKNOWN_MEMORY_PARTITION_MODE = 0,
+	AMDGPU_NPS1_PARTITION_MODE = 1,
+	AMDGPU_NPS2_PARTITION_MODE = 2,
+	AMDGPU_NPS3_PARTITION_MODE = 3,
+	AMDGPU_NPS4_PARTITION_MODE = 4,
+	AMDGPU_NPS6_PARTITION_MODE = 6,
+	AMDGPU_NPS8_PARTITION_MODE = 8,
+};
+
+#define AMDGPU_ALL_NPS_MASK                                                  \
+	(BIT(AMDGPU_NPS1_PARTITION_MODE) | BIT(AMDGPU_NPS2_PARTITION_MODE) | \
+	 BIT(AMDGPU_NPS3_PARTITION_MODE) | BIT(AMDGPU_NPS4_PARTITION_MODE) | \
+	 BIT(AMDGPU_NPS6_PARTITION_MODE) | BIT(AMDGPU_NPS8_PARTITION_MODE))
+
+#define AMDGPU_GMC_INIT_RESET_NPS  BIT(0)
+
+#define AMDGPU_MAX_MEM_RANGES 8
+
+#define AMDGPU_GMC9_FAULT_SOURCE_DATA_RETRY 0x80
+#define AMDGPU_GMC9_FAULT_SOURCE_DATA_READ  0x40
+#define AMDGPU_GMC9_FAULT_SOURCE_DATA_WRITE 0x20
+#define AMDGPU_GMC9_FAULT_SOURCE_DATA_EXE   0x10
+
+/*
+ * GMC page fault information
+ */
+struct amdgpu_gmc_fault {
+	uint64_t	timestamp:48;
+	uint64_t	next:AMDGPU_GMC_FAULT_RING_ORDER;
+	atomic64_t	key;
+	uint64_t	timestamp_expiry:48;
+};
+
+/*
+ * VMHUB structures, functions & helpers
+ */
+struct amdgpu_vmhub_funcs {
+	void (*print_l2_protection_fault_status)(struct amdgpu_device *adev,
+						 uint32_t status);
+	uint32_t (*get_invalidate_req)(unsigned int vmid, uint32_t flush_type);
+};
+
+struct amdgpu_vmhub {
+	uint32_t	ctx0_ptb_addr_lo32;
+	uint32_t	ctx0_ptb_addr_hi32;
+	uint32_t	vm_inv_eng0_sem;
+	uint32_t	vm_inv_eng0_req;
+	uint32_t	vm_inv_eng0_ack;
+	uint32_t	vm_context0_cntl;
+	uint32_t	vm_l2_pro_fault_status;
+	uint32_t	vm_l2_pro_fault_cntl;
+
+	/*
+	 * store the register distances between two continuous context domain
+	 * and invalidation engine.
+	 */
+	uint32_t	ctx_distance;
+	uint32_t	ctx_addr_distance; /* include LO32/HI32 */
+	uint32_t	eng_distance;
+	uint32_t	eng_addr_distance; /* include LO32/HI32 */
+
+	uint32_t        vm_cntx_cntl;
+	uint32_t	vm_cntx_cntl_vm_fault;
+	uint32_t	vm_l2_bank_select_reserved_cid2;
+
+	uint32_t	vm_contexts_disable;
+
+	bool		sdma_invalidation_workaround;
+
+	const struct amdgpu_vmhub_funcs *vmhub_funcs;
+};
+
+/*
+ * GPU MC structures, functions & helpers
+ */
+struct amdgpu_gmc_funcs {
+	/* flush the vm tlb via mmio */
+	void (*flush_gpu_tlb)(struct amdgpu_device *adev, uint32_t vmid,
+				uint32_t vmhub, uint32_t flush_type);
+	/* flush the vm tlb via pasid */
+	void (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid,
+				    uint32_t flush_type, bool all_hub,
+				    uint32_t inst);
+	/* flush the vm tlb via ring */
+	uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
+				       uint64_t pd_addr);
+	/* Change the VMID -> PASID mapping */
+	void (*emit_pasid_mapping)(struct amdgpu_ring *ring, unsigned vmid,
+				   unsigned pasid);
+	/* enable/disable PRT support */
+	void (*set_prt)(struct amdgpu_device *adev, bool enable);
+	/* get the pde for a given mc addr */
+	void (*get_vm_pde)(struct amdgpu_device *adev, int level,
+			   u64 *dst, u64 *flags);
+	/* get the pte flags to use for PTEs */
+	void (*get_vm_pte)(struct amdgpu_device *adev,
+			   struct amdgpu_vm *vm,
+			   struct amdgpu_bo *bo,
+			   uint32_t vm_flags,
+			   uint64_t *pte_flags);
+	/* override per-page pte flags */
+	void (*override_vm_pte_flags)(struct amdgpu_device *dev,
+				      struct amdgpu_vm *vm,
+				      uint64_t addr, uint64_t *flags);
+	/* get the amount of memory used by the vbios for pre-OS console */
+	unsigned int (*get_vbios_fb_size)(struct amdgpu_device *adev);
+	/* get the DCC buffer alignment */
+	unsigned int (*get_dcc_alignment)(struct amdgpu_device *adev);
+
+	enum amdgpu_memory_partition (*query_mem_partition_mode)(
+		struct amdgpu_device *adev);
+	/* Request NPS mode */
+	int (*request_mem_partition_mode)(struct amdgpu_device *adev,
+					  int nps_mode);
+	bool (*need_reset_on_init)(struct amdgpu_device *adev);
+};
+
+struct amdgpu_mem_partition_info {
+	union {
+		struct {
+			uint32_t fpfn;
+			uint32_t lpfn;
+		} range;
+		struct {
+			int node;
+		} numa;
+	};
+	uint64_t size;
+};
+
+#define INVALID_PFN    -1
+
+struct amdgpu_gmc_memrange {
+	uint64_t base_address;
+	uint64_t limit_address;
+	uint32_t flags;
+	int nid_mask;
+};
+
+enum amdgpu_gart_placement {
+	AMDGPU_GART_PLACEMENT_BEST_FIT = 0,
+	AMDGPU_GART_PLACEMENT_HIGH,
+	AMDGPU_GART_PLACEMENT_LOW,
+};
+
+struct amdgpu_gmc {
+	/* FB's physical address in MMIO space (for CPU to
+	 * map FB). This is different compared to the agp/
+	 * gart/vram_start/end field as the later is from
+	 * GPU's view and aper_base is from CPU's view.
+	 */
+	resource_size_t		aper_size;
+	resource_size_t		aper_base;
+	/* for some chips with <= 32MB we need to lie
+	 * about vram size near mc fb location */
+	u64			mc_vram_size;
+	u64			visible_vram_size;
+	/* AGP aperture start and end in MC address space
+	 * Driver find a hole in the MC address space
+	 * to place AGP by setting MC_VM_AGP_BOT/TOP registers
+	 * Under VMID0, logical address == MC address. AGP
+	 * aperture maps to physical bus or IOVA addressed.
+	 * AGP aperture is used to simulate FB in ZFB case.
+	 * AGP aperture is also used for page table in system
+	 * memory (mainly for APU).
+	 *
+	 */
+	u64			agp_size;
+	u64			agp_start;
+	u64			agp_end;
+	/* GART aperture start and end in MC address space
+	 * Driver find a hole in the MC address space
+	 * to place GART by setting VM_CONTEXT0_PAGE_TABLE_START/END_ADDR
+	 * registers
+	 * Under VMID0, logical address inside GART aperture will
+	 * be translated through gpuvm gart page table to access
+	 * paged system memory
+	 */
+	u64			gart_size;
+	u64			gart_start;
+	u64			gart_end;
+	/* Frame buffer aperture of this GPU device. Different from
+	 * fb_start (see below), this only covers the local GPU device.
+	 * If driver uses FB aperture to access FB, driver get fb_start from
+	 * MC_VM_FB_LOCATION_BASE (set by vbios) and calculate vram_start
+	 * of this local device by adding an offset inside the XGMI hive.
+	 * If driver uses GART table for VMID0 FB access, driver finds a hole in
+	 * VMID0's virtual address space to place the SYSVM aperture inside
+	 * which the first part is vram and the second part is gart (covering
+	 * system ram).
+	 */
+	u64			vram_start;
+	u64			vram_end;
+	/* FB region , it's same as local vram region in single GPU, in XGMI
+	 * configuration, this region covers all GPUs in the same hive ,
+	 * each GPU in the hive has the same view of this FB region .
+	 * GPU0's vram starts at offset (0 * segment size) ,
+	 * GPU1 starts at offset (1 * segment size), etc.
+	 */
+	u64			fb_start;
+	u64			fb_end;
+	unsigned		vram_width;
+	u64			real_vram_size;
+	int			vram_mtrr;
+	u64                     mc_mask;
+	const struct firmware   *fw;	/* MC firmware */
+	uint32_t                fw_version;
+	struct amdgpu_irq_src	vm_fault;
+	uint32_t		vram_type;
+	uint8_t			vram_vendor;
+	uint32_t                srbm_soft_reset;
+	bool			prt_warning;
+	uint32_t		sdpif_register;
+	/* apertures */
+	u64			shared_aperture_start;
+	u64			shared_aperture_end;
+	u64			private_aperture_start;
+	u64			private_aperture_end;
+	/* protects concurrent invalidation */
+	spinlock_t		invalidate_lock;
+	bool			translate_further;
+	struct kfd_vm_fault_info *vm_fault_info;
+	atomic_t		vm_fault_info_updated;
+
+	struct amdgpu_gmc_fault	fault_ring[AMDGPU_GMC_FAULT_RING_SIZE];
+	struct {
+		uint64_t	idx:AMDGPU_GMC_FAULT_RING_ORDER;
+	} fault_hash[AMDGPU_GMC_FAULT_HASH_SIZE];
+	uint64_t		last_fault:AMDGPU_GMC_FAULT_RING_ORDER;
+
+	bool tmz_enabled;
+	bool is_app_apu;
+
+	struct amdgpu_mem_partition_info *mem_partitions;
+	uint8_t num_mem_partitions;
+	const struct amdgpu_gmc_funcs	*gmc_funcs;
+	enum amdgpu_memory_partition	requested_nps_mode;
+	uint32_t supported_nps_modes;
+	uint32_t reset_flags;
+
+	struct amdgpu_xgmi xgmi;
+	struct amdgpu_irq_src	ecc_irq;
+	int noretry;
+	uint32_t xnack_flags;
+
+	uint32_t	vmid0_page_table_block_size;
+	uint32_t	vmid0_page_table_depth;
+	struct amdgpu_bo		*pdb0_bo;
+	/* CPU kmapped address of pdb0*/
+	void				*ptr_pdb0;
+
+	/* MALL size */
+	u64 mall_size;
+	uint32_t m_half_use;
+
+	/* number of UMC instances */
+	int num_umc;
+	/* mode2 save restore */
+	u64 VM_L2_CNTL;
+	u64 VM_L2_CNTL2;
+	u64 VM_DUMMY_PAGE_FAULT_CNTL;
+	u64 VM_DUMMY_PAGE_FAULT_ADDR_LO32;
+	u64 VM_DUMMY_PAGE_FAULT_ADDR_HI32;
+	u64 VM_L2_PROTECTION_FAULT_CNTL;
+	u64 VM_L2_PROTECTION_FAULT_CNTL2;
+	u64 VM_L2_PROTECTION_FAULT_MM_CNTL3;
+	u64 VM_L2_PROTECTION_FAULT_MM_CNTL4;
+	u64 VM_L2_PROTECTION_FAULT_ADDR_LO32;
+	u64 VM_L2_PROTECTION_FAULT_ADDR_HI32;
+	u64 VM_DEBUG;
+	u64 VM_L2_MM_GROUP_RT_CLASSES;
+	u64 VM_L2_BANK_SELECT_RESERVED_CID;
+	u64 VM_L2_BANK_SELECT_RESERVED_CID2;
+	u64 VM_L2_CACHE_PARITY_CNTL;
+	u64 VM_L2_IH_LOG_CNTL;
+	u64 VM_CONTEXT_CNTL[16];
+	u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_LO32[16];
+	u64 VM_CONTEXT_PAGE_TABLE_BASE_ADDR_HI32[16];
+	u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_LO32[16];
+	u64 VM_CONTEXT_PAGE_TABLE_START_ADDR_HI32[16];
+	u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_LO32[16];
+	u64 VM_CONTEXT_PAGE_TABLE_END_ADDR_HI32[16];
+	u64 MC_VM_MX_L1_TLB_CNTL;
+
+	u64 noretry_flags;
+
+	bool flush_tlb_needs_extra_type_0;
+	bool flush_tlb_needs_extra_type_2;
+	bool flush_pasid_uses_kiq;
+};
+
+#define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
+#define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
+#define amdgpu_gmc_get_vm_pde(adev, level, dst, flags) (adev)->gmc.gmc_funcs->get_vm_pde((adev), (level), (dst), (flags))
+#define amdgpu_gmc_get_vm_pte(adev, vm, bo, vm_flags, pte_flags) \
+	((adev)->gmc.gmc_funcs->get_vm_pte((adev), (vm), (bo), (vm_flags), \
+					   (pte_flags)))
+#define amdgpu_gmc_override_vm_pte_flags(adev, vm, addr, pte_flags)	\
+	(adev)->gmc.gmc_funcs->override_vm_pte_flags			\
+		((adev), (vm), (addr), (pte_flags))
+#define amdgpu_gmc_get_vbios_fb_size(adev) (adev)->gmc.gmc_funcs->get_vbios_fb_size((adev))
+#define amdgpu_gmc_get_dcc_alignment(adev) ({			\
+	typeof(adev) _adev = (adev);				\
+	_adev->gmc.gmc_funcs->get_dcc_alignment(_adev);		\
+})
+
+/**
+ * amdgpu_gmc_vram_full_visible - Check if full VRAM is visible through the BAR
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Returns:
+ * True if full VRAM is visible through the BAR
+ */
+static inline bool amdgpu_gmc_vram_full_visible(struct amdgpu_gmc *gmc)
+{
+	WARN_ON(gmc->real_vram_size < gmc->visible_vram_size);
+
+	return (gmc->real_vram_size == gmc->visible_vram_size);
+}
+
+/**
+ * amdgpu_gmc_sign_extend - sign extend the given gmc address
+ *
+ * @addr: address to extend
+ */
+static inline uint64_t amdgpu_gmc_sign_extend(uint64_t addr)
+{
+	if (addr >= AMDGPU_GMC_HOLE_START)
+		addr |= AMDGPU_GMC_HOLE_END;
+
+	return addr;
+}
+
+bool amdgpu_gmc_is_pdb0_enabled(struct amdgpu_device *adev);
+int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev);
+void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
+			       uint64_t *addr, uint64_t *flags);
+int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
+				uint32_t gpu_page_idx, uint64_t addr,
+				uint64_t flags);
+uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo);
+uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo);
+void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc);
+void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
+			      u64 base);
+void amdgpu_gmc_gart_location(struct amdgpu_device *adev,
+			      struct amdgpu_gmc *mc,
+			      enum amdgpu_gart_placement gart_placement);
+void amdgpu_gmc_agp_location(struct amdgpu_device *adev,
+			     struct amdgpu_gmc *mc);
+void amdgpu_gmc_set_agp_default(struct amdgpu_device *adev,
+				struct amdgpu_gmc *mc);
+bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev,
+			      struct amdgpu_ih_ring *ih, uint64_t addr,
+			      uint16_t pasid, uint64_t timestamp);
+void amdgpu_gmc_filter_faults_remove(struct amdgpu_device *adev, uint64_t addr,
+				     uint16_t pasid);
+int amdgpu_gmc_ras_sw_init(struct amdgpu_device *adev);
+int amdgpu_gmc_ras_late_init(struct amdgpu_device *adev);
+void amdgpu_gmc_ras_fini(struct amdgpu_device *adev);
+int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev);
+void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
+			      uint32_t vmhub, uint32_t flush_type);
+int amdgpu_gmc_flush_gpu_tlb_pasid(struct amdgpu_device *adev, uint16_t pasid,
+				   uint32_t flush_type, bool all_hub,
+				   uint32_t inst);
+void amdgpu_gmc_fw_reg_write_reg_wait(struct amdgpu_device *adev,
+				      uint32_t reg0, uint32_t reg1,
+				      uint32_t ref, uint32_t mask,
+				      uint32_t xcc_inst);
+
+extern void amdgpu_gmc_tmz_set(struct amdgpu_device *adev);
+extern void amdgpu_gmc_noretry_set(struct amdgpu_device *adev);
+
+extern void
+amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type,
+			      bool enable);
+
+void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev);
+
+void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev);
+uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr);
+uint64_t amdgpu_gmc_vram_pa(struct amdgpu_device *adev, struct amdgpu_bo *bo);
+int amdgpu_gmc_vram_checking(struct amdgpu_device *adev);
+int amdgpu_gmc_sysfs_init(struct amdgpu_device *adev);
+void amdgpu_gmc_sysfs_fini(struct amdgpu_device *adev);
+
+int amdgpu_gmc_get_nps_memranges(struct amdgpu_device *adev,
+				 struct amdgpu_mem_partition_info *mem_ranges,
+				 uint8_t *exp_ranges);
+
+int amdgpu_gmc_request_memory_partition(struct amdgpu_device *adev,
+					int nps_mode);
+void amdgpu_gmc_prepare_nps_mode_change(struct amdgpu_device *adev);
+bool amdgpu_gmc_need_reset_on_init(struct amdgpu_device *adev);
+enum amdgpu_memory_partition
+amdgpu_gmc_get_vf_memory_partition(struct amdgpu_device *adev);
+enum amdgpu_memory_partition
+amdgpu_gmc_get_memory_partition(struct amdgpu_device *adev, u32 *supp_modes);
+enum amdgpu_memory_partition
+amdgpu_gmc_query_memory_partition(struct amdgpu_device *adev);
+int amdgpu_gmc_init_mem_ranges(struct amdgpu_device *adev);
+void amdgpu_gmc_init_sw_mem_ranges(struct amdgpu_device *adev,
+				   struct amdgpu_mem_partition_info *mem_ranges);
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index f7d22c44034d..895c1e4c6747 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -22,182 +22,138 @@
  * Authors: Christian König
  */
 
-#include <drm/drmP.h>
+#include <drm/ttm/ttm_range_manager.h>
+
 #include "amdgpu.h"
 
-struct amdgpu_gtt_mgr {
-	struct drm_mm mm;
-	spinlock_t lock;
-	uint64_t available;
-};
+static inline struct amdgpu_gtt_mgr *
+to_gtt_mgr(struct ttm_resource_manager *man)
+{
+	return container_of(man, struct amdgpu_gtt_mgr, manager);
+}
 
 /**
- * amdgpu_gtt_mgr_init - init GTT manager and DRM MM
- *
- * @man: TTM memory type manager
- * @p_size: maximum size of GTT
+ * DOC: mem_info_gtt_total
  *
- * Allocate and initialize the GTT manager.
+ * The amdgpu driver provides a sysfs API for reporting current total size of
+ * the GTT.
+ * The file mem_info_gtt_total is used for this, and returns the total size of
+ * the GTT block, in bytes
  */
-static int amdgpu_gtt_mgr_init(struct ttm_mem_type_manager *man,
-			       unsigned long p_size)
+static ssize_t amdgpu_mem_info_gtt_total_show(struct device *dev,
+					      struct device_attribute *attr,
+					      char *buf)
 {
-	struct amdgpu_gtt_mgr *mgr;
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct ttm_resource_manager *man;
 
-	mgr = kzalloc(sizeof(*mgr), GFP_KERNEL);
-	if (!mgr)
-		return -ENOMEM;
-
-	drm_mm_init(&mgr->mm, 0, p_size);
-	spin_lock_init(&mgr->lock);
-	mgr->available = p_size;
-	man->priv = mgr;
-	return 0;
+	man = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT);
+	return sysfs_emit(buf, "%llu\n", man->size);
 }
 
 /**
- * amdgpu_gtt_mgr_fini - free and destroy GTT manager
- *
- * @man: TTM memory type manager
+ * DOC: mem_info_gtt_used
  *
- * Destroy and free the GTT manager, returns -EBUSY if ranges are still
- * allocated inside it.
+ * The amdgpu driver provides a sysfs API for reporting current total amount of
+ * used GTT.
+ * The file mem_info_gtt_used is used for this, and returns the current used
+ * size of the GTT block, in bytes
  */
-static int amdgpu_gtt_mgr_fini(struct ttm_mem_type_manager *man)
+static ssize_t amdgpu_mem_info_gtt_used_show(struct device *dev,
+					     struct device_attribute *attr,
+					     char *buf)
 {
-	struct amdgpu_gtt_mgr *mgr = man->priv;
-
-	spin_lock(&mgr->lock);
-	if (!drm_mm_clean(&mgr->mm)) {
-		spin_unlock(&mgr->lock);
-		return -EBUSY;
-	}
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct ttm_resource_manager *man = &adev->mman.gtt_mgr.manager;
 
-	drm_mm_takedown(&mgr->mm);
-	spin_unlock(&mgr->lock);
-	kfree(mgr);
-	man->priv = NULL;
-	return 0;
+	return sysfs_emit(buf, "%llu\n", ttm_resource_manager_usage(man));
 }
 
+static DEVICE_ATTR(mem_info_gtt_total, S_IRUGO,
+	           amdgpu_mem_info_gtt_total_show, NULL);
+static DEVICE_ATTR(mem_info_gtt_used, S_IRUGO,
+	           amdgpu_mem_info_gtt_used_show, NULL);
+
+static struct attribute *amdgpu_gtt_mgr_attributes[] = {
+	&dev_attr_mem_info_gtt_total.attr,
+	&dev_attr_mem_info_gtt_used.attr,
+	NULL
+};
+
+const struct attribute_group amdgpu_gtt_mgr_attr_group = {
+	.attrs = amdgpu_gtt_mgr_attributes
+};
+
 /**
- * amdgpu_gtt_mgr_alloc - allocate new ranges
+ * amdgpu_gtt_mgr_has_gart_addr - Check if mem has address space
  *
- * @man: TTM memory type manager
- * @tbo: TTM BO we need this range for
- * @place: placement flags and restrictions
- * @mem: the resulting mem object
+ * @res: the mem object to check
  *
- * Allocate the address space for a node.
+ * Check if a mem object has already address space allocated.
  */
-int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
-			 struct ttm_buffer_object *tbo,
-			 const struct ttm_place *place,
-			 struct ttm_mem_reg *mem)
+bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_resource *res)
 {
-	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_mm_node *node = mem->mm_node;
-	enum drm_mm_insert_mode mode;
-	unsigned long fpfn, lpfn;
-	int r;
-
-	if (node->start != AMDGPU_BO_INVALID_OFFSET)
-		return 0;
-
-	if (place)
-		fpfn = place->fpfn;
-	else
-		fpfn = 0;
+	struct ttm_range_mgr_node *node = to_ttm_range_mgr_node(res);
 
-	if (place && place->lpfn)
-		lpfn = place->lpfn;
-	else
-		lpfn = man->size;
-
-	mode = DRM_MM_INSERT_BEST;
-	if (place && place->flags & TTM_PL_FLAG_TOPDOWN)
-		mode = DRM_MM_INSERT_HIGH;
-
-	spin_lock(&mgr->lock);
-	r = drm_mm_insert_node_in_range(&mgr->mm, node,
-					mem->num_pages, mem->page_alignment, 0,
-					fpfn, lpfn, mode);
-	spin_unlock(&mgr->lock);
-
-	if (!r) {
-		mem->start = node->start;
-		if (&tbo->mem == mem)
-			tbo->offset = (tbo->mem.start << PAGE_SHIFT) +
-			    tbo->bdev->man[tbo->mem.mem_type].gpu_offset;
-	}
-
-	return r;
+	return drm_mm_node_allocated(&node->mm_nodes[0]);
 }
 
-void amdgpu_gtt_mgr_print(struct seq_file *m, struct ttm_mem_type_manager *man)
-{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
-	struct amdgpu_gtt_mgr *mgr = man->priv;
-
-	seq_printf(m, "man size:%llu pages, gtt available:%llu pages, usage:%lluMB\n",
-		   man->size, mgr->available, (u64)atomic64_read(&adev->gtt_usage) >> 20);
-
-}
 /**
  * amdgpu_gtt_mgr_new - allocate a new node
  *
  * @man: TTM memory type manager
  * @tbo: TTM BO we need this range for
  * @place: placement flags and restrictions
- * @mem: the resulting mem object
+ * @res: the resulting mem object
  *
  * Dummy, allocate the node but no space for it yet.
  */
-static int amdgpu_gtt_mgr_new(struct ttm_mem_type_manager *man,
+static int amdgpu_gtt_mgr_new(struct ttm_resource_manager *man,
 			      struct ttm_buffer_object *tbo,
 			      const struct ttm_place *place,
-			      struct ttm_mem_reg *mem)
+			      struct ttm_resource **res)
 {
-	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_mm_node *node;
+	struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
+	uint32_t num_pages = PFN_UP(tbo->base.size);
+	struct ttm_range_mgr_node *node;
 	int r;
 
-	spin_lock(&mgr->lock);
-	if (mgr->available < mem->num_pages) {
-		spin_unlock(&mgr->lock);
-		return 0;
-	}
-	mgr->available -= mem->num_pages;
-	spin_unlock(&mgr->lock);
+	node = kzalloc(struct_size(node, mm_nodes, 1), GFP_KERNEL);
+	if (!node)
+		return -ENOMEM;
 
-	node = kzalloc(sizeof(*node), GFP_KERNEL);
-	if (!node) {
-		r = -ENOMEM;
-		goto err_out;
+	ttm_resource_init(tbo, place, &node->base);
+	if (!(place->flags & TTM_PL_FLAG_TEMPORARY) &&
+	    ttm_resource_manager_usage(man) > man->size) {
+		r = -ENOSPC;
+		goto err_free;
 	}
 
-	node->start = AMDGPU_BO_INVALID_OFFSET;
-	node->size = mem->num_pages;
-	mem->mm_node = node;
-
-	if (place->fpfn || place->lpfn || place->flags & TTM_PL_FLAG_TOPDOWN) {
-		r = amdgpu_gtt_mgr_alloc(man, tbo, place, mem);
-		if (unlikely(r)) {
-			kfree(node);
-			mem->mm_node = NULL;
-			r = 0;
-			goto err_out;
-		}
+	if (place->lpfn) {
+		spin_lock(&mgr->lock);
+		r = drm_mm_insert_node_in_range(&mgr->mm, &node->mm_nodes[0],
+						num_pages, tbo->page_alignment,
+						0, place->fpfn, place->lpfn,
+						DRM_MM_INSERT_BEST);
+		spin_unlock(&mgr->lock);
+		if (unlikely(r))
+			goto err_free;
+
+		node->base.start = node->mm_nodes[0].start;
 	} else {
-		mem->start = node->start;
+		node->mm_nodes[0].start = 0;
+		node->mm_nodes[0].size = PFN_UP(node->base.size);
+		node->base.start = AMDGPU_BO_INVALID_OFFSET;
 	}
 
+	*res = &node->base;
 	return 0;
-err_out:
-	spin_lock(&mgr->lock);
-	mgr->available += mem->num_pages;
-	spin_unlock(&mgr->lock);
 
+err_free:
+	ttm_resource_fini(man, &node->base);
+	kfree(node);
 	return r;
 }
 
@@ -205,54 +161,163 @@ err_out:
  * amdgpu_gtt_mgr_del - free ranges
  *
  * @man: TTM memory type manager
- * @tbo: TTM BO we need this range for
- * @place: placement flags and restrictions
- * @mem: TTM memory object
+ * @res: TTM memory object
  *
  * Free the allocated GTT again.
  */
-static void amdgpu_gtt_mgr_del(struct ttm_mem_type_manager *man,
-			       struct ttm_mem_reg *mem)
+static void amdgpu_gtt_mgr_del(struct ttm_resource_manager *man,
+			       struct ttm_resource *res)
 {
-	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_mm_node *node = mem->mm_node;
-
-	if (!node)
-		return;
+	struct ttm_range_mgr_node *node = to_ttm_range_mgr_node(res);
+	struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
 
 	spin_lock(&mgr->lock);
-	if (node->start != AMDGPU_BO_INVALID_OFFSET)
-		drm_mm_remove_node(node);
-	mgr->available += mem->num_pages;
+	if (drm_mm_node_allocated(&node->mm_nodes[0]))
+		drm_mm_remove_node(&node->mm_nodes[0]);
 	spin_unlock(&mgr->lock);
 
+	ttm_resource_fini(man, res);
 	kfree(node);
-	mem->mm_node = NULL;
+}
+
+/**
+ * amdgpu_gtt_mgr_recover - re-init gart
+ *
+ * @mgr: amdgpu_gtt_mgr pointer
+ *
+ * Re-init the gart for each known BO in the GTT.
+ */
+void amdgpu_gtt_mgr_recover(struct amdgpu_gtt_mgr *mgr)
+{
+	struct ttm_range_mgr_node *node;
+	struct drm_mm_node *mm_node;
+	struct amdgpu_device *adev;
+
+	adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
+	spin_lock(&mgr->lock);
+	drm_mm_for_each_node(mm_node, &mgr->mm) {
+		node = container_of(mm_node, typeof(*node), mm_nodes[0]);
+		amdgpu_ttm_recover_gart(node->base.bo);
+	}
+	spin_unlock(&mgr->lock);
+}
+
+/**
+ * amdgpu_gtt_mgr_intersects - test for intersection
+ *
+ * @man: Our manager object
+ * @res: The resource to test
+ * @place: The place for the new allocation
+ * @size: The size of the new allocation
+ *
+ * Simplified intersection test, only interesting if we need GART or not.
+ */
+static bool amdgpu_gtt_mgr_intersects(struct ttm_resource_manager *man,
+				      struct ttm_resource *res,
+				      const struct ttm_place *place,
+				      size_t size)
+{
+	return !place->lpfn || amdgpu_gtt_mgr_has_gart_addr(res);
+}
+
+/**
+ * amdgpu_gtt_mgr_compatible - test for compatibility
+ *
+ * @man: Our manager object
+ * @res: The resource to test
+ * @place: The place for the new allocation
+ * @size: The size of the new allocation
+ *
+ * Simplified compatibility test.
+ */
+static bool amdgpu_gtt_mgr_compatible(struct ttm_resource_manager *man,
+				      struct ttm_resource *res,
+				      const struct ttm_place *place,
+				      size_t size)
+{
+	return !place->lpfn || amdgpu_gtt_mgr_has_gart_addr(res);
 }
 
 /**
  * amdgpu_gtt_mgr_debug - dump VRAM table
  *
  * @man: TTM memory type manager
- * @prefix: text prefix
+ * @printer: DRM printer to use
  *
  * Dump the table content using printk.
  */
-static void amdgpu_gtt_mgr_debug(struct ttm_mem_type_manager *man,
-				  const char *prefix)
+static void amdgpu_gtt_mgr_debug(struct ttm_resource_manager *man,
+				 struct drm_printer *printer)
 {
-	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_printer p = drm_debug_printer(prefix);
+	struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
 
 	spin_lock(&mgr->lock);
-	drm_mm_print(&mgr->mm, &p);
+	drm_mm_print(&mgr->mm, printer);
 	spin_unlock(&mgr->lock);
 }
 
-const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func = {
-	.init = amdgpu_gtt_mgr_init,
-	.takedown = amdgpu_gtt_mgr_fini,
-	.get_node = amdgpu_gtt_mgr_new,
-	.put_node = amdgpu_gtt_mgr_del,
+static const struct ttm_resource_manager_func amdgpu_gtt_mgr_func = {
+	.alloc = amdgpu_gtt_mgr_new,
+	.free = amdgpu_gtt_mgr_del,
+	.intersects = amdgpu_gtt_mgr_intersects,
+	.compatible = amdgpu_gtt_mgr_compatible,
 	.debug = amdgpu_gtt_mgr_debug
 };
+
+/**
+ * amdgpu_gtt_mgr_init - init GTT manager and DRM MM
+ *
+ * @adev: amdgpu_device pointer
+ * @gtt_size: maximum size of GTT
+ *
+ * Allocate and initialize the GTT manager.
+ */
+int amdgpu_gtt_mgr_init(struct amdgpu_device *adev, uint64_t gtt_size)
+{
+	struct amdgpu_gtt_mgr *mgr = &adev->mman.gtt_mgr;
+	struct ttm_resource_manager *man = &mgr->manager;
+	uint64_t start, size;
+
+	man->use_tt = true;
+	man->func = &amdgpu_gtt_mgr_func;
+
+	ttm_resource_manager_init(man, &adev->mman.bdev, gtt_size);
+
+	start = AMDGPU_GTT_MAX_TRANSFER_SIZE * AMDGPU_GTT_NUM_TRANSFER_WINDOWS;
+	start += amdgpu_vce_required_gart_pages(adev);
+	size = (adev->gmc.gart_size >> PAGE_SHIFT) - start;
+	drm_mm_init(&mgr->mm, start, size);
+	spin_lock_init(&mgr->lock);
+
+	ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_TT, &mgr->manager);
+	ttm_resource_manager_set_used(man, true);
+	return 0;
+}
+
+/**
+ * amdgpu_gtt_mgr_fini - free and destroy GTT manager
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Destroy and free the GTT manager, returns -EBUSY if ranges are still
+ * allocated inside it.
+ */
+void amdgpu_gtt_mgr_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_gtt_mgr *mgr = &adev->mman.gtt_mgr;
+	struct ttm_resource_manager *man = &mgr->manager;
+	int ret;
+
+	ttm_resource_manager_set_used(man, false);
+
+	ret = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
+	if (ret)
+		return;
+
+	spin_lock(&mgr->lock);
+	drm_mm_takedown(&mgr->mm);
+	spin_unlock(&mgr->lock);
+
+	ttm_resource_manager_cleanup(man);
+	ttm_set_driver_manager(&adev->mman.bdev, TTM_PL_TT, NULL);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_hdp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_hdp.c
new file mode 100644
index 000000000000..5a60d69a3e1f
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_hdp.c
@@ -0,0 +1,84 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include "amdgpu.h"
+#include "amdgpu_ras.h"
+#include <uapi/linux/kfd_ioctl.h>
+
+int amdgpu_hdp_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err;
+	struct amdgpu_hdp_ras *ras;
+
+	if (!adev->hdp.ras)
+		return 0;
+
+	ras = adev->hdp.ras;
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register hdp ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "hdp");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__HDP;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+	adev->hdp.ras_if = &ras->ras_block.ras_comm;
+
+	/* hdp ras follows amdgpu_ras_block_late_init_default for late init */
+	return 0;
+}
+
+void amdgpu_hdp_generic_flush(struct amdgpu_device *adev,
+			      struct amdgpu_ring *ring)
+{
+	if (!ring || !ring->funcs->emit_wreg) {
+		WREG32((adev->rmmio_remap.reg_offset +
+			KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >>
+			       2,
+		       0);
+		if (adev->nbio.funcs->get_memsize)
+			adev->nbio.funcs->get_memsize(adev);
+	} else {
+		amdgpu_ring_emit_wreg(ring,
+				      (adev->rmmio_remap.reg_offset +
+				       KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL) >>
+					      2,
+				      0);
+	}
+}
+
+void amdgpu_hdp_invalidate(struct amdgpu_device *adev, struct amdgpu_ring *ring)
+{
+	if (adev->asic_funcs && adev->asic_funcs->invalidate_hdp)
+		adev->asic_funcs->invalidate_hdp(adev, ring);
+	else if (adev->hdp.funcs && adev->hdp.funcs->invalidate_hdp)
+		adev->hdp.funcs->invalidate_hdp(adev, ring);
+}
+
+void amdgpu_hdp_flush(struct amdgpu_device *adev, struct amdgpu_ring *ring)
+{
+	if (adev->asic_funcs && adev->asic_funcs->flush_hdp)
+		adev->asic_funcs->flush_hdp(adev, ring);
+	else if (adev->hdp.funcs && adev->hdp.funcs->flush_hdp)
+		adev->hdp.funcs->flush_hdp(adev, ring);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_hdp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_hdp.h
new file mode 100644
index 000000000000..d9f488fa76b9
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_hdp.h
@@ -0,0 +1,53 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_HDP_H__
+#define __AMDGPU_HDP_H__
+#include "amdgpu_ras.h"
+
+struct amdgpu_hdp_ras {
+	struct amdgpu_ras_block_object ras_block;
+};
+
+struct amdgpu_hdp_funcs {
+	void (*flush_hdp)(struct amdgpu_device *adev, struct amdgpu_ring *ring);
+	void (*invalidate_hdp)(struct amdgpu_device *adev,
+			       struct amdgpu_ring *ring);
+	void (*update_clock_gating)(struct amdgpu_device *adev, bool enable);
+	void (*get_clock_gating_state)(struct amdgpu_device *adev, u64 *flags);
+	void (*init_registers)(struct amdgpu_device *adev);
+};
+
+struct amdgpu_hdp {
+	struct ras_common_if			*ras_if;
+	const struct amdgpu_hdp_funcs		*funcs;
+	struct amdgpu_hdp_ras	*ras;
+};
+
+int amdgpu_hdp_ras_sw_init(struct amdgpu_device *adev);
+void amdgpu_hdp_generic_flush(struct amdgpu_device *adev,
+			      struct amdgpu_ring *ring);
+void amdgpu_hdp_invalidate(struct amdgpu_device *adev,
+			   struct amdgpu_ring *ring);
+void amdgpu_hdp_flush(struct amdgpu_device *adev,
+		      struct amdgpu_ring *ring);
+#endif /* __AMDGPU_HDP_H__ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c
new file mode 100644
index 000000000000..90d26d820bac
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c
@@ -0,0 +1,293 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+/*
+ * Authors:
+ *    Christian König <christian.koenig@amd.com>
+ */
+
+/**
+ * DOC: MMU Notifier
+ *
+ * For coherent userptr handling registers an MMU notifier to inform the driver
+ * about updates on the page tables of a process.
+ *
+ * When somebody tries to invalidate the page tables we block the update until
+ * all operations on the pages in question are completed, then those pages are
+ * marked as accessed and also dirty if it wasn't a read only access.
+ *
+ * New command submissions using the userptrs in question are delayed until all
+ * page table invalidation are completed and we once more see a coherent process
+ * address space.
+ */
+
+#include <linux/firmware.h>
+#include <linux/module.h>
+#include <drm/drm.h>
+
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_hmm.h"
+
+#define MAX_WALK_BYTE	(2UL << 30)
+
+/**
+ * amdgpu_hmm_invalidate_gfx - callback to notify about mm change
+ *
+ * @mni: the range (mm) is about to update
+ * @range: details on the invalidation
+ * @cur_seq: Value to pass to mmu_interval_set_seq()
+ *
+ * Block for operations on BOs to finish and mark pages as accessed and
+ * potentially dirty.
+ */
+static bool amdgpu_hmm_invalidate_gfx(struct mmu_interval_notifier *mni,
+				      const struct mmu_notifier_range *range,
+				      unsigned long cur_seq)
+{
+	struct amdgpu_bo *bo = container_of(mni, struct amdgpu_bo, notifier);
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	long r;
+
+	if (!mmu_notifier_range_blockable(range))
+		return false;
+
+	mutex_lock(&adev->notifier_lock);
+
+	mmu_interval_set_seq(mni, cur_seq);
+
+	r = dma_resv_wait_timeout(bo->tbo.base.resv, DMA_RESV_USAGE_BOOKKEEP,
+				  false, MAX_SCHEDULE_TIMEOUT);
+	mutex_unlock(&adev->notifier_lock);
+	if (r <= 0)
+		DRM_ERROR("(%ld) failed to wait for user bo\n", r);
+	return true;
+}
+
+static const struct mmu_interval_notifier_ops amdgpu_hmm_gfx_ops = {
+	.invalidate = amdgpu_hmm_invalidate_gfx,
+};
+
+/**
+ * amdgpu_hmm_invalidate_hsa - callback to notify about mm change
+ *
+ * @mni: the range (mm) is about to update
+ * @range: details on the invalidation
+ * @cur_seq: Value to pass to mmu_interval_set_seq()
+ *
+ * We temporarily evict the BO attached to this range. This necessitates
+ * evicting all user-mode queues of the process.
+ */
+static bool amdgpu_hmm_invalidate_hsa(struct mmu_interval_notifier *mni,
+				      const struct mmu_notifier_range *range,
+				      unsigned long cur_seq)
+{
+	struct amdgpu_bo *bo = container_of(mni, struct amdgpu_bo, notifier);
+
+	if (!mmu_notifier_range_blockable(range))
+		return false;
+
+	amdgpu_amdkfd_evict_userptr(mni, cur_seq, bo->kfd_bo);
+
+	return true;
+}
+
+static const struct mmu_interval_notifier_ops amdgpu_hmm_hsa_ops = {
+	.invalidate = amdgpu_hmm_invalidate_hsa,
+};
+
+/**
+ * amdgpu_hmm_register - register a BO for notifier updates
+ *
+ * @bo: amdgpu buffer object
+ * @addr: userptr addr we should monitor
+ *
+ * Registers a mmu_notifier for the given BO at the specified address.
+ * Returns 0 on success, -ERRNO if anything goes wrong.
+ */
+int amdgpu_hmm_register(struct amdgpu_bo *bo, unsigned long addr)
+{
+	int r;
+
+	if (bo->kfd_bo)
+		r = mmu_interval_notifier_insert(&bo->notifier, current->mm,
+						    addr, amdgpu_bo_size(bo),
+						    &amdgpu_hmm_hsa_ops);
+	else
+		r = mmu_interval_notifier_insert(&bo->notifier, current->mm, addr,
+							amdgpu_bo_size(bo),
+							&amdgpu_hmm_gfx_ops);
+	if (r)
+		/*
+		 * Make sure amdgpu_hmm_unregister() doesn't call
+		 * mmu_interval_notifier_remove() when the notifier isn't properly
+		 * initialized.
+		 */
+		bo->notifier.mm = NULL;
+
+	return r;
+}
+
+/**
+ * amdgpu_hmm_unregister - unregister a BO for notifier updates
+ *
+ * @bo: amdgpu buffer object
+ *
+ * Remove any registration of mmu notifier updates from the buffer object.
+ */
+void amdgpu_hmm_unregister(struct amdgpu_bo *bo)
+{
+	if (!bo->notifier.mm)
+		return;
+	mmu_interval_notifier_remove(&bo->notifier);
+	bo->notifier.mm = NULL;
+}
+
+int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier,
+			       uint64_t start, uint64_t npages, bool readonly,
+			       void *owner,
+			       struct amdgpu_hmm_range *range)
+{
+	unsigned long end;
+	unsigned long timeout;
+	unsigned long *pfns;
+	int r = 0;
+	struct hmm_range *hmm_range = &range->hmm_range;
+
+	pfns = kvmalloc_array(npages, sizeof(*pfns), GFP_KERNEL);
+	if (unlikely(!pfns)) {
+		r = -ENOMEM;
+		goto out_free_range;
+	}
+
+	hmm_range->notifier = notifier;
+	hmm_range->default_flags = HMM_PFN_REQ_FAULT;
+	if (!readonly)
+		hmm_range->default_flags |= HMM_PFN_REQ_WRITE;
+	hmm_range->hmm_pfns = pfns;
+	hmm_range->start = start;
+	end = start + npages * PAGE_SIZE;
+	hmm_range->dev_private_owner = owner;
+
+	do {
+		hmm_range->end = min(hmm_range->start + MAX_WALK_BYTE, end);
+
+		pr_debug("hmm range: start = 0x%lx, end = 0x%lx",
+			hmm_range->start, hmm_range->end);
+
+		timeout = jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT);
+
+retry:
+		hmm_range->notifier_seq = mmu_interval_read_begin(notifier);
+		r = hmm_range_fault(hmm_range);
+		if (unlikely(r)) {
+			if (r == -EBUSY && !time_after(jiffies, timeout))
+				goto retry;
+			goto out_free_pfns;
+		}
+
+		if (hmm_range->end == end)
+			break;
+		hmm_range->hmm_pfns += MAX_WALK_BYTE >> PAGE_SHIFT;
+		hmm_range->start = hmm_range->end;
+	} while (hmm_range->end < end);
+
+	hmm_range->start = start;
+	hmm_range->hmm_pfns = pfns;
+
+	return 0;
+
+out_free_pfns:
+	kvfree(pfns);
+	hmm_range->hmm_pfns = NULL;
+out_free_range:
+	if (r == -EBUSY)
+		r = -EAGAIN;
+	return r;
+}
+
+/**
+ * amdgpu_hmm_range_valid - check if an HMM range is still valid
+ * @range: pointer to the &struct amdgpu_hmm_range to validate
+ *
+ * Determines whether the given HMM range @range is still valid by
+ * checking for invalidations via the MMU notifier sequence. This is
+ * typically used to verify that the range has not been invalidated
+ * by concurrent address space updates before it is accessed.
+ *
+ * Return:
+ * * true if @range is valid and can be used safely
+ * * false if @range is NULL or has been invalidated
+ */
+bool amdgpu_hmm_range_valid(struct amdgpu_hmm_range *range)
+{
+	if (!range)
+		return false;
+
+	return !mmu_interval_read_retry(range->hmm_range.notifier,
+					range->hmm_range.notifier_seq);
+}
+
+/**
+ * amdgpu_hmm_range_alloc - allocate and initialize an AMDGPU HMM range
+ * @bo: optional buffer object to associate with this HMM range
+ *
+ * Allocates memory for amdgpu_hmm_range and associates it with the @bo passed.
+ * The reference count of the @bo is incremented.
+ *
+ * Return:
+ * Pointer to a newly allocated struct amdgpu_hmm_range on success,
+ * or NULL if memory allocation fails.
+ */
+struct amdgpu_hmm_range *amdgpu_hmm_range_alloc(struct amdgpu_bo *bo)
+{
+	struct amdgpu_hmm_range *range;
+
+	range = kzalloc(sizeof(*range), GFP_KERNEL);
+	if (!range)
+		return NULL;
+
+	range->bo = amdgpu_bo_ref(bo);
+	return range;
+}
+
+/**
+ * amdgpu_hmm_range_free - release an AMDGPU HMM range
+ * @range: pointer to the range object to free
+ *
+ * Releases all resources held by @range, including the associated
+ * hmm_pfns and the dropping reference of associated bo if any.
+ *
+ * Return: void
+ */
+void amdgpu_hmm_range_free(struct amdgpu_hmm_range *range)
+{
+	if (!range)
+		return;
+
+	kvfree(range->hmm_range.hmm_pfns);
+	amdgpu_bo_unref(&range->bo);
+	kfree(range);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.h
new file mode 100644
index 000000000000..140bc9cd57b4
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Christian König
+ */
+#ifndef __AMDGPU_MN_H__
+#define __AMDGPU_MN_H__
+
+#include <linux/types.h>
+#include <linux/hmm.h>
+#include <linux/rwsem.h>
+#include <linux/workqueue.h>
+#include <linux/interval_tree.h>
+#include <linux/mmu_notifier.h>
+
+struct amdgpu_hmm_range {
+	struct hmm_range hmm_range;
+	struct amdgpu_bo *bo;
+};
+
+int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier,
+			       uint64_t start, uint64_t npages, bool readonly,
+			       void *owner,
+			       struct amdgpu_hmm_range *range);
+
+#if defined(CONFIG_HMM_MIRROR)
+bool amdgpu_hmm_range_valid(struct amdgpu_hmm_range *range);
+struct amdgpu_hmm_range *amdgpu_hmm_range_alloc(struct amdgpu_bo *bo);
+void amdgpu_hmm_range_free(struct amdgpu_hmm_range *range);
+int amdgpu_hmm_register(struct amdgpu_bo *bo, unsigned long addr);
+void amdgpu_hmm_unregister(struct amdgpu_bo *bo);
+#else
+static inline int amdgpu_hmm_register(struct amdgpu_bo *bo, unsigned long addr)
+{
+	DRM_WARN_ONCE("HMM_MIRROR kernel config option is not enabled, "
+		      "add CONFIG_ZONE_DEVICE=y in config file to fix this\n");
+	return -ENODEV;
+}
+
+static inline void amdgpu_hmm_unregister(struct amdgpu_bo *bo) {}
+
+static inline bool amdgpu_hmm_range_valid(struct amdgpu_hmm_range *range)
+{
+	return false;
+}
+
+static inline struct amdgpu_hmm_range *amdgpu_hmm_range_alloc(struct amdgpu_bo *bo)
+{
+	return NULL;
+}
+
+static inline void amdgpu_hmm_range_free(struct amdgpu_hmm_range *range) {}
+#endif
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.c
index f2739995c335..9cb72f0c5277 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.c
@@ -23,9 +23,9 @@
  * Authors: Dave Airlie
  *          Alex Deucher
  */
-#include <linux/export.h>
 
-#include <drm/drmP.h>
+#include <linux/pci.h>
+
 #include <drm/drm_edid.h>
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
@@ -39,7 +39,7 @@
 static int amdgpu_i2c_pre_xfer(struct i2c_adapter *i2c_adap)
 {
 	struct amdgpu_i2c_chan *i2c = i2c_get_adapdata(i2c_adap);
-	struct amdgpu_device *adev = i2c->dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(i2c->dev);
 	struct amdgpu_i2c_bus_rec *rec = &i2c->rec;
 	uint32_t temp;
 
@@ -81,7 +81,7 @@ static int amdgpu_i2c_pre_xfer(struct i2c_adapter *i2c_adap)
 static void amdgpu_i2c_post_xfer(struct i2c_adapter *i2c_adap)
 {
 	struct amdgpu_i2c_chan *i2c = i2c_get_adapdata(i2c_adap);
-	struct amdgpu_device *adev = i2c->dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(i2c->dev);
 	struct amdgpu_i2c_bus_rec *rec = &i2c->rec;
 	uint32_t temp;
 
@@ -100,7 +100,7 @@ static void amdgpu_i2c_post_xfer(struct i2c_adapter *i2c_adap)
 static int amdgpu_i2c_get_clock(void *i2c_priv)
 {
 	struct amdgpu_i2c_chan *i2c = i2c_priv;
-	struct amdgpu_device *adev = i2c->dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(i2c->dev);
 	struct amdgpu_i2c_bus_rec *rec = &i2c->rec;
 	uint32_t val;
 
@@ -115,7 +115,7 @@ static int amdgpu_i2c_get_clock(void *i2c_priv)
 static int amdgpu_i2c_get_data(void *i2c_priv)
 {
 	struct amdgpu_i2c_chan *i2c = i2c_priv;
-	struct amdgpu_device *adev = i2c->dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(i2c->dev);
 	struct amdgpu_i2c_bus_rec *rec = &i2c->rec;
 	uint32_t val;
 
@@ -129,7 +129,7 @@ static int amdgpu_i2c_get_data(void *i2c_priv)
 static void amdgpu_i2c_set_clock(void *i2c_priv, int clock)
 {
 	struct amdgpu_i2c_chan *i2c = i2c_priv;
-	struct amdgpu_device *adev = i2c->dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(i2c->dev);
 	struct amdgpu_i2c_bus_rec *rec = &i2c->rec;
 	uint32_t val;
 
@@ -142,7 +142,7 @@ static void amdgpu_i2c_set_clock(void *i2c_priv, int clock)
 static void amdgpu_i2c_set_data(void *i2c_priv, int data)
 {
 	struct amdgpu_i2c_chan *i2c = i2c_priv;
-	struct amdgpu_device *adev = i2c->dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(i2c->dev);
 	struct amdgpu_i2c_bus_rec *rec = &i2c->rec;
 	uint32_t val;
 
@@ -174,8 +174,7 @@ struct amdgpu_i2c_chan *amdgpu_i2c_create(struct drm_device *dev,
 
 	i2c->rec = *rec;
 	i2c->adapter.owner = THIS_MODULE;
-	i2c->adapter.class = I2C_CLASS_DDC;
-	i2c->adapter.dev.parent = &dev->pdev->dev;
+	i2c->adapter.dev.parent = dev->dev;
 	i2c->dev = dev;
 	i2c_set_adapdata(&i2c->adapter, i2c);
 	mutex_init(&i2c->mutex);
@@ -185,7 +184,7 @@ struct amdgpu_i2c_chan *amdgpu_i2c_create(struct drm_device *dev,
 		snprintf(i2c->adapter.name, sizeof(i2c->adapter.name),
 			 "AMDGPU i2c hw bus %s", name);
 		i2c->adapter.algo = &amdgpu_atombios_i2c_algo;
-		ret = i2c_add_adapter(&i2c->adapter);
+		ret = devm_i2c_add_adapter(dev->dev, &i2c->adapter);
 		if (ret)
 			goto out_free;
 	} else {
@@ -216,22 +215,23 @@ out_free:
 
 }
 
-void amdgpu_i2c_destroy(struct amdgpu_i2c_chan *i2c)
-{
-	if (!i2c)
-		return;
-	WARN_ON(i2c->has_aux);
-	i2c_del_adapter(&i2c->adapter);
-	kfree(i2c);
-}
-
-/* Add the default buses */
 void amdgpu_i2c_init(struct amdgpu_device *adev)
 {
-	if (amdgpu_hw_i2c)
-		DRM_INFO("hw_i2c forced on, you may experience display detection problems!\n");
-
-	amdgpu_atombios_i2c_init(adev);
+	if (!adev->is_atom_fw) {
+		if (!amdgpu_device_has_dc_support(adev)) {
+			amdgpu_atombios_i2c_init(adev);
+		} else {
+			switch (adev->asic_type) {
+			case CHIP_POLARIS10:
+			case CHIP_POLARIS11:
+			case CHIP_POLARIS12:
+				amdgpu_atombios_oem_i2c_init(adev, 0x97);
+				break;
+			default:
+				break;
+			}
+		}
+	}
 }
 
 /* remove all the buses */
@@ -239,28 +239,9 @@ void amdgpu_i2c_fini(struct amdgpu_device *adev)
 {
 	int i;
 
-	for (i = 0; i < AMDGPU_MAX_I2C_BUS; i++) {
-		if (adev->i2c_bus[i]) {
-			amdgpu_i2c_destroy(adev->i2c_bus[i]);
+	for (i = 0; i < AMDGPU_MAX_I2C_BUS; i++)
+		if (adev->i2c_bus[i])
 			adev->i2c_bus[i] = NULL;
-		}
-	}
-}
-
-/* Add additional buses */
-void amdgpu_i2c_add(struct amdgpu_device *adev,
-		    const struct amdgpu_i2c_bus_rec *rec,
-		    const char *name)
-{
-	struct drm_device *dev = adev->ddev;
-	int i;
-
-	for (i = 0; i < AMDGPU_MAX_I2C_BUS; i++) {
-		if (!adev->i2c_bus[i]) {
-			adev->i2c_bus[i] = amdgpu_i2c_create(dev, rec, name);
-			return;
-		}
-	}
 }
 
 /* looks up bus based on id */
@@ -279,7 +260,7 @@ amdgpu_i2c_lookup(struct amdgpu_device *adev,
 	return NULL;
 }
 
-static void amdgpu_i2c_get_byte(struct amdgpu_i2c_chan *i2c_bus,
+static int amdgpu_i2c_get_byte(struct amdgpu_i2c_chan *i2c_bus,
 				 u8 slave_addr,
 				 u8 addr,
 				 u8 *val)
@@ -304,16 +285,18 @@ static void amdgpu_i2c_get_byte(struct amdgpu_i2c_chan *i2c_bus,
 	out_buf[0] = addr;
 	out_buf[1] = 0;
 
-	if (i2c_transfer(&i2c_bus->adapter, msgs, 2) == 2) {
-		*val = in_buf[0];
-		DRM_DEBUG("val = 0x%02x\n", *val);
-	} else {
-		DRM_DEBUG("i2c 0x%02x 0x%02x read failed\n",
-			  addr, *val);
+	if (i2c_transfer(&i2c_bus->adapter, msgs, 2) != 2) {
+		DRM_DEBUG("i2c 0x%02x read failed\n", addr);
+		return -EIO;
 	}
+
+	*val = in_buf[0];
+	DRM_DEBUG("val = 0x%02x\n", *val);
+
+	return 0;
 }
 
-static void amdgpu_i2c_put_byte(struct amdgpu_i2c_chan *i2c_bus,
+static int amdgpu_i2c_put_byte(struct amdgpu_i2c_chan *i2c_bus,
 				 u8 slave_addr,
 				 u8 addr,
 				 u8 val)
@@ -329,16 +312,19 @@ static void amdgpu_i2c_put_byte(struct amdgpu_i2c_chan *i2c_bus,
 	out_buf[0] = addr;
 	out_buf[1] = val;
 
-	if (i2c_transfer(&i2c_bus->adapter, &msg, 1) != 1)
-		DRM_DEBUG("i2c 0x%02x 0x%02x write failed\n",
-			  addr, val);
+	if (i2c_transfer(&i2c_bus->adapter, &msg, 1) != 1) {
+		DRM_DEBUG("i2c 0x%02x 0x%02x write failed\n", addr, val);
+		return -EIO;
+	}
+
+	return 0;
 }
 
 /* ddc router switching */
 void
 amdgpu_i2c_router_select_ddc_port(const struct amdgpu_connector *amdgpu_connector)
 {
-	u8 val;
+	u8 val = 0;
 
 	if (!amdgpu_connector->router.ddc_valid)
 		return;
@@ -346,16 +332,18 @@ amdgpu_i2c_router_select_ddc_port(const struct amdgpu_connector *amdgpu_connecto
 	if (!amdgpu_connector->router_bus)
 		return;
 
-	amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
+	if (amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
 			    amdgpu_connector->router.i2c_addr,
-			    0x3, &val);
+			    0x3, &val))
+		return;
 	val &= ~amdgpu_connector->router.ddc_mux_control_pin;
 	amdgpu_i2c_put_byte(amdgpu_connector->router_bus,
 			    amdgpu_connector->router.i2c_addr,
 			    0x3, val);
-	amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
+	if (amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
 			    amdgpu_connector->router.i2c_addr,
-			    0x1, &val);
+			    0x1, &val))
+		return;
 	val &= ~amdgpu_connector->router.ddc_mux_control_pin;
 	val |= amdgpu_connector->router.ddc_mux_state;
 	amdgpu_i2c_put_byte(amdgpu_connector->router_bus,
@@ -375,16 +363,18 @@ amdgpu_i2c_router_select_cd_port(const struct amdgpu_connector *amdgpu_connector
 	if (!amdgpu_connector->router_bus)
 		return;
 
-	amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
+	if (amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
 			    amdgpu_connector->router.i2c_addr,
-			    0x3, &val);
+			    0x3, &val))
+		return;
 	val &= ~amdgpu_connector->router.cd_mux_control_pin;
 	amdgpu_i2c_put_byte(amdgpu_connector->router_bus,
 			    amdgpu_connector->router.i2c_addr,
 			    0x3, val);
-	amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
+	if (amdgpu_i2c_get_byte(amdgpu_connector->router_bus,
 			    amdgpu_connector->router.i2c_addr,
-			    0x1, &val);
+			    0x1, &val))
+		return;
 	val &= ~amdgpu_connector->router.cd_mux_control_pin;
 	val |= amdgpu_connector->router.cd_mux_state;
 	amdgpu_i2c_put_byte(amdgpu_connector->router_bus,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.h
index 63c2ff7499e1..1d3d3806e0dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_i2c.h
@@ -30,9 +30,6 @@ struct amdgpu_i2c_chan *amdgpu_i2c_create(struct drm_device *dev,
 void amdgpu_i2c_destroy(struct amdgpu_i2c_chan *i2c);
 void amdgpu_i2c_init(struct amdgpu_device *adev);
 void amdgpu_i2c_fini(struct amdgpu_device *adev);
-void amdgpu_i2c_add(struct amdgpu_device *adev,
-		    const struct amdgpu_i2c_bus_rec *rec,
-		    const char *name);
 struct amdgpu_i2c_chan *
 amdgpu_i2c_lookup(struct amdgpu_device *adev,
 		  const struct amdgpu_i2c_bus_rec *i2c_bus);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index 6e4ae0d983c2..586a58facca1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -28,12 +28,15 @@
  */
 #include <linux/seq_file.h>
 #include <linux/slab.h>
-#include <drm/drmP.h>
+
 #include <drm/amdgpu_drm.h>
+
 #include "amdgpu.h"
 #include "atom.h"
+#include "amdgpu_trace.h"
 
 #define AMDGPU_IB_TEST_TIMEOUT	msecs_to_jiffies(1000)
+#define AMDGPU_IB_TEST_GFX_XGMI_TIMEOUT	msecs_to_jiffies(2000)
 
 /*
  * IB
@@ -44,13 +47,14 @@
  * produce command buffers which are send to the kernel and
  * put in IBs for execution by the requested ring.
  */
-static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev);
 
 /**
  * amdgpu_ib_get - request an IB (Indirect Buffer)
  *
- * @ring: ring index the IB is associated with
+ * @adev: amdgpu_device pointer
+ * @vm: amdgpu_vm pointer
  * @size: requested IB size
+ * @pool_type: IB pool type (delayed, immediate, direct)
  * @ib: IB object returned
  *
  * Request an IB (all asics).  IBs are allocated using the
@@ -58,19 +62,22 @@ static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev);
  * Returns 0 on success, error on failure.
  */
 int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-		  unsigned size, struct amdgpu_ib *ib)
+		  unsigned int size, enum amdgpu_ib_pool_type pool_type,
+		  struct amdgpu_ib *ib)
 {
 	int r;
 
 	if (size) {
-		r = amdgpu_sa_bo_new(&adev->ring_tmp_bo,
-				      &ib->sa_bo, size, 256);
+		r = amdgpu_sa_bo_new(&adev->ib_pools[pool_type],
+				     &ib->sa_bo, size);
 		if (r) {
 			dev_err(adev->dev, "failed to get a new IB (%d)\n", r);
 			return r;
 		}
 
 		ib->ptr = amdgpu_sa_bo_cpu_addr(ib->sa_bo);
+		/* flush the cache before commit the IB */
+		ib->flags = AMDGPU_IB_FLAG_EMIT_MEM_SYNC;
 
 		if (!vm)
 			ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
@@ -82,24 +89,23 @@ int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 /**
  * amdgpu_ib_free - free an IB (Indirect Buffer)
  *
- * @adev: amdgpu_device pointer
  * @ib: IB object to free
  * @f: the fence SA bo need wait on for the ib alloation
  *
  * Free an IB (all asics).
  */
-void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib,
-		    struct dma_fence *f)
+void amdgpu_ib_free(struct amdgpu_ib *ib, struct dma_fence *f)
 {
-	amdgpu_sa_bo_free(adev, &ib->sa_bo, f);
+	amdgpu_sa_bo_free(&ib->sa_bo, f);
 }
 
 /**
  * amdgpu_ib_schedule - schedule an IB (Indirect Buffer) on the ring
  *
- * @adev: amdgpu_device pointer
+ * @ring: ring index the IB is associated with
  * @num_ibs: number of IBs to schedule
  * @ibs: IB objects to schedule
+ * @job: job to schedule
  * @f: fence created during this submission
  *
  * Schedule an IB on the associated ring (all asics).
@@ -115,19 +121,25 @@ void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib,
  * a CONST_IB), it will be put on the ring prior to the DE IB.  Prior
  * to SI there was just a DE IB.
  */
-int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
+int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned int num_ibs,
 		       struct amdgpu_ib *ibs, struct amdgpu_job *job,
 		       struct dma_fence **f)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_ib *ib = &ibs[0];
-	bool skip_preamble, need_ctx_switch;
-	unsigned patch_offset = ~0;
+	struct dma_fence *tmp = NULL;
+	struct amdgpu_fence *af;
+	bool need_ctx_switch;
 	struct amdgpu_vm *vm;
 	uint64_t fence_ctx;
 	uint32_t status = 0, alloc_size;
-
-	unsigned i;
+	unsigned int fence_flags = 0;
+	bool secure, init_shadow;
+	u64 shadow_va, csa_va, gds_va;
+	int vmid = AMDGPU_JOB_GET_VMID(job);
+	bool need_pipe_sync = false;
+	unsigned int cond_exec;
+	unsigned int i;
 	int r = 0;
 
 	if (num_ibs == 0)
@@ -136,20 +148,49 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	/* ring tests don't use a job */
 	if (job) {
 		vm = job->vm;
-		fence_ctx = job->fence_ctx;
+		fence_ctx = job->base.s_fence ?
+			job->base.s_fence->finished.context : 0;
+		shadow_va = job->shadow_va;
+		csa_va = job->csa_va;
+		gds_va = job->gds_va;
+		init_shadow = job->init_shadow;
+		af = job->hw_fence;
+		/* Save the context of the job for reset handling.
+		 * The driver needs this so it can skip the ring
+		 * contents for guilty contexts.
+		 */
+		af->context = fence_ctx;
+		/* the vm fence is also part of the job's context */
+		job->hw_vm_fence->context = fence_ctx;
 	} else {
 		vm = NULL;
 		fence_ctx = 0;
+		shadow_va = 0;
+		csa_va = 0;
+		gds_va = 0;
+		init_shadow = false;
+		af = kzalloc(sizeof(*af), GFP_ATOMIC);
+		if (!af)
+			return -ENOMEM;
 	}
 
-	if (!ring->ready) {
+	if (!ring->sched.ready) {
 		dev_err(adev->dev, "couldn't schedule ib on ring <%s>\n", ring->name);
-		return -EINVAL;
+		r = -EINVAL;
+		goto free_fence;
 	}
 
-	if (vm && !job->vm_id) {
+	if (vm && !job->vmid) {
 		dev_err(adev->dev, "VM IB without ID\n");
-		return -EINVAL;
+		r = -EINVAL;
+		goto free_fence;
+	}
+
+	if ((ib->flags & AMDGPU_IB_FLAGS_SECURE) &&
+	    (!ring->funcs->secure_submission_supported)) {
+		dev_err(adev->dev, "secure submissions not supported on ring <%s>\n", ring->name);
+		r = -EINVAL;
+		goto free_fence;
 	}
 
 	alloc_size = ring->funcs->emit_frame_size + num_ibs *
@@ -158,90 +199,146 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
 	r = amdgpu_ring_alloc(ring, alloc_size);
 	if (r) {
 		dev_err(adev->dev, "scheduling IB failed (%d).\n", r);
-		return r;
+		goto free_fence;
 	}
-	if (ring->funcs->emit_pipeline_sync && job && job->need_pipeline_sync)
-		amdgpu_ring_emit_pipeline_sync(ring);
 
-	if (vm) {
-		r = amdgpu_vm_flush(ring, job);
+	need_ctx_switch = ring->current_ctx != fence_ctx;
+	if (ring->funcs->emit_pipeline_sync && job &&
+	    ((tmp = amdgpu_sync_get_fence(&job->explicit_sync)) ||
+	     need_ctx_switch || amdgpu_vm_need_pipeline_sync(ring, job))) {
+
+		need_pipe_sync = true;
+
+		if (tmp)
+			trace_amdgpu_ib_pipe_sync(job, tmp);
+
+		dma_fence_put(tmp);
+	}
+
+	if ((ib->flags & AMDGPU_IB_FLAG_EMIT_MEM_SYNC) && ring->funcs->emit_mem_sync)
+		ring->funcs->emit_mem_sync(ring);
+
+	if (ring->funcs->emit_wave_limit &&
+	    ring->hw_prio == AMDGPU_GFX_PIPE_PRIO_HIGH)
+		ring->funcs->emit_wave_limit(ring, true);
+
+	if (ring->funcs->insert_start)
+		ring->funcs->insert_start(ring);
+
+	if (job) {
+		r = amdgpu_vm_flush(ring, job, need_pipe_sync);
 		if (r) {
 			amdgpu_ring_undo(ring);
 			return r;
 		}
 	}
 
+	amdgpu_ring_ib_begin(ring);
+
+	if (ring->funcs->emit_gfx_shadow)
+		amdgpu_ring_emit_gfx_shadow(ring, shadow_va, csa_va, gds_va,
+					    init_shadow, vmid);
+
 	if (ring->funcs->init_cond_exec)
-		patch_offset = amdgpu_ring_init_cond_exec(ring);
+		cond_exec = amdgpu_ring_init_cond_exec(ring,
+						       ring->cond_exe_gpu_addr);
 
-	if (ring->funcs->emit_hdp_flush
-#ifdef CONFIG_X86_64
-	    && !(adev->flags & AMD_IS_APU)
-#endif
-	   )
-		amdgpu_ring_emit_hdp_flush(ring);
+	amdgpu_device_flush_hdp(adev, ring);
+
+	if (need_ctx_switch)
+		status |= AMDGPU_HAVE_CTX_SWITCH;
 
-	skip_preamble = ring->current_ctx == fence_ctx;
-	need_ctx_switch = ring->current_ctx != fence_ctx;
 	if (job && ring->funcs->emit_cntxcntl) {
-		if (need_ctx_switch)
-			status |= AMDGPU_HAVE_CTX_SWITCH;
 		status |= job->preamble_status;
-
-		if (vm)
-			status |= AMDGPU_VM_DOMAIN;
+		status |= job->preemption_status;
 		amdgpu_ring_emit_cntxcntl(ring, status);
 	}
 
+	/* Setup initial TMZiness and send it off.
+	 */
+	secure = false;
+	if (job && ring->funcs->emit_frame_cntl) {
+		secure = ib->flags & AMDGPU_IB_FLAGS_SECURE;
+		amdgpu_ring_emit_frame_cntl(ring, true, secure);
+	}
+
 	for (i = 0; i < num_ibs; ++i) {
 		ib = &ibs[i];
 
-		/* drop preamble IBs if we don't have a context switch */
-		if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) &&
-			skip_preamble &&
-			!(status & AMDGPU_PREAMBLE_IB_PRESENT_FIRST) &&
-			!amdgpu_sriov_vf(adev)) /* for SRIOV preemption, Preamble CE ib must be inserted anyway */
-			continue;
+		if (job && ring->funcs->emit_frame_cntl) {
+			if (secure != !!(ib->flags & AMDGPU_IB_FLAGS_SECURE)) {
+				amdgpu_ring_emit_frame_cntl(ring, false, secure);
+				secure = !secure;
+				amdgpu_ring_emit_frame_cntl(ring, true, secure);
+			}
+		}
 
-		amdgpu_ring_emit_ib(ring, ib, job ? job->vm_id : 0,
-				    need_ctx_switch);
-		need_ctx_switch = false;
+		amdgpu_ring_emit_ib(ring, job, ib, status);
+		status &= ~AMDGPU_HAVE_CTX_SWITCH;
 	}
 
-	if (ring->funcs->emit_hdp_invalidate
-#ifdef CONFIG_X86_64
-	    && !(adev->flags & AMD_IS_APU)
-#endif
-	   )
-		amdgpu_ring_emit_hdp_invalidate(ring);
+	if (job && ring->funcs->emit_frame_cntl)
+		amdgpu_ring_emit_frame_cntl(ring, false, secure);
+
+	amdgpu_device_invalidate_hdp(adev, ring);
+
+	if (ib->flags & AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE)
+		fence_flags |= AMDGPU_FENCE_FLAG_TC_WB_ONLY;
+
+	/* wrap the last IB with fence */
+	if (job && job->uf_addr) {
+		amdgpu_ring_emit_fence(ring, job->uf_addr, job->uf_sequence,
+				       fence_flags | AMDGPU_FENCE_FLAG_64BIT);
+	}
 
-	r = amdgpu_fence_emit(ring, f);
+	if (ring->funcs->emit_gfx_shadow && ring->funcs->init_cond_exec) {
+		amdgpu_ring_emit_gfx_shadow(ring, 0, 0, 0, false, 0);
+		amdgpu_ring_init_cond_exec(ring, ring->cond_exe_gpu_addr);
+	}
+
+	r = amdgpu_fence_emit(ring, af, fence_flags);
 	if (r) {
 		dev_err(adev->dev, "failed to emit fence (%d)\n", r);
-		if (job && job->vm_id)
-			amdgpu_vm_reset_id(adev, ring->funcs->vmhub,
-					   job->vm_id);
+		if (job && job->vmid)
+			amdgpu_vmid_reset(adev, ring->vm_hub, job->vmid);
 		amdgpu_ring_undo(ring);
 		return r;
 	}
+	*f = &af->base;
+	/* get a ref for the job */
+	if (job)
+		dma_fence_get(*f);
 
 	if (ring->funcs->insert_end)
 		ring->funcs->insert_end(ring);
 
-	/* wrap the last IB with fence */
-	if (job && job->uf_addr) {
-		amdgpu_ring_emit_fence(ring, job->uf_addr, job->uf_sequence,
-				       AMDGPU_FENCE_FLAG_64BIT);
-	}
-
-	if (patch_offset != ~0 && ring->funcs->patch_cond_exec)
-		amdgpu_ring_patch_cond_exec(ring, patch_offset);
+	amdgpu_ring_patch_cond_exec(ring, cond_exec);
 
 	ring->current_ctx = fence_ctx;
-	if (vm && ring->funcs->emit_switch_buffer)
+	if (job && ring->funcs->emit_switch_buffer)
 		amdgpu_ring_emit_switch_buffer(ring);
+
+	if (ring->funcs->emit_wave_limit &&
+	    ring->hw_prio == AMDGPU_GFX_PIPE_PRIO_HIGH)
+		ring->funcs->emit_wave_limit(ring, false);
+
+	/* Save the wptr associated with this fence.
+	 * This must be last for resets to work properly
+	 * as we need to save the wptr associated with this
+	 * fence so we know what rings contents to backup
+	 * after we reset the queue.
+	 */
+	amdgpu_fence_save_wptr(af);
+
+	amdgpu_ring_ib_end(ring);
 	amdgpu_ring_commit(ring);
+
 	return 0;
+
+free_fence:
+	if (!job)
+		kfree(af);
+	return r;
 }
 
 /**
@@ -255,29 +352,26 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
  */
 int amdgpu_ib_pool_init(struct amdgpu_device *adev)
 {
-	int r;
+	int r, i;
 
-	if (adev->ib_pool_ready) {
+	if (adev->ib_pool_ready)
 		return 0;
-	}
-	r = amdgpu_sa_bo_manager_init(adev, &adev->ring_tmp_bo,
-				      AMDGPU_IB_POOL_SIZE*64*1024,
-				      AMDGPU_GPU_PAGE_SIZE,
-				      AMDGPU_GEM_DOMAIN_GTT);
-	if (r) {
-		return r;
-	}
 
-	r = amdgpu_sa_bo_manager_start(adev, &adev->ring_tmp_bo);
-	if (r) {
-		return r;
+	for (i = 0; i < AMDGPU_IB_POOL_MAX; i++) {
+		r = amdgpu_sa_bo_manager_init(adev, &adev->ib_pools[i],
+					      AMDGPU_IB_POOL_SIZE, 256,
+					      AMDGPU_GEM_DOMAIN_GTT);
+		if (r)
+			goto error;
 	}
-
 	adev->ib_pool_ready = true;
-	if (amdgpu_debugfs_sa_init(adev)) {
-		dev_err(adev->dev, "failed to register debugfs file for SA\n");
-	}
+
 	return 0;
+
+error:
+	while (i--)
+		amdgpu_sa_bo_manager_fini(adev, &adev->ib_pools[i]);
+	return r;
 }
 
 /**
@@ -290,11 +384,14 @@ int amdgpu_ib_pool_init(struct amdgpu_device *adev)
  */
 void amdgpu_ib_pool_fini(struct amdgpu_device *adev)
 {
-	if (adev->ib_pool_ready) {
-		amdgpu_sa_bo_manager_suspend(adev, &adev->ring_tmp_bo);
-		amdgpu_sa_bo_manager_fini(adev, &adev->ring_tmp_bo);
-		adev->ib_pool_ready = false;
-	}
+	int i;
+
+	if (!adev->ib_pool_ready)
+		return;
+
+	for (i = 0; i < AMDGPU_IB_POOL_MAX; i++)
+		amdgpu_sa_bo_manager_fini(adev, &adev->ib_pools[i]);
+	adev->ib_pool_ready = false;
 }
 
 /**
@@ -309,30 +406,74 @@ void amdgpu_ib_pool_fini(struct amdgpu_device *adev)
  */
 int amdgpu_ib_ring_tests(struct amdgpu_device *adev)
 {
-	unsigned i;
+	long tmo_gfx, tmo_mm;
 	int r, ret = 0;
+	unsigned int i;
+
+	tmo_mm = tmo_gfx = AMDGPU_IB_TEST_TIMEOUT;
+	if (amdgpu_sriov_vf(adev)) {
+		/* for MM engines in hypervisor side they are not scheduled together
+		 * with CP and SDMA engines, so even in exclusive mode MM engine could
+		 * still running on other VF thus the IB TEST TIMEOUT for MM engines
+		 * under SR-IOV should be set to a long time. 8 sec should be enough
+		 * for the MM comes back to this VF.
+		 */
+		tmo_mm = 8 * AMDGPU_IB_TEST_TIMEOUT;
+	}
 
-	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
+	if (amdgpu_sriov_runtime(adev)) {
+		/* for CP & SDMA engines since they are scheduled together so
+		 * need to make the timeout width enough to cover the time
+		 * cost waiting for it coming back under RUNTIME only
+		 */
+		tmo_gfx = 8 * AMDGPU_IB_TEST_TIMEOUT;
+	} else if (adev->gmc.xgmi.hive_id) {
+		tmo_gfx = AMDGPU_IB_TEST_GFX_XGMI_TIMEOUT;
+	}
+
+	for (i = 0; i < adev->num_rings; ++i) {
 		struct amdgpu_ring *ring = adev->rings[i];
+		long tmo;
 
-		if (!ring || !ring->ready)
+		/* KIQ rings don't have an IB test because we never submit IBs
+		 * to them and they have no interrupt support.
+		 */
+		if (!ring->sched.ready || !ring->funcs->test_ib)
 			continue;
 
-		r = amdgpu_ring_test_ib(ring, AMDGPU_IB_TEST_TIMEOUT);
-		if (r) {
-			ring->ready = false;
-
-			if (ring == &adev->gfx.gfx_ring[0]) {
-				/* oh, oh, that's really bad */
-				DRM_ERROR("amdgpu: failed testing IB on GFX ring (%d).\n", r);
-				adev->accel_working = false;
-				return r;
-
-			} else {
-				/* still not good, but we can live with it */
-				DRM_ERROR("amdgpu: failed testing IB on ring %d (%d).\n", i, r);
-				ret = r;
-			}
+		if (adev->enable_mes &&
+		    ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
+			continue;
+
+		/* MM engine need more time */
+		if (ring->funcs->type == AMDGPU_RING_TYPE_UVD ||
+			ring->funcs->type == AMDGPU_RING_TYPE_VCE ||
+			ring->funcs->type == AMDGPU_RING_TYPE_UVD_ENC ||
+			ring->funcs->type == AMDGPU_RING_TYPE_VCN_DEC ||
+			ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC ||
+			ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG)
+			tmo = tmo_mm;
+		else
+			tmo = tmo_gfx;
+
+		r = amdgpu_ring_test_ib(ring, tmo);
+		if (!r) {
+			DRM_DEV_DEBUG(adev->dev, "ib test on %s succeeded\n",
+				      ring->name);
+			continue;
+		}
+
+		ring->sched.ready = false;
+		DRM_DEV_ERROR(adev->dev, "IB test failed on %s (%d).\n",
+			  ring->name, r);
+
+		if (ring == &adev->gfx.gfx_ring[0]) {
+			/* oh, oh, that's really bad */
+			adev->accel_working = false;
+			return r;
+
+		} else {
+			ret = r;
 		}
 	}
 	return ret;
@@ -343,29 +484,34 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev)
  */
 #if defined(CONFIG_DEBUG_FS)
 
-static int amdgpu_debugfs_sa_info(struct seq_file *m, void *data)
+static int amdgpu_debugfs_sa_info_show(struct seq_file *m, void *unused)
 {
-	struct drm_info_node *node = (struct drm_info_node *) m->private;
-	struct drm_device *dev = node->minor->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = m->private;
 
-	amdgpu_sa_bo_dump_debug_info(&adev->ring_tmp_bo, m);
+	seq_puts(m, "--------------------- DELAYED ---------------------\n");
+	amdgpu_sa_bo_dump_debug_info(&adev->ib_pools[AMDGPU_IB_POOL_DELAYED],
+				     m);
+	seq_puts(m, "-------------------- IMMEDIATE --------------------\n");
+	amdgpu_sa_bo_dump_debug_info(&adev->ib_pools[AMDGPU_IB_POOL_IMMEDIATE],
+				     m);
+	seq_puts(m, "--------------------- DIRECT ----------------------\n");
+	amdgpu_sa_bo_dump_debug_info(&adev->ib_pools[AMDGPU_IB_POOL_DIRECT], m);
 
 	return 0;
-
 }
 
-static const struct drm_info_list amdgpu_debugfs_sa_list[] = {
-	{"amdgpu_sa_info", &amdgpu_debugfs_sa_info, 0, NULL},
-};
+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_sa_info);
 
 #endif
 
-static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev)
+void amdgpu_debugfs_sa_init(struct amdgpu_device *adev)
 {
 #if defined(CONFIG_DEBUG_FS)
-	return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_sa_list, 1);
-#else
-	return 0;
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+
+	debugfs_create_file("amdgpu_sa_info", 0444, root, adev,
+			    &amdgpu_debugfs_sa_info_fops);
+
 #endif
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
new file mode 100644
index 000000000000..9cab36322c16
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
@@ -0,0 +1,618 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include "amdgpu_ids.h"
+
+#include <linux/idr.h>
+#include <linux/dma-fence-array.h>
+
+
+#include "amdgpu.h"
+#include "amdgpu_trace.h"
+
+/*
+ * PASID manager
+ *
+ * PASIDs are global address space identifiers that can be shared
+ * between the GPU, an IOMMU and the driver. VMs on different devices
+ * may use the same PASID if they share the same address
+ * space. Therefore PASIDs are allocated using a global IDA. VMs are
+ * looked up from the PASID per amdgpu_device.
+ */
+static DEFINE_IDA(amdgpu_pasid_ida);
+
+/* Helper to free pasid from a fence callback */
+struct amdgpu_pasid_cb {
+	struct dma_fence_cb cb;
+	u32 pasid;
+};
+
+/**
+ * amdgpu_pasid_alloc - Allocate a PASID
+ * @bits: Maximum width of the PASID in bits, must be at least 1
+ *
+ * Allocates a PASID of the given width while keeping smaller PASIDs
+ * available if possible.
+ *
+ * Returns a positive integer on success. Returns %-EINVAL if bits==0.
+ * Returns %-ENOSPC if no PASID was available. Returns %-ENOMEM on
+ * memory allocation failure.
+ */
+int amdgpu_pasid_alloc(unsigned int bits)
+{
+	int pasid = -EINVAL;
+
+	for (bits = min(bits, 31U); bits > 0; bits--) {
+		pasid = ida_alloc_range(&amdgpu_pasid_ida, 1U << (bits - 1),
+					(1U << bits) - 1, GFP_KERNEL);
+		if (pasid != -ENOSPC)
+			break;
+	}
+
+	if (pasid >= 0)
+		trace_amdgpu_pasid_allocated(pasid);
+
+	return pasid;
+}
+
+/**
+ * amdgpu_pasid_free - Free a PASID
+ * @pasid: PASID to free
+ */
+void amdgpu_pasid_free(u32 pasid)
+{
+	trace_amdgpu_pasid_freed(pasid);
+	ida_free(&amdgpu_pasid_ida, pasid);
+}
+
+static void amdgpu_pasid_free_cb(struct dma_fence *fence,
+				 struct dma_fence_cb *_cb)
+{
+	struct amdgpu_pasid_cb *cb =
+		container_of(_cb, struct amdgpu_pasid_cb, cb);
+
+	amdgpu_pasid_free(cb->pasid);
+	dma_fence_put(fence);
+	kfree(cb);
+}
+
+/**
+ * amdgpu_pasid_free_delayed - free pasid when fences signal
+ *
+ * @resv: reservation object with the fences to wait for
+ * @pasid: pasid to free
+ *
+ * Free the pasid only after all the fences in resv are signaled.
+ */
+void amdgpu_pasid_free_delayed(struct dma_resv *resv,
+			       u32 pasid)
+{
+	struct amdgpu_pasid_cb *cb;
+	struct dma_fence *fence;
+	int r;
+
+	r = dma_resv_get_singleton(resv, DMA_RESV_USAGE_BOOKKEEP, &fence);
+	if (r)
+		goto fallback;
+
+	if (!fence) {
+		amdgpu_pasid_free(pasid);
+		return;
+	}
+
+	cb = kmalloc(sizeof(*cb), GFP_KERNEL);
+	if (!cb) {
+		/* Last resort when we are OOM */
+		dma_fence_wait(fence, false);
+		dma_fence_put(fence);
+		amdgpu_pasid_free(pasid);
+	} else {
+		cb->pasid = pasid;
+		if (dma_fence_add_callback(fence, &cb->cb,
+					   amdgpu_pasid_free_cb))
+			amdgpu_pasid_free_cb(fence, &cb->cb);
+	}
+
+	return;
+
+fallback:
+	/* Not enough memory for the delayed delete, as last resort
+	 * block for all the fences to complete.
+	 */
+	dma_resv_wait_timeout(resv, DMA_RESV_USAGE_BOOKKEEP,
+			      false, MAX_SCHEDULE_TIMEOUT);
+	amdgpu_pasid_free(pasid);
+}
+
+/*
+ * VMID manager
+ *
+ * VMIDs are a per VMHUB identifier for page tables handling.
+ */
+
+/**
+ * amdgpu_vmid_had_gpu_reset - check if reset occured since last use
+ *
+ * @adev: amdgpu_device pointer
+ * @id: VMID structure
+ *
+ * Check if GPU reset occured since last use of the VMID.
+ */
+bool amdgpu_vmid_had_gpu_reset(struct amdgpu_device *adev,
+			       struct amdgpu_vmid *id)
+{
+	return id->current_gpu_reset_count !=
+		atomic_read(&adev->gpu_reset_counter);
+}
+
+/* Check if we need to switch to another set of resources */
+static bool amdgpu_vmid_gds_switch_needed(struct amdgpu_vmid *id,
+					  struct amdgpu_job *job)
+{
+	return id->gds_base != job->gds_base ||
+		id->gds_size != job->gds_size ||
+		id->gws_base != job->gws_base ||
+		id->gws_size != job->gws_size ||
+		id->oa_base != job->oa_base ||
+		id->oa_size != job->oa_size;
+}
+
+/* Check if the id is compatible with the job */
+static bool amdgpu_vmid_compatible(struct amdgpu_vmid *id,
+				   struct amdgpu_job *job)
+{
+	return  id->pd_gpu_addr == job->vm_pd_addr &&
+		!amdgpu_vmid_gds_switch_needed(id, job);
+}
+
+/**
+ * amdgpu_vmid_grab_idle - grab idle VMID
+ *
+ * @ring: ring we want to submit job to
+ * @idle: resulting idle VMID
+ * @fence: fence to wait for if no id could be grabbed
+ *
+ * Try to find an idle VMID, if none is idle add a fence to wait to the sync
+ * object. Returns -ENOMEM when we are out of memory.
+ */
+static int amdgpu_vmid_grab_idle(struct amdgpu_ring *ring,
+				 struct amdgpu_vmid **idle,
+				 struct dma_fence **fence)
+{
+	struct amdgpu_device *adev = ring->adev;
+	unsigned vmhub = ring->vm_hub;
+	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
+
+	/* If anybody is waiting for a VMID let everybody wait for fairness */
+	if (!dma_fence_is_signaled(ring->vmid_wait)) {
+		*fence = dma_fence_get(ring->vmid_wait);
+		return 0;
+	}
+
+	/* Check if we have an idle VMID */
+	list_for_each_entry_reverse((*idle), &id_mgr->ids_lru, list) {
+		/* Don't use per engine and per process VMID at the same time */
+		struct amdgpu_ring *r = adev->vm_manager.concurrent_flush ?
+			NULL : ring;
+
+		*fence = amdgpu_sync_peek_fence(&(*idle)->active, r);
+		if (!(*fence))
+			return 0;
+	}
+
+	/*
+	 * If we can't find a idle VMID to use, wait on a fence from the least
+	 * recently used in the hope that it will be available soon.
+	 */
+	*idle = NULL;
+	dma_fence_put(ring->vmid_wait);
+	ring->vmid_wait = dma_fence_get(*fence);
+
+	/* This is the reference we return */
+	dma_fence_get(*fence);
+	return 0;
+}
+
+/**
+ * amdgpu_vmid_grab_reserved - try to assign reserved VMID
+ *
+ * @vm: vm to allocate id for
+ * @ring: ring we want to submit job to
+ * @job: job who wants to use the VMID
+ * @id: resulting VMID
+ * @fence: fence to wait for if no id could be grabbed
+ *
+ * Try to assign a reserved VMID.
+ */
+static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm,
+				     struct amdgpu_ring *ring,
+				     struct amdgpu_job *job,
+				     struct amdgpu_vmid **id,
+				     struct dma_fence **fence)
+{
+	struct amdgpu_device *adev = ring->adev;
+	unsigned vmhub = ring->vm_hub;
+	uint64_t fence_context = adev->fence_context + ring->idx;
+	bool needs_flush = vm->use_cpu_for_update;
+	uint64_t updates = amdgpu_vm_tlb_seq(vm);
+	int r;
+
+	*id = vm->reserved_vmid[vmhub];
+	if ((*id)->owner != vm->immediate.fence_context ||
+	    !amdgpu_vmid_compatible(*id, job) ||
+	    (*id)->flushed_updates < updates ||
+	    !(*id)->last_flush ||
+	    ((*id)->last_flush->context != fence_context &&
+	     !dma_fence_is_signaled((*id)->last_flush)))
+		needs_flush = true;
+
+	if ((*id)->owner != vm->immediate.fence_context ||
+	    (!adev->vm_manager.concurrent_flush && needs_flush)) {
+		struct dma_fence *tmp;
+
+		/* Don't use per engine and per process VMID at the
+		 * same time
+		 */
+		if (adev->vm_manager.concurrent_flush)
+			ring = NULL;
+
+		/* to prevent one context starved by another context */
+		(*id)->pd_gpu_addr = 0;
+		tmp = amdgpu_sync_peek_fence(&(*id)->active, ring);
+		if (tmp) {
+			*id = NULL;
+			*fence = dma_fence_get(tmp);
+			return 0;
+		}
+	}
+
+	/* Good we can use this VMID. Remember this submission as
+	* user of the VMID.
+	*/
+	r = amdgpu_sync_fence(&(*id)->active, &job->base.s_fence->finished,
+			      GFP_ATOMIC);
+	if (r)
+		return r;
+
+	job->vm_needs_flush = needs_flush;
+	job->spm_update_needed = true;
+	return 0;
+}
+
+/**
+ * amdgpu_vmid_grab_used - try to reuse a VMID
+ *
+ * @vm: vm to allocate id for
+ * @ring: ring we want to submit job to
+ * @job: job who wants to use the VMID
+ * @id: resulting VMID
+ *
+ * Try to reuse a VMID for this submission.
+ */
+static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm,
+				 struct amdgpu_ring *ring,
+				 struct amdgpu_job *job,
+				 struct amdgpu_vmid **id)
+{
+	struct amdgpu_device *adev = ring->adev;
+	unsigned vmhub = ring->vm_hub;
+	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
+	uint64_t fence_context = adev->fence_context + ring->idx;
+	uint64_t updates = amdgpu_vm_tlb_seq(vm);
+	int r;
+
+	job->vm_needs_flush = vm->use_cpu_for_update;
+
+	/* Check if we can use a VMID already assigned to this VM */
+	list_for_each_entry_reverse((*id), &id_mgr->ids_lru, list) {
+		bool needs_flush = vm->use_cpu_for_update;
+
+		/* Check all the prerequisites to using this VMID */
+		if ((*id)->owner != vm->immediate.fence_context)
+			continue;
+
+		if (!amdgpu_vmid_compatible(*id, job))
+			continue;
+
+		if (!(*id)->last_flush ||
+		    ((*id)->last_flush->context != fence_context &&
+		     !dma_fence_is_signaled((*id)->last_flush)))
+			needs_flush = true;
+
+		if ((*id)->flushed_updates < updates)
+			needs_flush = true;
+
+		if (needs_flush && !adev->vm_manager.concurrent_flush)
+			continue;
+
+		/* Good, we can use this VMID. Remember this submission as
+		 * user of the VMID.
+		 */
+		r = amdgpu_sync_fence(&(*id)->active,
+				      &job->base.s_fence->finished,
+				      GFP_ATOMIC);
+		if (r)
+			return r;
+
+		job->vm_needs_flush |= needs_flush;
+		return 0;
+	}
+
+	*id = NULL;
+	return 0;
+}
+
+/**
+ * amdgpu_vmid_grab - allocate the next free VMID
+ *
+ * @vm: vm to allocate id for
+ * @ring: ring we want to submit job to
+ * @job: job who wants to use the VMID
+ * @fence: fence to wait for if no id could be grabbed
+ *
+ * Allocate an id for the vm, adding fences to the sync obj as necessary.
+ */
+int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
+		     struct amdgpu_job *job, struct dma_fence **fence)
+{
+	struct amdgpu_device *adev = ring->adev;
+	unsigned vmhub = ring->vm_hub;
+	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
+	struct amdgpu_vmid *idle = NULL;
+	struct amdgpu_vmid *id = NULL;
+	int r = 0;
+
+	mutex_lock(&id_mgr->lock);
+	r = amdgpu_vmid_grab_idle(ring, &idle, fence);
+	if (r || !idle)
+		goto error;
+
+	if (amdgpu_vmid_uses_reserved(vm, vmhub)) {
+		r = amdgpu_vmid_grab_reserved(vm, ring, job, &id, fence);
+		if (r || !id)
+			goto error;
+	} else {
+		r = amdgpu_vmid_grab_used(vm, ring, job, &id);
+		if (r)
+			goto error;
+
+		if (!id) {
+			/* Still no ID to use? Then use the idle one found earlier */
+			id = idle;
+
+			/* Remember this submission as user of the VMID */
+			r = amdgpu_sync_fence(&id->active,
+					      &job->base.s_fence->finished,
+					      GFP_ATOMIC);
+			if (r)
+				goto error;
+
+			job->vm_needs_flush = true;
+		}
+
+		list_move_tail(&id->list, &id_mgr->ids_lru);
+	}
+
+	job->gds_switch_needed = amdgpu_vmid_gds_switch_needed(id, job);
+	if (job->vm_needs_flush) {
+		id->flushed_updates = amdgpu_vm_tlb_seq(vm);
+		dma_fence_put(id->last_flush);
+		id->last_flush = NULL;
+	}
+	job->vmid = id - id_mgr->ids;
+	job->pasid = vm->pasid;
+
+	id->gds_base = job->gds_base;
+	id->gds_size = job->gds_size;
+	id->gws_base = job->gws_base;
+	id->gws_size = job->gws_size;
+	id->oa_base = job->oa_base;
+	id->oa_size = job->oa_size;
+	id->pd_gpu_addr = job->vm_pd_addr;
+	id->owner = vm->immediate.fence_context;
+
+	trace_amdgpu_vm_grab_id(vm, ring, job);
+
+error:
+	mutex_unlock(&id_mgr->lock);
+	return r;
+}
+
+/*
+ * amdgpu_vmid_uses_reserved - check if a VM will use a reserved VMID
+ * @vm: the VM to check
+ * @vmhub: the VMHUB which will be used
+ *
+ * Returns: True if the VM will use a reserved VMID.
+ */
+bool amdgpu_vmid_uses_reserved(struct amdgpu_vm *vm, unsigned int vmhub)
+{
+	return vm->reserved_vmid[vmhub];
+}
+
+/*
+ * amdgpu_vmid_alloc_reserved - reserve a specific VMID for this vm
+ * @adev: amdgpu device structure
+ * @vm: the VM to reserve an ID for
+ * @vmhub: the VMHUB which should be used
+ *
+ * Mostly used to have a reserved VMID for debugging and SPM.
+ *
+ * Returns: 0 for success, -ENOENT if an ID is already reserved.
+ */
+int amdgpu_vmid_alloc_reserved(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			       unsigned vmhub)
+{
+	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
+	struct amdgpu_vmid *id;
+	int r = 0;
+
+	mutex_lock(&id_mgr->lock);
+	if (vm->reserved_vmid[vmhub])
+		goto unlock;
+	if (id_mgr->reserved_vmid) {
+		r = -ENOENT;
+		goto unlock;
+	}
+	/* Remove from normal round robin handling */
+	id = list_first_entry(&id_mgr->ids_lru, struct amdgpu_vmid, list);
+	list_del_init(&id->list);
+	vm->reserved_vmid[vmhub] = id;
+	id_mgr->reserved_vmid = true;
+	mutex_unlock(&id_mgr->lock);
+
+	return 0;
+unlock:
+	mutex_unlock(&id_mgr->lock);
+	return r;
+}
+
+/*
+ * amdgpu_vmid_free_reserved - free up a reserved VMID again
+ * @adev: amdgpu device structure
+ * @vm: the VM with the reserved ID
+ * @vmhub: the VMHUB which should be used
+ */
+void amdgpu_vmid_free_reserved(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			       unsigned vmhub)
+{
+	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
+
+	mutex_lock(&id_mgr->lock);
+	if (vm->reserved_vmid[vmhub]) {
+		list_add(&vm->reserved_vmid[vmhub]->list,
+			&id_mgr->ids_lru);
+		vm->reserved_vmid[vmhub] = NULL;
+		id_mgr->reserved_vmid = false;
+	}
+	mutex_unlock(&id_mgr->lock);
+}
+
+/**
+ * amdgpu_vmid_reset - reset VMID to zero
+ *
+ * @adev: amdgpu device structure
+ * @vmhub: vmhub type
+ * @vmid: vmid number to use
+ *
+ * Reset saved GDW, GWS and OA to force switch on next flush.
+ */
+void amdgpu_vmid_reset(struct amdgpu_device *adev, unsigned vmhub,
+		       unsigned vmid)
+{
+	struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
+	struct amdgpu_vmid *id = &id_mgr->ids[vmid];
+
+	mutex_lock(&id_mgr->lock);
+	id->owner = 0;
+	id->gds_base = 0;
+	id->gds_size = 0;
+	id->gws_base = 0;
+	id->gws_size = 0;
+	id->oa_base = 0;
+	id->oa_size = 0;
+	mutex_unlock(&id_mgr->lock);
+}
+
+/**
+ * amdgpu_vmid_reset_all - reset VMID to zero
+ *
+ * @adev: amdgpu device structure
+ *
+ * Reset VMID to force flush on next use
+ */
+void amdgpu_vmid_reset_all(struct amdgpu_device *adev)
+{
+	unsigned i, j;
+
+	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
+		struct amdgpu_vmid_mgr *id_mgr =
+			&adev->vm_manager.id_mgr[i];
+
+		for (j = 1; j < id_mgr->num_ids; ++j)
+			amdgpu_vmid_reset(adev, i, j);
+	}
+}
+
+/**
+ * amdgpu_vmid_mgr_init - init the VMID manager
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Initialize the VM manager structures
+ */
+void amdgpu_vmid_mgr_init(struct amdgpu_device *adev)
+{
+	unsigned i, j;
+
+	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
+		struct amdgpu_vmid_mgr *id_mgr =
+			&adev->vm_manager.id_mgr[i];
+
+		mutex_init(&id_mgr->lock);
+		INIT_LIST_HEAD(&id_mgr->ids_lru);
+
+		/* for GC <10, SDMA uses MMHUB so use first_kfd_vmid for both GC and MM */
+		if (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(10, 0, 0))
+			/* manage only VMIDs not used by KFD */
+			id_mgr->num_ids = adev->vm_manager.first_kfd_vmid;
+		else if (AMDGPU_IS_MMHUB0(i) ||
+			 AMDGPU_IS_MMHUB1(i))
+			id_mgr->num_ids = 16;
+		else
+			/* manage only VMIDs not used by KFD */
+			id_mgr->num_ids = adev->vm_manager.first_kfd_vmid;
+
+		/* skip over VMID 0, since it is the system VM */
+		for (j = 1; j < id_mgr->num_ids; ++j) {
+			amdgpu_vmid_reset(adev, i, j);
+			amdgpu_sync_create(&id_mgr->ids[j].active);
+			list_add_tail(&id_mgr->ids[j].list, &id_mgr->ids_lru);
+		}
+	}
+}
+
+/**
+ * amdgpu_vmid_mgr_fini - cleanup VM manager
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Cleanup the VM manager and free resources.
+ */
+void amdgpu_vmid_mgr_fini(struct amdgpu_device *adev)
+{
+	unsigned i, j;
+
+	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
+		struct amdgpu_vmid_mgr *id_mgr =
+			&adev->vm_manager.id_mgr[i];
+
+		mutex_destroy(&id_mgr->lock);
+		for (j = 0; j < AMDGPU_NUM_VMID; ++j) {
+			struct amdgpu_vmid *id = &id_mgr->ids[j];
+
+			amdgpu_sync_free(&id->active);
+			dma_fence_put(id->last_flush);
+			dma_fence_put(id->pasid_mapping);
+		}
+	}
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h
new file mode 100644
index 000000000000..b3649cd3af56
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h
@@ -0,0 +1,94 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_IDS_H__
+#define __AMDGPU_IDS_H__
+
+#include <linux/types.h>
+#include <linux/mutex.h>
+#include <linux/list.h>
+#include <linux/dma-fence.h>
+
+#include "amdgpu_sync.h"
+
+/* maximum number of VMIDs */
+#define AMDGPU_NUM_VMID	16
+
+struct amdgpu_device;
+struct amdgpu_vm;
+struct amdgpu_ring;
+struct amdgpu_sync;
+struct amdgpu_job;
+
+struct amdgpu_vmid {
+	struct list_head	list;
+	struct amdgpu_sync	active;
+	struct dma_fence	*last_flush;
+	uint64_t		owner;
+
+	uint64_t		pd_gpu_addr;
+	/* last flushed PD/PT update */
+	uint64_t		flushed_updates;
+
+	uint32_t                current_gpu_reset_count;
+
+	uint32_t		gds_base;
+	uint32_t		gds_size;
+	uint32_t		gws_base;
+	uint32_t		gws_size;
+	uint32_t		oa_base;
+	uint32_t		oa_size;
+
+	unsigned		pasid;
+	struct dma_fence	*pasid_mapping;
+};
+
+struct amdgpu_vmid_mgr {
+	struct mutex		lock;
+	unsigned		num_ids;
+	struct list_head	ids_lru;
+	struct amdgpu_vmid	ids[AMDGPU_NUM_VMID];
+	bool			reserved_vmid;
+};
+
+int amdgpu_pasid_alloc(unsigned int bits);
+void amdgpu_pasid_free(u32 pasid);
+void amdgpu_pasid_free_delayed(struct dma_resv *resv,
+			       u32 pasid);
+
+bool amdgpu_vmid_had_gpu_reset(struct amdgpu_device *adev,
+			       struct amdgpu_vmid *id);
+bool amdgpu_vmid_uses_reserved(struct amdgpu_vm *vm, unsigned int vmhub);
+int amdgpu_vmid_alloc_reserved(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			       unsigned vmhub);
+void amdgpu_vmid_free_reserved(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+			       unsigned vmhub);
+int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
+		     struct amdgpu_job *job, struct dma_fence **fence);
+void amdgpu_vmid_reset(struct amdgpu_device *adev, unsigned vmhub,
+		       unsigned vmid);
+void amdgpu_vmid_reset_all(struct amdgpu_device *adev);
+
+void amdgpu_vmid_mgr_init(struct amdgpu_device *adev);
+void amdgpu_vmid_mgr_fini(struct amdgpu_device *adev);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
index 3ab4c65ecc8b..a6419246e9c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
@@ -21,49 +21,26 @@
  *
  */
 
-#include <drm/drmP.h>
+#include <linux/dma-mapping.h>
+
 #include "amdgpu.h"
 #include "amdgpu_ih.h"
-#include "amdgpu_amdkfd.h"
-
-/**
- * amdgpu_ih_ring_alloc - allocate memory for the IH ring
- *
- * @adev: amdgpu_device pointer
- *
- * Allocate a ring buffer for the interrupt controller.
- * Returns 0 for success, errors for failure.
- */
-static int amdgpu_ih_ring_alloc(struct amdgpu_device *adev)
-{
-	int r;
-
-	/* Allocate ring buffer */
-	if (adev->irq.ih.ring_obj == NULL) {
-		r = amdgpu_bo_create_kernel(adev, adev->irq.ih.ring_size,
-					    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
-					    &adev->irq.ih.ring_obj,
-					    &adev->irq.ih.gpu_addr,
-					    (void **)&adev->irq.ih.ring);
-		if (r) {
-			DRM_ERROR("amdgpu: failed to create ih ring buffer (%d).\n", r);
-			return r;
-		}
-	}
-	return 0;
-}
+#include "amdgpu_reset.h"
 
 /**
  * amdgpu_ih_ring_init - initialize the IH state
  *
  * @adev: amdgpu_device pointer
+ * @ih: ih ring to initialize
+ * @ring_size: ring size to allocate
+ * @use_bus_addr: true when we can use dma_alloc_coherent
  *
  * Initializes the IH state and allocates a buffer
  * for the IH ring buffer.
  * Returns 0 for success, errors for failure.
  */
-int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size,
-			bool use_bus_addr)
+int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
+			unsigned ring_size, bool use_bus_addr)
 {
 	u32 rb_bufsz;
 	int r;
@@ -71,122 +48,270 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size,
 	/* Align ring size */
 	rb_bufsz = order_base_2(ring_size / 4);
 	ring_size = (1 << rb_bufsz) * 4;
-	adev->irq.ih.ring_size = ring_size;
-	adev->irq.ih.ptr_mask = adev->irq.ih.ring_size - 1;
-	adev->irq.ih.rptr = 0;
-	adev->irq.ih.use_bus_addr = use_bus_addr;
-
-	if (adev->irq.ih.use_bus_addr) {
-		if (!adev->irq.ih.ring) {
-			/* add 8 bytes for the rptr/wptr shadows and
-			 * add them to the end of the ring allocation.
-			 */
-			adev->irq.ih.ring = pci_alloc_consistent(adev->pdev,
-								 adev->irq.ih.ring_size + 8,
-								 &adev->irq.ih.rb_dma_addr);
-			if (adev->irq.ih.ring == NULL)
-				return -ENOMEM;
-			memset((void *)adev->irq.ih.ring, 0, adev->irq.ih.ring_size + 8);
-			adev->irq.ih.wptr_offs = (adev->irq.ih.ring_size / 4) + 0;
-			adev->irq.ih.rptr_offs = (adev->irq.ih.ring_size / 4) + 1;
-		}
-		return 0;
+	ih->ring_size = ring_size;
+	ih->ptr_mask = ih->ring_size - 1;
+	ih->rptr = 0;
+	ih->use_bus_addr = use_bus_addr;
+
+	if (use_bus_addr) {
+		dma_addr_t dma_addr;
+
+		if (ih->ring)
+			return 0;
+
+		/* add 8 bytes for the rptr/wptr shadows and
+		 * add them to the end of the ring allocation.
+		 */
+		ih->ring = dma_alloc_coherent(adev->dev, ih->ring_size + 8,
+					      &dma_addr, GFP_KERNEL);
+		if (ih->ring == NULL)
+			return -ENOMEM;
+
+		ih->gpu_addr = dma_addr;
+		ih->wptr_addr = dma_addr + ih->ring_size;
+		ih->wptr_cpu = &ih->ring[ih->ring_size / 4];
+		ih->rptr_addr = dma_addr + ih->ring_size + 4;
+		ih->rptr_cpu = &ih->ring[(ih->ring_size / 4) + 1];
 	} else {
-		r = amdgpu_wb_get(adev, &adev->irq.ih.wptr_offs);
+		unsigned wptr_offs, rptr_offs;
+
+		r = amdgpu_device_wb_get(adev, &wptr_offs);
+		if (r)
+			return r;
+
+		r = amdgpu_device_wb_get(adev, &rptr_offs);
 		if (r) {
-			dev_err(adev->dev, "(%d) ih wptr_offs wb alloc failed\n", r);
+			amdgpu_device_wb_free(adev, wptr_offs);
 			return r;
 		}
 
-		r = amdgpu_wb_get(adev, &adev->irq.ih.rptr_offs);
+		r = amdgpu_bo_create_kernel(adev, ih->ring_size, PAGE_SIZE,
+					    AMDGPU_GEM_DOMAIN_GTT,
+					    &ih->ring_obj, &ih->gpu_addr,
+					    (void **)&ih->ring);
 		if (r) {
-			amdgpu_wb_free(adev, adev->irq.ih.wptr_offs);
-			dev_err(adev->dev, "(%d) ih rptr_offs wb alloc failed\n", r);
+			amdgpu_device_wb_free(adev, rptr_offs);
+			amdgpu_device_wb_free(adev, wptr_offs);
 			return r;
 		}
 
-		return amdgpu_ih_ring_alloc(adev);
+		ih->wptr_addr = adev->wb.gpu_addr + wptr_offs * 4;
+		ih->wptr_cpu = &adev->wb.wb[wptr_offs];
+		ih->rptr_addr = adev->wb.gpu_addr + rptr_offs * 4;
+		ih->rptr_cpu = &adev->wb.wb[rptr_offs];
 	}
+
+	init_waitqueue_head(&ih->wait_process);
+	return 0;
 }
 
 /**
  * amdgpu_ih_ring_fini - tear down the IH state
  *
  * @adev: amdgpu_device pointer
+ * @ih: ih ring to tear down
  *
  * Tears down the IH state and frees buffer
  * used for the IH ring buffer.
  */
-void amdgpu_ih_ring_fini(struct amdgpu_device *adev)
+void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih)
 {
-	if (adev->irq.ih.use_bus_addr) {
-		if (adev->irq.ih.ring) {
-			/* add 8 bytes for the rptr/wptr shadows and
-			 * add them to the end of the ring allocation.
-			 */
-			pci_free_consistent(adev->pdev, adev->irq.ih.ring_size + 8,
-					    (void *)adev->irq.ih.ring,
-					    adev->irq.ih.rb_dma_addr);
-			adev->irq.ih.ring = NULL;
-		}
+
+	if (!ih->ring)
+		return;
+
+	if (ih->use_bus_addr) {
+
+		/* add 8 bytes for the rptr/wptr shadows and
+		 * add them to the end of the ring allocation.
+		 */
+		dma_free_coherent(adev->dev, ih->ring_size + 8,
+				  (void *)ih->ring, ih->gpu_addr);
+		ih->ring = NULL;
 	} else {
-		amdgpu_bo_free_kernel(&adev->irq.ih.ring_obj,
-				      &adev->irq.ih.gpu_addr,
-				      (void **)&adev->irq.ih.ring);
-		amdgpu_wb_free(adev, adev->irq.ih.wptr_offs);
-		amdgpu_wb_free(adev, adev->irq.ih.rptr_offs);
+		amdgpu_bo_free_kernel(&ih->ring_obj, &ih->gpu_addr,
+				      (void **)&ih->ring);
+		amdgpu_device_wb_free(adev, (ih->wptr_addr - ih->gpu_addr) / 4);
+		amdgpu_device_wb_free(adev, (ih->rptr_addr - ih->gpu_addr) / 4);
+	}
+}
+
+/**
+ * amdgpu_ih_ring_write - write IV to the ring buffer
+ *
+ * @adev: amdgpu_device pointer
+ * @ih: ih ring to write to
+ * @iv: the iv to write
+ * @num_dw: size of the iv in dw
+ *
+ * Writes an IV to the ring buffer using the CPU and increment the wptr.
+ * Used for testing and delegating IVs to a software ring.
+ */
+void amdgpu_ih_ring_write(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
+			  const uint32_t *iv, unsigned int num_dw)
+{
+	uint32_t wptr = le32_to_cpu(*ih->wptr_cpu) >> 2;
+	unsigned int i;
+
+	for (i = 0; i < num_dw; ++i)
+	        ih->ring[wptr++] = cpu_to_le32(iv[i]);
+
+	wptr <<= 2;
+	wptr &= ih->ptr_mask;
+
+	/* Only commit the new wptr if we don't overflow */
+	if (wptr != READ_ONCE(ih->rptr)) {
+		wmb();
+		WRITE_ONCE(*ih->wptr_cpu, cpu_to_le32(wptr));
+	} else if (adev->irq.retry_cam_enabled) {
+		dev_warn_once(adev->dev, "IH soft ring buffer overflow 0x%X, 0x%X\n",
+			      wptr, ih->rptr);
 	}
 }
 
 /**
+ * amdgpu_ih_wait_on_checkpoint_process_ts - wait to process IVs up to checkpoint
+ *
+ * @adev: amdgpu_device pointer
+ * @ih: ih ring to process
+ *
+ * Used to ensure ring has processed IVs up to the checkpoint write pointer.
+ */
+int amdgpu_ih_wait_on_checkpoint_process_ts(struct amdgpu_device *adev,
+					struct amdgpu_ih_ring *ih)
+{
+	uint32_t checkpoint_wptr;
+	uint64_t checkpoint_ts;
+	long timeout = HZ;
+
+	if (!ih->enabled || adev->shutdown)
+		return -ENODEV;
+
+	checkpoint_wptr = amdgpu_ih_get_wptr(adev, ih);
+	/* Order wptr with ring data. */
+	rmb();
+	checkpoint_ts = amdgpu_ih_decode_iv_ts(adev, ih, checkpoint_wptr, -1);
+
+	return wait_event_interruptible_timeout(ih->wait_process,
+		    amdgpu_ih_ts_after(checkpoint_ts, ih->processed_timestamp) ||
+		    ih->rptr == amdgpu_ih_get_wptr(adev, ih), timeout);
+}
+
+/**
  * amdgpu_ih_process - interrupt handler
  *
  * @adev: amdgpu_device pointer
+ * @ih: ih ring to process
  *
  * Interrupt hander (VI), walk the IH ring.
  * Returns irq process return code.
  */
-int amdgpu_ih_process(struct amdgpu_device *adev)
+int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih)
 {
-	struct amdgpu_iv_entry entry;
+	unsigned int count;
 	u32 wptr;
 
-	if (!adev->irq.ih.enabled || adev->shutdown)
+	if (!ih->enabled || adev->shutdown)
 		return IRQ_NONE;
 
-	wptr = amdgpu_ih_get_wptr(adev);
+	wptr = amdgpu_ih_get_wptr(adev, ih);
 
 restart_ih:
-	/* is somebody else already processing irqs? */
-	if (atomic_xchg(&adev->irq.ih.lock, 1))
-		return IRQ_NONE;
-
-	DRM_DEBUG("%s: rptr %d, wptr %d\n", __func__, adev->irq.ih.rptr, wptr);
+	count  = AMDGPU_IH_MAX_NUM_IVS;
+	dev_dbg(adev->dev, "%s: rptr %d, wptr %d\n", __func__, ih->rptr, wptr);
 
 	/* Order reading of wptr vs. reading of IH ring data */
 	rmb();
 
-	while (adev->irq.ih.rptr != wptr) {
-		u32 ring_index = adev->irq.ih.rptr >> 2;
-
-		/* Before dispatching irq to IP blocks, send it to amdkfd */
-		amdgpu_amdkfd_interrupt(adev,
-				(const void *) &adev->irq.ih.ring[ring_index]);
+	while (ih->rptr != wptr && --count) {
+		amdgpu_irq_dispatch(adev, ih);
+		ih->rptr &= ih->ptr_mask;
+	}
 
-		entry.iv_entry = (const uint32_t *)
-			&adev->irq.ih.ring[ring_index];
-		amdgpu_ih_decode_iv(adev, &entry);
-		adev->irq.ih.rptr &= adev->irq.ih.ptr_mask;
+	if (!ih->overflow)
+		amdgpu_ih_set_rptr(adev, ih);
 
-		amdgpu_irq_dispatch(adev, &entry);
-	}
-	amdgpu_ih_set_rptr(adev);
-	atomic_set(&adev->irq.ih.lock, 0);
+	wake_up_all(&ih->wait_process);
 
 	/* make sure wptr hasn't changed while processing */
-	wptr = amdgpu_ih_get_wptr(adev);
-	if (wptr != adev->irq.ih.rptr)
-		goto restart_ih;
+	wptr = amdgpu_ih_get_wptr(adev, ih);
+	if (wptr != ih->rptr)
+		if (!ih->overflow)
+			goto restart_ih;
+
+	if (ih->overflow)
+		if (amdgpu_sriov_runtime(adev))
+			WARN_ONCE(!amdgpu_reset_domain_schedule(adev->reset_domain,
+				   &adev->virt.flr_work),
+				  "Failed to queue work! at %s",
+				  __func__);
 
 	return IRQ_HANDLED;
 }
+
+/**
+ * amdgpu_ih_decode_iv_helper - decode an interrupt vector
+ *
+ * @adev: amdgpu_device pointer
+ * @ih: ih ring to process
+ * @entry: IV entry
+ *
+ * Decodes the interrupt vector at the current rptr
+ * position and also advance the position for Vega10
+ * and later GPUs.
+ */
+void amdgpu_ih_decode_iv_helper(struct amdgpu_device *adev,
+				struct amdgpu_ih_ring *ih,
+				struct amdgpu_iv_entry *entry)
+{
+	/* wptr/rptr are in bytes! */
+	u32 ring_index = ih->rptr >> 2;
+	uint32_t dw[8];
+
+	dw[0] = le32_to_cpu(ih->ring[ring_index + 0]);
+	dw[1] = le32_to_cpu(ih->ring[ring_index + 1]);
+	dw[2] = le32_to_cpu(ih->ring[ring_index + 2]);
+	dw[3] = le32_to_cpu(ih->ring[ring_index + 3]);
+	dw[4] = le32_to_cpu(ih->ring[ring_index + 4]);
+	dw[5] = le32_to_cpu(ih->ring[ring_index + 5]);
+	dw[6] = le32_to_cpu(ih->ring[ring_index + 6]);
+	dw[7] = le32_to_cpu(ih->ring[ring_index + 7]);
+
+	entry->client_id = dw[0] & 0xff;
+	entry->src_id = (dw[0] >> 8) & 0xff;
+	entry->ring_id = (dw[0] >> 16) & 0xff;
+	entry->vmid = (dw[0] >> 24) & 0xf;
+	entry->vmid_src = (dw[0] >> 31);
+	entry->timestamp = dw[1] | ((u64)(dw[2] & 0xffff) << 32);
+	entry->timestamp_src = dw[2] >> 31;
+	entry->pasid = dw[3] & 0xffff;
+	entry->node_id = (dw[3] >> 16) & 0xff;
+	entry->src_data[0] = dw[4];
+	entry->src_data[1] = dw[5];
+	entry->src_data[2] = dw[6];
+	entry->src_data[3] = dw[7];
+
+	/* wptr/rptr are in bytes! */
+	ih->rptr += 32;
+}
+
+uint64_t amdgpu_ih_decode_iv_ts_helper(struct amdgpu_ih_ring *ih, u32 rptr,
+				       signed int offset)
+{
+	uint32_t iv_size = 32;
+	uint32_t ring_index;
+	uint32_t dw1, dw2;
+
+	rptr += iv_size * offset;
+	ring_index = (rptr & ih->ptr_mask) >> 2;
+
+	dw1 = le32_to_cpu(ih->ring[ring_index + 1]);
+	dw2 = le32_to_cpu(ih->ring[ring_index + 2]);
+	return dw1 | ((u64)(dw2 & 0xffff) << 32);
+}
+
+const char *amdgpu_ih_ring_name(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih)
+{
+	return ih == &adev->irq.ih ? "ih" : ih == &adev->irq.ih_soft ? "sw ih" :
+	       ih == &adev->irq.ih1 ? "ih1" : ih == &adev->irq.ih2 ? "ih2" : "unknown";
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
index a3da1a122fc8..f58b6be7fccc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
@@ -24,89 +24,95 @@
 #ifndef __AMDGPU_IH_H__
 #define __AMDGPU_IH_H__
 
+/* Maximum number of IVs processed at once */
+#define AMDGPU_IH_MAX_NUM_IVS	32
+
+#define IH_RING_SIZE	(256 * 1024)
+#define IH_SW_RING_SIZE	(16 * 1024)	/* enough for 512 CAM entries */
+
 struct amdgpu_device;
- /*
-  * vega10+ IH clients
- */
-enum amdgpu_ih_clientid
-{
-    AMDGPU_IH_CLIENTID_IH	    = 0x00,
-    AMDGPU_IH_CLIENTID_ACP	    = 0x01,
-    AMDGPU_IH_CLIENTID_ATHUB	    = 0x02,
-    AMDGPU_IH_CLIENTID_BIF	    = 0x03,
-    AMDGPU_IH_CLIENTID_DCE	    = 0x04,
-    AMDGPU_IH_CLIENTID_ISP	    = 0x05,
-    AMDGPU_IH_CLIENTID_PCIE0	    = 0x06,
-    AMDGPU_IH_CLIENTID_RLC	    = 0x07,
-    AMDGPU_IH_CLIENTID_SDMA0	    = 0x08,
-    AMDGPU_IH_CLIENTID_SDMA1	    = 0x09,
-    AMDGPU_IH_CLIENTID_SE0SH	    = 0x0a,
-    AMDGPU_IH_CLIENTID_SE1SH	    = 0x0b,
-    AMDGPU_IH_CLIENTID_SE2SH	    = 0x0c,
-    AMDGPU_IH_CLIENTID_SE3SH	    = 0x0d,
-    AMDGPU_IH_CLIENTID_SYSHUB	    = 0x0e,
-    AMDGPU_IH_CLIENTID_THM	    = 0x0f,
-    AMDGPU_IH_CLIENTID_UVD	    = 0x10,
-    AMDGPU_IH_CLIENTID_VCE0	    = 0x11,
-    AMDGPU_IH_CLIENTID_VMC	    = 0x12,
-    AMDGPU_IH_CLIENTID_XDMA	    = 0x13,
-    AMDGPU_IH_CLIENTID_GRBM_CP	    = 0x14,
-    AMDGPU_IH_CLIENTID_ATS	    = 0x15,
-    AMDGPU_IH_CLIENTID_ROM_SMUIO    = 0x16,
-    AMDGPU_IH_CLIENTID_DF	    = 0x17,
-    AMDGPU_IH_CLIENTID_VCE1	    = 0x18,
-    AMDGPU_IH_CLIENTID_PWR	    = 0x19,
-    AMDGPU_IH_CLIENTID_UTCL2	    = 0x1b,
-    AMDGPU_IH_CLIENTID_EA	    = 0x1c,
-    AMDGPU_IH_CLIENTID_UTCL2LOG	    = 0x1d,
-    AMDGPU_IH_CLIENTID_MP0	    = 0x1e,
-    AMDGPU_IH_CLIENTID_MP1	    = 0x1f,
-
-    AMDGPU_IH_CLIENTID_MAX
+struct amdgpu_iv_entry;
 
+struct amdgpu_ih_regs {
+	uint32_t ih_rb_base;
+	uint32_t ih_rb_base_hi;
+	uint32_t ih_rb_cntl;
+	uint32_t ih_rb_wptr;
+	uint32_t ih_rb_rptr;
+	uint32_t ih_doorbell_rptr;
+	uint32_t ih_rb_wptr_addr_lo;
+	uint32_t ih_rb_wptr_addr_hi;
+	uint32_t psp_reg_id;
 };
 
-#define AMDGPU_IH_CLIENTID_LEGACY 0
-
 /*
  * R6xx+ IH ring
  */
 struct amdgpu_ih_ring {
-	struct amdgpu_bo	*ring_obj;
-	volatile uint32_t	*ring;
-	unsigned		rptr;
 	unsigned		ring_size;
-	uint64_t		gpu_addr;
 	uint32_t		ptr_mask;
-	atomic_t		lock;
-	bool                    enabled;
-	unsigned		wptr_offs;
-	unsigned		rptr_offs;
 	u32			doorbell_index;
 	bool			use_doorbell;
 	bool			use_bus_addr;
-	dma_addr_t		rb_dma_addr; /* only used when use_bus_addr = true */
+
+	struct amdgpu_bo	*ring_obj;
+	uint32_t		*ring;
+	uint64_t		gpu_addr;
+
+	uint64_t		wptr_addr;
+	uint32_t		*wptr_cpu;
+
+	uint64_t		rptr_addr;
+	uint32_t		*rptr_cpu;
+
+	bool                    enabled;
+	unsigned		rptr;
+	struct amdgpu_ih_regs	ih_regs;
+
+	/* For waiting on IH processing at checkpoint. */
+	wait_queue_head_t wait_process;
+	uint64_t		processed_timestamp;
+	bool overflow;
 };
 
-#define AMDGPU_IH_SRC_DATA_MAX_SIZE_DW 4
-
-struct amdgpu_iv_entry {
-	unsigned client_id;
-	unsigned src_id;
-	unsigned ring_id;
-	unsigned vm_id;
-	unsigned vm_id_src;
-	uint64_t timestamp;
-	unsigned timestamp_src;
-	unsigned pas_id;
-	unsigned pasid_src;
-	unsigned src_data[AMDGPU_IH_SRC_DATA_MAX_SIZE_DW];
-	const uint32_t *iv_entry;
+/* return true if time stamp t2 is after t1 with 48bit wrap around */
+#define amdgpu_ih_ts_after(t1, t2) \
+		(((int64_t)((t2) << 16) - (int64_t)((t1) << 16)) > 0LL)
+
+#define amdgpu_ih_ts_after_or_equal(t1, t2) \
+		(((int64_t)((t2) << 16) - (int64_t)((t1) << 16)) >= 0LL)
+
+/* provided by the ih block */
+struct amdgpu_ih_funcs {
+	/* ring read/write ptr handling, called from interrupt context */
+	u32 (*get_wptr)(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih);
+	void (*decode_iv)(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
+			  struct amdgpu_iv_entry *entry);
+	uint64_t (*decode_iv_ts)(struct amdgpu_ih_ring *ih, u32 rptr,
+				 signed int offset);
+	void (*set_rptr)(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih);
 };
 
-int amdgpu_ih_ring_init(struct amdgpu_device *adev, unsigned ring_size,
-			bool use_bus_addr);
-void amdgpu_ih_ring_fini(struct amdgpu_device *adev);
-int amdgpu_ih_process(struct amdgpu_device *adev);
+#define amdgpu_ih_get_wptr(adev, ih) (adev)->irq.ih_funcs->get_wptr((adev), (ih))
+#define amdgpu_ih_decode_iv(adev, iv) \
+	(adev)->irq.ih_funcs->decode_iv((adev), (ih), (iv))
+#define amdgpu_ih_decode_iv_ts(adev, ih, rptr, offset) \
+	(WARN_ON_ONCE(!(adev)->irq.ih_funcs->decode_iv_ts) ? 0 : \
+	(adev)->irq.ih_funcs->decode_iv_ts((ih), (rptr), (offset)))
+#define amdgpu_ih_set_rptr(adev, ih) (adev)->irq.ih_funcs->set_rptr((adev), (ih))
 
+int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
+			unsigned ring_size, bool use_bus_addr);
+void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih);
+void amdgpu_ih_ring_write(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
+			  const uint32_t *iv, unsigned int num_dw);
+int amdgpu_ih_wait_on_checkpoint_process_ts(struct amdgpu_device *adev,
+					    struct amdgpu_ih_ring *ih);
+int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih);
+void amdgpu_ih_decode_iv_helper(struct amdgpu_device *adev,
+				struct amdgpu_ih_ring *ih,
+				struct amdgpu_iv_entry *entry);
+uint64_t amdgpu_ih_decode_iv_ts_helper(struct amdgpu_ih_ring *ih, u32 rptr,
+				       signed int offset);
+const char *amdgpu_ih_ring_name(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih);
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_imu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_imu.h
new file mode 100644
index 000000000000..484e936812e4
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_imu.h
@@ -0,0 +1,58 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_IMU_H__
+#define __AMDGPU_IMU_H__
+
+enum imu_work_mode {
+	DEBUG_MODE,
+	MISSION_MODE
+};
+
+struct amdgpu_imu_funcs {
+    int (*init_microcode)(struct amdgpu_device *adev);
+    int (*load_microcode)(struct amdgpu_device *adev);
+    void (*setup_imu)(struct amdgpu_device *adev);
+    int (*start_imu)(struct amdgpu_device *adev);
+    void (*program_rlc_ram)(struct amdgpu_device *adev);
+    int (*wait_for_reset_status)(struct amdgpu_device *adev);
+};
+
+struct imu_rlc_ram_golden {
+    u32 hwip;
+    u32 instance;
+    u32 segment;
+    u32 reg;
+    u32 data;
+    u32 addr_mask;
+};
+
+#define IMU_RLC_RAM_GOLDEN_VALUE(ip, inst, reg, data, addr_mask) \
+    { ip##_HWIP, inst, reg##_BASE_IDX, reg, data, addr_mask }
+
+struct amdgpu_imu {
+    const struct amdgpu_imu_funcs *funcs;
+    enum imu_work_mode mode;
+};
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ioc32.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ioc32.c
index 26482914dc4b..a1cbd7c3deb2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ioc32.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ioc32.c
@@ -1,4 +1,4 @@
-/**
+/*
  * \file amdgpu_ioc32.c
  *
  * 32-bit ioctl compatibility routines for the AMDGPU DRM.
@@ -29,19 +29,17 @@
  */
 #include <linux/compat.h>
 
-#include <drm/drmP.h>
 #include <drm/amdgpu_drm.h>
+#include <drm/drm_ioctl.h>
+
 #include "amdgpu_drv.h"
 
 long amdgpu_kms_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 {
 	unsigned int nr = DRM_IOCTL_NR(cmd);
-	int ret;
 
 	if (nr < DRM_COMMAND_BASE)
 		return drm_compat_ioctl(filp, cmd, arg);
 
-	ret = amdgpu_drm_ioctl(filp, cmd, arg);
-
-	return ret;
+	return amdgpu_drm_ioctl(filp, cmd, arg);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ip.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ip.c
new file mode 100644
index 000000000000..99e1cf4fc955
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ip.c
@@ -0,0 +1,96 @@
+/*
+ * Copyright 2025 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "amdgpu.h"
+#include "amdgpu_ip.h"
+
+static int8_t amdgpu_logical_to_dev_inst(struct amdgpu_device *adev,
+					 enum amd_hw_ip_block_type block,
+					 int8_t inst)
+{
+	int8_t dev_inst;
+
+	switch (block) {
+	case GC_HWIP:
+	case SDMA0_HWIP:
+	/* Both JPEG and VCN as JPEG is only alias of VCN */
+	case VCN_HWIP:
+		dev_inst = adev->ip_map.dev_inst[block][inst];
+		break;
+	default:
+		/* For rest of the IPs, no look up required.
+		 * Assume 'logical instance == physical instance' for all configs. */
+		dev_inst = inst;
+		break;
+	}
+
+	return dev_inst;
+}
+
+static uint32_t amdgpu_logical_to_dev_mask(struct amdgpu_device *adev,
+					   enum amd_hw_ip_block_type block,
+					   uint32_t mask)
+{
+	uint32_t dev_mask = 0;
+	int8_t log_inst, dev_inst;
+
+	while (mask) {
+		log_inst = ffs(mask) - 1;
+		dev_inst = amdgpu_logical_to_dev_inst(adev, block, log_inst);
+		dev_mask |= (1 << dev_inst);
+		mask &= ~(1 << log_inst);
+	}
+
+	return dev_mask;
+}
+
+static void amdgpu_populate_ip_map(struct amdgpu_device *adev,
+				   enum amd_hw_ip_block_type ip_block,
+				   uint32_t inst_mask)
+{
+	int l = 0, i;
+
+	while (inst_mask) {
+		i = ffs(inst_mask) - 1;
+		adev->ip_map.dev_inst[ip_block][l++] = i;
+		inst_mask &= ~(1 << i);
+	}
+	for (; l < HWIP_MAX_INSTANCE; l++)
+		adev->ip_map.dev_inst[ip_block][l] = -1;
+}
+
+void amdgpu_ip_map_init(struct amdgpu_device *adev)
+{
+	u32 ip_map[][2] = {
+		{ GC_HWIP, adev->gfx.xcc_mask },
+		{ SDMA0_HWIP, adev->sdma.sdma_mask },
+		{ VCN_HWIP, adev->vcn.inst_mask },
+	};
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(ip_map); ++i)
+		amdgpu_populate_ip_map(adev, ip_map[i][0], ip_map[i][1]);
+
+	adev->ip_map.logical_to_dev_inst = amdgpu_logical_to_dev_inst;
+	adev->ip_map.logical_to_dev_mask = amdgpu_logical_to_dev_mask;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ip.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ip.h
new file mode 100644
index 000000000000..2490fd322aec
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ip.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright 2025 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_IP_H__
+#define __AMDGPU_IP_H__
+
+void amdgpu_ip_map_init(struct amdgpu_device *adev);
+
+#endif /* __AMDGPU_IP_H__ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c
index a6b7e367a860..8112ffc85995 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c
@@ -25,76 +25,110 @@
  *          Alex Deucher
  *          Jerome Glisse
  */
+
+/**
+ * DOC: Interrupt Handling
+ *
+ * Interrupts generated within GPU hardware raise interrupt requests that are
+ * passed to amdgpu IRQ handler which is responsible for detecting source and
+ * type of the interrupt and dispatching matching handlers. If handling an
+ * interrupt requires calling kernel functions that may sleep processing is
+ * dispatched to work handlers.
+ *
+ * If MSI functionality is not disabled by module parameter then MSI
+ * support will be enabled.
+ *
+ * For GPU interrupt sources that may be driven by another driver, IRQ domain
+ * support is used (with mapping between virtual and hardware IRQs).
+ */
+
 #include <linux/irq.h>
-#include <drm/drmP.h>
-#include <drm/drm_crtc_helper.h>
+#include <linux/pci.h>
+
+#include <drm/drm_vblank.h>
 #include <drm/amdgpu_drm.h>
+#include <drm/drm_drv.h>
 #include "amdgpu.h"
 #include "amdgpu_ih.h"
 #include "atom.h"
 #include "amdgpu_connectors.h"
 #include "amdgpu_trace.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_ras.h"
 
 #include <linux/pm_runtime.h>
 
+#ifdef CONFIG_DRM_AMD_DC
+#include "amdgpu_dm_irq.h"
+#endif
+
 #define AMDGPU_WAIT_IDLE_TIMEOUT 200
 
-/*
- * Handle hotplug events outside the interrupt handler proper.
- */
-/**
- * amdgpu_hotplug_work_func - display hotplug work handler
- *
- * @work: work struct
- *
- * This is the hot plug event work handler (all asics).
- * The work gets scheduled from the irq handler if there
- * was a hot plug interrupt.  It walks the connector table
- * and calls the hotplug handler for each one, then sends
- * a drm hotplug event to alert userspace.
- */
-static void amdgpu_hotplug_work_func(struct work_struct *work)
-{
-	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
-						  hotplug_work);
-	struct drm_device *dev = adev->ddev;
-	struct drm_mode_config *mode_config = &dev->mode_config;
-	struct drm_connector *connector;
-
-	mutex_lock(&mode_config->mutex);
-	list_for_each_entry(connector, &mode_config->connector_list, head)
-		amdgpu_connector_hotplug(connector);
-	mutex_unlock(&mode_config->mutex);
-	/* Just fire off a uevent and let userspace tell us what to do */
-	drm_helper_hpd_irq_event(dev);
-}
+const char *soc15_ih_clientid_name[] = {
+	"IH",
+	"SDMA2 or ACP",
+	"ATHUB",
+	"BIF",
+	"SDMA3 or DCE",
+	"SDMA4 or ISP",
+	"VMC1 or PCIE0",
+	"RLC",
+	"SDMA0",
+	"SDMA1",
+	"SE0SH",
+	"SE1SH",
+	"SE2SH",
+	"SE3SH",
+	"VCN1 or UVD1",
+	"THM",
+	"VCN or UVD",
+	"SDMA5 or VCE0",
+	"VMC",
+	"SDMA6 or XDMA",
+	"GRBM_CP",
+	"ATS",
+	"ROM_SMUIO",
+	"DF",
+	"SDMA7 or VCE1",
+	"PWR",
+	"reserved",
+	"UTCL2",
+	"EA",
+	"UTCL2LOG",
+	"MP0",
+	"MP1"
+};
+
+const int node_id_to_phys_map[NODEID_MAX] = {
+	[AID0_NODEID] = 0,
+	[XCD0_NODEID] = 0,
+	[XCD1_NODEID] = 1,
+	[AID1_NODEID] = 1,
+	[XCD2_NODEID] = 2,
+	[XCD3_NODEID] = 3,
+	[AID2_NODEID] = 2,
+	[XCD4_NODEID] = 4,
+	[XCD5_NODEID] = 5,
+	[AID3_NODEID] = 3,
+	[XCD6_NODEID] = 6,
+	[XCD7_NODEID] = 7,
+};
 
 /**
- * amdgpu_irq_reset_work_func - execute gpu reset
+ * amdgpu_irq_disable_all - disable *all* interrupts
  *
- * @work: work struct
+ * @adev: amdgpu device pointer
  *
- * Execute scheduled gpu reset (cayman+).
- * This function is called when the irq handler
- * thinks we need a gpu reset.
+ * Disable all types of interrupts from all sources.
  */
-static void amdgpu_irq_reset_work_func(struct work_struct *work)
-{
-	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
-						  reset_work);
-
-	amdgpu_gpu_reset(adev);
-}
-
-/* Disable *all* interrupts */
-static void amdgpu_irq_disable_all(struct amdgpu_device *adev)
+void amdgpu_irq_disable_all(struct amdgpu_device *adev)
 {
 	unsigned long irqflags;
-	unsigned i, j, k;
+	unsigned int i, j, k;
 	int r;
 
 	spin_lock_irqsave(&adev->irq.lock, irqflags);
-	for (i = 0; i < AMDGPU_IH_CLIENTID_MAX; ++i) {
+	for (i = 0; i < AMDGPU_IRQ_CLIENTID_MAX; ++i) {
 		if (!adev->irq.client[i].sources)
 			continue;
 
@@ -105,12 +139,12 @@ static void amdgpu_irq_disable_all(struct amdgpu_device *adev)
 				continue;
 
 			for (k = 0; k < src->num_types; ++k) {
-				atomic_set(&src->enabled_types[k], 0);
 				r = src->funcs->set(adev, src, k,
 						    AMDGPU_IRQ_STATE_DISABLE);
 				if (r)
-					DRM_ERROR("error disabling interrupt (%d)\n",
-						  r);
+					dev_err(adev->dev,
+						"error disabling interrupt (%d)\n",
+						r);
 			}
 		}
 	}
@@ -118,86 +152,89 @@ static void amdgpu_irq_disable_all(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_irq_preinstall - drm irq preinstall callback
+ * amdgpu_irq_handler - IRQ handler
  *
- * @dev: drm dev pointer
+ * @irq: IRQ number (unused)
+ * @arg: pointer to DRM device
  *
- * Gets the hw ready to enable irqs (all asics).
- * This function disables all interrupt sources on the GPU.
+ * IRQ handler for amdgpu driver (all ASICs).
+ *
+ * Returns:
+ * result of handling the IRQ, as defined by &irqreturn_t
  */
-void amdgpu_irq_preinstall(struct drm_device *dev)
+static irqreturn_t amdgpu_irq_handler(int irq, void *arg)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct drm_device *dev = (struct drm_device *) arg;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	irqreturn_t ret;
+
+	ret = amdgpu_ih_process(adev, &adev->irq.ih);
+	if (ret == IRQ_HANDLED)
+		pm_runtime_mark_last_busy(dev->dev);
 
-	/* Disable *all* interrupts */
-	amdgpu_irq_disable_all(adev);
-	/* Clear bits */
-	amdgpu_ih_process(adev);
+	amdgpu_ras_interrupt_fatal_error_handler(adev);
+
+	return ret;
 }
 
 /**
- * amdgpu_irq_postinstall - drm irq preinstall callback
+ * amdgpu_irq_handle_ih1 - kick of processing for IH1
  *
- * @dev: drm dev pointer
+ * @work: work structure in struct amdgpu_irq
  *
- * Handles stuff to be done after enabling irqs (all asics).
- * Returns 0 on success.
+ * Kick of processing IH ring 1.
  */
-int amdgpu_irq_postinstall(struct drm_device *dev)
+static void amdgpu_irq_handle_ih1(struct work_struct *work)
 {
-	dev->max_vblank_count = 0x00ffffff;
-	return 0;
+	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
+						  irq.ih1_work);
+
+	amdgpu_ih_process(adev, &adev->irq.ih1);
 }
 
 /**
- * amdgpu_irq_uninstall - drm irq uninstall callback
+ * amdgpu_irq_handle_ih2 - kick of processing for IH2
  *
- * @dev: drm dev pointer
+ * @work: work structure in struct amdgpu_irq
  *
- * This function disables all interrupt sources on the GPU (all asics).
+ * Kick of processing IH ring 2.
  */
-void amdgpu_irq_uninstall(struct drm_device *dev)
+static void amdgpu_irq_handle_ih2(struct work_struct *work)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
+						  irq.ih2_work);
 
-	if (adev == NULL) {
-		return;
-	}
-	amdgpu_irq_disable_all(adev);
+	amdgpu_ih_process(adev, &adev->irq.ih2);
 }
 
 /**
- * amdgpu_irq_handler - irq handler
+ * amdgpu_irq_handle_ih_soft - kick of processing for ih_soft
  *
- * @int irq, void *arg: args
+ * @work: work structure in struct amdgpu_irq
  *
- * This is the irq handler for the amdgpu driver (all asics).
+ * Kick of processing IH soft ring.
  */
-irqreturn_t amdgpu_irq_handler(int irq, void *arg)
+static void amdgpu_irq_handle_ih_soft(struct work_struct *work)
 {
-	struct drm_device *dev = (struct drm_device *) arg;
-	struct amdgpu_device *adev = dev->dev_private;
-	irqreturn_t ret;
+	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
+						  irq.ih_soft_work);
 
-	ret = amdgpu_ih_process(adev);
-	if (ret == IRQ_HANDLED)
-		pm_runtime_mark_last_busy(dev->dev);
-	return ret;
+	amdgpu_ih_process(adev, &adev->irq.ih_soft);
 }
 
 /**
- * amdgpu_msi_ok - asic specific msi checks
+ * amdgpu_msi_ok - check whether MSI functionality is enabled
  *
- * @adev: amdgpu device pointer
+ * @adev: amdgpu device pointer (unused)
  *
- * Handles asic specific MSI checks to determine if
- * MSIs should be enabled on a particular chip (all asics).
- * Returns true if MSIs should be enabled, false if MSIs
- * should not be enabled.
+ * Checks whether MSI functionality has been disabled via module parameter
+ * (all ASICs).
+ *
+ * Returns:
+ * *true* if MSIs are allowed to be enabled or *false* otherwise
  */
 static bool amdgpu_msi_ok(struct amdgpu_device *adev)
 {
-	/* force MSI on */
 	if (amdgpu_msi == 1)
 		return true;
 	else if (amdgpu_msi == 0)
@@ -206,73 +243,119 @@ static bool amdgpu_msi_ok(struct amdgpu_device *adev)
 	return true;
 }
 
+void amdgpu_restore_msix(struct amdgpu_device *adev)
+{
+	u16 ctrl;
+
+	pci_read_config_word(adev->pdev, adev->pdev->msix_cap + PCI_MSIX_FLAGS, &ctrl);
+	if (!(ctrl & PCI_MSIX_FLAGS_ENABLE))
+		return;
+
+	/* VF FLR */
+	ctrl &= ~PCI_MSIX_FLAGS_ENABLE;
+	pci_write_config_word(adev->pdev, adev->pdev->msix_cap + PCI_MSIX_FLAGS, ctrl);
+	ctrl |= PCI_MSIX_FLAGS_ENABLE;
+	pci_write_config_word(adev->pdev, adev->pdev->msix_cap + PCI_MSIX_FLAGS, ctrl);
+}
+
 /**
- * amdgpu_irq_init - init driver interrupt info
+ * amdgpu_irq_init - initialize interrupt handling
  *
  * @adev: amdgpu device pointer
  *
- * Sets up the work irq handlers, vblank init, MSIs, etc. (all asics).
- * Returns 0 for success, error for failure.
+ * Sets up work functions for hotplug and reset interrupts, enables MSI
+ * functionality, initializes vblank, hotplug and reset interrupt handling.
+ *
+ * Returns:
+ * 0 on success or error code on failure
  */
 int amdgpu_irq_init(struct amdgpu_device *adev)
 {
-	int r = 0;
+	unsigned int irq, flags;
+	int r;
 
 	spin_lock_init(&adev->irq.lock);
-	r = drm_vblank_init(adev->ddev, adev->mode_info.num_crtc);
-	if (r) {
-		return r;
-	}
 
-	/* enable msi */
+	/* Enable MSI if not disabled by module parameter */
 	adev->irq.msi_enabled = false;
 
+	if (!amdgpu_msi_ok(adev))
+		flags = PCI_IRQ_INTX;
+	else
+		flags = PCI_IRQ_ALL_TYPES;
+
+	/* we only need one vector */
+	r = pci_alloc_irq_vectors(adev->pdev, 1, 1, flags);
+	if (r < 0) {
+		dev_err(adev->dev, "Failed to alloc msi vectors\n");
+		return r;
+	}
+
 	if (amdgpu_msi_ok(adev)) {
-		int ret = pci_enable_msi(adev->pdev);
-		if (!ret) {
-			adev->irq.msi_enabled = true;
-			dev_info(adev->dev, "amdgpu: using MSI.\n");
-		}
+		adev->irq.msi_enabled = true;
+		dev_dbg(adev->dev, "using MSI/MSI-X.\n");
 	}
 
-	INIT_WORK(&adev->hotplug_work, amdgpu_hotplug_work_func);
-	INIT_WORK(&adev->reset_work, amdgpu_irq_reset_work_func);
+	INIT_WORK(&adev->irq.ih1_work, amdgpu_irq_handle_ih1);
+	INIT_WORK(&adev->irq.ih2_work, amdgpu_irq_handle_ih2);
+	INIT_WORK(&adev->irq.ih_soft_work, amdgpu_irq_handle_ih_soft);
+
+	/* Use vector 0 for MSI-X. */
+	r = pci_irq_vector(adev->pdev, 0);
+	if (r < 0)
+		goto free_vectors;
+	irq = r;
+
+	/* PCI devices require shared interrupts. */
+	r = request_irq(irq, amdgpu_irq_handler, IRQF_SHARED, adev_to_drm(adev)->driver->name,
+			adev_to_drm(adev));
+	if (r)
+		goto free_vectors;
 
 	adev->irq.installed = true;
-	r = drm_irq_install(adev->ddev, adev->ddev->pdev->irq);
-	if (r) {
+	adev->irq.irq = irq;
+	adev_to_drm(adev)->max_vblank_count = 0x00ffffff;
+
+	dev_dbg(adev->dev, "amdgpu: irq initialized.\n");
+	return 0;
+
+free_vectors:
+	if (adev->irq.msi_enabled)
+		pci_free_irq_vectors(adev->pdev);
+
+	adev->irq.msi_enabled = false;
+	return r;
+}
+
+void amdgpu_irq_fini_hw(struct amdgpu_device *adev)
+{
+	if (adev->irq.installed) {
+		free_irq(adev->irq.irq, adev_to_drm(adev));
 		adev->irq.installed = false;
-		flush_work(&adev->hotplug_work);
-		cancel_work_sync(&adev->reset_work);
-		return r;
+		if (adev->irq.msi_enabled)
+			pci_free_irq_vectors(adev->pdev);
 	}
 
-	DRM_INFO("amdgpu: irq initialized.\n");
-	return 0;
+	amdgpu_ih_ring_fini(adev, &adev->irq.ih_soft);
+	amdgpu_ih_ring_fini(adev, &adev->irq.ih);
+	amdgpu_ih_ring_fini(adev, &adev->irq.ih1);
+	amdgpu_ih_ring_fini(adev, &adev->irq.ih2);
 }
 
 /**
- * amdgpu_irq_fini - tear down driver interrupt info
+ * amdgpu_irq_fini_sw - shut down interrupt handling
  *
  * @adev: amdgpu device pointer
  *
- * Tears down the work irq handlers, vblank handlers, MSIs, etc. (all asics).
+ * Tears down work functions for hotplug and reset interrupts, disables MSI
+ * functionality, shuts down vblank, hotplug and reset interrupt handling,
+ * turns off interrupts from all sources (all ASICs).
  */
-void amdgpu_irq_fini(struct amdgpu_device *adev)
+void amdgpu_irq_fini_sw(struct amdgpu_device *adev)
 {
-	unsigned i, j;
+	unsigned int i, j;
 
-	drm_vblank_cleanup(adev->ddev);
-	if (adev->irq.installed) {
-		drm_irq_uninstall(adev->ddev);
-		adev->irq.installed = false;
-		if (adev->irq.msi_enabled)
-			pci_disable_msi(adev->pdev);
-		flush_work(&adev->hotplug_work);
-		cancel_work_sync(&adev->reset_work);
-	}
-
-	for (i = 0; i < AMDGPU_IH_CLIENTID_MAX; ++i) {
+	for (i = 0; i < AMDGPU_IRQ_CLIENTID_MAX; ++i) {
 		if (!adev->irq.client[i].sources)
 			continue;
 
@@ -284,29 +367,30 @@ void amdgpu_irq_fini(struct amdgpu_device *adev)
 
 			kfree(src->enabled_types);
 			src->enabled_types = NULL;
-			if (src->data) {
-				kfree(src->data);
-				kfree(src);
-				adev->irq.client[i].sources[j] = NULL;
-			}
 		}
 		kfree(adev->irq.client[i].sources);
+		adev->irq.client[i].sources = NULL;
 	}
 }
 
 /**
- * amdgpu_irq_add_id - register irq source
+ * amdgpu_irq_add_id - register IRQ source
  *
  * @adev: amdgpu device pointer
- * @src_id: source id for this source
- * @source: irq source
+ * @client_id: client id
+ * @src_id: source id
+ * @source: IRQ source pointer
+ *
+ * Registers IRQ source on a client.
  *
+ * Returns:
+ * 0 on success or error code otherwise
  */
 int amdgpu_irq_add_id(struct amdgpu_device *adev,
-		      unsigned client_id, unsigned src_id,
+		      unsigned int client_id, unsigned int src_id,
 		      struct amdgpu_irq_src *source)
 {
-	if (client_id >= AMDGPU_IH_CLIENTID_MAX)
+	if (client_id >= AMDGPU_IRQ_CLIENTID_MAX)
 		return -EINVAL;
 
 	if (src_id >= AMDGPU_MAX_IRQ_SRC_ID)
@@ -343,65 +427,107 @@ int amdgpu_irq_add_id(struct amdgpu_device *adev,
 }
 
 /**
- * amdgpu_irq_dispatch - dispatch irq to IP blocks
+ * amdgpu_irq_dispatch - dispatch IRQ to IP blocks
  *
  * @adev: amdgpu device pointer
- * @entry: interrupt vector
+ * @ih: interrupt ring instance
  *
- * Dispatches the irq to the different IP blocks
+ * Dispatches IRQ to IP blocks.
  */
 void amdgpu_irq_dispatch(struct amdgpu_device *adev,
-			 struct amdgpu_iv_entry *entry)
+			 struct amdgpu_ih_ring *ih)
 {
-	unsigned client_id = entry->client_id;
-	unsigned src_id = entry->src_id;
+	u32 ring_index = ih->rptr >> 2;
+	struct amdgpu_iv_entry entry;
+	unsigned int client_id, src_id;
 	struct amdgpu_irq_src *src;
+	bool handled = false;
 	int r;
 
-	trace_amdgpu_iv(entry);
+	entry.ih = ih;
+	entry.iv_entry = (const uint32_t *)&ih->ring[ring_index];
 
-	if (client_id >= AMDGPU_IH_CLIENTID_MAX) {
-		DRM_DEBUG("Invalid client_id in IV: %d\n", client_id);
-		return;
-	}
+	/*
+	 * timestamp is not supported on some legacy SOCs (cik, cz, iceland,
+	 * si and tonga), so initialize timestamp and timestamp_src to 0
+	 */
+	entry.timestamp = 0;
+	entry.timestamp_src = 0;
 
-	if (src_id >= AMDGPU_MAX_IRQ_SRC_ID) {
-		DRM_DEBUG("Invalid src_id in IV: %d\n", src_id);
-		return;
-	}
+	amdgpu_ih_decode_iv(adev, &entry);
 
-	if (adev->irq.virq[src_id]) {
-		generic_handle_irq(irq_find_mapping(adev->irq.domain, src_id));
-	} else {
-		if (!adev->irq.client[client_id].sources) {
-			DRM_DEBUG("Unregistered interrupt client_id: %d src_id: %d\n",
-				  client_id, src_id);
-			return;
-		}
+	trace_amdgpu_iv(ih - &adev->irq.ih, &entry);
 
-		src = adev->irq.client[client_id].sources[src_id];
-		if (!src) {
-			DRM_DEBUG("Unhandled interrupt src_id: %d\n", src_id);
-			return;
-		}
+	client_id = entry.client_id;
+	src_id = entry.src_id;
+
+	if (client_id >= AMDGPU_IRQ_CLIENTID_MAX) {
+		dev_dbg(adev->dev, "Invalid client_id in IV: %d\n", client_id);
+
+	} else	if (src_id >= AMDGPU_MAX_IRQ_SRC_ID) {
+		dev_dbg(adev->dev, "Invalid src_id in IV: %d\n", src_id);
+
+	} else if (((client_id == AMDGPU_IRQ_CLIENTID_LEGACY) ||
+		    (client_id == SOC15_IH_CLIENTID_ISP)) &&
+		   adev->irq.virq[src_id]) {
+		generic_handle_domain_irq(adev->irq.domain, src_id);
+
+	} else if (!adev->irq.client[client_id].sources) {
+		dev_dbg(adev->dev,
+			"Unregistered interrupt client_id: %d src_id: %d\n",
+			client_id, src_id);
+
+	} else if ((src = adev->irq.client[client_id].sources[src_id])) {
+		r = src->funcs->process(adev, src, &entry);
+		if (r < 0)
+			dev_err(adev->dev, "error processing interrupt (%d)\n",
+				r);
+		else if (r)
+			handled = true;
 
-		r = src->funcs->process(adev, src, entry);
-		if (r)
-			DRM_ERROR("error processing interrupt (%d)\n", r);
+	} else {
+		dev_dbg(adev->dev,
+			"Unregistered interrupt src_id: %d of client_id:%d\n",
+			src_id, client_id);
 	}
+
+	/* Send it to amdkfd as well if it isn't already handled */
+	if (!handled)
+		amdgpu_amdkfd_interrupt(adev, entry.iv_entry);
+
+	if (amdgpu_ih_ts_after(ih->processed_timestamp, entry.timestamp))
+		ih->processed_timestamp = entry.timestamp;
+}
+
+/**
+ * amdgpu_irq_delegate - delegate IV to soft IH ring
+ *
+ * @adev: amdgpu device pointer
+ * @entry: IV entry
+ * @num_dw: size of IV
+ *
+ * Delegate the IV to the soft IH ring and schedule processing of it. Used
+ * if the hardware delegation to IH1 or IH2 doesn't work for some reason.
+ */
+void amdgpu_irq_delegate(struct amdgpu_device *adev,
+			 struct amdgpu_iv_entry *entry,
+			 unsigned int num_dw)
+{
+	amdgpu_ih_ring_write(adev, &adev->irq.ih_soft, entry->iv_entry, num_dw);
+	schedule_work(&adev->irq.ih_soft_work);
 }
 
 /**
- * amdgpu_irq_update - update hw interrupt state
+ * amdgpu_irq_update - update hardware interrupt state
  *
  * @adev: amdgpu device pointer
- * @src: interrupt src you want to enable
- * @type: type of interrupt you want to update
+ * @src: interrupt source pointer
+ * @type: type of interrupt
  *
- * Updates the interrupt state for a specific src (all asics).
+ * Updates interrupt state for the specific source (all ASICs).
  */
 int amdgpu_irq_update(struct amdgpu_device *adev,
-			     struct amdgpu_irq_src *src, unsigned type)
+			     struct amdgpu_irq_src *src, unsigned int type)
 {
 	unsigned long irqflags;
 	enum amdgpu_interrupt_state state;
@@ -409,8 +535,9 @@ int amdgpu_irq_update(struct amdgpu_device *adev,
 
 	spin_lock_irqsave(&adev->irq.lock, irqflags);
 
-	/* we need to determine after taking the lock, otherwise
-	   we might disable just enabled interrupts again */
+	/* We need to determine after taking the lock, otherwise
+	 * we might disable just enabled interrupts again
+	 */
 	if (amdgpu_irq_enabled(adev, src, type))
 		state = AMDGPU_IRQ_STATE_ENABLE;
 	else
@@ -421,18 +548,29 @@ int amdgpu_irq_update(struct amdgpu_device *adev,
 	return r;
 }
 
+/**
+ * amdgpu_irq_gpu_reset_resume_helper - update interrupt states on all sources
+ *
+ * @adev: amdgpu device pointer
+ *
+ * Updates state of all types of interrupts on all sources on resume after
+ * reset.
+ */
 void amdgpu_irq_gpu_reset_resume_helper(struct amdgpu_device *adev)
 {
 	int i, j, k;
 
-	for (i = 0; i < AMDGPU_IH_CLIENTID_MAX; ++i) {
+	if (amdgpu_sriov_vf(adev) || amdgpu_passthrough(adev))
+		amdgpu_restore_msix(adev);
+
+	for (i = 0; i < AMDGPU_IRQ_CLIENTID_MAX; ++i) {
 		if (!adev->irq.client[i].sources)
 			continue;
 
 		for (j = 0; j < AMDGPU_MAX_IRQ_SRC_ID; ++j) {
 			struct amdgpu_irq_src *src = adev->irq.client[i].sources[j];
 
-			if (!src)
+			if (!src || !src->funcs || !src->funcs->set)
 				continue;
 			for (k = 0; k < src->num_types; k++)
 				amdgpu_irq_update(adev, src, k);
@@ -444,15 +582,18 @@ void amdgpu_irq_gpu_reset_resume_helper(struct amdgpu_device *adev)
  * amdgpu_irq_get - enable interrupt
  *
  * @adev: amdgpu device pointer
- * @src: interrupt src you want to enable
- * @type: type of interrupt you want to enable
+ * @src: interrupt source pointer
+ * @type: type of interrupt
  *
- * Enables the interrupt type for a specific src (all asics).
+ * Enables specified type of interrupt on the specified source (all ASICs).
+ *
+ * Returns:
+ * 0 on success or error code otherwise
  */
 int amdgpu_irq_get(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
-		   unsigned type)
+		   unsigned int type)
 {
-	if (!adev->ddev->irq_enabled)
+	if (!adev->irq.installed)
 		return -ENOENT;
 
 	if (type >= src->num_types)
@@ -471,15 +612,22 @@ int amdgpu_irq_get(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
  * amdgpu_irq_put - disable interrupt
  *
  * @adev: amdgpu device pointer
- * @src: interrupt src you want to disable
- * @type: type of interrupt you want to disable
+ * @src: interrupt source pointer
+ * @type: type of interrupt
+ *
+ * Enables specified type of interrupt on the specified source (all ASICs).
  *
- * Disables the interrupt type for a specific src (all asics).
+ * Returns:
+ * 0 on success or error code otherwise
  */
 int amdgpu_irq_put(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
-		   unsigned type)
+		   unsigned int type)
 {
-	if (!adev->ddev->irq_enabled)
+	/* When the threshold is reached,the interrupt source may not be enabled.return -EINVAL */
+	if (amdgpu_ras_is_rma(adev) && !amdgpu_irq_enabled(adev, src, type))
+		return -EINVAL;
+
+	if (!adev->irq.installed)
 		return -ENOENT;
 
 	if (type >= src->num_types)
@@ -488,6 +636,9 @@ int amdgpu_irq_put(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
 	if (!src->enabled_types || !src->funcs->set)
 		return -EINVAL;
 
+	if (WARN_ON(!amdgpu_irq_enabled(adev, src, type)))
+		return -EINVAL;
+
 	if (atomic_dec_and_test(&src->enabled_types[type]))
 		return amdgpu_irq_update(adev, src, type);
 
@@ -495,17 +646,22 @@ int amdgpu_irq_put(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
 }
 
 /**
- * amdgpu_irq_enabled - test if irq is enabled or not
+ * amdgpu_irq_enabled - check whether interrupt is enabled or not
  *
  * @adev: amdgpu device pointer
- * @idx: interrupt src you want to test
+ * @src: interrupt source pointer
+ * @type: type of interrupt
  *
- * Tests if the given interrupt source is enabled or not
+ * Checks whether the given type of interrupt is enabled on the given source.
+ *
+ * Returns:
+ * *true* if interrupt is enabled, *false* if interrupt is disabled or on
+ * invalid parameters
  */
 bool amdgpu_irq_enabled(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
-			unsigned type)
+			unsigned int type)
 {
-	if (!adev->ddev->irq_enabled)
+	if (!adev->irq.installed)
 		return false;
 
 	if (type >= src->num_types)
@@ -517,7 +673,7 @@ bool amdgpu_irq_enabled(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
 	return !!atomic_read(&src->enabled_types[type]);
 }
 
-/* gen irq */
+/* XXX: Generic IRQ handling */
 static void amdgpu_irq_mask(struct irq_data *irqd)
 {
 	/* XXX */
@@ -528,12 +684,26 @@ static void amdgpu_irq_unmask(struct irq_data *irqd)
 	/* XXX */
 }
 
+/* amdgpu hardware interrupt chip descriptor */
 static struct irq_chip amdgpu_irq_chip = {
 	.name = "amdgpu-ih",
 	.irq_mask = amdgpu_irq_mask,
 	.irq_unmask = amdgpu_irq_unmask,
 };
 
+/**
+ * amdgpu_irqdomain_map - create mapping between virtual and hardware IRQ numbers
+ *
+ * @d: amdgpu IRQ domain pointer (unused)
+ * @irq: virtual IRQ number
+ * @hwirq: hardware irq number
+ *
+ * Current implementation assigns simple interrupt handler to the given virtual
+ * IRQ.
+ *
+ * Returns:
+ * 0 on success or error code otherwise
+ */
 static int amdgpu_irqdomain_map(struct irq_domain *d,
 				unsigned int irq, irq_hw_number_t hwirq)
 {
@@ -545,24 +715,28 @@ static int amdgpu_irqdomain_map(struct irq_domain *d,
 	return 0;
 }
 
+/* Implementation of methods for amdgpu IRQ domain */
 static const struct irq_domain_ops amdgpu_hw_irqdomain_ops = {
 	.map = amdgpu_irqdomain_map,
 };
 
 /**
- * amdgpu_irq_add_domain - create a linear irq domain
+ * amdgpu_irq_add_domain - create a linear IRQ domain
  *
  * @adev: amdgpu device pointer
  *
- * Create an irq domain for GPU interrupt sources
+ * Creates an IRQ domain for GPU interrupt sources
  * that may be driven by another driver (e.g., ACP).
+ *
+ * Returns:
+ * 0 on success or error code otherwise
  */
 int amdgpu_irq_add_domain(struct amdgpu_device *adev)
 {
-	adev->irq.domain = irq_domain_add_linear(NULL, AMDGPU_MAX_IRQ_SRC_ID,
-						 &amdgpu_hw_irqdomain_ops, adev);
+	adev->irq.domain = irq_domain_create_linear(NULL, AMDGPU_MAX_IRQ_SRC_ID,
+						    &amdgpu_hw_irqdomain_ops, adev);
 	if (!adev->irq.domain) {
-		DRM_ERROR("GPU irq add domain failed\n");
+		dev_err(adev->dev, "GPU irq add domain failed\n");
 		return -ENODEV;
 	}
 
@@ -570,11 +744,11 @@ int amdgpu_irq_add_domain(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_irq_remove_domain - remove the irq domain
+ * amdgpu_irq_remove_domain - remove the IRQ domain
  *
  * @adev: amdgpu device pointer
  *
- * Remove the irq domain for GPU interrupt sources
+ * Removes the IRQ domain for GPU interrupt sources
  * that may be driven by another driver (e.g., ACP).
  */
 void amdgpu_irq_remove_domain(struct amdgpu_device *adev)
@@ -586,18 +760,19 @@ void amdgpu_irq_remove_domain(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_irq_create_mapping - create a mapping between a domain irq and a
- *                             Linux irq
+ * amdgpu_irq_create_mapping - create mapping between domain Linux IRQs
  *
  * @adev: amdgpu device pointer
  * @src_id: IH source id
  *
- * Create a mapping between a domain irq (GPU IH src id) and a Linux irq
+ * Creates mapping between a domain IRQ (GPU IH src id) and a Linux IRQ
  * Use this for components that generate a GPU interrupt, but are driven
  * by a different driver (e.g., ACP).
- * Returns the Linux irq.
+ *
+ * Returns:
+ * Linux IRQ
  */
-unsigned amdgpu_irq_create_mapping(struct amdgpu_device *adev, unsigned src_id)
+unsigned int amdgpu_irq_create_mapping(struct amdgpu_device *adev, unsigned int src_id)
 {
 	adev->irq.virq[src_id] = irq_create_mapping(adev->irq.domain, src_id);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h
index 0610cc4a9788..9f0417456abd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h
@@ -25,24 +25,43 @@
 #define __AMDGPU_IRQ_H__
 
 #include <linux/irqdomain.h>
+#include "soc15_ih_clientid.h"
 #include "amdgpu_ih.h"
 
-#define AMDGPU_MAX_IRQ_SRC_ID	0x100
+#define AMDGPU_MAX_IRQ_SRC_ID		0x100
 #define AMDGPU_MAX_IRQ_CLIENT_ID	0x100
 
+#define AMDGPU_IRQ_CLIENTID_LEGACY	0
+#define AMDGPU_IRQ_CLIENTID_MAX		SOC15_IH_CLIENTID_MAX
+
+#define AMDGPU_IRQ_SRC_DATA_MAX_SIZE_DW	4
+
 struct amdgpu_device;
-struct amdgpu_iv_entry;
 
 enum amdgpu_interrupt_state {
 	AMDGPU_IRQ_STATE_DISABLE,
 	AMDGPU_IRQ_STATE_ENABLE,
 };
 
+struct amdgpu_iv_entry {
+	struct amdgpu_ih_ring *ih;
+	unsigned client_id;
+	unsigned src_id;
+	unsigned ring_id;
+	unsigned vmid;
+	unsigned vmid_src;
+	uint64_t timestamp;
+	unsigned timestamp_src;
+	unsigned pasid;
+	unsigned node_id;
+	unsigned src_data[AMDGPU_IRQ_SRC_DATA_MAX_SIZE_DW];
+	const uint32_t *iv_entry;
+};
+
 struct amdgpu_irq_src {
 	unsigned				num_types;
 	atomic_t				*enabled_types;
 	const struct amdgpu_irq_src_funcs	*funcs;
-	void *data;
 };
 
 struct amdgpu_irq_client {
@@ -61,35 +80,59 @@ struct amdgpu_irq_src_funcs {
 
 struct amdgpu_irq {
 	bool				installed;
+	unsigned int			irq;
 	spinlock_t			lock;
 	/* interrupt sources */
-	struct amdgpu_irq_client	client[AMDGPU_IH_CLIENTID_MAX];
+	struct amdgpu_irq_client	client[AMDGPU_IRQ_CLIENTID_MAX];
 
 	/* status, etc. */
 	bool				msi_enabled; /* msi enabled */
 
-	/* interrupt ring */
-	struct amdgpu_ih_ring		ih;
-	const struct amdgpu_ih_funcs	*ih_funcs;
+	/* interrupt rings */
+	struct amdgpu_ih_ring		ih, ih1, ih2, ih_soft;
+	const struct amdgpu_ih_funcs    *ih_funcs;
+	struct work_struct		ih1_work, ih2_work, ih_soft_work;
+	struct amdgpu_irq_src		self_irq;
 
 	/* gen irq stuff */
 	struct irq_domain		*domain; /* GPU irq controller domain */
 	unsigned			virq[AMDGPU_MAX_IRQ_SRC_ID];
 	uint32_t                        srbm_soft_reset;
+	u32                             retry_cam_doorbell_index;
+	bool                            retry_cam_enabled;
 };
 
-void amdgpu_irq_preinstall(struct drm_device *dev);
-int amdgpu_irq_postinstall(struct drm_device *dev);
-void amdgpu_irq_uninstall(struct drm_device *dev);
-irqreturn_t amdgpu_irq_handler(int irq, void *arg);
+enum interrupt_node_id_per_aid {
+	AID0_NODEID = 0,
+	XCD0_NODEID = 1,
+	XCD1_NODEID = 2,
+	AID1_NODEID = 4,
+	XCD2_NODEID = 5,
+	XCD3_NODEID = 6,
+	AID2_NODEID = 8,
+	XCD4_NODEID = 9,
+	XCD5_NODEID = 10,
+	AID3_NODEID = 12,
+	XCD6_NODEID = 13,
+	XCD7_NODEID = 14,
+	NODEID_MAX,
+};
+
+extern const int node_id_to_phys_map[NODEID_MAX];
+
+void amdgpu_irq_disable_all(struct amdgpu_device *adev);
 
 int amdgpu_irq_init(struct amdgpu_device *adev);
-void amdgpu_irq_fini(struct amdgpu_device *adev);
+void amdgpu_irq_fini_sw(struct amdgpu_device *adev);
+void amdgpu_irq_fini_hw(struct amdgpu_device *adev);
 int amdgpu_irq_add_id(struct amdgpu_device *adev,
 		      unsigned client_id, unsigned src_id,
 		      struct amdgpu_irq_src *source);
 void amdgpu_irq_dispatch(struct amdgpu_device *adev,
-			 struct amdgpu_iv_entry *entry);
+			 struct amdgpu_ih_ring *ih);
+void amdgpu_irq_delegate(struct amdgpu_device *adev,
+			 struct amdgpu_iv_entry *entry,
+			 unsigned int num_dw);
 int amdgpu_irq_update(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
 		      unsigned type);
 int amdgpu_irq_get(struct amdgpu_device *adev, struct amdgpu_irq_src *src,
@@ -103,5 +146,6 @@ void amdgpu_irq_gpu_reset_resume_helper(struct amdgpu_device *adev);
 int amdgpu_irq_add_domain(struct amdgpu_device *adev);
 void amdgpu_irq_remove_domain(struct amdgpu_device *adev);
 unsigned amdgpu_irq_create_mapping(struct amdgpu_device *adev, unsigned src_id);
+void amdgpu_restore_msix(struct amdgpu_device *adev);
 
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_isp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_isp.c
new file mode 100644
index 000000000000..37270c4dab8d
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_isp.c
@@ -0,0 +1,345 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright (C) 2024 Advanced Micro Devices, Inc. All rights reserved.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+
+#include <linux/firmware.h>
+#include <linux/mfd/core.h>
+
+#include "amdgpu.h"
+#include "amdgpu_isp.h"
+#include "isp_v4_1_0.h"
+#include "isp_v4_1_1.h"
+
+#define ISP_MC_ADDR_ALIGN (1024 * 32)
+
+/**
+ * isp_hw_init - start and test isp block
+ *
+ * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
+ *
+ */
+static int isp_hw_init(struct amdgpu_ip_block *ip_block)
+{
+	struct amdgpu_device *adev = ip_block->adev;
+	struct amdgpu_isp *isp = &adev->isp;
+
+	if (isp->funcs->hw_init != NULL)
+		return isp->funcs->hw_init(isp);
+
+	return -ENODEV;
+}
+
+/**
+ * isp_hw_fini - stop the hardware block
+ *
+ * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
+ *
+ */
+static int isp_hw_fini(struct amdgpu_ip_block *ip_block)
+{
+	struct amdgpu_isp *isp = &ip_block->adev->isp;
+
+	if (isp->funcs->hw_fini != NULL)
+		return isp->funcs->hw_fini(isp);
+
+	return -ENODEV;
+}
+
+static int isp_load_fw_by_psp(struct amdgpu_device *adev)
+{
+	const struct common_firmware_header *hdr;
+	char ucode_prefix[10];
+	int r = 0;
+
+	/* get isp fw binary name and path */
+	amdgpu_ucode_ip_version_decode(adev, ISP_HWIP, ucode_prefix,
+				       sizeof(ucode_prefix));
+
+	/* read isp fw */
+	r = amdgpu_ucode_request(adev, &adev->isp.fw, AMDGPU_UCODE_OPTIONAL,
+				"amdgpu/%s.bin", ucode_prefix);
+	if (r) {
+		amdgpu_ucode_release(&adev->isp.fw);
+		return r;
+	}
+
+	hdr = (const struct common_firmware_header *)adev->isp.fw->data;
+
+	adev->firmware.ucode[AMDGPU_UCODE_ID_ISP].ucode_id =
+		AMDGPU_UCODE_ID_ISP;
+	adev->firmware.ucode[AMDGPU_UCODE_ID_ISP].fw = adev->isp.fw;
+
+	adev->firmware.fw_size +=
+		ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
+
+	return r;
+}
+
+static int isp_early_init(struct amdgpu_ip_block *ip_block)
+{
+
+	struct amdgpu_device *adev = ip_block->adev;
+	struct amdgpu_isp *isp = &adev->isp;
+
+	switch (amdgpu_ip_version(adev, ISP_HWIP, 0)) {
+	case IP_VERSION(4, 1, 0):
+		isp_v4_1_0_set_isp_funcs(isp);
+		break;
+	case IP_VERSION(4, 1, 1):
+		isp_v4_1_1_set_isp_funcs(isp);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	isp->adev = adev;
+	isp->parent = adev->dev;
+
+	if (isp_load_fw_by_psp(adev)) {
+		DRM_DEBUG_DRIVER("%s: isp fw load failed\n", __func__);
+		return -ENOENT;
+	}
+
+	return 0;
+}
+
+static bool isp_is_idle(struct amdgpu_ip_block *ip_block)
+{
+	return true;
+}
+
+static int isp_set_clockgating_state(struct amdgpu_ip_block *ip_block,
+				     enum amd_clockgating_state state)
+{
+	return 0;
+}
+
+static int isp_set_powergating_state(struct amdgpu_ip_block *ip_block,
+				     enum amd_powergating_state state)
+{
+	return 0;
+}
+
+static int is_valid_isp_device(struct device *isp_parent, struct device *amdgpu_dev)
+{
+	if (isp_parent != amdgpu_dev)
+		return -EINVAL;
+
+	return 0;
+}
+
+/**
+ * isp_user_buffer_alloc - create user buffer object (BO) for isp
+ *
+ * @dev: isp device handle
+ * @dmabuf: DMABUF handle for isp buffer allocated in system memory
+ * @buf_obj: GPU buffer object handle to initialize
+ * @buf_addr: GPU addr of the pinned BO to initialize
+ *
+ * Imports isp DMABUF to allocate and pin a user BO for isp internal use. It does
+ * GART alloc to generate GPU addr for BO to make it accessible through the
+ * GART aperture for ISP HW.
+ *
+ * This function is exported to allow the V4L2 isp device external to drm device
+ * to create and access the isp user BO.
+ *
+ * Returns:
+ * 0 on success, negative error code otherwise.
+ */
+int isp_user_buffer_alloc(struct device *dev, void *dmabuf,
+			  void **buf_obj, u64 *buf_addr)
+{
+	struct platform_device *ispdev = to_platform_device(dev);
+	const struct isp_platform_data *isp_pdata;
+	struct amdgpu_device *adev;
+	struct mfd_cell *mfd_cell;
+	struct amdgpu_bo *bo;
+	u64 gpu_addr;
+	int ret;
+
+	if (WARN_ON(!ispdev))
+		return -ENODEV;
+
+	if (WARN_ON(!buf_obj))
+		return -EINVAL;
+
+	if (WARN_ON(!buf_addr))
+		return -EINVAL;
+
+	mfd_cell = &ispdev->mfd_cell[0];
+	if (!mfd_cell)
+		return -ENODEV;
+
+	isp_pdata = mfd_cell->platform_data;
+	adev = isp_pdata->adev;
+
+	ret = is_valid_isp_device(ispdev->dev.parent, adev->dev);
+	if (ret)
+		return ret;
+
+	ret = amdgpu_bo_create_isp_user(adev, dmabuf,
+					AMDGPU_GEM_DOMAIN_GTT, &bo, &gpu_addr);
+	if (ret) {
+		drm_err(&adev->ddev, "failed to alloc gart user buffer (%d)", ret);
+		return ret;
+	}
+
+	*buf_obj = (void *)bo;
+	*buf_addr = gpu_addr;
+
+	return 0;
+}
+EXPORT_SYMBOL(isp_user_buffer_alloc);
+
+/**
+ * isp_user_buffer_free - free isp user buffer object (BO)
+ *
+ * @buf_obj: amdgpu isp user BO to free
+ *
+ * unpin and unref BO for isp internal use.
+ *
+ * This function is exported to allow the V4L2 isp device
+ * external to drm device to free the isp user BO.
+ */
+void isp_user_buffer_free(void *buf_obj)
+{
+	amdgpu_bo_free_isp_user(buf_obj);
+}
+EXPORT_SYMBOL(isp_user_buffer_free);
+
+/**
+ * isp_kernel_buffer_alloc - create kernel buffer object (BO) for isp
+ *
+ * @dev: isp device handle
+ * @size: size for the new BO
+ * @buf_obj: GPU BO handle to initialize
+ * @gpu_addr: GPU addr of the pinned BO
+ * @cpu_addr: CPU address mapping of BO
+ *
+ * Allocates and pins a kernel BO for internal isp firmware use.
+ *
+ * This function is exported to allow the V4L2 isp device
+ * external to drm device to create and access the kernel BO.
+ *
+ * Returns:
+ * 0 on success, negative error code otherwise.
+ */
+int isp_kernel_buffer_alloc(struct device *dev, u64 size,
+			    void **buf_obj, u64 *gpu_addr, void **cpu_addr)
+{
+	struct platform_device *ispdev = to_platform_device(dev);
+	struct amdgpu_bo **bo = (struct amdgpu_bo **)buf_obj;
+	const struct isp_platform_data *isp_pdata;
+	struct amdgpu_device *adev;
+	struct mfd_cell *mfd_cell;
+	int ret;
+
+	if (WARN_ON(!ispdev))
+		return -ENODEV;
+
+	if (WARN_ON(!buf_obj))
+		return -EINVAL;
+
+	if (WARN_ON(!gpu_addr))
+		return -EINVAL;
+
+	if (WARN_ON(!cpu_addr))
+		return -EINVAL;
+
+	mfd_cell = &ispdev->mfd_cell[0];
+	if (!mfd_cell)
+		return -ENODEV;
+
+	isp_pdata = mfd_cell->platform_data;
+	adev = isp_pdata->adev;
+
+	ret = is_valid_isp_device(ispdev->dev.parent, adev->dev);
+	if (ret)
+		return ret;
+
+	/* Ensure *bo is NULL so a new BO will be created */
+	*bo = NULL;
+	ret = amdgpu_bo_create_kernel(adev,
+				      size,
+				      ISP_MC_ADDR_ALIGN,
+				      AMDGPU_GEM_DOMAIN_GTT,
+				      bo,
+				      gpu_addr,
+				      cpu_addr);
+	if (!cpu_addr || ret) {
+		drm_err(&adev->ddev, "failed to alloc gart kernel buffer (%d)", ret);
+		return ret;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL(isp_kernel_buffer_alloc);
+
+/**
+ * isp_kernel_buffer_free - free isp kernel buffer object (BO)
+ *
+ * @buf_obj: amdgpu isp user BO to free
+ * @gpu_addr: GPU addr of isp kernel BO
+ * @cpu_addr: CPU addr of isp kernel BO
+ *
+ * unmaps and unpin a isp kernel BO.
+ *
+ * This function is exported to allow the V4L2 isp device
+ * external to drm device to free the kernel BO.
+ */
+void isp_kernel_buffer_free(void **buf_obj, u64 *gpu_addr, void **cpu_addr)
+{
+	struct amdgpu_bo **bo = (struct amdgpu_bo **)buf_obj;
+
+	amdgpu_bo_free_kernel(bo, gpu_addr, cpu_addr);
+}
+EXPORT_SYMBOL(isp_kernel_buffer_free);
+
+static const struct amd_ip_funcs isp_ip_funcs = {
+	.name = "isp_ip",
+	.early_init = isp_early_init,
+	.hw_init = isp_hw_init,
+	.hw_fini = isp_hw_fini,
+	.is_idle = isp_is_idle,
+	.set_clockgating_state = isp_set_clockgating_state,
+	.set_powergating_state = isp_set_powergating_state,
+};
+
+const struct amdgpu_ip_block_version isp_v4_1_0_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_ISP,
+	.major = 4,
+	.minor = 1,
+	.rev = 0,
+	.funcs = &isp_ip_funcs,
+};
+
+const struct amdgpu_ip_block_version isp_v4_1_1_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_ISP,
+	.major = 4,
+	.minor = 1,
+	.rev = 1,
+	.funcs = &isp_ip_funcs,
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_isp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_isp.h
new file mode 100644
index 000000000000..d6f4ffa4c97c
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_isp.h
@@ -0,0 +1,60 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright (C) 2024 Advanced Micro Devices, Inc. All rights reserved.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+
+#ifndef __AMDGPU_ISP_H__
+#define __AMDGPU_ISP_H__
+
+#include <drm/amd/isp.h>
+#include <linux/pm_domain.h>
+
+#define ISP_REGS_OFFSET_END 0x629A4
+
+struct amdgpu_isp;
+
+struct isp_funcs {
+	int (*hw_init)(struct amdgpu_isp *isp);
+	int (*hw_fini)(struct amdgpu_isp *isp);
+};
+
+struct amdgpu_isp {
+	struct device *parent;
+	struct amdgpu_device	*adev;
+	const struct isp_funcs	*funcs;
+	struct mfd_cell *isp_cell;
+	struct resource *isp_res;
+	struct resource *isp_i2c_res;
+	struct resource *isp_gpio_res;
+	struct isp_platform_data *isp_pdata;
+	unsigned int harvest_config;
+	const struct firmware	*fw;
+	struct generic_pm_domain ispgpd;
+};
+
+extern const struct amdgpu_ip_block_version isp_v4_1_0_ip_block;
+extern const struct amdgpu_ip_block_version isp_v4_1_1_ip_block;
+
+#endif /* __AMDGPU_ISP_H__ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 7570f2439a11..0a0dcbf0798d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -24,169 +24,484 @@
 #include <linux/kthread.h>
 #include <linux/wait.h>
 #include <linux/sched.h>
-#include <drm/drmP.h>
+
+#include <drm/drm_drv.h>
+
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
+#include "amdgpu_reset.h"
+#include "amdgpu_dev_coredump.h"
+#include "amdgpu_xgmi.h"
 
-static void amdgpu_job_timedout(struct amd_sched_job *s_job)
+static void amdgpu_job_do_core_dump(struct amdgpu_device *adev,
+				    struct amdgpu_job *job)
 {
-	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
+	int i;
 
-	DRM_ERROR("ring %s timeout, last signaled seq=%u, last emitted seq=%u\n",
-		  job->base.sched->name,
-		  atomic_read(&job->ring->fence_drv.last_seq),
-		  job->ring->fence_drv.sync_seq);
-	amdgpu_gpu_reset(job->adev);
+	dev_info(adev->dev, "Dumping IP State\n");
+	for (i = 0; i < adev->num_ip_blocks; i++)
+		if (adev->ip_blocks[i].version->funcs->dump_ip_state)
+			adev->ip_blocks[i].version->funcs
+				->dump_ip_state((void *)&adev->ip_blocks[i]);
+	dev_info(adev->dev, "Dumping IP State Completed\n");
+
+	amdgpu_coredump(adev, true, false, job);
 }
 
-int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
-		     struct amdgpu_job **job, struct amdgpu_vm *vm)
+static void amdgpu_job_core_dump(struct amdgpu_device *adev,
+				 struct amdgpu_job *job)
 {
-	size_t size = sizeof(struct amdgpu_job);
+	struct list_head device_list, *device_list_handle =  NULL;
+	struct amdgpu_device *tmp_adev = NULL;
+	struct amdgpu_hive_info *hive = NULL;
+
+	if (!amdgpu_sriov_vf(adev))
+		hive = amdgpu_get_xgmi_hive(adev);
+	if (hive)
+		mutex_lock(&hive->hive_lock);
+	/*
+	 * Reuse the logic in amdgpu_device_gpu_recover() to build list of
+	 * devices for code dump
+	 */
+	INIT_LIST_HEAD(&device_list);
+	if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1) && hive) {
+		list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head)
+			list_add_tail(&tmp_adev->reset_list, &device_list);
+		if (!list_is_first(&adev->reset_list, &device_list))
+			list_rotate_to_front(&adev->reset_list, &device_list);
+		device_list_handle = &device_list;
+	} else {
+		list_add_tail(&adev->reset_list, &device_list);
+		device_list_handle = &device_list;
+	}
+
+	/* Do the coredump for each device */
+	list_for_each_entry(tmp_adev, device_list_handle, reset_list)
+		amdgpu_job_do_core_dump(tmp_adev, job);
+
+	if (hive) {
+		mutex_unlock(&hive->hive_lock);
+		amdgpu_put_xgmi_hive(hive);
+	}
+}
+
+static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
+{
+	struct amdgpu_ring *ring = to_amdgpu_ring(s_job->sched);
+	struct amdgpu_job *job = to_amdgpu_job(s_job);
+	struct drm_wedge_task_info *info = NULL;
+	struct amdgpu_task_info *ti = NULL;
+	struct amdgpu_device *adev = ring->adev;
+	int idx, r;
+
+	if (!drm_dev_enter(adev_to_drm(adev), &idx)) {
+		dev_info(adev->dev, "%s - device unplugged skipping recovery on scheduler:%s",
+			 __func__, s_job->sched->name);
+
+		/* Effectively the job is aborted as the device is gone */
+		return DRM_GPU_SCHED_STAT_ENODEV;
+	}
+
+	/*
+	 * Do the coredump immediately after a job timeout to get a very
+	 * close dump/snapshot/representation of GPU's current error status
+	 * Skip it for SRIOV, since VF FLR will be triggered by host driver
+	 * before job timeout
+	 */
+	if (!amdgpu_sriov_vf(adev))
+		amdgpu_job_core_dump(adev, job);
+
+	if (amdgpu_gpu_recovery &&
+	    amdgpu_ring_is_reset_type_supported(ring, AMDGPU_RESET_TYPE_SOFT_RESET) &&
+	    amdgpu_ring_soft_recovery(ring, job->vmid, s_job->s_fence->parent)) {
+		dev_err(adev->dev, "ring %s timeout, but soft recovered\n",
+			s_job->sched->name);
+		goto exit;
+	}
+
+	dev_err(adev->dev, "ring %s timeout, signaled seq=%u, emitted seq=%u\n",
+		job->base.sched->name, atomic_read(&ring->fence_drv.last_seq),
+		ring->fence_drv.sync_seq);
+
+	ti = amdgpu_vm_get_task_info_pasid(ring->adev, job->pasid);
+	if (ti) {
+		amdgpu_vm_print_task_info(adev, ti);
+		info = &ti->task;
+	}
+
+	/* attempt a per ring reset */
+	if (amdgpu_gpu_recovery &&
+	    amdgpu_ring_is_reset_type_supported(ring, AMDGPU_RESET_TYPE_PER_QUEUE) &&
+	    ring->funcs->reset) {
+		dev_err(adev->dev, "Starting %s ring reset\n",
+			s_job->sched->name);
+		r = amdgpu_ring_reset(ring, job->vmid, job->hw_fence);
+		if (!r) {
+			atomic_inc(&ring->adev->gpu_reset_counter);
+			dev_err(adev->dev, "Ring %s reset succeeded\n",
+				ring->sched.name);
+			drm_dev_wedged_event(adev_to_drm(adev),
+					     DRM_WEDGE_RECOVERY_NONE, info);
+			goto exit;
+		}
+		dev_err(adev->dev, "Ring %s reset failed\n", ring->sched.name);
+	}
+
+	dma_fence_set_error(&s_job->s_fence->finished, -ETIME);
+
+	if (amdgpu_device_should_recover_gpu(ring->adev)) {
+		struct amdgpu_reset_context reset_context;
+		memset(&reset_context, 0, sizeof(reset_context));
+
+		reset_context.method = AMD_RESET_METHOD_NONE;
+		reset_context.reset_req_dev = adev;
+		reset_context.src = AMDGPU_RESET_SRC_JOB;
+		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
+
+		/*
+		 * To avoid an unnecessary extra coredump, as we have already
+		 * got the very close representation of GPU's error status
+		 */
+		set_bit(AMDGPU_SKIP_COREDUMP, &reset_context.flags);
+
+		r = amdgpu_device_gpu_recover(ring->adev, job, &reset_context);
+		if (r)
+			dev_err(adev->dev, "GPU Recovery Failed: %d\n", r);
+	} else {
+		drm_sched_suspend_timeout(&ring->sched);
+		if (amdgpu_sriov_vf(adev))
+			adev->virt.tdr_debug = true;
+	}
+
+exit:
+	amdgpu_vm_put_task_info(ti);
+	drm_dev_exit(idx);
+	return DRM_GPU_SCHED_STAT_RESET;
+}
+
+int amdgpu_job_alloc(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+		     struct drm_sched_entity *entity, void *owner,
+		     unsigned int num_ibs, struct amdgpu_job **job,
+		     u64 drm_client_id)
+{
+	struct amdgpu_fence *af;
+	int r;
 
 	if (num_ibs == 0)
 		return -EINVAL;
 
-	size += sizeof(struct amdgpu_ib) * num_ibs;
-
-	*job = kzalloc(size, GFP_KERNEL);
+	*job = kzalloc(struct_size(*job, ibs, num_ibs), GFP_KERNEL);
 	if (!*job)
 		return -ENOMEM;
 
-	(*job)->adev = adev;
+	af = kzalloc(sizeof(struct amdgpu_fence), GFP_KERNEL);
+	if (!af) {
+		r = -ENOMEM;
+		goto err_job;
+	}
+	(*job)->hw_fence = af;
+
+	af = kzalloc(sizeof(struct amdgpu_fence), GFP_KERNEL);
+	if (!af) {
+		r = -ENOMEM;
+		goto err_fence;
+	}
+	(*job)->hw_vm_fence = af;
+
 	(*job)->vm = vm;
-	(*job)->ibs = (void *)&(*job)[1];
-	(*job)->num_ibs = num_ibs;
-	(*job)->need_pipeline_sync = false;
 
-	amdgpu_sync_create(&(*job)->sync);
+	amdgpu_sync_create(&(*job)->explicit_sync);
+	(*job)->generation = amdgpu_vm_generation(adev, vm);
+	(*job)->vm_pd_addr = AMDGPU_BO_INVALID_OFFSET;
 
-	return 0;
+	if (!entity)
+		return 0;
+
+	return drm_sched_job_init(&(*job)->base, entity, 1, owner,
+				  drm_client_id);
+
+err_fence:
+	kfree((*job)->hw_fence);
+err_job:
+	kfree(*job);
+	*job = NULL;
+
+	return r;
 }
 
-int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
-			     struct amdgpu_job **job)
+int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev,
+			     struct drm_sched_entity *entity, void *owner,
+			     size_t size, enum amdgpu_ib_pool_type pool_type,
+			     struct amdgpu_job **job, u64 k_job_id)
 {
 	int r;
 
-	r = amdgpu_job_alloc(adev, 1, job, NULL);
+	r = amdgpu_job_alloc(adev, NULL, entity, owner, 1, job,
+			     k_job_id);
 	if (r)
 		return r;
 
-	r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]);
-	if (r)
+	(*job)->num_ibs = 1;
+	r = amdgpu_ib_get(adev, NULL, size, pool_type, &(*job)->ibs[0]);
+	if (r) {
+		if (entity)
+			drm_sched_job_cleanup(&(*job)->base);
+		kfree((*job)->hw_vm_fence);
+		kfree((*job)->hw_fence);
 		kfree(*job);
+		*job = NULL;
+	}
 
 	return r;
 }
 
+void amdgpu_job_set_resources(struct amdgpu_job *job, struct amdgpu_bo *gds,
+			      struct amdgpu_bo *gws, struct amdgpu_bo *oa)
+{
+	if (gds) {
+		job->gds_base = amdgpu_bo_gpu_offset(gds) >> PAGE_SHIFT;
+		job->gds_size = amdgpu_bo_size(gds) >> PAGE_SHIFT;
+	}
+	if (gws) {
+		job->gws_base = amdgpu_bo_gpu_offset(gws) >> PAGE_SHIFT;
+		job->gws_size = amdgpu_bo_size(gws) >> PAGE_SHIFT;
+	}
+	if (oa) {
+		job->oa_base = amdgpu_bo_gpu_offset(oa) >> PAGE_SHIFT;
+		job->oa_size = amdgpu_bo_size(oa) >> PAGE_SHIFT;
+	}
+}
+
 void amdgpu_job_free_resources(struct amdgpu_job *job)
 {
 	struct dma_fence *f;
 	unsigned i;
 
-	/* use sched fence if available */
-	f = job->base.s_fence ? &job->base.s_fence->finished : job->fence;
+	/* Check if any fences were initialized */
+	if (job->base.s_fence && job->base.s_fence->finished.ops)
+		f = &job->base.s_fence->finished;
+	else if (job->hw_fence && job->hw_fence->base.ops)
+		f = &job->hw_fence->base;
+	else
+		f = NULL;
 
 	for (i = 0; i < job->num_ibs; ++i)
-		amdgpu_ib_free(job->adev, &job->ibs[i], f);
+		amdgpu_ib_free(&job->ibs[i], f);
 }
 
-static void amdgpu_job_free_cb(struct amd_sched_job *s_job)
+static void amdgpu_job_free_cb(struct drm_sched_job *s_job)
 {
-	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
+	struct amdgpu_job *job = to_amdgpu_job(s_job);
+
+	drm_sched_job_cleanup(s_job);
+
+	amdgpu_sync_free(&job->explicit_sync);
+
+	if (job->hw_fence->base.ops)
+		dma_fence_put(&job->hw_fence->base);
+	else
+		kfree(job->hw_fence);
+	if (job->hw_vm_fence->base.ops)
+		dma_fence_put(&job->hw_vm_fence->base);
+	else
+		kfree(job->hw_vm_fence);
 
-	dma_fence_put(job->fence);
-	amdgpu_sync_free(&job->sync);
 	kfree(job);
 }
 
+void amdgpu_job_set_gang_leader(struct amdgpu_job *job,
+				struct amdgpu_job *leader)
+{
+	struct dma_fence *fence = &leader->base.s_fence->scheduled;
+
+	WARN_ON(job->gang_submit);
+
+	/*
+	 * Don't add a reference when we are the gang leader to avoid circle
+	 * dependency.
+	 */
+	if (job != leader)
+		dma_fence_get(fence);
+	job->gang_submit = fence;
+}
+
 void amdgpu_job_free(struct amdgpu_job *job)
 {
+	if (job->base.entity)
+		drm_sched_job_cleanup(&job->base);
+
 	amdgpu_job_free_resources(job);
+	amdgpu_sync_free(&job->explicit_sync);
+	if (job->gang_submit != &job->base.s_fence->scheduled)
+		dma_fence_put(job->gang_submit);
+
+	if (job->hw_fence->base.ops)
+		dma_fence_put(&job->hw_fence->base);
+	else
+		kfree(job->hw_fence);
+	if (job->hw_vm_fence->base.ops)
+		dma_fence_put(&job->hw_vm_fence->base);
+	else
+		kfree(job->hw_vm_fence);
 
-	dma_fence_put(job->fence);
-	amdgpu_sync_free(&job->sync);
 	kfree(job);
 }
 
-int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
-		      struct amd_sched_entity *entity, void *owner,
-		      struct dma_fence **f)
+struct dma_fence *amdgpu_job_submit(struct amdgpu_job *job)
+{
+	struct dma_fence *f;
+
+	drm_sched_job_arm(&job->base);
+	f = dma_fence_get(&job->base.s_fence->finished);
+	amdgpu_job_free_resources(job);
+	drm_sched_entity_push_job(&job->base);
+
+	return f;
+}
+
+int amdgpu_job_submit_direct(struct amdgpu_job *job, struct amdgpu_ring *ring,
+			     struct dma_fence **fence)
 {
 	int r;
-	job->ring = ring;
 
-	if (!f)
-		return -EINVAL;
+	job->base.sched = &ring->sched;
+	r = amdgpu_ib_schedule(ring, job->num_ibs, job->ibs, job, fence);
 
-	r = amd_sched_job_init(&job->base, &ring->sched, entity, owner);
 	if (r)
 		return r;
 
-	job->owner = owner;
-	job->fence_ctx = entity->fence_context;
-	*f = dma_fence_get(&job->base.s_fence->finished);
-	amdgpu_job_free_resources(job);
-	amd_sched_entity_push_job(&job->base);
-
+	amdgpu_job_free(job);
 	return 0;
 }
 
-static struct dma_fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
+static struct dma_fence *
+amdgpu_job_prepare_job(struct drm_sched_job *sched_job,
+		      struct drm_sched_entity *s_entity)
 {
+	struct amdgpu_ring *ring = to_amdgpu_ring(s_entity->rq->sched);
 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
-	struct amdgpu_vm *vm = job->vm;
-
-	struct dma_fence *fence = amdgpu_sync_get_fence(&job->sync);
+	struct dma_fence *fence;
+	int r;
 
-	while (fence == NULL && vm && !job->vm_id) {
-		struct amdgpu_ring *ring = job->ring;
-		int r;
+	r = drm_sched_entity_error(s_entity);
+	if (r)
+		goto error;
 
-		r = amdgpu_vm_grab_id(vm, ring, &job->sync,
-				      &job->base.s_fence->finished,
-				      job);
-		if (r)
-			DRM_ERROR("Error getting VM ID (%d)\n", r);
+	if (job->gang_submit) {
+		fence = amdgpu_device_switch_gang(ring->adev, job->gang_submit);
+		if (fence)
+			return fence;
+	}
 
-		fence = amdgpu_sync_get_fence(&job->sync);
+	fence = amdgpu_device_enforce_isolation(ring->adev, ring, job);
+	if (fence)
+		return fence;
+
+	if (job->vm && !job->vmid) {
+		r = amdgpu_vmid_grab(job->vm, ring, job, &fence);
+		if (r) {
+			dev_err(ring->adev->dev, "Error getting VM ID (%d)\n", r);
+			goto error;
+		}
+		return fence;
 	}
 
-	if (amd_sched_dependency_optimized(fence, sched_job->s_entity))
-		job->need_pipeline_sync = true;
+	return NULL;
 
-	return fence;
+error:
+	dma_fence_set_error(&job->base.s_fence->finished, r);
+	return NULL;
 }
 
-static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job)
+static struct dma_fence *amdgpu_job_run(struct drm_sched_job *sched_job)
 {
-	struct dma_fence *fence = NULL;
+	struct amdgpu_ring *ring = to_amdgpu_ring(sched_job->sched);
+	struct amdgpu_device *adev = ring->adev;
+	struct dma_fence *fence = NULL, *finished;
 	struct amdgpu_job *job;
-	int r;
+	int r = 0;
 
-	if (!sched_job) {
-		DRM_ERROR("job is null\n");
-		return NULL;
-	}
 	job = to_amdgpu_job(sched_job);
-
-	BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
+	finished = &job->base.s_fence->finished;
 
 	trace_amdgpu_sched_run_job(job);
-	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence);
-	if (r)
-		DRM_ERROR("Error scheduling IBs (%d)\n", r);
 
-	/* if gpu reset, hw fence will be replaced here */
-	dma_fence_put(job->fence);
-	job->fence = dma_fence_get(fence);
+	/* Skip job if VRAM is lost and never resubmit gangs */
+	if (job->generation != amdgpu_vm_generation(adev, job->vm) ||
+	    (job->job_run_counter && job->gang_submit))
+		dma_fence_set_error(finished, -ECANCELED);
+
+	if (finished->error < 0) {
+		dev_dbg(adev->dev, "Skip scheduling IBs in ring(%s)",
+			ring->name);
+	} else {
+		r = amdgpu_ib_schedule(ring, job->num_ibs, job->ibs, job,
+				       &fence);
+		if (r)
+			dev_err(adev->dev,
+				"Error scheduling IBs (%d) in ring(%s)", r,
+				ring->name);
+	}
+
+	job->job_run_counter++;
 	amdgpu_job_free_resources(job);
+
+	fence = r ? ERR_PTR(r) : fence;
 	return fence;
 }
 
-const struct amd_sched_backend_ops amdgpu_sched_ops = {
-	.dependency = amdgpu_job_dependency,
+/*
+ * This is a duplicate function from DRM scheduler sched_internal.h.
+ * Plan is to remove it when amdgpu_job_stop_all_jobs_on_sched is removed, due
+ * latter being incorrect and racy.
+ *
+ * See https://lore.kernel.org/amd-gfx/44edde63-7181-44fb-a4f7-94e50514f539@amd.com/
+ */
+static struct drm_sched_job *
+drm_sched_entity_queue_pop(struct drm_sched_entity *entity)
+{
+	struct spsc_node *node;
+
+	node = spsc_queue_pop(&entity->job_queue);
+	if (!node)
+		return NULL;
+
+	return container_of(node, struct drm_sched_job, queue_node);
+}
+
+void amdgpu_job_stop_all_jobs_on_sched(struct drm_gpu_scheduler *sched)
+{
+	struct drm_sched_job *s_job;
+	struct drm_sched_entity *s_entity = NULL;
+	int i;
+
+	/* Signal all jobs not yet scheduled */
+	for (i = DRM_SCHED_PRIORITY_KERNEL; i < sched->num_rqs; i++) {
+		struct drm_sched_rq *rq = sched->sched_rq[i];
+		spin_lock(&rq->lock);
+		list_for_each_entry(s_entity, &rq->entities, list) {
+			while ((s_job = drm_sched_entity_queue_pop(s_entity))) {
+				struct drm_sched_fence *s_fence = s_job->s_fence;
+
+				dma_fence_signal(&s_fence->scheduled);
+				dma_fence_set_error(&s_fence->finished, -EHWPOISON);
+				dma_fence_signal(&s_fence->finished);
+			}
+		}
+		spin_unlock(&rq->lock);
+	}
+
+	/* Signal all jobs already scheduled to HW */
+	list_for_each_entry(s_job, &sched->pending_list, list) {
+		struct drm_sched_fence *s_fence = s_job->s_fence;
+
+		dma_fence_set_error(&s_fence->finished, -EHWPOISON);
+		dma_fence_signal(&s_fence->finished);
+	}
+}
+
+const struct drm_sched_backend_ops amdgpu_sched_ops = {
+	.prepare_job = amdgpu_job_prepare_job,
 	.run_job = amdgpu_job_run,
 	.timedout_job = amdgpu_job_timedout,
 	.free_job = amdgpu_job_free_cb
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
new file mode 100644
index 000000000000..7abf069d17d4
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.h
@@ -0,0 +1,130 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_JOB_H__
+#define __AMDGPU_JOB_H__
+
+#include <drm/gpu_scheduler.h>
+#include "amdgpu_sync.h"
+#include "amdgpu_ring.h"
+
+/* bit set means command submit involves a preamble IB */
+#define AMDGPU_PREAMBLE_IB_PRESENT          (1 << 0)
+/* bit set means preamble IB is first presented in belonging context */
+#define AMDGPU_PREAMBLE_IB_PRESENT_FIRST    (1 << 1)
+/* bit set means context switch occured */
+#define AMDGPU_HAVE_CTX_SWITCH              (1 << 2)
+/* bit set means IB is preempted */
+#define AMDGPU_IB_PREEMPTED                 (1 << 3)
+
+#define to_amdgpu_job(sched_job)		\
+		container_of((sched_job), struct amdgpu_job, base)
+
+#define AMDGPU_JOB_GET_VMID(job) ((job) ? (job)->vmid : 0)
+
+struct amdgpu_fence;
+enum amdgpu_ib_pool_type;
+
+/* Internal kernel job ids. (decreasing values, starting from U64_MAX). */
+#define AMDGPU_KERNEL_JOB_ID_VM_UPDATE              (18446744073709551615ULL)
+#define AMDGPU_KERNEL_JOB_ID_VM_UPDATE_PDES         (18446744073709551614ULL)
+#define AMDGPU_KERNEL_JOB_ID_VM_UPDATE_RANGE        (18446744073709551613ULL)
+#define AMDGPU_KERNEL_JOB_ID_VM_PT_CLEAR            (18446744073709551612ULL)
+#define AMDGPU_KERNEL_JOB_ID_TTM_MAP_BUFFER         (18446744073709551611ULL)
+#define AMDGPU_KERNEL_JOB_ID_TTM_ACCESS_MEMORY_SDMA (18446744073709551610ULL)
+#define AMDGPU_KERNEL_JOB_ID_TTM_COPY_BUFFER        (18446744073709551609ULL)
+#define AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE       (18446744073709551608ULL)
+#define AMDGPU_KERNEL_JOB_ID_MOVE_BLIT              (18446744073709551607ULL)
+#define AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER       (18446744073709551606ULL)
+#define AMDGPU_KERNEL_JOB_ID_CLEANER_SHADER         (18446744073709551605ULL)
+#define AMDGPU_KERNEL_JOB_ID_FLUSH_GPU_TLB          (18446744073709551604ULL)
+#define AMDGPU_KERNEL_JOB_ID_KFD_GART_MAP           (18446744073709551603ULL)
+#define AMDGPU_KERNEL_JOB_ID_VCN_RING_TEST          (18446744073709551602ULL)
+
+struct amdgpu_job {
+	struct drm_sched_job    base;
+	struct amdgpu_vm	*vm;
+	struct amdgpu_sync	explicit_sync;
+	struct amdgpu_fence	*hw_fence;
+	struct amdgpu_fence	*hw_vm_fence;
+	struct dma_fence	*gang_submit;
+	uint32_t		preamble_status;
+	uint32_t                preemption_status;
+	bool                    vm_needs_flush;
+	bool			gds_switch_needed;
+	bool			spm_update_needed;
+	uint64_t		vm_pd_addr;
+	unsigned		vmid;
+	unsigned		pasid;
+	uint32_t		gds_base, gds_size;
+	uint32_t		gws_base, gws_size;
+	uint32_t		oa_base, oa_size;
+	uint64_t		generation;
+
+	/* user fence handling */
+	uint64_t		uf_addr;
+	uint64_t		uf_sequence;
+
+	/* virtual addresses for shadow/GDS/CSA */
+	uint64_t		shadow_va;
+	uint64_t		csa_va;
+	uint64_t		gds_va;
+	bool			init_shadow;
+
+	/* job_run_counter >= 1 means a resubmit job */
+	uint32_t		job_run_counter;
+
+	/* enforce isolation */
+	bool			enforce_isolation;
+	bool			run_cleaner_shader;
+
+	uint32_t		num_ibs;
+	struct amdgpu_ib	ibs[];
+};
+
+static inline struct amdgpu_ring *amdgpu_job_ring(struct amdgpu_job *job)
+{
+	return to_amdgpu_ring(job->base.entity->rq->sched);
+}
+
+int amdgpu_job_alloc(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+		     struct drm_sched_entity *entity, void *owner,
+		     unsigned int num_ibs, struct amdgpu_job **job,
+		     u64 drm_client_id);
+int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev,
+			     struct drm_sched_entity *entity, void *owner,
+			     size_t size, enum amdgpu_ib_pool_type pool_type,
+			     struct amdgpu_job **job,
+			     u64 k_job_id);
+void amdgpu_job_set_resources(struct amdgpu_job *job, struct amdgpu_bo *gds,
+			      struct amdgpu_bo *gws, struct amdgpu_bo *oa);
+void amdgpu_job_free_resources(struct amdgpu_job *job);
+void amdgpu_job_set_gang_leader(struct amdgpu_job *job,
+				struct amdgpu_job *leader);
+void amdgpu_job_free(struct amdgpu_job *job);
+struct dma_fence *amdgpu_job_submit(struct amdgpu_job *job);
+int amdgpu_job_submit_direct(struct amdgpu_job *job, struct amdgpu_ring *ring,
+			     struct dma_fence **fence);
+
+void amdgpu_job_stop_all_jobs_on_sched(struct drm_gpu_scheduler *sched);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
new file mode 100644
index 000000000000..63ee6ba6a931
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c
@@ -0,0 +1,607 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+
+#include "amdgpu.h"
+#include "amdgpu_jpeg.h"
+#include "amdgpu_pm.h"
+#include "soc15d.h"
+#include "soc15_common.h"
+
+#define JPEG_IDLE_TIMEOUT	msecs_to_jiffies(1000)
+
+static void amdgpu_jpeg_idle_work_handler(struct work_struct *work);
+static void amdgpu_jpeg_reg_dump_fini(struct amdgpu_device *adev);
+
+int amdgpu_jpeg_sw_init(struct amdgpu_device *adev)
+{
+	int i, r;
+
+	INIT_DELAYED_WORK(&adev->jpeg.idle_work, amdgpu_jpeg_idle_work_handler);
+	mutex_init(&adev->jpeg.jpeg_pg_lock);
+	atomic_set(&adev->jpeg.total_submission_cnt, 0);
+
+	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
+	    (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG))
+		adev->jpeg.indirect_sram = true;
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; i++) {
+		if (adev->jpeg.harvest_config & (1U << i))
+			continue;
+
+		if (adev->jpeg.indirect_sram) {
+			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
+					AMDGPU_GEM_DOMAIN_VRAM |
+					AMDGPU_GEM_DOMAIN_GTT,
+					&adev->jpeg.inst[i].dpg_sram_bo,
+					&adev->jpeg.inst[i].dpg_sram_gpu_addr,
+					&adev->jpeg.inst[i].dpg_sram_cpu_addr);
+			if (r) {
+				dev_err(adev->dev,
+				"JPEG %d (%d) failed to allocate DPG bo\n", i, r);
+				return r;
+			}
+		}
+	}
+
+	return 0;
+}
+
+int amdgpu_jpeg_sw_fini(struct amdgpu_device *adev)
+{
+	int i, j;
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+		if (adev->jpeg.harvest_config & (1U << i))
+			continue;
+
+		amdgpu_bo_free_kernel(
+			&adev->jpeg.inst[i].dpg_sram_bo,
+			&adev->jpeg.inst[i].dpg_sram_gpu_addr,
+			(void **)&adev->jpeg.inst[i].dpg_sram_cpu_addr);
+
+		for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j)
+			amdgpu_ring_fini(&adev->jpeg.inst[i].ring_dec[j]);
+	}
+
+	if (adev->jpeg.reg_list)
+		amdgpu_jpeg_reg_dump_fini(adev);
+
+	mutex_destroy(&adev->jpeg.jpeg_pg_lock);
+
+	return 0;
+}
+
+int amdgpu_jpeg_suspend(struct amdgpu_device *adev)
+{
+	cancel_delayed_work_sync(&adev->jpeg.idle_work);
+
+	return 0;
+}
+
+int amdgpu_jpeg_resume(struct amdgpu_device *adev)
+{
+	return 0;
+}
+
+static void amdgpu_jpeg_idle_work_handler(struct work_struct *work)
+{
+	struct amdgpu_device *adev =
+		container_of(work, struct amdgpu_device, jpeg.idle_work.work);
+	unsigned int fences = 0;
+	unsigned int i, j;
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+		if (adev->jpeg.harvest_config & (1U << i))
+			continue;
+
+		for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j)
+			fences += amdgpu_fence_count_emitted(&adev->jpeg.inst[i].ring_dec[j]);
+	}
+
+	if (!fences && !atomic_read(&adev->jpeg.total_submission_cnt)) {
+		mutex_lock(&adev->jpeg.jpeg_pg_lock);
+		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_JPEG,
+						       AMD_PG_STATE_GATE);
+		mutex_unlock(&adev->jpeg.jpeg_pg_lock);
+	} else
+		schedule_delayed_work(&adev->jpeg.idle_work, JPEG_IDLE_TIMEOUT);
+}
+
+void amdgpu_jpeg_ring_begin_use(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+
+	atomic_inc(&adev->jpeg.total_submission_cnt);
+	cancel_delayed_work_sync(&adev->jpeg.idle_work);
+
+	mutex_lock(&adev->jpeg.jpeg_pg_lock);
+	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_JPEG,
+						       AMD_PG_STATE_UNGATE);
+	mutex_unlock(&adev->jpeg.jpeg_pg_lock);
+}
+
+void amdgpu_jpeg_ring_end_use(struct amdgpu_ring *ring)
+{
+	atomic_dec(&ring->adev->jpeg.total_submission_cnt);
+	schedule_delayed_work(&ring->adev->jpeg.idle_work, JPEG_IDLE_TIMEOUT);
+}
+
+int amdgpu_jpeg_dec_ring_test_ring(struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = ring->adev;
+	uint32_t tmp = 0;
+	unsigned i;
+	int r;
+
+	/* JPEG in SRIOV does not support direct register read/write */
+	if (amdgpu_sriov_vf(adev))
+		return 0;
+
+	r = amdgpu_ring_alloc(ring, 3);
+	if (r)
+		return r;
+
+	WREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe], 0xCAFEDEAD);
+	/* Add a read register to make sure the write register is executed. */
+	RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe]);
+
+	amdgpu_ring_write(ring, PACKET0(adev->jpeg.internal.jpeg_pitch[ring->pipe], 0));
+	amdgpu_ring_write(ring, 0xABADCAFE);
+	amdgpu_ring_commit(ring);
+
+	for (i = 0; i < adev->usec_timeout; i++) {
+		tmp = RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe]);
+		if (tmp == 0xABADCAFE)
+			break;
+		udelay(1);
+	}
+
+	if (i >= adev->usec_timeout)
+		r = -ETIMEDOUT;
+
+	return r;
+}
+
+static int amdgpu_jpeg_dec_set_reg(struct amdgpu_ring *ring, uint32_t handle,
+		struct dma_fence **fence)
+{
+	struct amdgpu_device *adev = ring->adev;
+	struct amdgpu_job *job;
+	struct amdgpu_ib *ib;
+	struct dma_fence *f = NULL;
+	const unsigned ib_size_dw = 16;
+	int i, r;
+
+	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL, ib_size_dw * 4,
+				     AMDGPU_IB_POOL_DIRECT, &job,
+				     AMDGPU_KERNEL_JOB_ID_VCN_RING_TEST);
+	if (r)
+		return r;
+
+	ib = &job->ibs[0];
+
+	ib->ptr[0] = PACKETJ(adev->jpeg.internal.jpeg_pitch[ring->pipe], 0, 0, PACKETJ_TYPE0);
+	ib->ptr[1] = 0xDEADBEEF;
+	for (i = 2; i < 16; i += 2) {
+		ib->ptr[i] = PACKETJ(0, 0, 0, PACKETJ_TYPE6);
+		ib->ptr[i+1] = 0;
+	}
+	ib->length_dw = 16;
+
+	r = amdgpu_job_submit_direct(job, ring, &f);
+	if (r)
+		goto err;
+
+	if (fence)
+		*fence = dma_fence_get(f);
+	dma_fence_put(f);
+
+	return 0;
+
+err:
+	amdgpu_job_free(job);
+	return r;
+}
+
+int amdgpu_jpeg_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
+{
+	struct amdgpu_device *adev = ring->adev;
+	uint32_t tmp = 0;
+	unsigned i;
+	struct dma_fence *fence = NULL;
+	long r = 0;
+
+	r = amdgpu_jpeg_dec_set_reg(ring, 1, &fence);
+	if (r)
+		goto error;
+
+	r = dma_fence_wait_timeout(fence, false, timeout);
+	if (r == 0) {
+		r = -ETIMEDOUT;
+		goto error;
+	} else if (r < 0) {
+		goto error;
+	} else {
+		r = 0;
+	}
+
+	if (!amdgpu_sriov_vf(adev)) {
+		for (i = 0; i < adev->usec_timeout; i++) {
+			tmp = RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe]);
+			if (tmp == 0xDEADBEEF)
+				break;
+			udelay(1);
+			if (amdgpu_emu_mode == 1)
+				udelay(10);
+		}
+
+		if (i >= adev->usec_timeout)
+			r = -ETIMEDOUT;
+	}
+
+	dma_fence_put(fence);
+error:
+	return r;
+}
+
+int amdgpu_jpeg_process_poison_irq(struct amdgpu_device *adev,
+				struct amdgpu_irq_src *source,
+				struct amdgpu_iv_entry *entry)
+{
+	struct ras_common_if *ras_if = adev->jpeg.ras_if;
+	struct ras_dispatch_if ih_data = {
+		.entry = entry,
+	};
+
+	if (!ras_if)
+		return 0;
+
+	ih_data.head = *ras_if;
+	amdgpu_ras_interrupt_dispatch(adev, &ih_data);
+
+	return 0;
+}
+
+int amdgpu_jpeg_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
+{
+	int r, i;
+
+	r = amdgpu_ras_block_late_init(adev, ras_block);
+	if (r)
+		return r;
+
+	if (amdgpu_ras_is_supported(adev, ras_block->block)) {
+		for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+			if (adev->jpeg.harvest_config & (1 << i) ||
+			    !adev->jpeg.inst[i].ras_poison_irq.funcs)
+				continue;
+
+			r = amdgpu_irq_get(adev, &adev->jpeg.inst[i].ras_poison_irq, 0);
+			if (r)
+				goto late_fini;
+		}
+	}
+	return 0;
+
+late_fini:
+	amdgpu_ras_block_late_fini(adev, ras_block);
+	return r;
+}
+
+int amdgpu_jpeg_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err;
+	struct amdgpu_jpeg_ras *ras;
+
+	if (!adev->jpeg.ras)
+		return 0;
+
+	ras = adev->jpeg.ras;
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register jpeg ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "jpeg");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__JPEG;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
+	adev->jpeg.ras_if = &ras->ras_block.ras_comm;
+
+	if (!ras->ras_block.ras_late_init)
+		ras->ras_block.ras_late_init = amdgpu_jpeg_ras_late_init;
+
+	return 0;
+}
+
+int amdgpu_jpeg_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
+			       enum AMDGPU_UCODE_ID ucode_id)
+{
+	struct amdgpu_firmware_info ucode = {
+		.ucode_id = AMDGPU_UCODE_ID_JPEG_RAM,
+		.mc_addr = adev->jpeg.inst[inst_idx].dpg_sram_gpu_addr,
+		.ucode_size = ((uintptr_t)adev->jpeg.inst[inst_idx].dpg_sram_curr_addr -
+			      (uintptr_t)adev->jpeg.inst[inst_idx].dpg_sram_cpu_addr),
+	};
+
+	return psp_execute_ip_fw_load(&adev->psp, &ucode);
+}
+
+/*
+ * debugfs for to enable/disable jpeg job submission to specific core.
+ */
+#if defined(CONFIG_DEBUG_FS)
+static int amdgpu_debugfs_jpeg_sched_mask_set(void *data, u64 val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	u32 i, j;
+	u64 mask = 0;
+	struct amdgpu_ring *ring;
+
+	if (!adev)
+		return -ENODEV;
+
+	mask = (1ULL << (adev->jpeg.num_jpeg_inst * adev->jpeg.num_jpeg_rings)) - 1;
+	if ((val & mask) == 0)
+		return -EINVAL;
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+		for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j) {
+			ring = &adev->jpeg.inst[i].ring_dec[j];
+			if (val & (BIT_ULL((i * adev->jpeg.num_jpeg_rings) + j)))
+				ring->sched.ready = true;
+			else
+				ring->sched.ready = false;
+		}
+	}
+	/* publish sched.ready flag update effective immediately across smp */
+	smp_rmb();
+	return 0;
+}
+
+static int amdgpu_debugfs_jpeg_sched_mask_get(void *data, u64 *val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	u32 i, j;
+	u64 mask = 0;
+	struct amdgpu_ring *ring;
+
+	if (!adev)
+		return -ENODEV;
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+		for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j) {
+			ring = &adev->jpeg.inst[i].ring_dec[j];
+			if (ring->sched.ready)
+				mask |= 1ULL << ((i * adev->jpeg.num_jpeg_rings) + j);
+		}
+	}
+	*val = mask;
+	return 0;
+}
+
+DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_jpeg_sched_mask_fops,
+			 amdgpu_debugfs_jpeg_sched_mask_get,
+			 amdgpu_debugfs_jpeg_sched_mask_set, "%llx\n");
+
+#endif
+
+void amdgpu_debugfs_jpeg_sched_mask_init(struct amdgpu_device *adev)
+{
+#if defined(CONFIG_DEBUG_FS)
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+	char name[32];
+
+	if (!(adev->jpeg.num_jpeg_inst > 1) && !(adev->jpeg.num_jpeg_rings > 1))
+		return;
+	sprintf(name, "amdgpu_jpeg_sched_mask");
+	debugfs_create_file(name, 0600, root, adev,
+			    &amdgpu_debugfs_jpeg_sched_mask_fops);
+#endif
+}
+
+static ssize_t amdgpu_get_jpeg_reset_mask(struct device *dev,
+						struct device_attribute *attr,
+						char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	if (!adev)
+		return -ENODEV;
+
+	return amdgpu_show_reset_mask(buf, adev->jpeg.supported_reset);
+}
+
+static DEVICE_ATTR(jpeg_reset_mask, 0444,
+		   amdgpu_get_jpeg_reset_mask, NULL);
+
+int amdgpu_jpeg_sysfs_reset_mask_init(struct amdgpu_device *adev)
+{
+	int r = 0;
+
+	if (adev->jpeg.num_jpeg_inst) {
+		r = device_create_file(adev->dev, &dev_attr_jpeg_reset_mask);
+		if (r)
+			return r;
+	}
+
+	return r;
+}
+
+void amdgpu_jpeg_sysfs_reset_mask_fini(struct amdgpu_device *adev)
+{
+	if (adev->dev->kobj.sd) {
+		if (adev->jpeg.num_jpeg_inst)
+			device_remove_file(adev->dev, &dev_attr_jpeg_reset_mask);
+	}
+}
+
+int amdgpu_jpeg_reg_dump_init(struct amdgpu_device *adev,
+			       const struct amdgpu_hwip_reg_entry *reg, u32 count)
+{
+	adev->jpeg.ip_dump = kcalloc(adev->jpeg.num_jpeg_inst * count,
+				     sizeof(uint32_t), GFP_KERNEL);
+	if (!adev->jpeg.ip_dump) {
+		dev_err(adev->dev,
+			"Failed to allocate memory for JPEG IP Dump\n");
+		return -ENOMEM;
+	}
+	adev->jpeg.reg_list = reg;
+	adev->jpeg.reg_count = count;
+
+	return 0;
+}
+
+static void amdgpu_jpeg_reg_dump_fini(struct amdgpu_device *adev)
+{
+	kfree(adev->jpeg.ip_dump);
+	adev->jpeg.reg_list = NULL;
+	adev->jpeg.reg_count = 0;
+}
+
+void amdgpu_jpeg_dump_ip_state(struct amdgpu_ip_block *ip_block)
+{
+	struct amdgpu_device *adev = ip_block->adev;
+	u32 inst_off, inst_id, is_powered;
+	int i, j;
+
+	if (!adev->jpeg.ip_dump)
+		return;
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; i++) {
+		if (adev->jpeg.harvest_config & (1 << i))
+			continue;
+
+		inst_id = GET_INST(JPEG, i);
+		inst_off = i * adev->jpeg.reg_count;
+		/* check power status from UVD_JPEG_POWER_STATUS */
+		adev->jpeg.ip_dump[inst_off] =
+			RREG32(SOC15_REG_ENTRY_OFFSET_INST(adev->jpeg.reg_list[0],
+							   inst_id));
+		is_powered = ((adev->jpeg.ip_dump[inst_off] & 0x1) != 1);
+
+		if (is_powered)
+			for (j = 1; j < adev->jpeg.reg_count; j++)
+				adev->jpeg.ip_dump[inst_off + j] =
+					RREG32(SOC15_REG_ENTRY_OFFSET_INST(adev->jpeg.reg_list[j],
+									   inst_id));
+	}
+}
+
+void amdgpu_jpeg_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p)
+{
+	struct amdgpu_device *adev = ip_block->adev;
+	u32 inst_off, is_powered;
+	int i, j;
+
+	if (!adev->jpeg.ip_dump)
+		return;
+
+	drm_printf(p, "num_instances:%d\n", adev->jpeg.num_jpeg_inst);
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; i++) {
+		if (adev->jpeg.harvest_config & (1 << i)) {
+			drm_printf(p, "\nHarvested Instance:JPEG%d Skipping dump\n", i);
+			continue;
+		}
+
+		inst_off = i * adev->jpeg.reg_count;
+		is_powered = ((adev->jpeg.ip_dump[inst_off] & 0x1) != 1);
+
+		if (is_powered) {
+			drm_printf(p, "Active Instance:JPEG%d\n", i);
+			for (j = 0; j < adev->jpeg.reg_count; j++)
+				drm_printf(p, "%-50s \t 0x%08x\n", adev->jpeg.reg_list[j].reg_name,
+					   adev->jpeg.ip_dump[inst_off + j]);
+		} else
+			drm_printf(p, "\nInactive Instance:JPEG%d\n", i);
+	}
+}
+
+static inline bool amdgpu_jpeg_reg_valid(u32 reg)
+{
+	if (reg < JPEG_REG_RANGE_START || reg > JPEG_REG_RANGE_END ||
+	    (reg >= JPEG_ATOMIC_RANGE_START && reg <= JPEG_ATOMIC_RANGE_END))
+		return false;
+	else
+		return true;
+}
+
+/**
+ * amdgpu_jpeg_dec_parse_cs - command submission parser
+ *
+ * @parser: Command submission parser context
+ * @job: the job to parse
+ * @ib: the IB to parse
+ *
+ * Parse the command stream, return -EINVAL for invalid packet,
+ * 0 otherwise
+ */
+
+int amdgpu_jpeg_dec_parse_cs(struct amdgpu_cs_parser *parser,
+			      struct amdgpu_job *job,
+			      struct amdgpu_ib *ib)
+{
+	u32 i, reg, res, cond, type;
+	struct amdgpu_device *adev = parser->adev;
+
+	for (i = 0; i < ib->length_dw ; i += 2) {
+		reg  = CP_PACKETJ_GET_REG(ib->ptr[i]);
+		res  = CP_PACKETJ_GET_RES(ib->ptr[i]);
+		cond = CP_PACKETJ_GET_COND(ib->ptr[i]);
+		type = CP_PACKETJ_GET_TYPE(ib->ptr[i]);
+
+		if (res) /* only support 0 at the moment */
+			return -EINVAL;
+
+		switch (type) {
+		case PACKETJ_TYPE0:
+			if (cond != PACKETJ_CONDITION_CHECK0 ||
+			    !amdgpu_jpeg_reg_valid(reg)) {
+				dev_err(adev->dev, "Invalid packet [0x%08x]!\n", ib->ptr[i]);
+				return -EINVAL;
+			}
+			break;
+		case PACKETJ_TYPE3:
+			if (cond != PACKETJ_CONDITION_CHECK3 ||
+			    !amdgpu_jpeg_reg_valid(reg)) {
+				dev_err(adev->dev, "Invalid packet [0x%08x]!\n", ib->ptr[i]);
+				return -EINVAL;
+			}
+			break;
+		case PACKETJ_TYPE6:
+			if (ib->ptr[i] == CP_PACKETJ_NOP)
+				continue;
+			dev_err(adev->dev, "Invalid packet [0x%08x]!\n", ib->ptr[i]);
+			return -EINVAL;
+		default:
+			dev_err(adev->dev, "Unknown packet type %d !\n", type);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
new file mode 100644
index 000000000000..346ae0ab09d3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
@@ -0,0 +1,184 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_JPEG_H__
+#define __AMDGPU_JPEG_H__
+
+#include "amdgpu_ras.h"
+#include "amdgpu_cs.h"
+
+#define AMDGPU_MAX_JPEG_INSTANCES	4
+#define AMDGPU_MAX_JPEG_RINGS           10
+#define AMDGPU_MAX_JPEG_RINGS_4_0_3     8
+
+#define JPEG_REG_RANGE_START            0x4000
+#define JPEG_REG_RANGE_END              0x41c2
+#define JPEG_ATOMIC_RANGE_START         0x4120
+#define JPEG_ATOMIC_RANGE_END           0x412A
+
+
+#define AMDGPU_JPEG_HARVEST_JPEG0 (1 << 0)
+#define AMDGPU_JPEG_HARVEST_JPEG1 (1 << 1)
+
+#define WREG32_SOC15_JPEG_DPG_MODE(inst_idx, offset, value, indirect)			\
+	do {										\
+		if (!indirect) {							\
+			WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
+				     mmUVD_DPG_LMA_DATA, value);			\
+			WREG32_SOC15(							\
+				JPEG, GET_INST(JPEG, inst_idx),				\
+				mmUVD_DPG_LMA_CTL,					\
+				(UVD_DPG_LMA_CTL__READ_WRITE_MASK |			\
+				 offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT |	\
+				 indirect << UVD_DPG_LMA_CTL__SRAM_SEL__SHIFT));	\
+		} else {								\
+			*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ =		\
+				offset;							\
+			*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ =		\
+				value;							\
+		}									\
+	} while (0)
+
+#define RREG32_SOC15_JPEG_DPG_MODE(inst_idx, offset, mask_en)					\
+	({											\
+		WREG32_SOC15(JPEG, inst_idx, mmUVD_DPG_LMA_CTL,					\
+			(0x0 << UVD_DPG_LMA_CTL__READ_WRITE__SHIFT |				\
+			mask_en << UVD_DPG_LMA_CTL__MASK_EN__SHIFT |				\
+			offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT));			\
+		RREG32_SOC15(JPEG, inst_idx, mmUVD_DPG_LMA_DATA);				\
+	})
+
+#define WREG32_SOC24_JPEG_DPG_MODE(inst_idx, offset, value, indirect)		\
+	do {									\
+		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
+			     regUVD_DPG_LMA_DATA, value);			\
+		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
+			     regUVD_DPG_LMA_MASK, 0xFFFFFFFF);			\
+		WREG32_SOC15(							\
+			JPEG, GET_INST(JPEG, inst_idx),				\
+			regUVD_DPG_LMA_CTL,					\
+			(UVD_DPG_LMA_CTL__READ_WRITE_MASK |			\
+			 offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT |	\
+			 indirect << UVD_DPG_LMA_CTL__SRAM_SEL__SHIFT));	\
+	} while (0)
+
+#define RREG32_SOC24_JPEG_DPG_MODE(inst_idx, offset, mask_en)			\
+	do {									\
+		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
+			regUVD_DPG_LMA_MASK, 0xFFFFFFFF);			\
+		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
+			regUVD_DPG_LMA_CTL,					\
+			(UVD_DPG_LMA_CTL__MASK_EN_MASK |			\
+			offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT));	\
+		RREG32_SOC15(JPEG, inst_idx, regUVD_DPG_LMA_DATA);		\
+	} while (0)
+
+#define ADD_SOC24_JPEG_TO_DPG_SRAM(inst_idx, offset, value, indirect)		\
+	do {									\
+		*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ = offset;	\
+		*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ = value;	\
+	} while (0)
+
+struct amdgpu_hwip_reg_entry;
+
+enum amdgpu_jpeg_caps {
+	AMDGPU_JPEG_RRMT_ENABLED,
+};
+
+#define AMDGPU_JPEG_CAPS(caps) BIT(AMDGPU_JPEG_##caps)
+
+struct amdgpu_jpeg_reg{
+	unsigned jpeg_pitch[AMDGPU_MAX_JPEG_RINGS];
+};
+
+struct amdgpu_jpeg_inst {
+	struct amdgpu_ring ring_dec[AMDGPU_MAX_JPEG_RINGS];
+	struct amdgpu_irq_src irq;
+	struct amdgpu_irq_src ras_poison_irq;
+	struct amdgpu_jpeg_reg external;
+	struct amdgpu_bo	*dpg_sram_bo;
+	struct dpg_pause_state	pause_state;
+	void			*dpg_sram_cpu_addr;
+	uint64_t		dpg_sram_gpu_addr;
+	uint32_t		*dpg_sram_curr_addr;
+	uint8_t aid_id;
+};
+
+struct amdgpu_jpeg_ras {
+	struct amdgpu_ras_block_object ras_block;
+};
+
+struct amdgpu_jpeg {
+	uint8_t	num_jpeg_inst;
+	struct amdgpu_jpeg_inst inst[AMDGPU_MAX_JPEG_INSTANCES];
+	unsigned num_jpeg_rings;
+	struct amdgpu_jpeg_reg internal;
+	unsigned harvest_config;
+	struct delayed_work idle_work;
+	enum amd_powergating_state cur_state;
+	struct mutex jpeg_pg_lock;
+	atomic_t total_submission_cnt;
+	struct ras_common_if	*ras_if;
+	struct amdgpu_jpeg_ras	*ras;
+
+	uint16_t inst_mask;
+	uint8_t num_inst_per_aid;
+	bool	indirect_sram;
+	uint32_t supported_reset;
+	uint32_t caps;
+	u32 *ip_dump;
+	u32 reg_count;
+	const struct amdgpu_hwip_reg_entry *reg_list;
+};
+
+int amdgpu_jpeg_sw_init(struct amdgpu_device *adev);
+int amdgpu_jpeg_sw_fini(struct amdgpu_device *adev);
+int amdgpu_jpeg_suspend(struct amdgpu_device *adev);
+int amdgpu_jpeg_resume(struct amdgpu_device *adev);
+
+void amdgpu_jpeg_ring_begin_use(struct amdgpu_ring *ring);
+void amdgpu_jpeg_ring_end_use(struct amdgpu_ring *ring);
+
+int amdgpu_jpeg_dec_ring_test_ring(struct amdgpu_ring *ring);
+int amdgpu_jpeg_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout);
+
+int amdgpu_jpeg_process_poison_irq(struct amdgpu_device *adev,
+				struct amdgpu_irq_src *source,
+				struct amdgpu_iv_entry *entry);
+int amdgpu_jpeg_ras_late_init(struct amdgpu_device *adev,
+				struct ras_common_if *ras_block);
+int amdgpu_jpeg_ras_sw_init(struct amdgpu_device *adev);
+int amdgpu_jpeg_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
+			       enum AMDGPU_UCODE_ID ucode_id);
+void amdgpu_debugfs_jpeg_sched_mask_init(struct amdgpu_device *adev);
+int amdgpu_jpeg_sysfs_reset_mask_init(struct amdgpu_device *adev);
+void amdgpu_jpeg_sysfs_reset_mask_fini(struct amdgpu_device *adev);
+int amdgpu_jpeg_reg_dump_init(struct amdgpu_device *adev,
+			       const struct amdgpu_hwip_reg_entry *reg, u32 count);
+void amdgpu_jpeg_dump_ip_state(struct amdgpu_ip_block *ip_block);
+void amdgpu_jpeg_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p);
+int amdgpu_jpeg_dec_parse_cs(struct amdgpu_cs_parser *parser,
+			     struct amdgpu_job *job,
+			     struct amdgpu_ib *ib);
+
+#endif /*__AMDGPU_JPEG_H__*/
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 96c341670782..6ee77f431d56 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -25,16 +25,51 @@
  *          Alex Deucher
  *          Jerome Glisse
  */
-#include <drm/drmP.h>
+
 #include "amdgpu.h"
 #include <drm/amdgpu_drm.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_fb_helper.h>
 #include "amdgpu_uvd.h"
 #include "amdgpu_vce.h"
+#include "atom.h"
 
 #include <linux/vga_switcheroo.h>
 #include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/pci.h>
 #include <linux/pm_runtime.h>
 #include "amdgpu_amdkfd.h"
+#include "amdgpu_gem.h"
+#include "amdgpu_display.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_reset.h"
+#include "amd_pcie.h"
+#include "amdgpu_userq.h"
+
+void amdgpu_unregister_gpu_instance(struct amdgpu_device *adev)
+{
+	struct amdgpu_gpu_instance *gpu_instance;
+	int i;
+
+	mutex_lock(&mgpu_info.mutex);
+
+	for (i = 0; i < mgpu_info.num_gpu; i++) {
+		gpu_instance = &(mgpu_info.gpu_ins[i]);
+		if (gpu_instance->adev == adev) {
+			mgpu_info.gpu_ins[i] =
+				mgpu_info.gpu_ins[mgpu_info.num_gpu - 1];
+			mgpu_info.num_gpu--;
+			if (adev->flags & AMD_IS_APU)
+				mgpu_info.num_apu--;
+			else
+				mgpu_info.num_dgpu--;
+			break;
+		}
+	}
+
+	mutex_unlock(&mgpu_info.mutex);
+}
 
 /**
  * amdgpu_driver_unload_kms - Main unload function for KMS.
@@ -46,60 +81,63 @@
  */
 void amdgpu_driver_unload_kms(struct drm_device *dev)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 
 	if (adev == NULL)
 		return;
 
+	amdgpu_unregister_gpu_instance(adev);
+
 	if (adev->rmmio == NULL)
-		goto done_free;
+		return;
 
-	if (amdgpu_sriov_vf(adev))
-		amdgpu_virt_request_full_gpu(adev, false);
+	if (amdgpu_acpi_smart_shift_update(adev, AMDGPU_SS_DRV_UNLOAD))
+		DRM_WARN("smart shift update failed\n");
 
-	if (amdgpu_device_is_px(dev)) {
-		pm_runtime_get_sync(dev->dev);
-		pm_runtime_forbid(dev->dev);
-	}
+	amdgpu_acpi_fini(adev);
+	amdgpu_device_fini_hw(adev);
+}
+
+void amdgpu_register_gpu_instance(struct amdgpu_device *adev)
+{
+	struct amdgpu_gpu_instance *gpu_instance;
 
-	amdgpu_amdkfd_device_fini(adev);
+	mutex_lock(&mgpu_info.mutex);
 
-	amdgpu_acpi_fini(adev);
+	if (mgpu_info.num_gpu >= MAX_GPU_INSTANCE) {
+		DRM_ERROR("Cannot register more gpu instance\n");
+		mutex_unlock(&mgpu_info.mutex);
+		return;
+	}
 
-	amdgpu_device_fini(adev);
+	gpu_instance = &(mgpu_info.gpu_ins[mgpu_info.num_gpu]);
+	gpu_instance->adev = adev;
+	gpu_instance->mgpu_fan_enabled = 0;
 
-done_free:
-	kfree(adev);
-	dev->dev_private = NULL;
+	mgpu_info.num_gpu++;
+	if (adev->flags & AMD_IS_APU)
+		mgpu_info.num_apu++;
+	else
+		mgpu_info.num_dgpu++;
+
+	mutex_unlock(&mgpu_info.mutex);
 }
 
 /**
  * amdgpu_driver_load_kms - Main load function for KMS.
  *
- * @dev: drm dev pointer
+ * @adev: pointer to struct amdgpu_device
  * @flags: device flags
  *
  * This is the main load function for KMS (all asics).
  * Returns 0 on success, error on failure.
  */
-int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
+int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags)
 {
-	struct amdgpu_device *adev;
+	struct drm_device *dev;
 	int r, acpi_status;
 
-	adev = kzalloc(sizeof(struct amdgpu_device), GFP_KERNEL);
-	if (adev == NULL) {
-		return -ENOMEM;
-	}
-	dev->dev_private = (void *)adev;
-
-	if ((amdgpu_runtime_pm != 0) &&
-	    amdgpu_has_atpx() &&
-	    (amdgpu_is_atpx_hybrid() ||
-	     amdgpu_has_atpx_dgpu_power_cntl()) &&
-	    ((flags & AMD_IS_APU) == 0) &&
-	    !pci_is_thunderbolt_attached(dev->pdev))
-		flags |= AMD_IS_PX;
+	dev = adev_to_drm(adev);
 
 	/* amdgpu_device_init should report only fatal error
 	 * like memory allocation failure or iomapping failure,
@@ -107,49 +145,70 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
 	 * properly initialize the GPU MC controller and permit
 	 * VRAM allocation
 	 */
-	r = amdgpu_device_init(adev, dev, dev->pdev, flags);
+	r = amdgpu_device_init(adev, flags);
 	if (r) {
-		dev_err(&dev->pdev->dev, "Fatal error during GPU init\n");
+		dev_err(dev->dev, "Fatal error during GPU init\n");
 		goto out;
 	}
 
+	amdgpu_device_detect_runtime_pm_mode(adev);
+
 	/* Call ACPI methods: require modeset init
 	 * but failure is not fatal
 	 */
-	if (!r) {
-		acpi_status = amdgpu_acpi_init(adev);
-		if (acpi_status)
-		dev_dbg(&dev->pdev->dev,
-				"Error during ACPI methods call\n");
-	}
 
-	amdgpu_amdkfd_load_interface(adev);
-	amdgpu_amdkfd_device_probe(adev);
-	amdgpu_amdkfd_device_init(adev);
-
-	if (amdgpu_device_is_px(dev)) {
-		pm_runtime_use_autosuspend(dev->dev);
-		pm_runtime_set_autosuspend_delay(dev->dev, 5000);
-		pm_runtime_set_active(dev->dev);
-		pm_runtime_allow(dev->dev);
-		pm_runtime_mark_last_busy(dev->dev);
-		pm_runtime_put_autosuspend(dev->dev);
-	}
+	acpi_status = amdgpu_acpi_init(adev);
+	if (acpi_status)
+		dev_dbg(dev->dev, "Error during ACPI methods call\n");
 
-	if (amdgpu_sriov_vf(adev))
-		amdgpu_virt_release_full_gpu(adev, true);
+	if (amdgpu_acpi_smart_shift_update(adev, AMDGPU_SS_DRV_LOAD))
+		DRM_WARN("smart shift update failed\n");
 
 out:
-	if (r) {
-		/* balance pm_runtime_get_sync in amdgpu_driver_unload_kms */
-		if (adev->rmmio && amdgpu_device_is_px(dev))
-			pm_runtime_put_noidle(dev->dev);
+	if (r)
 		amdgpu_driver_unload_kms(dev);
-	}
 
 	return r;
 }
 
+static enum amd_ip_block_type
+	amdgpu_ip_get_block_type(struct amdgpu_device *adev, uint32_t ip)
+{
+	enum amd_ip_block_type type;
+
+	switch (ip) {
+	case AMDGPU_HW_IP_GFX:
+		type = AMD_IP_BLOCK_TYPE_GFX;
+		break;
+	case AMDGPU_HW_IP_COMPUTE:
+		type = AMD_IP_BLOCK_TYPE_GFX;
+		break;
+	case AMDGPU_HW_IP_DMA:
+		type = AMD_IP_BLOCK_TYPE_SDMA;
+		break;
+	case AMDGPU_HW_IP_UVD:
+	case AMDGPU_HW_IP_UVD_ENC:
+		type = AMD_IP_BLOCK_TYPE_UVD;
+		break;
+	case AMDGPU_HW_IP_VCE:
+		type = AMD_IP_BLOCK_TYPE_VCE;
+		break;
+	case AMDGPU_HW_IP_VCN_DEC:
+	case AMDGPU_HW_IP_VCN_ENC:
+		type = AMD_IP_BLOCK_TYPE_VCN;
+		break;
+	case AMDGPU_HW_IP_VCN_JPEG:
+		type = (amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_JPEG)) ?
+				   AMD_IP_BLOCK_TYPE_JPEG : AMD_IP_BLOCK_TYPE_VCN;
+		break;
+	default:
+		type = AMD_IP_BLOCK_TYPE_NUM;
+		break;
+	}
+
+	return type;
+}
+
 static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 				struct drm_amdgpu_query_fw *query_fw,
 				struct amdgpu_device *adev)
@@ -163,8 +222,12 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 		fw_info->ver = adev->uvd.fw_version;
 		fw_info->feature = 0;
 		break;
+	case AMDGPU_INFO_FW_VCN:
+		fw_info->ver = adev->vcn.fw_version;
+		fw_info->feature = 0;
+		break;
 	case AMDGPU_INFO_FW_GMC:
-		fw_info->ver = adev->mc.fw_version;
+		fw_info->ver = adev->gmc.fw_version;
 		fw_info->feature = 0;
 		break;
 	case AMDGPU_INFO_FW_GFX_ME:
@@ -183,6 +246,26 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 		fw_info->ver = adev->gfx.rlc_fw_version;
 		fw_info->feature = adev->gfx.rlc_feature_version;
 		break;
+	case AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_CNTL:
+		fw_info->ver = adev->gfx.rlc_srlc_fw_version;
+		fw_info->feature = adev->gfx.rlc_srlc_feature_version;
+		break;
+	case AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_GPM_MEM:
+		fw_info->ver = adev->gfx.rlc_srlg_fw_version;
+		fw_info->feature = adev->gfx.rlc_srlg_feature_version;
+		break;
+	case AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM:
+		fw_info->ver = adev->gfx.rlc_srls_fw_version;
+		fw_info->feature = adev->gfx.rlc_srls_feature_version;
+		break;
+	case AMDGPU_INFO_FW_GFX_RLCP:
+		fw_info->ver = adev->gfx.rlcp_ucode_version;
+		fw_info->feature = adev->gfx.rlcp_ucode_feature_version;
+		break;
+	case AMDGPU_INFO_FW_GFX_RLCV:
+		fw_info->ver = adev->gfx.rlcv_ucode_version;
+		fw_info->feature = adev->gfx.rlcv_ucode_feature_version;
+		break;
 	case AMDGPU_INFO_FW_GFX_MEC:
 		if (query_fw->index == 0) {
 			fw_info->ver = adev->gfx.mec_fw_version;
@@ -197,6 +280,43 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 		fw_info->ver = adev->pm.fw_version;
 		fw_info->feature = 0;
 		break;
+	case AMDGPU_INFO_FW_TA:
+		switch (query_fw->index) {
+		case TA_FW_TYPE_PSP_XGMI:
+			fw_info->ver = adev->psp.xgmi_context.context.bin_desc.fw_version;
+			fw_info->feature = adev->psp.xgmi_context.context
+						   .bin_desc.feature_version;
+			break;
+		case TA_FW_TYPE_PSP_RAS:
+			fw_info->ver = adev->psp.ras_context.context.bin_desc.fw_version;
+			fw_info->feature = adev->psp.ras_context.context
+						   .bin_desc.feature_version;
+			break;
+		case TA_FW_TYPE_PSP_HDCP:
+			fw_info->ver = adev->psp.hdcp_context.context.bin_desc.fw_version;
+			fw_info->feature = adev->psp.hdcp_context.context
+						   .bin_desc.feature_version;
+			break;
+		case TA_FW_TYPE_PSP_DTM:
+			fw_info->ver = adev->psp.dtm_context.context.bin_desc.fw_version;
+			fw_info->feature = adev->psp.dtm_context.context
+						   .bin_desc.feature_version;
+			break;
+		case TA_FW_TYPE_PSP_RAP:
+			fw_info->ver = adev->psp.rap_context.context.bin_desc.fw_version;
+			fw_info->feature = adev->psp.rap_context.context
+						   .bin_desc.feature_version;
+			break;
+		case TA_FW_TYPE_PSP_SECUREDISPLAY:
+			fw_info->ver = adev->psp.securedisplay_context.context.bin_desc.fw_version;
+			fw_info->feature =
+				adev->psp.securedisplay_context.context.bin_desc
+					.feature_version;
+			break;
+		default:
+			return -EINVAL;
+		}
+		break;
 	case AMDGPU_INFO_FW_SDMA:
 		if (query_fw->index >= adev->sdma.num_instances)
 			return -EINVAL;
@@ -204,16 +324,274 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 		fw_info->feature = adev->sdma.instance[query_fw->index].feature_version;
 		break;
 	case AMDGPU_INFO_FW_SOS:
-		fw_info->ver = adev->psp.sos_fw_version;
-		fw_info->feature = adev->psp.sos_feature_version;
+		fw_info->ver = adev->psp.sos.fw_version;
+		fw_info->feature = adev->psp.sos.feature_version;
 		break;
 	case AMDGPU_INFO_FW_ASD:
-		fw_info->ver = adev->psp.asd_fw_version;
-		fw_info->feature = adev->psp.asd_feature_version;
+		fw_info->ver = adev->psp.asd_context.bin_desc.fw_version;
+		fw_info->feature = adev->psp.asd_context.bin_desc.feature_version;
+		break;
+	case AMDGPU_INFO_FW_DMCU:
+		fw_info->ver = adev->dm.dmcu_fw_version;
+		fw_info->feature = 0;
+		break;
+	case AMDGPU_INFO_FW_DMCUB:
+		fw_info->ver = adev->dm.dmcub_fw_version;
+		fw_info->feature = 0;
+		break;
+	case AMDGPU_INFO_FW_TOC:
+		fw_info->ver = adev->psp.toc.fw_version;
+		fw_info->feature = adev->psp.toc.feature_version;
+		break;
+	case AMDGPU_INFO_FW_CAP:
+		fw_info->ver = adev->psp.cap_fw_version;
+		fw_info->feature = adev->psp.cap_feature_version;
+		break;
+	case AMDGPU_INFO_FW_MES_KIQ:
+		fw_info->ver = adev->mes.kiq_version & AMDGPU_MES_VERSION_MASK;
+		fw_info->feature = (adev->mes.kiq_version & AMDGPU_MES_FEAT_VERSION_MASK)
+					>> AMDGPU_MES_FEAT_VERSION_SHIFT;
+		break;
+	case AMDGPU_INFO_FW_MES:
+		fw_info->ver = adev->mes.sched_version & AMDGPU_MES_VERSION_MASK;
+		fw_info->feature = (adev->mes.sched_version & AMDGPU_MES_FEAT_VERSION_MASK)
+					>> AMDGPU_MES_FEAT_VERSION_SHIFT;
+		break;
+	case AMDGPU_INFO_FW_IMU:
+		fw_info->ver = adev->gfx.imu_fw_version;
+		fw_info->feature = 0;
+		break;
+	case AMDGPU_INFO_FW_VPE:
+		fw_info->ver = adev->vpe.fw_version;
+		fw_info->feature = adev->vpe.feature_version;
+		break;
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int amdgpu_userq_metadata_info_gfx(struct amdgpu_device *adev,
+					  struct drm_amdgpu_info *info,
+					  struct drm_amdgpu_info_uq_metadata_gfx *meta)
+{
+	int ret = -EOPNOTSUPP;
+
+	if (adev->gfx.funcs->get_gfx_shadow_info) {
+		struct amdgpu_gfx_shadow_info shadow = {};
+
+		adev->gfx.funcs->get_gfx_shadow_info(adev, &shadow, true);
+		meta->shadow_size = shadow.shadow_size;
+		meta->shadow_alignment = shadow.shadow_alignment;
+		meta->csa_size = shadow.csa_size;
+		meta->csa_alignment = shadow.csa_alignment;
+		ret = 0;
+	}
+
+	return ret;
+}
+
+static int amdgpu_hw_ip_info(struct amdgpu_device *adev,
+			     struct drm_amdgpu_info *info,
+			     struct drm_amdgpu_info_hw_ip *result)
+{
+	uint32_t ib_start_alignment = 0;
+	uint32_t ib_size_alignment = 0;
+	enum amd_ip_block_type type;
+	unsigned int num_rings = 0;
+	uint32_t num_slots = 0;
+	unsigned int i, j;
+
+	if (info->query_hw_ip.ip_instance >= AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
+		return -EINVAL;
+
+	switch (info->query_hw_ip.type) {
+	case AMDGPU_HW_IP_GFX:
+		type = AMD_IP_BLOCK_TYPE_GFX;
+		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
+			if (adev->gfx.gfx_ring[i].sched.ready &&
+			    !adev->gfx.gfx_ring[i].no_user_submission)
+				++num_rings;
+
+		if (!adev->gfx.disable_uq) {
+			for (i = 0; i < AMDGPU_MES_MAX_GFX_PIPES; i++)
+				num_slots += hweight32(adev->mes.gfx_hqd_mask[i]);
+		}
+
+		ib_start_alignment = 32;
+		ib_size_alignment = 32;
+		break;
+	case AMDGPU_HW_IP_COMPUTE:
+		type = AMD_IP_BLOCK_TYPE_GFX;
+		for (i = 0; i < adev->gfx.num_compute_rings; i++)
+			if (adev->gfx.compute_ring[i].sched.ready &&
+			    !adev->gfx.compute_ring[i].no_user_submission)
+				++num_rings;
+
+		if (!adev->sdma.disable_uq) {
+			for (i = 0; i < AMDGPU_MES_MAX_COMPUTE_PIPES; i++)
+				num_slots += hweight32(adev->mes.compute_hqd_mask[i]);
+		}
+
+		ib_start_alignment = 32;
+		ib_size_alignment = 32;
+		break;
+	case AMDGPU_HW_IP_DMA:
+		type = AMD_IP_BLOCK_TYPE_SDMA;
+		for (i = 0; i < adev->sdma.num_instances; i++)
+			if (adev->sdma.instance[i].ring.sched.ready &&
+			    !adev->sdma.instance[i].ring.no_user_submission)
+				++num_rings;
+
+		if (!adev->gfx.disable_uq) {
+			for (i = 0; i < AMDGPU_MES_MAX_SDMA_PIPES; i++)
+				num_slots += hweight32(adev->mes.sdma_hqd_mask[i]);
+		}
+
+		ib_start_alignment = 256;
+		ib_size_alignment = 4;
+		break;
+	case AMDGPU_HW_IP_UVD:
+		type = AMD_IP_BLOCK_TYPE_UVD;
+		for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+			if (adev->uvd.harvest_config & (1 << i))
+				continue;
+
+			if (adev->uvd.inst[i].ring.sched.ready &&
+			    !adev->uvd.inst[i].ring.no_user_submission)
+				++num_rings;
+		}
+		ib_start_alignment = 256;
+		ib_size_alignment = 64;
+		break;
+	case AMDGPU_HW_IP_VCE:
+		type = AMD_IP_BLOCK_TYPE_VCE;
+		for (i = 0; i < adev->vce.num_rings; i++)
+			if (adev->vce.ring[i].sched.ready &&
+			    !adev->vce.ring[i].no_user_submission)
+				++num_rings;
+		ib_start_alignment = 256;
+		ib_size_alignment = 4;
+		break;
+	case AMDGPU_HW_IP_UVD_ENC:
+		type = AMD_IP_BLOCK_TYPE_UVD;
+		for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+			if (adev->uvd.harvest_config & (1 << i))
+				continue;
+
+			for (j = 0; j < adev->uvd.num_enc_rings; j++)
+				if (adev->uvd.inst[i].ring_enc[j].sched.ready &&
+				    !adev->uvd.inst[i].ring_enc[j].no_user_submission)
+					++num_rings;
+		}
+		ib_start_alignment = 256;
+		ib_size_alignment = 4;
+		break;
+	case AMDGPU_HW_IP_VCN_DEC:
+		type = AMD_IP_BLOCK_TYPE_VCN;
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+			if (adev->vcn.harvest_config & (1 << i))
+				continue;
+
+			if (adev->vcn.inst[i].ring_dec.sched.ready &&
+			    !adev->vcn.inst[i].ring_dec.no_user_submission)
+				++num_rings;
+		}
+		ib_start_alignment = 256;
+		ib_size_alignment = 64;
+		break;
+	case AMDGPU_HW_IP_VCN_ENC:
+		type = AMD_IP_BLOCK_TYPE_VCN;
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+			if (adev->vcn.harvest_config & (1 << i))
+				continue;
+
+			for (j = 0; j < adev->vcn.inst[i].num_enc_rings; j++)
+				if (adev->vcn.inst[i].ring_enc[j].sched.ready &&
+				    !adev->vcn.inst[i].ring_enc[j].no_user_submission)
+					++num_rings;
+		}
+		ib_start_alignment = 256;
+		ib_size_alignment = 4;
+		break;
+	case AMDGPU_HW_IP_VCN_JPEG:
+		type = (amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_JPEG)) ?
+			AMD_IP_BLOCK_TYPE_JPEG : AMD_IP_BLOCK_TYPE_VCN;
+
+		for (i = 0; i < adev->jpeg.num_jpeg_inst; i++) {
+			if (adev->jpeg.harvest_config & (1 << i))
+				continue;
+
+			for (j = 0; j < adev->jpeg.num_jpeg_rings; j++)
+				if (adev->jpeg.inst[i].ring_dec[j].sched.ready &&
+				    !adev->jpeg.inst[i].ring_dec[j].no_user_submission)
+					++num_rings;
+		}
+		ib_start_alignment = 256;
+		ib_size_alignment = 64;
+		break;
+	case AMDGPU_HW_IP_VPE:
+		type = AMD_IP_BLOCK_TYPE_VPE;
+		if (adev->vpe.ring.sched.ready &&
+		    !adev->vpe.ring.no_user_submission)
+			++num_rings;
+		ib_start_alignment = 256;
+		ib_size_alignment = 4;
 		break;
 	default:
 		return -EINVAL;
 	}
+
+	for (i = 0; i < adev->num_ip_blocks; i++)
+		if (adev->ip_blocks[i].version->type == type &&
+		    adev->ip_blocks[i].status.valid)
+			break;
+
+	if (i == adev->num_ip_blocks)
+		return 0;
+
+	num_rings = min(amdgpu_ctx_num_entities[info->query_hw_ip.type],
+			num_rings);
+
+	result->hw_ip_version_major = adev->ip_blocks[i].version->major;
+	result->hw_ip_version_minor = adev->ip_blocks[i].version->minor;
+
+	if (adev->asic_type >= CHIP_VEGA10) {
+		switch (type) {
+		case AMD_IP_BLOCK_TYPE_GFX:
+			result->ip_discovery_version =
+				IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, GC_HWIP, 0));
+			break;
+		case AMD_IP_BLOCK_TYPE_SDMA:
+			result->ip_discovery_version =
+				IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, SDMA0_HWIP, 0));
+			break;
+		case AMD_IP_BLOCK_TYPE_UVD:
+		case AMD_IP_BLOCK_TYPE_VCN:
+		case AMD_IP_BLOCK_TYPE_JPEG:
+			result->ip_discovery_version =
+				IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, UVD_HWIP, 0));
+			break;
+		case AMD_IP_BLOCK_TYPE_VCE:
+			result->ip_discovery_version =
+				IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, VCE_HWIP, 0));
+			break;
+		case AMD_IP_BLOCK_TYPE_VPE:
+			result->ip_discovery_version =
+				IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, VPE_HWIP, 0));
+			break;
+		default:
+			result->ip_discovery_version = 0;
+			break;
+		}
+	} else {
+		result->ip_discovery_version = 0;
+	}
+	result->capabilities_flags = 0;
+	result->available_rings = (1 << num_rings) - 1;
+	result->userq_num_slots = num_slots;
+	result->ib_start_alignment = ib_start_alignment;
+	result->ib_size_alignment = ib_size_alignment;
 	return 0;
 }
 
@@ -223,7 +601,7 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 /**
  * amdgpu_info_ioctl - answer a device specific request.
  *
- * @adev: amdgpu device pointer
+ * @dev: drm device pointer
  * @data: request object
  * @filp: drm filp
  *
@@ -232,17 +610,22 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
  * etc. (all asics).
  * Returns 0 on success, -EINVAL on failure.
  */
-static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
+int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct drm_amdgpu_info *info = data;
 	struct amdgpu_mode_info *minfo = &adev->mode_info;
 	void __user *out = (void __user *)(uintptr_t)info->return_pointer;
+	struct amdgpu_fpriv *fpriv;
+	struct amdgpu_ip_block *ip_block;
+	enum amd_ip_block_type type;
+	struct amdgpu_xcp *xcp;
+	u32 count, inst_mask;
 	uint32_t size = info->return_size;
 	struct drm_crtc *crtc;
 	uint32_t ui32 = 0;
 	uint64_t ui64 = 0;
-	int i, found;
+	int i, found, ret;
 	int ui32_size = sizeof(ui32);
 
 	if (!info->return_size || !info->return_pointer)
@@ -257,6 +640,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			crtc = (struct drm_crtc *)minfo->crtcs[i];
 			if (crtc && crtc->base.id == info->mode_crtc.id) {
 				struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
+
 				ui32 = amdgpu_crtc->crtc_id;
 				found = 1;
 				break;
@@ -269,108 +653,82 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		return copy_to_user(out, &ui32, min(size, 4u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_HW_IP_INFO: {
 		struct drm_amdgpu_info_hw_ip ip = {};
-		enum amd_ip_block_type type;
-		uint32_t ring_mask = 0;
-		uint32_t ib_start_alignment = 0;
-		uint32_t ib_size_alignment = 0;
 
-		if (info->query_hw_ip.ip_instance >= AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
-			return -EINVAL;
+		ret = amdgpu_hw_ip_info(adev, info, &ip);
+		if (ret)
+			return ret;
 
-		switch (info->query_hw_ip.type) {
-		case AMDGPU_HW_IP_GFX:
-			type = AMD_IP_BLOCK_TYPE_GFX;
-			for (i = 0; i < adev->gfx.num_gfx_rings; i++)
-				ring_mask |= ((adev->gfx.gfx_ring[i].ready ? 1 : 0) << i);
-			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
-			ib_size_alignment = 8;
-			break;
-		case AMDGPU_HW_IP_COMPUTE:
-			type = AMD_IP_BLOCK_TYPE_GFX;
-			for (i = 0; i < adev->gfx.num_compute_rings; i++)
-				ring_mask |= ((adev->gfx.compute_ring[i].ready ? 1 : 0) << i);
-			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
-			ib_size_alignment = 8;
-			break;
-		case AMDGPU_HW_IP_DMA:
-			type = AMD_IP_BLOCK_TYPE_SDMA;
-			for (i = 0; i < adev->sdma.num_instances; i++)
-				ring_mask |= ((adev->sdma.instance[i].ring.ready ? 1 : 0) << i);
-			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
-			ib_size_alignment = 1;
-			break;
-		case AMDGPU_HW_IP_UVD:
-			type = AMD_IP_BLOCK_TYPE_UVD;
-			ring_mask = adev->uvd.ring.ready ? 1 : 0;
-			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
-			ib_size_alignment = 16;
-			break;
-		case AMDGPU_HW_IP_VCE:
-			type = AMD_IP_BLOCK_TYPE_VCE;
-			for (i = 0; i < adev->vce.num_rings; i++)
-				ring_mask |= ((adev->vce.ring[i].ready ? 1 : 0) << i);
-			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
-			ib_size_alignment = 1;
-			break;
-		case AMDGPU_HW_IP_UVD_ENC:
-			type = AMD_IP_BLOCK_TYPE_UVD;
-			for (i = 0; i < adev->uvd.num_enc_rings; i++)
-				ring_mask |= ((adev->uvd.ring_enc[i].ready ? 1 : 0) << i);
-			ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
-			ib_size_alignment = 1;
-			break;
-		default:
+		ret = copy_to_user(out, &ip, min_t(size_t, size, sizeof(ip)));
+		return ret ? -EFAULT : 0;
+	}
+	case AMDGPU_INFO_HW_IP_COUNT: {
+		fpriv = (struct amdgpu_fpriv *)filp->driver_priv;
+		type = amdgpu_ip_get_block_type(adev, info->query_hw_ip.type);
+		ip_block = amdgpu_device_ip_get_ip_block(adev, type);
+
+		if (!ip_block || !ip_block->status.valid)
 			return -EINVAL;
-		}
 
-		for (i = 0; i < adev->num_ip_blocks; i++) {
-			if (adev->ip_blocks[i].version->type == type &&
-			    adev->ip_blocks[i].status.valid) {
-				ip.hw_ip_version_major = adev->ip_blocks[i].version->major;
-				ip.hw_ip_version_minor = adev->ip_blocks[i].version->minor;
-				ip.capabilities_flags = 0;
-				ip.available_rings = ring_mask;
-				ip.ib_start_alignment = ib_start_alignment;
-				ip.ib_size_alignment = ib_size_alignment;
+		if (adev->xcp_mgr && adev->xcp_mgr->num_xcps > 0 &&
+		    fpriv->xcp_id < adev->xcp_mgr->num_xcps) {
+			xcp = &adev->xcp_mgr->xcp[fpriv->xcp_id];
+			switch (type) {
+			case AMD_IP_BLOCK_TYPE_GFX:
+				ret = amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_GFX, &inst_mask);
+				if (ret)
+					return ret;
+				count = hweight32(inst_mask);
+				break;
+			case AMD_IP_BLOCK_TYPE_SDMA:
+				ret = amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_SDMA, &inst_mask);
+				if (ret)
+					return ret;
+				count = hweight32(inst_mask);
+				break;
+			case AMD_IP_BLOCK_TYPE_JPEG:
+				ret = amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_VCN, &inst_mask);
+				if (ret)
+					return ret;
+				count = hweight32(inst_mask) * adev->jpeg.num_jpeg_rings;
 				break;
+			case AMD_IP_BLOCK_TYPE_VCN:
+				ret = amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_VCN, &inst_mask);
+				if (ret)
+					return ret;
+				count = hweight32(inst_mask);
+				break;
+			default:
+				return -EINVAL;
 			}
+
+			return copy_to_user(out, &count, min(size, 4u)) ? -EFAULT : 0;
 		}
-		return copy_to_user(out, &ip,
-				    min((size_t)size, sizeof(ip))) ? -EFAULT : 0;
-	}
-	case AMDGPU_INFO_HW_IP_COUNT: {
-		enum amd_ip_block_type type;
-		uint32_t count = 0;
 
-		switch (info->query_hw_ip.type) {
-		case AMDGPU_HW_IP_GFX:
-			type = AMD_IP_BLOCK_TYPE_GFX;
-			break;
-		case AMDGPU_HW_IP_COMPUTE:
-			type = AMD_IP_BLOCK_TYPE_GFX;
+		switch (type) {
+		case AMD_IP_BLOCK_TYPE_GFX:
+		case AMD_IP_BLOCK_TYPE_VCE:
+			count = 1;
 			break;
-		case AMDGPU_HW_IP_DMA:
-			type = AMD_IP_BLOCK_TYPE_SDMA;
+		case AMD_IP_BLOCK_TYPE_SDMA:
+			count = adev->sdma.num_instances;
 			break;
-		case AMDGPU_HW_IP_UVD:
-			type = AMD_IP_BLOCK_TYPE_UVD;
+		case AMD_IP_BLOCK_TYPE_JPEG:
+			count = adev->jpeg.num_jpeg_inst * adev->jpeg.num_jpeg_rings;
 			break;
-		case AMDGPU_HW_IP_VCE:
-			type = AMD_IP_BLOCK_TYPE_VCE;
+		case AMD_IP_BLOCK_TYPE_VCN:
+			count = adev->vcn.num_vcn_inst;
 			break;
-		case AMDGPU_HW_IP_UVD_ENC:
-			type = AMD_IP_BLOCK_TYPE_UVD;
+		case AMD_IP_BLOCK_TYPE_UVD:
+			count = adev->uvd.num_uvd_inst;
 			break;
+		/* For all other IP block types not listed in the switch statement
+		 * the ip status is valid here and the instance count is one.
+		 */
 		default:
-			return -EINVAL;
+			count = 1;
+			break;
 		}
 
-		for (i = 0; i < adev->num_ip_blocks; i++)
-			if (adev->ip_blocks[i].version->type == type &&
-			    adev->ip_blocks[i].status.valid &&
-			    count < AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
-				count++;
-
 		return copy_to_user(out, &count, min(size, 4u)) ? -EFAULT : 0;
 	}
 	case AMDGPU_INFO_TIMESTAMP:
@@ -378,7 +736,6 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_FW_VERSION: {
 		struct drm_amdgpu_info_firmware fw_info;
-		int ret;
 
 		/* We only support one instance of each IP block right now. */
 		if (info->query_fw.ip_instance != 0)
@@ -397,65 +754,76 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 	case AMDGPU_INFO_NUM_EVICTIONS:
 		ui64 = atomic64_read(&adev->num_evictions);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
+	case AMDGPU_INFO_NUM_VRAM_CPU_PAGE_FAULTS:
+		ui64 = atomic64_read(&adev->num_vram_cpu_page_faults);
+		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_VRAM_USAGE:
-		ui64 = atomic64_read(&adev->vram_usage);
+		ui64 = ttm_resource_manager_used(&adev->mman.vram_mgr.manager) ?
+			ttm_resource_manager_usage(&adev->mman.vram_mgr.manager) : 0;
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_VIS_VRAM_USAGE:
-		ui64 = atomic64_read(&adev->vram_vis_usage);
+		ui64 = amdgpu_vram_mgr_vis_usage(&adev->mman.vram_mgr);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_GTT_USAGE:
-		ui64 = atomic64_read(&adev->gtt_usage);
+		ui64 = ttm_resource_manager_usage(&adev->mman.gtt_mgr.manager);
 		return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
 	case AMDGPU_INFO_GDS_CONFIG: {
 		struct drm_amdgpu_info_gds gds_info;
 
 		memset(&gds_info, 0, sizeof(gds_info));
-		gds_info.gds_gfx_partition_size = adev->gds.mem.gfx_partition_size >> AMDGPU_GDS_SHIFT;
-		gds_info.compute_partition_size = adev->gds.mem.cs_partition_size >> AMDGPU_GDS_SHIFT;
-		gds_info.gds_total_size = adev->gds.mem.total_size >> AMDGPU_GDS_SHIFT;
-		gds_info.gws_per_gfx_partition = adev->gds.gws.gfx_partition_size >> AMDGPU_GWS_SHIFT;
-		gds_info.gws_per_compute_partition = adev->gds.gws.cs_partition_size >> AMDGPU_GWS_SHIFT;
-		gds_info.oa_per_gfx_partition = adev->gds.oa.gfx_partition_size >> AMDGPU_OA_SHIFT;
-		gds_info.oa_per_compute_partition = adev->gds.oa.cs_partition_size >> AMDGPU_OA_SHIFT;
+		gds_info.compute_partition_size = adev->gds.gds_size;
+		gds_info.gds_total_size = adev->gds.gds_size;
+		gds_info.gws_per_compute_partition = adev->gds.gws_size;
+		gds_info.oa_per_compute_partition = adev->gds.oa_size;
 		return copy_to_user(out, &gds_info,
 				    min((size_t)size, sizeof(gds_info))) ? -EFAULT : 0;
 	}
 	case AMDGPU_INFO_VRAM_GTT: {
 		struct drm_amdgpu_info_vram_gtt vram_gtt;
 
-		vram_gtt.vram_size = adev->mc.real_vram_size;
-		vram_gtt.vram_size -= adev->vram_pin_size;
-		vram_gtt.vram_cpu_accessible_size = adev->mc.visible_vram_size;
-		vram_gtt.vram_cpu_accessible_size -= (adev->vram_pin_size - adev->invisible_pin_size);
-		vram_gtt.gtt_size  = adev->mc.gtt_size;
-		vram_gtt.gtt_size -= adev->gart_pin_size;
+		vram_gtt.vram_size = adev->gmc.real_vram_size -
+			atomic64_read(&adev->vram_pin_size) -
+			AMDGPU_VM_RESERVED_VRAM;
+		vram_gtt.vram_cpu_accessible_size =
+			min(adev->gmc.visible_vram_size -
+			    atomic64_read(&adev->visible_pin_size),
+			    vram_gtt.vram_size);
+		vram_gtt.gtt_size = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT)->size;
+		vram_gtt.gtt_size -= atomic64_read(&adev->gart_pin_size);
 		return copy_to_user(out, &vram_gtt,
 				    min((size_t)size, sizeof(vram_gtt))) ? -EFAULT : 0;
 	}
 	case AMDGPU_INFO_MEMORY: {
 		struct drm_amdgpu_memory_info mem;
+		struct ttm_resource_manager *gtt_man =
+			&adev->mman.gtt_mgr.manager;
+		struct ttm_resource_manager *vram_man =
+			&adev->mman.vram_mgr.manager;
 
 		memset(&mem, 0, sizeof(mem));
-		mem.vram.total_heap_size = adev->mc.real_vram_size;
-		mem.vram.usable_heap_size =
-			adev->mc.real_vram_size - adev->vram_pin_size;
-		mem.vram.heap_usage = atomic64_read(&adev->vram_usage);
+		mem.vram.total_heap_size = adev->gmc.real_vram_size;
+		mem.vram.usable_heap_size = adev->gmc.real_vram_size -
+			atomic64_read(&adev->vram_pin_size) -
+			AMDGPU_VM_RESERVED_VRAM;
+		mem.vram.heap_usage = ttm_resource_manager_used(&adev->mman.vram_mgr.manager) ?
+				ttm_resource_manager_usage(vram_man) : 0;
 		mem.vram.max_allocation = mem.vram.usable_heap_size * 3 / 4;
 
 		mem.cpu_accessible_vram.total_heap_size =
-			adev->mc.visible_vram_size;
+			adev->gmc.visible_vram_size;
 		mem.cpu_accessible_vram.usable_heap_size =
-			adev->mc.visible_vram_size -
-			(adev->vram_pin_size - adev->invisible_pin_size);
+			min(adev->gmc.visible_vram_size -
+			    atomic64_read(&adev->visible_pin_size),
+			    mem.vram.usable_heap_size);
 		mem.cpu_accessible_vram.heap_usage =
-			atomic64_read(&adev->vram_vis_usage);
+			amdgpu_vram_mgr_vis_usage(&adev->mman.vram_mgr);
 		mem.cpu_accessible_vram.max_allocation =
 			mem.cpu_accessible_vram.usable_heap_size * 3 / 4;
 
-		mem.gtt.total_heap_size = adev->mc.gtt_size;
-		mem.gtt.usable_heap_size =
-			adev->mc.gtt_size - adev->gart_pin_size;
-		mem.gtt.heap_usage = atomic64_read(&adev->gtt_usage);
+		mem.gtt.total_heap_size = gtt_man->size;
+		mem.gtt.usable_heap_size = mem.gtt.total_heap_size -
+			atomic64_read(&adev->gart_pin_size);
+		mem.gtt.heap_usage = ttm_resource_manager_usage(gtt_man);
 		mem.gtt.max_allocation = mem.gtt.usable_heap_size * 3 / 4;
 
 		return copy_to_user(out, &mem,
@@ -463,110 +831,218 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 				    ? -EFAULT : 0;
 	}
 	case AMDGPU_INFO_READ_MMR_REG: {
-		unsigned n, alloc_size;
+		int ret = 0;
+		unsigned int n, alloc_size;
 		uint32_t *regs;
-		unsigned se_num = (info->read_mmr_reg.instance >>
+		unsigned int se_num = (info->read_mmr_reg.instance >>
 				   AMDGPU_INFO_MMR_SE_INDEX_SHIFT) &
 				  AMDGPU_INFO_MMR_SE_INDEX_MASK;
-		unsigned sh_num = (info->read_mmr_reg.instance >>
+		unsigned int sh_num = (info->read_mmr_reg.instance >>
 				   AMDGPU_INFO_MMR_SH_INDEX_SHIFT) &
 				  AMDGPU_INFO_MMR_SH_INDEX_MASK;
 
+		if (!down_read_trylock(&adev->reset_domain->sem))
+			return -ENOENT;
+
 		/* set full masks if the userspace set all bits
-		 * in the bitfields */
-		if (se_num == AMDGPU_INFO_MMR_SE_INDEX_MASK)
+		 * in the bitfields
+		 */
+		if (se_num == AMDGPU_INFO_MMR_SE_INDEX_MASK) {
 			se_num = 0xffffffff;
-		if (sh_num == AMDGPU_INFO_MMR_SH_INDEX_MASK)
+		} else if (se_num >= AMDGPU_GFX_MAX_SE) {
+			ret = -EINVAL;
+			goto out;
+		}
+
+		if (sh_num == AMDGPU_INFO_MMR_SH_INDEX_MASK) {
 			sh_num = 0xffffffff;
+		} else if (sh_num >= AMDGPU_GFX_MAX_SH_PER_SE) {
+			ret = -EINVAL;
+			goto out;
+		}
+
+		if (info->read_mmr_reg.count > 128) {
+			ret = -EINVAL;
+			goto out;
+		}
 
 		regs = kmalloc_array(info->read_mmr_reg.count, sizeof(*regs), GFP_KERNEL);
-		if (!regs)
-			return -ENOMEM;
+		if (!regs) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
 		alloc_size = info->read_mmr_reg.count * sizeof(*regs);
 
-		for (i = 0; i < info->read_mmr_reg.count; i++)
+		amdgpu_gfx_off_ctrl(adev, false);
+		for (i = 0; i < info->read_mmr_reg.count; i++) {
 			if (amdgpu_asic_read_register(adev, se_num, sh_num,
 						      info->read_mmr_reg.dword_offset + i,
 						      &regs[i])) {
 				DRM_DEBUG_KMS("unallowed offset %#x\n",
 					      info->read_mmr_reg.dword_offset + i);
 				kfree(regs);
-				return -EFAULT;
+				amdgpu_gfx_off_ctrl(adev, true);
+				ret = -EFAULT;
+				goto out;
 			}
+		}
+		amdgpu_gfx_off_ctrl(adev, true);
 		n = copy_to_user(out, regs, min(size, alloc_size));
 		kfree(regs);
-		return n ? -EFAULT : 0;
+		ret = (n ? -EFAULT : 0);
+out:
+		up_read(&adev->reset_domain->sem);
+		return ret;
 	}
 	case AMDGPU_INFO_DEV_INFO: {
-		struct drm_amdgpu_info_device dev_info = {};
-
-		dev_info.device_id = dev->pdev->device;
-		dev_info.chip_rev = adev->rev_id;
-		dev_info.external_rev = adev->external_rev_id;
-		dev_info.pci_rev = dev->pdev->revision;
-		dev_info.family = adev->family;
-		dev_info.num_shader_engines = adev->gfx.config.max_shader_engines;
-		dev_info.num_shader_arrays_per_engine = adev->gfx.config.max_sh_per_se;
+		struct drm_amdgpu_info_device *dev_info;
+		uint64_t vm_size;
+		uint32_t pcie_gen_mask, pcie_width_mask;
+
+		dev_info = kzalloc(sizeof(*dev_info), GFP_KERNEL);
+		if (!dev_info)
+			return -ENOMEM;
+
+		dev_info->device_id = adev->pdev->device;
+		dev_info->chip_rev = adev->rev_id;
+		dev_info->external_rev = adev->external_rev_id;
+		dev_info->pci_rev = adev->pdev->revision;
+		dev_info->family = adev->family;
+		dev_info->num_shader_engines = adev->gfx.config.max_shader_engines;
+		dev_info->num_shader_arrays_per_engine = adev->gfx.config.max_sh_per_se;
 		/* return all clocks in KHz */
-		dev_info.gpu_counter_freq = amdgpu_asic_get_xclk(adev) * 10;
+		dev_info->gpu_counter_freq = amdgpu_asic_get_xclk(adev) * 10;
 		if (adev->pm.dpm_enabled) {
-			dev_info.max_engine_clock = amdgpu_dpm_get_sclk(adev, false) * 10;
-			dev_info.max_memory_clock = amdgpu_dpm_get_mclk(adev, false) * 10;
+			dev_info->max_engine_clock = amdgpu_dpm_get_sclk(adev, false) * 10;
+			dev_info->max_memory_clock = amdgpu_dpm_get_mclk(adev, false) * 10;
+			dev_info->min_engine_clock = amdgpu_dpm_get_sclk(adev, true) * 10;
+			dev_info->min_memory_clock = amdgpu_dpm_get_mclk(adev, true) * 10;
 		} else {
-			dev_info.max_engine_clock = adev->pm.default_sclk * 10;
-			dev_info.max_memory_clock = adev->pm.default_mclk * 10;
+			dev_info->max_engine_clock =
+				dev_info->min_engine_clock =
+					adev->clock.default_sclk * 10;
+			dev_info->max_memory_clock =
+				dev_info->min_memory_clock =
+					adev->clock.default_mclk * 10;
 		}
-		dev_info.enabled_rb_pipes_mask = adev->gfx.config.backend_enable_mask;
-		dev_info.num_rb_pipes = adev->gfx.config.max_backends_per_se *
+		dev_info->enabled_rb_pipes_mask = adev->gfx.config.backend_enable_mask;
+		dev_info->num_rb_pipes = adev->gfx.config.max_backends_per_se *
 			adev->gfx.config.max_shader_engines;
-		dev_info.num_hw_gfx_contexts = adev->gfx.config.max_hw_contexts;
-		dev_info._pad = 0;
-		dev_info.ids_flags = 0;
+		dev_info->num_hw_gfx_contexts = adev->gfx.config.max_hw_contexts;
+		dev_info->ids_flags = 0;
 		if (adev->flags & AMD_IS_APU)
-			dev_info.ids_flags |= AMDGPU_IDS_FLAGS_FUSION;
-		if (amdgpu_sriov_vf(adev))
-			dev_info.ids_flags |= AMDGPU_IDS_FLAGS_PREEMPTION;
-		dev_info.virtual_address_offset = AMDGPU_VA_RESERVED_SIZE;
-		dev_info.virtual_address_max = (uint64_t)adev->vm_manager.max_pfn * AMDGPU_GPU_PAGE_SIZE;
-		dev_info.virtual_address_alignment = max((int)PAGE_SIZE, AMDGPU_GPU_PAGE_SIZE);
-		dev_info.pte_fragment_size = (1 << AMDGPU_LOG2_PAGES_PER_FRAG) *
-					     AMDGPU_GPU_PAGE_SIZE;
-		dev_info.gart_page_size = AMDGPU_GPU_PAGE_SIZE;
-
-		dev_info.cu_active_number = adev->gfx.cu_info.number;
-		dev_info.cu_ao_mask = adev->gfx.cu_info.ao_cu_mask;
-		dev_info.ce_ram_size = adev->gfx.ce_ram_size;
-		memcpy(&dev_info.cu_bitmap[0], &adev->gfx.cu_info.bitmap[0],
-		       sizeof(adev->gfx.cu_info.bitmap));
-		dev_info.vram_type = adev->mc.vram_type;
-		dev_info.vram_bit_width = adev->mc.vram_width;
-		dev_info.vce_harvest_config = adev->vce.harvest_config;
-		dev_info.gc_double_offchip_lds_buf =
+			dev_info->ids_flags |= AMDGPU_IDS_FLAGS_FUSION;
+		if (adev->gfx.mcbp)
+			dev_info->ids_flags |= AMDGPU_IDS_FLAGS_PREEMPTION;
+		if (amdgpu_is_tmz(adev))
+			dev_info->ids_flags |= AMDGPU_IDS_FLAGS_TMZ;
+		if (adev->gfx.config.ta_cntl2_truncate_coord_mode)
+			dev_info->ids_flags |= AMDGPU_IDS_FLAGS_CONFORMANT_TRUNC_COORD;
+
+		/* Gang submit is not supported under SRIOV currently */
+		if (!amdgpu_sriov_vf(adev))
+			dev_info->ids_flags |= AMDGPU_IDS_FLAGS_GANG_SUBMIT;
+
+		if (amdgpu_passthrough(adev))
+			dev_info->ids_flags |= (AMDGPU_IDS_FLAGS_MODE_PT <<
+						AMDGPU_IDS_FLAGS_MODE_SHIFT) &
+						AMDGPU_IDS_FLAGS_MODE_MASK;
+		else if (amdgpu_sriov_vf(adev))
+			dev_info->ids_flags |= (AMDGPU_IDS_FLAGS_MODE_VF <<
+						AMDGPU_IDS_FLAGS_MODE_SHIFT) &
+						AMDGPU_IDS_FLAGS_MODE_MASK;
+
+		vm_size = adev->vm_manager.max_pfn * AMDGPU_GPU_PAGE_SIZE;
+		vm_size -= AMDGPU_VA_RESERVED_TOP;
+
+		/* Older VCE FW versions are buggy and can handle only 40bits */
+		if (adev->vce.fw_version &&
+		    adev->vce.fw_version < AMDGPU_VCE_FW_53_45)
+			vm_size = min(vm_size, 1ULL << 40);
+
+		dev_info->virtual_address_offset = AMDGPU_VA_RESERVED_BOTTOM;
+		dev_info->virtual_address_max =
+			min(vm_size, AMDGPU_GMC_HOLE_START);
+
+		if (vm_size > AMDGPU_GMC_HOLE_START) {
+			dev_info->high_va_offset = AMDGPU_GMC_HOLE_END;
+			dev_info->high_va_max = AMDGPU_GMC_HOLE_END | vm_size;
+		}
+		dev_info->virtual_address_alignment = max_t(u32, PAGE_SIZE, AMDGPU_GPU_PAGE_SIZE);
+		dev_info->pte_fragment_size = (1 << adev->vm_manager.fragment_size) * AMDGPU_GPU_PAGE_SIZE;
+		dev_info->gart_page_size = max_t(u32, PAGE_SIZE, AMDGPU_GPU_PAGE_SIZE);
+		dev_info->cu_active_number = adev->gfx.cu_info.number;
+		dev_info->cu_ao_mask = adev->gfx.cu_info.ao_cu_mask;
+		dev_info->ce_ram_size = adev->gfx.ce_ram_size;
+		memcpy(&dev_info->cu_ao_bitmap[0], &adev->gfx.cu_info.ao_cu_bitmap[0],
+		       sizeof(adev->gfx.cu_info.ao_cu_bitmap));
+		memcpy(&dev_info->cu_bitmap[0], &adev->gfx.cu_info.bitmap[0],
+		       sizeof(dev_info->cu_bitmap));
+		dev_info->vram_type = adev->gmc.vram_type;
+		dev_info->vram_bit_width = adev->gmc.vram_width;
+		dev_info->vce_harvest_config = adev->vce.harvest_config;
+		dev_info->gc_double_offchip_lds_buf =
 			adev->gfx.config.double_offchip_lds_buf;
-
-		if (amdgpu_ngg) {
-			dev_info.prim_buf_gpu_addr = adev->gfx.ngg.buf[NGG_PRIM].gpu_addr;
-			dev_info.prim_buf_size = adev->gfx.ngg.buf[NGG_PRIM].size;
-			dev_info.pos_buf_gpu_addr = adev->gfx.ngg.buf[NGG_POS].gpu_addr;
-			dev_info.pos_buf_size = adev->gfx.ngg.buf[NGG_POS].size;
-			dev_info.cntl_sb_buf_gpu_addr = adev->gfx.ngg.buf[NGG_CNTL].gpu_addr;
-			dev_info.cntl_sb_buf_size = adev->gfx.ngg.buf[NGG_CNTL].size;
-			dev_info.param_buf_gpu_addr = adev->gfx.ngg.buf[NGG_PARAM].gpu_addr;
-			dev_info.param_buf_size = adev->gfx.ngg.buf[NGG_PARAM].size;
+		dev_info->wave_front_size = adev->gfx.cu_info.wave_front_size;
+		dev_info->num_shader_visible_vgprs = adev->gfx.config.max_gprs;
+		dev_info->num_cu_per_sh = adev->gfx.config.max_cu_per_sh;
+		dev_info->num_tcc_blocks = adev->gfx.config.max_texture_channel_caches;
+		dev_info->gs_vgt_table_depth = adev->gfx.config.gs_vgt_table_depth;
+		dev_info->gs_prim_buffer_depth = adev->gfx.config.gs_prim_buffer_depth;
+		dev_info->max_gs_waves_per_vgt = adev->gfx.config.max_gs_threads;
+
+		if (adev->family >= AMDGPU_FAMILY_NV)
+			dev_info->pa_sc_tile_steering_override =
+				adev->gfx.config.pa_sc_tile_steering_override;
+
+		dev_info->tcc_disabled_mask = adev->gfx.config.tcc_disabled_mask;
+
+		/* Combine the chip gen mask with the platform (CPU/mobo) mask. */
+		pcie_gen_mask = adev->pm.pcie_gen_mask &
+			(adev->pm.pcie_gen_mask >> CAIL_PCIE_LINK_SPEED_SUPPORT_SHIFT);
+		pcie_width_mask = adev->pm.pcie_mlw_mask &
+			(adev->pm.pcie_mlw_mask >> CAIL_PCIE_LINK_WIDTH_SUPPORT_SHIFT);
+		dev_info->pcie_gen = fls(pcie_gen_mask);
+		dev_info->pcie_num_lanes =
+			pcie_width_mask & CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X32 ? 32 :
+			pcie_width_mask & CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X16 ? 16 :
+			pcie_width_mask & CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X12 ? 12 :
+			pcie_width_mask & CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X8 ? 8 :
+			pcie_width_mask & CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X4 ? 4 :
+			pcie_width_mask & CAIL_ASIC_PCIE_LINK_WIDTH_SUPPORT_X2 ? 2 : 1;
+
+		dev_info->tcp_cache_size = adev->gfx.config.gc_tcp_l1_size;
+		dev_info->num_sqc_per_wgp = adev->gfx.config.gc_num_sqc_per_wgp;
+		dev_info->sqc_data_cache_size = adev->gfx.config.gc_l1_data_cache_size_per_sqc;
+		dev_info->sqc_inst_cache_size = adev->gfx.config.gc_l1_instruction_cache_size_per_sqc;
+		dev_info->gl1c_cache_size = adev->gfx.config.gc_gl1c_size_per_instance *
+					    adev->gfx.config.gc_gl1c_per_sa;
+		dev_info->gl2c_cache_size = adev->gfx.config.gc_gl2c_per_gpu;
+		dev_info->mall_size = adev->gmc.mall_size;
+
+
+		if (adev->gfx.funcs->get_gfx_shadow_info) {
+			struct amdgpu_gfx_shadow_info shadow_info;
+
+			ret = amdgpu_gfx_get_gfx_shadow_info(adev, &shadow_info);
+			if (!ret) {
+				dev_info->shadow_size = shadow_info.shadow_size;
+				dev_info->shadow_alignment = shadow_info.shadow_alignment;
+				dev_info->csa_size = shadow_info.csa_size;
+				dev_info->csa_alignment = shadow_info.csa_alignment;
+			}
 		}
-		dev_info.wave_front_size = adev->gfx.cu_info.wave_front_size;
-		dev_info.num_shader_visible_vgprs = adev->gfx.config.max_gprs;
-		dev_info.num_cu_per_sh = adev->gfx.config.max_cu_per_sh;
-		dev_info.num_tcc_blocks = adev->gfx.config.max_texture_channel_caches;
-		dev_info.gs_vgt_table_depth = adev->gfx.config.gs_vgt_table_depth;
-		dev_info.gs_prim_buffer_depth = adev->gfx.config.gs_prim_buffer_depth;
-		dev_info.max_gs_waves_per_vgt = adev->gfx.config.max_gs_threads;
-
-		return copy_to_user(out, &dev_info,
-				    min((size_t)size, sizeof(dev_info))) ? -EFAULT : 0;
+
+		dev_info->userq_ip_mask = amdgpu_userq_get_supported_ip_mask(adev);
+
+		ret = copy_to_user(out, dev_info,
+				   min((size_t)size, sizeof(*dev_info))) ? -EFAULT : 0;
+		kfree(dev_info);
+		return ret;
 	}
 	case AMDGPU_INFO_VCE_CLOCK_TABLE: {
-		unsigned i;
+		unsigned int i;
 		struct drm_amdgpu_info_vce_clock_table vce_clk_table = {};
 		struct amd_vce_state *vce_state;
 
@@ -603,6 +1079,26 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 					    min((size_t)size, (size_t)(bios_size - bios_offset)))
 					? -EFAULT : 0;
 		}
+		case AMDGPU_INFO_VBIOS_INFO: {
+			struct drm_amdgpu_info_vbios vbios_info = {};
+			struct atom_context *atom_context;
+
+			atom_context = adev->mode_info.atom_context;
+			if (atom_context) {
+				memcpy(vbios_info.name, atom_context->name,
+				       sizeof(atom_context->name));
+				memcpy(vbios_info.vbios_pn, atom_context->vbios_pn,
+				       sizeof(atom_context->vbios_pn));
+				vbios_info.version = atom_context->version;
+				memcpy(vbios_info.vbios_ver_str, atom_context->vbios_ver_str,
+				       sizeof(atom_context->vbios_ver_str));
+				memcpy(vbios_info.date, atom_context->date,
+				       sizeof(atom_context->date));
+			}
+
+			return copy_to_user(out, &vbios_info,
+						min((size_t)size, sizeof(vbios_info))) ? -EFAULT : 0;
+		}
 		default:
 			DRM_DEBUG_KMS("Invalid request %d\n",
 					info->vbios_info.type);
@@ -631,10 +1127,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		}
 	}
 	case AMDGPU_INFO_SENSOR: {
-		struct pp_gpu_power query = {0};
-		int query_size = sizeof(query);
-
-		if (amdgpu_dpm == 0)
+		if (!adev->pm.dpm_enabled)
 			return -ENOENT;
 
 		switch (info->sensor_info.type) {
@@ -675,11 +1168,25 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		case AMDGPU_INFO_SENSOR_GPU_AVG_POWER:
 			/* get average GPU power */
 			if (amdgpu_dpm_read_sensor(adev,
-						   AMDGPU_PP_SENSOR_GPU_POWER,
-						   (void *)&query, &query_size)) {
+						   AMDGPU_PP_SENSOR_GPU_AVG_POWER,
+						   (void *)&ui32, &ui32_size)) {
+				/* fall back to input power for backwards compat */
+				if (amdgpu_dpm_read_sensor(adev,
+							   AMDGPU_PP_SENSOR_GPU_INPUT_POWER,
+							   (void *)&ui32, &ui32_size)) {
+					return -EINVAL;
+				}
+			}
+			ui32 >>= 8;
+			break;
+		case AMDGPU_INFO_SENSOR_GPU_INPUT_POWER:
+			/* get input GPU power */
+			if (amdgpu_dpm_read_sensor(adev,
+						   AMDGPU_PP_SENSOR_GPU_INPUT_POWER,
+						   (void *)&ui32, &ui32_size)) {
 				return -EINVAL;
 			}
-			ui32 = query.average_gpu_power >> 8;
+			ui32 >>= 8;
 			break;
 		case AMDGPU_INFO_SENSOR_VDDNB:
 			/* get VDDNB in millivolts */
@@ -697,6 +1204,42 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 				return -EINVAL;
 			}
 			break;
+		case AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_SCLK:
+			/* get stable pstate sclk in Mhz */
+			if (amdgpu_dpm_read_sensor(adev,
+						   AMDGPU_PP_SENSOR_STABLE_PSTATE_SCLK,
+						   (void *)&ui32, &ui32_size)) {
+				return -EINVAL;
+			}
+			ui32 /= 100;
+			break;
+		case AMDGPU_INFO_SENSOR_STABLE_PSTATE_GFX_MCLK:
+			/* get stable pstate mclk in Mhz */
+			if (amdgpu_dpm_read_sensor(adev,
+						   AMDGPU_PP_SENSOR_STABLE_PSTATE_MCLK,
+						   (void *)&ui32, &ui32_size)) {
+				return -EINVAL;
+			}
+			ui32 /= 100;
+			break;
+		case AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_SCLK:
+			/* get peak pstate sclk in Mhz */
+			if (amdgpu_dpm_read_sensor(adev,
+						   AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK,
+						   (void *)&ui32, &ui32_size)) {
+				return -EINVAL;
+			}
+			ui32 /= 100;
+			break;
+		case AMDGPU_INFO_SENSOR_PEAK_PSTATE_GFX_MCLK:
+			/* get peak pstate mclk in Mhz */
+			if (amdgpu_dpm_read_sensor(adev,
+						   AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK,
+						   (void *)&ui32, &ui32_size)) {
+				return -EINVAL;
+			}
+			ui32 /= 100;
+			break;
 		default:
 			DRM_DEBUG_KMS("Invalid request %d\n",
 				      info->sensor_info.type);
@@ -704,6 +1247,126 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 		}
 		return copy_to_user(out, &ui32, min(size, 4u)) ? -EFAULT : 0;
 	}
+	case AMDGPU_INFO_VRAM_LOST_COUNTER:
+		ui32 = atomic_read(&adev->vram_lost_counter);
+		return copy_to_user(out, &ui32, min(size, 4u)) ? -EFAULT : 0;
+	case AMDGPU_INFO_RAS_ENABLED_FEATURES: {
+		struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+		uint64_t ras_mask;
+
+		if (!ras)
+			return -EINVAL;
+		ras_mask = (uint64_t)adev->ras_enabled << 32 | ras->features;
+
+		return copy_to_user(out, &ras_mask,
+				min_t(u64, size, sizeof(ras_mask))) ?
+			-EFAULT : 0;
+	}
+	case AMDGPU_INFO_VIDEO_CAPS: {
+		const struct amdgpu_video_codecs *codecs;
+		struct drm_amdgpu_info_video_caps *caps;
+		int r;
+
+		if (!adev->asic_funcs->query_video_codecs)
+			return -EINVAL;
+
+		switch (info->video_cap.type) {
+		case AMDGPU_INFO_VIDEO_CAPS_DECODE:
+			r = amdgpu_asic_query_video_codecs(adev, false, &codecs);
+			if (r)
+				return -EINVAL;
+			break;
+		case AMDGPU_INFO_VIDEO_CAPS_ENCODE:
+			r = amdgpu_asic_query_video_codecs(adev, true, &codecs);
+			if (r)
+				return -EINVAL;
+			break;
+		default:
+			DRM_DEBUG_KMS("Invalid request %d\n",
+				      info->video_cap.type);
+			return -EINVAL;
+		}
+
+		caps = kzalloc(sizeof(*caps), GFP_KERNEL);
+		if (!caps)
+			return -ENOMEM;
+
+		for (i = 0; i < codecs->codec_count; i++) {
+			int idx = codecs->codec_array[i].codec_type;
+
+			switch (idx) {
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2:
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4:
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1:
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC:
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC:
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG:
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9:
+			case AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_AV1:
+				caps->codec_info[idx].valid = 1;
+				caps->codec_info[idx].max_width =
+					codecs->codec_array[i].max_width;
+				caps->codec_info[idx].max_height =
+					codecs->codec_array[i].max_height;
+				caps->codec_info[idx].max_pixels_per_frame =
+					codecs->codec_array[i].max_pixels_per_frame;
+				caps->codec_info[idx].max_level =
+					codecs->codec_array[i].max_level;
+				break;
+			default:
+				break;
+			}
+		}
+		r = copy_to_user(out, caps,
+				 min((size_t)size, sizeof(*caps))) ? -EFAULT : 0;
+		kfree(caps);
+		return r;
+	}
+	case AMDGPU_INFO_MAX_IBS: {
+		uint32_t max_ibs[AMDGPU_HW_IP_NUM];
+
+		for (i = 0; i < AMDGPU_HW_IP_NUM; ++i)
+			max_ibs[i] = amdgpu_ring_max_ibs(i);
+
+		return copy_to_user(out, max_ibs,
+				    min((size_t)size, sizeof(max_ibs))) ? -EFAULT : 0;
+	}
+	case AMDGPU_INFO_GPUVM_FAULT: {
+		struct amdgpu_fpriv *fpriv = filp->driver_priv;
+		struct amdgpu_vm *vm = &fpriv->vm;
+		struct drm_amdgpu_info_gpuvm_fault gpuvm_fault;
+		unsigned long flags;
+
+		if (!vm)
+			return -EINVAL;
+
+		memset(&gpuvm_fault, 0, sizeof(gpuvm_fault));
+
+		xa_lock_irqsave(&adev->vm_manager.pasids, flags);
+		gpuvm_fault.addr = vm->fault_info.addr;
+		gpuvm_fault.status = vm->fault_info.status;
+		gpuvm_fault.vmhub = vm->fault_info.vmhub;
+		xa_unlock_irqrestore(&adev->vm_manager.pasids, flags);
+
+		return copy_to_user(out, &gpuvm_fault,
+				    min((size_t)size, sizeof(gpuvm_fault))) ? -EFAULT : 0;
+	}
+	case AMDGPU_INFO_UQ_FW_AREAS: {
+		struct drm_amdgpu_info_uq_metadata meta_info = {};
+
+		switch (info->query_hw_ip.type) {
+		case AMDGPU_HW_IP_GFX:
+			ret = amdgpu_userq_metadata_info_gfx(adev, info, &meta_info.gfx);
+			if (ret)
+				return ret;
+
+			ret = copy_to_user(out, &meta_info,
+						min((size_t)size, sizeof(meta_info))) ? -EFAULT : 0;
+			return 0;
+		default:
+			return -EINVAL;
+		}
+	}
 	default:
 		DRM_DEBUG_KMS("Invalid request %d\n", info->query);
 		return -EINVAL;
@@ -711,25 +1374,6 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 	return 0;
 }
 
-
-/*
- * Outdated mess for old drm with Xorg being in charge (void function now).
- */
-/**
- * amdgpu_driver_lastclose_kms - drm callback for last close
- *
- * @dev: drm dev pointer
- *
- * Switch vga_switcheroo state after last close (all asics).
- */
-void amdgpu_driver_lastclose_kms(struct drm_device *dev)
-{
-	struct amdgpu_device *adev = dev->dev_private;
-
-	amdgpu_fbdev_restore_mode(adev);
-	vga_switcheroo_process_delayed_switch();
-}
-
 /**
  * amdgpu_driver_open_kms - drm callback for open
  *
@@ -741,15 +1385,24 @@ void amdgpu_driver_lastclose_kms(struct drm_device *dev)
  */
 int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_fpriv *fpriv;
-	int r;
+	int r, pasid;
+
+	/* Ensure IB tests are run on ring */
+	flush_delayed_work(&adev->delayed_init_work);
+
+
+	if (amdgpu_ras_intr_triggered()) {
+		DRM_ERROR("RAS Intr triggered, device disabled!!");
+		return -EHWPOISON;
+	}
 
 	file_priv->driver_priv = NULL;
 
 	r = pm_runtime_get_sync(dev->dev);
 	if (r < 0)
-		return r;
+		goto pm_put;
 
 	fpriv = kzalloc(sizeof(*fpriv), GFP_KERNEL);
 	if (unlikely(!fpriv)) {
@@ -757,35 +1410,68 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
 		goto out_suspend;
 	}
 
-	r = amdgpu_vm_init(adev, &fpriv->vm);
-	if (r) {
-		kfree(fpriv);
-		goto out_suspend;
+	pasid = amdgpu_pasid_alloc(16);
+	if (pasid < 0) {
+		dev_warn(adev->dev, "No more PASIDs available!");
+		pasid = 0;
 	}
 
+	r = amdgpu_xcp_open_device(adev, fpriv, file_priv);
+	if (r)
+		goto error_pasid;
+
+	amdgpu_debugfs_vm_init(file_priv);
+
+	r = amdgpu_vm_init(adev, &fpriv->vm, fpriv->xcp_id, pasid);
+	if (r)
+		goto error_pasid;
+
 	fpriv->prt_va = amdgpu_vm_bo_add(adev, &fpriv->vm, NULL);
 	if (!fpriv->prt_va) {
 		r = -ENOMEM;
-		amdgpu_vm_fini(adev, &fpriv->vm);
-		kfree(fpriv);
-		goto out_suspend;
+		goto error_vm;
 	}
 
-	if (amdgpu_sriov_vf(adev)) {
-		r = amdgpu_map_static_csa(adev, &fpriv->vm);
+	if (adev->gfx.mcbp) {
+		uint64_t csa_addr = amdgpu_csa_vaddr(adev) & AMDGPU_GMC_HOLE_MASK;
+
+		r = amdgpu_map_static_csa(adev, &fpriv->vm, adev->virt.csa_obj,
+						&fpriv->csa_va, csa_addr, AMDGPU_CSA_SIZE);
 		if (r)
-			goto out_suspend;
+			goto error_vm;
 	}
 
+	r = amdgpu_seq64_map(adev, &fpriv->vm, &fpriv->seq64_va);
+	if (r)
+		goto error_vm;
+
 	mutex_init(&fpriv->bo_list_lock);
-	idr_init(&fpriv->bo_list_handles);
+	idr_init_base(&fpriv->bo_list_handles, 1);
+
+	r = amdgpu_userq_mgr_init(&fpriv->userq_mgr, file_priv, adev);
+	if (r)
+		DRM_WARN("Can't setup usermode queues, use legacy workload submission only\n");
 
-	amdgpu_ctx_mgr_init(&fpriv->ctx_mgr);
+	r = amdgpu_eviction_fence_init(&fpriv->evf_mgr);
+	if (r)
+		goto error_vm;
+
+	amdgpu_ctx_mgr_init(&fpriv->ctx_mgr, adev);
 
 	file_priv->driver_priv = fpriv;
+	goto out_suspend;
+
+error_vm:
+	amdgpu_vm_fini(adev, &fpriv->vm);
+
+error_pasid:
+	if (pasid)
+		amdgpu_pasid_free(pasid);
+
+	kfree(fpriv);
 
 out_suspend:
-	pm_runtime_mark_last_busy(dev->dev);
+pm_put:
 	pm_runtime_put_autosuspend(dev->dev);
 
 	return r;
@@ -802,9 +1488,11 @@ out_suspend:
 void amdgpu_driver_postclose_kms(struct drm_device *dev,
 				 struct drm_file *file_priv)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
 	struct amdgpu_bo_list *list;
+	struct amdgpu_bo *pd;
+	u32 pasid;
 	int handle;
 
 	if (!fpriv)
@@ -812,25 +1500,37 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
 
 	pm_runtime_get_sync(dev->dev);
 
-	amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
+	if (amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_UVD) != NULL)
+		amdgpu_uvd_free_handles(adev, file_priv);
+	if (amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_VCE) != NULL)
+		amdgpu_vce_free_handles(adev, file_priv);
+
+	if (fpriv->csa_va) {
+		uint64_t csa_addr = amdgpu_csa_vaddr(adev) & AMDGPU_GMC_HOLE_MASK;
 
-	amdgpu_uvd_free_handles(adev, file_priv);
-	amdgpu_vce_free_handles(adev, file_priv);
+		WARN_ON(amdgpu_unmap_static_csa(adev, &fpriv->vm, adev->virt.csa_obj,
+						fpriv->csa_va, csa_addr));
+		fpriv->csa_va = NULL;
+	}
 
-	amdgpu_vm_bo_rmv(adev, fpriv->prt_va);
+	amdgpu_seq64_unmap(adev, fpriv);
 
-	if (amdgpu_sriov_vf(adev)) {
-		/* TODO: how to handle reserve failure */
-		BUG_ON(amdgpu_bo_reserve(adev->virt.csa_obj, true));
-		amdgpu_vm_bo_rmv(adev, fpriv->vm.csa_bo_va);
-		fpriv->vm.csa_bo_va = NULL;
-		amdgpu_bo_unreserve(adev->virt.csa_obj);
+	pasid = fpriv->vm.pasid;
+	pd = amdgpu_bo_ref(fpriv->vm.root.bo);
+	if (!WARN_ON(amdgpu_bo_reserve(pd, true))) {
+		amdgpu_vm_bo_del(adev, fpriv->prt_va);
+		amdgpu_bo_unreserve(pd);
 	}
 
+	amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
 	amdgpu_vm_fini(adev, &fpriv->vm);
 
+	if (pasid)
+		amdgpu_pasid_free_delayed(pd->tbo.base.resv, pasid);
+	amdgpu_bo_unref(&pd);
+
 	idr_for_each_entry(&fpriv->bo_list_handles, list, handle)
-		amdgpu_bo_list_free(list);
+		amdgpu_bo_list_put(list);
 
 	idr_destroy(&fpriv->bo_list_handles);
 	mutex_destroy(&fpriv->bo_list_lock);
@@ -838,25 +1538,34 @@ void amdgpu_driver_postclose_kms(struct drm_device *dev,
 	kfree(fpriv);
 	file_priv->driver_priv = NULL;
 
-	pm_runtime_mark_last_busy(dev->dev);
 	pm_runtime_put_autosuspend(dev->dev);
 }
 
+
+void amdgpu_driver_release_kms(struct drm_device *dev)
+{
+	struct amdgpu_device *adev = drm_to_adev(dev);
+
+	amdgpu_device_fini_sw(adev);
+	pci_set_drvdata(adev->pdev, NULL);
+}
+
 /*
  * VBlank related functions.
  */
 /**
  * amdgpu_get_vblank_counter_kms - get frame count
  *
- * @dev: drm dev pointer
- * @pipe: crtc to get the frame count from
+ * @crtc: crtc to get the frame count from
  *
  * Gets the frame count on the requested crtc (all asics).
  * Returns frame count on success, -EINVAL on failure.
  */
-u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe)
+u32 amdgpu_get_vblank_counter_kms(struct drm_crtc *crtc)
 {
-	struct amdgpu_device *adev = dev->dev_private;
+	struct drm_device *dev = crtc->dev;
+	unsigned int pipe = crtc->index;
+	struct amdgpu_device *adev = drm_to_adev(dev);
 	int vpos, hpos, stat;
 	u32 count;
 
@@ -879,11 +1588,11 @@ u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe)
 		 */
 		do {
 			count = amdgpu_display_vblank_get_counter(adev, pipe);
-			/* Ask amdgpu_get_crtc_scanoutpos to return vpos as
-			 * distance to start of vblank, instead of regular
-			 * vertical scanout pos.
+			/* Ask amdgpu_display_get_crtc_scanoutpos to return
+			 * vpos as distance to start of vblank, instead of
+			 * regular vertical scanout pos.
 			 */
-			stat = amdgpu_get_crtc_scanoutpos(
+			stat = amdgpu_display_get_crtc_scanoutpos(
 				dev, pipe, GET_DISTANCE_TO_VBLANKSTART,
 				&vpos, &hpos, NULL, NULL,
 				&adev->mode_info.crtcs[pipe]->base.hwmode);
@@ -915,16 +1624,17 @@ u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe)
 /**
  * amdgpu_enable_vblank_kms - enable vblank interrupt
  *
- * @dev: drm dev pointer
- * @pipe: crtc to enable vblank interrupt for
+ * @crtc: crtc to enable vblank interrupt for
  *
  * Enable the interrupt on the requested crtc (all asics).
  * Returns 0 on success, -EINVAL on failure.
  */
-int amdgpu_enable_vblank_kms(struct drm_device *dev, unsigned int pipe)
+int amdgpu_enable_vblank_kms(struct drm_crtc *crtc)
 {
-	struct amdgpu_device *adev = dev->dev_private;
-	int idx = amdgpu_crtc_idx_to_irq_type(adev, pipe);
+	struct drm_device *dev = crtc->dev;
+	unsigned int pipe = crtc->index;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	int idx = amdgpu_display_crtc_idx_to_irq_type(adev, pipe);
 
 	return amdgpu_irq_get(adev, &adev->crtc_irq, idx);
 }
@@ -932,92 +1642,45 @@ int amdgpu_enable_vblank_kms(struct drm_device *dev, unsigned int pipe)
 /**
  * amdgpu_disable_vblank_kms - disable vblank interrupt
  *
- * @dev: drm dev pointer
- * @pipe: crtc to disable vblank interrupt for
+ * @crtc: crtc to disable vblank interrupt for
  *
  * Disable the interrupt on the requested crtc (all asics).
  */
-void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
+void amdgpu_disable_vblank_kms(struct drm_crtc *crtc)
 {
-	struct amdgpu_device *adev = dev->dev_private;
-	int idx = amdgpu_crtc_idx_to_irq_type(adev, pipe);
+	struct drm_device *dev = crtc->dev;
+	unsigned int pipe = crtc->index;
+	struct amdgpu_device *adev = drm_to_adev(dev);
+	int idx = amdgpu_display_crtc_idx_to_irq_type(adev, pipe);
 
 	amdgpu_irq_put(adev, &adev->crtc_irq, idx);
 }
 
-/**
- * amdgpu_get_vblank_timestamp_kms - get vblank timestamp
- *
- * @dev: drm dev pointer
- * @crtc: crtc to get the timestamp for
- * @max_error: max error
- * @vblank_time: time value
- * @flags: flags passed to the driver
- *
- * Gets the timestamp on the requested crtc based on the
- * scanout position.  (all asics).
- * Returns postive status flags on success, negative error on failure.
- */
-int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, unsigned int pipe,
-				    int *max_error,
-				    struct timeval *vblank_time,
-				    unsigned flags)
-{
-	struct drm_crtc *crtc;
-	struct amdgpu_device *adev = dev->dev_private;
-
-	if (pipe >= dev->num_crtcs) {
-		DRM_ERROR("Invalid crtc %u\n", pipe);
-		return -EINVAL;
-	}
-
-	/* Get associated drm_crtc: */
-	crtc = &adev->mode_info.crtcs[pipe]->base;
-	if (!crtc) {
-		/* This can occur on driver load if some component fails to
-		 * initialize completely and driver is unloaded */
-		DRM_ERROR("Uninitialized crtc %d\n", pipe);
-		return -EINVAL;
-	}
-
-	/* Helper routine in DRM core does all the work: */
-	return drm_calc_vbltimestamp_from_scanoutpos(dev, pipe, max_error,
-						     vblank_time, flags,
-						     &crtc->hwmode);
-}
-
-const struct drm_ioctl_desc amdgpu_ioctls_kms[] = {
-	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_CREATE, amdgpu_gem_create_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_CTX, amdgpu_ctx_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_BO_LIST, amdgpu_bo_list_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	/* KMS */
-	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_MMAP, amdgpu_gem_mmap_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_WAIT_IDLE, amdgpu_gem_wait_idle_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_CS, amdgpu_cs_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_INFO, amdgpu_info_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_CS, amdgpu_cs_wait_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_FENCES, amdgpu_cs_wait_fences_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_METADATA, amdgpu_gem_metadata_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
-};
-const int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);
-
 /*
  * Debugfs info
  */
 #if defined(CONFIG_DEBUG_FS)
 
-static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
+static int amdgpu_debugfs_firmware_info_show(struct seq_file *m, void *unused)
 {
-	struct drm_info_node *node = (struct drm_info_node *) m->private;
-	struct drm_device *dev = node->minor->dev;
-	struct amdgpu_device *adev = dev->dev_private;
+	struct amdgpu_device *adev = m->private;
 	struct drm_amdgpu_info_firmware fw_info;
 	struct drm_amdgpu_query_fw query_fw;
+	struct atom_context *ctx = adev->mode_info.atom_context;
+	uint8_t smu_program, smu_major, smu_minor, smu_debug;
 	int ret, i;
 
+	static const char *ta_fw_name[TA_FW_TYPE_MAX_INDEX] = {
+#define TA_FW_NAME(type)[TA_FW_TYPE_PSP_##type] = #type
+		TA_FW_NAME(XGMI),
+		TA_FW_NAME(RAS),
+		TA_FW_NAME(HDCP),
+		TA_FW_NAME(DTM),
+		TA_FW_NAME(RAP),
+		TA_FW_NAME(SECUREDISPLAY),
+#undef TA_FW_NAME
+	};
+
 	/* VCE */
 	query_fw.fw_type = AMDGPU_INFO_FW_VCE;
 	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
@@ -1074,6 +1737,46 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 	seq_printf(m, "RLC feature version: %u, firmware version: 0x%08x\n",
 		   fw_info.feature, fw_info.ver);
 
+	/* RLC SAVE RESTORE LIST CNTL */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_CNTL;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLC SRLC feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* RLC SAVE RESTORE LIST GPM MEM */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_GPM_MEM;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLC SRLG feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* RLC SAVE RESTORE LIST SRM MEM */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLC SRLS feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* RLCP */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLCP;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLCP feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* RLCV */
+	query_fw.fw_type = AMDGPU_INFO_FW_GFX_RLCV;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "RLCV feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
 	/* MEC */
 	query_fw.fw_type = AMDGPU_INFO_FW_GFX_MEC;
 	query_fw.index = 0;
@@ -1084,8 +1787,7 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 		   fw_info.feature, fw_info.ver);
 
 	/* MEC2 */
-	if (adev->asic_type == CHIP_KAVERI ||
-	    (adev->asic_type > CHIP_TOPAZ && adev->asic_type != CHIP_STONEY)) {
+	if (adev->gfx.mec2_fw) {
 		query_fw.index = 1;
 		ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
 		if (ret)
@@ -1094,6 +1796,15 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 			   fw_info.feature, fw_info.ver);
 	}
 
+	/* IMU */
+	query_fw.fw_type = AMDGPU_INFO_FW_IMU;
+	query_fw.index = 0;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "IMU feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
 	/* PSP SOS */
 	query_fw.fw_type = AMDGPU_INFO_FW_SOS;
 	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
@@ -1111,13 +1822,28 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 	seq_printf(m, "ASD feature version: %u, firmware version: 0x%08x\n",
 		   fw_info.feature, fw_info.ver);
 
+	query_fw.fw_type = AMDGPU_INFO_FW_TA;
+	for (i = TA_FW_TYPE_PSP_XGMI; i < TA_FW_TYPE_MAX_INDEX; i++) {
+		query_fw.index = i;
+		ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+		if (ret)
+			continue;
+
+		seq_printf(m, "TA %s feature version: 0x%08x, firmware version: 0x%08x\n",
+			   ta_fw_name[i], fw_info.feature, fw_info.ver);
+	}
+
 	/* SMC */
 	query_fw.fw_type = AMDGPU_INFO_FW_SMC;
 	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
 	if (ret)
 		return ret;
-	seq_printf(m, "SMC feature version: %u, firmware version: 0x%08x\n",
-		   fw_info.feature, fw_info.ver);
+	smu_program = (fw_info.ver >> 24) & 0xff;
+	smu_major = (fw_info.ver >> 16) & 0xff;
+	smu_minor = (fw_info.ver >> 8) & 0xff;
+	smu_debug = (fw_info.ver >> 0) & 0xff;
+	seq_printf(m, "SMC feature version: %u, program: %d, firmware version: 0x%08x (%d.%d.%d)\n",
+		   fw_info.feature, smu_program, fw_info.ver, smu_major, smu_minor, smu_debug);
 
 	/* SDMA */
 	query_fw.fw_type = AMDGPU_INFO_FW_SDMA;
@@ -1130,20 +1856,89 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 			   i, fw_info.feature, fw_info.ver);
 	}
 
+	/* VCN */
+	query_fw.fw_type = AMDGPU_INFO_FW_VCN;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "VCN feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* DMCU */
+	query_fw.fw_type = AMDGPU_INFO_FW_DMCU;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "DMCU feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* DMCUB */
+	query_fw.fw_type = AMDGPU_INFO_FW_DMCUB;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "DMCUB feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* TOC */
+	query_fw.fw_type = AMDGPU_INFO_FW_TOC;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "TOC feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* CAP */
+	if (adev->psp.cap_fw) {
+		query_fw.fw_type = AMDGPU_INFO_FW_CAP;
+		ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+		if (ret)
+			return ret;
+		seq_printf(m, "CAP feature version: %u, firmware version: 0x%08x\n",
+				fw_info.feature, fw_info.ver);
+	}
+
+	/* MES_KIQ */
+	query_fw.fw_type = AMDGPU_INFO_FW_MES_KIQ;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "MES_KIQ feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* MES */
+	query_fw.fw_type = AMDGPU_INFO_FW_MES;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "MES feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	/* VPE */
+	query_fw.fw_type = AMDGPU_INFO_FW_VPE;
+	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+	if (ret)
+		return ret;
+	seq_printf(m, "VPE feature version: %u, firmware version: 0x%08x\n",
+		   fw_info.feature, fw_info.ver);
+
+	seq_printf(m, "VBIOS version: %s\n", ctx->vbios_pn);
+
 	return 0;
 }
 
-static const struct drm_info_list amdgpu_firmware_info_list[] = {
-	{"amdgpu_firmware_info", amdgpu_debugfs_firmware_info, 0, NULL},
-};
+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_firmware_info);
+
 #endif
 
-int amdgpu_debugfs_firmware_init(struct amdgpu_device *adev)
+void amdgpu_debugfs_firmware_init(struct amdgpu_device *adev)
 {
 #if defined(CONFIG_DEBUG_FS)
-	return amdgpu_debugfs_add_files(adev, amdgpu_firmware_info_list,
-					ARRAY_SIZE(amdgpu_firmware_info_list));
-#else
-	return 0;
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+
+	debugfs_create_file("amdgpu_firmware_info", 0444, root,
+			    adev, &amdgpu_debugfs_firmware_info_fops);
+
 #endif
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_lsdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_lsdma.c
new file mode 100644
index 000000000000..4d1d4994ea3f
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_lsdma.c
@@ -0,0 +1,91 @@
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "amdgpu.h"
+#include "amdgpu_lsdma.h"
+
+#define AMDGPU_LSDMA_MAX_SIZE	0x2000000ULL
+
+int amdgpu_lsdma_wait_for(struct amdgpu_device *adev,
+			  uint32_t reg_index, uint32_t reg_val,
+			  uint32_t mask)
+{
+	uint32_t val;
+	int i;
+
+	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32(reg_index);
+		if ((val & mask) == reg_val)
+			return 0;
+		udelay(1);
+	}
+
+	return -ETIME;
+}
+
+int amdgpu_lsdma_copy_mem(struct amdgpu_device *adev,
+			  uint64_t src_addr,
+			  uint64_t dst_addr,
+			  uint64_t mem_size)
+{
+	int ret;
+
+	if (mem_size == 0)
+		return -EINVAL;
+
+	while (mem_size > 0) {
+		uint64_t current_copy_size = min(mem_size, AMDGPU_LSDMA_MAX_SIZE);
+
+		ret = adev->lsdma.funcs->copy_mem(adev, src_addr, dst_addr, current_copy_size);
+		if (ret)
+			return ret;
+		src_addr += current_copy_size;
+		dst_addr += current_copy_size;
+		mem_size -= current_copy_size;
+	}
+
+	return 0;
+}
+
+int amdgpu_lsdma_fill_mem(struct amdgpu_device *adev,
+			  uint64_t dst_addr,
+			  uint32_t data,
+			  uint64_t mem_size)
+{
+	int ret;
+
+	if (mem_size == 0)
+		return -EINVAL;
+
+	while (mem_size > 0) {
+		uint64_t current_fill_size = min(mem_size, AMDGPU_LSDMA_MAX_SIZE);
+
+		ret = adev->lsdma.funcs->fill_mem(adev, dst_addr, data, current_fill_size);
+		if (ret)
+			return ret;
+		dst_addr += current_fill_size;
+		mem_size -= current_fill_size;
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_lsdma.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_lsdma.h
new file mode 100644
index 000000000000..c61ba58c5ee0
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_lsdma.h
@@ -0,0 +1,46 @@
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_LSDMA_H__
+#define __AMDGPU_LSDMA_H__
+
+struct amdgpu_lsdma {
+	const struct amdgpu_lsdma_funcs      *funcs;
+};
+
+struct amdgpu_lsdma_funcs {
+	int (*copy_mem)(struct amdgpu_device *adev, uint64_t src_addr,
+			uint64_t dst_addr, uint64_t size);
+	int (*fill_mem)(struct amdgpu_device *adev, uint64_t dst_addr,
+			uint32_t data, uint64_t size);
+	void (*update_memory_power_gating)(struct amdgpu_device *adev, bool enable);
+};
+
+int amdgpu_lsdma_copy_mem(struct amdgpu_device *adev, uint64_t src_addr,
+			  uint64_t dst_addr, uint64_t mem_size);
+int amdgpu_lsdma_fill_mem(struct amdgpu_device *adev, uint64_t dst_addr,
+			  uint32_t data, uint64_t mem_size);
+int amdgpu_lsdma_wait_for(struct amdgpu_device *adev, uint32_t reg_index,
+			  uint32_t reg_val, uint32_t mask);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mca.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mca.c
new file mode 100644
index 000000000000..3ca03b5e0f91
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mca.c
@@ -0,0 +1,630 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include "amdgpu_ras.h"
+#include "amdgpu.h"
+#include "amdgpu_mca.h"
+
+#include "umc/umc_6_7_0_offset.h"
+#include "umc/umc_6_7_0_sh_mask.h"
+
+static bool amdgpu_mca_is_deferred_error(struct amdgpu_device *adev,
+					uint64_t mc_status)
+{
+	if (adev->umc.ras->check_ecc_err_status)
+		return adev->umc.ras->check_ecc_err_status(adev,
+				AMDGPU_MCA_ERROR_TYPE_DE, &mc_status);
+
+	return false;
+}
+
+void amdgpu_mca_query_correctable_error_count(struct amdgpu_device *adev,
+					      uint64_t mc_status_addr,
+					      unsigned long *error_count)
+{
+	uint64_t mc_status = RREG64_PCIE(mc_status_addr);
+
+	if (REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
+	    REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, CECC) == 1)
+		*error_count += 1;
+}
+
+void amdgpu_mca_query_uncorrectable_error_count(struct amdgpu_device *adev,
+						uint64_t mc_status_addr,
+						unsigned long *error_count)
+{
+	uint64_t mc_status = RREG64_PCIE(mc_status_addr);
+
+	if ((REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1) &&
+	    (REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Deferred) == 1 ||
+	    REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
+	    REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
+	    REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
+	    REG_GET_FIELD(mc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1))
+		*error_count += 1;
+}
+
+void amdgpu_mca_reset_error_count(struct amdgpu_device *adev,
+				  uint64_t mc_status_addr)
+{
+	WREG64_PCIE(mc_status_addr, 0x0ULL);
+}
+
+void amdgpu_mca_query_ras_error_count(struct amdgpu_device *adev,
+				      uint64_t mc_status_addr,
+				      void *ras_error_status)
+{
+	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
+
+	amdgpu_mca_query_correctable_error_count(adev, mc_status_addr, &(err_data->ce_count));
+	amdgpu_mca_query_uncorrectable_error_count(adev, mc_status_addr, &(err_data->ue_count));
+
+	amdgpu_mca_reset_error_count(adev, mc_status_addr);
+}
+
+int amdgpu_mca_mp0_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err;
+	struct amdgpu_mca_ras_block *ras;
+
+	if (!adev->mca.mp0.ras)
+		return 0;
+
+	ras = adev->mca.mp0.ras;
+
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register mca.mp0 ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "mca.mp0");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+	adev->mca.mp0.ras_if = &ras->ras_block.ras_comm;
+
+	return 0;
+}
+
+int amdgpu_mca_mp1_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err;
+	struct amdgpu_mca_ras_block *ras;
+
+	if (!adev->mca.mp1.ras)
+		return 0;
+
+	ras = adev->mca.mp1.ras;
+
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register mca.mp1 ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "mca.mp1");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+	adev->mca.mp1.ras_if = &ras->ras_block.ras_comm;
+
+	return 0;
+}
+
+int amdgpu_mca_mpio_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err;
+	struct amdgpu_mca_ras_block *ras;
+
+	if (!adev->mca.mpio.ras)
+		return 0;
+
+	ras = adev->mca.mpio.ras;
+
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register mca.mpio ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "mca.mpio");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MCA;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+	adev->mca.mpio.ras_if = &ras->ras_block.ras_comm;
+
+	return 0;
+}
+
+static void amdgpu_mca_bank_set_init(struct mca_bank_set *mca_set)
+{
+	if (!mca_set)
+		return;
+
+	memset(mca_set, 0, sizeof(*mca_set));
+	INIT_LIST_HEAD(&mca_set->list);
+}
+
+static int amdgpu_mca_bank_set_add_entry(struct mca_bank_set *mca_set, struct mca_bank_entry *entry)
+{
+	struct mca_bank_node *node;
+
+	if (!entry)
+		return -EINVAL;
+
+	node = kvzalloc(sizeof(*node), GFP_KERNEL);
+	if (!node)
+		return -ENOMEM;
+
+	memcpy(&node->entry, entry, sizeof(*entry));
+
+	INIT_LIST_HEAD(&node->node);
+	list_add_tail(&node->node, &mca_set->list);
+
+	mca_set->nr_entries++;
+
+	return 0;
+}
+
+static int amdgpu_mca_bank_set_merge(struct mca_bank_set *mca_set, struct mca_bank_set *new)
+{
+	struct mca_bank_node *node;
+
+	list_for_each_entry(node, &new->list, node)
+		amdgpu_mca_bank_set_add_entry(mca_set, &node->entry);
+
+	return 0;
+}
+
+static void amdgpu_mca_bank_set_remove_node(struct mca_bank_set *mca_set, struct mca_bank_node *node)
+{
+	if (!node)
+		return;
+
+	list_del(&node->node);
+	kvfree(node);
+
+	mca_set->nr_entries--;
+}
+
+static void amdgpu_mca_bank_set_release(struct mca_bank_set *mca_set)
+{
+	struct mca_bank_node *node, *tmp;
+
+	if (list_empty(&mca_set->list))
+		return;
+
+	list_for_each_entry_safe(node, tmp, &mca_set->list, node)
+		amdgpu_mca_bank_set_remove_node(mca_set, node);
+}
+
+void amdgpu_mca_smu_init_funcs(struct amdgpu_device *adev, const struct amdgpu_mca_smu_funcs *mca_funcs)
+{
+	struct amdgpu_mca *mca = &adev->mca;
+
+	mca->mca_funcs = mca_funcs;
+}
+
+int amdgpu_mca_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_mca *mca = &adev->mca;
+	struct mca_bank_cache *mca_cache;
+	int i;
+
+	atomic_set(&mca->ue_update_flag, 0);
+
+	for (i = 0; i < ARRAY_SIZE(mca->mca_caches); i++) {
+		mca_cache = &mca->mca_caches[i];
+		mutex_init(&mca_cache->lock);
+		amdgpu_mca_bank_set_init(&mca_cache->mca_set);
+	}
+
+	return 0;
+}
+
+void amdgpu_mca_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_mca *mca = &adev->mca;
+	struct mca_bank_cache *mca_cache;
+	int i;
+
+	atomic_set(&mca->ue_update_flag, 0);
+
+	for (i = 0; i < ARRAY_SIZE(mca->mca_caches); i++) {
+		mca_cache = &mca->mca_caches[i];
+		amdgpu_mca_bank_set_release(&mca_cache->mca_set);
+		mutex_destroy(&mca_cache->lock);
+	}
+}
+
+int amdgpu_mca_reset(struct amdgpu_device *adev)
+{
+	amdgpu_mca_fini(adev);
+
+	return amdgpu_mca_init(adev);
+}
+
+int amdgpu_mca_smu_set_debug_mode(struct amdgpu_device *adev, bool enable)
+{
+	const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
+
+	if (mca_funcs && mca_funcs->mca_set_debug_mode)
+		return mca_funcs->mca_set_debug_mode(adev, enable);
+
+	return -EOPNOTSUPP;
+}
+
+static void amdgpu_mca_smu_mca_bank_dump(struct amdgpu_device *adev, int idx, struct mca_bank_entry *entry,
+					 struct ras_query_context *qctx)
+{
+	u64 event_id = qctx ? qctx->evid.event_id : RAS_EVENT_INVALID_ID;
+
+	RAS_EVENT_LOG(adev, event_id, HW_ERR "Accelerator Check Architecture events logged\n");
+	RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].STATUS=0x%016llx\n",
+		      idx, entry->regs[MCA_REG_IDX_STATUS]);
+	RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].ADDR=0x%016llx\n",
+		      idx, entry->regs[MCA_REG_IDX_ADDR]);
+	RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].MISC0=0x%016llx\n",
+		      idx, entry->regs[MCA_REG_IDX_MISC0]);
+	RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].IPID=0x%016llx\n",
+		      idx, entry->regs[MCA_REG_IDX_IPID]);
+	RAS_EVENT_LOG(adev, event_id, HW_ERR "aca entry[%02d].SYND=0x%016llx\n",
+		      idx, entry->regs[MCA_REG_IDX_SYND]);
+}
+
+static int amdgpu_mca_smu_get_valid_mca_count(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, uint32_t *count)
+{
+	const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
+
+	if (!count)
+		return -EINVAL;
+
+	if (mca_funcs && mca_funcs->mca_get_valid_mca_count)
+		return mca_funcs->mca_get_valid_mca_count(adev, type, count);
+
+	return -EOPNOTSUPP;
+}
+
+static int amdgpu_mca_smu_get_mca_entry(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
+					int idx, struct mca_bank_entry *entry)
+{
+	const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
+	int count;
+
+	if (!mca_funcs || !mca_funcs->mca_get_mca_entry)
+		return -EOPNOTSUPP;
+
+	switch (type) {
+	case AMDGPU_MCA_ERROR_TYPE_UE:
+		count = mca_funcs->max_ue_count;
+		break;
+	case AMDGPU_MCA_ERROR_TYPE_CE:
+		count = mca_funcs->max_ce_count;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (idx >= count)
+		return -EINVAL;
+
+	return mca_funcs->mca_get_mca_entry(adev, type, idx, entry);
+}
+
+static bool amdgpu_mca_bank_should_update(struct amdgpu_device *adev, enum amdgpu_mca_error_type type)
+{
+	struct amdgpu_mca *mca = &adev->mca;
+	bool ret = true;
+
+	/*
+	 * Because the UE Valid MCA count will only be cleared after reset,
+	 * in order to avoid repeated counting of the error count,
+	 * the aca bank is only updated once during the gpu recovery stage.
+	 */
+	if (type == AMDGPU_MCA_ERROR_TYPE_UE) {
+		if (amdgpu_ras_intr_triggered())
+			ret = atomic_cmpxchg(&mca->ue_update_flag, 0, 1) == 0;
+		else
+			atomic_set(&mca->ue_update_flag, 0);
+	}
+
+	return ret;
+}
+
+static bool amdgpu_mca_bank_should_dump(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
+					struct mca_bank_entry *entry)
+{
+	bool ret;
+
+	switch (type) {
+	case AMDGPU_MCA_ERROR_TYPE_CE:
+		ret = amdgpu_mca_is_deferred_error(adev, entry->regs[MCA_REG_IDX_STATUS]);
+		break;
+	case AMDGPU_MCA_ERROR_TYPE_UE:
+	default:
+		ret = true;
+		break;
+	}
+
+	return ret;
+}
+
+static int amdgpu_mca_smu_get_mca_set(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, struct mca_bank_set *mca_set,
+				      struct ras_query_context *qctx)
+{
+	struct mca_bank_entry entry;
+	uint32_t count = 0, i;
+	int ret;
+
+	if (!mca_set)
+		return -EINVAL;
+
+	if (!amdgpu_mca_bank_should_update(adev, type))
+		return 0;
+
+	ret = amdgpu_mca_smu_get_valid_mca_count(adev, type, &count);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < count; i++) {
+		memset(&entry, 0, sizeof(entry));
+		ret = amdgpu_mca_smu_get_mca_entry(adev, type, i, &entry);
+		if (ret)
+			return ret;
+
+		amdgpu_mca_bank_set_add_entry(mca_set, &entry);
+
+		if (amdgpu_mca_bank_should_dump(adev, type, &entry))
+			amdgpu_mca_smu_mca_bank_dump(adev, i, &entry, qctx);
+	}
+
+	return 0;
+}
+
+static int amdgpu_mca_smu_parse_mca_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
+						enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count)
+{
+	const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
+
+	if (!count || !entry)
+		return -EINVAL;
+
+	if (!mca_funcs || !mca_funcs->mca_parse_mca_error_count)
+		return -EOPNOTSUPP;
+
+	return mca_funcs->mca_parse_mca_error_count(adev, blk, type, entry, count);
+}
+
+static int amdgpu_mca_dispatch_mca_set(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type,
+				       struct mca_bank_set *mca_set, struct ras_err_data *err_data)
+{
+	struct amdgpu_smuio_mcm_config_info mcm_info;
+	struct mca_bank_node *node, *tmp;
+	struct mca_bank_entry *entry;
+	uint32_t count;
+	int ret;
+
+	if (!mca_set)
+		return -EINVAL;
+
+	if (!mca_set->nr_entries)
+		return 0;
+
+	list_for_each_entry_safe(node, tmp, &mca_set->list, node) {
+		entry = &node->entry;
+
+		count = 0;
+		ret = amdgpu_mca_smu_parse_mca_error_count(adev, blk, type, entry, &count);
+		if (ret && ret != -EOPNOTSUPP)
+			return ret;
+
+		if (!count)
+			continue;
+
+		memset(&mcm_info, 0, sizeof(mcm_info));
+
+		mcm_info.socket_id = entry->info.socket_id;
+		mcm_info.die_id = entry->info.aid;
+
+		if (type == AMDGPU_MCA_ERROR_TYPE_UE) {
+			amdgpu_ras_error_statistic_ue_count(err_data,
+							    &mcm_info, (uint64_t)count);
+		} else {
+			if (amdgpu_mca_is_deferred_error(adev, entry->regs[MCA_REG_IDX_STATUS]))
+				amdgpu_ras_error_statistic_de_count(err_data,
+								    &mcm_info, (uint64_t)count);
+			else
+				amdgpu_ras_error_statistic_ce_count(err_data,
+								    &mcm_info, (uint64_t)count);
+		}
+
+		amdgpu_mca_bank_set_remove_node(mca_set, node);
+	}
+
+	return 0;
+}
+
+static int amdgpu_mca_add_mca_set_to_cache(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, struct mca_bank_set *new)
+{
+	struct mca_bank_cache *mca_cache = &adev->mca.mca_caches[type];
+	int ret;
+
+	mutex_lock(&mca_cache->lock);
+	ret = amdgpu_mca_bank_set_merge(&mca_cache->mca_set, new);
+	mutex_unlock(&mca_cache->lock);
+
+	return ret;
+}
+
+int amdgpu_mca_smu_log_ras_error(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type,
+				 struct ras_err_data *err_data, struct ras_query_context *qctx)
+{
+	struct mca_bank_set mca_set;
+	struct mca_bank_cache *mca_cache = &adev->mca.mca_caches[type];
+	int ret;
+
+	amdgpu_mca_bank_set_init(&mca_set);
+
+	ret = amdgpu_mca_smu_get_mca_set(adev, type, &mca_set, qctx);
+	if (ret)
+		goto out_mca_release;
+
+	ret = amdgpu_mca_dispatch_mca_set(adev, blk, type, &mca_set, err_data);
+	if (ret)
+		goto out_mca_release;
+
+	/* add remain mca bank to mca cache */
+	if (mca_set.nr_entries) {
+		ret = amdgpu_mca_add_mca_set_to_cache(adev, type, &mca_set);
+		if (ret)
+			goto out_mca_release;
+	}
+
+	/* dispatch mca set again if mca cache has valid data */
+	mutex_lock(&mca_cache->lock);
+	if (mca_cache->mca_set.nr_entries)
+		ret = amdgpu_mca_dispatch_mca_set(adev, blk, type, &mca_cache->mca_set, err_data);
+	mutex_unlock(&mca_cache->lock);
+
+out_mca_release:
+	amdgpu_mca_bank_set_release(&mca_set);
+
+	return ret;
+}
+
+#if defined(CONFIG_DEBUG_FS)
+static int amdgpu_mca_smu_debug_mode_set(void *data, u64 val)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)data;
+	int ret;
+
+	ret = amdgpu_ras_set_mca_debug_mode(adev, val ? true : false);
+	if (ret)
+		return ret;
+
+	dev_info(adev->dev, "amdgpu set smu mca debug mode %s success\n", val ? "on" : "off");
+
+	return 0;
+}
+
+static void mca_dump_entry(struct seq_file *m, struct mca_bank_entry *entry)
+{
+	int i, idx = entry->idx;
+	int reg_idx_array[] = {
+		MCA_REG_IDX_STATUS,
+		MCA_REG_IDX_ADDR,
+		MCA_REG_IDX_MISC0,
+		MCA_REG_IDX_IPID,
+		MCA_REG_IDX_SYND,
+	};
+
+	seq_printf(m, "mca entry[%d].type: %s\n", idx, entry->type == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE");
+	seq_printf(m, "mca entry[%d].ip: %d\n", idx, entry->ip);
+	seq_printf(m, "mca entry[%d].info: socketid:%d aid:%d hwid:0x%03x mcatype:0x%04x\n",
+		   idx, entry->info.socket_id, entry->info.aid, entry->info.hwid, entry->info.mcatype);
+
+	for (i = 0; i < ARRAY_SIZE(reg_idx_array); i++)
+		seq_printf(m, "mca entry[%d].regs[%d]: 0x%016llx\n", idx, reg_idx_array[i], entry->regs[reg_idx_array[i]]);
+}
+
+static int mca_dump_show(struct seq_file *m, enum amdgpu_mca_error_type type)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+	struct mca_bank_node *node;
+	struct mca_bank_set mca_set;
+	struct ras_query_context qctx;
+	int ret;
+
+	amdgpu_mca_bank_set_init(&mca_set);
+
+	qctx.evid.event_id = RAS_EVENT_INVALID_ID;
+	ret = amdgpu_mca_smu_get_mca_set(adev, type, &mca_set, &qctx);
+	if (ret)
+		goto err_free_mca_set;
+
+	seq_printf(m, "amdgpu smu %s valid mca count: %d\n",
+		   type == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE", mca_set.nr_entries);
+
+	if (!mca_set.nr_entries)
+		goto err_free_mca_set;
+
+	list_for_each_entry(node, &mca_set.list, node)
+		mca_dump_entry(m, &node->entry);
+
+	/* add mca bank to mca bank cache */
+	ret = amdgpu_mca_add_mca_set_to_cache(adev, type, &mca_set);
+
+err_free_mca_set:
+	amdgpu_mca_bank_set_release(&mca_set);
+
+	return ret;
+}
+
+static int mca_dump_ce_show(struct seq_file *m, void *unused)
+{
+	return mca_dump_show(m, AMDGPU_MCA_ERROR_TYPE_CE);
+}
+
+static int mca_dump_ce_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, mca_dump_ce_show, inode->i_private);
+}
+
+static const struct file_operations mca_ce_dump_debug_fops = {
+	.owner = THIS_MODULE,
+	.open = mca_dump_ce_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = single_release,
+};
+
+static int mca_dump_ue_show(struct seq_file *m, void *unused)
+{
+	return mca_dump_show(m, AMDGPU_MCA_ERROR_TYPE_UE);
+}
+
+static int mca_dump_ue_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, mca_dump_ue_show, inode->i_private);
+}
+
+static const struct file_operations mca_ue_dump_debug_fops = {
+	.owner = THIS_MODULE,
+	.open = mca_dump_ue_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = single_release,
+};
+
+DEFINE_DEBUGFS_ATTRIBUTE(mca_debug_mode_fops, NULL, amdgpu_mca_smu_debug_mode_set, "%llu\n");
+#endif
+
+void amdgpu_mca_smu_debugfs_init(struct amdgpu_device *adev, struct dentry *root)
+{
+#if defined(CONFIG_DEBUG_FS)
+	if (!root)
+		return;
+
+	debugfs_create_file("mca_debug_mode", 0200, root, adev, &mca_debug_mode_fops);
+	debugfs_create_file("mca_ue_dump", 0400, root, adev, &mca_ue_dump_debug_fops);
+	debugfs_create_file("mca_ce_dump", 0400, root, adev, &mca_ce_dump_debug_fops);
+#endif
+}
+
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mca.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mca.h
new file mode 100644
index 000000000000..e80323ff90c1
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mca.h
@@ -0,0 +1,169 @@
+/*
+ * Copyright (C) 2021  Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#ifndef __AMDGPU_MCA_H__
+#define __AMDGPU_MCA_H__
+
+#include "amdgpu_ras.h"
+
+#define MCA_MAX_REGS_COUNT	(16)
+
+#define MCA_REG_FIELD(x, h, l)			(((x) & GENMASK_ULL(h, l)) >> l)
+#define MCA_REG__STATUS__VAL(x)			MCA_REG_FIELD(x, 63, 63)
+#define MCA_REG__STATUS__OVERFLOW(x)		MCA_REG_FIELD(x, 62, 62)
+#define MCA_REG__STATUS__UC(x)			MCA_REG_FIELD(x, 61, 61)
+#define MCA_REG__STATUS__EN(x)			MCA_REG_FIELD(x, 60, 60)
+#define MCA_REG__STATUS__MISCV(x)		MCA_REG_FIELD(x, 59, 59)
+#define MCA_REG__STATUS__ADDRV(x)		MCA_REG_FIELD(x, 58, 58)
+#define MCA_REG__STATUS__PCC(x)			MCA_REG_FIELD(x, 57, 57)
+#define MCA_REG__STATUS__ERRCOREIDVAL(x)	MCA_REG_FIELD(x, 56, 56)
+#define MCA_REG__STATUS__TCC(x)			MCA_REG_FIELD(x, 55, 55)
+#define MCA_REG__STATUS__SYNDV(x)		MCA_REG_FIELD(x, 53, 53)
+#define MCA_REG__STATUS__CECC(x)		MCA_REG_FIELD(x, 46, 46)
+#define MCA_REG__STATUS__UECC(x)		MCA_REG_FIELD(x, 45, 45)
+#define MCA_REG__STATUS__DEFERRED(x)		MCA_REG_FIELD(x, 44, 44)
+#define MCA_REG__STATUS__POISON(x)		MCA_REG_FIELD(x, 43, 43)
+#define MCA_REG__STATUS__SCRUB(x)		MCA_REG_FIELD(x, 40, 40)
+#define MCA_REG__STATUS__ERRCOREID(x)		MCA_REG_FIELD(x, 37, 32)
+#define MCA_REG__STATUS__ADDRLSB(x)		MCA_REG_FIELD(x, 29, 24)
+#define MCA_REG__STATUS__ERRORCODEEXT(x)	MCA_REG_FIELD(x, 21, 16)
+#define MCA_REG__STATUS__ERRORCODE(x)		MCA_REG_FIELD(x, 15, 0)
+
+#define MCA_REG__MISC0__ERRCNT(x)		MCA_REG_FIELD(x, 43, 32)
+
+#define MCA_REG__SYND__ERRORINFORMATION(x)	MCA_REG_FIELD(x, 17, 0)
+
+enum amdgpu_mca_ip {
+	AMDGPU_MCA_IP_UNKNOW = -1,
+	AMDGPU_MCA_IP_PSP = 0,
+	AMDGPU_MCA_IP_SDMA,
+	AMDGPU_MCA_IP_GC,
+	AMDGPU_MCA_IP_SMU,
+	AMDGPU_MCA_IP_MP5,
+	AMDGPU_MCA_IP_UMC,
+	AMDGPU_MCA_IP_PCS_XGMI,
+	AMDGPU_MCA_IP_COUNT,
+};
+
+enum amdgpu_mca_error_type {
+	AMDGPU_MCA_ERROR_TYPE_UE = 0,
+	AMDGPU_MCA_ERROR_TYPE_CE,
+	AMDGPU_MCA_ERROR_TYPE_DE,
+};
+
+struct amdgpu_mca_ras_block {
+	struct amdgpu_ras_block_object ras_block;
+};
+
+struct amdgpu_mca_ras {
+	struct ras_common_if *ras_if;
+	struct amdgpu_mca_ras_block *ras;
+};
+
+struct mca_bank_set {
+	int nr_entries;
+	struct list_head list;
+};
+
+struct mca_bank_cache {
+	struct mca_bank_set mca_set;
+	struct mutex lock;
+};
+
+struct amdgpu_mca {
+	struct amdgpu_mca_ras mp0;
+	struct amdgpu_mca_ras mp1;
+	struct amdgpu_mca_ras mpio;
+	const struct amdgpu_mca_smu_funcs *mca_funcs;
+	struct mca_bank_cache mca_caches[AMDGPU_MCA_ERROR_TYPE_DE];
+	atomic_t ue_update_flag;
+};
+
+enum mca_reg_idx {
+	MCA_REG_IDX_STATUS		= 1,
+	MCA_REG_IDX_ADDR		= 2,
+	MCA_REG_IDX_MISC0		= 3,
+	MCA_REG_IDX_IPID		= 5,
+	MCA_REG_IDX_SYND		= 6,
+	MCA_REG_IDX_COUNT		= 16,
+};
+
+struct mca_bank_info {
+	int socket_id;
+	int aid;
+	int hwid;
+	int mcatype;
+};
+
+struct mca_bank_entry {
+	int idx;
+	enum amdgpu_mca_error_type type;
+	enum amdgpu_mca_ip ip;
+	struct mca_bank_info info;
+	uint64_t regs[MCA_MAX_REGS_COUNT];
+};
+
+struct mca_bank_node {
+	struct mca_bank_entry entry;
+	struct list_head node;
+};
+
+struct amdgpu_mca_smu_funcs {
+	int max_ue_count;
+	int max_ce_count;
+	int (*mca_set_debug_mode)(struct amdgpu_device *adev, bool enable);
+	int (*mca_parse_mca_error_count)(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type,
+					 struct mca_bank_entry *entry, uint32_t *count);
+	int (*mca_get_valid_mca_count)(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
+				       uint32_t *count);
+	int (*mca_get_mca_entry)(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
+				 int idx, struct mca_bank_entry *entry);
+};
+
+void amdgpu_mca_query_correctable_error_count(struct amdgpu_device *adev,
+					      uint64_t mc_status_addr,
+					      unsigned long *error_count);
+
+void amdgpu_mca_query_uncorrectable_error_count(struct amdgpu_device *adev,
+						uint64_t mc_status_addr,
+						unsigned long *error_count);
+
+void amdgpu_mca_reset_error_count(struct amdgpu_device *adev,
+				  uint64_t mc_status_addr);
+
+void amdgpu_mca_query_ras_error_count(struct amdgpu_device *adev,
+				      uint64_t mc_status_addr,
+				      void *ras_error_status);
+int amdgpu_mca_mp0_ras_sw_init(struct amdgpu_device *adev);
+int amdgpu_mca_mp1_ras_sw_init(struct amdgpu_device *adev);
+int amdgpu_mca_mpio_ras_sw_init(struct amdgpu_device *adev);
+
+void amdgpu_mca_smu_init_funcs(struct amdgpu_device *adev, const struct amdgpu_mca_smu_funcs *mca_funcs);
+int amdgpu_mca_init(struct amdgpu_device *adev);
+void amdgpu_mca_fini(struct amdgpu_device *adev);
+int amdgpu_mca_reset(struct amdgpu_device *adev);
+int amdgpu_mca_smu_set_debug_mode(struct amdgpu_device *adev, bool enable);
+int amdgpu_mca_smu_get_mca_set_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
+					   enum amdgpu_mca_error_type type, uint32_t *total);
+void amdgpu_mca_smu_debugfs_init(struct amdgpu_device *adev, struct dentry *root);
+int amdgpu_mca_smu_log_ras_error(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type,
+				 struct ras_err_data *err_data, struct ras_query_context *qctx);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
new file mode 100644
index 000000000000..9c182ce501af
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
@@ -0,0 +1,784 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/firmware.h>
+#include <drm/drm_exec.h>
+
+#include "amdgpu_mes.h"
+#include "amdgpu.h"
+#include "soc15_common.h"
+#include "amdgpu_mes_ctx.h"
+
+#define AMDGPU_MES_MAX_NUM_OF_QUEUES_PER_PROCESS 1024
+#define AMDGPU_ONE_DOORBELL_SIZE 8
+
+int amdgpu_mes_doorbell_process_slice(struct amdgpu_device *adev)
+{
+	return roundup(AMDGPU_ONE_DOORBELL_SIZE *
+		       AMDGPU_MES_MAX_NUM_OF_QUEUES_PER_PROCESS,
+		       PAGE_SIZE);
+}
+
+static int amdgpu_mes_doorbell_init(struct amdgpu_device *adev)
+{
+	int i;
+	struct amdgpu_mes *mes = &adev->mes;
+
+	/* Bitmap for dynamic allocation of kernel doorbells */
+	mes->doorbell_bitmap = bitmap_zalloc(PAGE_SIZE / sizeof(u32), GFP_KERNEL);
+	if (!mes->doorbell_bitmap) {
+		dev_err(adev->dev, "Failed to allocate MES doorbell bitmap\n");
+		return -ENOMEM;
+	}
+
+	mes->num_mes_dbs = PAGE_SIZE / AMDGPU_ONE_DOORBELL_SIZE;
+	for (i = 0; i < AMDGPU_MES_PRIORITY_NUM_LEVELS; i++) {
+		adev->mes.aggregated_doorbells[i] = mes->db_start_dw_offset + i * 2;
+		set_bit(i, mes->doorbell_bitmap);
+	}
+
+	return 0;
+}
+
+static int amdgpu_mes_event_log_init(struct amdgpu_device *adev)
+{
+	int r;
+
+	if (!amdgpu_mes_log_enable)
+		return 0;
+
+	r = amdgpu_bo_create_kernel(adev, adev->mes.event_log_size, PAGE_SIZE,
+				    AMDGPU_GEM_DOMAIN_VRAM,
+				    &adev->mes.event_log_gpu_obj,
+				    &adev->mes.event_log_gpu_addr,
+				    &adev->mes.event_log_cpu_addr);
+	if (r) {
+		dev_warn(adev->dev, "failed to create MES event log buffer (%d)", r);
+		return r;
+	}
+
+	memset(adev->mes.event_log_cpu_addr, 0, adev->mes.event_log_size);
+
+	return  0;
+
+}
+
+static void amdgpu_mes_doorbell_free(struct amdgpu_device *adev)
+{
+	bitmap_free(adev->mes.doorbell_bitmap);
+}
+
+int amdgpu_mes_init(struct amdgpu_device *adev)
+{
+	int i, r, num_pipes;
+
+	adev->mes.adev = adev;
+
+	idr_init(&adev->mes.pasid_idr);
+	idr_init(&adev->mes.gang_id_idr);
+	idr_init(&adev->mes.queue_id_idr);
+	ida_init(&adev->mes.doorbell_ida);
+	spin_lock_init(&adev->mes.queue_id_lock);
+	mutex_init(&adev->mes.mutex_hidden);
+
+	for (i = 0; i < AMDGPU_MAX_MES_PIPES; i++)
+		spin_lock_init(&adev->mes.ring_lock[i]);
+
+	adev->mes.total_max_queue = AMDGPU_FENCE_MES_QUEUE_ID_MASK;
+	adev->mes.vmid_mask_mmhub = 0xFF00;
+	adev->mes.vmid_mask_gfxhub = adev->gfx.disable_kq ? 0xFFFE : 0xFF00;
+
+	num_pipes = adev->gfx.me.num_pipe_per_me * adev->gfx.me.num_me;
+	if (num_pipes > AMDGPU_MES_MAX_GFX_PIPES)
+		dev_warn(adev->dev, "more gfx pipes than supported by MES! (%d vs %d)\n",
+			 num_pipes, AMDGPU_MES_MAX_GFX_PIPES);
+
+	for (i = 0; i < AMDGPU_MES_MAX_GFX_PIPES; i++) {
+		if (i >= num_pipes)
+			break;
+		if (amdgpu_ip_version(adev, GC_HWIP, 0) >=
+		    IP_VERSION(12, 0, 0))
+			/*
+			 * GFX V12 has only one GFX pipe, but 8 queues in it.
+			 * GFX pipe 0 queue 0 is being used by Kernel queue.
+			 * Set GFX pipe 0 queue 1-7 for MES scheduling
+			 * mask = 1111 1110b
+			 */
+			adev->mes.gfx_hqd_mask[i] = adev->gfx.disable_kq ? 0xFF : 0xFE;
+		else
+			/*
+			 * GFX pipe 0 queue 0 is being used by Kernel queue.
+			 * Set GFX pipe 0 queue 1 for MES scheduling
+			 * mask = 10b
+			 */
+			adev->mes.gfx_hqd_mask[i] = adev->gfx.disable_kq ? 0x3 : 0x2;
+	}
+
+	num_pipes = adev->gfx.mec.num_pipe_per_mec * adev->gfx.mec.num_mec;
+	if (num_pipes > AMDGPU_MES_MAX_COMPUTE_PIPES)
+		dev_warn(adev->dev, "more compute pipes than supported by MES! (%d vs %d)\n",
+			 num_pipes, AMDGPU_MES_MAX_COMPUTE_PIPES);
+
+	for (i = 0; i < AMDGPU_MES_MAX_COMPUTE_PIPES; i++) {
+		if (i >= num_pipes)
+			break;
+		adev->mes.compute_hqd_mask[i] = adev->gfx.disable_kq ? 0xF : 0xC;
+	}
+
+	num_pipes = adev->sdma.num_instances;
+	if (num_pipes > AMDGPU_MES_MAX_SDMA_PIPES)
+		dev_warn(adev->dev, "more SDMA pipes than supported by MES! (%d vs %d)\n",
+			 num_pipes, AMDGPU_MES_MAX_SDMA_PIPES);
+
+	for (i = 0; i < AMDGPU_MES_MAX_SDMA_PIPES; i++) {
+		if (i >= num_pipes)
+			break;
+		adev->mes.sdma_hqd_mask[i] = 0xfc;
+	}
+
+	for (i = 0; i < AMDGPU_MAX_MES_PIPES; i++) {
+		r = amdgpu_device_wb_get(adev, &adev->mes.sch_ctx_offs[i]);
+		if (r) {
+			dev_err(adev->dev,
+				"(%d) ring trail_fence_offs wb alloc failed\n",
+				r);
+			goto error;
+		}
+		adev->mes.sch_ctx_gpu_addr[i] =
+			adev->wb.gpu_addr + (adev->mes.sch_ctx_offs[i] * 4);
+		adev->mes.sch_ctx_ptr[i] =
+			(uint64_t *)&adev->wb.wb[adev->mes.sch_ctx_offs[i]];
+
+		r = amdgpu_device_wb_get(adev,
+				 &adev->mes.query_status_fence_offs[i]);
+		if (r) {
+			dev_err(adev->dev,
+			      "(%d) query_status_fence_offs wb alloc failed\n",
+			      r);
+			goto error;
+		}
+		adev->mes.query_status_fence_gpu_addr[i] = adev->wb.gpu_addr +
+			(adev->mes.query_status_fence_offs[i] * 4);
+		adev->mes.query_status_fence_ptr[i] =
+			(uint64_t *)&adev->wb.wb[adev->mes.query_status_fence_offs[i]];
+	}
+
+	r = amdgpu_mes_doorbell_init(adev);
+	if (r)
+		goto error;
+
+	r = amdgpu_mes_event_log_init(adev);
+	if (r)
+		goto error_doorbell;
+
+	if (adev->mes.hung_queue_db_array_size) {
+		r = amdgpu_bo_create_kernel(adev,
+					    adev->mes.hung_queue_db_array_size * sizeof(u32),
+					    PAGE_SIZE,
+					    AMDGPU_GEM_DOMAIN_GTT,
+					    &adev->mes.hung_queue_db_array_gpu_obj,
+					    &adev->mes.hung_queue_db_array_gpu_addr,
+					    &adev->mes.hung_queue_db_array_cpu_addr);
+		if (r) {
+			dev_warn(adev->dev, "failed to create MES hung db array buffer (%d)", r);
+			goto error_doorbell;
+		}
+	}
+
+	return 0;
+
+error_doorbell:
+	amdgpu_mes_doorbell_free(adev);
+error:
+	for (i = 0; i < AMDGPU_MAX_MES_PIPES; i++) {
+		if (adev->mes.sch_ctx_ptr[i])
+			amdgpu_device_wb_free(adev, adev->mes.sch_ctx_offs[i]);
+		if (adev->mes.query_status_fence_ptr[i])
+			amdgpu_device_wb_free(adev,
+				      adev->mes.query_status_fence_offs[i]);
+	}
+
+	idr_destroy(&adev->mes.pasid_idr);
+	idr_destroy(&adev->mes.gang_id_idr);
+	idr_destroy(&adev->mes.queue_id_idr);
+	ida_destroy(&adev->mes.doorbell_ida);
+	mutex_destroy(&adev->mes.mutex_hidden);
+	return r;
+}
+
+void amdgpu_mes_fini(struct amdgpu_device *adev)
+{
+	int i;
+
+	amdgpu_bo_free_kernel(&adev->mes.hung_queue_db_array_gpu_obj,
+			      &adev->mes.hung_queue_db_array_gpu_addr,
+			      &adev->mes.hung_queue_db_array_cpu_addr);
+
+	amdgpu_bo_free_kernel(&adev->mes.event_log_gpu_obj,
+			      &adev->mes.event_log_gpu_addr,
+			      &adev->mes.event_log_cpu_addr);
+
+	for (i = 0; i < AMDGPU_MAX_MES_PIPES; i++) {
+		if (adev->mes.sch_ctx_ptr[i])
+			amdgpu_device_wb_free(adev, adev->mes.sch_ctx_offs[i]);
+		if (adev->mes.query_status_fence_ptr[i])
+			amdgpu_device_wb_free(adev,
+				      adev->mes.query_status_fence_offs[i]);
+	}
+
+	amdgpu_mes_doorbell_free(adev);
+
+	idr_destroy(&adev->mes.pasid_idr);
+	idr_destroy(&adev->mes.gang_id_idr);
+	idr_destroy(&adev->mes.queue_id_idr);
+	ida_destroy(&adev->mes.doorbell_ida);
+	mutex_destroy(&adev->mes.mutex_hidden);
+}
+
+int amdgpu_mes_suspend(struct amdgpu_device *adev)
+{
+	struct mes_suspend_gang_input input;
+	int r;
+
+	if (!amdgpu_mes_suspend_resume_all_supported(adev))
+		return 0;
+
+	memset(&input, 0x0, sizeof(struct mes_suspend_gang_input));
+	input.suspend_all_gangs = 1;
+
+	/*
+	 * Avoid taking any other locks under MES lock to avoid circular
+	 * lock dependencies.
+	 */
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->suspend_gang(&adev->mes, &input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to suspend all gangs");
+
+	return r;
+}
+
+int amdgpu_mes_resume(struct amdgpu_device *adev)
+{
+	struct mes_resume_gang_input input;
+	int r;
+
+	if (!amdgpu_mes_suspend_resume_all_supported(adev))
+		return 0;
+
+	memset(&input, 0x0, sizeof(struct mes_resume_gang_input));
+	input.resume_all_gangs = 1;
+
+	/*
+	 * Avoid taking any other locks under MES lock to avoid circular
+	 * lock dependencies.
+	 */
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->resume_gang(&adev->mes, &input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to resume all gangs");
+
+	return r;
+}
+
+int amdgpu_mes_map_legacy_queue(struct amdgpu_device *adev,
+				struct amdgpu_ring *ring)
+{
+	struct mes_map_legacy_queue_input queue_input;
+	int r;
+
+	memset(&queue_input, 0, sizeof(queue_input));
+
+	queue_input.queue_type = ring->funcs->type;
+	queue_input.doorbell_offset = ring->doorbell_index;
+	queue_input.pipe_id = ring->pipe;
+	queue_input.queue_id = ring->queue;
+	queue_input.mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
+	queue_input.wptr_addr = ring->wptr_gpu_addr;
+
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->map_legacy_queue(&adev->mes, &queue_input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to map legacy queue\n");
+
+	return r;
+}
+
+int amdgpu_mes_unmap_legacy_queue(struct amdgpu_device *adev,
+				  struct amdgpu_ring *ring,
+				  enum amdgpu_unmap_queues_action action,
+				  u64 gpu_addr, u64 seq)
+{
+	struct mes_unmap_legacy_queue_input queue_input;
+	int r;
+
+	queue_input.action = action;
+	queue_input.queue_type = ring->funcs->type;
+	queue_input.doorbell_offset = ring->doorbell_index;
+	queue_input.pipe_id = ring->pipe;
+	queue_input.queue_id = ring->queue;
+	queue_input.trail_fence_addr = gpu_addr;
+	queue_input.trail_fence_data = seq;
+
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->unmap_legacy_queue(&adev->mes, &queue_input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to unmap legacy queue\n");
+
+	return r;
+}
+
+int amdgpu_mes_reset_legacy_queue(struct amdgpu_device *adev,
+				  struct amdgpu_ring *ring,
+				  unsigned int vmid,
+				  bool use_mmio)
+{
+	struct mes_reset_queue_input queue_input;
+	int r;
+
+	memset(&queue_input, 0, sizeof(queue_input));
+
+	queue_input.queue_type = ring->funcs->type;
+	queue_input.doorbell_offset = ring->doorbell_index;
+	queue_input.me_id = ring->me;
+	queue_input.pipe_id = ring->pipe;
+	queue_input.queue_id = ring->queue;
+	queue_input.mqd_addr = ring->mqd_obj ? amdgpu_bo_gpu_offset(ring->mqd_obj) : 0;
+	queue_input.wptr_addr = ring->wptr_gpu_addr;
+	queue_input.vmid = vmid;
+	queue_input.use_mmio = use_mmio;
+	queue_input.is_kq = true;
+	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX)
+		queue_input.legacy_gfx = true;
+
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->reset_hw_queue(&adev->mes, &queue_input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to reset legacy queue\n");
+
+	return r;
+}
+
+int amdgpu_mes_get_hung_queue_db_array_size(struct amdgpu_device *adev)
+{
+	return adev->mes.hung_queue_db_array_size;
+}
+
+int amdgpu_mes_detect_and_reset_hung_queues(struct amdgpu_device *adev,
+					    int queue_type,
+					    bool detect_only,
+					    unsigned int *hung_db_num,
+					    u32 *hung_db_array)
+
+{
+	struct mes_detect_and_reset_queue_input input;
+	u32 *db_array = adev->mes.hung_queue_db_array_cpu_addr;
+	int r, i;
+
+	if (!hung_db_num || !hung_db_array)
+		return -EINVAL;
+
+	if ((queue_type != AMDGPU_RING_TYPE_GFX) &&
+	    (queue_type != AMDGPU_RING_TYPE_COMPUTE) &&
+	    (queue_type != AMDGPU_RING_TYPE_SDMA))
+		return -EINVAL;
+
+	/* Clear the doorbell array before detection */
+	memset(adev->mes.hung_queue_db_array_cpu_addr, AMDGPU_MES_INVALID_DB_OFFSET,
+		adev->mes.hung_queue_db_array_size * sizeof(u32));
+	input.queue_type = queue_type;
+	input.detect_only = detect_only;
+
+	r = adev->mes.funcs->detect_and_reset_hung_queues(&adev->mes,
+							  &input);
+	if (r) {
+		dev_err(adev->dev, "failed to detect and reset\n");
+	} else {
+		*hung_db_num = 0;
+		for (i = 0; i < adev->mes.hung_queue_hqd_info_offset; i++) {
+			if (db_array[i] != AMDGPU_MES_INVALID_DB_OFFSET) {
+				hung_db_array[i] = db_array[i];
+				*hung_db_num += 1;
+			}
+		}
+
+		/*
+		 * TODO: return HQD info for MES scheduled user compute queue reset cases
+		 * stored in hung_db_array hqd info offset to full array size
+		 */
+	}
+
+	return r;
+}
+
+uint32_t amdgpu_mes_rreg(struct amdgpu_device *adev, uint32_t reg)
+{
+	struct mes_misc_op_input op_input;
+	int r, val = 0;
+	uint32_t addr_offset = 0;
+	uint64_t read_val_gpu_addr;
+	uint32_t *read_val_ptr;
+
+	if (amdgpu_device_wb_get(adev, &addr_offset)) {
+		dev_err(adev->dev, "critical bug! too many mes readers\n");
+		goto error;
+	}
+	read_val_gpu_addr = adev->wb.gpu_addr + (addr_offset * 4);
+	read_val_ptr = (uint32_t *)&adev->wb.wb[addr_offset];
+	op_input.op = MES_MISC_OP_READ_REG;
+	op_input.read_reg.reg_offset = reg;
+	op_input.read_reg.buffer_addr = read_val_gpu_addr;
+
+	if (!adev->mes.funcs->misc_op) {
+		dev_err(adev->dev, "mes rreg is not supported!\n");
+		goto error;
+	}
+
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->misc_op(&adev->mes, &op_input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to read reg (0x%x)\n", reg);
+	else
+		val = *(read_val_ptr);
+
+error:
+	if (addr_offset)
+		amdgpu_device_wb_free(adev, addr_offset);
+	return val;
+}
+
+int amdgpu_mes_wreg(struct amdgpu_device *adev,
+		    uint32_t reg, uint32_t val)
+{
+	struct mes_misc_op_input op_input;
+	int r;
+
+	op_input.op = MES_MISC_OP_WRITE_REG;
+	op_input.write_reg.reg_offset = reg;
+	op_input.write_reg.reg_value = val;
+
+	if (!adev->mes.funcs->misc_op) {
+		dev_err(adev->dev, "mes wreg is not supported!\n");
+		r = -EINVAL;
+		goto error;
+	}
+
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->misc_op(&adev->mes, &op_input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to write reg (0x%x)\n", reg);
+
+error:
+	return r;
+}
+
+int amdgpu_mes_reg_write_reg_wait(struct amdgpu_device *adev,
+				  uint32_t reg0, uint32_t reg1,
+				  uint32_t ref, uint32_t mask)
+{
+	struct mes_misc_op_input op_input;
+	int r;
+
+	op_input.op = MES_MISC_OP_WRM_REG_WR_WAIT;
+	op_input.wrm_reg.reg0 = reg0;
+	op_input.wrm_reg.reg1 = reg1;
+	op_input.wrm_reg.ref = ref;
+	op_input.wrm_reg.mask = mask;
+
+	if (!adev->mes.funcs->misc_op) {
+		dev_err(adev->dev, "mes reg_write_reg_wait is not supported!\n");
+		r = -EINVAL;
+		goto error;
+	}
+
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->misc_op(&adev->mes, &op_input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to reg_write_reg_wait\n");
+
+error:
+	return r;
+}
+
+int amdgpu_mes_hdp_flush(struct amdgpu_device *adev)
+{
+	uint32_t hdp_flush_req_offset, hdp_flush_done_offset, ref_and_mask;
+
+	hdp_flush_req_offset = adev->nbio.funcs->get_hdp_flush_req_offset(adev);
+	hdp_flush_done_offset = adev->nbio.funcs->get_hdp_flush_done_offset(adev);
+	ref_and_mask = adev->nbio.hdp_flush_reg->ref_and_mask_cp0;
+
+	return amdgpu_mes_reg_write_reg_wait(adev, hdp_flush_req_offset, hdp_flush_done_offset,
+					     ref_and_mask, ref_and_mask);
+}
+
+int amdgpu_mes_set_shader_debugger(struct amdgpu_device *adev,
+				uint64_t process_context_addr,
+				uint32_t spi_gdbg_per_vmid_cntl,
+				const uint32_t *tcp_watch_cntl,
+				uint32_t flags,
+				bool trap_en)
+{
+	struct mes_misc_op_input op_input = {0};
+	int r;
+
+	if (!adev->mes.funcs->misc_op) {
+		dev_err(adev->dev,
+			"mes set shader debugger is not supported!\n");
+		return -EINVAL;
+	}
+
+	op_input.op = MES_MISC_OP_SET_SHADER_DEBUGGER;
+	op_input.set_shader_debugger.process_context_addr = process_context_addr;
+	op_input.set_shader_debugger.flags.u32all = flags;
+
+	/* use amdgpu mes_flush_shader_debugger instead */
+	if (op_input.set_shader_debugger.flags.process_ctx_flush)
+		return -EINVAL;
+
+	op_input.set_shader_debugger.spi_gdbg_per_vmid_cntl = spi_gdbg_per_vmid_cntl;
+	memcpy(op_input.set_shader_debugger.tcp_watch_cntl, tcp_watch_cntl,
+			sizeof(op_input.set_shader_debugger.tcp_watch_cntl));
+
+	if (((adev->mes.sched_version & AMDGPU_MES_API_VERSION_MASK) >>
+			AMDGPU_MES_API_VERSION_SHIFT) >= 14)
+		op_input.set_shader_debugger.trap_en = trap_en;
+
+	amdgpu_mes_lock(&adev->mes);
+
+	r = adev->mes.funcs->misc_op(&adev->mes, &op_input);
+	if (r)
+		dev_err(adev->dev, "failed to set_shader_debugger\n");
+
+	amdgpu_mes_unlock(&adev->mes);
+
+	return r;
+}
+
+int amdgpu_mes_flush_shader_debugger(struct amdgpu_device *adev,
+				     uint64_t process_context_addr)
+{
+	struct mes_misc_op_input op_input = {0};
+	int r;
+
+	if (!adev->mes.funcs->misc_op) {
+		dev_err(adev->dev,
+			"mes flush shader debugger is not supported!\n");
+		return -EINVAL;
+	}
+
+	op_input.op = MES_MISC_OP_SET_SHADER_DEBUGGER;
+	op_input.set_shader_debugger.process_context_addr = process_context_addr;
+	op_input.set_shader_debugger.flags.process_ctx_flush = true;
+
+	amdgpu_mes_lock(&adev->mes);
+
+	r = adev->mes.funcs->misc_op(&adev->mes, &op_input);
+	if (r)
+		dev_err(adev->dev, "failed to set_shader_debugger\n");
+
+	amdgpu_mes_unlock(&adev->mes);
+
+	return r;
+}
+
+uint32_t amdgpu_mes_get_aggregated_doorbell_index(struct amdgpu_device *adev,
+						   enum amdgpu_mes_priority_level prio)
+{
+	return adev->mes.aggregated_doorbells[prio];
+}
+
+int amdgpu_mes_init_microcode(struct amdgpu_device *adev, int pipe)
+{
+	const struct mes_firmware_header_v1_0 *mes_hdr;
+	struct amdgpu_firmware_info *info;
+	char ucode_prefix[30];
+	char fw_name[50];
+	bool need_retry = false;
+	u32 *ucode_ptr;
+	int r;
+
+	amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix,
+				       sizeof(ucode_prefix));
+	if (adev->enable_uni_mes) {
+		snprintf(fw_name, sizeof(fw_name),
+			 "amdgpu/%s_uni_mes.bin", ucode_prefix);
+	} else if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0) &&
+	    amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(12, 0, 0)) {
+		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mes%s.bin",
+			 ucode_prefix,
+			 pipe == AMDGPU_MES_SCHED_PIPE ? "_2" : "1");
+		need_retry = true;
+	} else {
+		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mes%s.bin",
+			 ucode_prefix,
+			 pipe == AMDGPU_MES_SCHED_PIPE ? "" : "1");
+	}
+
+	r = amdgpu_ucode_request(adev, &adev->mes.fw[pipe], AMDGPU_UCODE_REQUIRED,
+				 "%s", fw_name);
+	if (r && need_retry && pipe == AMDGPU_MES_SCHED_PIPE) {
+		dev_info(adev->dev, "try to fall back to %s_mes.bin\n", ucode_prefix);
+		r = amdgpu_ucode_request(adev, &adev->mes.fw[pipe],
+					 AMDGPU_UCODE_REQUIRED,
+					 "amdgpu/%s_mes.bin", ucode_prefix);
+	}
+
+	if (r)
+		goto out;
+
+	mes_hdr = (const struct mes_firmware_header_v1_0 *)
+		adev->mes.fw[pipe]->data;
+	adev->mes.uc_start_addr[pipe] =
+		le32_to_cpu(mes_hdr->mes_uc_start_addr_lo) |
+		((uint64_t)(le32_to_cpu(mes_hdr->mes_uc_start_addr_hi)) << 32);
+	adev->mes.data_start_addr[pipe] =
+		le32_to_cpu(mes_hdr->mes_data_start_addr_lo) |
+		((uint64_t)(le32_to_cpu(mes_hdr->mes_data_start_addr_hi)) << 32);
+	ucode_ptr = (u32 *)(adev->mes.fw[pipe]->data +
+			  sizeof(union amdgpu_firmware_header));
+	adev->mes.fw_version[pipe] =
+		le32_to_cpu(ucode_ptr[24]) & AMDGPU_MES_VERSION_MASK;
+
+	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+		int ucode, ucode_data;
+
+		if (pipe == AMDGPU_MES_SCHED_PIPE) {
+			ucode = AMDGPU_UCODE_ID_CP_MES;
+			ucode_data = AMDGPU_UCODE_ID_CP_MES_DATA;
+		} else {
+			ucode = AMDGPU_UCODE_ID_CP_MES1;
+			ucode_data = AMDGPU_UCODE_ID_CP_MES1_DATA;
+		}
+
+		info = &adev->firmware.ucode[ucode];
+		info->ucode_id = ucode;
+		info->fw = adev->mes.fw[pipe];
+		adev->firmware.fw_size +=
+			ALIGN(le32_to_cpu(mes_hdr->mes_ucode_size_bytes),
+			      PAGE_SIZE);
+
+		info = &adev->firmware.ucode[ucode_data];
+		info->ucode_id = ucode_data;
+		info->fw = adev->mes.fw[pipe];
+		adev->firmware.fw_size +=
+			ALIGN(le32_to_cpu(mes_hdr->mes_ucode_data_size_bytes),
+			      PAGE_SIZE);
+	}
+
+	return 0;
+out:
+	amdgpu_ucode_release(&adev->mes.fw[pipe]);
+	return r;
+}
+
+bool amdgpu_mes_suspend_resume_all_supported(struct amdgpu_device *adev)
+{
+	uint32_t mes_rev = adev->mes.sched_version & AMDGPU_MES_VERSION_MASK;
+
+	return ((amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0) &&
+		 amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(12, 0, 0) &&
+		 mes_rev >= 0x63) ||
+		amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(12, 0, 0));
+}
+
+/* Fix me -- node_id is used to identify the correct MES instances in the future */
+static int amdgpu_mes_set_enforce_isolation(struct amdgpu_device *adev,
+					    uint32_t node_id, bool enable)
+{
+	struct mes_misc_op_input op_input = {0};
+	int r;
+
+	op_input.op = MES_MISC_OP_CHANGE_CONFIG;
+	op_input.change_config.option.limit_single_process = enable ? 1 : 0;
+
+	if (!adev->mes.funcs->misc_op) {
+		dev_err(adev->dev, "mes change config is not supported!\n");
+		r = -EINVAL;
+		goto error;
+	}
+
+	amdgpu_mes_lock(&adev->mes);
+	r = adev->mes.funcs->misc_op(&adev->mes, &op_input);
+	amdgpu_mes_unlock(&adev->mes);
+	if (r)
+		dev_err(adev->dev, "failed to change_config.\n");
+
+error:
+	return r;
+}
+
+int amdgpu_mes_update_enforce_isolation(struct amdgpu_device *adev)
+{
+	int i, r = 0;
+
+	if (adev->enable_mes && adev->gfx.enable_cleaner_shader) {
+		mutex_lock(&adev->enforce_isolation_mutex);
+		for (i = 0; i < (adev->xcp_mgr ? adev->xcp_mgr->num_xcps : 1); i++) {
+			if (adev->enforce_isolation[i] == AMDGPU_ENFORCE_ISOLATION_ENABLE)
+				r |= amdgpu_mes_set_enforce_isolation(adev, i, true);
+			else
+				r |= amdgpu_mes_set_enforce_isolation(adev, i, false);
+		}
+		mutex_unlock(&adev->enforce_isolation_mutex);
+	}
+	return r;
+}
+
+#if defined(CONFIG_DEBUG_FS)
+
+static int amdgpu_debugfs_mes_event_log_show(struct seq_file *m, void *unused)
+{
+	struct amdgpu_device *adev = m->private;
+	uint32_t *mem = (uint32_t *)(adev->mes.event_log_cpu_addr);
+
+	seq_hex_dump(m, "", DUMP_PREFIX_OFFSET, 32, 4,
+		     mem, adev->mes.event_log_size, false);
+
+	return 0;
+}
+
+DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_mes_event_log);
+
+#endif
+
+void amdgpu_debugfs_mes_event_log_init(struct amdgpu_device *adev)
+{
+
+#if defined(CONFIG_DEBUG_FS)
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+	struct dentry *root = minor->debugfs_root;
+	if (adev->enable_mes && amdgpu_mes_log_enable)
+		debugfs_create_file("amdgpu_mes_event_log", 0444, root,
+				    adev, &amdgpu_debugfs_mes_event_log_fops);
+
+#endif
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
new file mode 100644
index 000000000000..e989225b354b
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
@@ -0,0 +1,509 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_MES_H__
+#define __AMDGPU_MES_H__
+
+#include "amdgpu_irq.h"
+#include "kgd_kfd_interface.h"
+#include "amdgpu_gfx.h"
+#include "amdgpu_doorbell.h"
+#include <linux/sched/mm.h>
+
+#define AMDGPU_MES_MAX_COMPUTE_PIPES        8
+#define AMDGPU_MES_MAX_GFX_PIPES            2
+#define AMDGPU_MES_MAX_SDMA_PIPES           2
+
+#define AMDGPU_MES_API_VERSION_SHIFT	12
+#define AMDGPU_MES_FEAT_VERSION_SHIFT	24
+
+#define AMDGPU_MES_VERSION_MASK		0x00000fff
+#define AMDGPU_MES_API_VERSION_MASK	0x00fff000
+#define AMDGPU_MES_FEAT_VERSION_MASK	0xff000000
+#define AMDGPU_MES_MSCRATCH_SIZE	0x40000
+#define AMDGPU_MES_INVALID_DB_OFFSET	0xffffffff
+
+enum amdgpu_mes_priority_level {
+	AMDGPU_MES_PRIORITY_LEVEL_LOW       = 0,
+	AMDGPU_MES_PRIORITY_LEVEL_NORMAL    = 1,
+	AMDGPU_MES_PRIORITY_LEVEL_MEDIUM    = 2,
+	AMDGPU_MES_PRIORITY_LEVEL_HIGH      = 3,
+	AMDGPU_MES_PRIORITY_LEVEL_REALTIME  = 4,
+	AMDGPU_MES_PRIORITY_NUM_LEVELS
+};
+
+#define AMDGPU_MES_PROC_CTX_SIZE 0x1000 /* one page area */
+#define AMDGPU_MES_GANG_CTX_SIZE 0x1000 /* one page area */
+
+struct amdgpu_mes_funcs;
+
+enum amdgpu_mes_pipe {
+	AMDGPU_MES_SCHED_PIPE = 0,
+	AMDGPU_MES_KIQ_PIPE,
+	AMDGPU_MAX_MES_PIPES = 2,
+};
+
+struct amdgpu_mes {
+	struct amdgpu_device            *adev;
+
+	struct mutex                    mutex_hidden;
+
+	struct idr                      pasid_idr;
+	struct idr                      gang_id_idr;
+	struct idr                      queue_id_idr;
+	struct ida                      doorbell_ida;
+
+	spinlock_t                      queue_id_lock;
+
+	uint32_t			sched_version;
+	uint32_t			kiq_version;
+	uint32_t			fw_version[AMDGPU_MAX_MES_PIPES];
+	bool                            enable_legacy_queue_map;
+
+	uint32_t                        total_max_queue;
+	uint32_t                        max_doorbell_slices;
+
+	uint64_t                        default_process_quantum;
+	uint64_t                        default_gang_quantum;
+
+	struct amdgpu_ring              ring[AMDGPU_MAX_MES_PIPES];
+	spinlock_t                      ring_lock[AMDGPU_MAX_MES_PIPES];
+
+	const struct firmware           *fw[AMDGPU_MAX_MES_PIPES];
+
+	/* mes ucode */
+	struct amdgpu_bo		*ucode_fw_obj[AMDGPU_MAX_MES_PIPES];
+	uint64_t			ucode_fw_gpu_addr[AMDGPU_MAX_MES_PIPES];
+	uint32_t			*ucode_fw_ptr[AMDGPU_MAX_MES_PIPES];
+	uint64_t                        uc_start_addr[AMDGPU_MAX_MES_PIPES];
+
+	/* mes ucode data */
+	struct amdgpu_bo		*data_fw_obj[AMDGPU_MAX_MES_PIPES];
+	uint64_t			data_fw_gpu_addr[AMDGPU_MAX_MES_PIPES];
+	uint32_t			*data_fw_ptr[AMDGPU_MAX_MES_PIPES];
+	uint64_t                        data_start_addr[AMDGPU_MAX_MES_PIPES];
+
+	/* eop gpu obj */
+	struct amdgpu_bo		*eop_gpu_obj[AMDGPU_MAX_MES_PIPES];
+	uint64_t                        eop_gpu_addr[AMDGPU_MAX_MES_PIPES];
+
+	void                            *mqd_backup[AMDGPU_MAX_MES_PIPES];
+	struct amdgpu_irq_src	        irq[AMDGPU_MAX_MES_PIPES];
+
+	uint32_t                        vmid_mask_gfxhub;
+	uint32_t                        vmid_mask_mmhub;
+	uint32_t                        gfx_hqd_mask[AMDGPU_MES_MAX_GFX_PIPES];
+	uint32_t                        compute_hqd_mask[AMDGPU_MES_MAX_COMPUTE_PIPES];
+	uint32_t                        sdma_hqd_mask[AMDGPU_MES_MAX_SDMA_PIPES];
+	uint32_t                        aggregated_doorbells[AMDGPU_MES_PRIORITY_NUM_LEVELS];
+	uint32_t                        sch_ctx_offs[AMDGPU_MAX_MES_PIPES];
+	uint64_t			sch_ctx_gpu_addr[AMDGPU_MAX_MES_PIPES];
+	uint64_t			*sch_ctx_ptr[AMDGPU_MAX_MES_PIPES];
+	uint32_t			query_status_fence_offs[AMDGPU_MAX_MES_PIPES];
+	uint64_t			query_status_fence_gpu_addr[AMDGPU_MAX_MES_PIPES];
+	uint64_t			*query_status_fence_ptr[AMDGPU_MAX_MES_PIPES];
+
+	uint32_t			saved_flags;
+
+	/* initialize kiq pipe */
+	int                             (*kiq_hw_init)(struct amdgpu_device *adev);
+	int                             (*kiq_hw_fini)(struct amdgpu_device *adev);
+
+	/* MES doorbells */
+	uint32_t			db_start_dw_offset;
+	uint32_t			num_mes_dbs;
+	unsigned long			*doorbell_bitmap;
+
+	/* MES event log buffer */
+	uint32_t			event_log_size;
+	struct amdgpu_bo	*event_log_gpu_obj;
+	uint64_t			event_log_gpu_addr;
+	void				*event_log_cpu_addr;
+
+	/* ip specific functions */
+	const struct amdgpu_mes_funcs   *funcs;
+
+	/* mes resource_1 bo*/
+	struct amdgpu_bo    *resource_1[AMDGPU_MAX_MES_PIPES];
+	uint64_t            resource_1_gpu_addr[AMDGPU_MAX_MES_PIPES];
+	void                *resource_1_addr[AMDGPU_MAX_MES_PIPES];
+
+	int				hung_queue_db_array_size;
+	int				hung_queue_hqd_info_offset;
+	struct amdgpu_bo		*hung_queue_db_array_gpu_obj;
+	uint64_t			hung_queue_db_array_gpu_addr;
+	void				*hung_queue_db_array_cpu_addr;
+};
+
+struct amdgpu_mes_gang {
+	int 				gang_id;
+	int 				priority;
+	int 				inprocess_gang_priority;
+	int 				global_priority_level;
+	struct list_head 		list;
+	struct amdgpu_mes_process 	*process;
+	struct amdgpu_bo 		*gang_ctx_bo;
+	uint64_t 			gang_ctx_gpu_addr;
+	void 				*gang_ctx_cpu_ptr;
+	uint64_t 			gang_quantum;
+	struct list_head 		queue_list;
+};
+
+struct amdgpu_mes_queue {
+	struct list_head 		list;
+	struct amdgpu_mes_gang 		*gang;
+	int 				queue_id;
+	uint64_t 			doorbell_off;
+	struct amdgpu_bo		*mqd_obj;
+	void				*mqd_cpu_ptr;
+	uint64_t 			mqd_gpu_addr;
+	uint64_t 			wptr_gpu_addr;
+	int 				queue_type;
+	int 				paging;
+	struct amdgpu_ring 		*ring;
+};
+
+struct amdgpu_mes_queue_properties {
+	int 			queue_type;
+	uint64_t                hqd_base_gpu_addr;
+	uint64_t                rptr_gpu_addr;
+	uint64_t                wptr_gpu_addr;
+	uint64_t                wptr_mc_addr;
+	uint32_t                queue_size;
+	uint64_t                eop_gpu_addr;
+	uint32_t                hqd_pipe_priority;
+	uint32_t                hqd_queue_priority;
+	bool 			paging;
+	struct amdgpu_ring 	*ring;
+	/* out */
+	uint64_t       		doorbell_off;
+};
+
+struct amdgpu_mes_gang_properties {
+	uint32_t 	priority;
+	uint32_t 	gang_quantum;
+	uint32_t 	inprocess_gang_priority;
+	uint32_t 	priority_level;
+	int 		global_priority_level;
+};
+
+struct mes_add_queue_input {
+	uint32_t	process_id;
+	uint64_t	page_table_base_addr;
+	uint64_t	process_va_start;
+	uint64_t	process_va_end;
+	uint64_t	process_quantum;
+	uint64_t	process_context_addr;
+	uint64_t	gang_quantum;
+	uint64_t	gang_context_addr;
+	uint32_t	inprocess_gang_priority;
+	uint32_t	gang_global_priority_level;
+	uint32_t	doorbell_offset;
+	uint64_t	mqd_addr;
+	uint64_t	wptr_addr;
+	uint64_t	wptr_mc_addr;
+	uint32_t	queue_type;
+	uint32_t	paging;
+	uint32_t        gws_base;
+	uint32_t        gws_size;
+	uint64_t	tba_addr;
+	uint64_t	tma_addr;
+	uint32_t	trap_en;
+	uint32_t	skip_process_ctx_clear;
+	uint32_t	is_kfd_process;
+	uint32_t	is_aql_queue;
+	uint32_t	queue_size;
+	uint32_t	exclusively_scheduled;
+};
+
+struct mes_remove_queue_input {
+	uint32_t	doorbell_offset;
+	uint64_t	gang_context_addr;
+	bool		remove_queue_after_reset;
+};
+
+struct mes_map_legacy_queue_input {
+	uint32_t                           queue_type;
+	uint32_t                           doorbell_offset;
+	uint32_t                           pipe_id;
+	uint32_t                           queue_id;
+	uint64_t                           mqd_addr;
+	uint64_t                           wptr_addr;
+};
+
+struct mes_unmap_legacy_queue_input {
+	enum amdgpu_unmap_queues_action    action;
+	uint32_t                           queue_type;
+	uint32_t                           doorbell_offset;
+	uint32_t                           pipe_id;
+	uint32_t                           queue_id;
+	uint64_t                           trail_fence_addr;
+	uint64_t                           trail_fence_data;
+};
+
+struct mes_suspend_gang_input {
+	bool		suspend_all_gangs;
+	uint64_t	gang_context_addr;
+	uint64_t	suspend_fence_addr;
+	uint32_t	suspend_fence_value;
+};
+
+struct mes_resume_gang_input {
+	bool		resume_all_gangs;
+	uint64_t	gang_context_addr;
+};
+
+struct mes_reset_queue_input {
+	uint32_t                           queue_type;
+	uint32_t                           doorbell_offset;
+	bool                               use_mmio;
+	uint32_t                           me_id;
+	uint32_t                           pipe_id;
+	uint32_t                           queue_id;
+	uint64_t                           mqd_addr;
+	uint64_t                           wptr_addr;
+	uint32_t                           vmid;
+	bool                               legacy_gfx;
+	bool                               is_kq;
+};
+
+struct mes_detect_and_reset_queue_input {
+	uint32_t                           queue_type;
+	bool                               detect_only;
+};
+
+struct mes_inv_tlbs_pasid_input {
+	uint32_t        xcc_id;
+	uint16_t        pasid;
+	uint8_t         hub_id;
+	uint8_t         flush_type;
+};
+
+enum mes_misc_opcode {
+	MES_MISC_OP_WRITE_REG,
+	MES_MISC_OP_READ_REG,
+	MES_MISC_OP_WRM_REG_WAIT,
+	MES_MISC_OP_WRM_REG_WR_WAIT,
+	MES_MISC_OP_SET_SHADER_DEBUGGER,
+	MES_MISC_OP_CHANGE_CONFIG,
+};
+
+struct mes_misc_op_input {
+	enum mes_misc_opcode op;
+
+	union {
+		struct {
+			uint32_t                  reg_offset;
+			uint64_t                  buffer_addr;
+		} read_reg;
+
+		struct {
+			uint32_t                  reg_offset;
+			uint32_t                  reg_value;
+		} write_reg;
+
+		struct {
+			uint32_t                   ref;
+			uint32_t                   mask;
+			uint32_t                   reg0;
+			uint32_t                   reg1;
+		} wrm_reg;
+
+		struct {
+			uint64_t process_context_addr;
+			union {
+				struct {
+					uint32_t single_memop : 1;
+					uint32_t single_alu_op : 1;
+					uint32_t reserved: 29;
+					uint32_t process_ctx_flush: 1;
+				};
+				uint32_t u32all;
+			} flags;
+			uint32_t spi_gdbg_per_vmid_cntl;
+			uint32_t tcp_watch_cntl[4];
+			uint32_t trap_en;
+		} set_shader_debugger;
+
+		struct {
+			union {
+				struct {
+					uint32_t limit_single_process : 1;
+					uint32_t enable_hws_logging_buffer : 1;
+					uint32_t reserved : 30;
+				};
+				uint32_t all;
+			} option;
+			struct {
+				uint32_t tdr_level;
+				uint32_t tdr_delay;
+			} tdr_config;
+		} change_config;
+	};
+};
+
+struct amdgpu_mes_funcs {
+	int (*add_hw_queue)(struct amdgpu_mes *mes,
+			    struct mes_add_queue_input *input);
+
+	int (*remove_hw_queue)(struct amdgpu_mes *mes,
+			       struct mes_remove_queue_input *input);
+
+	int (*map_legacy_queue)(struct amdgpu_mes *mes,
+				struct mes_map_legacy_queue_input *input);
+
+	int (*unmap_legacy_queue)(struct amdgpu_mes *mes,
+				  struct mes_unmap_legacy_queue_input *input);
+
+	int (*suspend_gang)(struct amdgpu_mes *mes,
+			    struct mes_suspend_gang_input *input);
+
+	int (*resume_gang)(struct amdgpu_mes *mes,
+			   struct mes_resume_gang_input *input);
+
+	int (*misc_op)(struct amdgpu_mes *mes,
+		       struct mes_misc_op_input *input);
+
+	int (*reset_hw_queue)(struct amdgpu_mes *mes,
+			      struct mes_reset_queue_input *input);
+
+	int (*detect_and_reset_hung_queues)(struct amdgpu_mes *mes,
+			      struct mes_detect_and_reset_queue_input *input);
+
+
+	int (*invalidate_tlbs_pasid)(struct amdgpu_mes *mes,
+			      struct mes_inv_tlbs_pasid_input *input);
+};
+
+#define amdgpu_mes_kiq_hw_init(adev) (adev)->mes.kiq_hw_init((adev))
+#define amdgpu_mes_kiq_hw_fini(adev) (adev)->mes.kiq_hw_fini((adev))
+
+int amdgpu_mes_init_microcode(struct amdgpu_device *adev, int pipe);
+int amdgpu_mes_init(struct amdgpu_device *adev);
+void amdgpu_mes_fini(struct amdgpu_device *adev);
+
+int amdgpu_mes_suspend(struct amdgpu_device *adev);
+int amdgpu_mes_resume(struct amdgpu_device *adev);
+
+int amdgpu_mes_map_legacy_queue(struct amdgpu_device *adev,
+				struct amdgpu_ring *ring);
+int amdgpu_mes_unmap_legacy_queue(struct amdgpu_device *adev,
+				  struct amdgpu_ring *ring,
+				  enum amdgpu_unmap_queues_action action,
+				  u64 gpu_addr, u64 seq);
+int amdgpu_mes_reset_legacy_queue(struct amdgpu_device *adev,
+				  struct amdgpu_ring *ring,
+				  unsigned int vmid,
+				  bool use_mmio);
+
+int amdgpu_mes_get_hung_queue_db_array_size(struct amdgpu_device *adev);
+int amdgpu_mes_detect_and_reset_hung_queues(struct amdgpu_device *adev,
+					    int queue_type,
+					    bool detect_only,
+					    unsigned int *hung_db_num,
+					    u32 *hung_db_array);
+
+uint32_t amdgpu_mes_rreg(struct amdgpu_device *adev, uint32_t reg);
+int amdgpu_mes_wreg(struct amdgpu_device *adev,
+		    uint32_t reg, uint32_t val);
+int amdgpu_mes_reg_write_reg_wait(struct amdgpu_device *adev,
+				  uint32_t reg0, uint32_t reg1,
+				  uint32_t ref, uint32_t mask);
+int amdgpu_mes_hdp_flush(struct amdgpu_device *adev);
+int amdgpu_mes_set_shader_debugger(struct amdgpu_device *adev,
+				uint64_t process_context_addr,
+				uint32_t spi_gdbg_per_vmid_cntl,
+				const uint32_t *tcp_watch_cntl,
+				uint32_t flags,
+				bool trap_en);
+int amdgpu_mes_flush_shader_debugger(struct amdgpu_device *adev,
+				uint64_t process_context_addr);
+
+uint32_t amdgpu_mes_get_aggregated_doorbell_index(struct amdgpu_device *adev,
+						   enum amdgpu_mes_priority_level prio);
+
+int amdgpu_mes_doorbell_process_slice(struct amdgpu_device *adev);
+
+/*
+ * MES lock can be taken in MMU notifiers.
+ *
+ * A bit more detail about why to set no-FS reclaim with MES lock:
+ *
+ * The purpose of the MMU notifier is to stop GPU access to memory so
+ * that the Linux VM subsystem can move pages around safely. This is
+ * done by preempting user mode queues for the affected process. When
+ * MES is used, MES lock needs to be taken to preempt the queues.
+ *
+ * The MMU notifier callback entry point in the driver is
+ * amdgpu_mn_invalidate_range_start_hsa. The relevant call chain from
+ * there is:
+ * amdgpu_amdkfd_evict_userptr -> kgd2kfd_quiesce_mm ->
+ * kfd_process_evict_queues -> pdd->dev->dqm->ops.evict_process_queues
+ *
+ * The last part of the chain is a function pointer where we take the
+ * MES lock.
+ *
+ * The problem with taking locks in the MMU notifier is, that MMU
+ * notifiers can be called in reclaim-FS context. That's where the
+ * kernel frees up pages to make room for new page allocations under
+ * memory pressure. While we are running in reclaim-FS context, we must
+ * not trigger another memory reclaim operation because that would
+ * recursively reenter the reclaim code and cause a deadlock. The
+ * memalloc_nofs_save/restore calls guarantee that.
+ *
+ * In addition we also need to avoid lock dependencies on other locks taken
+ * under the MES lock, for example reservation locks. Here is a possible
+ * scenario of a deadlock:
+ * Thread A: takes and holds reservation lock | triggers reclaim-FS |
+ * MMU notifier | blocks trying to take MES lock
+ * Thread B: takes and holds MES lock | blocks trying to take reservation lock
+ *
+ * In this scenario Thread B gets involved in a deadlock even without
+ * triggering a reclaim-FS operation itself.
+ * To fix this and break the lock dependency chain you'd need to either:
+ * 1. protect reservation locks with memalloc_nofs_save/restore, or
+ * 2. avoid taking reservation locks under the MES lock.
+ *
+ * Reservation locks are taken all over the kernel in different subsystems, we
+ * have no control over them and their lock dependencies.So the only workable
+ * solution is to avoid taking other locks under the MES lock.
+ * As a result, make sure no reclaim-FS happens while holding this lock anywhere
+ * to prevent deadlocks when an MMU notifier runs in reclaim-FS context.
+ */
+static inline void amdgpu_mes_lock(struct amdgpu_mes *mes)
+{
+	mutex_lock(&mes->mutex_hidden);
+	mes->saved_flags = memalloc_noreclaim_save();
+}
+
+static inline void amdgpu_mes_unlock(struct amdgpu_mes *mes)
+{
+	memalloc_noreclaim_restore(mes->saved_flags);
+	mutex_unlock(&mes->mutex_hidden);
+}
+
+bool amdgpu_mes_suspend_resume_all_supported(struct amdgpu_device *adev);
+
+int amdgpu_mes_update_enforce_isolation(struct amdgpu_device *adev);
+
+#endif /* __AMDGPU_MES_H__ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes_ctx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes_ctx.h
new file mode 100644
index 000000000000..912a5be2ece6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes_ctx.h
@@ -0,0 +1,122 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_MES_CTX_H__
+#define __AMDGPU_MES_CTX_H__
+
+#include "v10_structs.h"
+
+enum {
+	AMDGPU_MES_CTX_RPTR_OFFS = 0,
+	AMDGPU_MES_CTX_WPTR_OFFS,
+	AMDGPU_MES_CTX_FENCE_OFFS,
+	AMDGPU_MES_CTX_COND_EXE_OFFS,
+	AMDGPU_MES_CTX_TRAIL_FENCE_OFFS,
+	AMDGPU_MES_CTX_MAX_OFFS,
+};
+
+enum {
+	AMDGPU_MES_CTX_RING_OFFS = AMDGPU_MES_CTX_MAX_OFFS,
+	AMDGPU_MES_CTX_IB_OFFS,
+	AMDGPU_MES_CTX_PADDING_OFFS,
+};
+
+#define AMDGPU_MES_CTX_MAX_GFX_RINGS            1
+#define AMDGPU_MES_CTX_MAX_COMPUTE_RINGS        4
+#define AMDGPU_MES_CTX_MAX_SDMA_RINGS           2
+#define AMDGPU_MES_CTX_MAX_RINGS					\
+	(AMDGPU_MES_CTX_MAX_GFX_RINGS +					\
+	 AMDGPU_MES_CTX_MAX_COMPUTE_RINGS +				\
+	 AMDGPU_MES_CTX_MAX_SDMA_RINGS)
+
+#define AMDGPU_CSA_SDMA_SIZE    64
+#define GFX10_MEC_HPD_SIZE	2048
+
+struct amdgpu_wb_slot {
+	uint32_t data[8];
+};
+
+struct amdgpu_mes_ctx_meta_data {
+	struct {
+		uint8_t ring[PAGE_SIZE * 4];
+
+		/* gfx csa */
+		struct v10_gfx_meta_data gfx_meta_data;
+
+		uint8_t gds_backup[64 * 1024];
+
+		struct amdgpu_wb_slot slots[AMDGPU_MES_CTX_MAX_OFFS];
+
+		/* only for ib test */
+		uint32_t ib[256] __aligned(256);
+
+		uint32_t padding[64];
+
+	} __aligned(PAGE_SIZE) gfx[AMDGPU_MES_CTX_MAX_GFX_RINGS];
+
+	struct {
+		uint8_t ring[PAGE_SIZE * 4];
+
+		uint8_t mec_hpd[GFX10_MEC_HPD_SIZE];
+
+		struct amdgpu_wb_slot slots[AMDGPU_MES_CTX_MAX_OFFS];
+
+		/* only for ib test */
+		uint32_t ib[256] __aligned(256);
+
+		uint32_t padding[64];
+
+	} __aligned(PAGE_SIZE) compute[AMDGPU_MES_CTX_MAX_COMPUTE_RINGS];
+
+	struct {
+		uint8_t ring[PAGE_SIZE * 4];
+
+		/* sdma csa for mcbp */
+		uint8_t sdma_meta_data[AMDGPU_CSA_SDMA_SIZE];
+
+		struct amdgpu_wb_slot slots[AMDGPU_MES_CTX_MAX_OFFS];
+
+		/* only for ib test */
+		uint32_t ib[256] __aligned(256);
+
+		uint32_t padding[64];
+
+	} __aligned(PAGE_SIZE) sdma[AMDGPU_MES_CTX_MAX_SDMA_RINGS];
+};
+
+struct amdgpu_mes_ctx_data {
+	struct amdgpu_bo	*meta_data_obj;
+	uint64_t                meta_data_gpu_addr;
+	uint64_t                meta_data_mc_addr;
+	struct amdgpu_bo_va	*meta_data_va;
+	void                    *meta_data_ptr;
+	uint32_t                gang_ids[AMDGPU_HW_IP_DMA+1];
+};
+
+#define AMDGPU_FENCE_MES_QUEUE_FLAG     0x1000000u
+#define AMDGPU_FENCE_MES_QUEUE_ID_MASK  (AMDGPU_FENCE_MES_QUEUE_FLAG - 1)
+
+#define AMDGPU_FENCE_MES_QUEUE_FLAG     0x1000000u
+#define AMDGPU_FENCE_MES_QUEUE_ID_MASK  (AMDGPU_FENCE_MES_QUEUE_FLAG - 1)
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mmhub.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mmhub.c
new file mode 100644
index 000000000000..0f6b1021fef3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mmhub.c
@@ -0,0 +1,46 @@
+/*
+ * Copyright (C) 2023  Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#include "amdgpu.h"
+#include "amdgpu_ras.h"
+
+int amdgpu_mmhub_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err;
+	struct amdgpu_mmhub_ras *ras;
+
+	if (!adev->mmhub.ras)
+		return 0;
+
+	ras = adev->mmhub.ras;
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register mmhub ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "mmhub");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__MMHUB;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+	adev->mmhub.ras_if = &ras->ras_block.ras_comm;
+
+	/* mmhub ras follows amdgpu_ras_block_late_init_default for late init */
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mmhub.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mmhub.h
new file mode 100644
index 000000000000..1ca9d4ed8063
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mmhub.h
@@ -0,0 +1,77 @@
+/*
+ * Copyright (C) 2019  Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#ifndef __AMDGPU_MMHUB_H__
+#define __AMDGPU_MMHUB_H__
+
+enum amdgpu_mmhub_ras_memory_id {
+	AMDGPU_MMHUB_WGMI_PAGEMEM = 0,
+	AMDGPU_MMHUB_RGMI_PAGEMEM = 1,
+	AMDGPU_MMHUB_WDRAM_PAGEMEM = 2,
+	AMDGPU_MMHUB_RDRAM_PAGEMEM = 3,
+	AMDGPU_MMHUB_WIO_CMDMEM = 4,
+	AMDGPU_MMHUB_RIO_CMDMEM = 5,
+	AMDGPU_MMHUB_WGMI_CMDMEM = 6,
+	AMDGPU_MMHUB_RGMI_CMDMEM = 7,
+	AMDGPU_MMHUB_WDRAM_CMDMEM = 8,
+	AMDGPU_MMHUB_RDRAM_CMDMEM = 9,
+	AMDGPU_MMHUB_MAM_DMEM0 = 10,
+	AMDGPU_MMHUB_MAM_DMEM1 = 11,
+	AMDGPU_MMHUB_MAM_DMEM2 = 12,
+	AMDGPU_MMHUB_MAM_DMEM3 = 13,
+	AMDGPU_MMHUB_WRET_TAGMEM = 19,
+	AMDGPU_MMHUB_RRET_TAGMEM = 20,
+	AMDGPU_MMHUB_WIO_DATAMEM = 21,
+	AMDGPU_MMHUB_WGMI_DATAMEM = 22,
+	AMDGPU_MMHUB_WDRAM_DATAMEM = 23,
+	AMDGPU_MMHUB_MEMORY_BLOCK_LAST,
+};
+
+struct amdgpu_mmhub_ras {
+	struct amdgpu_ras_block_object ras_block;
+};
+
+struct amdgpu_mmhub_funcs {
+	u64 (*get_fb_location)(struct amdgpu_device *adev);
+	u64 (*get_mc_fb_offset)(struct amdgpu_device *adev);
+	void (*init)(struct amdgpu_device *adev);
+	int (*gart_enable)(struct amdgpu_device *adev);
+	void (*set_fault_enable_default)(struct amdgpu_device *adev,
+			bool value);
+	void (*gart_disable)(struct amdgpu_device *adev);
+	int (*set_clockgating)(struct amdgpu_device *adev,
+			       enum amd_clockgating_state state);
+	void (*get_clockgating)(struct amdgpu_device *adev, u64 *flags);
+	void (*setup_vm_pt_regs)(struct amdgpu_device *adev, uint32_t vmid,
+				uint64_t page_table_base);
+	void (*update_power_gating)(struct amdgpu_device *adev,
+                                bool enable);
+};
+
+struct amdgpu_mmhub {
+	struct ras_common_if *ras_if;
+	const struct amdgpu_mmhub_funcs *funcs;
+	struct amdgpu_mmhub_ras  *ras;
+};
+
+int amdgpu_mmhub_ras_sw_init(struct amdgpu_device *adev);
+
+#endif
+
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
deleted file mode 100644
index 38f739fb727b..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
+++ /dev/null
@@ -1,373 +0,0 @@
-/*
- * Copyright 2014 Advanced Micro Devices, Inc.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- */
-/*
- * Authors:
- *    Christian König <christian.koenig@amd.com>
- */
-
-#include <linux/firmware.h>
-#include <linux/module.h>
-#include <linux/mmu_notifier.h>
-#include <linux/interval_tree.h>
-#include <drm/drmP.h>
-#include <drm/drm.h>
-
-#include "amdgpu.h"
-
-struct amdgpu_mn {
-	/* constant after initialisation */
-	struct amdgpu_device	*adev;
-	struct mm_struct	*mm;
-	struct mmu_notifier	mn;
-
-	/* only used on destruction */
-	struct work_struct	work;
-
-	/* protected by adev->mn_lock */
-	struct hlist_node	node;
-
-	/* objects protected by lock */
-	struct mutex		lock;
-	struct rb_root		objects;
-};
-
-struct amdgpu_mn_node {
-	struct interval_tree_node	it;
-	struct list_head		bos;
-};
-
-/**
- * amdgpu_mn_destroy - destroy the rmn
- *
- * @work: previously sheduled work item
- *
- * Lazy destroys the notifier from a work item
- */
-static void amdgpu_mn_destroy(struct work_struct *work)
-{
-	struct amdgpu_mn *rmn = container_of(work, struct amdgpu_mn, work);
-	struct amdgpu_device *adev = rmn->adev;
-	struct amdgpu_mn_node *node, *next_node;
-	struct amdgpu_bo *bo, *next_bo;
-
-	mutex_lock(&adev->mn_lock);
-	mutex_lock(&rmn->lock);
-	hash_del(&rmn->node);
-	rbtree_postorder_for_each_entry_safe(node, next_node, &rmn->objects,
-					     it.rb) {
-		list_for_each_entry_safe(bo, next_bo, &node->bos, mn_list) {
-			bo->mn = NULL;
-			list_del_init(&bo->mn_list);
-		}
-		kfree(node);
-	}
-	mutex_unlock(&rmn->lock);
-	mutex_unlock(&adev->mn_lock);
-	mmu_notifier_unregister_no_release(&rmn->mn, rmn->mm);
-	kfree(rmn);
-}
-
-/**
- * amdgpu_mn_release - callback to notify about mm destruction
- *
- * @mn: our notifier
- * @mn: the mm this callback is about
- *
- * Shedule a work item to lazy destroy our notifier.
- */
-static void amdgpu_mn_release(struct mmu_notifier *mn,
-			      struct mm_struct *mm)
-{
-	struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn);
-	INIT_WORK(&rmn->work, amdgpu_mn_destroy);
-	schedule_work(&rmn->work);
-}
-
-/**
- * amdgpu_mn_invalidate_node - unmap all BOs of a node
- *
- * @node: the node with the BOs to unmap
- *
- * We block for all BOs and unmap them by move them
- * into system domain again.
- */
-static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node,
-				      unsigned long start,
-				      unsigned long end)
-{
-	struct amdgpu_bo *bo;
-	long r;
-
-	list_for_each_entry(bo, &node->bos, mn_list) {
-
-		if (!amdgpu_ttm_tt_affect_userptr(bo->tbo.ttm, start, end))
-			continue;
-
-		r = amdgpu_bo_reserve(bo, true);
-		if (r) {
-			DRM_ERROR("(%ld) failed to reserve user bo\n", r);
-			continue;
-		}
-
-		r = reservation_object_wait_timeout_rcu(bo->tbo.resv,
-			true, false, MAX_SCHEDULE_TIMEOUT);
-		if (r <= 0)
-			DRM_ERROR("(%ld) failed to wait for user bo\n", r);
-
-		amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
-		r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-		if (r)
-			DRM_ERROR("(%ld) failed to validate user bo\n", r);
-
-		amdgpu_bo_unreserve(bo);
-	}
-}
-
-/**
- * amdgpu_mn_invalidate_page - callback to notify about mm change
- *
- * @mn: our notifier
- * @mn: the mm this callback is about
- * @address: address of invalidate page
- *
- * Invalidation of a single page. Blocks for all BOs mapping it
- * and unmap them by move them into system domain again.
- */
-static void amdgpu_mn_invalidate_page(struct mmu_notifier *mn,
-				      struct mm_struct *mm,
-				      unsigned long address)
-{
-	struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn);
-	struct interval_tree_node *it;
-
-	mutex_lock(&rmn->lock);
-
-	it = interval_tree_iter_first(&rmn->objects, address, address);
-	if (it) {
-		struct amdgpu_mn_node *node;
-
-		node = container_of(it, struct amdgpu_mn_node, it);
-		amdgpu_mn_invalidate_node(node, address, address);
-	}
-
-	mutex_unlock(&rmn->lock);
-}
-
-/**
- * amdgpu_mn_invalidate_range_start - callback to notify about mm change
- *
- * @mn: our notifier
- * @mn: the mm this callback is about
- * @start: start of updated range
- * @end: end of updated range
- *
- * We block for all BOs between start and end to be idle and
- * unmap them by move them into system domain again.
- */
-static void amdgpu_mn_invalidate_range_start(struct mmu_notifier *mn,
-					     struct mm_struct *mm,
-					     unsigned long start,
-					     unsigned long end)
-{
-	struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn);
-	struct interval_tree_node *it;
-
-	/* notification is exclusive, but interval is inclusive */
-	end -= 1;
-
-	mutex_lock(&rmn->lock);
-
-	it = interval_tree_iter_first(&rmn->objects, start, end);
-	while (it) {
-		struct amdgpu_mn_node *node;
-
-		node = container_of(it, struct amdgpu_mn_node, it);
-		it = interval_tree_iter_next(it, start, end);
-
-		amdgpu_mn_invalidate_node(node, start, end);
-	}
-
-	mutex_unlock(&rmn->lock);
-}
-
-static const struct mmu_notifier_ops amdgpu_mn_ops = {
-	.release = amdgpu_mn_release,
-	.invalidate_page = amdgpu_mn_invalidate_page,
-	.invalidate_range_start = amdgpu_mn_invalidate_range_start,
-};
-
-/**
- * amdgpu_mn_get - create notifier context
- *
- * @adev: amdgpu device pointer
- *
- * Creates a notifier context for current->mm.
- */
-static struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev)
-{
-	struct mm_struct *mm = current->mm;
-	struct amdgpu_mn *rmn;
-	int r;
-
-	mutex_lock(&adev->mn_lock);
-	if (down_write_killable(&mm->mmap_sem)) {
-		mutex_unlock(&adev->mn_lock);
-		return ERR_PTR(-EINTR);
-	}
-
-	hash_for_each_possible(adev->mn_hash, rmn, node, (unsigned long)mm)
-		if (rmn->mm == mm)
-			goto release_locks;
-
-	rmn = kzalloc(sizeof(*rmn), GFP_KERNEL);
-	if (!rmn) {
-		rmn = ERR_PTR(-ENOMEM);
-		goto release_locks;
-	}
-
-	rmn->adev = adev;
-	rmn->mm = mm;
-	rmn->mn.ops = &amdgpu_mn_ops;
-	mutex_init(&rmn->lock);
-	rmn->objects = RB_ROOT;
-
-	r = __mmu_notifier_register(&rmn->mn, mm);
-	if (r)
-		goto free_rmn;
-
-	hash_add(adev->mn_hash, &rmn->node, (unsigned long)mm);
-
-release_locks:
-	up_write(&mm->mmap_sem);
-	mutex_unlock(&adev->mn_lock);
-
-	return rmn;
-
-free_rmn:
-	up_write(&mm->mmap_sem);
-	mutex_unlock(&adev->mn_lock);
-	kfree(rmn);
-
-	return ERR_PTR(r);
-}
-
-/**
- * amdgpu_mn_register - register a BO for notifier updates
- *
- * @bo: amdgpu buffer object
- * @addr: userptr addr we should monitor
- *
- * Registers an MMU notifier for the given BO at the specified address.
- * Returns 0 on success, -ERRNO if anything goes wrong.
- */
-int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr)
-{
-	unsigned long end = addr + amdgpu_bo_size(bo) - 1;
-	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	struct amdgpu_mn *rmn;
-	struct amdgpu_mn_node *node = NULL;
-	struct list_head bos;
-	struct interval_tree_node *it;
-
-	rmn = amdgpu_mn_get(adev);
-	if (IS_ERR(rmn))
-		return PTR_ERR(rmn);
-
-	INIT_LIST_HEAD(&bos);
-
-	mutex_lock(&rmn->lock);
-
-	while ((it = interval_tree_iter_first(&rmn->objects, addr, end))) {
-		kfree(node);
-		node = container_of(it, struct amdgpu_mn_node, it);
-		interval_tree_remove(&node->it, &rmn->objects);
-		addr = min(it->start, addr);
-		end = max(it->last, end);
-		list_splice(&node->bos, &bos);
-	}
-
-	if (!node) {
-		node = kmalloc(sizeof(struct amdgpu_mn_node), GFP_KERNEL);
-		if (!node) {
-			mutex_unlock(&rmn->lock);
-			return -ENOMEM;
-		}
-	}
-
-	bo->mn = rmn;
-
-	node->it.start = addr;
-	node->it.last = end;
-	INIT_LIST_HEAD(&node->bos);
-	list_splice(&bos, &node->bos);
-	list_add(&bo->mn_list, &node->bos);
-
-	interval_tree_insert(&node->it, &rmn->objects);
-
-	mutex_unlock(&rmn->lock);
-
-	return 0;
-}
-
-/**
- * amdgpu_mn_unregister - unregister a BO for notifier updates
- *
- * @bo: amdgpu buffer object
- *
- * Remove any registration of MMU notifier updates from the buffer object.
- */
-void amdgpu_mn_unregister(struct amdgpu_bo *bo)
-{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	struct amdgpu_mn *rmn;
-	struct list_head *head;
-
-	mutex_lock(&adev->mn_lock);
-
-	rmn = bo->mn;
-	if (rmn == NULL) {
-		mutex_unlock(&adev->mn_lock);
-		return;
-	}
-
-	mutex_lock(&rmn->lock);
-
-	/* save the next list entry for later */
-	head = bo->mn_list.next;
-
-	bo->mn = NULL;
-	list_del(&bo->mn_list);
-
-	if (list_empty(head)) {
-		struct amdgpu_mn_node *node;
-		node = container_of(head, struct amdgpu_mn_node, bos);
-		interval_tree_remove(&node->it, &rmn->objects);
-		kfree(node);
-	}
-
-	mutex_unlock(&rmn->lock);
-	mutex_unlock(&adev->mn_lock);
-}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h
index dbd10618ec20..dc8d2f52c7d6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h
@@ -30,32 +30,39 @@
 #ifndef AMDGPU_MODE_H
 #define AMDGPU_MODE_H
 
+#include <drm/display/drm_dp_helper.h>
 #include <drm/drm_crtc.h>
-#include <drm/drm_edid.h>
 #include <drm/drm_encoder.h>
-#include <drm/drm_dp_helper.h>
 #include <drm/drm_fixed.h>
-#include <drm/drm_crtc_helper.h>
-#include <drm/drm_fb_helper.h>
-#include <drm/drm_plane_helper.h>
+#include <drm/drm_framebuffer.h>
+#include <drm/drm_probe_helper.h>
 #include <linux/i2c.h>
 #include <linux/i2c-algo-bit.h>
 #include <linux/hrtimer.h>
 #include "amdgpu_irq.h"
 
+#include <drm/display/drm_dp_mst_helper.h>
+#include "modules/inc/mod_freesync.h"
+#include "amdgpu_dm_irq_params.h"
+
 struct amdgpu_bo;
 struct amdgpu_device;
 struct amdgpu_encoder;
 struct amdgpu_router;
 struct amdgpu_hpd;
+struct edid;
+struct drm_edid;
 
 #define to_amdgpu_crtc(x) container_of(x, struct amdgpu_crtc, base)
 #define to_amdgpu_connector(x) container_of(x, struct amdgpu_connector, base)
 #define to_amdgpu_encoder(x) container_of(x, struct amdgpu_encoder, base)
 #define to_amdgpu_framebuffer(x) container_of(x, struct amdgpu_framebuffer, base)
 
+#define to_dm_plane_state(x)	container_of(x, struct dm_plane_state, base)
+
 #define AMDGPU_MAX_HPD_PINS 6
 #define AMDGPU_MAX_CRTCS 6
+#define AMDGPU_MAX_PLANES 6
 #define AMDGPU_MAX_AFMT_BLOCKS 9
 
 enum amdgpu_rmx_type {
@@ -81,7 +88,6 @@ enum amdgpu_hpd_id {
 	AMDGPU_HPD_4,
 	AMDGPU_HPD_5,
 	AMDGPU_HPD_6,
-	AMDGPU_HPD_LAST,
 	AMDGPU_HPD_NONE = 0xff,
 };
 
@@ -98,7 +104,6 @@ enum amdgpu_crtc_irq {
 	AMDGPU_CRTC_IRQ_VLINE4,
 	AMDGPU_CRTC_IRQ_VLINE5,
 	AMDGPU_CRTC_IRQ_VLINE6,
-	AMDGPU_CRTC_IRQ_LAST,
 	AMDGPU_CRTC_IRQ_NONE = 0xff
 };
 
@@ -109,7 +114,6 @@ enum amdgpu_pageflip_irq {
 	AMDGPU_PAGEFLIP_IRQ_D4,
 	AMDGPU_PAGEFLIP_IRQ_D5,
 	AMDGPU_PAGEFLIP_IRQ_D6,
-	AMDGPU_PAGEFLIP_IRQ_LAST,
 	AMDGPU_PAGEFLIP_IRQ_NONE = 0xff
 };
 
@@ -227,8 +231,6 @@ struct amdgpu_i2c_chan {
 	struct mutex mutex;
 };
 
-struct amdgpu_fbdev;
-
 struct amdgpu_afmt {
 	bool enabled;
 	int offset;
@@ -257,21 +259,11 @@ struct amdgpu_audio {
 	int num_pins;
 };
 
-struct amdgpu_mode_mc_save {
-	u32 vga_render_control;
-	u32 vga_hdp_control;
-	bool crtc_enabled[AMDGPU_MAX_CRTCS];
-};
-
 struct amdgpu_display_funcs {
-	/* vga render */
-	void (*set_vga_render_state)(struct amdgpu_device *adev, bool render);
 	/* display watermarks */
 	void (*bandwidth_update)(struct amdgpu_device *adev);
 	/* get frame count */
 	u32 (*vblank_get_counter)(struct amdgpu_device *adev, int crtc);
-	/* wait for vblank */
-	void (*vblank_wait)(struct amdgpu_device *adev, int crtc);
 	/* set backlight level */
 	void (*backlight_set_level)(struct amdgpu_encoder *amdgpu_encoder,
 				    u8 level);
@@ -300,10 +292,19 @@ struct amdgpu_display_funcs {
 			      uint16_t connector_object_id,
 			      struct amdgpu_hpd *hpd,
 			      struct amdgpu_router *router);
-	void (*stop_mc_access)(struct amdgpu_device *adev,
-			       struct amdgpu_mode_mc_save *save);
-	void (*resume_mc_access)(struct amdgpu_device *adev,
-				 struct amdgpu_mode_mc_save *save);
+
+
+};
+
+struct amdgpu_framebuffer {
+	struct drm_framebuffer base;
+
+	uint64_t tiling_flags;
+	bool tmz_surface;
+	bool gfx12_dcc;
+
+	/* caching for later use */
+	uint64_t address;
 };
 
 struct amdgpu_mode_info {
@@ -311,6 +312,7 @@ struct amdgpu_mode_info {
 	struct card_info *atom_card_info;
 	bool mode_config_initialized;
 	struct amdgpu_crtc *crtcs[AMDGPU_MAX_CRTCS];
+	struct drm_plane *planes[AMDGPU_MAX_PLANES];
 	struct amdgpu_afmt *afmt[AMDGPU_MAX_AFMT_BLOCKS];
 	/* DVI-I properties */
 	struct drm_property *coherent_mode_property;
@@ -324,35 +326,124 @@ struct amdgpu_mode_info {
 	struct drm_property *audio_property;
 	/* FMT dithering */
 	struct drm_property *dither_property;
+	/* Adaptive Backlight Modulation (power feature) */
+	struct drm_property *abm_level_property;
 	/* hardcoded DFP edid from BIOS */
-	struct edid *bios_hardcoded_edid;
-	int bios_hardcoded_edid_size;
+	const struct drm_edid *bios_hardcoded_edid;
 
-	/* pointer to fbdev info structure */
-	struct amdgpu_fbdev *rfbdev;
 	/* firmware flags */
-	u16 firmware_flags;
+	u32 firmware_flags;
 	/* pointer to backlight encoder */
 	struct amdgpu_encoder *bl_encoder;
+	u8 bl_level; /* saved backlight level */
 	struct amdgpu_audio	audio; /* audio stuff */
 	int			num_crtc; /* number of crtcs */
 	int			num_hpd; /* number of hpd pins */
 	int			num_dig; /* number of dig blocks */
+	bool			gpu_vm_support; /* supports display from GTT */
 	int			disp_priority;
 	const struct amdgpu_display_funcs *funcs;
+	const enum drm_plane_type *plane_type;
+
+	/* Driver-private color mgmt props */
+
+	/* @plane_degamma_lut_property: Plane property to set a degamma LUT to
+	 * convert encoded values to light linear values before sampling or
+	 * blending.
+	 */
+	struct drm_property *plane_degamma_lut_property;
+	/* @plane_degamma_lut_size_property: Plane property to define the max
+	 * size of degamma LUT as supported by the driver (read-only).
+	 */
+	struct drm_property *plane_degamma_lut_size_property;
+	/**
+	 * @plane_degamma_tf_property: Plane pre-defined transfer function to
+	 * to go from scanout/encoded values to linear values.
+	 */
+	struct drm_property *plane_degamma_tf_property;
+	/**
+	 * @plane_hdr_mult_property:
+	 */
+	struct drm_property *plane_hdr_mult_property;
+
+	struct drm_property *plane_ctm_property;
+	/**
+	 * @shaper_lut_property: Plane property to set pre-blending shaper LUT
+	 * that converts color content before 3D LUT. If
+	 * plane_shaper_tf_property != Identity TF, AMD color module will
+	 * combine the user LUT values with pre-defined TF into the LUT
+	 * parameters to be programmed.
+	 */
+	struct drm_property *plane_shaper_lut_property;
+	/**
+	 * @shaper_lut_size_property: Plane property for the size of
+	 * pre-blending shaper LUT as supported by the driver (read-only).
+	 */
+	struct drm_property *plane_shaper_lut_size_property;
+	/**
+	 * @plane_shaper_tf_property: Plane property to set a predefined
+	 * transfer function for pre-blending shaper (before applying 3D LUT)
+	 * with or without LUT. There is no shaper ROM, but we can use AMD
+	 * color modules to program LUT parameters from predefined TF (or
+	 * from a combination of pre-defined TF and the custom 1D LUT).
+	 */
+	struct drm_property *plane_shaper_tf_property;
+	/**
+	 * @plane_lut3d_property: Plane property for color transformation using
+	 * a 3D LUT (pre-blending), a three-dimensional array where each
+	 * element is an RGB triplet. Each dimension has the size of
+	 * lut3d_size. The array contains samples from the approximated
+	 * function. On AMD, values between samples are estimated by
+	 * tetrahedral interpolation. The array is accessed with three indices,
+	 * one for each input dimension (color channel), blue being the
+	 * outermost dimension, red the innermost.
+	 */
+	struct drm_property *plane_lut3d_property;
+	/**
+	 * @plane_degamma_lut_size_property: Plane property to define the max
+	 * size of 3D LUT as supported by the driver (read-only). The max size
+	 * is the max size of one dimension and, therefore, the max number of
+	 * entries for 3D LUT array is the 3D LUT size cubed;
+	 */
+	struct drm_property *plane_lut3d_size_property;
+	/**
+	 * @plane_blend_lut_property: Plane property for output gamma before
+	 * blending. Userspace set a blend LUT to convert colors after 3D LUT
+	 * conversion. It works as a post-3DLUT 1D LUT. With shaper LUT, they
+	 * are sandwiching 3D LUT with two 1D LUT. If plane_blend_tf_property
+	 * != Identity TF, AMD color module will combine the user LUT values
+	 * with pre-defined TF into the LUT parameters to be programmed.
+	 */
+	struct drm_property *plane_blend_lut_property;
+	/**
+	 * @plane_blend_lut_size_property: Plane property to define the max
+	 * size of blend LUT as supported by the driver (read-only).
+	 */
+	struct drm_property *plane_blend_lut_size_property;
+	/**
+	 * @plane_blend_tf_property: Plane property to set a predefined
+	 * transfer function for pre-blending blend/out_gamma (after applying
+	 * 3D LUT) with or without LUT. There is no blend ROM, but we can use
+	 * AMD color modules to program LUT parameters from predefined TF (or
+	 * from a combination of pre-defined TF and the custom 1D LUT).
+	 */
+	struct drm_property *plane_blend_tf_property;
+	/* @regamma_tf_property: Transfer function for CRTC regamma
+	 * (post-blending). Possible values are defined by `enum
+	 * amdgpu_transfer_function`. There is no regamma ROM, but we can use
+	 * AMD color modules to program LUT parameters from predefined TF (or
+	 * from a combination of pre-defined TF and the custom 1D LUT).
+	 */
+	struct drm_property *regamma_tf_property;
 };
 
 #define AMDGPU_MAX_BL_LEVEL 0xFF
 
-#if defined(CONFIG_BACKLIGHT_CLASS_DEVICE) || defined(CONFIG_BACKLIGHT_CLASS_DEVICE_MODULE)
-
 struct amdgpu_backlight_privdata {
 	struct amdgpu_encoder *encoder;
 	uint8_t negative;
 };
 
-#endif
-
 struct amdgpu_atom_ss {
 	uint16_t percentage;
 	uint16_t percentage_divider;
@@ -369,7 +460,6 @@ struct amdgpu_atom_ss {
 struct amdgpu_crtc {
 	struct drm_crtc base;
 	int crtc_id;
-	u16 lut_r[256], lut_g[256], lut_b[256];
 	bool enabled;
 	bool can_tile;
 	uint32_t crtc_offset;
@@ -394,6 +484,8 @@ struct amdgpu_crtc {
 	struct amdgpu_flip_work *pflip_works;
 	enum amdgpu_flip_status pflip_status;
 	int deferred_flip_completion;
+	/* parameters access from DM IRQ handler */
+	struct dm_irq_params dm_irq_params;
 	/* pll sharing */
 	struct amdgpu_atom_ss ss;
 	bool ss_enabled;
@@ -406,13 +498,18 @@ struct amdgpu_crtc {
 	struct drm_connector *connector;
 	/* for dpm */
 	u32 line_time;
-	u32 wm_low;
-	u32 wm_high;
 	u32 lb_vblank_lead_lines;
 	struct drm_display_mode hw_mode;
 	/* for virtual dce */
 	struct hrtimer vblank_timer;
 	enum amdgpu_interrupt_state vsync_timer_enabled;
+
+	int otg_inst;
+	struct drm_pending_vblank_event *event;
+
+	bool wb_pending;
+	bool wb_enabled;
+	struct drm_writeback_connector *wb_conn;
 };
 
 struct amdgpu_encoder_atom_dig {
@@ -455,6 +552,7 @@ struct amdgpu_encoder {
 struct amdgpu_connector_atom_dig {
 	/* displayport */
 	u8 dpcd[DP_RECEIVER_CAP_SIZE];
+	u8 downstream_ports[DP_MAX_DOWNSTREAM_PORTS];
 	u8 dp_sink_type;
 	int dp_clock;
 	int dp_lane_count;
@@ -502,6 +600,20 @@ enum amdgpu_connector_dither {
 	AMDGPU_FMT_DITHER_ENABLE = 1,
 };
 
+struct amdgpu_dm_dp_aux {
+	struct drm_dp_aux aux;
+	struct ddc_service *ddc_service;
+};
+
+struct amdgpu_i2c_adapter {
+	struct i2c_adapter base;
+
+	struct ddc_service *ddc_service;
+	bool oem;
+};
+
+#define TO_DM_AUX(x) container_of((x), struct amdgpu_dm_dp_aux, aux)
+
 struct amdgpu_connector {
 	struct drm_connector base;
 	uint32_t connector_id;
@@ -516,6 +628,7 @@ struct amdgpu_connector {
 	void *con_priv;
 	bool dac_load_detect;
 	bool detected_by_load; /* if the connection status was determined by load */
+	bool detected_hpd_without_ddc; /* if an HPD signal was detected on DVI, but ddc probing failed */
 	uint16_t connector_object_id;
 	struct amdgpu_hpd hpd;
 	struct amdgpu_router router;
@@ -525,15 +638,25 @@ struct amdgpu_connector {
 	unsigned pixelclock_for_modeset;
 };
 
-struct amdgpu_framebuffer {
-	struct drm_framebuffer base;
-	struct drm_gem_object *obj;
+/* TODO: start to use this struct and remove same field from base one */
+struct amdgpu_mst_connector {
+	struct amdgpu_connector base;
+
+	struct drm_dp_mst_topology_mgr mst_mgr;
+	struct amdgpu_dm_dp_aux dm_dp_aux;
+	struct drm_dp_mst_port *mst_output_port;
+	struct amdgpu_connector *mst_root;
+	bool is_mst_connector;
+	struct amdgpu_encoder *mst_encoder;
 };
 
 #define ENCODER_MODE_IS_DP(em) (((em) == ATOM_ENCODER_MODE_DP) || \
 				((em) == ATOM_ENCODER_MODE_DP_MST))
 
-/* Driver internal use only flags of amdgpu_get_crtc_scanoutpos() */
+/* Driver internal use only flags of amdgpu_display_get_crtc_scanoutpos() */
+#define DRM_SCANOUTPOS_VALID        (1 << 0)
+#define DRM_SCANOUTPOS_IN_VBLANK    (1 << 1)
+#define DRM_SCANOUTPOS_ACCURATE     (1 << 2)
 #define USE_REAL_VBLANKSTART		(1 << 30)
 #define GET_DISTANCE_TO_VBLANKSTART	(1 << 31)
 
@@ -549,54 +672,42 @@ bool amdgpu_dig_monitor_is_duallink(struct drm_encoder *encoder,
 u16 amdgpu_encoder_get_dp_bridge_encoder_id(struct drm_encoder *encoder);
 struct drm_encoder *amdgpu_get_external_encoder(struct drm_encoder *encoder);
 
-bool amdgpu_ddc_probe(struct amdgpu_connector *amdgpu_connector, bool use_aux);
+bool amdgpu_display_ddc_probe(struct amdgpu_connector *amdgpu_connector,
+			      bool use_aux);
 
 void amdgpu_encoder_set_active_device(struct drm_encoder *encoder);
 
-int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe,
-			       unsigned int flags, int *vpos, int *hpos,
-			       ktime_t *stime, ktime_t *etime,
-			       const struct drm_display_mode *mode);
-
-int amdgpu_framebuffer_init(struct drm_device *dev,
-			     struct amdgpu_framebuffer *rfb,
-			     const struct drm_mode_fb_cmd2 *mode_cmd,
-			     struct drm_gem_object *obj);
+int amdgpu_display_get_crtc_scanoutpos(struct drm_device *dev,
+			unsigned int pipe, unsigned int flags, int *vpos,
+			int *hpos, ktime_t *stime, ktime_t *etime,
+			const struct drm_display_mode *mode);
 
 int amdgpufb_remove(struct drm_device *dev, struct drm_framebuffer *fb);
 
 void amdgpu_enc_destroy(struct drm_encoder *encoder);
 void amdgpu_copy_fb(struct drm_device *dev, struct drm_gem_object *dst_obj);
-bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
-					const struct drm_display_mode *mode,
-					struct drm_display_mode *adjusted_mode);
+bool amdgpu_display_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
+				const struct drm_display_mode *mode,
+				struct drm_display_mode *adjusted_mode);
 void amdgpu_panel_mode_fixup(struct drm_encoder *encoder,
 			     struct drm_display_mode *adjusted_mode);
-int amdgpu_crtc_idx_to_irq_type(struct amdgpu_device *adev, int crtc);
-
-/* fbdev layer */
-int amdgpu_fbdev_init(struct amdgpu_device *adev);
-void amdgpu_fbdev_fini(struct amdgpu_device *adev);
-void amdgpu_fbdev_set_suspend(struct amdgpu_device *adev, int state);
-int amdgpu_fbdev_total_size(struct amdgpu_device *adev);
-bool amdgpu_fbdev_robj_is_fb(struct amdgpu_device *adev, struct amdgpu_bo *robj);
-void amdgpu_fbdev_restore_mode(struct amdgpu_device *adev);
-
-void amdgpu_fb_output_poll_changed(struct amdgpu_device *adev);
-
+int amdgpu_display_crtc_idx_to_irq_type(struct amdgpu_device *adev, int crtc);
 
-int amdgpu_align_pitch(struct amdgpu_device *adev, int width, int bpp, bool tiled);
+bool amdgpu_crtc_get_scanout_position(struct drm_crtc *crtc,
+			bool in_vblank_irq, int *vpos,
+			int *hpos, ktime_t *stime, ktime_t *etime,
+			const struct drm_display_mode *mode);
 
 /* amdgpu_display.c */
-void amdgpu_print_display_setup(struct drm_device *dev);
-int amdgpu_modeset_create_props(struct amdgpu_device *adev);
-int amdgpu_crtc_set_config(struct drm_mode_set *set,
-			   struct drm_modeset_acquire_ctx *ctx);
-int amdgpu_crtc_page_flip_target(struct drm_crtc *crtc,
-				 struct drm_framebuffer *fb,
-				 struct drm_pending_vblank_event *event,
-				 uint32_t page_flip_flags, uint32_t target,
-				 struct drm_modeset_acquire_ctx *ctx);
+void amdgpu_display_print_display_setup(struct drm_device *dev);
+int amdgpu_display_modeset_create_props(struct amdgpu_device *adev);
+int amdgpu_display_crtc_set_config(struct drm_mode_set *set,
+				   struct drm_modeset_acquire_ctx *ctx);
+int amdgpu_display_crtc_page_flip_target(struct drm_crtc *crtc,
+				struct drm_framebuffer *fb,
+				struct drm_pending_vblank_event *event,
+				uint32_t page_flip_flags, uint32_t target,
+				struct drm_modeset_acquire_ctx *ctx);
 extern const struct drm_mode_config_funcs amdgpu_mode_funcs;
 
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.c
new file mode 100644
index 000000000000..a974265837f0
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.c
@@ -0,0 +1,86 @@
+/*
+ * Copyright (C) 2019  Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include "amdgpu.h"
+#include "amdgpu_ras.h"
+
+int amdgpu_nbio_ras_sw_init(struct amdgpu_device *adev)
+{
+	int err;
+	struct amdgpu_nbio_ras *ras;
+
+	if (!adev->nbio.ras)
+		return 0;
+
+	ras = adev->nbio.ras;
+	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
+	if (err) {
+		dev_err(adev->dev, "Failed to register pcie_bif ras block!\n");
+		return err;
+	}
+
+	strcpy(ras->ras_block.ras_comm.name, "pcie_bif");
+	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__PCIE_BIF;
+	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+	adev->nbio.ras_if = &ras->ras_block.ras_comm;
+
+	return 0;
+}
+
+u64 amdgpu_nbio_get_pcie_replay_count(struct amdgpu_device *adev)
+{
+	if (adev->nbio.funcs && adev->nbio.funcs->get_pcie_replay_count)
+		return adev->nbio.funcs->get_pcie_replay_count(adev);
+
+	return 0;
+}
+
+bool amdgpu_nbio_is_replay_cnt_supported(struct amdgpu_device *adev)
+{
+	if (amdgpu_sriov_vf(adev) || !adev->asic_funcs ||
+	    !adev->asic_funcs->get_pcie_replay_count ||
+	    (!adev->nbio.funcs || !adev->nbio.funcs->get_pcie_replay_count))
+		return false;
+
+	return true;
+}
+
+int amdgpu_nbio_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
+{
+	int r;
+	r = amdgpu_ras_block_late_init(adev, ras_block);
+	if (r)
+		return r;
+
+	if (amdgpu_ras_is_supported(adev, ras_block->block)) {
+		r = amdgpu_irq_get(adev, &adev->nbio.ras_controller_irq, 0);
+		if (r)
+			goto late_fini;
+		r = amdgpu_irq_get(adev, &adev->nbio.ras_err_event_athub_irq, 0);
+		if (r)
+			goto late_fini;
+	}
+
+	return 0;
+late_fini:
+	amdgpu_ras_block_late_fini(adev, ras_block);
+	return r;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.h
new file mode 100644
index 000000000000..b528de6a01f6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_nbio.h
@@ -0,0 +1,124 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#ifndef __AMDGPU_NBIO_H__
+#define __AMDGPU_NBIO_H__
+
+/*
+ * amdgpu nbio functions
+ */
+struct nbio_hdp_flush_reg {
+	u32 ref_and_mask_cp0;
+	u32 ref_and_mask_cp1;
+	u32 ref_and_mask_cp2;
+	u32 ref_and_mask_cp3;
+	u32 ref_and_mask_cp4;
+	u32 ref_and_mask_cp5;
+	u32 ref_and_mask_cp6;
+	u32 ref_and_mask_cp7;
+	u32 ref_and_mask_cp8;
+	u32 ref_and_mask_cp9;
+	u32 ref_and_mask_sdma0;
+	u32 ref_and_mask_sdma1;
+	u32 ref_and_mask_sdma2;
+	u32 ref_and_mask_sdma3;
+	u32 ref_and_mask_sdma4;
+	u32 ref_and_mask_sdma5;
+	u32 ref_and_mask_sdma6;
+	u32 ref_and_mask_sdma7;
+};
+
+struct amdgpu_nbio_ras {
+	struct amdgpu_ras_block_object ras_block;
+	void (*handle_ras_controller_intr_no_bifring)(struct amdgpu_device *adev);
+	void (*handle_ras_err_event_athub_intr_no_bifring)(struct amdgpu_device *adev);
+	int (*init_ras_controller_interrupt)(struct amdgpu_device *adev);
+	int (*init_ras_err_event_athub_interrupt)(struct amdgpu_device *adev);
+};
+
+struct amdgpu_nbio_funcs {
+	const struct nbio_hdp_flush_reg *hdp_flush_reg;
+	u32 (*get_hdp_flush_req_offset)(struct amdgpu_device *adev);
+	u32 (*get_hdp_flush_done_offset)(struct amdgpu_device *adev);
+	u32 (*get_pcie_index_offset)(struct amdgpu_device *adev);
+	u32 (*get_pcie_data_offset)(struct amdgpu_device *adev);
+	u32 (*get_pcie_index_hi_offset)(struct amdgpu_device *adev);
+	u32 (*get_pcie_port_index_offset)(struct amdgpu_device *adev);
+	u32 (*get_pcie_port_data_offset)(struct amdgpu_device *adev);
+	u32 (*get_rev_id)(struct amdgpu_device *adev);
+	void (*mc_access_enable)(struct amdgpu_device *adev, bool enable);
+	u32 (*get_memsize)(struct amdgpu_device *adev);
+	void (*sdma_doorbell_range)(struct amdgpu_device *adev, int instance,
+			bool use_doorbell, int doorbell_index, int doorbell_size);
+	void (*vpe_doorbell_range)(struct amdgpu_device *adev, int instance,
+			bool use_doorbell, int doorbell_index, int doorbell_size);
+	void (*vcn_doorbell_range)(struct amdgpu_device *adev, bool use_doorbell,
+				   int doorbell_index, int instance);
+	void (*gc_doorbell_init)(struct amdgpu_device *adev);
+	void (*enable_doorbell_aperture)(struct amdgpu_device *adev,
+					 bool enable);
+	void (*enable_doorbell_selfring_aperture)(struct amdgpu_device *adev,
+						  bool enable);
+	void (*ih_doorbell_range)(struct amdgpu_device *adev,
+				  bool use_doorbell, int doorbell_index);
+	void (*enable_doorbell_interrupt)(struct amdgpu_device *adev,
+					  bool enable);
+	void (*update_medium_grain_clock_gating)(struct amdgpu_device *adev,
+						 bool enable);
+	void (*update_medium_grain_light_sleep)(struct amdgpu_device *adev,
+						bool enable);
+	void (*get_clockgating_state)(struct amdgpu_device *adev,
+				      u64 *flags);
+	void (*ih_control)(struct amdgpu_device *adev);
+	void (*init_registers)(struct amdgpu_device *adev);
+	void (*remap_hdp_registers)(struct amdgpu_device *adev);
+	void (*enable_aspm)(struct amdgpu_device *adev,
+			    bool enable);
+	void (*program_aspm)(struct amdgpu_device *adev);
+	void (*apply_lc_spc_mode_wa)(struct amdgpu_device *adev);
+	void (*apply_l1_link_width_reconfig_wa)(struct amdgpu_device *adev);
+	void (*clear_doorbell_interrupt)(struct amdgpu_device *adev);
+	u32 (*get_rom_offset)(struct amdgpu_device *adev);
+	int (*get_compute_partition_mode)(struct amdgpu_device *adev);
+	u32 (*get_memory_partition_mode)(struct amdgpu_device *adev,
+					 u32 *supp_modes);
+	bool (*is_nps_switch_requested)(struct amdgpu_device *adev);
+	u64 (*get_pcie_replay_count)(struct amdgpu_device *adev);
+	void (*set_reg_remap)(struct amdgpu_device *adev);
+};
+
+struct amdgpu_nbio {
+	const struct nbio_hdp_flush_reg *hdp_flush_reg;
+	struct amdgpu_irq_src ras_controller_irq;
+	struct amdgpu_irq_src ras_err_event_athub_irq;
+	struct ras_common_if *ras_if;
+	const struct amdgpu_nbio_funcs *funcs;
+	struct amdgpu_nbio_ras  *ras;
+};
+
+int amdgpu_nbio_ras_sw_init(struct amdgpu_device *adev);
+int amdgpu_nbio_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block);
+u64 amdgpu_nbio_get_pcie_replay_count(struct amdgpu_device *adev);
+
+bool amdgpu_nbio_is_replay_cnt_supported(struct amdgpu_device *adev);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 365883d7948d..e08f58de4b17 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -31,191 +31,294 @@
  */
 #include <linux/list.h>
 #include <linux/slab.h>
-#include <drm/drmP.h>
+#include <linux/dma-buf.h>
+#include <linux/export.h>
+
+#include <drm/drm_drv.h>
 #include <drm/amdgpu_drm.h>
 #include <drm/drm_cache.h>
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_vram_mgr.h"
+#include "amdgpu_vm.h"
+#include "amdgpu_dma_buf.h"
 
+/**
+ * DOC: amdgpu_object
+ *
+ * This defines the interfaces to operate on an &amdgpu_bo buffer object which
+ * represents memory used by driver (VRAM, system memory, etc.). The driver
+ * provides DRM/GEM APIs to userspace. DRM/GEM APIs then use these interfaces
+ * to create/destroy/set buffer object which are then managed by the kernel TTM
+ * memory manager.
+ * The interfaces are also used internally by kernel clients, including gfx,
+ * uvd, etc. for kernel managed allocations used by the GPU.
+ *
+ */
 
-
-static u64 amdgpu_get_vis_part_size(struct amdgpu_device *adev,
-						struct ttm_mem_reg *mem)
-{
-	if (mem->start << PAGE_SHIFT >= adev->mc.visible_vram_size)
-		return 0;
-
-	return ((mem->start << PAGE_SHIFT) + mem->size) >
-		adev->mc.visible_vram_size ?
-		adev->mc.visible_vram_size - (mem->start << PAGE_SHIFT) :
-		mem->size;
-}
-
-static void amdgpu_update_memory_usage(struct amdgpu_device *adev,
-		       struct ttm_mem_reg *old_mem,
-		       struct ttm_mem_reg *new_mem)
+static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
 {
-	u64 vis_size;
-	if (!adev)
-		return;
+	struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
 
-	if (new_mem) {
-		switch (new_mem->mem_type) {
-		case TTM_PL_TT:
-			atomic64_add(new_mem->size, &adev->gtt_usage);
-			break;
-		case TTM_PL_VRAM:
-			atomic64_add(new_mem->size, &adev->vram_usage);
-			vis_size = amdgpu_get_vis_part_size(adev, new_mem);
-			atomic64_add(vis_size, &adev->vram_vis_usage);
-			break;
-		}
-	}
+	amdgpu_bo_kunmap(bo);
 
-	if (old_mem) {
-		switch (old_mem->mem_type) {
-		case TTM_PL_TT:
-			atomic64_sub(old_mem->size, &adev->gtt_usage);
-			break;
-		case TTM_PL_VRAM:
-			atomic64_sub(old_mem->size, &adev->vram_usage);
-			vis_size = amdgpu_get_vis_part_size(adev, old_mem);
-			atomic64_sub(vis_size, &adev->vram_vis_usage);
-			break;
-		}
-	}
+	if (drm_gem_is_imported(&bo->tbo.base))
+		drm_prime_gem_destroy(&bo->tbo.base, bo->tbo.sg);
+	drm_gem_object_release(&bo->tbo.base);
+	amdgpu_bo_unref(&bo->parent);
+	kvfree(bo);
 }
 
-static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo)
+static void amdgpu_bo_user_destroy(struct ttm_buffer_object *tbo)
 {
-	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-	struct amdgpu_bo *bo;
+	struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+	struct amdgpu_bo_user *ubo;
 
-	bo = container_of(tbo, struct amdgpu_bo, tbo);
-
-	amdgpu_update_memory_usage(adev, &bo->tbo.mem, NULL);
-
-	drm_gem_object_release(&bo->gem_base);
-	amdgpu_bo_unref(&bo->parent);
-	if (!list_empty(&bo->shadow_list)) {
-		mutex_lock(&adev->shadow_list_lock);
-		list_del_init(&bo->shadow_list);
-		mutex_unlock(&adev->shadow_list_lock);
-	}
-	kfree(bo->metadata);
-	kfree(bo);
+	ubo = to_amdgpu_bo_user(bo);
+	kfree(ubo->metadata);
+	amdgpu_bo_destroy(tbo);
 }
 
-bool amdgpu_ttm_bo_is_amdgpu_bo(struct ttm_buffer_object *bo)
+/**
+ * amdgpu_bo_is_amdgpu_bo - check if the buffer object is an &amdgpu_bo
+ * @bo: buffer object to be checked
+ *
+ * Uses destroy function associated with the object to determine if this is
+ * an &amdgpu_bo.
+ *
+ * Returns:
+ * true if the object belongs to &amdgpu_bo, false if not.
+ */
+bool amdgpu_bo_is_amdgpu_bo(struct ttm_buffer_object *bo)
 {
-	if (bo->destroy == &amdgpu_ttm_bo_destroy)
+	if (bo->destroy == &amdgpu_bo_destroy ||
+	    bo->destroy == &amdgpu_bo_user_destroy)
 		return true;
+
 	return false;
 }
 
-static void amdgpu_ttm_placement_init(struct amdgpu_device *adev,
-				      struct ttm_placement *placement,
-				      struct ttm_place *places,
-				      u32 domain, u64 flags)
+/**
+ * amdgpu_bo_placement_from_domain - set buffer's placement
+ * @abo: &amdgpu_bo buffer object whose placement is to be set
+ * @domain: requested domain
+ *
+ * Sets buffer's placement according to requested domain and the buffer's
+ * flags.
+ */
+void amdgpu_bo_placement_from_domain(struct amdgpu_bo *abo, u32 domain)
 {
+	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
+	struct ttm_placement *placement = &abo->placement;
+	struct ttm_place *places = abo->placements;
+	u64 flags = abo->flags;
 	u32 c = 0;
 
 	if (domain & AMDGPU_GEM_DOMAIN_VRAM) {
-		unsigned visible_pfn = adev->mc.visible_vram_size >> PAGE_SHIFT;
-
-		places[c].fpfn = 0;
-		places[c].lpfn = 0;
-		places[c].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
-			TTM_PL_FLAG_VRAM;
+		unsigned int visible_pfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
+		int8_t mem_id = KFD_XCP_MEM_ID(adev, abo->xcp_id);
+
+		if (adev->gmc.mem_partitions && mem_id >= 0) {
+			places[c].fpfn = adev->gmc.mem_partitions[mem_id].range.fpfn;
+			/*
+			 * memory partition range lpfn is inclusive start + size - 1
+			 * TTM place lpfn is exclusive start + size
+			 */
+			places[c].lpfn = adev->gmc.mem_partitions[mem_id].range.lpfn + 1;
+		} else {
+			places[c].fpfn = 0;
+			places[c].lpfn = 0;
+		}
+		places[c].mem_type = TTM_PL_VRAM;
+		places[c].flags = 0;
 
 		if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
-			places[c].lpfn = visible_pfn;
+			places[c].lpfn = min_not_zero(places[c].lpfn, visible_pfn);
 		else
 			places[c].flags |= TTM_PL_FLAG_TOPDOWN;
 
-		if (flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
+		if (abo->tbo.type == ttm_bo_type_kernel &&
+		    flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
 			places[c].flags |= TTM_PL_FLAG_CONTIGUOUS;
+
+		c++;
+	}
+
+	if (domain & AMDGPU_GEM_DOMAIN_DOORBELL) {
+		places[c].fpfn = 0;
+		places[c].lpfn = 0;
+		places[c].mem_type = AMDGPU_PL_DOORBELL;
+		places[c].flags = 0;
+		c++;
+	}
+
+	if (domain & AMDGPU_GEM_DOMAIN_MMIO_REMAP) {
+		places[c].fpfn = 0;
+		places[c].lpfn = 0;
+		places[c].mem_type = AMDGPU_PL_MMIO_REMAP;
+		places[c].flags = 0;
 		c++;
 	}
 
 	if (domain & AMDGPU_GEM_DOMAIN_GTT) {
 		places[c].fpfn = 0;
 		places[c].lpfn = 0;
-		places[c].flags = TTM_PL_FLAG_TT;
-		if (flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
-			places[c].flags |= TTM_PL_FLAG_WC |
-				TTM_PL_FLAG_UNCACHED;
-		else
-			places[c].flags |= TTM_PL_FLAG_CACHED;
+		places[c].mem_type =
+			abo->flags & AMDGPU_GEM_CREATE_PREEMPTIBLE ?
+			AMDGPU_PL_PREEMPT : TTM_PL_TT;
+		places[c].flags = 0;
+		/*
+		 * When GTT is just an alternative to VRAM make sure that we
+		 * only use it as fallback and still try to fill up VRAM first.
+		 */
+		if (abo->tbo.resource && !(adev->flags & AMD_IS_APU) &&
+		    domain & abo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM)
+			places[c].flags |= TTM_PL_FLAG_FALLBACK;
 		c++;
 	}
 
 	if (domain & AMDGPU_GEM_DOMAIN_CPU) {
 		places[c].fpfn = 0;
 		places[c].lpfn = 0;
-		places[c].flags = TTM_PL_FLAG_SYSTEM;
-		if (flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
-			places[c].flags |= TTM_PL_FLAG_WC |
-				TTM_PL_FLAG_UNCACHED;
-		else
-			places[c].flags |= TTM_PL_FLAG_CACHED;
+		places[c].mem_type = TTM_PL_SYSTEM;
+		places[c].flags = 0;
 		c++;
 	}
 
 	if (domain & AMDGPU_GEM_DOMAIN_GDS) {
 		places[c].fpfn = 0;
 		places[c].lpfn = 0;
-		places[c].flags = TTM_PL_FLAG_UNCACHED | AMDGPU_PL_FLAG_GDS;
+		places[c].mem_type = AMDGPU_PL_GDS;
+		places[c].flags = 0;
 		c++;
 	}
 
 	if (domain & AMDGPU_GEM_DOMAIN_GWS) {
 		places[c].fpfn = 0;
 		places[c].lpfn = 0;
-		places[c].flags = TTM_PL_FLAG_UNCACHED | AMDGPU_PL_FLAG_GWS;
+		places[c].mem_type = AMDGPU_PL_GWS;
+		places[c].flags = 0;
 		c++;
 	}
 
 	if (domain & AMDGPU_GEM_DOMAIN_OA) {
 		places[c].fpfn = 0;
 		places[c].lpfn = 0;
-		places[c].flags = TTM_PL_FLAG_UNCACHED | AMDGPU_PL_FLAG_OA;
+		places[c].mem_type = AMDGPU_PL_OA;
+		places[c].flags = 0;
 		c++;
 	}
 
 	if (!c) {
 		places[c].fpfn = 0;
 		places[c].lpfn = 0;
-		places[c].flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM;
+		places[c].mem_type = TTM_PL_SYSTEM;
+		places[c].flags = 0;
 		c++;
 	}
 
+	BUG_ON(c > AMDGPU_BO_MAX_PLACEMENTS);
+
 	placement->num_placement = c;
 	placement->placement = places;
-
-	placement->num_busy_placement = c;
-	placement->busy_placement = places;
 }
 
-void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *abo, u32 domain)
+/**
+ * amdgpu_bo_create_reserved - create reserved BO for kernel use
+ *
+ * @adev: amdgpu device object
+ * @size: size for the new BO
+ * @align: alignment for the new BO
+ * @domain: where to place it
+ * @bo_ptr: used to initialize BOs in structures
+ * @gpu_addr: GPU addr of the pinned BO
+ * @cpu_addr: optional CPU address mapping
+ *
+ * Allocates and pins a BO for kernel internal use, and returns it still
+ * reserved.
+ *
+ * Note: For bo_ptr new BO is only created if bo_ptr points to NULL.
+ *
+ * Returns:
+ * 0 on success, negative error code otherwise.
+ */
+int amdgpu_bo_create_reserved(struct amdgpu_device *adev,
+			      unsigned long size, int align,
+			      u32 domain, struct amdgpu_bo **bo_ptr,
+			      u64 *gpu_addr, void **cpu_addr)
 {
-	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
+	struct amdgpu_bo_param bp;
+	bool free = false;
+	int r;
 
-	amdgpu_ttm_placement_init(adev, &abo->placement, abo->placements,
-				  domain, abo->flags);
-}
+	if (!size) {
+		amdgpu_bo_unref(bo_ptr);
+		return 0;
+	}
 
-static void amdgpu_fill_placement_to_bo(struct amdgpu_bo *bo,
-					struct ttm_placement *placement)
-{
-	BUG_ON(placement->num_placement > (AMDGPU_GEM_DOMAIN_MAX + 1));
-
-	memcpy(bo->placements, placement->placement,
-	       placement->num_placement * sizeof(struct ttm_place));
-	bo->placement.num_placement = placement->num_placement;
-	bo->placement.num_busy_placement = placement->num_busy_placement;
-	bo->placement.placement = bo->placements;
-	bo->placement.busy_placement = bo->placements;
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = align;
+	bp.domain = domain;
+	bp.flags = cpu_addr ? AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED
+		: AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
+	bp.flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+	bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+
+	if (!*bo_ptr) {
+		r = amdgpu_bo_create(adev, &bp, bo_ptr);
+		if (r) {
+			dev_err(adev->dev, "(%d) failed to allocate kernel bo\n",
+				r);
+			return r;
+		}
+		free = true;
+	}
+
+	r = amdgpu_bo_reserve(*bo_ptr, false);
+	if (r) {
+		dev_err(adev->dev, "(%d) failed to reserve kernel bo\n", r);
+		goto error_free;
+	}
+
+	r = amdgpu_bo_pin(*bo_ptr, domain);
+	if (r) {
+		dev_err(adev->dev, "(%d) kernel bo pin failed\n", r);
+		goto error_unreserve;
+	}
+
+	r = amdgpu_ttm_alloc_gart(&(*bo_ptr)->tbo);
+	if (r) {
+		dev_err(adev->dev, "%p bind failed\n", *bo_ptr);
+		goto error_unpin;
+	}
+
+	if (gpu_addr)
+		*gpu_addr = amdgpu_bo_gpu_offset(*bo_ptr);
+
+	if (cpu_addr) {
+		r = amdgpu_bo_kmap(*bo_ptr, cpu_addr);
+		if (r) {
+			dev_err(adev->dev, "(%d) kernel bo map failed\n", r);
+			goto error_unpin;
+		}
+	}
+
+	return 0;
+
+error_unpin:
+	amdgpu_bo_unpin(*bo_ptr);
+error_unreserve:
+	amdgpu_bo_unreserve(*bo_ptr);
+
+error_free:
+	if (free)
+		amdgpu_bo_unref(bo_ptr);
+
+	return r;
 }
 
 /**
@@ -225,13 +328,19 @@ static void amdgpu_fill_placement_to_bo(struct amdgpu_bo *bo,
  * @size: size for the new BO
  * @align: alignment for the new BO
  * @domain: where to place it
- * @bo_ptr: resulting BO
+ * @bo_ptr:  used to initialize BOs in structures
  * @gpu_addr: GPU addr of the pinned BO
  * @cpu_addr: optional CPU address mapping
  *
  * Allocates and pins a BO for kernel internal use.
  *
- * Returns 0 on success, negative error code otherwise.
+ * This function is exported to allow the V4L2 isp device
+ * external to drm device to create and access the kernel BO.
+ *
+ * Note: For bo_ptr new BO is only created if bo_ptr points to NULL.
+ *
+ * Returns:
+ * 0 on success, negative error code otherwise.
  */
 int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
 			    unsigned long size, int align,
@@ -240,45 +349,150 @@ int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
 {
 	int r;
 
-	r = amdgpu_bo_create(adev, size, align, true, domain,
-			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-			     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
-			     NULL, NULL, bo_ptr);
-	if (r) {
-		dev_err(adev->dev, "(%d) failed to allocate kernel bo\n", r);
+	r = amdgpu_bo_create_reserved(adev, size, align, domain, bo_ptr,
+				      gpu_addr, cpu_addr);
+
+	if (r)
 		return r;
+
+	if (*bo_ptr)
+		amdgpu_bo_unreserve(*bo_ptr);
+
+	return 0;
+}
+
+/**
+ * amdgpu_bo_create_isp_user - create user BO for isp
+ *
+ * @adev: amdgpu device object
+ * @dma_buf: DMABUF handle for isp buffer
+ * @domain: where to place it
+ * @bo:  used to initialize BOs in structures
+ * @gpu_addr: GPU addr of the pinned BO
+ *
+ * Imports isp DMABUF to allocate and pin a user BO for isp internal use. It does
+ * GART alloc to generate gpu_addr for BO to make it accessible through the
+ * GART aperture for ISP HW.
+ *
+ * This function is exported to allow the V4L2 isp device external to drm device
+ * to create and access the isp user BO.
+ *
+ * Returns:
+ * 0 on success, negative error code otherwise.
+ */
+int amdgpu_bo_create_isp_user(struct amdgpu_device *adev,
+			   struct dma_buf *dma_buf, u32 domain, struct amdgpu_bo **bo,
+			   u64 *gpu_addr)
+
+{
+	struct drm_gem_object *gem_obj;
+	int r;
+
+	gem_obj = amdgpu_gem_prime_import(&adev->ddev, dma_buf);
+	*bo = gem_to_amdgpu_bo(gem_obj);
+	if (!(*bo)) {
+		dev_err(adev->dev, "failed to get valid isp user bo\n");
+		return -EINVAL;
 	}
 
-	r = amdgpu_bo_reserve(*bo_ptr, false);
+	r = amdgpu_bo_reserve(*bo, false);
 	if (r) {
-		dev_err(adev->dev, "(%d) failed to reserve kernel bo\n", r);
-		goto error_free;
+		dev_err(adev->dev, "(%d) failed to reserve isp user bo\n", r);
+		return r;
 	}
 
-	r = amdgpu_bo_pin(*bo_ptr, domain, gpu_addr);
+	r = amdgpu_bo_pin(*bo, domain);
 	if (r) {
-		dev_err(adev->dev, "(%d) kernel bo pin failed\n", r);
+		dev_err(adev->dev, "(%d) isp user bo pin failed\n", r);
 		goto error_unreserve;
 	}
 
-	if (cpu_addr) {
-		r = amdgpu_bo_kmap(*bo_ptr, cpu_addr);
-		if (r) {
-			dev_err(adev->dev, "(%d) kernel bo map failed\n", r);
-			goto error_unreserve;
-		}
+	r = amdgpu_ttm_alloc_gart(&(*bo)->tbo);
+	if (r) {
+		dev_err(adev->dev, "%p bind failed\n", *bo);
+		goto error_unpin;
 	}
 
-	amdgpu_bo_unreserve(*bo_ptr);
+	if (!WARN_ON(!gpu_addr))
+		*gpu_addr = amdgpu_bo_gpu_offset(*bo);
+
+	amdgpu_bo_unreserve(*bo);
 
 	return 0;
 
+error_unpin:
+	amdgpu_bo_unpin(*bo);
 error_unreserve:
+	amdgpu_bo_unreserve(*bo);
+	amdgpu_bo_unref(bo);
+
+	return r;
+}
+
+/**
+ * amdgpu_bo_create_kernel_at - create BO for kernel use at specific location
+ *
+ * @adev: amdgpu device object
+ * @offset: offset of the BO
+ * @size: size of the BO
+ * @bo_ptr:  used to initialize BOs in structures
+ * @cpu_addr: optional CPU address mapping
+ *
+ * Creates a kernel BO at a specific offset in VRAM.
+ *
+ * Returns:
+ * 0 on success, negative error code otherwise.
+ */
+int amdgpu_bo_create_kernel_at(struct amdgpu_device *adev,
+			       uint64_t offset, uint64_t size,
+			       struct amdgpu_bo **bo_ptr, void **cpu_addr)
+{
+	struct ttm_operation_ctx ctx = { false, false };
+	unsigned int i;
+	int r;
+
+	offset &= PAGE_MASK;
+	size = ALIGN(size, PAGE_SIZE);
+
+	r = amdgpu_bo_create_reserved(adev, size, PAGE_SIZE,
+				      AMDGPU_GEM_DOMAIN_VRAM, bo_ptr, NULL,
+				      cpu_addr);
+	if (r)
+		return r;
+
+	if ((*bo_ptr) == NULL)
+		return 0;
+
+	/*
+	 * Remove the original mem node and create a new one at the request
+	 * position.
+	 */
+	if (cpu_addr)
+		amdgpu_bo_kunmap(*bo_ptr);
+
+	ttm_resource_free(&(*bo_ptr)->tbo, &(*bo_ptr)->tbo.resource);
+
+	for (i = 0; i < (*bo_ptr)->placement.num_placement; ++i) {
+		(*bo_ptr)->placements[i].fpfn = offset >> PAGE_SHIFT;
+		(*bo_ptr)->placements[i].lpfn = (offset + size) >> PAGE_SHIFT;
+	}
+	r = ttm_bo_mem_space(&(*bo_ptr)->tbo, &(*bo_ptr)->placement,
+			     &(*bo_ptr)->tbo.resource, &ctx);
+	if (r)
+		goto error;
+
+	if (cpu_addr) {
+		r = amdgpu_bo_kmap(*bo_ptr, cpu_addr);
+		if (r)
+			goto error;
+	}
+
 	amdgpu_bo_unreserve(*bo_ptr);
+	return 0;
 
-error_free:
+error:
+	amdgpu_bo_unreserve(*bo_ptr);
 	amdgpu_bo_unref(bo_ptr);
-
 	return r;
 }
 
@@ -286,8 +500,13 @@ error_free:
  * amdgpu_bo_free_kernel - free BO for kernel use
  *
  * @bo: amdgpu BO to free
+ * @gpu_addr: pointer to where the BO's GPU memory space address was stored
+ * @cpu_addr: pointer to where the BO's CPU memory space address was stored
  *
  * unmaps and unpin a BO for kernel internal use.
+ *
+ * This function is exported to allow the V4L2 isp device
+ * external to drm device to free the kernel BO.
  */
 void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
 			   void **cpu_addr)
@@ -295,6 +514,8 @@ void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
 	if (*bo == NULL)
 		return;
 
+	WARN_ON(amdgpu_ttm_adev((*bo)->tbo.bdev)->in_suspend);
+
 	if (likely(amdgpu_bo_reserve(*bo, true) == 0)) {
 		if (cpu_addr)
 			amdgpu_bo_kunmap(*bo);
@@ -311,63 +532,67 @@ void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
 		*cpu_addr = NULL;
 }
 
-int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
-				unsigned long size, int byte_align,
-				bool kernel, u32 domain, u64 flags,
-				struct sg_table *sg,
-				struct ttm_placement *placement,
-				struct reservation_object *resv,
-				struct amdgpu_bo **bo_ptr)
+/**
+ * amdgpu_bo_free_isp_user - free BO for isp use
+ *
+ * @bo: amdgpu isp user BO to free
+ *
+ * unpin and unref BO for isp internal use.
+ *
+ * This function is exported to allow the V4L2 isp device
+ * external to drm device to free the isp user BO.
+ */
+void amdgpu_bo_free_isp_user(struct amdgpu_bo *bo)
 {
-	struct amdgpu_bo *bo;
-	enum ttm_bo_type type;
-	unsigned long page_align;
-	u64 initial_bytes_moved;
-	size_t acc_size;
-	int r;
-
-	page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
-	size = ALIGN(size, PAGE_SIZE);
+	if (bo == NULL)
+		return;
 
-	if (kernel) {
-		type = ttm_bo_type_kernel;
-	} else if (sg) {
-		type = ttm_bo_type_sg;
-	} else {
-		type = ttm_bo_type_device;
+	if (amdgpu_bo_reserve(bo, true) == 0) {
+		amdgpu_bo_unpin(bo);
+		amdgpu_bo_unreserve(bo);
 	}
-	*bo_ptr = NULL;
+	amdgpu_bo_unref(&bo);
+}
 
-	acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size,
-				       sizeof(struct amdgpu_bo));
+/* Validate bo size is bit bigger than the request domain */
+static bool amdgpu_bo_validate_size(struct amdgpu_device *adev,
+					  unsigned long size, u32 domain)
+{
+	struct ttm_resource_manager *man = NULL;
 
-	bo = kzalloc(sizeof(struct amdgpu_bo), GFP_KERNEL);
-	if (bo == NULL)
-		return -ENOMEM;
-	r = drm_gem_object_init(adev->ddev, &bo->gem_base, size);
-	if (unlikely(r)) {
-		kfree(bo);
-		return r;
+	/*
+	 * If GTT is part of requested domains the check must succeed to
+	 * allow fall back to GTT.
+	 */
+	if (domain & AMDGPU_GEM_DOMAIN_GTT)
+		man = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT);
+	else if (domain & AMDGPU_GEM_DOMAIN_VRAM)
+		man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
+	else
+		return true;
+
+	if (!man) {
+		if (domain & AMDGPU_GEM_DOMAIN_GTT)
+			WARN_ON_ONCE("GTT domain requested but GTT mem manager uninitialized");
+		return false;
 	}
-	INIT_LIST_HEAD(&bo->shadow_list);
-	INIT_LIST_HEAD(&bo->va);
-	bo->prefered_domains = domain & (AMDGPU_GEM_DOMAIN_VRAM |
-					 AMDGPU_GEM_DOMAIN_GTT |
-					 AMDGPU_GEM_DOMAIN_CPU |
-					 AMDGPU_GEM_DOMAIN_GDS |
-					 AMDGPU_GEM_DOMAIN_GWS |
-					 AMDGPU_GEM_DOMAIN_OA);
-	bo->allowed_domains = bo->prefered_domains;
-	if (!kernel && bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
-		bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
 
-	bo->flags = flags;
+	/* TODO add more domains checks, such as AMDGPU_GEM_DOMAIN_CPU, _DOMAIN_DOORBELL */
+	if (size < man->size)
+		return true;
+
+	DRM_DEBUG("BO size %lu > total memory in domain: %llu\n", size, man->size);
+	return false;
+}
+
+bool amdgpu_bo_support_uswc(u64 bo_flags)
+{
 
 #ifdef CONFIG_X86_32
 	/* XXX: Write-combined CPU mappings of GTT seem broken on 32-bit
 	 * See https://bugs.freedesktop.org/show_bug.cgi?id=84627
 	 */
-	bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+	return false;
 #elif defined(CONFIG_X86) && !defined(CONFIG_X86_PAT)
 	/* Don't try to enable write-combining when it can't work, or things
 	 * may be slow
@@ -379,422 +604,439 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 	 thanks to write-combining
 #endif
 
-	if (bo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
+	if (bo_flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
 		DRM_INFO_ONCE("Please enable CONFIG_MTRR and CONFIG_X86_PAT for "
 			      "better performance thanks to write-combining\n");
-	bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+	return false;
 #else
 	/* For architectures that don't support WC memory,
 	 * mask out the WC flag from the BO
 	 */
 	if (!drm_arch_can_wc_memory())
-		bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+		return false;
+
+	return true;
 #endif
+}
 
-	amdgpu_fill_placement_to_bo(bo, placement);
-	/* Kernel allocation are uninterruptible */
+/**
+ * amdgpu_bo_create - create an &amdgpu_bo buffer object
+ * @adev: amdgpu device object
+ * @bp: parameters to be used for the buffer object
+ * @bo_ptr: pointer to the buffer object pointer
+ *
+ * Creates an &amdgpu_bo buffer object.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+int amdgpu_bo_create(struct amdgpu_device *adev,
+			       struct amdgpu_bo_param *bp,
+			       struct amdgpu_bo **bo_ptr)
+{
+	struct ttm_operation_ctx ctx = {
+		.interruptible = (bp->type != ttm_bo_type_kernel),
+		.no_wait_gpu = bp->no_wait_gpu,
+		/* We opt to avoid OOM on system pages allocations */
+		.gfp_retry_mayfail = true,
+		.allow_res_evict = bp->type != ttm_bo_type_kernel,
+		.resv = bp->resv
+	};
+	struct amdgpu_bo *bo;
+	unsigned long page_align, size = bp->size;
+	int r;
+
+	/* Note that GDS/GWS/OA allocates 1 page per byte/resource. */
+	if (bp->domain & (AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA)) {
+		/* GWS and OA don't need any alignment. */
+		page_align = bp->byte_align;
+		size <<= PAGE_SHIFT;
+
+	} else if (bp->domain & AMDGPU_GEM_DOMAIN_GDS) {
+		/* Both size and alignment must be a multiple of 4. */
+		page_align = ALIGN(bp->byte_align, 4);
+		size = ALIGN(size, 4) << PAGE_SHIFT;
+	} else {
+		/* Memory should be aligned at least to a page size. */
+		page_align = ALIGN(bp->byte_align, PAGE_SIZE) >> PAGE_SHIFT;
+		size = ALIGN(size, PAGE_SIZE);
+	}
+
+	if (!amdgpu_bo_validate_size(adev, size, bp->domain))
+		return -ENOMEM;
 
-	initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
-	r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, size, type,
-				 &bo->placement, page_align, !kernel, NULL,
-				 acc_size, sg, resv, &amdgpu_ttm_bo_destroy);
-	amdgpu_cs_report_moved_bytes(adev,
-		atomic64_read(&adev->num_bytes_moved) - initial_bytes_moved);
+	BUG_ON(bp->bo_ptr_size < sizeof(struct amdgpu_bo));
+
+	*bo_ptr = NULL;
+	bo = kvzalloc(bp->bo_ptr_size, GFP_KERNEL);
+	if (bo == NULL)
+		return -ENOMEM;
+	drm_gem_private_object_init(adev_to_drm(adev), &bo->tbo.base, size);
+	bo->tbo.base.funcs = &amdgpu_gem_object_funcs;
+	bo->vm_bo = NULL;
+	bo->preferred_domains = bp->preferred_domain ? bp->preferred_domain :
+		bp->domain;
+	bo->allowed_domains = bo->preferred_domains;
+	if (bp->type != ttm_bo_type_kernel &&
+	    !(bp->flags & AMDGPU_GEM_CREATE_DISCARDABLE) &&
+	    bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM)
+		bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT;
+
+	bo->flags = bp->flags;
+
+	if (adev->gmc.mem_partitions)
+		/* For GPUs with spatial partitioning, bo->xcp_id=-1 means any partition */
+		bo->xcp_id = bp->xcp_id_plus1 - 1;
+	else
+		/* For GPUs without spatial partitioning */
+		bo->xcp_id = 0;
 
+	if (!amdgpu_bo_support_uswc(bo->flags))
+		bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+
+	bo->tbo.bdev = &adev->mman.bdev;
+	if (bp->domain & (AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA |
+			  AMDGPU_GEM_DOMAIN_GDS))
+		amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+	else
+		amdgpu_bo_placement_from_domain(bo, bp->domain);
+	if (bp->type == ttm_bo_type_kernel)
+		bo->tbo.priority = 2;
+	else if (!(bp->flags & AMDGPU_GEM_CREATE_DISCARDABLE))
+		bo->tbo.priority = 1;
+
+	if (!bp->destroy)
+		bp->destroy = &amdgpu_bo_destroy;
+
+	r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, bp->type,
+				 &bo->placement, page_align, &ctx,  NULL,
+				 bp->resv, bp->destroy);
 	if (unlikely(r != 0))
 		return r;
 
-	if (kernel)
-		bo->tbo.priority = 1;
+	if (!amdgpu_gmc_vram_full_visible(&adev->gmc) &&
+	    amdgpu_res_cpu_visible(adev, bo->tbo.resource))
+		amdgpu_cs_report_moved_bytes(adev, ctx.bytes_moved,
+					     ctx.bytes_moved);
+	else
+		amdgpu_cs_report_moved_bytes(adev, ctx.bytes_moved, 0);
 
-	if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
-	    bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) {
+	if (bp->flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
+	    bo->tbo.resource->mem_type == TTM_PL_VRAM) {
 		struct dma_fence *fence;
 
-		r = amdgpu_fill_buffer(bo, 0, bo->tbo.resv, &fence);
+		r = amdgpu_ttm_clear_buffer(bo, bo->tbo.base.resv, &fence);
 		if (unlikely(r))
 			goto fail_unreserve;
 
-		amdgpu_bo_fence(bo, fence, false);
-		dma_fence_put(bo->tbo.moving);
-		bo->tbo.moving = dma_fence_get(fence);
+		dma_resv_add_fence(bo->tbo.base.resv, fence,
+				   DMA_RESV_USAGE_KERNEL);
 		dma_fence_put(fence);
 	}
-	if (!resv)
+	if (!bp->resv)
 		amdgpu_bo_unreserve(bo);
 	*bo_ptr = bo;
 
 	trace_amdgpu_bo_create(bo);
 
+	/* Treat CPU_ACCESS_REQUIRED only as a hint if given by UMD */
+	if (bp->type == ttm_bo_type_device)
+		bo->flags &= ~AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+
 	return 0;
 
 fail_unreserve:
-	if (!resv)
-		ww_mutex_unlock(&bo->tbo.resv->lock);
+	if (!bp->resv)
+		dma_resv_unlock(bo->tbo.base.resv);
 	amdgpu_bo_unref(&bo);
 	return r;
 }
 
-static int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-				   unsigned long size, int byte_align,
-				   struct amdgpu_bo *bo)
-{
-	struct ttm_placement placement = {0};
-	struct ttm_place placements[AMDGPU_GEM_DOMAIN_MAX + 1];
-	int r;
-
-	if (bo->shadow)
-		return 0;
-
-	bo->flags |= AMDGPU_GEM_CREATE_SHADOW;
-	memset(&placements, 0,
-	       (AMDGPU_GEM_DOMAIN_MAX + 1) * sizeof(struct ttm_place));
-
-	amdgpu_ttm_placement_init(adev, &placement,
-				  placements, AMDGPU_GEM_DOMAIN_GTT,
-				  AMDGPU_GEM_CREATE_CPU_GTT_USWC);
-
-	r = amdgpu_bo_create_restricted(adev, size, byte_align, true,
-					AMDGPU_GEM_DOMAIN_GTT,
-					AMDGPU_GEM_CREATE_CPU_GTT_USWC,
-					NULL, &placement,
-					bo->tbo.resv,
-					&bo->shadow);
-	if (!r) {
-		bo->shadow->parent = amdgpu_bo_ref(bo);
-		mutex_lock(&adev->shadow_list_lock);
-		list_add_tail(&bo->shadow_list, &adev->shadow_list);
-		mutex_unlock(&adev->shadow_list_lock);
-	}
-
-	return r;
-}
+/**
+ * amdgpu_bo_create_user - create an &amdgpu_bo_user buffer object
+ * @adev: amdgpu device object
+ * @bp: parameters to be used for the buffer object
+ * @ubo_ptr: pointer to the buffer object pointer
+ *
+ * Create a BO to be used by user application;
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
 
-int amdgpu_bo_create(struct amdgpu_device *adev,
-		     unsigned long size, int byte_align,
-		     bool kernel, u32 domain, u64 flags,
-		     struct sg_table *sg,
-		     struct reservation_object *resv,
-		     struct amdgpu_bo **bo_ptr)
+int amdgpu_bo_create_user(struct amdgpu_device *adev,
+			  struct amdgpu_bo_param *bp,
+			  struct amdgpu_bo_user **ubo_ptr)
 {
-	struct ttm_placement placement = {0};
-	struct ttm_place placements[AMDGPU_GEM_DOMAIN_MAX + 1];
+	struct amdgpu_bo *bo_ptr;
 	int r;
 
-	memset(&placements, 0,
-	       (AMDGPU_GEM_DOMAIN_MAX + 1) * sizeof(struct ttm_place));
-
-	amdgpu_ttm_placement_init(adev, &placement,
-				  placements, domain, flags);
-
-	r = amdgpu_bo_create_restricted(adev, size, byte_align, kernel,
-					domain, flags, sg, &placement,
-					resv, bo_ptr);
+	bp->bo_ptr_size = sizeof(struct amdgpu_bo_user);
+	bp->destroy = &amdgpu_bo_user_destroy;
+	r = amdgpu_bo_create(adev, bp, &bo_ptr);
 	if (r)
 		return r;
 
-	if (amdgpu_need_backup(adev) && (flags & AMDGPU_GEM_CREATE_SHADOW)) {
-		if (!resv) {
-			r = ww_mutex_lock(&(*bo_ptr)->tbo.resv->lock, NULL);
-			WARN_ON(r != 0);
-		}
-
-		r = amdgpu_bo_create_shadow(adev, size, byte_align, (*bo_ptr));
-
-		if (!resv)
-			ww_mutex_unlock(&(*bo_ptr)->tbo.resv->lock);
-
-		if (r)
-			amdgpu_bo_unref(bo_ptr);
-	}
-
+	*ubo_ptr = to_amdgpu_bo_user(bo_ptr);
 	return r;
 }
 
-int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
-			       struct amdgpu_ring *ring,
-			       struct amdgpu_bo *bo,
-			       struct reservation_object *resv,
-			       struct dma_fence **fence,
-			       bool direct)
-
-{
-	struct amdgpu_bo *shadow = bo->shadow;
-	uint64_t bo_addr, shadow_addr;
-	int r;
-
-	if (!shadow)
-		return -EINVAL;
-
-	bo_addr = amdgpu_bo_gpu_offset(bo);
-	shadow_addr = amdgpu_bo_gpu_offset(bo->shadow);
-
-	r = reservation_object_reserve_shared(bo->tbo.resv);
-	if (r)
-		goto err;
-
-	r = amdgpu_copy_buffer(ring, bo_addr, shadow_addr,
-			       amdgpu_bo_size(bo), resv, fence,
-			       direct);
-	if (!r)
-		amdgpu_bo_fence(bo, *fence, true);
-
-err:
-	return r;
-}
-
-int amdgpu_bo_validate(struct amdgpu_bo *bo)
-{
-	uint32_t domain;
-	int r;
-
-	if (bo->pin_count)
-		return 0;
-
-	domain = bo->prefered_domains;
-
-retry:
-	amdgpu_ttm_placement_from_domain(bo, domain);
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-	if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
-		domain = bo->allowed_domains;
-		goto retry;
-	}
-
-	return r;
-}
-
-int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
-				  struct amdgpu_ring *ring,
-				  struct amdgpu_bo *bo,
-				  struct reservation_object *resv,
-				  struct dma_fence **fence,
-				  bool direct)
+/**
+ * amdgpu_bo_create_vm - create an &amdgpu_bo_vm buffer object
+ * @adev: amdgpu device object
+ * @bp: parameters to be used for the buffer object
+ * @vmbo_ptr: pointer to the buffer object pointer
+ *
+ * Create a BO to be for GPUVM.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
 
+int amdgpu_bo_create_vm(struct amdgpu_device *adev,
+			struct amdgpu_bo_param *bp,
+			struct amdgpu_bo_vm **vmbo_ptr)
 {
-	struct amdgpu_bo *shadow = bo->shadow;
-	uint64_t bo_addr, shadow_addr;
+	struct amdgpu_bo *bo_ptr;
 	int r;
 
-	if (!shadow)
-		return -EINVAL;
-
-	bo_addr = amdgpu_bo_gpu_offset(bo);
-	shadow_addr = amdgpu_bo_gpu_offset(bo->shadow);
-
-	r = reservation_object_reserve_shared(bo->tbo.resv);
+	/* bo_ptr_size will be determined by the caller and it depends on
+	 * num of amdgpu_vm_pt entries.
+	 */
+	BUG_ON(bp->bo_ptr_size < sizeof(struct amdgpu_bo_vm));
+	r = amdgpu_bo_create(adev, bp, &bo_ptr);
 	if (r)
-		goto err;
-
-	r = amdgpu_copy_buffer(ring, shadow_addr, bo_addr,
-			       amdgpu_bo_size(bo), resv, fence,
-			       direct);
-	if (!r)
-		amdgpu_bo_fence(bo, *fence, true);
+		return r;
 
-err:
+	*vmbo_ptr = to_amdgpu_bo_vm(bo_ptr);
 	return r;
 }
 
+/**
+ * amdgpu_bo_kmap - map an &amdgpu_bo buffer object
+ * @bo: &amdgpu_bo buffer object to be mapped
+ * @ptr: kernel virtual address to be returned
+ *
+ * Calls ttm_bo_kmap() to set up the kernel virtual mapping; calls
+ * amdgpu_bo_kptr() to get the kernel virtual address.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
 int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr)
 {
-	bool is_iomem;
+	void *kptr;
 	long r;
 
 	if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
 		return -EPERM;
 
-	if (bo->kptr) {
-		if (ptr) {
-			*ptr = bo->kptr;
-		}
-		return 0;
-	}
-
-	r = reservation_object_wait_timeout_rcu(bo->tbo.resv, false, false,
-						MAX_SCHEDULE_TIMEOUT);
+	r = dma_resv_wait_timeout(bo->tbo.base.resv, DMA_RESV_USAGE_KERNEL,
+				  false, MAX_SCHEDULE_TIMEOUT);
 	if (r < 0)
 		return r;
 
-	r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap);
+	kptr = amdgpu_bo_kptr(bo);
+	if (kptr) {
+		if (ptr)
+			*ptr = kptr;
+		return 0;
+	}
+
+	r = ttm_bo_kmap(&bo->tbo, 0, PFN_UP(bo->tbo.base.size), &bo->kmap);
 	if (r)
 		return r;
 
-	bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
 	if (ptr)
-		*ptr = bo->kptr;
+		*ptr = amdgpu_bo_kptr(bo);
 
 	return 0;
 }
 
+/**
+ * amdgpu_bo_kptr - returns a kernel virtual address of the buffer object
+ * @bo: &amdgpu_bo buffer object
+ *
+ * Calls ttm_kmap_obj_virtual() to get the kernel virtual address
+ *
+ * Returns:
+ * the virtual address of a buffer object area.
+ */
+void *amdgpu_bo_kptr(struct amdgpu_bo *bo)
+{
+	bool is_iomem;
+
+	return ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
+}
+
+/**
+ * amdgpu_bo_kunmap - unmap an &amdgpu_bo buffer object
+ * @bo: &amdgpu_bo buffer object to be unmapped
+ *
+ * Unmaps a kernel map set up by amdgpu_bo_kmap().
+ */
 void amdgpu_bo_kunmap(struct amdgpu_bo *bo)
 {
-	if (bo->kptr == NULL)
-		return;
-	bo->kptr = NULL;
-	ttm_bo_kunmap(&bo->kmap);
+	if (bo->kmap.bo)
+		ttm_bo_kunmap(&bo->kmap);
 }
 
+/**
+ * amdgpu_bo_ref - reference an &amdgpu_bo buffer object
+ * @bo: &amdgpu_bo buffer object
+ *
+ * References the contained &ttm_buffer_object.
+ *
+ * Returns:
+ * a refcounted pointer to the &amdgpu_bo buffer object.
+ */
 struct amdgpu_bo *amdgpu_bo_ref(struct amdgpu_bo *bo)
 {
 	if (bo == NULL)
 		return NULL;
 
-	ttm_bo_reference(&bo->tbo);
+	drm_gem_object_get(&bo->tbo.base);
 	return bo;
 }
 
+/**
+ * amdgpu_bo_unref - unreference an &amdgpu_bo buffer object
+ * @bo: &amdgpu_bo buffer object
+ *
+ * Unreferences the contained &ttm_buffer_object and clear the pointer
+ */
 void amdgpu_bo_unref(struct amdgpu_bo **bo)
 {
-	struct ttm_buffer_object *tbo;
-
 	if ((*bo) == NULL)
 		return;
 
-	tbo = &((*bo)->tbo);
-	ttm_bo_unref(&tbo);
-	if (tbo == NULL)
-		*bo = NULL;
+	drm_gem_object_put(&(*bo)->tbo.base);
+	*bo = NULL;
 }
 
-int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
-			     u64 min_offset, u64 max_offset,
-			     u64 *gpu_addr)
+/**
+ * amdgpu_bo_pin - pin an &amdgpu_bo buffer object
+ * @bo: &amdgpu_bo buffer object to be pinned
+ * @domain: domain to be pinned to
+ *
+ * Pins the buffer object according to requested domain. If the memory is
+ * unbound gart memory, binds the pages into gart table. Adjusts pin_count and
+ * pin_size accordingly.
+ *
+ * Pinning means to lock pages in memory along with keeping them at a fixed
+ * offset. It is required when a buffer can not be moved, for example, when
+ * a display buffer is being scanned out.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct ttm_operation_ctx ctx = { false, false };
 	int r, i;
-	unsigned fpfn, lpfn;
 
 	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm))
 		return -EPERM;
 
-	if (WARN_ON_ONCE(min_offset > max_offset))
-		return -EINVAL;
+	/* Check domain to be pinned to against preferred domains */
+	if (bo->preferred_domains & domain)
+		domain = bo->preferred_domains & domain;
 
 	/* A shared bo cannot be migrated to VRAM */
-	if (bo->prime_shared_count && (domain == AMDGPU_GEM_DOMAIN_VRAM))
-		return -EINVAL;
+	if (drm_gem_is_imported(&bo->tbo.base)) {
+		if (domain & AMDGPU_GEM_DOMAIN_GTT)
+			domain = AMDGPU_GEM_DOMAIN_GTT;
+		else
+			return -EINVAL;
+	}
 
-	if (bo->pin_count) {
-		uint32_t mem_type = bo->tbo.mem.mem_type;
+	if (bo->tbo.pin_count) {
+		uint32_t mem_type = bo->tbo.resource->mem_type;
+		uint32_t mem_flags = bo->tbo.resource->placement;
 
-		if (domain != amdgpu_mem_type_to_domain(mem_type))
+		if (!(domain & amdgpu_mem_type_to_domain(mem_type)))
 			return -EINVAL;
 
-		bo->pin_count++;
-		if (gpu_addr)
-			*gpu_addr = amdgpu_bo_gpu_offset(bo);
-
-		if (max_offset != 0) {
-			u64 domain_start = bo->tbo.bdev->man[mem_type].gpu_offset;
-			WARN_ON_ONCE(max_offset <
-				     (amdgpu_bo_gpu_offset(bo) - domain_start));
-		}
+		if ((mem_type == TTM_PL_VRAM) &&
+		    (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS) &&
+		    !(mem_flags & TTM_PL_FLAG_CONTIGUOUS))
+			return -EINVAL;
 
+		ttm_bo_pin(&bo->tbo);
 		return 0;
 	}
 
-	bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-	amdgpu_ttm_placement_from_domain(bo, domain);
+	/* This assumes only APU display buffers are pinned with (VRAM|GTT).
+	 * See function amdgpu_display_supported_domains()
+	 */
+	domain = amdgpu_bo_get_preferred_domain(adev, domain);
+
+	if (drm_gem_is_imported(&bo->tbo.base))
+		dma_buf_pin(bo->tbo.base.import_attach);
+
+	/* force to pin into visible video ram */
+	if (!(bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS))
+		bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+	amdgpu_bo_placement_from_domain(bo, domain);
 	for (i = 0; i < bo->placement.num_placement; i++) {
-		/* force to pin into visible video ram */
-		if ((bo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
-		    !(bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) &&
-		    (!max_offset || max_offset >
-		     adev->mc.visible_vram_size)) {
-			if (WARN_ON_ONCE(min_offset >
-					 adev->mc.visible_vram_size))
-				return -EINVAL;
-			fpfn = min_offset >> PAGE_SHIFT;
-			lpfn = adev->mc.visible_vram_size >> PAGE_SHIFT;
-		} else {
-			fpfn = min_offset >> PAGE_SHIFT;
-			lpfn = max_offset >> PAGE_SHIFT;
-		}
-		if (fpfn > bo->placements[i].fpfn)
-			bo->placements[i].fpfn = fpfn;
-		if (!bo->placements[i].lpfn ||
-		    (lpfn && lpfn < bo->placements[i].lpfn))
-			bo->placements[i].lpfn = lpfn;
-		bo->placements[i].flags |= TTM_PL_FLAG_NO_EVICT;
+		if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS &&
+		    bo->placements[i].mem_type == TTM_PL_VRAM)
+			bo->placements[i].flags |= TTM_PL_FLAG_CONTIGUOUS;
 	}
 
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
+	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
 	if (unlikely(r)) {
 		dev_err(adev->dev, "%p pin failed\n", bo);
 		goto error;
 	}
-	r = amdgpu_ttm_bind(&bo->tbo, &bo->tbo.mem);
-	if (unlikely(r)) {
-		dev_err(adev->dev, "%p bind failed\n", bo);
-		goto error;
-	}
 
-	bo->pin_count = 1;
-	if (gpu_addr != NULL)
-		*gpu_addr = amdgpu_bo_gpu_offset(bo);
-	if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
-		adev->vram_pin_size += amdgpu_bo_size(bo);
-		if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
-			adev->invisible_pin_size += amdgpu_bo_size(bo);
-	} else if (domain == AMDGPU_GEM_DOMAIN_GTT) {
-		adev->gart_pin_size += amdgpu_bo_size(bo);
+	ttm_bo_pin(&bo->tbo);
+
+	if (bo->tbo.resource->mem_type == TTM_PL_VRAM) {
+		atomic64_add(amdgpu_bo_size(bo), &adev->vram_pin_size);
+		atomic64_add(amdgpu_vram_mgr_bo_visible_size(bo),
+			     &adev->visible_pin_size);
+	} else if (bo->tbo.resource->mem_type == TTM_PL_TT) {
+		atomic64_add(amdgpu_bo_size(bo), &adev->gart_pin_size);
 	}
 
 error:
 	return r;
 }
 
-int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain, u64 *gpu_addr)
-{
-	return amdgpu_bo_pin_restricted(bo, domain, 0, 0, gpu_addr);
-}
-
-int amdgpu_bo_unpin(struct amdgpu_bo *bo)
+/**
+ * amdgpu_bo_unpin - unpin an &amdgpu_bo buffer object
+ * @bo: &amdgpu_bo buffer object to be unpinned
+ *
+ * Decreases the pin_count, and clears the flags if pin_count reaches 0.
+ * Changes placement and pin size accordingly.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+void amdgpu_bo_unpin(struct amdgpu_bo *bo)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	int r, i;
 
-	if (!bo->pin_count) {
-		dev_warn(adev->dev, "%p unpin not necessary\n", bo);
-		return 0;
-	}
-	bo->pin_count--;
-	if (bo->pin_count)
-		return 0;
-	for (i = 0; i < bo->placement.num_placement; i++) {
-		bo->placements[i].lpfn = 0;
-		bo->placements[i].flags &= ~TTM_PL_FLAG_NO_EVICT;
-	}
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-	if (unlikely(r)) {
-		dev_err(adev->dev, "%p validate failed for unpin\n", bo);
-		goto error;
-	}
+	ttm_bo_unpin(&bo->tbo);
+	if (bo->tbo.pin_count)
+		return;
 
-	if (bo->tbo.mem.mem_type == TTM_PL_VRAM) {
-		adev->vram_pin_size -= amdgpu_bo_size(bo);
-		if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
-			adev->invisible_pin_size -= amdgpu_bo_size(bo);
-	} else if (bo->tbo.mem.mem_type == TTM_PL_TT) {
-		adev->gart_pin_size -= amdgpu_bo_size(bo);
-	}
+	if (drm_gem_is_imported(&bo->tbo.base))
+		dma_buf_unpin(bo->tbo.base.import_attach);
 
-error:
-	return r;
-}
-
-int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
-{
-	/* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
-	if (0 && (adev->flags & AMD_IS_APU)) {
-		/* Useless to evict on IGP chips */
-		return 0;
+	if (bo->tbo.resource->mem_type == TTM_PL_VRAM) {
+		atomic64_sub(amdgpu_bo_size(bo), &adev->vram_pin_size);
+		atomic64_sub(amdgpu_vram_mgr_bo_visible_size(bo),
+			     &adev->visible_pin_size);
+	} else if (bo->tbo.resource->mem_type == TTM_PL_TT) {
+		atomic64_sub(amdgpu_bo_size(bo), &adev->gart_pin_size);
 	}
-	return ttm_bo_evict_mm(&adev->mman.bdev, TTM_PL_VRAM);
+
 }
 
-static const char *amdgpu_vram_names[] = {
+static const char * const amdgpu_vram_names[] = {
 	"UNKNOWN",
 	"GDDR1",
 	"DDR2",
@@ -802,69 +1044,143 @@ static const char *amdgpu_vram_names[] = {
 	"GDDR4",
 	"GDDR5",
 	"HBM",
-	"DDR3"
+	"DDR3",
+	"DDR4",
+	"GDDR6",
+	"DDR5",
+	"LPDDR4",
+	"LPDDR5",
+	"HBM3E"
 };
 
+/**
+ * amdgpu_bo_init - initialize memory manager
+ * @adev: amdgpu device object
+ *
+ * Calls amdgpu_ttm_init() to initialize amdgpu memory manager.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
 int amdgpu_bo_init(struct amdgpu_device *adev)
 {
-	/* reserve PAT memory space to WC for VRAM */
-	arch_io_reserve_memtype_wc(adev->mc.aper_base,
-				   adev->mc.aper_size);
+	/* On A+A platform, VRAM can be mapped as WB */
+	if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) {
+		/* reserve PAT memory space to WC for VRAM */
+		int r = arch_io_reserve_memtype_wc(adev->gmc.aper_base,
+				adev->gmc.aper_size);
+
+		if (r) {
+			DRM_ERROR("Unable to set WC memtype for the aperture base\n");
+			return r;
+		}
+
+		/* Add an MTRR for the VRAM */
+		adev->gmc.vram_mtrr = arch_phys_wc_add(adev->gmc.aper_base,
+				adev->gmc.aper_size);
+	}
 
-	/* Add an MTRR for the VRAM */
-	adev->mc.vram_mtrr = arch_phys_wc_add(adev->mc.aper_base,
-					      adev->mc.aper_size);
 	DRM_INFO("Detected VRAM RAM=%lluM, BAR=%lluM\n",
-		adev->mc.mc_vram_size >> 20,
-		(unsigned long long)adev->mc.aper_size >> 20);
+		 adev->gmc.mc_vram_size >> 20,
+		 (unsigned long long)adev->gmc.aper_size >> 20);
 	DRM_INFO("RAM width %dbits %s\n",
-		 adev->mc.vram_width, amdgpu_vram_names[adev->mc.vram_type]);
+		 adev->gmc.vram_width, amdgpu_vram_names[adev->gmc.vram_type]);
 	return amdgpu_ttm_init(adev);
 }
 
+/**
+ * amdgpu_bo_fini - tear down memory manager
+ * @adev: amdgpu device object
+ *
+ * Reverses amdgpu_bo_init() to tear down memory manager.
+ */
 void amdgpu_bo_fini(struct amdgpu_device *adev)
 {
+	int idx;
+
 	amdgpu_ttm_fini(adev);
-	arch_phys_wc_del(adev->mc.vram_mtrr);
-	arch_io_free_memtype_wc(adev->mc.aper_base, adev->mc.aper_size);
-}
 
-int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
-			     struct vm_area_struct *vma)
-{
-	return ttm_fbdev_mmap(vma, &bo->tbo);
+	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
+		if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) {
+			arch_phys_wc_del(adev->gmc.vram_mtrr);
+			arch_io_free_memtype_wc(adev->gmc.aper_base, adev->gmc.aper_size);
+		}
+		drm_dev_exit(idx);
+	}
 }
 
+/**
+ * amdgpu_bo_set_tiling_flags - set tiling flags
+ * @bo: &amdgpu_bo buffer object
+ * @tiling_flags: new flags
+ *
+ * Sets buffer object's tiling flags with the new one. Used by GEM ioctl or
+ * kernel driver to set the tiling flags on a buffer.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
 int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct amdgpu_bo_user *ubo;
 
+	BUG_ON(bo->tbo.type == ttm_bo_type_kernel);
 	if (adev->family <= AMDGPU_FAMILY_CZ &&
 	    AMDGPU_TILING_GET(tiling_flags, TILE_SPLIT) > 6)
 		return -EINVAL;
 
-	bo->tiling_flags = tiling_flags;
+	ubo = to_amdgpu_bo_user(bo);
+	ubo->tiling_flags = tiling_flags;
 	return 0;
 }
 
+/**
+ * amdgpu_bo_get_tiling_flags - get tiling flags
+ * @bo: &amdgpu_bo buffer object
+ * @tiling_flags: returned flags
+ *
+ * Gets buffer object's tiling flags. Used by GEM ioctl or kernel driver to
+ * set the tiling flags on a buffer.
+ */
 void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64 *tiling_flags)
 {
-	lockdep_assert_held(&bo->tbo.resv->lock.base);
+	struct amdgpu_bo_user *ubo;
+
+	BUG_ON(bo->tbo.type == ttm_bo_type_kernel);
+	dma_resv_assert_held(bo->tbo.base.resv);
+	ubo = to_amdgpu_bo_user(bo);
 
 	if (tiling_flags)
-		*tiling_flags = bo->tiling_flags;
+		*tiling_flags = ubo->tiling_flags;
 }
 
-int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
-			    uint32_t metadata_size, uint64_t flags)
+/**
+ * amdgpu_bo_set_metadata - set metadata
+ * @bo: &amdgpu_bo buffer object
+ * @metadata: new metadata
+ * @metadata_size: size of the new metadata
+ * @flags: flags of the new metadata
+ *
+ * Sets buffer object's metadata, its size and flags.
+ * Used via GEM ioctl.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+int amdgpu_bo_set_metadata(struct amdgpu_bo *bo, void *metadata,
+			   u32 metadata_size, uint64_t flags)
 {
+	struct amdgpu_bo_user *ubo;
 	void *buffer;
 
+	BUG_ON(bo->tbo.type == ttm_bo_type_kernel);
+	ubo = to_amdgpu_bo_user(bo);
 	if (!metadata_size) {
-		if (bo->metadata_size) {
-			kfree(bo->metadata);
-			bo->metadata = NULL;
-			bo->metadata_size = 0;
+		if (ubo->metadata_size) {
+			kfree(ubo->metadata);
+			ubo->metadata = NULL;
+			ubo->metadata_size = 0;
 		}
 		return 0;
 	}
@@ -876,112 +1192,196 @@ int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
 	if (buffer == NULL)
 		return -ENOMEM;
 
-	kfree(bo->metadata);
-	bo->metadata_flags = flags;
-	bo->metadata = buffer;
-	bo->metadata_size = metadata_size;
+	kfree(ubo->metadata);
+	ubo->metadata_flags = flags;
+	ubo->metadata = buffer;
+	ubo->metadata_size = metadata_size;
 
 	return 0;
 }
 
+/**
+ * amdgpu_bo_get_metadata - get metadata
+ * @bo: &amdgpu_bo buffer object
+ * @buffer: returned metadata
+ * @buffer_size: size of the buffer
+ * @metadata_size: size of the returned metadata
+ * @flags: flags of the returned metadata
+ *
+ * Gets buffer object's metadata, its size and flags. buffer_size shall not be
+ * less than metadata_size.
+ * Used via GEM ioctl.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
 int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
 			   size_t buffer_size, uint32_t *metadata_size,
 			   uint64_t *flags)
 {
+	struct amdgpu_bo_user *ubo;
+
 	if (!buffer && !metadata_size)
 		return -EINVAL;
 
+	BUG_ON(bo->tbo.type == ttm_bo_type_kernel);
+	ubo = to_amdgpu_bo_user(bo);
+	if (metadata_size)
+		*metadata_size = ubo->metadata_size;
+
 	if (buffer) {
-		if (buffer_size < bo->metadata_size)
+		if (buffer_size < ubo->metadata_size)
 			return -EINVAL;
 
-		if (bo->metadata_size)
-			memcpy(buffer, bo->metadata, bo->metadata_size);
+		if (ubo->metadata_size)
+			memcpy(buffer, ubo->metadata, ubo->metadata_size);
 	}
 
-	if (metadata_size)
-		*metadata_size = bo->metadata_size;
 	if (flags)
-		*flags = bo->metadata_flags;
+		*flags = ubo->metadata_flags;
 
 	return 0;
 }
 
+/**
+ * amdgpu_bo_move_notify - notification about a memory move
+ * @bo: pointer to a buffer object
+ * @evict: if this move is evicting the buffer from the graphics address space
+ * @new_mem: new resource for backing the BO
+ *
+ * Marks the corresponding &amdgpu_bo buffer object as invalid, also performs
+ * bookkeeping.
+ * TTM driver callback which is called when ttm moves a buffer.
+ */
 void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 			   bool evict,
-			   struct ttm_mem_reg *new_mem)
+			   struct ttm_resource *new_mem)
 {
-	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
+	struct ttm_resource *old_mem = bo->resource;
 	struct amdgpu_bo *abo;
-	struct ttm_mem_reg *old_mem = &bo->mem;
 
-	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
+	if (!amdgpu_bo_is_amdgpu_bo(bo))
 		return;
 
-	abo = container_of(bo, struct amdgpu_bo, tbo);
-	amdgpu_vm_bo_invalidate(adev, abo);
+	abo = ttm_to_amdgpu_bo(bo);
+	amdgpu_vm_bo_move(abo, new_mem, evict);
 
-	/* remember the eviction */
-	if (evict)
-		atomic64_inc(&adev->num_evictions);
+	amdgpu_bo_kunmap(abo);
 
-	/* update statistics */
-	if (!new_mem)
-		return;
+	if (abo->tbo.base.dma_buf && !drm_gem_is_imported(&abo->tbo.base) &&
+	    old_mem && old_mem->mem_type != TTM_PL_SYSTEM)
+		dma_buf_move_notify(abo->tbo.base.dma_buf);
 
 	/* move_notify is called before move happens */
-	amdgpu_update_memory_usage(adev, &bo->mem, new_mem);
-
-	trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
+	trace_amdgpu_bo_move(abo, new_mem ? new_mem->mem_type : -1,
+			     old_mem ? old_mem->mem_type : -1);
 }
 
-int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
+/**
+ * amdgpu_bo_release_notify - notification about a BO being released
+ * @bo: pointer to a buffer object
+ *
+ * Wipes VRAM buffers whose contents should not be leaked before the
+ * memory is released.
+ */
+void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
+	struct dma_fence *fence = NULL;
 	struct amdgpu_bo *abo;
-	unsigned long offset, size, lpfn;
-	int i, r;
+	int r;
 
-	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
-		return 0;
+	if (!amdgpu_bo_is_amdgpu_bo(bo))
+		return;
 
-	abo = container_of(bo, struct amdgpu_bo, tbo);
-	if (bo->mem.mem_type != TTM_PL_VRAM)
-		return 0;
+	abo = ttm_to_amdgpu_bo(bo);
+
+	WARN_ON(abo->vm_bo);
+
+	if (abo->kfd_bo)
+		amdgpu_amdkfd_release_notify(abo);
+
+	/*
+	 * We lock the private dma_resv object here and since the BO is about to
+	 * be released nobody else should have a pointer to it.
+	 * So when this locking here fails something is wrong with the reference
+	 * counting.
+	 */
+	if (WARN_ON_ONCE(!dma_resv_trylock(&bo->base._resv)))
+		return;
+
+	amdgpu_amdkfd_remove_all_eviction_fences(abo);
+
+	if (!bo->resource || bo->resource->mem_type != TTM_PL_VRAM ||
+	    !(abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE) ||
+	    adev->in_suspend || drm_dev_is_unplugged(adev_to_drm(adev)))
+		goto out;
 
-	size = bo->mem.num_pages << PAGE_SHIFT;
-	offset = bo->mem.start << PAGE_SHIFT;
-	/* TODO: figure out how to map scattered VRAM to the CPU */
-	if ((offset + size) <= adev->mc.visible_vram_size)
+	r = dma_resv_reserve_fences(&bo->base._resv, 1);
+	if (r)
+		goto out;
+
+	r = amdgpu_fill_buffer(abo, 0, &bo->base._resv, &fence, true,
+			       AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
+	if (WARN_ON(r))
+		goto out;
+
+	amdgpu_vram_mgr_set_cleared(bo->resource);
+	dma_resv_add_fence(&bo->base._resv, fence, DMA_RESV_USAGE_KERNEL);
+	dma_fence_put(fence);
+
+out:
+	dma_resv_unlock(&bo->base._resv);
+}
+
+/**
+ * amdgpu_bo_fault_reserve_notify - notification about a memory fault
+ * @bo: pointer to a buffer object
+ *
+ * Notifies the driver we are taking a fault on this BO and have reserved it,
+ * also performs bookkeeping.
+ * TTM driver callback for dealing with vm faults.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+vm_fault_t amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
+	struct ttm_operation_ctx ctx = { false, false };
+	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
+	int r;
+
+	/* Remember that this BO was accessed by the CPU */
+	abo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+
+	if (amdgpu_res_cpu_visible(adev, bo->resource))
 		return 0;
 
 	/* Can't move a pinned BO to visible VRAM */
-	if (abo->pin_count > 0)
-		return -EINVAL;
+	if (abo->tbo.pin_count > 0)
+		return VM_FAULT_SIGBUS;
 
 	/* hurrah the memory is not visible ! */
-	amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM);
-	lpfn =	adev->mc.visible_vram_size >> PAGE_SHIFT;
-	for (i = 0; i < abo->placement.num_placement; i++) {
-		/* Force into visible VRAM */
-		if ((abo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
-		    (!abo->placements[i].lpfn ||
-		     abo->placements[i].lpfn > lpfn))
-			abo->placements[i].lpfn = lpfn;
-	}
-	r = ttm_bo_validate(bo, &abo->placement, false, false);
-	if (unlikely(r == -ENOMEM)) {
-		amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT);
-		return ttm_bo_validate(bo, &abo->placement, false, false);
-	} else if (unlikely(r != 0)) {
-		return r;
-	}
+	atomic64_inc(&adev->num_vram_cpu_page_faults);
+	amdgpu_bo_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM |
+					AMDGPU_GEM_DOMAIN_GTT);
+
+	/* Avoid costly evictions; only set GTT as a busy placement */
+	abo->placements[0].flags |= TTM_PL_FLAG_DESIRED;
+
+	r = ttm_bo_validate(bo, &abo->placement, &ctx);
+	if (unlikely(r == -EBUSY || r == -ERESTARTSYS))
+		return VM_FAULT_NOPAGE;
+	else if (unlikely(r))
+		return VM_FAULT_SIGBUS;
 
-	offset = bo->mem.start << PAGE_SHIFT;
 	/* this should never happen */
-	if ((offset + size) > adev->mc.visible_vram_size)
-		return -EINVAL;
+	if (bo->resource->mem_type == TTM_PL_VRAM &&
+	    !amdgpu_res_cpu_visible(adev, bo->resource))
+		return VM_FAULT_SIGBUS;
 
+	ttm_bo_move_to_lru_tail_unlocked(bo);
 	return 0;
 }
 
@@ -996,33 +1396,293 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
 void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
 		     bool shared)
 {
-	struct reservation_object *resv = bo->tbo.resv;
+	struct dma_resv *resv = bo->tbo.base.resv;
+	int r;
 
-	if (shared)
-		reservation_object_add_shared_fence(resv, fence);
-	else
-		reservation_object_add_excl_fence(resv, fence);
+	r = dma_resv_reserve_fences(resv, 1);
+	if (r) {
+		/* As last resort on OOM we block for the fence */
+		dma_fence_wait(fence, false);
+		return;
+	}
+
+	dma_resv_add_fence(resv, fence, shared ? DMA_RESV_USAGE_READ :
+			   DMA_RESV_USAGE_WRITE);
+}
+
+/**
+ * amdgpu_bo_sync_wait_resv - Wait for BO reservation fences
+ *
+ * @adev: amdgpu device pointer
+ * @resv: reservation object to sync to
+ * @sync_mode: synchronization mode
+ * @owner: fence owner
+ * @intr: Whether the wait is interruptible
+ *
+ * Extract the fences from the reservation object and waits for them to finish.
+ *
+ * Returns:
+ * 0 on success, errno otherwise.
+ */
+int amdgpu_bo_sync_wait_resv(struct amdgpu_device *adev, struct dma_resv *resv,
+			     enum amdgpu_sync_mode sync_mode, void *owner,
+			     bool intr)
+{
+	struct amdgpu_sync sync;
+	int r;
+
+	amdgpu_sync_create(&sync);
+	amdgpu_sync_resv(adev, &sync, resv, sync_mode, owner);
+	r = amdgpu_sync_wait(&sync, intr);
+	amdgpu_sync_free(&sync);
+	return r;
+}
+
+/**
+ * amdgpu_bo_sync_wait - Wrapper for amdgpu_bo_sync_wait_resv
+ * @bo: buffer object to wait for
+ * @owner: fence owner
+ * @intr: Whether the wait is interruptible
+ *
+ * Wrapper to wait for fences in a BO.
+ * Returns:
+ * 0 on success, errno otherwise.
+ */
+int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+
+	return amdgpu_bo_sync_wait_resv(adev, bo->tbo.base.resv,
+					AMDGPU_SYNC_NE_OWNER, owner, intr);
 }
 
 /**
  * amdgpu_bo_gpu_offset - return GPU offset of bo
  * @bo:	amdgpu object for which we query the offset
  *
- * Returns current GPU offset of the object.
- *
  * Note: object should either be pinned or reserved when calling this
  * function, it might be useful to add check for this for debugging.
+ *
+ * Returns:
+ * current GPU offset of the object.
  */
 u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
 {
-	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_SYSTEM);
-	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_TT &&
-		     !amdgpu_ttm_is_bound(bo->tbo.ttm));
-	WARN_ON_ONCE(!ww_mutex_is_locked(&bo->tbo.resv->lock) &&
-		     !bo->pin_count);
-	WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
-	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
+	WARN_ON_ONCE(bo->tbo.resource->mem_type == TTM_PL_SYSTEM);
+	WARN_ON_ONCE(!dma_resv_is_locked(bo->tbo.base.resv) &&
+		     !bo->tbo.pin_count && bo->tbo.type != ttm_bo_type_kernel);
+	WARN_ON_ONCE(bo->tbo.resource->start == AMDGPU_BO_INVALID_OFFSET);
+	WARN_ON_ONCE(bo->tbo.resource->mem_type == TTM_PL_VRAM &&
 		     !(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
 
-	return bo->tbo.offset;
+	return amdgpu_bo_gpu_offset_no_check(bo);
+}
+
+/**
+ * amdgpu_bo_fb_aper_addr - return FB aperture GPU offset of the VRAM bo
+ * @bo:	amdgpu VRAM buffer object for which we query the offset
+ *
+ * Returns:
+ * current FB aperture GPU offset of the object.
+ */
+u64 amdgpu_bo_fb_aper_addr(struct amdgpu_bo *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	uint64_t offset, fb_base;
+
+	WARN_ON_ONCE(bo->tbo.resource->mem_type != TTM_PL_VRAM);
+
+	fb_base = adev->gmc.fb_start;
+	fb_base += adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
+	offset = (bo->tbo.resource->start << PAGE_SHIFT) + fb_base;
+	return amdgpu_gmc_sign_extend(offset);
+}
+
+/**
+ * amdgpu_bo_gpu_offset_no_check - return GPU offset of bo
+ * @bo:	amdgpu object for which we query the offset
+ *
+ * Returns:
+ * current GPU offset of the object without raising warnings.
+ */
+u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	uint64_t offset = AMDGPU_BO_INVALID_OFFSET;
+
+	if (bo->tbo.resource->mem_type == TTM_PL_TT)
+		offset = amdgpu_gmc_agp_addr(&bo->tbo);
+
+	if (offset == AMDGPU_BO_INVALID_OFFSET)
+		offset = (bo->tbo.resource->start << PAGE_SHIFT) +
+			amdgpu_ttm_domain_start(adev, bo->tbo.resource->mem_type);
+
+	return amdgpu_gmc_sign_extend(offset);
 }
+
+/**
+ * amdgpu_bo_mem_stats_placement - bo placement for memory accounting
+ * @bo:	the buffer object we should look at
+ *
+ * BO can have multiple preferred placements, to avoid double counting we want
+ * to file it under a single placement for memory stats.
+ * Luckily, if we take the highest set bit in preferred_domains the result is
+ * quite sensible.
+ *
+ * Returns:
+ * Which of the placements should the BO be accounted under.
+ */
+uint32_t amdgpu_bo_mem_stats_placement(struct amdgpu_bo *bo)
+{
+	uint32_t domain = bo->preferred_domains & AMDGPU_GEM_DOMAIN_MASK;
+
+	if (!domain)
+		return TTM_PL_SYSTEM;
+
+	switch (rounddown_pow_of_two(domain)) {
+	case AMDGPU_GEM_DOMAIN_CPU:
+		return TTM_PL_SYSTEM;
+	case AMDGPU_GEM_DOMAIN_GTT:
+		return TTM_PL_TT;
+	case AMDGPU_GEM_DOMAIN_VRAM:
+		return TTM_PL_VRAM;
+	case AMDGPU_GEM_DOMAIN_GDS:
+		return AMDGPU_PL_GDS;
+	case AMDGPU_GEM_DOMAIN_GWS:
+		return AMDGPU_PL_GWS;
+	case AMDGPU_GEM_DOMAIN_OA:
+		return AMDGPU_PL_OA;
+	case AMDGPU_GEM_DOMAIN_DOORBELL:
+		return AMDGPU_PL_DOORBELL;
+	case AMDGPU_GEM_DOMAIN_MMIO_REMAP:
+		return AMDGPU_PL_MMIO_REMAP;
+	default:
+		return TTM_PL_SYSTEM;
+	}
+}
+
+/**
+ * amdgpu_bo_get_preferred_domain - get preferred domain
+ * @adev: amdgpu device object
+ * @domain: allowed :ref:`memory domains <amdgpu_memory_domains>`
+ *
+ * Returns:
+ * Which of the allowed domains is preferred for allocating the BO.
+ */
+uint32_t amdgpu_bo_get_preferred_domain(struct amdgpu_device *adev,
+					    uint32_t domain)
+{
+	if ((domain == (AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT)) &&
+	    ((adev->asic_type == CHIP_CARRIZO) || (adev->asic_type == CHIP_STONEY))) {
+		domain = AMDGPU_GEM_DOMAIN_VRAM;
+		if (adev->gmc.real_vram_size <= AMDGPU_SG_THRESHOLD)
+			domain = AMDGPU_GEM_DOMAIN_GTT;
+	}
+	return domain;
+}
+
+#if defined(CONFIG_DEBUG_FS)
+#define amdgpu_bo_print_flag(m, bo, flag)		        \
+	do {							\
+		if (bo->flags & (AMDGPU_GEM_CREATE_ ## flag)) {	\
+			seq_printf((m), " " #flag);		\
+		}						\
+	} while (0)
+
+/**
+ * amdgpu_bo_print_info - print BO info in debugfs file
+ *
+ * @id: Index or Id of the BO
+ * @bo: Requested BO for printing info
+ * @m: debugfs file
+ *
+ * Print BO information in debugfs file
+ *
+ * Returns:
+ * Size of the BO in bytes.
+ */
+u64 amdgpu_bo_print_info(int id, struct amdgpu_bo *bo, struct seq_file *m)
+{
+	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+	struct dma_buf_attachment *attachment;
+	struct dma_buf *dma_buf;
+	const char *placement;
+	unsigned int pin_count;
+	u64 size;
+
+	if (dma_resv_trylock(bo->tbo.base.resv)) {
+		if (!bo->tbo.resource) {
+			placement = "NONE";
+		} else {
+			switch (bo->tbo.resource->mem_type) {
+			case TTM_PL_VRAM:
+				if (amdgpu_res_cpu_visible(adev, bo->tbo.resource))
+					placement = "VRAM VISIBLE";
+				else
+					placement = "VRAM";
+				break;
+			case TTM_PL_TT:
+				placement = "GTT";
+				break;
+			case AMDGPU_PL_GDS:
+				placement = "GDS";
+				break;
+			case AMDGPU_PL_GWS:
+				placement = "GWS";
+				break;
+			case AMDGPU_PL_OA:
+				placement = "OA";
+				break;
+			case AMDGPU_PL_PREEMPT:
+				placement = "PREEMPTIBLE";
+				break;
+			case AMDGPU_PL_DOORBELL:
+				placement = "DOORBELL";
+				break;
+			case AMDGPU_PL_MMIO_REMAP:
+				placement = "MMIO REMAP";
+				break;
+			case TTM_PL_SYSTEM:
+			default:
+				placement = "CPU";
+				break;
+			}
+		}
+		dma_resv_unlock(bo->tbo.base.resv);
+	} else {
+		placement = "UNKNOWN";
+	}
+
+	size = amdgpu_bo_size(bo);
+	seq_printf(m, "\t\t0x%08x: %12lld byte %s",
+			id, size, placement);
+
+	pin_count = READ_ONCE(bo->tbo.pin_count);
+	if (pin_count)
+		seq_printf(m, " pin count %d", pin_count);
+
+	dma_buf = READ_ONCE(bo->tbo.base.dma_buf);
+	attachment = READ_ONCE(bo->tbo.base.import_attach);
+
+	if (attachment)
+		seq_printf(m, " imported from ino:%lu", file_inode(dma_buf->file)->i_ino);
+	else if (dma_buf)
+		seq_printf(m, " exported as ino:%lu", file_inode(dma_buf->file)->i_ino);
+
+	amdgpu_bo_print_flag(m, bo, CPU_ACCESS_REQUIRED);
+	amdgpu_bo_print_flag(m, bo, NO_CPU_ACCESS);
+	amdgpu_bo_print_flag(m, bo, CPU_GTT_USWC);
+	amdgpu_bo_print_flag(m, bo, VRAM_CLEARED);
+	amdgpu_bo_print_flag(m, bo, VRAM_CONTIGUOUS);
+	amdgpu_bo_print_flag(m, bo, VM_ALWAYS_VALID);
+	amdgpu_bo_print_flag(m, bo, EXPLICIT_SYNC);
+	/* Add the gem obj resv fence dump*/
+	if (dma_resv_trylock(bo->tbo.base.resv)) {
+		dma_resv_describe(bo->tbo.base.resv, m);
+		dma_resv_unlock(bo->tbo.base.resv);
+	}
+	seq_puts(m, "\n");
+
+	return size;
+}
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 382485115b06..52c2d1731aab 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -30,8 +30,120 @@
 
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
+#include "amdgpu_res_cursor.h"
+
+#ifdef CONFIG_MMU_NOTIFIER
+#include <linux/mmu_notifier.h>
+#endif
 
 #define AMDGPU_BO_INVALID_OFFSET	LONG_MAX
+#define AMDGPU_BO_MAX_PLACEMENTS	3
+
+/* BO flag to indicate a KFD userptr BO */
+#define AMDGPU_AMDKFD_CREATE_USERPTR_BO	(1ULL << 63)
+
+#define to_amdgpu_bo_user(abo) container_of((abo), struct amdgpu_bo_user, bo)
+#define to_amdgpu_bo_vm(abo) container_of((abo), struct amdgpu_bo_vm, bo)
+
+struct amdgpu_bo_param {
+	unsigned long			size;
+	int				byte_align;
+	u32				bo_ptr_size;
+	u32				domain;
+	u32				preferred_domain;
+	u64				flags;
+	enum ttm_bo_type		type;
+	bool				no_wait_gpu;
+	struct dma_resv			*resv;
+	void				(*destroy)(struct ttm_buffer_object *bo);
+	/* xcp partition number plus 1, 0 means any partition */
+	int8_t				xcp_id_plus1;
+};
+
+/* bo virtual addresses in a vm */
+struct amdgpu_bo_va_mapping {
+	struct amdgpu_bo_va		*bo_va;
+	struct list_head		list;
+	struct rb_node			rb;
+	uint64_t			start;
+	uint64_t			last;
+	uint64_t			__subtree_last;
+	uint64_t			offset;
+	uint32_t			flags;
+};
+
+/* User space allocated BO in a VM */
+struct amdgpu_bo_va {
+	struct amdgpu_vm_bo_base	base;
+
+	/* protected by bo being reserved */
+	unsigned			ref_count;
+
+	/* all other members protected by the VM PD being reserved */
+	struct dma_fence	        *last_pt_update;
+
+	/* mappings for this bo_va */
+	struct list_head		invalids;
+	struct list_head		valids;
+
+	/* If the mappings are cleared or filled */
+	bool				cleared;
+
+	bool				is_xgmi;
+
+	/*
+	 * protected by vm reservation lock
+	 * if non-zero, cannot unmap from GPU because user queues may still access it
+	 */
+	unsigned int			queue_refcount;
+	atomic_t			userq_va_mapped;
+};
+
+struct amdgpu_bo {
+	/* Protected by tbo.reserved */
+	u32				preferred_domains;
+	u32				allowed_domains;
+	struct ttm_place		placements[AMDGPU_BO_MAX_PLACEMENTS];
+	struct ttm_placement		placement;
+	struct ttm_buffer_object	tbo;
+	struct ttm_bo_kmap_obj		kmap;
+	u64				flags;
+	/* per VM structure for page tables and with virtual addresses */
+	struct amdgpu_vm_bo_base	*vm_bo;
+	/* Constant after initialization */
+	struct amdgpu_bo		*parent;
+
+#ifdef CONFIG_MMU_NOTIFIER
+	struct mmu_interval_notifier	notifier;
+#endif
+	struct kgd_mem                  *kfd_bo;
+
+	/*
+	 * For GPUs with spatial partitioning, xcp partition number, -1 means
+	 * any partition. For other ASICs without spatial partition, always 0
+	 * for memory accounting.
+	 */
+	int8_t				xcp_id;
+};
+
+struct amdgpu_bo_user {
+	struct amdgpu_bo		bo;
+	u64				tiling_flags;
+	u64				metadata_flags;
+	void				*metadata;
+	u32				metadata_size;
+
+};
+
+struct amdgpu_bo_vm {
+	struct amdgpu_bo		bo;
+	struct amdgpu_vm_bo_base        entries[];
+};
+
+static inline struct amdgpu_bo *ttm_to_amdgpu_bo(struct ttm_buffer_object *tbo)
+{
+	return container_of(tbo, struct amdgpu_bo, tbo);
+}
 
 /**
  * amdgpu_mem_type_to_domain - return domain corresponding to mem_type
@@ -54,6 +166,10 @@ static inline unsigned amdgpu_mem_type_to_domain(u32 mem_type)
 		return AMDGPU_GEM_DOMAIN_GWS;
 	case AMDGPU_PL_OA:
 		return AMDGPU_GEM_DOMAIN_OA;
+	case AMDGPU_PL_DOORBELL:
+		return AMDGPU_GEM_DOMAIN_DOORBELL;
+	case AMDGPU_PL_MMIO_REMAP:
+		return AMDGPU_GEM_DOMAIN_MMIO_REMAP;
 	default:
 		break;
 	}
@@ -90,17 +206,17 @@ static inline void amdgpu_bo_unreserve(struct amdgpu_bo *bo)
 
 static inline unsigned long amdgpu_bo_size(struct amdgpu_bo *bo)
 {
-	return bo->tbo.num_pages << PAGE_SHIFT;
+	return bo->tbo.base.size;
 }
 
 static inline unsigned amdgpu_bo_ngpu_pages(struct amdgpu_bo *bo)
 {
-	return (bo->tbo.num_pages << PAGE_SHIFT) / AMDGPU_GPU_PAGE_SIZE;
+	return bo->tbo.base.size / AMDGPU_GPU_PAGE_SIZE;
 }
 
 static inline unsigned amdgpu_bo_gpu_page_alignment(struct amdgpu_bo *bo)
 {
-	return (bo->tbo.mem.page_alignment << PAGE_SHIFT) / AMDGPU_GPU_PAGE_SIZE;
+	return (bo->tbo.page_alignment << PAGE_SHIFT) / AMDGPU_GPU_PAGE_SIZE;
 }
 
 /**
@@ -111,51 +227,67 @@ static inline unsigned amdgpu_bo_gpu_page_alignment(struct amdgpu_bo *bo)
  */
 static inline u64 amdgpu_bo_mmap_offset(struct amdgpu_bo *bo)
 {
-	return drm_vma_node_offset_addr(&bo->tbo.vma_node);
+	return drm_vma_node_offset_addr(&bo->tbo.base.vma_node);
 }
 
 /**
- * amdgpu_bo_gpu_accessible - return whether the bo is currently in memory that
- * is accessible to the GPU.
+ * amdgpu_bo_explicit_sync - return whether the bo is explicitly synced
  */
-static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
+static inline bool amdgpu_bo_explicit_sync(struct amdgpu_bo *bo)
 {
-	return bo->tbo.mem.mem_type != TTM_PL_SYSTEM;
+	return bo->flags & AMDGPU_GEM_CREATE_EXPLICIT_SYNC;
 }
 
+/**
+ * amdgpu_bo_encrypted - test if the BO is encrypted
+ * @bo: pointer to a buffer object
+ *
+ * Return true if the buffer object is encrypted, false otherwise.
+ */
+static inline bool amdgpu_bo_encrypted(struct amdgpu_bo *bo)
+{
+	return bo->flags & AMDGPU_GEM_CREATE_ENCRYPTED;
+}
+
+bool amdgpu_bo_is_amdgpu_bo(struct ttm_buffer_object *bo);
+void amdgpu_bo_placement_from_domain(struct amdgpu_bo *abo, u32 domain);
+
 int amdgpu_bo_create(struct amdgpu_device *adev,
-			    unsigned long size, int byte_align,
-			    bool kernel, u32 domain, u64 flags,
-			    struct sg_table *sg,
-			    struct reservation_object *resv,
-			    struct amdgpu_bo **bo_ptr);
-int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
-				unsigned long size, int byte_align,
-				bool kernel, u32 domain, u64 flags,
-				struct sg_table *sg,
-				struct ttm_placement *placement,
-			        struct reservation_object *resv,
-				struct amdgpu_bo **bo_ptr);
+		     struct amdgpu_bo_param *bp,
+		     struct amdgpu_bo **bo_ptr);
+int amdgpu_bo_create_reserved(struct amdgpu_device *adev,
+			      unsigned long size, int align,
+			      u32 domain, struct amdgpu_bo **bo_ptr,
+			      u64 *gpu_addr, void **cpu_addr);
 int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
 			    unsigned long size, int align,
 			    u32 domain, struct amdgpu_bo **bo_ptr,
 			    u64 *gpu_addr, void **cpu_addr);
+int amdgpu_bo_create_isp_user(struct amdgpu_device *adev,
+			   struct dma_buf *dbuf, u32 domain,
+			   struct amdgpu_bo **bo,
+			   u64 *gpu_addr);
+int amdgpu_bo_create_kernel_at(struct amdgpu_device *adev,
+			       uint64_t offset, uint64_t size,
+			       struct amdgpu_bo **bo_ptr, void **cpu_addr);
+int amdgpu_bo_create_user(struct amdgpu_device *adev,
+			  struct amdgpu_bo_param *bp,
+			  struct amdgpu_bo_user **ubo_ptr);
+int amdgpu_bo_create_vm(struct amdgpu_device *adev,
+			struct amdgpu_bo_param *bp,
+			struct amdgpu_bo_vm **ubo_ptr);
 void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
 			   void **cpu_addr);
+void amdgpu_bo_free_isp_user(struct amdgpu_bo *bo);
 int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr);
+void *amdgpu_bo_kptr(struct amdgpu_bo *bo);
 void amdgpu_bo_kunmap(struct amdgpu_bo *bo);
 struct amdgpu_bo *amdgpu_bo_ref(struct amdgpu_bo *bo);
 void amdgpu_bo_unref(struct amdgpu_bo **bo);
-int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain, u64 *gpu_addr);
-int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
-			     u64 min_offset, u64 max_offset,
-			     u64 *gpu_addr);
-int amdgpu_bo_unpin(struct amdgpu_bo *bo);
-int amdgpu_bo_evict_vram(struct amdgpu_device *adev);
+int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain);
+void amdgpu_bo_unpin(struct amdgpu_bo *bo);
 int amdgpu_bo_init(struct amdgpu_device *adev);
 void amdgpu_bo_fini(struct amdgpu_device *adev);
-int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
-				struct vm_area_struct *vma);
 int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
 void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64 *tiling_flags);
 int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
@@ -165,37 +297,41 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
 			   uint64_t *flags);
 void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
 			   bool evict,
-			   struct ttm_mem_reg *new_mem);
-int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
+			   struct ttm_resource *new_mem);
+void amdgpu_bo_release_notify(struct ttm_buffer_object *bo);
+vm_fault_t amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
 void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
 		     bool shared);
+int amdgpu_bo_sync_wait_resv(struct amdgpu_device *adev, struct dma_resv *resv,
+			     enum amdgpu_sync_mode sync_mode, void *owner,
+			     bool intr);
+int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr);
 u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
-int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
-			       struct amdgpu_ring *ring,
-			       struct amdgpu_bo *bo,
-			       struct reservation_object *resv,
-			       struct dma_fence **fence, bool direct);
-int amdgpu_bo_validate(struct amdgpu_bo *bo);
-int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
-				  struct amdgpu_ring *ring,
-				  struct amdgpu_bo *bo,
-				  struct reservation_object *resv,
-				  struct dma_fence **fence,
-				  bool direct);
-
+u64 amdgpu_bo_fb_aper_addr(struct amdgpu_bo *bo);
+u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
+uint32_t amdgpu_bo_mem_stats_placement(struct amdgpu_bo *bo);
+uint32_t amdgpu_bo_get_preferred_domain(struct amdgpu_device *adev,
+					    uint32_t domain);
 
 /*
  * sub allocation
  */
+static inline struct amdgpu_sa_manager *
+to_amdgpu_sa_manager(struct drm_suballoc_manager *manager)
+{
+	return container_of(manager, struct amdgpu_sa_manager, base);
+}
 
-static inline uint64_t amdgpu_sa_bo_gpu_addr(struct amdgpu_sa_bo *sa_bo)
+static inline uint64_t amdgpu_sa_bo_gpu_addr(struct drm_suballoc *sa_bo)
 {
-	return sa_bo->manager->gpu_addr + sa_bo->soffset;
+	return to_amdgpu_sa_manager(sa_bo->manager)->gpu_addr +
+		drm_suballoc_soffset(sa_bo);
 }
 
-static inline void * amdgpu_sa_bo_cpu_addr(struct amdgpu_sa_bo *sa_bo)
+static inline void *amdgpu_sa_bo_cpu_addr(struct drm_suballoc *sa_bo)
 {
-	return sa_bo->manager->cpu_ptr + sa_bo->soffset;
+	return to_amdgpu_sa_manager(sa_bo->manager)->cpu_ptr +
+		drm_suballoc_soffset(sa_bo);
 }
 
 int amdgpu_sa_bo_manager_init(struct amdgpu_device *adev,
@@ -205,18 +341,19 @@ void amdgpu_sa_bo_manager_fini(struct amdgpu_device *adev,
 				      struct amdgpu_sa_manager *sa_manager);
 int amdgpu_sa_bo_manager_start(struct amdgpu_device *adev,
 				      struct amdgpu_sa_manager *sa_manager);
-int amdgpu_sa_bo_manager_suspend(struct amdgpu_device *adev,
-					struct amdgpu_sa_manager *sa_manager);
 int amdgpu_sa_bo_new(struct amdgpu_sa_manager *sa_manager,
-		     struct amdgpu_sa_bo **sa_bo,
-		     unsigned size, unsigned align);
-void amdgpu_sa_bo_free(struct amdgpu_device *adev,
-			      struct amdgpu_sa_bo **sa_bo,
-			      struct dma_fence *fence);
+		     struct drm_suballoc **sa_bo,
+		     unsigned int size);
+void amdgpu_sa_bo_free(struct drm_suballoc **sa_bo,
+		       struct dma_fence *fence);
 #if defined(CONFIG_DEBUG_FS)
 void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager,
 					 struct seq_file *m);
+u64 amdgpu_bo_print_info(int id, struct amdgpu_bo *bo, struct seq_file *m);
 #endif
+void amdgpu_debugfs_sa_init(struct amdgpu_device *adev);
+
+bool amdgpu_bo_support_uswc(u64 bo_flags);
 
 
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.c
index 8e67c1210d7c..675aa138ea11 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.c
@@ -20,7 +20,7 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  *
  */
-#include <drm/drmP.h>
+
 #include <drm/amdgpu_drm.h>
 #include "amdgpu.h"
 #include "atom.h"
@@ -69,6 +69,7 @@ static void amdgpu_pll_reduce_ratio(unsigned *nom, unsigned *den,
 /**
  * amdgpu_pll_get_fb_ref_div - feedback and ref divider calculation
  *
+ * @adev: amdgpu_device pointer
  * @nom: nominator
  * @den: denominator
  * @post_div: post divider
@@ -80,15 +81,20 @@ static void amdgpu_pll_reduce_ratio(unsigned *nom, unsigned *den,
  * Calculate feedback and reference divider for a given post divider. Makes
  * sure we stay within the limits.
  */
-static void amdgpu_pll_get_fb_ref_div(unsigned nom, unsigned den, unsigned post_div,
-				      unsigned fb_div_max, unsigned ref_div_max,
-				      unsigned *fb_div, unsigned *ref_div)
+static void amdgpu_pll_get_fb_ref_div(struct amdgpu_device *adev, unsigned int nom,
+				      unsigned int den, unsigned int post_div,
+				      unsigned int fb_div_max, unsigned int ref_div_max,
+				      unsigned int *fb_div, unsigned int *ref_div)
 {
+
 	/* limit reference * post divider to a maximum */
-	ref_div_max = min(128 / post_div, ref_div_max);
+	if (adev->family == AMDGPU_FAMILY_SI)
+		ref_div_max = min(100 / post_div, ref_div_max);
+	else
+		ref_div_max = min(128 / post_div, ref_div_max);
 
 	/* get matching reference and feedback divider */
-	*ref_div = min(max(DIV_ROUND_CLOSEST(den, post_div), 1u), ref_div_max);
+	*ref_div = clamp(DIV_ROUND_CLOSEST(den, post_div), 1u, ref_div_max);
 	*fb_div = DIV_ROUND_CLOSEST(nom * *ref_div * post_div, den);
 
 	/* limit fb divider to its maximum */
@@ -101,17 +107,20 @@ static void amdgpu_pll_get_fb_ref_div(unsigned nom, unsigned den, unsigned post_
 /**
  * amdgpu_pll_compute - compute PLL paramaters
  *
+ * @adev: amdgpu_device pointer
  * @pll: information about the PLL
+ * @freq: requested frequency
  * @dot_clock_p: resulting pixel clock
- * fb_div_p: resulting feedback divider
- * frac_fb_div_p: fractional part of the feedback divider
- * ref_div_p: resulting reference divider
- * post_div_p: resulting reference divider
+ * @fb_div_p: resulting feedback divider
+ * @frac_fb_div_p: fractional part of the feedback divider
+ * @ref_div_p: resulting reference divider
+ * @post_div_p: resulting reference divider
  *
  * Try to calculate the PLL parameters to generate the given frequency:
  * dot_clock = (ref_freq * feedback_div) / (ref_div * post_div)
  */
-void amdgpu_pll_compute(struct amdgpu_pll *pll,
+void amdgpu_pll_compute(struct amdgpu_device *adev,
+			struct amdgpu_pll *pll,
 			u32 freq,
 			u32 *dot_clock_p,
 			u32 *fb_div_p,
@@ -198,7 +207,7 @@ void amdgpu_pll_compute(struct amdgpu_pll *pll,
 
 	for (post_div = post_div_min; post_div <= post_div_max; ++post_div) {
 		unsigned diff;
-		amdgpu_pll_get_fb_ref_div(nom, den, post_div, fb_div_max,
+		amdgpu_pll_get_fb_ref_div(adev, nom, den, post_div, fb_div_max,
 					  ref_div_max, &fb_div, &ref_div);
 		diff = abs(target_clock - (pll->reference_freq * fb_div) /
 			(ref_div * post_div));
@@ -213,7 +222,7 @@ void amdgpu_pll_compute(struct amdgpu_pll *pll,
 	post_div = post_div_best;
 
 	/* get the feedback and reference divider for the optimal value */
-	amdgpu_pll_get_fb_ref_div(nom, den, post_div, fb_div_max, ref_div_max,
+	amdgpu_pll_get_fb_ref_div(adev, nom, den, post_div, fb_div_max, ref_div_max,
 				  &fb_div, &ref_div);
 
 	/* reduce the numbers to a simpler ratio once more */
@@ -308,7 +317,6 @@ int amdgpu_pll_get_shared_dp_ppll(struct drm_crtc *crtc)
  * amdgpu_pll_get_shared_nondp_ppll - return the PPLL used by another non-DP crtc
  *
  * @crtc: drm crtc
- * @encoder: drm encoder
  *
  * Returns the PPLL (Pixel PLL) used by another non-DP crtc/encoder which can
  * be shared (i.e., same clock).
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.h
index db6136f68b82..44a583d6c9b4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pll.h
@@ -24,7 +24,8 @@
 #ifndef __AMDGPU_PLL_H__
 #define __AMDGPU_PLL_H__
 
-void amdgpu_pll_compute(struct amdgpu_pll *pll,
+void amdgpu_pll_compute(struct amdgpu_device *adev,
+			 struct amdgpu_pll *pll,
 			 u32 freq,
 			 u32 *dot_clock_p,
 			 u32 *fb_div_p,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c
deleted file mode 100644
index 7df503aedb69..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c
+++ /dev/null
@@ -1,1659 +0,0 @@
-/*
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors: Rafał Miłecki <zajec5@gmail.com>
- *          Alex Deucher <alexdeucher@gmail.com>
- */
-#include <drm/drmP.h>
-#include "amdgpu.h"
-#include "amdgpu_drv.h"
-#include "amdgpu_pm.h"
-#include "amdgpu_dpm.h"
-#include "atom.h"
-#include <linux/power_supply.h>
-#include <linux/hwmon.h>
-#include <linux/hwmon-sysfs.h>
-
-#include "amd_powerplay.h"
-
-static int amdgpu_debugfs_pm_init(struct amdgpu_device *adev);
-
-static const struct cg_flag_name clocks[] = {
-	{AMD_CG_SUPPORT_GFX_MGCG, "Graphics Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_GFX_MGLS, "Graphics Medium Grain memory Light Sleep"},
-	{AMD_CG_SUPPORT_GFX_CGCG, "Graphics Coarse Grain Clock Gating"},
-	{AMD_CG_SUPPORT_GFX_CGLS, "Graphics Coarse Grain memory Light Sleep"},
-	{AMD_CG_SUPPORT_GFX_CGTS, "Graphics Coarse Grain Tree Shader Clock Gating"},
-	{AMD_CG_SUPPORT_GFX_CGTS_LS, "Graphics Coarse Grain Tree Shader Light Sleep"},
-	{AMD_CG_SUPPORT_GFX_CP_LS, "Graphics Command Processor Light Sleep"},
-	{AMD_CG_SUPPORT_GFX_RLC_LS, "Graphics Run List Controller Light Sleep"},
-	{AMD_CG_SUPPORT_GFX_3D_CGCG, "Graphics 3D Coarse Grain Clock Gating"},
-	{AMD_CG_SUPPORT_GFX_3D_CGLS, "Graphics 3D Coarse Grain memory Light Sleep"},
-	{AMD_CG_SUPPORT_MC_LS, "Memory Controller Light Sleep"},
-	{AMD_CG_SUPPORT_MC_MGCG, "Memory Controller Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_SDMA_LS, "System Direct Memory Access Light Sleep"},
-	{AMD_CG_SUPPORT_SDMA_MGCG, "System Direct Memory Access Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_BIF_MGCG, "Bus Interface Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_BIF_LS, "Bus Interface Light Sleep"},
-	{AMD_CG_SUPPORT_UVD_MGCG, "Unified Video Decoder Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_VCE_MGCG, "Video Compression Engine Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_HDP_LS, "Host Data Path Light Sleep"},
-	{AMD_CG_SUPPORT_HDP_MGCG, "Host Data Path Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_DRM_MGCG, "Digital Right Management Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_DRM_LS, "Digital Right Management Light Sleep"},
-	{AMD_CG_SUPPORT_ROM_MGCG, "Rom Medium Grain Clock Gating"},
-	{AMD_CG_SUPPORT_DF_MGCG, "Data Fabric Medium Grain Clock Gating"},
-	{0, NULL},
-};
-
-void amdgpu_pm_acpi_event_handler(struct amdgpu_device *adev)
-{
-	if (adev->pp_enabled)
-		/* TODO */
-		return;
-
-	if (adev->pm.dpm_enabled) {
-		mutex_lock(&adev->pm.mutex);
-		if (power_supply_is_system_supplied() > 0)
-			adev->pm.dpm.ac_power = true;
-		else
-			adev->pm.dpm.ac_power = false;
-		if (adev->pm.funcs->enable_bapm)
-			amdgpu_dpm_enable_bapm(adev, adev->pm.dpm.ac_power);
-		mutex_unlock(&adev->pm.mutex);
-	}
-}
-
-static ssize_t amdgpu_get_dpm_state(struct device *dev,
-				    struct device_attribute *attr,
-				    char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	enum amd_pm_state_type pm;
-
-	if (adev->pp_enabled) {
-		pm = amdgpu_dpm_get_current_power_state(adev);
-	} else
-		pm = adev->pm.dpm.user_state;
-
-	return snprintf(buf, PAGE_SIZE, "%s\n",
-			(pm == POWER_STATE_TYPE_BATTERY) ? "battery" :
-			(pm == POWER_STATE_TYPE_BALANCED) ? "balanced" : "performance");
-}
-
-static ssize_t amdgpu_set_dpm_state(struct device *dev,
-				    struct device_attribute *attr,
-				    const char *buf,
-				    size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	enum amd_pm_state_type  state;
-
-	if (strncmp("battery", buf, strlen("battery")) == 0)
-		state = POWER_STATE_TYPE_BATTERY;
-	else if (strncmp("balanced", buf, strlen("balanced")) == 0)
-		state = POWER_STATE_TYPE_BALANCED;
-	else if (strncmp("performance", buf, strlen("performance")) == 0)
-		state = POWER_STATE_TYPE_PERFORMANCE;
-	else {
-		count = -EINVAL;
-		goto fail;
-	}
-
-	if (adev->pp_enabled) {
-		amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_ENABLE_USER_STATE, &state, NULL);
-	} else {
-		mutex_lock(&adev->pm.mutex);
-		adev->pm.dpm.user_state = state;
-		mutex_unlock(&adev->pm.mutex);
-
-		/* Can't set dpm state when the card is off */
-		if (!(adev->flags & AMD_IS_PX) ||
-		    (ddev->switch_power_state == DRM_SWITCH_POWER_ON))
-			amdgpu_pm_compute_clocks(adev);
-	}
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_dpm_forced_performance_level(struct device *dev,
-						struct device_attribute *attr,
-								char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	enum amd_dpm_forced_level level;
-
-	if  ((adev->flags & AMD_IS_PX) &&
-	     (ddev->switch_power_state != DRM_SWITCH_POWER_ON))
-		return snprintf(buf, PAGE_SIZE, "off\n");
-
-	level = amdgpu_dpm_get_performance_level(adev);
-	return snprintf(buf, PAGE_SIZE, "%s\n",
-			(level == AMD_DPM_FORCED_LEVEL_AUTO) ? "auto" :
-			(level == AMD_DPM_FORCED_LEVEL_LOW) ? "low" :
-			(level == AMD_DPM_FORCED_LEVEL_HIGH) ? "high" :
-			(level == AMD_DPM_FORCED_LEVEL_MANUAL) ? "manual" :
-			(level == AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD) ? "profile_standard" :
-			(level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK) ? "profile_min_sclk" :
-			(level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK) ? "profile_min_mclk" :
-			(level == AMD_DPM_FORCED_LEVEL_PROFILE_PEAK) ? "profile_peak" :
-			"unknown");
-}
-
-static ssize_t amdgpu_set_dpm_forced_performance_level(struct device *dev,
-						       struct device_attribute *attr,
-						       const char *buf,
-						       size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	enum amd_dpm_forced_level level;
-	enum amd_dpm_forced_level current_level;
-	int ret = 0;
-
-	/* Can't force performance level when the card is off */
-	if  ((adev->flags & AMD_IS_PX) &&
-	     (ddev->switch_power_state != DRM_SWITCH_POWER_ON))
-		return -EINVAL;
-
-	current_level = amdgpu_dpm_get_performance_level(adev);
-
-	if (strncmp("low", buf, strlen("low")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_LOW;
-	} else if (strncmp("high", buf, strlen("high")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_HIGH;
-	} else if (strncmp("auto", buf, strlen("auto")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_AUTO;
-	} else if (strncmp("manual", buf, strlen("manual")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_MANUAL;
-	} else if (strncmp("profile_exit", buf, strlen("profile_exit")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_PROFILE_EXIT;
-	} else if (strncmp("profile_standard", buf, strlen("profile_standard")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD;
-	} else if (strncmp("profile_min_sclk", buf, strlen("profile_min_sclk")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK;
-	} else if (strncmp("profile_min_mclk", buf, strlen("profile_min_mclk")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK;
-	} else if (strncmp("profile_peak", buf, strlen("profile_peak")) == 0) {
-		level = AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
-	}  else {
-		count = -EINVAL;
-		goto fail;
-	}
-
-	if (current_level == level)
-		return count;
-
-	if (adev->pp_enabled)
-		amdgpu_dpm_force_performance_level(adev, level);
-	else {
-		mutex_lock(&adev->pm.mutex);
-		if (adev->pm.dpm.thermal_active) {
-			count = -EINVAL;
-			mutex_unlock(&adev->pm.mutex);
-			goto fail;
-		}
-		ret = amdgpu_dpm_force_performance_level(adev, level);
-		if (ret)
-			count = -EINVAL;
-		else
-			adev->pm.dpm.forced_level = level;
-		mutex_unlock(&adev->pm.mutex);
-	}
-
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_num_states(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	struct pp_states_info data;
-	int i, buf_len;
-
-	if (adev->pp_enabled)
-		amdgpu_dpm_get_pp_num_states(adev, &data);
-
-	buf_len = snprintf(buf, PAGE_SIZE, "states: %d\n", data.nums);
-	for (i = 0; i < data.nums; i++)
-		buf_len += snprintf(buf + buf_len, PAGE_SIZE, "%d %s\n", i,
-				(data.states[i] == POWER_STATE_TYPE_INTERNAL_BOOT) ? "boot" :
-				(data.states[i] == POWER_STATE_TYPE_BATTERY) ? "battery" :
-				(data.states[i] == POWER_STATE_TYPE_BALANCED) ? "balanced" :
-				(data.states[i] == POWER_STATE_TYPE_PERFORMANCE) ? "performance" : "default");
-
-	return buf_len;
-}
-
-static ssize_t amdgpu_get_pp_cur_state(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	struct pp_states_info data;
-	enum amd_pm_state_type pm = 0;
-	int i = 0;
-
-	if (adev->pp_enabled) {
-
-		pm = amdgpu_dpm_get_current_power_state(adev);
-		amdgpu_dpm_get_pp_num_states(adev, &data);
-
-		for (i = 0; i < data.nums; i++) {
-			if (pm == data.states[i])
-				break;
-		}
-
-		if (i == data.nums)
-			i = -EINVAL;
-	}
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", i);
-}
-
-static ssize_t amdgpu_get_pp_force_state(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	struct pp_states_info data;
-	enum amd_pm_state_type pm = 0;
-	int i;
-
-	if (adev->pp_force_state_enabled && adev->pp_enabled) {
-		pm = amdgpu_dpm_get_current_power_state(adev);
-		amdgpu_dpm_get_pp_num_states(adev, &data);
-
-		for (i = 0; i < data.nums; i++) {
-			if (pm == data.states[i])
-				break;
-		}
-
-		if (i == data.nums)
-			i = -EINVAL;
-
-		return snprintf(buf, PAGE_SIZE, "%d\n", i);
-
-	} else
-		return snprintf(buf, PAGE_SIZE, "\n");
-}
-
-static ssize_t amdgpu_set_pp_force_state(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	enum amd_pm_state_type state = 0;
-	unsigned long idx;
-	int ret;
-
-	if (strlen(buf) == 1)
-		adev->pp_force_state_enabled = false;
-	else if (adev->pp_enabled) {
-		struct pp_states_info data;
-
-		ret = kstrtoul(buf, 0, &idx);
-		if (ret || idx >= ARRAY_SIZE(data.states)) {
-			count = -EINVAL;
-			goto fail;
-		}
-
-		amdgpu_dpm_get_pp_num_states(adev, &data);
-		state = data.states[idx];
-		/* only set user selected power states */
-		if (state != POWER_STATE_TYPE_INTERNAL_BOOT &&
-		    state != POWER_STATE_TYPE_DEFAULT) {
-			amdgpu_dpm_dispatch_task(adev,
-					AMD_PP_EVENT_ENABLE_USER_STATE, &state, NULL);
-			adev->pp_force_state_enabled = true;
-		}
-	}
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_table(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	char *table = NULL;
-	int size;
-
-	if (adev->pp_enabled)
-		size = amdgpu_dpm_get_pp_table(adev, &table);
-	else
-		return 0;
-
-	if (size >= PAGE_SIZE)
-		size = PAGE_SIZE - 1;
-
-	memcpy(buf, table, size);
-
-	return size;
-}
-
-static ssize_t amdgpu_set_pp_table(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-
-	if (adev->pp_enabled)
-		amdgpu_dpm_set_pp_table(adev, buf, count);
-
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_dpm_sclk(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	ssize_t size = 0;
-
-	if (adev->pp_enabled)
-		size = amdgpu_dpm_print_clock_levels(adev, PP_SCLK, buf);
-	else if (adev->pm.funcs->print_clock_levels)
-		size = adev->pm.funcs->print_clock_levels(adev, PP_SCLK, buf);
-
-	return size;
-}
-
-static ssize_t amdgpu_set_pp_dpm_sclk(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	int ret;
-	long level;
-	uint32_t i, mask = 0;
-	char sub_str[2];
-
-	for (i = 0; i < strlen(buf); i++) {
-		if (*(buf + i) == '\n')
-			continue;
-		sub_str[0] = *(buf + i);
-		sub_str[1] = '\0';
-		ret = kstrtol(sub_str, 0, &level);
-
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		mask |= 1 << level;
-	}
-
-	if (adev->pp_enabled)
-		amdgpu_dpm_force_clock_level(adev, PP_SCLK, mask);
-	else if (adev->pm.funcs->force_clock_level)
-		adev->pm.funcs->force_clock_level(adev, PP_SCLK, mask);
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_dpm_mclk(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	ssize_t size = 0;
-
-	if (adev->pp_enabled)
-		size = amdgpu_dpm_print_clock_levels(adev, PP_MCLK, buf);
-	else if (adev->pm.funcs->print_clock_levels)
-		size = adev->pm.funcs->print_clock_levels(adev, PP_MCLK, buf);
-
-	return size;
-}
-
-static ssize_t amdgpu_set_pp_dpm_mclk(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	int ret;
-	long level;
-	uint32_t i, mask = 0;
-	char sub_str[2];
-
-	for (i = 0; i < strlen(buf); i++) {
-		if (*(buf + i) == '\n')
-			continue;
-		sub_str[0] = *(buf + i);
-		sub_str[1] = '\0';
-		ret = kstrtol(sub_str, 0, &level);
-
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		mask |= 1 << level;
-	}
-
-	if (adev->pp_enabled)
-		amdgpu_dpm_force_clock_level(adev, PP_MCLK, mask);
-	else if (adev->pm.funcs->force_clock_level)
-		adev->pm.funcs->force_clock_level(adev, PP_MCLK, mask);
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_dpm_pcie(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	ssize_t size = 0;
-
-	if (adev->pp_enabled)
-		size = amdgpu_dpm_print_clock_levels(adev, PP_PCIE, buf);
-	else if (adev->pm.funcs->print_clock_levels)
-		size = adev->pm.funcs->print_clock_levels(adev, PP_PCIE, buf);
-
-	return size;
-}
-
-static ssize_t amdgpu_set_pp_dpm_pcie(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	int ret;
-	long level;
-	uint32_t i, mask = 0;
-	char sub_str[2];
-
-	for (i = 0; i < strlen(buf); i++) {
-		if (*(buf + i) == '\n')
-			continue;
-		sub_str[0] = *(buf + i);
-		sub_str[1] = '\0';
-		ret = kstrtol(sub_str, 0, &level);
-
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		mask |= 1 << level;
-	}
-
-	if (adev->pp_enabled)
-		amdgpu_dpm_force_clock_level(adev, PP_PCIE, mask);
-	else if (adev->pm.funcs->force_clock_level)
-		adev->pm.funcs->force_clock_level(adev, PP_PCIE, mask);
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_sclk_od(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	uint32_t value = 0;
-
-	if (adev->pp_enabled)
-		value = amdgpu_dpm_get_sclk_od(adev);
-	else if (adev->pm.funcs->get_sclk_od)
-		value = adev->pm.funcs->get_sclk_od(adev);
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", value);
-}
-
-static ssize_t amdgpu_set_pp_sclk_od(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	int ret;
-	long int value;
-
-	ret = kstrtol(buf, 0, &value);
-
-	if (ret) {
-		count = -EINVAL;
-		goto fail;
-	}
-
-	if (adev->pp_enabled) {
-		amdgpu_dpm_set_sclk_od(adev, (uint32_t)value);
-		amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_READJUST_POWER_STATE, NULL, NULL);
-	} else if (adev->pm.funcs->set_sclk_od) {
-		adev->pm.funcs->set_sclk_od(adev, (uint32_t)value);
-		adev->pm.dpm.current_ps = adev->pm.dpm.boot_ps;
-		amdgpu_pm_compute_clocks(adev);
-	}
-
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_mclk_od(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	uint32_t value = 0;
-
-	if (adev->pp_enabled)
-		value = amdgpu_dpm_get_mclk_od(adev);
-	else if (adev->pm.funcs->get_mclk_od)
-		value = adev->pm.funcs->get_mclk_od(adev);
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", value);
-}
-
-static ssize_t amdgpu_set_pp_mclk_od(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	int ret;
-	long int value;
-
-	ret = kstrtol(buf, 0, &value);
-
-	if (ret) {
-		count = -EINVAL;
-		goto fail;
-	}
-
-	if (adev->pp_enabled) {
-		amdgpu_dpm_set_mclk_od(adev, (uint32_t)value);
-		amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_READJUST_POWER_STATE, NULL, NULL);
-	} else if (adev->pm.funcs->set_mclk_od) {
-		adev->pm.funcs->set_mclk_od(adev, (uint32_t)value);
-		adev->pm.dpm.current_ps = adev->pm.dpm.boot_ps;
-		amdgpu_pm_compute_clocks(adev);
-	}
-
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_get_pp_power_profile(struct device *dev,
-		char *buf, struct amd_pp_profile *query)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	int ret = 0;
-
-	if (adev->pp_enabled)
-		ret = amdgpu_dpm_get_power_profile_state(
-				adev, query);
-	else if (adev->pm.funcs->get_power_profile_state)
-		ret = adev->pm.funcs->get_power_profile_state(
-				adev, query);
-
-	if (ret)
-		return ret;
-
-	return snprintf(buf, PAGE_SIZE,
-			"%d %d %d %d %d\n",
-			query->min_sclk / 100,
-			query->min_mclk / 100,
-			query->activity_threshold,
-			query->up_hyst,
-			query->down_hyst);
-}
-
-static ssize_t amdgpu_get_pp_gfx_power_profile(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct amd_pp_profile query = {0};
-
-	query.type = AMD_PP_GFX_PROFILE;
-
-	return amdgpu_get_pp_power_profile(dev, buf, &query);
-}
-
-static ssize_t amdgpu_get_pp_compute_power_profile(struct device *dev,
-		struct device_attribute *attr,
-		char *buf)
-{
-	struct amd_pp_profile query = {0};
-
-	query.type = AMD_PP_COMPUTE_PROFILE;
-
-	return amdgpu_get_pp_power_profile(dev, buf, &query);
-}
-
-static ssize_t amdgpu_set_pp_power_profile(struct device *dev,
-		const char *buf,
-		size_t count,
-		struct amd_pp_profile *request)
-{
-	struct drm_device *ddev = dev_get_drvdata(dev);
-	struct amdgpu_device *adev = ddev->dev_private;
-	uint32_t loop = 0;
-	char *sub_str, buf_cpy[128], *tmp_str;
-	const char delimiter[3] = {' ', '\n', '\0'};
-	long int value;
-	int ret = 0;
-
-	if (strncmp("reset", buf, strlen("reset")) == 0) {
-		if (adev->pp_enabled)
-			ret = amdgpu_dpm_reset_power_profile_state(
-					adev, request);
-		else if (adev->pm.funcs->reset_power_profile_state)
-			ret = adev->pm.funcs->reset_power_profile_state(
-					adev, request);
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		return count;
-	}
-
-	if (strncmp("set", buf, strlen("set")) == 0) {
-		if (adev->pp_enabled)
-			ret = amdgpu_dpm_set_power_profile_state(
-					adev, request);
-		else if (adev->pm.funcs->set_power_profile_state)
-			ret = adev->pm.funcs->set_power_profile_state(
-					adev, request);
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-		return count;
-	}
-
-	if (count + 1 >= 128) {
-		count = -EINVAL;
-		goto fail;
-	}
-
-	memcpy(buf_cpy, buf, count + 1);
-	tmp_str = buf_cpy;
-
-	while (tmp_str[0]) {
-		sub_str = strsep(&tmp_str, delimiter);
-		ret = kstrtol(sub_str, 0, &value);
-		if (ret) {
-			count = -EINVAL;
-			goto fail;
-		}
-
-		switch (loop) {
-		case 0:
-			/* input unit MHz convert to dpm table unit 10KHz*/
-			request->min_sclk = (uint32_t)value * 100;
-			break;
-		case 1:
-			/* input unit MHz convert to dpm table unit 10KHz*/
-			request->min_mclk = (uint32_t)value * 100;
-			break;
-		case 2:
-			request->activity_threshold = (uint16_t)value;
-			break;
-		case 3:
-			request->up_hyst = (uint8_t)value;
-			break;
-		case 4:
-			request->down_hyst = (uint8_t)value;
-			break;
-		default:
-			break;
-		}
-
-		loop++;
-	}
-
-	if (adev->pp_enabled)
-		ret = amdgpu_dpm_set_power_profile_state(
-				adev, request);
-	else if (adev->pm.funcs->set_power_profile_state)
-		ret = adev->pm.funcs->set_power_profile_state(
-				adev, request);
-
-	if (ret)
-		count = -EINVAL;
-
-fail:
-	return count;
-}
-
-static ssize_t amdgpu_set_pp_gfx_power_profile(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct amd_pp_profile request = {0};
-
-	request.type = AMD_PP_GFX_PROFILE;
-
-	return amdgpu_set_pp_power_profile(dev, buf, count, &request);
-}
-
-static ssize_t amdgpu_set_pp_compute_power_profile(struct device *dev,
-		struct device_attribute *attr,
-		const char *buf,
-		size_t count)
-{
-	struct amd_pp_profile request = {0};
-
-	request.type = AMD_PP_COMPUTE_PROFILE;
-
-	return amdgpu_set_pp_power_profile(dev, buf, count, &request);
-}
-
-static DEVICE_ATTR(power_dpm_state, S_IRUGO | S_IWUSR, amdgpu_get_dpm_state, amdgpu_set_dpm_state);
-static DEVICE_ATTR(power_dpm_force_performance_level, S_IRUGO | S_IWUSR,
-		   amdgpu_get_dpm_forced_performance_level,
-		   amdgpu_set_dpm_forced_performance_level);
-static DEVICE_ATTR(pp_num_states, S_IRUGO, amdgpu_get_pp_num_states, NULL);
-static DEVICE_ATTR(pp_cur_state, S_IRUGO, amdgpu_get_pp_cur_state, NULL);
-static DEVICE_ATTR(pp_force_state, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_force_state,
-		amdgpu_set_pp_force_state);
-static DEVICE_ATTR(pp_table, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_table,
-		amdgpu_set_pp_table);
-static DEVICE_ATTR(pp_dpm_sclk, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_dpm_sclk,
-		amdgpu_set_pp_dpm_sclk);
-static DEVICE_ATTR(pp_dpm_mclk, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_dpm_mclk,
-		amdgpu_set_pp_dpm_mclk);
-static DEVICE_ATTR(pp_dpm_pcie, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_dpm_pcie,
-		amdgpu_set_pp_dpm_pcie);
-static DEVICE_ATTR(pp_sclk_od, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_sclk_od,
-		amdgpu_set_pp_sclk_od);
-static DEVICE_ATTR(pp_mclk_od, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_mclk_od,
-		amdgpu_set_pp_mclk_od);
-static DEVICE_ATTR(pp_gfx_power_profile, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_gfx_power_profile,
-		amdgpu_set_pp_gfx_power_profile);
-static DEVICE_ATTR(pp_compute_power_profile, S_IRUGO | S_IWUSR,
-		amdgpu_get_pp_compute_power_profile,
-		amdgpu_set_pp_compute_power_profile);
-
-static ssize_t amdgpu_hwmon_show_temp(struct device *dev,
-				      struct device_attribute *attr,
-				      char *buf)
-{
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	struct drm_device *ddev = adev->ddev;
-	int temp;
-
-	/* Can't get temperature when the card is off */
-	if  ((adev->flags & AMD_IS_PX) &&
-	     (ddev->switch_power_state != DRM_SWITCH_POWER_ON))
-		return -EINVAL;
-
-	if (!adev->pp_enabled && !adev->pm.funcs->get_temperature)
-		temp = 0;
-	else
-		temp = amdgpu_dpm_get_temperature(adev);
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", temp);
-}
-
-static ssize_t amdgpu_hwmon_show_temp_thresh(struct device *dev,
-					     struct device_attribute *attr,
-					     char *buf)
-{
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	int hyst = to_sensor_dev_attr(attr)->index;
-	int temp;
-
-	if (hyst)
-		temp = adev->pm.dpm.thermal.min_temp;
-	else
-		temp = adev->pm.dpm.thermal.max_temp;
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", temp);
-}
-
-static ssize_t amdgpu_hwmon_get_pwm1_enable(struct device *dev,
-					    struct device_attribute *attr,
-					    char *buf)
-{
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	u32 pwm_mode = 0;
-
-	if (!adev->pp_enabled && !adev->pm.funcs->get_fan_control_mode)
-		return -EINVAL;
-
-	pwm_mode = amdgpu_dpm_get_fan_control_mode(adev);
-
-	return sprintf(buf, "%i\n", pwm_mode);
-}
-
-static ssize_t amdgpu_hwmon_set_pwm1_enable(struct device *dev,
-					    struct device_attribute *attr,
-					    const char *buf,
-					    size_t count)
-{
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	int err;
-	int value;
-
-	if (!adev->pp_enabled && !adev->pm.funcs->set_fan_control_mode)
-		return -EINVAL;
-
-	err = kstrtoint(buf, 10, &value);
-	if (err)
-		return err;
-
-	amdgpu_dpm_set_fan_control_mode(adev, value);
-
-	return count;
-}
-
-static ssize_t amdgpu_hwmon_get_pwm1_min(struct device *dev,
-					 struct device_attribute *attr,
-					 char *buf)
-{
-	return sprintf(buf, "%i\n", 0);
-}
-
-static ssize_t amdgpu_hwmon_get_pwm1_max(struct device *dev,
-					 struct device_attribute *attr,
-					 char *buf)
-{
-	return sprintf(buf, "%i\n", 255);
-}
-
-static ssize_t amdgpu_hwmon_set_pwm1(struct device *dev,
-				     struct device_attribute *attr,
-				     const char *buf, size_t count)
-{
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	int err;
-	u32 value;
-
-	err = kstrtou32(buf, 10, &value);
-	if (err)
-		return err;
-
-	value = (value * 100) / 255;
-
-	err = amdgpu_dpm_set_fan_speed_percent(adev, value);
-	if (err)
-		return err;
-
-	return count;
-}
-
-static ssize_t amdgpu_hwmon_get_pwm1(struct device *dev,
-				     struct device_attribute *attr,
-				     char *buf)
-{
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	int err;
-	u32 speed;
-
-	err = amdgpu_dpm_get_fan_speed_percent(adev, &speed);
-	if (err)
-		return err;
-
-	speed = (speed * 255) / 100;
-
-	return sprintf(buf, "%i\n", speed);
-}
-
-static ssize_t amdgpu_hwmon_get_fan1_input(struct device *dev,
-					   struct device_attribute *attr,
-					   char *buf)
-{
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	int err;
-	u32 speed;
-
-	err = amdgpu_dpm_get_fan_speed_rpm(adev, &speed);
-	if (err)
-		return err;
-
-	return sprintf(buf, "%i\n", speed);
-}
-
-static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO, amdgpu_hwmon_show_temp, NULL, 0);
-static SENSOR_DEVICE_ATTR(temp1_crit, S_IRUGO, amdgpu_hwmon_show_temp_thresh, NULL, 0);
-static SENSOR_DEVICE_ATTR(temp1_crit_hyst, S_IRUGO, amdgpu_hwmon_show_temp_thresh, NULL, 1);
-static SENSOR_DEVICE_ATTR(pwm1, S_IRUGO | S_IWUSR, amdgpu_hwmon_get_pwm1, amdgpu_hwmon_set_pwm1, 0);
-static SENSOR_DEVICE_ATTR(pwm1_enable, S_IRUGO | S_IWUSR, amdgpu_hwmon_get_pwm1_enable, amdgpu_hwmon_set_pwm1_enable, 0);
-static SENSOR_DEVICE_ATTR(pwm1_min, S_IRUGO, amdgpu_hwmon_get_pwm1_min, NULL, 0);
-static SENSOR_DEVICE_ATTR(pwm1_max, S_IRUGO, amdgpu_hwmon_get_pwm1_max, NULL, 0);
-static SENSOR_DEVICE_ATTR(fan1_input, S_IRUGO, amdgpu_hwmon_get_fan1_input, NULL, 0);
-
-static struct attribute *hwmon_attributes[] = {
-	&sensor_dev_attr_temp1_input.dev_attr.attr,
-	&sensor_dev_attr_temp1_crit.dev_attr.attr,
-	&sensor_dev_attr_temp1_crit_hyst.dev_attr.attr,
-	&sensor_dev_attr_pwm1.dev_attr.attr,
-	&sensor_dev_attr_pwm1_enable.dev_attr.attr,
-	&sensor_dev_attr_pwm1_min.dev_attr.attr,
-	&sensor_dev_attr_pwm1_max.dev_attr.attr,
-	&sensor_dev_attr_fan1_input.dev_attr.attr,
-	NULL
-};
-
-static umode_t hwmon_attributes_visible(struct kobject *kobj,
-					struct attribute *attr, int index)
-{
-	struct device *dev = kobj_to_dev(kobj);
-	struct amdgpu_device *adev = dev_get_drvdata(dev);
-	umode_t effective_mode = attr->mode;
-
-	/* Skip limit attributes if DPM is not enabled */
-	if (!adev->pm.dpm_enabled &&
-	    (attr == &sensor_dev_attr_temp1_crit.dev_attr.attr ||
-	     attr == &sensor_dev_attr_temp1_crit_hyst.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1_min.dev_attr.attr))
-		return 0;
-
-	if (adev->pp_enabled)
-		return effective_mode;
-
-	/* Skip fan attributes if fan is not present */
-	if (adev->pm.no_fan &&
-	    (attr == &sensor_dev_attr_pwm1.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1_min.dev_attr.attr))
-		return 0;
-
-	/* mask fan attributes if we have no bindings for this asic to expose */
-	if ((!adev->pm.funcs->get_fan_speed_percent &&
-	     attr == &sensor_dev_attr_pwm1.dev_attr.attr) || /* can't query fan */
-	    (!adev->pm.funcs->get_fan_control_mode &&
-	     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr)) /* can't query state */
-		effective_mode &= ~S_IRUGO;
-
-	if ((!adev->pm.funcs->set_fan_speed_percent &&
-	     attr == &sensor_dev_attr_pwm1.dev_attr.attr) || /* can't manage fan */
-	    (!adev->pm.funcs->set_fan_control_mode &&
-	     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr)) /* can't manage state */
-		effective_mode &= ~S_IWUSR;
-
-	/* hide max/min values if we can't both query and manage the fan */
-	if ((!adev->pm.funcs->set_fan_speed_percent &&
-	     !adev->pm.funcs->get_fan_speed_percent) &&
-	    (attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
-	     attr == &sensor_dev_attr_pwm1_min.dev_attr.attr))
-		return 0;
-
-	/* requires powerplay */
-	if (attr == &sensor_dev_attr_fan1_input.dev_attr.attr)
-		return 0;
-
-	return effective_mode;
-}
-
-static const struct attribute_group hwmon_attrgroup = {
-	.attrs = hwmon_attributes,
-	.is_visible = hwmon_attributes_visible,
-};
-
-static const struct attribute_group *hwmon_groups[] = {
-	&hwmon_attrgroup,
-	NULL
-};
-
-void amdgpu_dpm_thermal_work_handler(struct work_struct *work)
-{
-	struct amdgpu_device *adev =
-		container_of(work, struct amdgpu_device,
-			     pm.dpm.thermal.work);
-	/* switch to the thermal state */
-	enum amd_pm_state_type dpm_state = POWER_STATE_TYPE_INTERNAL_THERMAL;
-
-	if (!adev->pm.dpm_enabled)
-		return;
-
-	if (adev->pm.funcs->get_temperature) {
-		int temp = amdgpu_dpm_get_temperature(adev);
-
-		if (temp < adev->pm.dpm.thermal.min_temp)
-			/* switch back the user state */
-			dpm_state = adev->pm.dpm.user_state;
-	} else {
-		if (adev->pm.dpm.thermal.high_to_low)
-			/* switch back the user state */
-			dpm_state = adev->pm.dpm.user_state;
-	}
-	mutex_lock(&adev->pm.mutex);
-	if (dpm_state == POWER_STATE_TYPE_INTERNAL_THERMAL)
-		adev->pm.dpm.thermal_active = true;
-	else
-		adev->pm.dpm.thermal_active = false;
-	adev->pm.dpm.state = dpm_state;
-	mutex_unlock(&adev->pm.mutex);
-
-	amdgpu_pm_compute_clocks(adev);
-}
-
-static struct amdgpu_ps *amdgpu_dpm_pick_power_state(struct amdgpu_device *adev,
-						     enum amd_pm_state_type dpm_state)
-{
-	int i;
-	struct amdgpu_ps *ps;
-	u32 ui_class;
-	bool single_display = (adev->pm.dpm.new_active_crtc_count < 2) ?
-		true : false;
-
-	/* check if the vblank period is too short to adjust the mclk */
-	if (single_display && adev->pm.funcs->vblank_too_short) {
-		if (amdgpu_dpm_vblank_too_short(adev))
-			single_display = false;
-	}
-
-	/* certain older asics have a separare 3D performance state,
-	 * so try that first if the user selected performance
-	 */
-	if (dpm_state == POWER_STATE_TYPE_PERFORMANCE)
-		dpm_state = POWER_STATE_TYPE_INTERNAL_3DPERF;
-	/* balanced states don't exist at the moment */
-	if (dpm_state == POWER_STATE_TYPE_BALANCED)
-		dpm_state = POWER_STATE_TYPE_PERFORMANCE;
-
-restart_search:
-	/* Pick the best power state based on current conditions */
-	for (i = 0; i < adev->pm.dpm.num_ps; i++) {
-		ps = &adev->pm.dpm.ps[i];
-		ui_class = ps->class & ATOM_PPLIB_CLASSIFICATION_UI_MASK;
-		switch (dpm_state) {
-		/* user states */
-		case POWER_STATE_TYPE_BATTERY:
-			if (ui_class == ATOM_PPLIB_CLASSIFICATION_UI_BATTERY) {
-				if (ps->caps & ATOM_PPLIB_SINGLE_DISPLAY_ONLY) {
-					if (single_display)
-						return ps;
-				} else
-					return ps;
-			}
-			break;
-		case POWER_STATE_TYPE_BALANCED:
-			if (ui_class == ATOM_PPLIB_CLASSIFICATION_UI_BALANCED) {
-				if (ps->caps & ATOM_PPLIB_SINGLE_DISPLAY_ONLY) {
-					if (single_display)
-						return ps;
-				} else
-					return ps;
-			}
-			break;
-		case POWER_STATE_TYPE_PERFORMANCE:
-			if (ui_class == ATOM_PPLIB_CLASSIFICATION_UI_PERFORMANCE) {
-				if (ps->caps & ATOM_PPLIB_SINGLE_DISPLAY_ONLY) {
-					if (single_display)
-						return ps;
-				} else
-					return ps;
-			}
-			break;
-		/* internal states */
-		case POWER_STATE_TYPE_INTERNAL_UVD:
-			if (adev->pm.dpm.uvd_ps)
-				return adev->pm.dpm.uvd_ps;
-			else
-				break;
-		case POWER_STATE_TYPE_INTERNAL_UVD_SD:
-			if (ps->class & ATOM_PPLIB_CLASSIFICATION_SDSTATE)
-				return ps;
-			break;
-		case POWER_STATE_TYPE_INTERNAL_UVD_HD:
-			if (ps->class & ATOM_PPLIB_CLASSIFICATION_HDSTATE)
-				return ps;
-			break;
-		case POWER_STATE_TYPE_INTERNAL_UVD_HD2:
-			if (ps->class & ATOM_PPLIB_CLASSIFICATION_HD2STATE)
-				return ps;
-			break;
-		case POWER_STATE_TYPE_INTERNAL_UVD_MVC:
-			if (ps->class2 & ATOM_PPLIB_CLASSIFICATION2_MVC)
-				return ps;
-			break;
-		case POWER_STATE_TYPE_INTERNAL_BOOT:
-			return adev->pm.dpm.boot_ps;
-		case POWER_STATE_TYPE_INTERNAL_THERMAL:
-			if (ps->class & ATOM_PPLIB_CLASSIFICATION_THERMAL)
-				return ps;
-			break;
-		case POWER_STATE_TYPE_INTERNAL_ACPI:
-			if (ps->class & ATOM_PPLIB_CLASSIFICATION_ACPI)
-				return ps;
-			break;
-		case POWER_STATE_TYPE_INTERNAL_ULV:
-			if (ps->class2 & ATOM_PPLIB_CLASSIFICATION2_ULV)
-				return ps;
-			break;
-		case POWER_STATE_TYPE_INTERNAL_3DPERF:
-			if (ps->class & ATOM_PPLIB_CLASSIFICATION_3DPERFORMANCE)
-				return ps;
-			break;
-		default:
-			break;
-		}
-	}
-	/* use a fallback state if we didn't match */
-	switch (dpm_state) {
-	case POWER_STATE_TYPE_INTERNAL_UVD_SD:
-		dpm_state = POWER_STATE_TYPE_INTERNAL_UVD_HD;
-		goto restart_search;
-	case POWER_STATE_TYPE_INTERNAL_UVD_HD:
-	case POWER_STATE_TYPE_INTERNAL_UVD_HD2:
-	case POWER_STATE_TYPE_INTERNAL_UVD_MVC:
-		if (adev->pm.dpm.uvd_ps) {
-			return adev->pm.dpm.uvd_ps;
-		} else {
-			dpm_state = POWER_STATE_TYPE_PERFORMANCE;
-			goto restart_search;
-		}
-	case POWER_STATE_TYPE_INTERNAL_THERMAL:
-		dpm_state = POWER_STATE_TYPE_INTERNAL_ACPI;
-		goto restart_search;
-	case POWER_STATE_TYPE_INTERNAL_ACPI:
-		dpm_state = POWER_STATE_TYPE_BATTERY;
-		goto restart_search;
-	case POWER_STATE_TYPE_BATTERY:
-	case POWER_STATE_TYPE_BALANCED:
-	case POWER_STATE_TYPE_INTERNAL_3DPERF:
-		dpm_state = POWER_STATE_TYPE_PERFORMANCE;
-		goto restart_search;
-	default:
-		break;
-	}
-
-	return NULL;
-}
-
-static void amdgpu_dpm_change_power_state_locked(struct amdgpu_device *adev)
-{
-	struct amdgpu_ps *ps;
-	enum amd_pm_state_type dpm_state;
-	int ret;
-	bool equal;
-
-	/* if dpm init failed */
-	if (!adev->pm.dpm_enabled)
-		return;
-
-	if (adev->pm.dpm.user_state != adev->pm.dpm.state) {
-		/* add other state override checks here */
-		if ((!adev->pm.dpm.thermal_active) &&
-		    (!adev->pm.dpm.uvd_active))
-			adev->pm.dpm.state = adev->pm.dpm.user_state;
-	}
-	dpm_state = adev->pm.dpm.state;
-
-	ps = amdgpu_dpm_pick_power_state(adev, dpm_state);
-	if (ps)
-		adev->pm.dpm.requested_ps = ps;
-	else
-		return;
-
-	if (amdgpu_dpm == 1) {
-		printk("switching from power state:\n");
-		amdgpu_dpm_print_power_state(adev, adev->pm.dpm.current_ps);
-		printk("switching to power state:\n");
-		amdgpu_dpm_print_power_state(adev, adev->pm.dpm.requested_ps);
-	}
-
-	/* update whether vce is active */
-	ps->vce_active = adev->pm.dpm.vce_active;
-
-	amdgpu_dpm_display_configuration_changed(adev);
-
-	ret = amdgpu_dpm_pre_set_power_state(adev);
-	if (ret)
-		return;
-
-	if ((0 != amgdpu_dpm_check_state_equal(adev, adev->pm.dpm.current_ps, adev->pm.dpm.requested_ps, &equal)))
-		equal = false;
-
-	if (equal)
-		return;
-
-	amdgpu_dpm_set_power_state(adev);
-	amdgpu_dpm_post_set_power_state(adev);
-
-	adev->pm.dpm.current_active_crtcs = adev->pm.dpm.new_active_crtcs;
-	adev->pm.dpm.current_active_crtc_count = adev->pm.dpm.new_active_crtc_count;
-
-	if (adev->pm.funcs->force_performance_level) {
-		if (adev->pm.dpm.thermal_active) {
-			enum amd_dpm_forced_level level = adev->pm.dpm.forced_level;
-			/* force low perf level for thermal */
-			amdgpu_dpm_force_performance_level(adev, AMD_DPM_FORCED_LEVEL_LOW);
-			/* save the user's level */
-			adev->pm.dpm.forced_level = level;
-		} else {
-			/* otherwise, user selected level */
-			amdgpu_dpm_force_performance_level(adev, adev->pm.dpm.forced_level);
-		}
-	}
-}
-
-void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable)
-{
-	if (adev->pp_enabled || adev->pm.funcs->powergate_uvd) {
-		/* enable/disable UVD */
-		mutex_lock(&adev->pm.mutex);
-		amdgpu_dpm_powergate_uvd(adev, !enable);
-		mutex_unlock(&adev->pm.mutex);
-	} else {
-		if (enable) {
-			mutex_lock(&adev->pm.mutex);
-			adev->pm.dpm.uvd_active = true;
-			adev->pm.dpm.state = POWER_STATE_TYPE_INTERNAL_UVD;
-			mutex_unlock(&adev->pm.mutex);
-		} else {
-			mutex_lock(&adev->pm.mutex);
-			adev->pm.dpm.uvd_active = false;
-			mutex_unlock(&adev->pm.mutex);
-		}
-		amdgpu_pm_compute_clocks(adev);
-	}
-}
-
-void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable)
-{
-	if (adev->pp_enabled || adev->pm.funcs->powergate_vce) {
-		/* enable/disable VCE */
-		mutex_lock(&adev->pm.mutex);
-		amdgpu_dpm_powergate_vce(adev, !enable);
-		mutex_unlock(&adev->pm.mutex);
-	} else {
-		if (enable) {
-			mutex_lock(&adev->pm.mutex);
-			adev->pm.dpm.vce_active = true;
-			/* XXX select vce level based on ring/task */
-			adev->pm.dpm.vce_level = AMD_VCE_LEVEL_AC_ALL;
-			mutex_unlock(&adev->pm.mutex);
-			amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
-							AMD_CG_STATE_UNGATE);
-			amdgpu_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
-							AMD_PG_STATE_UNGATE);
-			amdgpu_pm_compute_clocks(adev);
-		} else {
-			amdgpu_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
-							AMD_PG_STATE_GATE);
-			amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
-							AMD_CG_STATE_GATE);
-			mutex_lock(&adev->pm.mutex);
-			adev->pm.dpm.vce_active = false;
-			mutex_unlock(&adev->pm.mutex);
-			amdgpu_pm_compute_clocks(adev);
-		}
-
-	}
-}
-
-void amdgpu_pm_print_power_states(struct amdgpu_device *adev)
-{
-	int i;
-
-	if (adev->pp_enabled)
-		/* TO DO */
-		return;
-
-	for (i = 0; i < adev->pm.dpm.num_ps; i++)
-		amdgpu_dpm_print_power_state(adev, &adev->pm.dpm.ps[i]);
-
-}
-
-int amdgpu_pm_sysfs_init(struct amdgpu_device *adev)
-{
-	int ret;
-
-	if (adev->pm.sysfs_initialized)
-		return 0;
-
-	if (!adev->pp_enabled) {
-		if (adev->pm.funcs->get_temperature == NULL)
-			return 0;
-	}
-
-	adev->pm.int_hwmon_dev = hwmon_device_register_with_groups(adev->dev,
-								   DRIVER_NAME, adev,
-								   hwmon_groups);
-	if (IS_ERR(adev->pm.int_hwmon_dev)) {
-		ret = PTR_ERR(adev->pm.int_hwmon_dev);
-		dev_err(adev->dev,
-			"Unable to register hwmon device: %d\n", ret);
-		return ret;
-	}
-
-	ret = device_create_file(adev->dev, &dev_attr_power_dpm_state);
-	if (ret) {
-		DRM_ERROR("failed to create device file for dpm state\n");
-		return ret;
-	}
-	ret = device_create_file(adev->dev, &dev_attr_power_dpm_force_performance_level);
-	if (ret) {
-		DRM_ERROR("failed to create device file for dpm state\n");
-		return ret;
-	}
-
-	if (adev->pp_enabled) {
-		ret = device_create_file(adev->dev, &dev_attr_pp_num_states);
-		if (ret) {
-			DRM_ERROR("failed to create device file pp_num_states\n");
-			return ret;
-		}
-		ret = device_create_file(adev->dev, &dev_attr_pp_cur_state);
-		if (ret) {
-			DRM_ERROR("failed to create device file pp_cur_state\n");
-			return ret;
-		}
-		ret = device_create_file(adev->dev, &dev_attr_pp_force_state);
-		if (ret) {
-			DRM_ERROR("failed to create device file pp_force_state\n");
-			return ret;
-		}
-		ret = device_create_file(adev->dev, &dev_attr_pp_table);
-		if (ret) {
-			DRM_ERROR("failed to create device file pp_table\n");
-			return ret;
-		}
-	}
-
-	ret = device_create_file(adev->dev, &dev_attr_pp_dpm_sclk);
-	if (ret) {
-		DRM_ERROR("failed to create device file pp_dpm_sclk\n");
-		return ret;
-	}
-	ret = device_create_file(adev->dev, &dev_attr_pp_dpm_mclk);
-	if (ret) {
-		DRM_ERROR("failed to create device file pp_dpm_mclk\n");
-		return ret;
-	}
-	ret = device_create_file(adev->dev, &dev_attr_pp_dpm_pcie);
-	if (ret) {
-		DRM_ERROR("failed to create device file pp_dpm_pcie\n");
-		return ret;
-	}
-	ret = device_create_file(adev->dev, &dev_attr_pp_sclk_od);
-	if (ret) {
-		DRM_ERROR("failed to create device file pp_sclk_od\n");
-		return ret;
-	}
-	ret = device_create_file(adev->dev, &dev_attr_pp_mclk_od);
-	if (ret) {
-		DRM_ERROR("failed to create device file pp_mclk_od\n");
-		return ret;
-	}
-	ret = device_create_file(adev->dev,
-			&dev_attr_pp_gfx_power_profile);
-	if (ret) {
-		DRM_ERROR("failed to create device file	"
-				"pp_gfx_power_profile\n");
-		return ret;
-	}
-	ret = device_create_file(adev->dev,
-			&dev_attr_pp_compute_power_profile);
-	if (ret) {
-		DRM_ERROR("failed to create device file	"
-				"pp_compute_power_profile\n");
-		return ret;
-	}
-
-	ret = amdgpu_debugfs_pm_init(adev);
-	if (ret) {
-		DRM_ERROR("Failed to register debugfs file for dpm!\n");
-		return ret;
-	}
-
-	adev->pm.sysfs_initialized = true;
-
-	return 0;
-}
-
-void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev)
-{
-	if (adev->pm.int_hwmon_dev)
-		hwmon_device_unregister(adev->pm.int_hwmon_dev);
-	device_remove_file(adev->dev, &dev_attr_power_dpm_state);
-	device_remove_file(adev->dev, &dev_attr_power_dpm_force_performance_level);
-	if (adev->pp_enabled) {
-		device_remove_file(adev->dev, &dev_attr_pp_num_states);
-		device_remove_file(adev->dev, &dev_attr_pp_cur_state);
-		device_remove_file(adev->dev, &dev_attr_pp_force_state);
-		device_remove_file(adev->dev, &dev_attr_pp_table);
-	}
-	device_remove_file(adev->dev, &dev_attr_pp_dpm_sclk);
-	device_remove_file(adev->dev, &dev_attr_pp_dpm_mclk);
-	device_remove_file(adev->dev, &dev_attr_pp_dpm_pcie);
-	device_remove_file(adev->dev, &dev_attr_pp_sclk_od);
-	device_remove_file(adev->dev, &dev_attr_pp_mclk_od);
-	device_remove_file(adev->dev,
-			&dev_attr_pp_gfx_power_profile);
-	device_remove_file(adev->dev,
-			&dev_attr_pp_compute_power_profile);
-}
-
-void amdgpu_pm_compute_clocks(struct amdgpu_device *adev)
-{
-	struct drm_device *ddev = adev->ddev;
-	struct drm_crtc *crtc;
-	struct amdgpu_crtc *amdgpu_crtc;
-	int i = 0;
-
-	if (!adev->pm.dpm_enabled)
-		return;
-
-	if (adev->mode_info.num_crtc)
-		amdgpu_display_bandwidth_update(adev);
-
-	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		if (ring && ring->ready)
-			amdgpu_fence_wait_empty(ring);
-	}
-
-	if (adev->pp_enabled) {
-		amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_DISPLAY_CONFIG_CHANGE, NULL, NULL);
-	} else {
-		mutex_lock(&adev->pm.mutex);
-		adev->pm.dpm.new_active_crtcs = 0;
-		adev->pm.dpm.new_active_crtc_count = 0;
-		if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) {
-			list_for_each_entry(crtc,
-					    &ddev->mode_config.crtc_list, head) {
-				amdgpu_crtc = to_amdgpu_crtc(crtc);
-				if (crtc->enabled) {
-					adev->pm.dpm.new_active_crtcs |= (1 << amdgpu_crtc->crtc_id);
-					adev->pm.dpm.new_active_crtc_count++;
-				}
-			}
-		}
-		/* update battery/ac status */
-		if (power_supply_is_system_supplied() > 0)
-			adev->pm.dpm.ac_power = true;
-		else
-			adev->pm.dpm.ac_power = false;
-
-		amdgpu_dpm_change_power_state_locked(adev);
-
-		mutex_unlock(&adev->pm.mutex);
-	}
-}
-
-/*
- * Debugfs info
- */
-#if defined(CONFIG_DEBUG_FS)
-
-static int amdgpu_debugfs_pm_info_pp(struct seq_file *m, struct amdgpu_device *adev)
-{
-	uint32_t value;
-	struct pp_gpu_power query = {0};
-	int size;
-
-	/* sanity check PP is enabled */
-	if (!(adev->powerplay.pp_funcs &&
-	      adev->powerplay.pp_funcs->read_sensor))
-	      return -EINVAL;
-
-	/* GPU Clocks */
-	size = sizeof(value);
-	seq_printf(m, "GFX Clocks and Power:\n");
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GFX_MCLK, (void *)&value, &size))
-		seq_printf(m, "\t%u MHz (MCLK)\n", value/100);
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GFX_SCLK, (void *)&value, &size))
-		seq_printf(m, "\t%u MHz (SCLK)\n", value/100);
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VDDGFX, (void *)&value, &size))
-		seq_printf(m, "\t%u mV (VDDGFX)\n", value);
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VDDNB, (void *)&value, &size))
-		seq_printf(m, "\t%u mV (VDDNB)\n", value);
-	size = sizeof(query);
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_POWER, (void *)&query, &size)) {
-		seq_printf(m, "\t%u.%u W (VDDC)\n", query.vddc_power >> 8,
-				query.vddc_power & 0xff);
-		seq_printf(m, "\t%u.%u W (VDDCI)\n", query.vddci_power >> 8,
-				query.vddci_power & 0xff);
-		seq_printf(m, "\t%u.%u W (max GPU)\n", query.max_gpu_power >> 8,
-				query.max_gpu_power & 0xff);
-		seq_printf(m, "\t%u.%u W (average GPU)\n", query.average_gpu_power >> 8,
-				query.average_gpu_power & 0xff);
-	}
-	size = sizeof(value);
-	seq_printf(m, "\n");
-
-	/* GPU Temp */
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_TEMP, (void *)&value, &size))
-		seq_printf(m, "GPU Temperature: %u C\n", value/1000);
-
-	/* GPU Load */
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_LOAD, (void *)&value, &size))
-		seq_printf(m, "GPU Load: %u %%\n", value);
-	seq_printf(m, "\n");
-
-	/* UVD clocks */
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_POWER, (void *)&value, &size)) {
-		if (!value) {
-			seq_printf(m, "UVD: Disabled\n");
-		} else {
-			seq_printf(m, "UVD: Enabled\n");
-			if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_DCLK, (void *)&value, &size))
-				seq_printf(m, "\t%u MHz (DCLK)\n", value/100);
-			if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_VCLK, (void *)&value, &size))
-				seq_printf(m, "\t%u MHz (VCLK)\n", value/100);
-		}
-	}
-	seq_printf(m, "\n");
-
-	/* VCE clocks */
-	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VCE_POWER, (void *)&value, &size)) {
-		if (!value) {
-			seq_printf(m, "VCE: Disabled\n");
-		} else {
-			seq_printf(m, "VCE: Enabled\n");
-			if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VCE_ECCLK, (void *)&value, &size))
-				seq_printf(m, "\t%u MHz (ECCLK)\n", value/100);
-		}
-	}
-
-	return 0;
-}
-
-static void amdgpu_parse_cg_state(struct seq_file *m, u32 flags)
-{
-	int i;
-
-	for (i = 0; clocks[i].flag; i++)
-		seq_printf(m, "\t%s: %s\n", clocks[i].name,
-			   (flags & clocks[i].flag) ? "On" : "Off");
-}
-
-static int amdgpu_debugfs_pm_info(struct seq_file *m, void *data)
-{
-	struct drm_info_node *node = (struct drm_info_node *) m->private;
-	struct drm_device *dev = node->minor->dev;
-	struct amdgpu_device *adev = dev->dev_private;
-	struct drm_device *ddev = adev->ddev;
-	u32 flags = 0;
-
-	amdgpu_get_clockgating_state(adev, &flags);
-	seq_printf(m, "Clock Gating Flags Mask: 0x%x\n", flags);
-	amdgpu_parse_cg_state(m, flags);
-	seq_printf(m, "\n");
-
-	if (!adev->pm.dpm_enabled) {
-		seq_printf(m, "dpm not enabled\n");
-		return 0;
-	}
-	if  ((adev->flags & AMD_IS_PX) &&
-	     (ddev->switch_power_state != DRM_SWITCH_POWER_ON)) {
-		seq_printf(m, "PX asic powered off\n");
-	} else if (adev->pp_enabled) {
-		return amdgpu_debugfs_pm_info_pp(m, adev);
-	} else {
-		mutex_lock(&adev->pm.mutex);
-		if (adev->pm.funcs->debugfs_print_current_performance_level)
-			adev->pm.funcs->debugfs_print_current_performance_level(adev, m);
-		else
-			seq_printf(m, "Debugfs support not implemented for this asic\n");
-		mutex_unlock(&adev->pm.mutex);
-	}
-
-	return 0;
-}
-
-static const struct drm_info_list amdgpu_pm_info_list[] = {
-	{"amdgpu_pm_info", amdgpu_debugfs_pm_info, 0, NULL},
-};
-#endif
-
-static int amdgpu_debugfs_pm_init(struct amdgpu_device *adev)
-{
-#if defined(CONFIG_DEBUG_FS)
-	return amdgpu_debugfs_add_files(adev, amdgpu_pm_info_list, ARRAY_SIZE(amdgpu_pm_info_list));
-#else
-	return 0;
-#endif
-}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.h
deleted file mode 100644
index c19c4d138751..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.h
+++ /dev/null
@@ -1,41 +0,0 @@
-/*
- * Copyright 2014 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- */
-
-#ifndef __AMDGPU_PM_H__
-#define __AMDGPU_PM_H__
-
-struct cg_flag_name
-{
-	u32 flag;
-	const char *name;
-};
-
-int amdgpu_pm_sysfs_init(struct amdgpu_device *adev);
-void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev);
-void amdgpu_pm_print_power_states(struct amdgpu_device *adev);
-void amdgpu_pm_compute_clocks(struct amdgpu_device *adev);
-void amdgpu_dpm_thermal_work_handler(struct work_struct *work);
-void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable);
-void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable);
-
-#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c
new file mode 100644
index 000000000000..6e91ea1de5aa
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c
@@ -0,0 +1,680 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/perf_event.h>
+#include <linux/init.h>
+#include "amdgpu.h"
+#include "amdgpu_pmu.h"
+
+#define PMU_NAME_SIZE 32
+#define NUM_FORMATS_AMDGPU_PMU		4
+#define NUM_FORMATS_DF_VEGA20		3
+#define NUM_EVENTS_DF_VEGA20		8
+#define NUM_EVENT_TYPES_VEGA20		1
+#define NUM_EVENTS_VEGA20_XGMI		2
+#define NUM_EVENTS_VEGA20_MAX		NUM_EVENTS_VEGA20_XGMI
+#define NUM_EVENT_TYPES_ARCTURUS	1
+#define NUM_EVENTS_ARCTURUS_XGMI	6
+#define NUM_EVENTS_ARCTURUS_MAX		NUM_EVENTS_ARCTURUS_XGMI
+
+struct amdgpu_pmu_event_attribute {
+	struct device_attribute attr;
+	const char *event_str;
+	unsigned int type;
+};
+
+/* record to keep track of pmu entry per pmu type per device */
+struct amdgpu_pmu_entry {
+	struct list_head entry;
+	struct amdgpu_device *adev;
+	struct pmu pmu;
+	unsigned int pmu_perf_type;
+	char *pmu_type_name;
+	char *pmu_file_prefix;
+	struct attribute_group fmt_attr_group;
+	struct amdgpu_pmu_event_attribute *fmt_attr;
+	struct attribute_group evt_attr_group;
+	struct amdgpu_pmu_event_attribute *evt_attr;
+};
+
+static ssize_t amdgpu_pmu_event_show(struct device *dev,
+				struct device_attribute *attr, char *buf)
+{
+	struct amdgpu_pmu_event_attribute *amdgpu_pmu_attr;
+
+	amdgpu_pmu_attr = container_of(attr, struct amdgpu_pmu_event_attribute,
+									attr);
+
+	if (!amdgpu_pmu_attr->type)
+		return sprintf(buf, "%s\n", amdgpu_pmu_attr->event_str);
+
+	return sprintf(buf, "%s,type=0x%x\n",
+			amdgpu_pmu_attr->event_str, amdgpu_pmu_attr->type);
+}
+
+static LIST_HEAD(amdgpu_pmu_list);
+
+
+struct amdgpu_pmu_attr {
+	const char *name;
+	const char *config;
+};
+
+struct amdgpu_pmu_type {
+	const unsigned int type;
+	const unsigned int num_of_type;
+};
+
+struct amdgpu_pmu_config {
+	struct amdgpu_pmu_attr *formats;
+	unsigned int num_formats;
+	struct amdgpu_pmu_attr *events;
+	unsigned int num_events;
+	struct amdgpu_pmu_type *types;
+	unsigned int num_types;
+};
+
+/*
+ * Events fall under two categories:
+ *  - PMU typed
+ *    Events in /sys/bus/event_source/devices/amdgpu_<pmu_type>_<dev_num> have
+ *    performance counter operations handled by one IP <pmu_type>.  Formats and
+ *    events should be defined by <pmu_type>_<asic_type>_formats and
+ *    <pmu_type>_<asic_type>_events respectively.
+ *
+ *  - Event config typed
+ *    Events in /sys/bus/event_source/devices/amdgpu_<dev_num> have performance
+ *    counter operations that can be handled by multiple IPs dictated by their
+ *    "type" format field.  Formats and events should be defined by
+ *    amdgpu_pmu_formats and <asic_type>_events respectively.  Format field
+ *    "type" is generated in amdgpu_pmu_event_show and defined in
+ *    <asic_type>_event_config_types.
+ */
+
+static struct amdgpu_pmu_attr amdgpu_pmu_formats[NUM_FORMATS_AMDGPU_PMU] = {
+	{ .name = "event", .config = "config:0-7" },
+	{ .name = "instance", .config = "config:8-15" },
+	{ .name = "umask", .config = "config:16-23"},
+	{ .name = "type", .config = "config:56-63"}
+};
+
+/* Vega20 events */
+static struct amdgpu_pmu_attr vega20_events[NUM_EVENTS_VEGA20_MAX] = {
+	{ .name = "xgmi_link0_data_outbound",
+			.config = "event=0x7,instance=0x46,umask=0x2" },
+	{ .name = "xgmi_link1_data_outbound",
+			.config = "event=0x7,instance=0x47,umask=0x2" }
+};
+
+static struct amdgpu_pmu_type vega20_types[NUM_EVENT_TYPES_VEGA20] = {
+	{ .type = AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI,
+					.num_of_type = NUM_EVENTS_VEGA20_XGMI }
+};
+
+static struct amdgpu_pmu_config vega20_config = {
+	.formats = amdgpu_pmu_formats,
+	.num_formats = ARRAY_SIZE(amdgpu_pmu_formats),
+	.events = vega20_events,
+	.num_events = ARRAY_SIZE(vega20_events),
+	.types = vega20_types,
+	.num_types = ARRAY_SIZE(vega20_types)
+};
+
+/* Vega20 data fabric (DF) events */
+static struct amdgpu_pmu_attr df_vega20_formats[NUM_FORMATS_DF_VEGA20] = {
+	{ .name = "event", .config = "config:0-7" },
+	{ .name = "instance", .config = "config:8-15" },
+	{ .name = "umask", .config = "config:16-23"}
+};
+
+static struct amdgpu_pmu_attr df_vega20_events[NUM_EVENTS_DF_VEGA20] = {
+	{ .name = "cake0_pcsout_txdata",
+			.config = "event=0x7,instance=0x46,umask=0x2" },
+	{ .name = "cake1_pcsout_txdata",
+			.config = "event=0x7,instance=0x47,umask=0x2" },
+	{ .name = "cake0_pcsout_txmeta",
+			.config = "event=0x7,instance=0x46,umask=0x4" },
+	{ .name = "cake1_pcsout_txmeta",
+			.config = "event=0x7,instance=0x47,umask=0x4" },
+	{ .name = "cake0_ftiinstat_reqalloc",
+			.config = "event=0xb,instance=0x46,umask=0x4" },
+	{ .name = "cake1_ftiinstat_reqalloc",
+			.config = "event=0xb,instance=0x47,umask=0x4" },
+	{ .name = "cake0_ftiinstat_rspalloc",
+			.config = "event=0xb,instance=0x46,umask=0x8" },
+	{ .name = "cake1_ftiinstat_rspalloc",
+			.config = "event=0xb,instance=0x47,umask=0x8" }
+};
+
+static struct amdgpu_pmu_config df_vega20_config = {
+	.formats = df_vega20_formats,
+	.num_formats = ARRAY_SIZE(df_vega20_formats),
+	.events = df_vega20_events,
+	.num_events = ARRAY_SIZE(df_vega20_events),
+	.types = NULL,
+	.num_types = 0
+};
+
+/* Arcturus events */
+static struct amdgpu_pmu_attr arcturus_events[NUM_EVENTS_ARCTURUS_MAX] = {
+	{ .name = "xgmi_link0_data_outbound",
+			.config = "event=0x7,instance=0x4b,umask=0x2" },
+	{ .name = "xgmi_link1_data_outbound",
+			.config = "event=0x7,instance=0x4c,umask=0x2" },
+	{ .name = "xgmi_link2_data_outbound",
+			.config = "event=0x7,instance=0x4d,umask=0x2" },
+	{ .name = "xgmi_link3_data_outbound",
+			.config = "event=0x7,instance=0x4e,umask=0x2" },
+	{ .name = "xgmi_link4_data_outbound",
+			.config = "event=0x7,instance=0x4f,umask=0x2" },
+	{ .name = "xgmi_link5_data_outbound",
+			.config = "event=0x7,instance=0x50,umask=0x2" }
+};
+
+static struct amdgpu_pmu_type arcturus_types[NUM_EVENT_TYPES_ARCTURUS] = {
+	{ .type = AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI,
+				.num_of_type = NUM_EVENTS_ARCTURUS_XGMI }
+};
+
+static struct amdgpu_pmu_config arcturus_config = {
+	.formats = amdgpu_pmu_formats,
+	.num_formats = ARRAY_SIZE(amdgpu_pmu_formats),
+	.events = arcturus_events,
+	.num_events = ARRAY_SIZE(arcturus_events),
+	.types = arcturus_types,
+	.num_types = ARRAY_SIZE(arcturus_types)
+};
+
+/* initialize perf counter */
+static int amdgpu_perf_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	/* test the event attr type check for PMU enumeration */
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/* update the hw_perf_event struct with config data */
+	hwc->config = event->attr.config;
+	hwc->config_base = AMDGPU_PMU_PERF_TYPE_NONE;
+
+	return 0;
+}
+
+/* start perf counter */
+static void amdgpu_perf_start(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct amdgpu_pmu_entry *pe = container_of(event->pmu,
+						  struct amdgpu_pmu_entry,
+						  pmu);
+	int target_cntr = 0;
+
+	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
+		return;
+
+	if ((!pe->adev->df.funcs) ||
+	    (!pe->adev->df.funcs->pmc_start))
+		return;
+
+	WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+	hwc->state = 0;
+
+	switch (hwc->config_base) {
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_DF:
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI:
+		if (!(flags & PERF_EF_RELOAD)) {
+			target_cntr = pe->adev->df.funcs->pmc_start(pe->adev,
+						hwc->config, 0 /* unused */,
+						1 /* add counter */);
+			if (target_cntr < 0)
+				break;
+
+			hwc->idx = target_cntr;
+		}
+
+		pe->adev->df.funcs->pmc_start(pe->adev, hwc->config,
+								hwc->idx, 0);
+		break;
+	default:
+		break;
+	}
+
+	perf_event_update_userpage(event);
+}
+
+/* read perf counter */
+static void amdgpu_perf_read(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct amdgpu_pmu_entry *pe = container_of(event->pmu,
+						  struct amdgpu_pmu_entry,
+						  pmu);
+	u64 count, prev;
+
+	if ((!pe->adev->df.funcs) ||
+	    (!pe->adev->df.funcs->pmc_get_count))
+		return;
+
+	prev = local64_read(&hwc->prev_count);
+	do {
+		switch (hwc->config_base) {
+		case AMDGPU_PMU_EVENT_CONFIG_TYPE_DF:
+		case AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI:
+			pe->adev->df.funcs->pmc_get_count(pe->adev,
+						hwc->config, hwc->idx, &count);
+			break;
+		default:
+			count = 0;
+			break;
+		}
+	} while (!local64_try_cmpxchg(&hwc->prev_count, &prev, count));
+
+	local64_add(count - prev, &event->count);
+}
+
+/* stop perf counter */
+static void amdgpu_perf_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct amdgpu_pmu_entry *pe = container_of(event->pmu,
+						  struct amdgpu_pmu_entry,
+						  pmu);
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	if ((!pe->adev->df.funcs) ||
+	    (!pe->adev->df.funcs->pmc_stop))
+		return;
+
+	switch (hwc->config_base) {
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_DF:
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI:
+		pe->adev->df.funcs->pmc_stop(pe->adev, hwc->config, hwc->idx,
+									0);
+		break;
+	default:
+		break;
+	}
+
+	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+	hwc->state |= PERF_HES_STOPPED;
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	amdgpu_perf_read(event);
+	hwc->state |= PERF_HES_UPTODATE;
+}
+
+/* add perf counter */
+static int amdgpu_perf_add(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int retval = 0, target_cntr;
+	struct amdgpu_pmu_entry *pe = container_of(event->pmu,
+						  struct amdgpu_pmu_entry,
+						  pmu);
+
+	if ((!pe->adev->df.funcs) ||
+	    (!pe->adev->df.funcs->pmc_start))
+		return -EINVAL;
+
+	switch (pe->pmu_perf_type) {
+	case AMDGPU_PMU_PERF_TYPE_DF:
+		hwc->config_base = AMDGPU_PMU_EVENT_CONFIG_TYPE_DF;
+		break;
+	case AMDGPU_PMU_PERF_TYPE_ALL:
+		hwc->config_base = (hwc->config >>
+					AMDGPU_PMU_EVENT_CONFIG_TYPE_SHIFT) &
+					AMDGPU_PMU_EVENT_CONFIG_TYPE_MASK;
+		break;
+	}
+
+	event->hw.state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+
+	switch (hwc->config_base) {
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_DF:
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI:
+		target_cntr = pe->adev->df.funcs->pmc_start(pe->adev,
+						hwc->config, 0 /* unused */,
+						1 /* add counter */);
+		if (target_cntr < 0)
+			retval = target_cntr;
+		else
+			hwc->idx = target_cntr;
+
+		break;
+	default:
+		return 0;
+	}
+
+	if (retval)
+		return retval;
+
+	if (flags & PERF_EF_START)
+		amdgpu_perf_start(event, PERF_EF_RELOAD);
+
+	return retval;
+}
+
+/* delete perf counter  */
+static void amdgpu_perf_del(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct amdgpu_pmu_entry *pe = container_of(event->pmu,
+						  struct amdgpu_pmu_entry,
+						  pmu);
+	if ((!pe->adev->df.funcs) ||
+	    (!pe->adev->df.funcs->pmc_stop))
+		return;
+
+	amdgpu_perf_stop(event, PERF_EF_UPDATE);
+
+	switch (hwc->config_base) {
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_DF:
+	case AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI:
+		pe->adev->df.funcs->pmc_stop(pe->adev, hwc->config, hwc->idx,
+									1);
+		break;
+	default:
+		break;
+	}
+
+	perf_event_update_userpage(event);
+}
+
+static void amdgpu_pmu_create_event_attrs_by_type(
+				struct attribute_group *attr_group,
+				struct amdgpu_pmu_event_attribute *pmu_attr,
+				struct amdgpu_pmu_attr events[],
+				int s_offset,
+				int e_offset,
+				unsigned int type)
+{
+	int i;
+
+	pmu_attr += s_offset;
+
+	for (i = s_offset; i < e_offset; i++) {
+		attr_group->attrs[i] = &pmu_attr->attr.attr;
+		sysfs_attr_init(&pmu_attr->attr.attr);
+		pmu_attr->attr.attr.name = events[i].name;
+		pmu_attr->attr.attr.mode = 0444;
+		pmu_attr->attr.show = amdgpu_pmu_event_show;
+		pmu_attr->event_str = events[i].config;
+		pmu_attr->type = type;
+		pmu_attr++;
+	}
+}
+
+static void amdgpu_pmu_create_attrs(struct attribute_group *attr_group,
+				struct amdgpu_pmu_event_attribute *pmu_attr,
+				struct amdgpu_pmu_attr events[],
+				int num_events)
+{
+	amdgpu_pmu_create_event_attrs_by_type(attr_group, pmu_attr, events, 0,
+				num_events, AMDGPU_PMU_EVENT_CONFIG_TYPE_NONE);
+}
+
+
+static int amdgpu_pmu_alloc_pmu_attrs(
+				struct attribute_group *fmt_attr_group,
+				struct amdgpu_pmu_event_attribute **fmt_attr,
+				struct attribute_group *evt_attr_group,
+				struct amdgpu_pmu_event_attribute **evt_attr,
+				struct amdgpu_pmu_config *config)
+{
+	*fmt_attr = kcalloc(config->num_formats, sizeof(**fmt_attr),
+								GFP_KERNEL);
+
+	if (!(*fmt_attr))
+		return -ENOMEM;
+
+	fmt_attr_group->attrs = kcalloc(config->num_formats + 1,
+				sizeof(*fmt_attr_group->attrs), GFP_KERNEL);
+
+	if (!fmt_attr_group->attrs)
+		goto err_fmt_attr_grp;
+
+	*evt_attr = kcalloc(config->num_events, sizeof(**evt_attr), GFP_KERNEL);
+
+	if (!(*evt_attr))
+		goto err_evt_attr;
+
+	evt_attr_group->attrs = kcalloc(config->num_events + 1,
+				sizeof(*evt_attr_group->attrs), GFP_KERNEL);
+
+	if (!evt_attr_group->attrs)
+		goto err_evt_attr_grp;
+
+	return 0;
+err_evt_attr_grp:
+	kfree(*evt_attr);
+err_evt_attr:
+	kfree(fmt_attr_group->attrs);
+err_fmt_attr_grp:
+	kfree(*fmt_attr);
+	return -ENOMEM;
+}
+
+/* init pmu tracking per pmu type */
+static int init_pmu_entry_by_type_and_add(struct amdgpu_pmu_entry *pmu_entry,
+			struct amdgpu_pmu_config *config)
+{
+	const struct attribute_group *attr_groups[] = {
+		&pmu_entry->fmt_attr_group,
+		&pmu_entry->evt_attr_group,
+		NULL
+	};
+	char pmu_name[PMU_NAME_SIZE];
+	int ret = 0, total_num_events = 0;
+
+	pmu_entry->pmu = (struct pmu){
+		.event_init = amdgpu_perf_event_init,
+		.add = amdgpu_perf_add,
+		.del = amdgpu_perf_del,
+		.start = amdgpu_perf_start,
+		.stop = amdgpu_perf_stop,
+		.read = amdgpu_perf_read,
+		.task_ctx_nr = perf_invalid_context,
+	};
+
+	ret = amdgpu_pmu_alloc_pmu_attrs(&pmu_entry->fmt_attr_group,
+					&pmu_entry->fmt_attr,
+					&pmu_entry->evt_attr_group,
+					&pmu_entry->evt_attr,
+					config);
+
+	if (ret)
+		goto err_out;
+
+	amdgpu_pmu_create_attrs(&pmu_entry->fmt_attr_group, pmu_entry->fmt_attr,
+					config->formats, config->num_formats);
+
+	if (pmu_entry->pmu_perf_type == AMDGPU_PMU_PERF_TYPE_ALL) {
+		int i;
+
+		for (i = 0; i < config->num_types; i++) {
+			amdgpu_pmu_create_event_attrs_by_type(
+					&pmu_entry->evt_attr_group,
+					pmu_entry->evt_attr,
+					config->events,
+					total_num_events,
+					total_num_events +
+						config->types[i].num_of_type,
+					config->types[i].type);
+			total_num_events += config->types[i].num_of_type;
+		}
+	} else {
+		amdgpu_pmu_create_attrs(&pmu_entry->evt_attr_group,
+					pmu_entry->evt_attr,
+					config->events, config->num_events);
+		total_num_events = config->num_events;
+	}
+
+	pmu_entry->pmu.attr_groups = kmemdup(attr_groups, sizeof(attr_groups),
+								GFP_KERNEL);
+
+	if (!pmu_entry->pmu.attr_groups) {
+		ret = -ENOMEM;
+		goto err_attr_group;
+	}
+
+	snprintf(pmu_name, PMU_NAME_SIZE, "%s_%d", pmu_entry->pmu_file_prefix,
+				adev_to_drm(pmu_entry->adev)->primary->index);
+
+	ret = perf_pmu_register(&pmu_entry->pmu, pmu_name, -1);
+
+	if (ret)
+		goto err_register;
+
+	if (pmu_entry->pmu_perf_type != AMDGPU_PMU_PERF_TYPE_ALL)
+		pr_info("Detected AMDGPU %s Counters. # of Counters = %d.\n",
+				pmu_entry->pmu_type_name, total_num_events);
+	else
+		pr_info("Detected AMDGPU %d Perf Events.\n", total_num_events);
+
+
+	list_add_tail(&pmu_entry->entry, &amdgpu_pmu_list);
+
+	return 0;
+err_register:
+	kfree(pmu_entry->pmu.attr_groups);
+err_attr_group:
+	kfree(pmu_entry->fmt_attr_group.attrs);
+	kfree(pmu_entry->fmt_attr);
+	kfree(pmu_entry->evt_attr_group.attrs);
+	kfree(pmu_entry->evt_attr);
+err_out:
+	pr_warn("Error initializing AMDGPU %s PMUs.\n",
+						pmu_entry->pmu_type_name);
+	return ret;
+}
+
+/* destroy all pmu data associated with target device */
+void amdgpu_pmu_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_pmu_entry *pe, *temp;
+
+	list_for_each_entry_safe(pe, temp, &amdgpu_pmu_list, entry) {
+		if (pe->adev != adev)
+			continue;
+		list_del(&pe->entry);
+		perf_pmu_unregister(&pe->pmu);
+		kfree(pe->pmu.attr_groups);
+		kfree(pe->fmt_attr_group.attrs);
+		kfree(pe->fmt_attr);
+		kfree(pe->evt_attr_group.attrs);
+		kfree(pe->evt_attr);
+		kfree(pe);
+	}
+}
+
+static struct amdgpu_pmu_entry *create_pmu_entry(struct amdgpu_device *adev,
+						unsigned int pmu_type,
+						char *pmu_type_name,
+						char *pmu_file_prefix)
+{
+	struct amdgpu_pmu_entry *pmu_entry;
+
+	pmu_entry = kzalloc(sizeof(struct amdgpu_pmu_entry), GFP_KERNEL);
+
+	if (!pmu_entry)
+		return pmu_entry;
+
+	pmu_entry->adev = adev;
+	pmu_entry->fmt_attr_group.name = "format";
+	pmu_entry->fmt_attr_group.attrs = NULL;
+	pmu_entry->evt_attr_group.name = "events";
+	pmu_entry->evt_attr_group.attrs = NULL;
+	pmu_entry->pmu_perf_type = pmu_type;
+	pmu_entry->pmu_type_name = pmu_type_name;
+	pmu_entry->pmu_file_prefix = pmu_file_prefix;
+
+	return pmu_entry;
+}
+
+/* init amdgpu_pmu */
+int amdgpu_pmu_init(struct amdgpu_device *adev)
+{
+	int ret = 0;
+	struct amdgpu_pmu_entry *pmu_entry, *pmu_entry_df;
+
+	switch (adev->asic_type) {
+	case CHIP_VEGA20:
+		pmu_entry_df = create_pmu_entry(adev, AMDGPU_PMU_PERF_TYPE_DF,
+						"DF", "amdgpu_df");
+
+		if (!pmu_entry_df)
+			return -ENOMEM;
+
+		ret = init_pmu_entry_by_type_and_add(pmu_entry_df,
+							&df_vega20_config);
+
+		if (ret) {
+			kfree(pmu_entry_df);
+			return ret;
+		}
+
+		pmu_entry = create_pmu_entry(adev, AMDGPU_PMU_PERF_TYPE_ALL,
+						"", "amdgpu");
+
+		if (!pmu_entry) {
+			amdgpu_pmu_fini(adev);
+			return -ENOMEM;
+		}
+
+		ret = init_pmu_entry_by_type_and_add(pmu_entry,
+							&vega20_config);
+
+		if (ret) {
+			kfree(pmu_entry);
+			amdgpu_pmu_fini(adev);
+			return ret;
+		}
+
+		break;
+	case CHIP_ARCTURUS:
+		pmu_entry = create_pmu_entry(adev, AMDGPU_PMU_PERF_TYPE_ALL,
+						"", "amdgpu");
+		if (!pmu_entry)
+			return -ENOMEM;
+
+		ret = init_pmu_entry_by_type_and_add(pmu_entry,
+							&arcturus_config);
+
+		if (ret) {
+			kfree(pmu_entry);
+			return -ENOMEM;
+		}
+
+		break;
+
+	default:
+		return 0;
+	}
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.h
new file mode 100644
index 000000000000..6882dc48c5d8
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.h
@@ -0,0 +1,57 @@
+/*
+ * Copyright 2019 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef _AMDGPU_PMU_H_
+#define _AMDGPU_PMU_H_
+
+/* PMU types. */
+enum amdgpu_pmu_perf_type {
+	AMDGPU_PMU_PERF_TYPE_NONE = 0,
+	AMDGPU_PMU_PERF_TYPE_DF,
+	AMDGPU_PMU_PERF_TYPE_ALL
+};
+
+/*
+ * PMU type AMDGPU_PMU_PERF_TYPE_ALL can hold events of different "type"
+ * configurations.  Event config types are parsed from the 64-bit raw
+ * config (See EVENT_CONFIG_TYPE_SHIFT and EVENT_CONFIG_TYPE_MASK) and
+ * are registered into the HW perf events config_base.
+ *
+ * PMU types with only a single event configuration type
+ * (non-AMDGPU_PMU_PERF_TYPE_ALL) have their event config type auto generated
+ * when the performance counter is added.
+ */
+enum amdgpu_pmu_event_config_type {
+	AMDGPU_PMU_EVENT_CONFIG_TYPE_NONE = 0,
+	AMDGPU_PMU_EVENT_CONFIG_TYPE_DF,
+	AMDGPU_PMU_EVENT_CONFIG_TYPE_XGMI,
+	AMDGPU_PMU_EVENT_CONFIG_TYPE_MAX
+};
+
+#define AMDGPU_PMU_EVENT_CONFIG_TYPE_SHIFT	56
+#define AMDGPU_PMU_EVENT_CONFIG_TYPE_MASK	0xff
+
+int amdgpu_pmu_init(struct amdgpu_device *adev);
+void amdgpu_pmu_fini(struct amdgpu_device *adev);
+
+#endif /* _AMDGPU_PMU_H_ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.c
deleted file mode 100644
index f5ae871aa11c..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.c
+++ /dev/null
@@ -1,319 +0,0 @@
-/*
- * Copyright 2015 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors: AMD
- *
- */
-#include "atom.h"
-#include "amdgpu.h"
-#include "amd_shared.h"
-#include <linux/module.h>
-#include <linux/moduleparam.h>
-#include "amdgpu_pm.h"
-#include <drm/amdgpu_drm.h>
-#include "amdgpu_powerplay.h"
-#include "si_dpm.h"
-#include "cik_dpm.h"
-#include "vi_dpm.h"
-
-static int amdgpu_create_pp_handle(struct amdgpu_device *adev)
-{
-	struct amd_pp_init pp_init;
-	struct amd_powerplay *amd_pp;
-	int ret;
-
-	amd_pp = &(adev->powerplay);
-	pp_init.chip_family = adev->family;
-	pp_init.chip_id = adev->asic_type;
-	pp_init.pm_en = (amdgpu_dpm != 0 && !amdgpu_sriov_vf(adev)) ? true : false;
-	pp_init.feature_mask = amdgpu_pp_feature_mask;
-	pp_init.device = amdgpu_cgs_create_device(adev);
-	ret = amd_powerplay_create(&pp_init, &(amd_pp->pp_handle));
-	if (ret)
-		return -EINVAL;
-	return 0;
-}
-
-static int amdgpu_pp_early_init(void *handle)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-	struct amd_powerplay *amd_pp;
-	int ret = 0;
-
-	amd_pp = &(adev->powerplay);
-	adev->pp_enabled = false;
-	amd_pp->pp_handle = (void *)adev;
-
-	switch (adev->asic_type) {
-	case CHIP_POLARIS11:
-	case CHIP_POLARIS10:
-	case CHIP_POLARIS12:
-	case CHIP_TONGA:
-	case CHIP_FIJI:
-	case CHIP_TOPAZ:
-	case CHIP_CARRIZO:
-	case CHIP_STONEY:
-	case CHIP_VEGA10:
-		adev->pp_enabled = true;
-		if (amdgpu_create_pp_handle(adev))
-			return -EINVAL;
-		amd_pp->ip_funcs = &pp_ip_funcs;
-		amd_pp->pp_funcs = &pp_dpm_funcs;
-		break;
-	/* These chips don't have powerplay implemenations */
-#ifdef CONFIG_DRM_AMDGPU_SI
-	case CHIP_TAHITI:
-	case CHIP_PITCAIRN:
-	case CHIP_VERDE:
-	case CHIP_OLAND:
-	case CHIP_HAINAN:
-		amd_pp->ip_funcs = &si_dpm_ip_funcs;
-	break;
-#endif
-#ifdef CONFIG_DRM_AMDGPU_CIK
-	case CHIP_BONAIRE:
-	case CHIP_HAWAII:
-		amd_pp->ip_funcs = &ci_dpm_ip_funcs;
-		break;
-	case CHIP_KABINI:
-	case CHIP_MULLINS:
-	case CHIP_KAVERI:
-		amd_pp->ip_funcs = &kv_dpm_ip_funcs;
-		break;
-#endif
-	default:
-		ret = -EINVAL;
-		break;
-	}
-
-	if (adev->powerplay.ip_funcs->early_init)
-		ret = adev->powerplay.ip_funcs->early_init(
-					adev->powerplay.pp_handle);
-
-	if (ret == PP_DPM_DISABLED) {
-		adev->pm.dpm_enabled = false;
-		return 0;
-	}
-	return ret;
-}
-
-
-static int amdgpu_pp_late_init(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->late_init)
-		ret = adev->powerplay.ip_funcs->late_init(
-					adev->powerplay.pp_handle);
-
-	if (adev->pp_enabled && adev->pm.dpm_enabled) {
-		amdgpu_pm_sysfs_init(adev);
-		amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_COMPLETE_INIT, NULL, NULL);
-	}
-
-	return ret;
-}
-
-static int amdgpu_pp_sw_init(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->sw_init)
-		ret = adev->powerplay.ip_funcs->sw_init(
-					adev->powerplay.pp_handle);
-
-	return ret;
-}
-
-static int amdgpu_pp_sw_fini(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->sw_fini)
-		ret = adev->powerplay.ip_funcs->sw_fini(
-					adev->powerplay.pp_handle);
-	if (ret)
-		return ret;
-
-	return ret;
-}
-
-static int amdgpu_pp_hw_init(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->pp_enabled && adev->firmware.load_type == AMDGPU_FW_LOAD_SMU)
-		amdgpu_ucode_init_bo(adev);
-
-	if (adev->powerplay.ip_funcs->hw_init)
-		ret = adev->powerplay.ip_funcs->hw_init(
-					adev->powerplay.pp_handle);
-
-	if (ret == PP_DPM_DISABLED) {
-		adev->pm.dpm_enabled = false;
-		return 0;
-	}
-
-	if ((amdgpu_dpm != 0) && !amdgpu_sriov_vf(adev))
-		adev->pm.dpm_enabled = true;
-
-	return ret;
-}
-
-static int amdgpu_pp_hw_fini(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->hw_fini)
-		ret = adev->powerplay.ip_funcs->hw_fini(
-					adev->powerplay.pp_handle);
-
-	if (adev->pp_enabled && adev->firmware.load_type == AMDGPU_FW_LOAD_SMU)
-		amdgpu_ucode_fini_bo(adev);
-
-	return ret;
-}
-
-static void amdgpu_pp_late_fini(void *handle)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->late_fini)
-		adev->powerplay.ip_funcs->late_fini(
-			  adev->powerplay.pp_handle);
-
-	if (adev->pp_enabled && adev->pm.dpm_enabled)
-		amdgpu_pm_sysfs_fini(adev);
-
-	amd_powerplay_destroy(adev->powerplay.pp_handle);
-}
-
-static int amdgpu_pp_suspend(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->suspend)
-		ret = adev->powerplay.ip_funcs->suspend(
-					 adev->powerplay.pp_handle);
-	return ret;
-}
-
-static int amdgpu_pp_resume(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->resume)
-		ret = adev->powerplay.ip_funcs->resume(
-					adev->powerplay.pp_handle);
-	return ret;
-}
-
-static int amdgpu_pp_set_clockgating_state(void *handle,
-					enum amd_clockgating_state state)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->set_clockgating_state)
-		ret = adev->powerplay.ip_funcs->set_clockgating_state(
-				adev->powerplay.pp_handle, state);
-	return ret;
-}
-
-static int amdgpu_pp_set_powergating_state(void *handle,
-					enum amd_powergating_state state)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->set_powergating_state)
-		ret = adev->powerplay.ip_funcs->set_powergating_state(
-				 adev->powerplay.pp_handle, state);
-	return ret;
-}
-
-
-static bool amdgpu_pp_is_idle(void *handle)
-{
-	bool ret = true;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->is_idle)
-		ret = adev->powerplay.ip_funcs->is_idle(
-					adev->powerplay.pp_handle);
-	return ret;
-}
-
-static int amdgpu_pp_wait_for_idle(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->wait_for_idle)
-		ret = adev->powerplay.ip_funcs->wait_for_idle(
-					adev->powerplay.pp_handle);
-	return ret;
-}
-
-static int amdgpu_pp_soft_reset(void *handle)
-{
-	int ret = 0;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	if (adev->powerplay.ip_funcs->soft_reset)
-		ret = adev->powerplay.ip_funcs->soft_reset(
-					adev->powerplay.pp_handle);
-	return ret;
-}
-
-static const struct amd_ip_funcs amdgpu_pp_ip_funcs = {
-	.name = "amdgpu_powerplay",
-	.early_init = amdgpu_pp_early_init,
-	.late_init = amdgpu_pp_late_init,
-	.sw_init = amdgpu_pp_sw_init,
-	.sw_fini = amdgpu_pp_sw_fini,
-	.hw_init = amdgpu_pp_hw_init,
-	.hw_fini = amdgpu_pp_hw_fini,
-	.late_fini = amdgpu_pp_late_fini,
-	.suspend = amdgpu_pp_suspend,
-	.resume = amdgpu_pp_resume,
-	.is_idle = amdgpu_pp_is_idle,
-	.wait_for_idle = amdgpu_pp_wait_for_idle,
-	.soft_reset = amdgpu_pp_soft_reset,
-	.set_clockgating_state = amdgpu_pp_set_clockgating_state,
-	.set_powergating_state = amdgpu_pp_set_powergating_state,
-};
-
-const struct amdgpu_ip_block_version amdgpu_pp_ip_block =
-{
-	.type = AMD_IP_BLOCK_TYPE_SMC,
-	.major = 1,
-	.minor = 0,
-	.rev = 0,
-	.funcs = &amdgpu_pp_ip_funcs,
-};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.h
deleted file mode 100644
index c0c4bfdcdb14..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.h
+++ /dev/null
@@ -1,33 +0,0 @@
-/*
- *  Copyright 2015 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors: AMD
- *
- */
-
-#ifndef __AMDGPU_POWERPLAY_H__
-#define __AMDGPU_POWERPLAY_H__
-
-#include "amd_shared.h"
-
-extern const struct amdgpu_ip_block_version amdgpu_pp_ip_block;
-
-#endif /* __AMDGPU_POWERPLAY_H__ */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_preempt_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_preempt_mgr.c
new file mode 100644
index 000000000000..34b5e22b44e5
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_preempt_mgr.c
@@ -0,0 +1,145 @@
+// SPDX-License-Identifier: GPL-2.0 OR MIT
+/*
+ * Copyright 2016-2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Christian König, Felix Kuehling
+ */
+
+#include "amdgpu.h"
+
+/**
+ * DOC: mem_info_preempt_used
+ *
+ * The amdgpu driver provides a sysfs API for reporting current total amount of
+ * used preemptible memory.
+ * The file mem_info_preempt_used is used for this, and returns the current
+ * used size of the preemptible block, in bytes
+ */
+static ssize_t mem_info_preempt_used_show(struct device *dev,
+					  struct device_attribute *attr,
+					  char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct ttm_resource_manager *man = &adev->mman.preempt_mgr;
+
+	return sysfs_emit(buf, "%llu\n", ttm_resource_manager_usage(man));
+}
+
+static DEVICE_ATTR_RO(mem_info_preempt_used);
+
+/**
+ * amdgpu_preempt_mgr_new - allocate a new node
+ *
+ * @man: TTM memory type manager
+ * @tbo: TTM BO we need this range for
+ * @place: placement flags and restrictions
+ * @res: TTM memory object
+ *
+ * Dummy, just count the space used without allocating resources or any limit.
+ */
+static int amdgpu_preempt_mgr_new(struct ttm_resource_manager *man,
+				  struct ttm_buffer_object *tbo,
+				  const struct ttm_place *place,
+				  struct ttm_resource **res)
+{
+	*res = kzalloc(sizeof(**res), GFP_KERNEL);
+	if (!*res)
+		return -ENOMEM;
+
+	ttm_resource_init(tbo, place, *res);
+	(*res)->start = AMDGPU_BO_INVALID_OFFSET;
+	return 0;
+}
+
+/**
+ * amdgpu_preempt_mgr_del - free ranges
+ *
+ * @man: TTM memory type manager
+ * @res: TTM memory object
+ *
+ * Free the allocated GTT again.
+ */
+static void amdgpu_preempt_mgr_del(struct ttm_resource_manager *man,
+				   struct ttm_resource *res)
+{
+	ttm_resource_fini(man, res);
+	kfree(res);
+}
+
+static const struct ttm_resource_manager_func amdgpu_preempt_mgr_func = {
+	.alloc = amdgpu_preempt_mgr_new,
+	.free = amdgpu_preempt_mgr_del,
+};
+
+/**
+ * amdgpu_preempt_mgr_init - init PREEMPT manager and DRM MM
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Allocate and initialize the GTT manager.
+ */
+int amdgpu_preempt_mgr_init(struct amdgpu_device *adev)
+{
+	struct ttm_resource_manager *man = &adev->mman.preempt_mgr;
+	int ret;
+
+	man->use_tt = true;
+	man->func = &amdgpu_preempt_mgr_func;
+
+	ttm_resource_manager_init(man, &adev->mman.bdev, (1 << 30));
+
+	ret = device_create_file(adev->dev, &dev_attr_mem_info_preempt_used);
+	if (ret) {
+		DRM_ERROR("Failed to create device file mem_info_preempt_used\n");
+		return ret;
+	}
+
+	ttm_set_driver_manager(&adev->mman.bdev, AMDGPU_PL_PREEMPT, man);
+	ttm_resource_manager_set_used(man, true);
+	return 0;
+}
+
+/**
+ * amdgpu_preempt_mgr_fini - free and destroy GTT manager
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Destroy and free the GTT manager, returns -EBUSY if ranges are still
+ * allocated inside it.
+ */
+void amdgpu_preempt_mgr_fini(struct amdgpu_device *adev)
+{
+	struct ttm_resource_manager *man = &adev->mman.preempt_mgr;
+	int ret;
+
+	ttm_resource_manager_set_used(man, false);
+
+	ret = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
+	if (ret)
+		return;
+
+	if (adev->dev->kobj.sd)
+		device_remove_file(adev->dev, &dev_attr_mem_info_preempt_used);
+
+	ttm_resource_manager_cleanup(man);
+	ttm_set_driver_manager(&adev->mman.bdev, AMDGPU_PL_PREEMPT, NULL);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
deleted file mode 100644
index 6bdc866570ab..000000000000
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c
+++ /dev/null
@@ -1,143 +0,0 @@
-/*
- * Copyright 2012 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * based on nouveau_prime.c
- *
- * Authors: Alex Deucher
- */
-#include <drm/drmP.h>
-
-#include "amdgpu.h"
-#include <drm/amdgpu_drm.h>
-#include <linux/dma-buf.h>
-
-struct sg_table *amdgpu_gem_prime_get_sg_table(struct drm_gem_object *obj)
-{
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
-	int npages = bo->tbo.num_pages;
-
-	return drm_prime_pages_to_sg(bo->tbo.ttm->pages, npages);
-}
-
-void *amdgpu_gem_prime_vmap(struct drm_gem_object *obj)
-{
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
-	int ret;
-
-	ret = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages,
-			  &bo->dma_buf_vmap);
-	if (ret)
-		return ERR_PTR(ret);
-
-	return bo->dma_buf_vmap.virtual;
-}
-
-void amdgpu_gem_prime_vunmap(struct drm_gem_object *obj, void *vaddr)
-{
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
-
-	ttm_bo_kunmap(&bo->dma_buf_vmap);
-}
-
-struct drm_gem_object *
-amdgpu_gem_prime_import_sg_table(struct drm_device *dev,
-				 struct dma_buf_attachment *attach,
-				 struct sg_table *sg)
-{
-	struct reservation_object *resv = attach->dmabuf->resv;
-	struct amdgpu_device *adev = dev->dev_private;
-	struct amdgpu_bo *bo;
-	int ret;
-
-	ww_mutex_lock(&resv->lock, NULL);
-	ret = amdgpu_bo_create(adev, attach->dmabuf->size, PAGE_SIZE, false,
-			       AMDGPU_GEM_DOMAIN_GTT, 0, sg, resv, &bo);
-	ww_mutex_unlock(&resv->lock);
-	if (ret)
-		return ERR_PTR(ret);
-
-	bo->prime_shared_count = 1;
-	return &bo->gem_base;
-}
-
-int amdgpu_gem_prime_pin(struct drm_gem_object *obj)
-{
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
-	long ret = 0;
-
-	ret = amdgpu_bo_reserve(bo, false);
-	if (unlikely(ret != 0))
-		return ret;
-
-	/*
-	 * Wait for all shared fences to complete before we switch to future
-	 * use of exclusive fence on this prime shared bo.
-	 */
-	ret = reservation_object_wait_timeout_rcu(bo->tbo.resv, true, false,
-						  MAX_SCHEDULE_TIMEOUT);
-	if (unlikely(ret < 0)) {
-		DRM_DEBUG_PRIME("Fence wait failed: %li\n", ret);
-		amdgpu_bo_unreserve(bo);
-		return ret;
-	}
-
-	/* pin buffer into GTT */
-	ret = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT, NULL);
-	if (likely(ret == 0))
-		bo->prime_shared_count++;
-
-	amdgpu_bo_unreserve(bo);
-	return ret;
-}
-
-void amdgpu_gem_prime_unpin(struct drm_gem_object *obj)
-{
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
-	int ret = 0;
-
-	ret = amdgpu_bo_reserve(bo, true);
-	if (unlikely(ret != 0))
-		return;
-
-	amdgpu_bo_unpin(bo);
-	if (bo->prime_shared_count)
-		bo->prime_shared_count--;
-	amdgpu_bo_unreserve(bo);
-}
-
-struct reservation_object *amdgpu_gem_prime_res_obj(struct drm_gem_object *obj)
-{
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
-
-	return bo->tbo.resv;
-}
-
-struct dma_buf *amdgpu_gem_prime_export(struct drm_device *dev,
-					struct drm_gem_object *gobj,
-					int flags)
-{
-	struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj);
-
-	if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm))
-		return ERR_PTR(-EPERM);
-
-	return drm_gem_prime_export(dev, gobj, flags);
-}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
index ac5e92e5d59d..0b10497d487c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
@@ -24,42 +24,240 @@
  */
 
 #include <linux/firmware.h>
-#include "drmP.h"
+#include <drm/drm_drv.h>
+
 #include "amdgpu.h"
 #include "amdgpu_psp.h"
 #include "amdgpu_ucode.h"
+#include "amdgpu_xgmi.h"
 #include "soc15_common.h"
 #include "psp_v3_1.h"
+#include "psp_v10_0.h"
+#include "psp_v11_0.h"
+#include "psp_v11_0_8.h"
+#include "psp_v12_0.h"
+#include "psp_v13_0.h"
+#include "psp_v13_0_4.h"
+#include "psp_v14_0.h"
+
+#include "amdgpu_ras.h"
+#include "amdgpu_securedisplay.h"
+#include "amdgpu_atomfirmware.h"
 
-static void psp_set_funcs(struct amdgpu_device *adev);
+#define AMD_VBIOS_FILE_MAX_SIZE_B      (1024*1024*16)
 
-static int psp_early_init(void *handle)
+static int psp_load_smu_fw(struct psp_context *psp);
+static int psp_rap_terminate(struct psp_context *psp);
+static int psp_securedisplay_terminate(struct psp_context *psp);
+
+static int psp_ring_init(struct psp_context *psp,
+			 enum psp_ring_type ring_type)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	int ret = 0;
+	struct psp_ring *ring;
+	struct amdgpu_device *adev = psp->adev;
 
-	psp_set_funcs(adev);
+	ring = &psp->km_ring;
+
+	ring->ring_type = ring_type;
+
+	/* allocate 4k Page of Local Frame Buffer memory for ring */
+	ring->ring_size = 0x1000;
+	ret = amdgpu_bo_create_kernel(adev, ring->ring_size, PAGE_SIZE,
+				      AMDGPU_GEM_DOMAIN_VRAM |
+				      AMDGPU_GEM_DOMAIN_GTT,
+				      &adev->firmware.rbuf,
+				      &ring->ring_mem_mc_addr,
+				      (void **)&ring->ring_mem);
+	if (ret) {
+		ring->ring_size = 0;
+		return ret;
+	}
 
 	return 0;
 }
 
-static int psp_sw_init(void *handle)
+/*
+ * Due to DF Cstate management centralized to PMFW, the firmware
+ * loading sequence will be updated as below:
+ *   - Load KDB
+ *   - Load SYS_DRV
+ *   - Load tOS
+ *   - Load PMFW
+ *   - Setup TMR
+ *   - Load other non-psp fw
+ *   - Load ASD
+ *   - Load XGMI/RAS/HDCP/DTM TA if any
+ *
+ * This new sequence is required for
+ *   - Arcturus and onwards
+ */
+static void psp_check_pmfw_centralized_cstate_management(struct psp_context *psp)
+{
+	struct amdgpu_device *adev = psp->adev;
+
+	if (amdgpu_sriov_vf(adev)) {
+		psp->pmfw_centralized_cstate_management = false;
+		return;
+	}
+
+	switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 4):
+	case IP_VERSION(11, 0, 5):
+	case IP_VERSION(11, 0, 7):
+	case IP_VERSION(11, 0, 9):
+	case IP_VERSION(11, 0, 11):
+	case IP_VERSION(11, 0, 12):
+	case IP_VERSION(11, 0, 13):
+	case IP_VERSION(13, 0, 0):
+	case IP_VERSION(13, 0, 2):
+	case IP_VERSION(13, 0, 7):
+		psp->pmfw_centralized_cstate_management = true;
+		break;
+	default:
+		psp->pmfw_centralized_cstate_management = false;
+		break;
+	}
+}
+
+static int psp_init_sriov_microcode(struct psp_context *psp)
+{
+	struct amdgpu_device *adev = psp->adev;
+	char ucode_prefix[30];
+	int ret = 0;
+
+	amdgpu_ucode_ip_version_decode(adev, MP0_HWIP, ucode_prefix, sizeof(ucode_prefix));
+
+	switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
+	case IP_VERSION(9, 0, 0):
+	case IP_VERSION(11, 0, 7):
+	case IP_VERSION(11, 0, 9):
+		adev->virt.autoload_ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
+		ret = psp_init_cap_microcode(psp, ucode_prefix);
+		break;
+	case IP_VERSION(13, 0, 2):
+		adev->virt.autoload_ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
+		ret = psp_init_cap_microcode(psp, ucode_prefix);
+		ret &= psp_init_ta_microcode(psp, ucode_prefix);
+		break;
+	case IP_VERSION(13, 0, 0):
+		adev->virt.autoload_ucode_id = 0;
+		break;
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 14):
+		ret = psp_init_cap_microcode(psp, ucode_prefix);
+		ret &= psp_init_ta_microcode(psp, ucode_prefix);
+		break;
+	case IP_VERSION(13, 0, 10):
+		adev->virt.autoload_ucode_id = AMDGPU_UCODE_ID_CP_MES1_DATA;
+		ret = psp_init_cap_microcode(psp, ucode_prefix);
+		break;
+	case IP_VERSION(13, 0, 12):
+		ret = psp_init_ta_microcode(psp, ucode_prefix);
+		break;
+	default:
+		return -EINVAL;
+	}
+	return ret;
+}
+
+static int psp_early_init(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 	struct psp_context *psp = &adev->psp;
-	int ret;
 
-	switch (adev->asic_type) {
-	case CHIP_VEGA10:
-		psp->init_microcode = psp_v3_1_init_microcode;
-		psp->bootloader_load_sysdrv = psp_v3_1_bootloader_load_sysdrv;
-		psp->bootloader_load_sos = psp_v3_1_bootloader_load_sos;
-		psp->prep_cmd_buf = psp_v3_1_prep_cmd_buf;
-		psp->ring_init = psp_v3_1_ring_init;
-		psp->ring_create = psp_v3_1_ring_create;
-		psp->ring_destroy = psp_v3_1_ring_destroy;
-		psp->cmd_submit = psp_v3_1_cmd_submit;
-		psp->compare_sram_data = psp_v3_1_compare_sram_data;
-		psp->smu_reload_quirk = psp_v3_1_smu_reload_quirk;
+	psp->autoload_supported = true;
+	psp->boot_time_tmr = true;
+
+	switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
+	case IP_VERSION(9, 0, 0):
+		psp_v3_1_set_psp_funcs(psp);
+		psp->autoload_supported = false;
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(10, 0, 0):
+	case IP_VERSION(10, 0, 1):
+		psp_v10_0_set_psp_funcs(psp);
+		psp->autoload_supported = false;
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(11, 0, 2):
+	case IP_VERSION(11, 0, 4):
+		psp_v11_0_set_psp_funcs(psp);
+		psp->autoload_supported = false;
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(11, 0, 0):
+	case IP_VERSION(11, 0, 7):
+		adev->psp.sup_pd_fw_up = !amdgpu_sriov_vf(adev);
+		fallthrough;
+	case IP_VERSION(11, 0, 5):
+	case IP_VERSION(11, 0, 9):
+	case IP_VERSION(11, 0, 11):
+	case IP_VERSION(11, 5, 0):
+	case IP_VERSION(11, 5, 2):
+	case IP_VERSION(11, 0, 12):
+	case IP_VERSION(11, 0, 13):
+		psp_v11_0_set_psp_funcs(psp);
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(11, 0, 3):
+	case IP_VERSION(12, 0, 1):
+		psp_v12_0_set_psp_funcs(psp);
+		psp->autoload_supported = false;
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(13, 0, 2):
+		psp->boot_time_tmr = false;
+		fallthrough;
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 14):
+		psp_v13_0_set_psp_funcs(psp);
+		psp->autoload_supported = false;
+		break;
+	case IP_VERSION(13, 0, 12):
+		psp_v13_0_set_psp_funcs(psp);
+		psp->autoload_supported = false;
+		adev->psp.sup_ifwi_up = !amdgpu_sriov_vf(adev);
+		break;
+	case IP_VERSION(13, 0, 1):
+	case IP_VERSION(13, 0, 3):
+	case IP_VERSION(13, 0, 5):
+	case IP_VERSION(13, 0, 8):
+	case IP_VERSION(13, 0, 11):
+	case IP_VERSION(14, 0, 0):
+	case IP_VERSION(14, 0, 1):
+	case IP_VERSION(14, 0, 4):
+		psp_v13_0_set_psp_funcs(psp);
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(11, 0, 8):
+		if (adev->apu_flags & AMD_APU_IS_CYAN_SKILLFISH2) {
+			psp_v11_0_8_set_psp_funcs(psp);
+		}
+		psp->autoload_supported = false;
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(13, 0, 0):
+	case IP_VERSION(13, 0, 7):
+	case IP_VERSION(13, 0, 10):
+		psp_v13_0_set_psp_funcs(psp);
+		adev->psp.sup_ifwi_up = !amdgpu_sriov_vf(adev);
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(13, 0, 4):
+		psp_v13_0_4_set_psp_funcs(psp);
+		psp->boot_time_tmr = false;
+		break;
+	case IP_VERSION(14, 0, 2):
+	case IP_VERSION(14, 0, 3):
+		adev->psp.sup_ifwi_up = !amdgpu_sriov_vf(adev);
+		psp_v14_0_set_psp_funcs(psp);
+		break;
+	case IP_VERSION(14, 0, 5):
+		psp_v14_0_set_psp_funcs(psp);
+		psp->boot_time_tmr = false;
 		break;
 	default:
 		return -EINVAL;
@@ -67,30 +265,330 @@ static int psp_sw_init(void *handle)
 
 	psp->adev = adev;
 
-	ret = psp_init_microcode(psp);
-	if (ret) {
-		DRM_ERROR("Failed to load psp firmware!\n");
-		return ret;
+	adev->psp_timeout = 20000;
+
+	psp_check_pmfw_centralized_cstate_management(psp);
+
+	if (amdgpu_sriov_vf(adev))
+		return psp_init_sriov_microcode(psp);
+	else
+		return psp_init_microcode(psp);
+}
+
+void psp_ta_free_shared_buf(struct ta_mem_context *mem_ctx)
+{
+	amdgpu_bo_free_kernel(&mem_ctx->shared_bo, &mem_ctx->shared_mc_addr,
+			      &mem_ctx->shared_buf);
+	mem_ctx->shared_bo = NULL;
+}
+
+static void psp_free_shared_bufs(struct psp_context *psp)
+{
+	void *tmr_buf;
+	void **pptr;
+
+	/* free TMR memory buffer */
+	pptr = amdgpu_sriov_vf(psp->adev) ? &tmr_buf : NULL;
+	amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr, pptr);
+	psp->tmr_bo = NULL;
+
+	/* free xgmi shared memory */
+	psp_ta_free_shared_buf(&psp->xgmi_context.context.mem_context);
+
+	/* free ras shared memory */
+	psp_ta_free_shared_buf(&psp->ras_context.context.mem_context);
+
+	/* free hdcp shared memory */
+	psp_ta_free_shared_buf(&psp->hdcp_context.context.mem_context);
+
+	/* free dtm shared memory */
+	psp_ta_free_shared_buf(&psp->dtm_context.context.mem_context);
+
+	/* free rap shared memory */
+	psp_ta_free_shared_buf(&psp->rap_context.context.mem_context);
+
+	/* free securedisplay shared memory */
+	psp_ta_free_shared_buf(&psp->securedisplay_context.context.mem_context);
+
+
+}
+
+static void psp_memory_training_fini(struct psp_context *psp)
+{
+	struct psp_memory_training_context *ctx = &psp->mem_train_ctx;
+
+	ctx->init = PSP_MEM_TRAIN_NOT_SUPPORT;
+	kfree(ctx->sys_cache);
+	ctx->sys_cache = NULL;
+}
+
+static int psp_memory_training_init(struct psp_context *psp)
+{
+	int ret;
+	struct psp_memory_training_context *ctx = &psp->mem_train_ctx;
+
+	if (ctx->init != PSP_MEM_TRAIN_RESERVE_SUCCESS) {
+		dev_dbg(psp->adev->dev, "memory training is not supported!\n");
+		return 0;
 	}
 
+	ctx->sys_cache = kzalloc(ctx->train_data_size, GFP_KERNEL);
+	if (ctx->sys_cache == NULL) {
+		dev_err(psp->adev->dev, "alloc mem_train_ctx.sys_cache failed!\n");
+		ret = -ENOMEM;
+		goto Err_out;
+	}
+
+	dev_dbg(psp->adev->dev,
+		"train_data_size:%llx,p2c_train_data_offset:%llx,c2p_train_data_offset:%llx.\n",
+		ctx->train_data_size,
+		ctx->p2c_train_data_offset,
+		ctx->c2p_train_data_offset);
+	ctx->init = PSP_MEM_TRAIN_INIT_SUCCESS;
 	return 0;
+
+Err_out:
+	psp_memory_training_fini(psp);
+	return ret;
 }
 
-static int psp_sw_fini(void *handle)
+/*
+ * Helper funciton to query psp runtime database entry
+ *
+ * @adev: amdgpu_device pointer
+ * @entry_type: the type of psp runtime database entry
+ * @db_entry: runtime database entry pointer
+ *
+ * Return false if runtime database doesn't exit or entry is invalid
+ * or true if the specific database entry is found, and copy to @db_entry
+ */
+static bool psp_get_runtime_db_entry(struct amdgpu_device *adev,
+				     enum psp_runtime_entry_type entry_type,
+				     void *db_entry)
 {
+	uint64_t db_header_pos, db_dir_pos;
+	struct psp_runtime_data_header db_header = {0};
+	struct psp_runtime_data_directory db_dir = {0};
+	bool ret = false;
+	int i;
+
+	if (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 6) ||
+	    amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 12) ||
+	    amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 14))
+		return false;
+
+	db_header_pos = adev->gmc.mc_vram_size - PSP_RUNTIME_DB_OFFSET;
+	db_dir_pos = db_header_pos + sizeof(struct psp_runtime_data_header);
+
+	/* read runtime db header from vram */
+	amdgpu_device_vram_access(adev, db_header_pos, (uint32_t *)&db_header,
+			sizeof(struct psp_runtime_data_header), false);
+
+	if (db_header.cookie != PSP_RUNTIME_DB_COOKIE_ID) {
+		/* runtime db doesn't exist, exit */
+		dev_dbg(adev->dev, "PSP runtime database doesn't exist\n");
+		return false;
+	}
+
+	/* read runtime database entry from vram */
+	amdgpu_device_vram_access(adev, db_dir_pos, (uint32_t *)&db_dir,
+			sizeof(struct psp_runtime_data_directory), false);
+
+	if (db_dir.entry_count >= PSP_RUNTIME_DB_DIAG_ENTRY_MAX_COUNT) {
+		/* invalid db entry count, exit */
+		dev_warn(adev->dev, "Invalid PSP runtime database entry count\n");
+		return false;
+	}
+
+	/* look up for requested entry type */
+	for (i = 0; i < db_dir.entry_count && !ret; i++) {
+		if (db_dir.entry_list[i].entry_type == entry_type) {
+			switch (entry_type) {
+			case PSP_RUNTIME_ENTRY_TYPE_BOOT_CONFIG:
+				if (db_dir.entry_list[i].size < sizeof(struct psp_runtime_boot_cfg_entry)) {
+					/* invalid db entry size */
+					dev_warn(adev->dev, "Invalid PSP runtime database boot cfg entry size\n");
+					return false;
+				}
+				/* read runtime database entry */
+				amdgpu_device_vram_access(adev, db_header_pos + db_dir.entry_list[i].offset,
+							  (uint32_t *)db_entry, sizeof(struct psp_runtime_boot_cfg_entry), false);
+				ret = true;
+				break;
+			case PSP_RUNTIME_ENTRY_TYPE_PPTABLE_ERR_STATUS:
+				if (db_dir.entry_list[i].size < sizeof(struct psp_runtime_scpm_entry)) {
+					/* invalid db entry size */
+					dev_warn(adev->dev, "Invalid PSP runtime database scpm entry size\n");
+					return false;
+				}
+				/* read runtime database entry */
+				amdgpu_device_vram_access(adev, db_header_pos + db_dir.entry_list[i].offset,
+							  (uint32_t *)db_entry, sizeof(struct psp_runtime_scpm_entry), false);
+				ret = true;
+				break;
+			default:
+				ret = false;
+				break;
+			}
+		}
+	}
+
+	return ret;
+}
+
+static int psp_sw_init(struct amdgpu_ip_block *ip_block)
+{
+	struct amdgpu_device *adev = ip_block->adev;
+	struct psp_context *psp = &adev->psp;
+	int ret;
+	struct psp_runtime_boot_cfg_entry boot_cfg_entry;
+	struct psp_memory_training_context *mem_training_ctx = &psp->mem_train_ctx;
+	struct psp_runtime_scpm_entry scpm_entry;
+
+	psp->cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
+	if (!psp->cmd) {
+		dev_err(adev->dev, "Failed to allocate memory to command buffer!\n");
+		return -ENOMEM;
+	}
+
+	adev->psp.xgmi_context.supports_extended_data =
+		!adev->gmc.xgmi.connected_to_cpu &&
+		amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 2);
+
+	memset(&scpm_entry, 0, sizeof(scpm_entry));
+	if ((psp_get_runtime_db_entry(adev,
+				PSP_RUNTIME_ENTRY_TYPE_PPTABLE_ERR_STATUS,
+				&scpm_entry)) &&
+	    (scpm_entry.scpm_status != SCPM_DISABLE)) {
+		adev->scpm_enabled = true;
+		adev->scpm_status = scpm_entry.scpm_status;
+	} else {
+		adev->scpm_enabled = false;
+		adev->scpm_status = SCPM_DISABLE;
+	}
+
+	/* TODO: stop gpu driver services and print alarm if scpm is enabled with error status */
+
+	memset(&boot_cfg_entry, 0, sizeof(boot_cfg_entry));
+	if (psp_get_runtime_db_entry(adev,
+				PSP_RUNTIME_ENTRY_TYPE_BOOT_CONFIG,
+				&boot_cfg_entry)) {
+		psp->boot_cfg_bitmask = boot_cfg_entry.boot_cfg_bitmask;
+		if ((psp->boot_cfg_bitmask) &
+		    BOOT_CFG_FEATURE_TWO_STAGE_DRAM_TRAINING) {
+			/* If psp runtime database exists, then
+			 * only enable two stage memory training
+			 * when TWO_STAGE_DRAM_TRAINING bit is set
+			 * in runtime database
+			 */
+			mem_training_ctx->enable_mem_training = true;
+		}
+
+	} else {
+		/* If psp runtime database doesn't exist or is
+		 * invalid, force enable two stage memory training
+		 */
+		mem_training_ctx->enable_mem_training = true;
+	}
+
+	if (mem_training_ctx->enable_mem_training) {
+		ret = psp_memory_training_init(psp);
+		if (ret) {
+			dev_err(adev->dev, "Failed to initialize memory training!\n");
+			return ret;
+		}
+
+		ret = psp_mem_training(psp, PSP_MEM_TRAIN_COLD_BOOT);
+		if (ret) {
+			dev_err(adev->dev, "Failed to process memory training!\n");
+			return ret;
+		}
+	}
+
+	ret = amdgpu_bo_create_kernel(adev, PSP_1_MEG, PSP_1_MEG,
+				      (amdgpu_sriov_vf(adev) || adev->debug_use_vram_fw_buf) ?
+				      AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT,
+				      &psp->fw_pri_bo,
+				      &psp->fw_pri_mc_addr,
+				      &psp->fw_pri_buf);
+	if (ret)
+		return ret;
+
+	ret = amdgpu_bo_create_kernel(adev, PSP_FENCE_BUFFER_SIZE, PAGE_SIZE,
+				      AMDGPU_GEM_DOMAIN_VRAM |
+				      AMDGPU_GEM_DOMAIN_GTT,
+				      &psp->fence_buf_bo,
+				      &psp->fence_buf_mc_addr,
+				      &psp->fence_buf);
+	if (ret)
+		goto failed1;
+
+	ret = amdgpu_bo_create_kernel(adev, PSP_CMD_BUFFER_SIZE, PAGE_SIZE,
+				      AMDGPU_GEM_DOMAIN_VRAM |
+				      AMDGPU_GEM_DOMAIN_GTT,
+				      &psp->cmd_buf_bo, &psp->cmd_buf_mc_addr,
+				      (void **)&psp->cmd_buf_mem);
+	if (ret)
+		goto failed2;
+
+	return 0;
+
+failed2:
+	amdgpu_bo_free_kernel(&psp->fence_buf_bo,
+			      &psp->fence_buf_mc_addr, &psp->fence_buf);
+failed1:
+	amdgpu_bo_free_kernel(&psp->fw_pri_bo,
+			      &psp->fw_pri_mc_addr, &psp->fw_pri_buf);
+	return ret;
+}
+
+static int psp_sw_fini(struct amdgpu_ip_block *ip_block)
+{
+	struct amdgpu_device *adev = ip_block->adev;
+	struct psp_context *psp = &adev->psp;
+
+	psp_memory_training_fini(psp);
+
+	amdgpu_ucode_release(&psp->sos_fw);
+	amdgpu_ucode_release(&psp->asd_fw);
+	amdgpu_ucode_release(&psp->ta_fw);
+	amdgpu_ucode_release(&psp->cap_fw);
+	amdgpu_ucode_release(&psp->toc_fw);
+
+	kfree(psp->cmd);
+	psp->cmd = NULL;
+
+	psp_free_shared_bufs(psp);
+
+	if (psp->km_ring.ring_mem)
+		amdgpu_bo_free_kernel(&adev->firmware.rbuf,
+				      &psp->km_ring.ring_mem_mc_addr,
+				      (void **)&psp->km_ring.ring_mem);
+
+	amdgpu_bo_free_kernel(&psp->fw_pri_bo,
+			      &psp->fw_pri_mc_addr, &psp->fw_pri_buf);
+	amdgpu_bo_free_kernel(&psp->fence_buf_bo,
+			      &psp->fence_buf_mc_addr, &psp->fence_buf);
+	amdgpu_bo_free_kernel(&psp->cmd_buf_bo, &psp->cmd_buf_mc_addr,
+			      (void **)&psp->cmd_buf_mem);
+
 	return 0;
 }
 
-int psp_wait_for(struct psp_context *psp, uint32_t reg_index,
-		 uint32_t reg_val, uint32_t mask, bool check_changed)
+int psp_wait_for(struct psp_context *psp, uint32_t reg_index, uint32_t reg_val,
+		 uint32_t mask, uint32_t flags)
 {
+	bool check_changed = flags & PSP_WAITREG_CHANGED;
+	bool verbose = !(flags & PSP_WAITREG_NOVERBOSE);
 	uint32_t val;
 	int i;
 	struct amdgpu_device *adev = psp->adev;
 
-	val = RREG32(reg_index);
+	if (psp->adev->no_hw_access)
+		return 0;
 
 	for (i = 0; i < adev->usec_timeout; i++) {
+		val = RREG32(reg_index);
 		if (check_changed) {
 			if (val != reg_val)
 				return 0;
@@ -101,215 +599,2508 @@ int psp_wait_for(struct psp_context *psp, uint32_t reg_index,
 		udelay(1);
 	}
 
+	if (verbose)
+		dev_err(adev->dev,
+			"psp reg (0x%x) wait timed out, mask: %x, read: %x exp: %x",
+			reg_index, mask, val, reg_val);
+
 	return -ETIME;
 }
 
+int psp_wait_for_spirom_update(struct psp_context *psp, uint32_t reg_index,
+			       uint32_t reg_val, uint32_t mask, uint32_t msec_timeout)
+{
+	uint32_t val;
+	int i;
+	struct amdgpu_device *adev = psp->adev;
+
+	if (psp->adev->no_hw_access)
+		return 0;
+
+	for (i = 0; i < msec_timeout; i++) {
+		val = RREG32(reg_index);
+		if ((val & mask) == reg_val)
+			return 0;
+		msleep(1);
+	}
+
+	return -ETIME;
+}
+
+static const char *psp_gfx_cmd_name(enum psp_gfx_cmd_id cmd_id)
+{
+	switch (cmd_id) {
+	case GFX_CMD_ID_LOAD_TA:
+		return "LOAD_TA";
+	case GFX_CMD_ID_UNLOAD_TA:
+		return "UNLOAD_TA";
+	case GFX_CMD_ID_INVOKE_CMD:
+		return "INVOKE_CMD";
+	case GFX_CMD_ID_LOAD_ASD:
+		return "LOAD_ASD";
+	case GFX_CMD_ID_SETUP_TMR:
+		return "SETUP_TMR";
+	case GFX_CMD_ID_LOAD_IP_FW:
+		return "LOAD_IP_FW";
+	case GFX_CMD_ID_DESTROY_TMR:
+		return "DESTROY_TMR";
+	case GFX_CMD_ID_SAVE_RESTORE:
+		return "SAVE_RESTORE_IP_FW";
+	case GFX_CMD_ID_SETUP_VMR:
+		return "SETUP_VMR";
+	case GFX_CMD_ID_DESTROY_VMR:
+		return "DESTROY_VMR";
+	case GFX_CMD_ID_PROG_REG:
+		return "PROG_REG";
+	case GFX_CMD_ID_GET_FW_ATTESTATION:
+		return "GET_FW_ATTESTATION";
+	case GFX_CMD_ID_LOAD_TOC:
+		return "ID_LOAD_TOC";
+	case GFX_CMD_ID_AUTOLOAD_RLC:
+		return "AUTOLOAD_RLC";
+	case GFX_CMD_ID_BOOT_CFG:
+		return "BOOT_CFG";
+	case GFX_CMD_ID_CONFIG_SQ_PERFMON:
+		return "CONFIG_SQ_PERFMON";
+	case GFX_CMD_ID_FB_FW_RESERV_ADDR:
+		return "FB_FW_RESERV_ADDR";
+	case GFX_CMD_ID_FB_FW_RESERV_EXT_ADDR:
+		return "FB_FW_RESERV_EXT_ADDR";
+	case GFX_CMD_ID_SRIOV_SPATIAL_PART:
+		return "SPATIAL_PARTITION";
+	case GFX_CMD_ID_FB_NPS_MODE:
+		return "NPS_MODE_CHANGE";
+	default:
+		return "UNKNOWN CMD";
+	}
+}
+
+static bool psp_err_warn(struct psp_context *psp)
+{
+	struct psp_gfx_cmd_resp *cmd = psp->cmd_buf_mem;
+
+	/* This response indicates reg list is already loaded */
+	if (amdgpu_ip_version(psp->adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 2) &&
+	    cmd->cmd_id == GFX_CMD_ID_LOAD_IP_FW &&
+	    cmd->cmd.cmd_load_ip_fw.fw_type == GFX_FW_TYPE_REG_LIST &&
+	    cmd->resp.status == TEE_ERROR_CANCEL)
+		return false;
+
+	return true;
+}
+
 static int
 psp_cmd_submit_buf(struct psp_context *psp,
 		   struct amdgpu_firmware_info *ucode,
-		   struct psp_gfx_cmd_resp *cmd, uint64_t fence_mc_addr,
-		   int index)
+		   struct psp_gfx_cmd_resp *cmd, uint64_t fence_mc_addr)
 {
 	int ret;
-	struct amdgpu_bo *cmd_buf_bo;
-	uint64_t cmd_buf_mc_addr;
-	struct psp_gfx_cmd_resp *cmd_buf_mem;
-	struct amdgpu_device *adev = psp->adev;
+	int index;
+	int timeout = psp->adev->psp_timeout;
+	bool ras_intr = false;
+	bool skip_unsupport = false;
 
-	ret = amdgpu_bo_create_kernel(adev, PSP_CMD_BUFFER_SIZE, PAGE_SIZE,
-				      AMDGPU_GEM_DOMAIN_VRAM,
-				      &cmd_buf_bo, &cmd_buf_mc_addr,
-				      (void **)&cmd_buf_mem);
-	if (ret)
-		return ret;
+	if (psp->adev->no_hw_access)
+		return 0;
 
-	memset(cmd_buf_mem, 0, PSP_CMD_BUFFER_SIZE);
+	memset(psp->cmd_buf_mem, 0, PSP_CMD_BUFFER_SIZE);
 
-	memcpy(cmd_buf_mem, cmd, sizeof(struct psp_gfx_cmd_resp));
+	memcpy(psp->cmd_buf_mem, cmd, sizeof(struct psp_gfx_cmd_resp));
 
-	ret = psp_cmd_submit(psp, ucode, cmd_buf_mc_addr,
-			     fence_mc_addr, index);
+	index = atomic_inc_return(&psp->fence_value);
+	ret = psp_ring_cmd_submit(psp, psp->cmd_buf_mc_addr, fence_mc_addr, index);
+	if (ret) {
+		atomic_dec(&psp->fence_value);
+		goto exit;
+	}
 
+	amdgpu_device_invalidate_hdp(psp->adev, NULL);
 	while (*((unsigned int *)psp->fence_buf) != index) {
-		msleep(1);
+		if (--timeout == 0)
+			break;
+		/*
+		 * Shouldn't wait for timeout when err_event_athub occurs,
+		 * because gpu reset thread triggered and lock resource should
+		 * be released for psp resume sequence.
+		 */
+		ras_intr = amdgpu_ras_intr_triggered();
+		if (ras_intr)
+			break;
+		usleep_range(10, 100);
+		amdgpu_device_invalidate_hdp(psp->adev, NULL);
+	}
+
+	/* We allow TEE_ERROR_NOT_SUPPORTED for VMR command and PSP_ERR_UNKNOWN_COMMAND in SRIOV */
+	skip_unsupport = (psp->cmd_buf_mem->resp.status == TEE_ERROR_NOT_SUPPORTED ||
+		psp->cmd_buf_mem->resp.status == PSP_ERR_UNKNOWN_COMMAND) && amdgpu_sriov_vf(psp->adev);
+
+	memcpy(&cmd->resp, &psp->cmd_buf_mem->resp, sizeof(struct psp_gfx_resp));
+
+	/* In some cases, psp response status is not 0 even there is no
+	 * problem while the command is submitted. Some version of PSP FW
+	 * doesn't write 0 to that field.
+	 * So here we would like to only print a warning instead of an error
+	 * during psp initialization to avoid breaking hw_init and it doesn't
+	 * return -EINVAL.
+	 */
+	if (!skip_unsupport && (psp->cmd_buf_mem->resp.status || !timeout) && !ras_intr) {
+		if (ucode)
+			dev_warn(psp->adev->dev,
+				 "failed to load ucode %s(0x%X) ",
+				 amdgpu_ucode_name(ucode->ucode_id), ucode->ucode_id);
+		if (psp_err_warn(psp))
+			dev_warn(
+				psp->adev->dev,
+				"psp gfx command %s(0x%X) failed and response status is (0x%X)\n",
+				psp_gfx_cmd_name(psp->cmd_buf_mem->cmd_id),
+				psp->cmd_buf_mem->cmd_id,
+				psp->cmd_buf_mem->resp.status);
+		/* If any firmware (including CAP) load fails under SRIOV, it should
+		 * return failure to stop the VF from initializing.
+		 * Also return failure in case of timeout
+		 */
+		if ((ucode && amdgpu_sriov_vf(psp->adev)) || !timeout) {
+			ret = -EINVAL;
+			goto exit;
+		}
 	}
 
-	amdgpu_bo_free_kernel(&cmd_buf_bo,
-			      &cmd_buf_mc_addr,
-			      (void **)&cmd_buf_mem);
+	if (ucode) {
+		ucode->tmr_mc_addr_lo = psp->cmd_buf_mem->resp.fw_addr_lo;
+		ucode->tmr_mc_addr_hi = psp->cmd_buf_mem->resp.fw_addr_hi;
+	}
 
+exit:
 	return ret;
 }
 
-static void psp_prep_tmr_cmd_buf(struct psp_gfx_cmd_resp *cmd,
-				 uint64_t tmr_mc, uint32_t size)
+static struct psp_gfx_cmd_resp *acquire_psp_cmd_buf(struct psp_context *psp)
+{
+	struct psp_gfx_cmd_resp *cmd = psp->cmd;
+
+	mutex_lock(&psp->mutex);
+
+	memset(cmd, 0, sizeof(struct psp_gfx_cmd_resp));
+
+	return cmd;
+}
+
+static void release_psp_cmd_buf(struct psp_context *psp)
 {
-	cmd->cmd_id = GFX_CMD_ID_SETUP_TMR;
-	cmd->cmd.cmd_setup_tmr.buf_phy_addr_lo = (uint32_t)tmr_mc;
-	cmd->cmd.cmd_setup_tmr.buf_phy_addr_hi = (uint32_t)(tmr_mc >> 32);
+	mutex_unlock(&psp->mutex);
+}
+
+static void psp_prep_tmr_cmd_buf(struct psp_context *psp,
+				 struct psp_gfx_cmd_resp *cmd,
+				 uint64_t tmr_mc, struct amdgpu_bo *tmr_bo)
+{
+	struct amdgpu_device *adev = psp->adev;
+	uint32_t size = 0;
+	uint64_t tmr_pa = 0;
+
+	if (tmr_bo) {
+		size = amdgpu_bo_size(tmr_bo);
+		tmr_pa = amdgpu_gmc_vram_pa(adev, tmr_bo);
+	}
+
+	if (amdgpu_sriov_vf(psp->adev))
+		cmd->cmd_id = GFX_CMD_ID_SETUP_VMR;
+	else
+		cmd->cmd_id = GFX_CMD_ID_SETUP_TMR;
+	cmd->cmd.cmd_setup_tmr.buf_phy_addr_lo = lower_32_bits(tmr_mc);
+	cmd->cmd.cmd_setup_tmr.buf_phy_addr_hi = upper_32_bits(tmr_mc);
 	cmd->cmd.cmd_setup_tmr.buf_size = size;
+	cmd->cmd.cmd_setup_tmr.bitfield.virt_phy_addr = 1;
+	cmd->cmd.cmd_setup_tmr.system_phy_addr_lo = lower_32_bits(tmr_pa);
+	cmd->cmd.cmd_setup_tmr.system_phy_addr_hi = upper_32_bits(tmr_pa);
+}
+
+static void psp_prep_load_toc_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+				      uint64_t pri_buf_mc, uint32_t size)
+{
+	cmd->cmd_id = GFX_CMD_ID_LOAD_TOC;
+	cmd->cmd.cmd_load_toc.toc_phy_addr_lo = lower_32_bits(pri_buf_mc);
+	cmd->cmd.cmd_load_toc.toc_phy_addr_hi = upper_32_bits(pri_buf_mc);
+	cmd->cmd.cmd_load_toc.toc_size = size;
+}
+
+/* Issue LOAD TOC cmd to PSP to part toc and calculate tmr size needed */
+static int psp_load_toc(struct psp_context *psp,
+			uint32_t *tmr_size)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd = acquire_psp_cmd_buf(psp);
+
+	/* Copy toc to psp firmware private buffer */
+	psp_copy_fw(psp, psp->toc.start_addr, psp->toc.size_bytes);
+
+	psp_prep_load_toc_cmd_buf(cmd, psp->fw_pri_mc_addr, psp->toc.size_bytes);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+				 psp->fence_buf_mc_addr);
+	if (!ret)
+		*tmr_size = psp->cmd_buf_mem->resp.tmr_size;
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
 }
 
 /* Set up Trusted Memory Region */
 static int psp_tmr_init(struct psp_context *psp)
 {
-	int ret;
+	int ret = 0;
+	int tmr_size;
+	void *tmr_buf;
+	void **pptr;
 
 	/*
-	 * Allocate 3M memory aligned to 1M from Frame Buffer (local
-	 * physical).
+	 * According to HW engineer, they prefer the TMR address be "naturally
+	 * aligned" , e.g. the start address be an integer divide of TMR size.
 	 *
 	 * Note: this memory need be reserved till the driver
 	 * uninitializes.
 	 */
-	ret = amdgpu_bo_create_kernel(psp->adev, 0x300000, 0x100000,
-				      AMDGPU_GEM_DOMAIN_VRAM,
-				      &psp->tmr_bo, &psp->tmr_mc_addr, &psp->tmr_buf);
+	tmr_size = PSP_TMR_SIZE(psp->adev);
+
+	/* For ASICs support RLC autoload, psp will parse the toc
+	 * and calculate the total size of TMR needed
+	 */
+	if (!amdgpu_sriov_vf(psp->adev) &&
+	    psp->toc.start_addr &&
+	    psp->toc.size_bytes &&
+	    psp->fw_pri_buf) {
+		ret = psp_load_toc(psp, &tmr_size);
+		if (ret) {
+			dev_err(psp->adev->dev, "Failed to load toc\n");
+			return ret;
+		}
+	}
+
+	if (!psp->tmr_bo && !psp->boot_time_tmr) {
+		pptr = amdgpu_sriov_vf(psp->adev) ? &tmr_buf : NULL;
+		ret = amdgpu_bo_create_kernel(psp->adev, tmr_size,
+					      PSP_TMR_ALIGNMENT,
+					      AMDGPU_GEM_DOMAIN_GTT | AMDGPU_GEM_DOMAIN_VRAM,
+					      &psp->tmr_bo, &psp->tmr_mc_addr,
+					      pptr);
+	}
+	if (amdgpu_virt_xgmi_migrate_enabled(psp->adev) && psp->tmr_bo)
+		psp->tmr_mc_addr = amdgpu_bo_fb_aper_addr(psp->tmr_bo);
 
 	return ret;
 }
 
+static bool psp_skip_tmr(struct psp_context *psp)
+{
+	switch (amdgpu_ip_version(psp->adev, MP0_HWIP, 0)) {
+	case IP_VERSION(11, 0, 9):
+	case IP_VERSION(11, 0, 7):
+	case IP_VERSION(13, 0, 2):
+	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 10):
+	case IP_VERSION(13, 0, 12):
+	case IP_VERSION(13, 0, 14):
+		return true;
+	default:
+		return false;
+	}
+}
+
 static int psp_tmr_load(struct psp_context *psp)
 {
 	int ret;
 	struct psp_gfx_cmd_resp *cmd;
 
-	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
-	if (!cmd)
-		return -ENOMEM;
+	/* For Navi12 and CHIP_SIENNA_CICHLID SRIOV, do not set up TMR.
+	 * Already set up by host driver.
+	 */
+	if (amdgpu_sriov_vf(psp->adev) && psp_skip_tmr(psp))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
 
-	psp_prep_tmr_cmd_buf(cmd, psp->tmr_mc_addr, 0x300000);
+	psp_prep_tmr_cmd_buf(psp, cmd, psp->tmr_mc_addr, psp->tmr_bo);
+	if (psp->tmr_bo)
+		dev_info(psp->adev->dev, "reserve 0x%lx from 0x%llx for PSP TMR\n",
+			 amdgpu_bo_size(psp->tmr_bo), psp->tmr_mc_addr);
 
 	ret = psp_cmd_submit_buf(psp, NULL, cmd,
-				 psp->fence_buf_mc_addr, 1);
+				 psp->fence_buf_mc_addr);
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static void psp_prep_tmr_unload_cmd_buf(struct psp_context *psp,
+					struct psp_gfx_cmd_resp *cmd)
+{
+	if (amdgpu_sriov_vf(psp->adev))
+		cmd->cmd_id = GFX_CMD_ID_DESTROY_VMR;
+	else
+		cmd->cmd_id = GFX_CMD_ID_DESTROY_TMR;
+}
+
+static int psp_tmr_unload(struct psp_context *psp)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd;
+
+	/* skip TMR unload for Navi12 and CHIP_SIENNA_CICHLID SRIOV,
+	 * as TMR is not loaded at all
+	 */
+	if (amdgpu_sriov_vf(psp->adev) && psp_skip_tmr(psp))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	psp_prep_tmr_unload_cmd_buf(psp, cmd);
+	dev_dbg(psp->adev->dev, "free PSP TMR buffer\n");
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+				 psp->fence_buf_mc_addr);
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static int psp_tmr_terminate(struct psp_context *psp)
+{
+	return psp_tmr_unload(psp);
+}
+
+int psp_get_fw_attestation_records_addr(struct psp_context *psp,
+					uint64_t *output_ptr)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd;
+
+	if (!output_ptr)
+		return -EINVAL;
+
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id = GFX_CMD_ID_GET_FW_ATTESTATION;
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+				 psp->fence_buf_mc_addr);
+
+	if (!ret) {
+		*output_ptr = ((uint64_t)cmd->resp.uresp.fwar_db_info.fwar_db_addr_lo) +
+			      ((uint64_t)cmd->resp.uresp.fwar_db_info.fwar_db_addr_hi << 32);
+	}
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static int psp_get_fw_reservation_info(struct psp_context *psp,
+						   uint32_t cmd_id,
+						   uint64_t *addr,
+						   uint32_t *size)
+{
+	int ret;
+	uint32_t status;
+	struct psp_gfx_cmd_resp *cmd;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id = cmd_id;
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+				 psp->fence_buf_mc_addr);
+	if (ret) {
+		release_psp_cmd_buf(psp);
+		return ret;
+	}
+
+	status = cmd->resp.status;
+	if (status == PSP_ERR_UNKNOWN_COMMAND) {
+		release_psp_cmd_buf(psp);
+		*addr = 0;
+		*size = 0;
+		return 0;
+	}
+
+	*addr = (uint64_t)cmd->resp.uresp.fw_reserve_info.reserve_base_address_hi << 32 |
+		cmd->resp.uresp.fw_reserve_info.reserve_base_address_lo;
+	*size = cmd->resp.uresp.fw_reserve_info.reserve_size;
+
+	release_psp_cmd_buf(psp);
+
+	return 0;
+}
+
+int psp_update_fw_reservation(struct psp_context *psp)
+{
+	int ret;
+	uint64_t reserv_addr, reserv_addr_ext;
+	uint32_t reserv_size, reserv_size_ext, mp0_ip_ver;
+	struct amdgpu_device *adev = psp->adev;
+
+	mp0_ip_ver = amdgpu_ip_version(adev, MP0_HWIP, 0);
+
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	switch (mp0_ip_ver) {
+	case IP_VERSION(14, 0, 2):
+		if (adev->psp.sos.fw_version < 0x3b0e0d)
+			return 0;
+		break;
+
+	case IP_VERSION(14, 0, 3):
+		if (adev->psp.sos.fw_version < 0x3a0e14)
+			return 0;
+		break;
+
+	default:
+		return 0;
+	}
+
+	ret = psp_get_fw_reservation_info(psp, GFX_CMD_ID_FB_FW_RESERV_ADDR, &reserv_addr, &reserv_size);
 	if (ret)
-		goto failed;
+		return ret;
+	ret = psp_get_fw_reservation_info(psp, GFX_CMD_ID_FB_FW_RESERV_EXT_ADDR, &reserv_addr_ext, &reserv_size_ext);
+	if (ret)
+		return ret;
+
+	if (reserv_addr != adev->gmc.real_vram_size - reserv_size) {
+		dev_warn(adev->dev, "reserve fw region is not valid!\n");
+		return 0;
+	}
+
+	amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory, NULL, NULL);
 
-	kfree(cmd);
+	reserv_size = roundup(reserv_size, SZ_1M);
+
+	ret = amdgpu_bo_create_kernel_at(adev, reserv_addr, reserv_size, &adev->mman.fw_reserved_memory, NULL);
+	if (ret) {
+		dev_err(adev->dev, "reserve fw region failed(%d)!\n", ret);
+		amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory, NULL, NULL);
+		return ret;
+	}
+
+	reserv_size_ext = roundup(reserv_size_ext, SZ_1M);
+
+	ret = amdgpu_bo_create_kernel_at(adev, reserv_addr_ext, reserv_size_ext,
+					 &adev->mman.fw_reserved_memory_extend, NULL);
+	if (ret) {
+		dev_err(adev->dev, "reserve extend fw region failed(%d)!\n", ret);
+		amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory_extend, NULL, NULL);
+		return ret;
+	}
 
 	return 0;
+}
+
+static int psp_boot_config_get(struct amdgpu_device *adev, uint32_t *boot_cfg)
+{
+	struct psp_context *psp = &adev->psp;
+	struct psp_gfx_cmd_resp *cmd;
+	int ret;
+
+	if (amdgpu_sriov_vf(adev))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id = GFX_CMD_ID_BOOT_CFG;
+	cmd->cmd.boot_cfg.sub_cmd = BOOTCFG_CMD_GET;
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+	if (!ret) {
+		*boot_cfg =
+			(cmd->resp.uresp.boot_cfg.boot_cfg & BOOT_CONFIG_GECC) ? 1 : 0;
+	}
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static int psp_boot_config_set(struct amdgpu_device *adev, uint32_t boot_cfg)
+{
+	int ret;
+	struct psp_context *psp = &adev->psp;
+	struct psp_gfx_cmd_resp *cmd;
+
+	if (amdgpu_sriov_vf(adev))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id = GFX_CMD_ID_BOOT_CFG;
+	cmd->cmd.boot_cfg.sub_cmd = BOOTCFG_CMD_SET;
+	cmd->cmd.boot_cfg.boot_config = boot_cfg;
+	cmd->cmd.boot_cfg.boot_config_valid = boot_cfg;
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+
+	release_psp_cmd_buf(psp);
 
-failed:
-	kfree(cmd);
 	return ret;
 }
 
-static void psp_prep_asd_cmd_buf(struct psp_gfx_cmd_resp *cmd,
-				 uint64_t asd_mc, uint64_t asd_mc_shared,
-				 uint32_t size, uint32_t shared_size)
+static int psp_rl_load(struct amdgpu_device *adev)
 {
-	cmd->cmd_id = GFX_CMD_ID_LOAD_ASD;
-	cmd->cmd.cmd_load_ta.app_phy_addr_lo = lower_32_bits(asd_mc);
-	cmd->cmd.cmd_load_ta.app_phy_addr_hi = upper_32_bits(asd_mc);
-	cmd->cmd.cmd_load_ta.app_len = size;
+	int ret;
+	struct psp_context *psp = &adev->psp;
+	struct psp_gfx_cmd_resp *cmd;
 
-	cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_lo = lower_32_bits(asd_mc_shared);
-	cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_hi = upper_32_bits(asd_mc_shared);
-	cmd->cmd.cmd_load_ta.cmd_buf_len = shared_size;
+	if (!is_psp_fw_valid(psp->rl))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	memset(psp->fw_pri_buf, 0, PSP_1_MEG);
+	memcpy(psp->fw_pri_buf, psp->rl.start_addr, psp->rl.size_bytes);
+
+	cmd->cmd_id = GFX_CMD_ID_LOAD_IP_FW;
+	cmd->cmd.cmd_load_ip_fw.fw_phy_addr_lo = lower_32_bits(psp->fw_pri_mc_addr);
+	cmd->cmd.cmd_load_ip_fw.fw_phy_addr_hi = upper_32_bits(psp->fw_pri_mc_addr);
+	cmd->cmd.cmd_load_ip_fw.fw_size = psp->rl.size_bytes;
+	cmd->cmd.cmd_load_ip_fw.fw_type = GFX_FW_TYPE_REG_LIST;
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
 }
 
-static int psp_asd_init(struct psp_context *psp)
+int psp_memory_partition(struct psp_context *psp, int mode)
 {
+	struct psp_gfx_cmd_resp *cmd;
 	int ret;
 
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id = GFX_CMD_ID_FB_NPS_MODE;
+	cmd->cmd.cmd_memory_part.mode = mode;
+
+	dev_info(psp->adev->dev,
+		 "Requesting %d memory partition change through PSP", mode);
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+	if (ret)
+		dev_err(psp->adev->dev,
+			"PSP request failed to change to NPS%d mode\n", mode);
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+int psp_spatial_partition(struct psp_context *psp, int mode)
+{
+	struct psp_gfx_cmd_resp *cmd;
+	int ret;
+
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id = GFX_CMD_ID_SRIOV_SPATIAL_PART;
+	cmd->cmd.cmd_spatial_part.mode = mode;
+
+	dev_info(psp->adev->dev, "Requesting %d partitions through PSP", mode);
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static int psp_asd_initialize(struct psp_context *psp)
+{
+	int ret;
+
+	/* If PSP version doesn't match ASD version, asd loading will be failed.
+	 * add workaround to bypass it for sriov now.
+	 * TODO: add version check to make it common
+	 */
+	if (amdgpu_sriov_vf(psp->adev) || !psp->asd_context.bin_desc.size_bytes)
+		return 0;
+
+	/* bypass asd if display hardware is not available */
+	if (!amdgpu_device_has_display_hardware(psp->adev) &&
+	    amdgpu_ip_version(psp->adev, MP0_HWIP, 0) >= IP_VERSION(13, 0, 10))
+		return 0;
+
+	psp->asd_context.mem_context.shared_mc_addr  = 0;
+	psp->asd_context.mem_context.shared_mem_size = PSP_ASD_SHARED_MEM_SIZE;
+	psp->asd_context.ta_load_type                = GFX_CMD_ID_LOAD_ASD;
+
+	ret = psp_ta_load(psp, &psp->asd_context);
+	if (!ret)
+		psp->asd_context.initialized = true;
+
+	return ret;
+}
+
+static void psp_prep_ta_unload_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+				       uint32_t session_id)
+{
+	cmd->cmd_id = GFX_CMD_ID_UNLOAD_TA;
+	cmd->cmd.cmd_unload_ta.session_id = session_id;
+}
+
+int psp_ta_unload(struct psp_context *psp, struct ta_context *context)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd = acquire_psp_cmd_buf(psp);
+
+	psp_prep_ta_unload_cmd_buf(cmd, context->session_id);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+
+	context->resp_status = cmd->resp.status;
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static int psp_asd_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->asd_context.initialized)
+		return 0;
+
+	ret = psp_ta_unload(psp, &psp->asd_context);
+	if (!ret)
+		psp->asd_context.initialized = false;
+
+	return ret;
+}
+
+static void psp_prep_reg_prog_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+		uint32_t id, uint32_t value)
+{
+	cmd->cmd_id = GFX_CMD_ID_PROG_REG;
+	cmd->cmd.cmd_setup_reg_prog.reg_value = value;
+	cmd->cmd.cmd_setup_reg_prog.reg_id = id;
+}
+
+int psp_reg_program(struct psp_context *psp, enum psp_reg_prog_id reg,
+		uint32_t value)
+{
+	struct psp_gfx_cmd_resp *cmd;
+	int ret = 0;
+
+	if (reg >= PSP_REG_LAST)
+		return -EINVAL;
+
+	cmd = acquire_psp_cmd_buf(psp);
+
+	psp_prep_reg_prog_cmd_buf(cmd, reg, value);
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+	if (ret)
+		dev_err(psp->adev->dev, "PSP failed to program reg id %d\n", reg);
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static void psp_prep_ta_load_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+				     uint64_t ta_bin_mc,
+				     struct ta_context *context)
+{
+	cmd->cmd_id				= context->ta_load_type;
+	cmd->cmd.cmd_load_ta.app_phy_addr_lo	= lower_32_bits(ta_bin_mc);
+	cmd->cmd.cmd_load_ta.app_phy_addr_hi	= upper_32_bits(ta_bin_mc);
+	cmd->cmd.cmd_load_ta.app_len		= context->bin_desc.size_bytes;
+
+	cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_lo =
+		lower_32_bits(context->mem_context.shared_mc_addr);
+	cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_hi =
+		upper_32_bits(context->mem_context.shared_mc_addr);
+	cmd->cmd.cmd_load_ta.cmd_buf_len = context->mem_context.shared_mem_size;
+}
+
+int psp_ta_init_shared_buf(struct psp_context *psp,
+				  struct ta_mem_context *mem_ctx)
+{
 	/*
 	 * Allocate 16k memory aligned to 4k from Frame Buffer (local
-	 * physical) for shared ASD <-> Driver
+	 * physical) for ta to host memory
 	 */
-	ret = amdgpu_bo_create_kernel(psp->adev, PSP_ASD_SHARED_MEM_SIZE,
-				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM,
-				      &psp->asd_shared_bo,
-				      &psp->asd_shared_mc_addr,
-				      &psp->asd_shared_buf);
+	return amdgpu_bo_create_kernel(psp->adev, mem_ctx->shared_mem_size,
+				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM |
+				      AMDGPU_GEM_DOMAIN_GTT,
+				      &mem_ctx->shared_bo,
+				      &mem_ctx->shared_mc_addr,
+				      &mem_ctx->shared_buf);
+}
+
+static void psp_prep_ta_invoke_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+				       uint32_t ta_cmd_id,
+				       uint32_t session_id)
+{
+	cmd->cmd_id				= GFX_CMD_ID_INVOKE_CMD;
+	cmd->cmd.cmd_invoke_cmd.session_id	= session_id;
+	cmd->cmd.cmd_invoke_cmd.ta_cmd_id	= ta_cmd_id;
+}
+
+int psp_ta_invoke(struct psp_context *psp,
+		  uint32_t ta_cmd_id,
+		  struct ta_context *context)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd = acquire_psp_cmd_buf(psp);
+
+	psp_prep_ta_invoke_cmd_buf(cmd, ta_cmd_id, context->session_id);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+				 psp->fence_buf_mc_addr);
+
+	context->resp_status = cmd->resp.status;
+
+	release_psp_cmd_buf(psp);
 
 	return ret;
 }
 
-static int psp_asd_load(struct psp_context *psp)
+int psp_ta_load(struct psp_context *psp, struct ta_context *context)
 {
 	int ret;
 	struct psp_gfx_cmd_resp *cmd;
 
-	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
-	if (!cmd)
-		return -ENOMEM;
+	cmd = acquire_psp_cmd_buf(psp);
 
-	memset(psp->fw_pri_buf, 0, PSP_1_MEG);
-	memcpy(psp->fw_pri_buf, psp->asd_start_addr, psp->asd_ucode_size);
+	psp_copy_fw(psp, context->bin_desc.start_addr,
+		    context->bin_desc.size_bytes);
 
-	psp_prep_asd_cmd_buf(cmd, psp->fw_pri_mc_addr, psp->asd_shared_mc_addr,
-			     psp->asd_ucode_size, PSP_ASD_SHARED_MEM_SIZE);
+	if (amdgpu_virt_xgmi_migrate_enabled(psp->adev) &&
+		context->mem_context.shared_bo)
+		context->mem_context.shared_mc_addr =
+			amdgpu_bo_fb_aper_addr(context->mem_context.shared_bo);
+
+	psp_prep_ta_load_cmd_buf(cmd, psp->fw_pri_mc_addr, context);
 
 	ret = psp_cmd_submit_buf(psp, NULL, cmd,
-				 psp->fence_buf_mc_addr, 2);
+				 psp->fence_buf_mc_addr);
+
+	context->resp_status = cmd->resp.status;
+
+	if (!ret)
+		context->session_id = cmd->resp.session_id;
 
-	kfree(cmd);
+	release_psp_cmd_buf(psp);
 
 	return ret;
 }
 
-static int psp_hw_start(struct psp_context *psp)
+int psp_xgmi_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
+{
+	return psp_ta_invoke(psp, ta_cmd_id, &psp->xgmi_context.context);
+}
+
+int psp_xgmi_terminate(struct psp_context *psp)
 {
 	int ret;
+	struct amdgpu_device *adev = psp->adev;
+
+	/* XGMI TA unload currently is not supported on Arcturus/Aldebaran A+A */
+	if (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(11, 0, 4) ||
+	    (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 2) &&
+	     adev->gmc.xgmi.connected_to_cpu))
+		return 0;
 
-	ret = psp_bootloader_load_sysdrv(psp);
+	if (!psp->xgmi_context.context.initialized)
+		return 0;
+
+	ret = psp_ta_unload(psp, &psp->xgmi_context.context);
+
+	psp->xgmi_context.context.initialized = false;
+
+	return ret;
+}
+
+int psp_xgmi_initialize(struct psp_context *psp, bool set_extended_data, bool load_ta)
+{
+	struct ta_xgmi_shared_memory *xgmi_cmd;
+	int ret;
+
+	if (!psp->ta_fw ||
+	    !psp->xgmi_context.context.bin_desc.size_bytes ||
+	    !psp->xgmi_context.context.bin_desc.start_addr)
+		return -ENOENT;
+
+	if (!load_ta)
+		goto invoke;
+
+	psp->xgmi_context.context.mem_context.shared_mem_size = PSP_XGMI_SHARED_MEM_SIZE;
+	psp->xgmi_context.context.ta_load_type = GFX_CMD_ID_LOAD_TA;
+
+	if (!psp->xgmi_context.context.mem_context.shared_buf) {
+		ret = psp_ta_init_shared_buf(psp, &psp->xgmi_context.context.mem_context);
+		if (ret)
+			return ret;
+	}
+
+	/* Load XGMI TA */
+	ret = psp_ta_load(psp, &psp->xgmi_context.context);
+	if (!ret)
+		psp->xgmi_context.context.initialized = true;
+	else
+		return ret;
+
+invoke:
+	/* Initialize XGMI session */
+	xgmi_cmd = (struct ta_xgmi_shared_memory *)(psp->xgmi_context.context.mem_context.shared_buf);
+	memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory));
+	xgmi_cmd->flag_extend_link_record = set_extended_data;
+	xgmi_cmd->cmd_id = TA_COMMAND_XGMI__INITIALIZE;
+
+	ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id);
+	/* note down the capbility flag for XGMI TA */
+	psp->xgmi_context.xgmi_ta_caps = xgmi_cmd->caps_flag;
+
+	return ret;
+}
+
+int psp_xgmi_get_hive_id(struct psp_context *psp, uint64_t *hive_id)
+{
+	struct ta_xgmi_shared_memory *xgmi_cmd;
+	int ret;
+
+	xgmi_cmd = (struct ta_xgmi_shared_memory *)psp->xgmi_context.context.mem_context.shared_buf;
+	memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory));
+
+	xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_HIVE_ID;
+
+	/* Invoke xgmi ta to get hive id */
+	ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id);
 	if (ret)
 		return ret;
 
-	ret = psp_bootloader_load_sos(psp);
+	*hive_id = xgmi_cmd->xgmi_out_message.get_hive_id.hive_id;
+
+	return 0;
+}
+
+int psp_xgmi_get_node_id(struct psp_context *psp, uint64_t *node_id)
+{
+	struct ta_xgmi_shared_memory *xgmi_cmd;
+	int ret;
+
+	xgmi_cmd = (struct ta_xgmi_shared_memory *)psp->xgmi_context.context.mem_context.shared_buf;
+	memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory));
+
+	xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_NODE_ID;
+
+	/* Invoke xgmi ta to get the node id */
+	ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id);
 	if (ret)
 		return ret;
 
-	ret = psp_ring_create(psp, PSP_RING_TYPE__KM);
+	*node_id = xgmi_cmd->xgmi_out_message.get_node_id.node_id;
+
+	return 0;
+}
+
+static bool psp_xgmi_peer_link_info_supported(struct psp_context *psp)
+{
+	return (amdgpu_ip_version(psp->adev, MP0_HWIP, 0) ==
+			IP_VERSION(13, 0, 2) &&
+		psp->xgmi_context.context.bin_desc.fw_version >= 0x2000000b) ||
+	       amdgpu_ip_version(psp->adev, MP0_HWIP, 0) >=
+		       IP_VERSION(13, 0, 6);
+}
+
+/*
+ * Chips that support extended topology information require the driver to
+ * reflect topology information in the opposite direction.  This is
+ * because the TA has already exceeded its link record limit and if the
+ * TA holds bi-directional information, the driver would have to do
+ * multiple fetches instead of just two.
+ */
+static void psp_xgmi_reflect_topology_info(struct psp_context *psp,
+					struct psp_xgmi_node_info node_info)
+{
+	struct amdgpu_device *mirror_adev;
+	struct amdgpu_hive_info *hive;
+	uint64_t src_node_id = psp->adev->gmc.xgmi.node_id;
+	uint64_t dst_node_id = node_info.node_id;
+	uint8_t dst_num_hops = node_info.num_hops;
+	uint8_t dst_is_sharing_enabled = node_info.is_sharing_enabled;
+	uint8_t dst_num_links = node_info.num_links;
+
+	hive = amdgpu_get_xgmi_hive(psp->adev);
+	if (WARN_ON(!hive))
+		return;
+
+	list_for_each_entry(mirror_adev, &hive->device_list, gmc.xgmi.head) {
+		struct psp_xgmi_topology_info *mirror_top_info;
+		int j;
+
+		if (mirror_adev->gmc.xgmi.node_id != dst_node_id)
+			continue;
+
+		mirror_top_info = &mirror_adev->psp.xgmi_context.top_info;
+		for (j = 0; j < mirror_top_info->num_nodes; j++) {
+			if (mirror_top_info->nodes[j].node_id != src_node_id)
+				continue;
+
+			mirror_top_info->nodes[j].num_hops = dst_num_hops;
+			mirror_top_info->nodes[j].is_sharing_enabled = dst_is_sharing_enabled;
+			/* prevent 0 num_links value re-reflection since reflection
+			 * criteria is based on num_hops (direct or indirect).
+			 */
+			if (dst_num_links) {
+				mirror_top_info->nodes[j].num_links = dst_num_links;
+				/* swap src and dst due to frame of reference */
+				for (int k = 0; k < dst_num_links; k++) {
+					mirror_top_info->nodes[j].port_num[k].src_xgmi_port_num =
+						node_info.port_num[k].dst_xgmi_port_num;
+					mirror_top_info->nodes[j].port_num[k].dst_xgmi_port_num =
+						node_info.port_num[k].src_xgmi_port_num;
+				}
+			}
+
+			break;
+		}
+
+		break;
+	}
+
+	amdgpu_put_xgmi_hive(hive);
+}
+
+int psp_xgmi_get_topology_info(struct psp_context *psp,
+			       int number_devices,
+			       struct psp_xgmi_topology_info *topology,
+			       bool get_extended_data)
+{
+	struct ta_xgmi_shared_memory *xgmi_cmd;
+	struct ta_xgmi_cmd_get_topology_info_input *topology_info_input;
+	struct ta_xgmi_cmd_get_topology_info_output *topology_info_output;
+	int i;
+	int ret;
+
+	if (!topology || topology->num_nodes > TA_XGMI__MAX_CONNECTED_NODES)
+		return -EINVAL;
+
+	xgmi_cmd = (struct ta_xgmi_shared_memory *)psp->xgmi_context.context.mem_context.shared_buf;
+	memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory));
+	xgmi_cmd->flag_extend_link_record = get_extended_data;
+
+	/* Fill in the shared memory with topology information as input */
+	topology_info_input = &xgmi_cmd->xgmi_in_message.get_topology_info;
+	xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_TOPOLOGY_INFO;
+	topology_info_input->num_nodes = number_devices;
+
+	for (i = 0; i < topology_info_input->num_nodes; i++) {
+		topology_info_input->nodes[i].node_id = topology->nodes[i].node_id;
+		topology_info_input->nodes[i].num_hops = topology->nodes[i].num_hops;
+		topology_info_input->nodes[i].is_sharing_enabled = topology->nodes[i].is_sharing_enabled;
+		topology_info_input->nodes[i].sdma_engine = topology->nodes[i].sdma_engine;
+	}
+
+	/* Invoke xgmi ta to get the topology information */
+	ret = psp_xgmi_invoke(psp, TA_COMMAND_XGMI__GET_TOPOLOGY_INFO);
 	if (ret)
 		return ret;
 
-	ret = psp_tmr_load(psp);
+	/* Read the output topology information from the shared memory */
+	topology_info_output = &xgmi_cmd->xgmi_out_message.get_topology_info;
+	topology->num_nodes = xgmi_cmd->xgmi_out_message.get_topology_info.num_nodes;
+	for (i = 0; i < topology->num_nodes; i++) {
+		/* extended data will either be 0 or equal to non-extended data */
+		if (topology_info_output->nodes[i].num_hops)
+			topology->nodes[i].num_hops = topology_info_output->nodes[i].num_hops;
+
+		/* non-extended data gets everything here so no need to update */
+		if (!get_extended_data) {
+			topology->nodes[i].node_id = topology_info_output->nodes[i].node_id;
+			topology->nodes[i].is_sharing_enabled =
+					topology_info_output->nodes[i].is_sharing_enabled;
+			topology->nodes[i].sdma_engine =
+					topology_info_output->nodes[i].sdma_engine;
+		}
+
+	}
+
+	/* Invoke xgmi ta again to get the link information */
+	if (psp_xgmi_peer_link_info_supported(psp)) {
+		struct ta_xgmi_cmd_get_peer_link_info *link_info_output;
+		struct ta_xgmi_cmd_get_extend_peer_link_info *link_extend_info_output;
+		bool requires_reflection =
+			(psp->xgmi_context.supports_extended_data &&
+			 get_extended_data) ||
+			amdgpu_ip_version(psp->adev, MP0_HWIP, 0) ==
+				IP_VERSION(13, 0, 6) ||
+			amdgpu_ip_version(psp->adev, MP0_HWIP, 0) ==
+				IP_VERSION(13, 0, 14) ||
+			amdgpu_sriov_vf(psp->adev);
+		bool ta_port_num_support = psp->xgmi_context.xgmi_ta_caps & EXTEND_PEER_LINK_INFO_CMD_FLAG ||
+			amdgpu_sriov_xgmi_ta_ext_peer_link_en(psp->adev);
+
+		/* popluate the shared output buffer rather than the cmd input buffer
+		 * with node_ids as the input for GET_PEER_LINKS command execution.
+		 * This is required for GET_PEER_LINKS per xgmi ta implementation.
+		 * The same requirement for GET_EXTEND_PEER_LINKS command.
+		 */
+		if (ta_port_num_support) {
+			link_extend_info_output = &xgmi_cmd->xgmi_out_message.get_extend_link_info;
+
+			for (i = 0; i < topology->num_nodes; i++)
+				link_extend_info_output->nodes[i].node_id = topology->nodes[i].node_id;
+
+			link_extend_info_output->num_nodes = topology->num_nodes;
+			xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_EXTEND_PEER_LINKS;
+		} else {
+			link_info_output = &xgmi_cmd->xgmi_out_message.get_link_info;
+
+			for (i = 0; i < topology->num_nodes; i++)
+				link_info_output->nodes[i].node_id = topology->nodes[i].node_id;
+
+			link_info_output->num_nodes = topology->num_nodes;
+			xgmi_cmd->cmd_id = TA_COMMAND_XGMI__GET_PEER_LINKS;
+		}
+
+		ret = psp_xgmi_invoke(psp, xgmi_cmd->cmd_id);
+		if (ret)
+			return ret;
+
+		for (i = 0; i < topology->num_nodes; i++) {
+			uint8_t node_num_links = ta_port_num_support ?
+				link_extend_info_output->nodes[i].num_links : link_info_output->nodes[i].num_links;
+			/* accumulate num_links on extended data */
+			if (get_extended_data) {
+				topology->nodes[i].num_links = topology->nodes[i].num_links + node_num_links;
+			} else {
+				topology->nodes[i].num_links = (requires_reflection && topology->nodes[i].num_links) ?
+								topology->nodes[i].num_links : node_num_links;
+			}
+			/* popluate the connected port num info if supported and available */
+			if (ta_port_num_support && topology->nodes[i].num_links) {
+				memcpy(topology->nodes[i].port_num, link_extend_info_output->nodes[i].port_num,
+				       sizeof(struct xgmi_connected_port_num) * TA_XGMI__MAX_PORT_NUM);
+			}
+
+			/* reflect the topology information for bi-directionality */
+			if (requires_reflection && topology->nodes[i].num_hops)
+				psp_xgmi_reflect_topology_info(psp, topology->nodes[i]);
+		}
+	}
+
+	return 0;
+}
+
+int psp_xgmi_set_topology_info(struct psp_context *psp,
+			       int number_devices,
+			       struct psp_xgmi_topology_info *topology)
+{
+	struct ta_xgmi_shared_memory *xgmi_cmd;
+	struct ta_xgmi_cmd_get_topology_info_input *topology_info_input;
+	int i;
+
+	if (!topology || topology->num_nodes > TA_XGMI__MAX_CONNECTED_NODES)
+		return -EINVAL;
+
+	xgmi_cmd = (struct ta_xgmi_shared_memory *)psp->xgmi_context.context.mem_context.shared_buf;
+	memset(xgmi_cmd, 0, sizeof(struct ta_xgmi_shared_memory));
+
+	topology_info_input = &xgmi_cmd->xgmi_in_message.get_topology_info;
+	xgmi_cmd->cmd_id = TA_COMMAND_XGMI__SET_TOPOLOGY_INFO;
+	topology_info_input->num_nodes = number_devices;
+
+	for (i = 0; i < topology_info_input->num_nodes; i++) {
+		topology_info_input->nodes[i].node_id = topology->nodes[i].node_id;
+		topology_info_input->nodes[i].num_hops = topology->nodes[i].num_hops;
+		topology_info_input->nodes[i].is_sharing_enabled = 1;
+		topology_info_input->nodes[i].sdma_engine = topology->nodes[i].sdma_engine;
+	}
+
+	/* Invoke xgmi ta to set topology information */
+	return psp_xgmi_invoke(psp, TA_COMMAND_XGMI__SET_TOPOLOGY_INFO);
+}
+
+// ras begin
+static void psp_ras_ta_check_status(struct psp_context *psp)
+{
+	struct ta_ras_shared_memory *ras_cmd =
+		(struct ta_ras_shared_memory *)psp->ras_context.context.mem_context.shared_buf;
+
+	switch (ras_cmd->ras_status) {
+	case TA_RAS_STATUS__ERROR_UNSUPPORTED_IP:
+		dev_warn(psp->adev->dev,
+			 "RAS WARNING: cmd failed due to unsupported ip\n");
+		break;
+	case TA_RAS_STATUS__ERROR_UNSUPPORTED_ERROR_INJ:
+		dev_warn(psp->adev->dev,
+			 "RAS WARNING: cmd failed due to unsupported error injection\n");
+		break;
+	case TA_RAS_STATUS__SUCCESS:
+		break;
+	case TA_RAS_STATUS__TEE_ERROR_ACCESS_DENIED:
+		if (ras_cmd->cmd_id == TA_RAS_COMMAND__TRIGGER_ERROR)
+			dev_warn(psp->adev->dev,
+				 "RAS WARNING: Inject error to critical region is not allowed\n");
+		break;
+	default:
+		dev_warn(psp->adev->dev,
+			 "RAS WARNING: ras status = 0x%X\n", ras_cmd->ras_status);
+		break;
+	}
+}
+
+static int psp_ras_send_cmd(struct psp_context *psp,
+		enum ras_command cmd_id, void *in, void *out)
+{
+	struct ta_ras_shared_memory *ras_cmd;
+	uint32_t cmd = cmd_id;
+	int ret = 0;
+
+	if (!in)
+		return -EINVAL;
+
+	mutex_lock(&psp->ras_context.mutex);
+	ras_cmd = (struct ta_ras_shared_memory *)psp->ras_context.context.mem_context.shared_buf;
+	memset(ras_cmd, 0, sizeof(struct ta_ras_shared_memory));
+
+	switch (cmd) {
+	case TA_RAS_COMMAND__ENABLE_FEATURES:
+	case TA_RAS_COMMAND__DISABLE_FEATURES:
+		memcpy(&ras_cmd->ras_in_message,
+			in, sizeof(ras_cmd->ras_in_message));
+		break;
+	case TA_RAS_COMMAND__TRIGGER_ERROR:
+		memcpy(&ras_cmd->ras_in_message.trigger_error,
+			in, sizeof(ras_cmd->ras_in_message.trigger_error));
+		break;
+	case TA_RAS_COMMAND__QUERY_ADDRESS:
+		memcpy(&ras_cmd->ras_in_message.address,
+			in, sizeof(ras_cmd->ras_in_message.address));
+		break;
+	default:
+		dev_err(psp->adev->dev, "Invalid ras cmd id: %u\n", cmd);
+		ret = -EINVAL;
+		goto err_out;
+	}
+
+	ras_cmd->cmd_id = cmd;
+	ret = psp_ras_invoke(psp, ras_cmd->cmd_id);
+
+	switch (cmd) {
+	case TA_RAS_COMMAND__TRIGGER_ERROR:
+		if (!ret && out)
+			memcpy(out, &ras_cmd->ras_status, sizeof(ras_cmd->ras_status));
+		break;
+	case TA_RAS_COMMAND__QUERY_ADDRESS:
+		if (ret || ras_cmd->ras_status || psp->cmd_buf_mem->resp.status)
+			ret = -EINVAL;
+		else if (out)
+			memcpy(out,
+				&ras_cmd->ras_out_message.address,
+				sizeof(ras_cmd->ras_out_message.address));
+		break;
+	default:
+		break;
+	}
+
+err_out:
+	mutex_unlock(&psp->ras_context.mutex);
+
+	return ret;
+}
+
+int psp_ras_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
+{
+	struct ta_ras_shared_memory *ras_cmd;
+	int ret;
+
+	ras_cmd = (struct ta_ras_shared_memory *)psp->ras_context.context.mem_context.shared_buf;
+
+	/*
+	 * TODO: bypass the loading in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	ret = psp_ta_invoke(psp, ta_cmd_id, &psp->ras_context.context);
+
+	if (amdgpu_ras_intr_triggered())
+		return ret;
+
+	if (ras_cmd->if_version > RAS_TA_HOST_IF_VER) {
+		dev_warn(psp->adev->dev, "RAS: Unsupported Interface\n");
+		return -EINVAL;
+	}
+
+	if (!ret) {
+		if (ras_cmd->ras_out_message.flags.err_inject_switch_disable_flag) {
+			dev_warn(psp->adev->dev, "ECC switch disabled\n");
+
+			ras_cmd->ras_status = TA_RAS_STATUS__ERROR_RAS_NOT_AVAILABLE;
+		} else if (ras_cmd->ras_out_message.flags.reg_access_failure_flag)
+			dev_warn(psp->adev->dev,
+				 "RAS internal register access blocked\n");
+
+		psp_ras_ta_check_status(psp);
+	}
+
+	return ret;
+}
+
+int psp_ras_enable_features(struct psp_context *psp,
+		union ta_ras_cmd_input *info, bool enable)
+{
+	enum ras_command cmd_id;
+	int ret;
+
+	if (!psp->ras_context.context.initialized || !info)
+		return -EINVAL;
+
+	cmd_id = enable ?
+		TA_RAS_COMMAND__ENABLE_FEATURES : TA_RAS_COMMAND__DISABLE_FEATURES;
+	ret = psp_ras_send_cmd(psp, cmd_id, info, NULL);
+	if (ret)
+		return -EINVAL;
+
+	return 0;
+}
+
+int psp_ras_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * TODO: bypass the terminate in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->ras_context.context.initialized)
+		return 0;
+
+	ret = psp_ta_unload(psp, &psp->ras_context.context);
+
+	psp->ras_context.context.initialized = false;
+
+	mutex_destroy(&psp->ras_context.mutex);
+
+	return ret;
+}
+
+int psp_ras_initialize(struct psp_context *psp)
+{
+	int ret;
+	uint32_t boot_cfg = 0xFF;
+	struct amdgpu_device *adev = psp->adev;
+	struct ta_ras_shared_memory *ras_cmd;
+
+	/*
+	 * TODO: bypass the initialize in sriov for now
+	 */
+	if (amdgpu_sriov_vf(adev))
+		return 0;
+
+	if (!adev->psp.ras_context.context.bin_desc.size_bytes ||
+	    !adev->psp.ras_context.context.bin_desc.start_addr) {
+		dev_info(adev->dev, "RAS: optional ras ta ucode is not available\n");
+		return 0;
+	}
+
+	if (amdgpu_atomfirmware_dynamic_boot_config_supported(adev)) {
+		/* query GECC enablement status from boot config
+		 * boot_cfg: 1: GECC is enabled or 0: GECC is disabled
+		 */
+		ret = psp_boot_config_get(adev, &boot_cfg);
+		if (ret)
+			dev_warn(adev->dev, "PSP get boot config failed\n");
+
+		if (boot_cfg == 1 && !adev->ras_default_ecc_enabled &&
+		    amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC)) {
+			dev_warn(adev->dev, "GECC is currently enabled, which may affect performance\n");
+			dev_warn(adev->dev,
+				"To disable GECC, please reboot the system and load the amdgpu driver with the parameter amdgpu_ras_enable=0\n");
+		} else {
+			if ((adev->ras_default_ecc_enabled || amdgpu_ras_enable == 1) &&
+				amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC)) {
+				if (boot_cfg == 1) {
+					dev_info(adev->dev, "GECC is enabled\n");
+				} else {
+					/* enable GECC in next boot cycle if it is disabled
+					 * in boot config, or force enable GECC if failed to
+					 * get boot configuration
+					 */
+					ret = psp_boot_config_set(adev, BOOT_CONFIG_GECC);
+					if (ret)
+						dev_warn(adev->dev, "PSP set boot config failed\n");
+					else
+						dev_warn(adev->dev, "GECC will be enabled in next boot cycle\n");
+				}
+			} else {
+				if (!boot_cfg) {
+					if (!adev->ras_default_ecc_enabled &&
+					    amdgpu_ras_enable != 1 &&
+					    amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC))
+						dev_warn(adev->dev, "GECC is disabled, set amdgpu_ras_enable=1 to enable GECC in next boot cycle if needed\n");
+					else
+						dev_info(adev->dev, "GECC is disabled\n");
+				} else {
+					/* disable GECC in next boot cycle if ras is
+					 * disabled by module parameter amdgpu_ras_enable
+					 * and/or amdgpu_ras_mask, or boot_config_get call
+					 * is failed
+					 */
+					ret = psp_boot_config_set(adev, 0);
+					if (ret)
+						dev_warn(adev->dev, "PSP set boot config failed\n");
+					else
+						dev_warn(adev->dev, "GECC will be disabled in next boot cycle if set amdgpu_ras_enable and/or amdgpu_ras_mask to 0x0\n");
+				}
+			}
+		}
+	}
+
+	psp->ras_context.context.mem_context.shared_mem_size = PSP_RAS_SHARED_MEM_SIZE;
+	psp->ras_context.context.ta_load_type = GFX_CMD_ID_LOAD_TA;
+
+	if (!psp->ras_context.context.mem_context.shared_buf) {
+		ret = psp_ta_init_shared_buf(psp, &psp->ras_context.context.mem_context);
+		if (ret)
+			return ret;
+	}
+
+	ras_cmd = (struct ta_ras_shared_memory *)psp->ras_context.context.mem_context.shared_buf;
+	memset(ras_cmd, 0, sizeof(struct ta_ras_shared_memory));
+
+	if (amdgpu_ras_is_poison_mode_supported(adev))
+		ras_cmd->ras_in_message.init_flags.poison_mode_en = 1;
+	if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu)
+		ras_cmd->ras_in_message.init_flags.dgpu_mode = 1;
+	ras_cmd->ras_in_message.init_flags.xcc_mask =
+		adev->gfx.xcc_mask;
+	ras_cmd->ras_in_message.init_flags.channel_dis_num = hweight32(adev->gmc.m_half_use) * 2;
+	if (adev->gmc.gmc_funcs->query_mem_partition_mode)
+		ras_cmd->ras_in_message.init_flags.nps_mode =
+			adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
+	ras_cmd->ras_in_message.init_flags.active_umc_mask = adev->umc.active_mask;
+
+	ret = psp_ta_load(psp, &psp->ras_context.context);
+
+	if (!ret && !ras_cmd->ras_status) {
+		psp->ras_context.context.initialized = true;
+		mutex_init(&psp->ras_context.mutex);
+	} else {
+		if (ras_cmd->ras_status)
+			dev_warn(adev->dev, "RAS Init Status: 0x%X\n", ras_cmd->ras_status);
+
+		/* fail to load RAS TA */
+		psp->ras_context.context.initialized = false;
+	}
+
+	return ret;
+}
+
+int psp_ras_trigger_error(struct psp_context *psp,
+			  struct ta_ras_trigger_error_input *info, uint32_t instance_mask)
+{
+	struct amdgpu_device *adev = psp->adev;
+	int ret;
+	uint32_t dev_mask;
+	uint32_t ras_status = 0;
+
+	if (!psp->ras_context.context.initialized || !info)
+		return -EINVAL;
+
+	switch (info->block_id) {
+	case TA_RAS_BLOCK__GFX:
+		dev_mask = GET_MASK(GC, instance_mask);
+		break;
+	case TA_RAS_BLOCK__SDMA:
+		dev_mask = GET_MASK(SDMA0, instance_mask);
+		break;
+	case TA_RAS_BLOCK__VCN:
+	case TA_RAS_BLOCK__JPEG:
+		dev_mask = GET_MASK(VCN, instance_mask);
+		break;
+	default:
+		dev_mask = instance_mask;
+		break;
+	}
+
+	/* reuse sub_block_index for backward compatibility */
+	dev_mask <<= AMDGPU_RAS_INST_SHIFT;
+	dev_mask &= AMDGPU_RAS_INST_MASK;
+	info->sub_block_index |= dev_mask;
+
+	ret = psp_ras_send_cmd(psp,
+			TA_RAS_COMMAND__TRIGGER_ERROR, info, &ras_status);
 	if (ret)
+		return -EINVAL;
+
+	/* If err_event_athub occurs error inject was successful, however
+	 *  return status from TA is no long reliable
+	 */
+	if (amdgpu_ras_intr_triggered())
+		return 0;
+
+	if (ras_status == TA_RAS_STATUS__TEE_ERROR_ACCESS_DENIED)
+		return -EACCES;
+	else if (ras_status)
+		return -EINVAL;
+
+	return 0;
+}
+
+int psp_ras_query_address(struct psp_context *psp,
+			  struct ta_ras_query_address_input *addr_in,
+			  struct ta_ras_query_address_output *addr_out)
+{
+	int ret;
+
+	if (!psp->ras_context.context.initialized ||
+		!addr_in || !addr_out)
+		return -EINVAL;
+
+	ret = psp_ras_send_cmd(psp,
+			TA_RAS_COMMAND__QUERY_ADDRESS, addr_in, addr_out);
+
+	return ret;
+}
+// ras end
+
+// HDCP start
+static int psp_hdcp_initialize(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * TODO: bypass the initialize in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	/* bypass hdcp initialization if dmu is harvested */
+	if (!amdgpu_device_has_display_hardware(psp->adev))
+		return 0;
+
+	if (!psp->hdcp_context.context.bin_desc.size_bytes ||
+	    !psp->hdcp_context.context.bin_desc.start_addr) {
+		dev_info(psp->adev->dev, "HDCP: optional hdcp ta ucode is not available\n");
+		return 0;
+	}
+
+	psp->hdcp_context.context.mem_context.shared_mem_size = PSP_HDCP_SHARED_MEM_SIZE;
+	psp->hdcp_context.context.ta_load_type = GFX_CMD_ID_LOAD_TA;
+
+	if (!psp->hdcp_context.context.mem_context.shared_buf) {
+		ret = psp_ta_init_shared_buf(psp, &psp->hdcp_context.context.mem_context);
+		if (ret)
+			return ret;
+	}
+
+	ret = psp_ta_load(psp, &psp->hdcp_context.context);
+	if (!ret) {
+		psp->hdcp_context.context.initialized = true;
+		mutex_init(&psp->hdcp_context.mutex);
+	}
+
+	return ret;
+}
+
+int psp_hdcp_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
+{
+	/*
+	 * TODO: bypass the loading in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->hdcp_context.context.initialized)
+		return 0;
+
+	return psp_ta_invoke(psp, ta_cmd_id, &psp->hdcp_context.context);
+}
+
+static int psp_hdcp_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * TODO: bypass the terminate in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->hdcp_context.context.initialized)
+		return 0;
+
+	ret = psp_ta_unload(psp, &psp->hdcp_context.context);
+
+	psp->hdcp_context.context.initialized = false;
+
+	return ret;
+}
+// HDCP end
+
+// DTM start
+static int psp_dtm_initialize(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * TODO: bypass the initialize in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	/* bypass dtm initialization if dmu is harvested */
+	if (!amdgpu_device_has_display_hardware(psp->adev))
+		return 0;
+
+	if (!psp->dtm_context.context.bin_desc.size_bytes ||
+	    !psp->dtm_context.context.bin_desc.start_addr) {
+		dev_info(psp->adev->dev, "DTM: optional dtm ta ucode is not available\n");
+		return 0;
+	}
+
+	psp->dtm_context.context.mem_context.shared_mem_size = PSP_DTM_SHARED_MEM_SIZE;
+	psp->dtm_context.context.ta_load_type = GFX_CMD_ID_LOAD_TA;
+
+	if (!psp->dtm_context.context.mem_context.shared_buf) {
+		ret = psp_ta_init_shared_buf(psp, &psp->dtm_context.context.mem_context);
+		if (ret)
+			return ret;
+	}
+
+	ret = psp_ta_load(psp, &psp->dtm_context.context);
+	if (!ret) {
+		psp->dtm_context.context.initialized = true;
+		mutex_init(&psp->dtm_context.mutex);
+	}
+
+	return ret;
+}
+
+int psp_dtm_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
+{
+	/*
+	 * TODO: bypass the loading in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->dtm_context.context.initialized)
+		return 0;
+
+	return psp_ta_invoke(psp, ta_cmd_id, &psp->dtm_context.context);
+}
+
+static int psp_dtm_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * TODO: bypass the terminate in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->dtm_context.context.initialized)
+		return 0;
+
+	ret = psp_ta_unload(psp, &psp->dtm_context.context);
+
+	psp->dtm_context.context.initialized = false;
+
+	return ret;
+}
+// DTM end
+
+// RAP start
+static int psp_rap_initialize(struct psp_context *psp)
+{
+	int ret;
+	enum ta_rap_status status = TA_RAP_STATUS__SUCCESS;
+
+	/*
+	 * TODO: bypass the initialize in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->rap_context.context.bin_desc.size_bytes ||
+	    !psp->rap_context.context.bin_desc.start_addr) {
+		dev_info(psp->adev->dev, "RAP: optional rap ta ucode is not available\n");
+		return 0;
+	}
+
+	psp->rap_context.context.mem_context.shared_mem_size = PSP_RAP_SHARED_MEM_SIZE;
+	psp->rap_context.context.ta_load_type = GFX_CMD_ID_LOAD_TA;
+
+	if (!psp->rap_context.context.mem_context.shared_buf) {
+		ret = psp_ta_init_shared_buf(psp, &psp->rap_context.context.mem_context);
+		if (ret)
+			return ret;
+	}
+
+	ret = psp_ta_load(psp, &psp->rap_context.context);
+	if (!ret) {
+		psp->rap_context.context.initialized = true;
+		mutex_init(&psp->rap_context.mutex);
+	} else
+		return ret;
+
+	ret = psp_rap_invoke(psp, TA_CMD_RAP__INITIALIZE, &status);
+	if (ret || status != TA_RAP_STATUS__SUCCESS) {
+		psp_rap_terminate(psp);
+		/* free rap shared memory */
+		psp_ta_free_shared_buf(&psp->rap_context.context.mem_context);
+
+		dev_warn(psp->adev->dev, "RAP TA initialize fail (%d) status %d.\n",
+			 ret, status);
+
 		return ret;
+	}
+
+	return 0;
+}
+
+static int psp_rap_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	if (!psp->rap_context.context.initialized)
+		return 0;
+
+	ret = psp_ta_unload(psp, &psp->rap_context.context);
+
+	psp->rap_context.context.initialized = false;
+
+	return ret;
+}
+
+int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id, enum ta_rap_status *status)
+{
+	struct ta_rap_shared_memory *rap_cmd;
+	int ret = 0;
+
+	if (!psp->rap_context.context.initialized)
+		return 0;
+
+	if (ta_cmd_id != TA_CMD_RAP__INITIALIZE &&
+	    ta_cmd_id != TA_CMD_RAP__VALIDATE_L0)
+		return -EINVAL;
+
+	mutex_lock(&psp->rap_context.mutex);
 
-	ret = psp_asd_load(psp);
+	rap_cmd = (struct ta_rap_shared_memory *)
+		  psp->rap_context.context.mem_context.shared_buf;
+	memset(rap_cmd, 0, sizeof(struct ta_rap_shared_memory));
+
+	rap_cmd->cmd_id = ta_cmd_id;
+	rap_cmd->validation_method_id = METHOD_A;
+
+	ret = psp_ta_invoke(psp, rap_cmd->cmd_id, &psp->rap_context.context);
 	if (ret)
+		goto out_unlock;
+
+	if (status)
+		*status = rap_cmd->rap_status;
+
+out_unlock:
+	mutex_unlock(&psp->rap_context.mutex);
+
+	return ret;
+}
+// RAP end
+
+/* securedisplay start */
+static int psp_securedisplay_initialize(struct psp_context *psp)
+{
+	int ret;
+	struct ta_securedisplay_cmd *securedisplay_cmd;
+
+	/*
+	 * TODO: bypass the initialize in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	/* bypass securedisplay initialization if dmu is harvested */
+	if (!amdgpu_device_has_display_hardware(psp->adev))
+		return 0;
+
+	if (!psp->securedisplay_context.context.bin_desc.size_bytes ||
+	    !psp->securedisplay_context.context.bin_desc.start_addr) {
+		dev_info(psp->adev->dev,
+			 "SECUREDISPLAY: optional securedisplay ta ucode is not available\n");
+		return 0;
+	}
+
+	psp->securedisplay_context.context.mem_context.shared_mem_size =
+		PSP_SECUREDISPLAY_SHARED_MEM_SIZE;
+	psp->securedisplay_context.context.ta_load_type = GFX_CMD_ID_LOAD_TA;
+
+	if (!psp->securedisplay_context.context.initialized) {
+		ret = psp_ta_init_shared_buf(psp,
+					     &psp->securedisplay_context.context.mem_context);
+		if (ret)
+			return ret;
+	}
+
+	ret = psp_ta_load(psp, &psp->securedisplay_context.context);
+	if (!ret && !psp->securedisplay_context.context.resp_status) {
+		psp->securedisplay_context.context.initialized = true;
+		mutex_init(&psp->securedisplay_context.mutex);
+	} else {
+		/* don't try again */
+		psp->securedisplay_context.context.bin_desc.size_bytes = 0;
 		return ret;
+	}
+
+	mutex_lock(&psp->securedisplay_context.mutex);
+
+	psp_prep_securedisplay_cmd_buf(psp, &securedisplay_cmd,
+			TA_SECUREDISPLAY_COMMAND__QUERY_TA);
+
+	ret = psp_securedisplay_invoke(psp, TA_SECUREDISPLAY_COMMAND__QUERY_TA);
+
+	mutex_unlock(&psp->securedisplay_context.mutex);
+
+	if (ret) {
+		psp_securedisplay_terminate(psp);
+		/* free securedisplay shared memory */
+		psp_ta_free_shared_buf(&psp->securedisplay_context.context.mem_context);
+		dev_err(psp->adev->dev, "SECUREDISPLAY TA initialize fail.\n");
+		return -EINVAL;
+	}
+
+	if (securedisplay_cmd->status != TA_SECUREDISPLAY_STATUS__SUCCESS) {
+		psp_securedisplay_parse_resp_status(psp, securedisplay_cmd->status);
+		dev_err(psp->adev->dev, "SECUREDISPLAY: query securedisplay TA failed. ret 0x%x\n",
+			securedisplay_cmd->securedisplay_out_message.query_ta.query_cmd_ret);
+		/* don't try again */
+		psp->securedisplay_context.context.bin_desc.size_bytes = 0;
+	}
+
+	return 0;
+}
+
+static int psp_securedisplay_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * TODO:bypass the terminate in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->securedisplay_context.context.initialized)
+		return 0;
+
+	ret = psp_ta_unload(psp, &psp->securedisplay_context.context);
+
+	psp->securedisplay_context.context.initialized = false;
+
+	return ret;
+}
+
+int psp_securedisplay_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
+{
+	int ret;
+
+	if (!psp->securedisplay_context.context.initialized)
+		return -EINVAL;
+
+	if (ta_cmd_id != TA_SECUREDISPLAY_COMMAND__QUERY_TA &&
+	    ta_cmd_id != TA_SECUREDISPLAY_COMMAND__SEND_ROI_CRC &&
+	    ta_cmd_id != TA_SECUREDISPLAY_COMMAND__SEND_ROI_CRC_V2)
+		return -EINVAL;
+
+	ret = psp_ta_invoke(psp, ta_cmd_id, &psp->securedisplay_context.context);
+
+	return ret;
+}
+/* SECUREDISPLAY end */
+
+int amdgpu_psp_wait_for_bootloader(struct amdgpu_device *adev)
+{
+	struct psp_context *psp = &adev->psp;
+	int ret = 0;
+
+	if (!amdgpu_sriov_vf(adev) && psp->funcs && psp->funcs->wait_for_bootloader != NULL)
+		ret = psp->funcs->wait_for_bootloader(psp);
+
+	return ret;
+}
+
+bool amdgpu_psp_get_ras_capability(struct psp_context *psp)
+{
+	if (psp->funcs &&
+	    psp->funcs->get_ras_capability) {
+		return psp->funcs->get_ras_capability(psp);
+	} else {
+		return false;
+	}
+}
+
+bool amdgpu_psp_tos_reload_needed(struct amdgpu_device *adev)
+{
+	struct psp_context *psp = &adev->psp;
+
+	if (amdgpu_sriov_vf(adev) || (adev->flags & AMD_IS_APU))
+		return false;
+
+	if (psp->funcs && psp->funcs->is_reload_needed)
+		return psp->funcs->is_reload_needed(psp);
+
+	return false;
+}
+
+static void psp_update_gpu_addresses(struct amdgpu_device *adev)
+{
+	struct psp_context *psp = &adev->psp;
+
+	if (psp->cmd_buf_bo && psp->cmd_buf_mem) {
+		psp->fw_pri_mc_addr = amdgpu_bo_fb_aper_addr(psp->fw_pri_bo);
+		psp->fence_buf_mc_addr = amdgpu_bo_fb_aper_addr(psp->fence_buf_bo);
+		psp->cmd_buf_mc_addr = amdgpu_bo_fb_aper_addr(psp->cmd_buf_bo);
+	}
+	if (adev->firmware.rbuf && psp->km_ring.ring_mem)
+		psp->km_ring.ring_mem_mc_addr = amdgpu_bo_fb_aper_addr(adev->firmware.rbuf);
+}
+
+static int psp_hw_start(struct psp_context *psp)
+{
+	struct amdgpu_device *adev = psp->adev;
+	int ret;
+
+	if (amdgpu_virt_xgmi_migrate_enabled(adev))
+		psp_update_gpu_addresses(adev);
+
+	if (!amdgpu_sriov_vf(adev)) {
+		if ((is_psp_fw_valid(psp->kdb)) &&
+		    (psp->funcs->bootloader_load_kdb != NULL)) {
+			ret = psp_bootloader_load_kdb(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load kdb failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->spl)) &&
+		    (psp->funcs->bootloader_load_spl != NULL)) {
+			ret = psp_bootloader_load_spl(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load spl failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->sys)) &&
+		    (psp->funcs->bootloader_load_sysdrv != NULL)) {
+			ret = psp_bootloader_load_sysdrv(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load sys drv failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->soc_drv)) &&
+		    (psp->funcs->bootloader_load_soc_drv != NULL)) {
+			ret = psp_bootloader_load_soc_drv(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load soc drv failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->intf_drv)) &&
+		    (psp->funcs->bootloader_load_intf_drv != NULL)) {
+			ret = psp_bootloader_load_intf_drv(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load intf drv failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->dbg_drv)) &&
+		    (psp->funcs->bootloader_load_dbg_drv != NULL)) {
+			ret = psp_bootloader_load_dbg_drv(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load dbg drv failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->ras_drv)) &&
+		    (psp->funcs->bootloader_load_ras_drv != NULL)) {
+			ret = psp_bootloader_load_ras_drv(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load ras_drv failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->ipkeymgr_drv)) &&
+		    (psp->funcs->bootloader_load_ipkeymgr_drv != NULL)) {
+			ret = psp_bootloader_load_ipkeymgr_drv(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load ipkeymgr_drv failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->spdm_drv)) &&
+		    (psp->funcs->bootloader_load_spdm_drv != NULL)) {
+			ret = psp_bootloader_load_spdm_drv(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load spdm_drv failed!\n");
+				return ret;
+			}
+		}
+
+		if ((is_psp_fw_valid(psp->sos)) &&
+		    (psp->funcs->bootloader_load_sos != NULL)) {
+			ret = psp_bootloader_load_sos(psp);
+			if (ret) {
+				dev_err(adev->dev, "PSP load sos failed!\n");
+				return ret;
+			}
+		}
+	}
+
+	ret = psp_ring_create(psp, PSP_RING_TYPE__KM);
+	if (ret) {
+		dev_err(adev->dev, "PSP create ring failed!\n");
+		return ret;
+	}
+
+	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
+		ret = psp_update_fw_reservation(psp);
+		if (ret) {
+			dev_err(adev->dev, "update fw reservation failed!\n");
+			return ret;
+		}
+	}
+
+	if (amdgpu_sriov_vf(adev) && amdgpu_in_reset(adev))
+		goto skip_pin_bo;
+
+	if (!psp->boot_time_tmr || psp->autoload_supported) {
+		ret = psp_tmr_init(psp);
+		if (ret) {
+			dev_err(adev->dev, "PSP tmr init failed!\n");
+			return ret;
+		}
+	}
+
+skip_pin_bo:
+	/*
+	 * For ASICs with DF Cstate management centralized
+	 * to PMFW, TMR setup should be performed after PMFW
+	 * loaded and before other non-psp firmware loaded.
+	 */
+	if (psp->pmfw_centralized_cstate_management) {
+		ret = psp_load_smu_fw(psp);
+		if (ret)
+			return ret;
+	}
+
+	if (!psp->boot_time_tmr || !psp->autoload_supported) {
+		ret = psp_tmr_load(psp);
+		if (ret) {
+			dev_err(adev->dev, "PSP load tmr failed!\n");
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+static int psp_get_fw_type(struct amdgpu_firmware_info *ucode,
+			   enum psp_gfx_fw_type *type)
+{
+	switch (ucode->ucode_id) {
+	case AMDGPU_UCODE_ID_CAP:
+		*type = GFX_FW_TYPE_CAP;
+		break;
+	case AMDGPU_UCODE_ID_SDMA0:
+		*type = GFX_FW_TYPE_SDMA0;
+		break;
+	case AMDGPU_UCODE_ID_SDMA1:
+		*type = GFX_FW_TYPE_SDMA1;
+		break;
+	case AMDGPU_UCODE_ID_SDMA2:
+		*type = GFX_FW_TYPE_SDMA2;
+		break;
+	case AMDGPU_UCODE_ID_SDMA3:
+		*type = GFX_FW_TYPE_SDMA3;
+		break;
+	case AMDGPU_UCODE_ID_SDMA4:
+		*type = GFX_FW_TYPE_SDMA4;
+		break;
+	case AMDGPU_UCODE_ID_SDMA5:
+		*type = GFX_FW_TYPE_SDMA5;
+		break;
+	case AMDGPU_UCODE_ID_SDMA6:
+		*type = GFX_FW_TYPE_SDMA6;
+		break;
+	case AMDGPU_UCODE_ID_SDMA7:
+		*type = GFX_FW_TYPE_SDMA7;
+		break;
+	case AMDGPU_UCODE_ID_CP_MES:
+		*type = GFX_FW_TYPE_CP_MES;
+		break;
+	case AMDGPU_UCODE_ID_CP_MES_DATA:
+		*type = GFX_FW_TYPE_MES_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_MES1:
+		*type = GFX_FW_TYPE_CP_MES_KIQ;
+		break;
+	case AMDGPU_UCODE_ID_CP_MES1_DATA:
+		*type = GFX_FW_TYPE_MES_KIQ_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_CE:
+		*type = GFX_FW_TYPE_CP_CE;
+		break;
+	case AMDGPU_UCODE_ID_CP_PFP:
+		*type = GFX_FW_TYPE_CP_PFP;
+		break;
+	case AMDGPU_UCODE_ID_CP_ME:
+		*type = GFX_FW_TYPE_CP_ME;
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC1:
+		*type = GFX_FW_TYPE_CP_MEC;
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC1_JT:
+		*type = GFX_FW_TYPE_CP_MEC_ME1;
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC2:
+		*type = GFX_FW_TYPE_CP_MEC;
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC2_JT:
+		*type = GFX_FW_TYPE_CP_MEC_ME2;
+		break;
+	case AMDGPU_UCODE_ID_RLC_P:
+		*type = GFX_FW_TYPE_RLC_P;
+		break;
+	case AMDGPU_UCODE_ID_RLC_V:
+		*type = GFX_FW_TYPE_RLC_V;
+		break;
+	case AMDGPU_UCODE_ID_RLC_G:
+		*type = GFX_FW_TYPE_RLC_G;
+		break;
+	case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL:
+		*type = GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_CNTL;
+		break;
+	case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM:
+		*type = GFX_FW_TYPE_RLC_RESTORE_LIST_GPM_MEM;
+		break;
+	case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM:
+		*type = GFX_FW_TYPE_RLC_RESTORE_LIST_SRM_MEM;
+		break;
+	case AMDGPU_UCODE_ID_RLC_IRAM:
+		*type = GFX_FW_TYPE_RLC_IRAM;
+		break;
+	case AMDGPU_UCODE_ID_RLC_DRAM:
+		*type = GFX_FW_TYPE_RLC_DRAM_BOOT;
+		break;
+	case AMDGPU_UCODE_ID_GLOBAL_TAP_DELAYS:
+		*type = GFX_FW_TYPE_GLOBAL_TAP_DELAYS;
+		break;
+	case AMDGPU_UCODE_ID_SE0_TAP_DELAYS:
+		*type = GFX_FW_TYPE_SE0_TAP_DELAYS;
+		break;
+	case AMDGPU_UCODE_ID_SE1_TAP_DELAYS:
+		*type = GFX_FW_TYPE_SE1_TAP_DELAYS;
+		break;
+	case AMDGPU_UCODE_ID_SE2_TAP_DELAYS:
+		*type = GFX_FW_TYPE_SE2_TAP_DELAYS;
+		break;
+	case AMDGPU_UCODE_ID_SE3_TAP_DELAYS:
+		*type = GFX_FW_TYPE_SE3_TAP_DELAYS;
+		break;
+	case AMDGPU_UCODE_ID_SMC:
+		*type = GFX_FW_TYPE_SMU;
+		break;
+	case AMDGPU_UCODE_ID_PPTABLE:
+		*type = GFX_FW_TYPE_PPTABLE;
+		break;
+	case AMDGPU_UCODE_ID_UVD:
+		*type = GFX_FW_TYPE_UVD;
+		break;
+	case AMDGPU_UCODE_ID_UVD1:
+		*type = GFX_FW_TYPE_UVD1;
+		break;
+	case AMDGPU_UCODE_ID_VCE:
+		*type = GFX_FW_TYPE_VCE;
+		break;
+	case AMDGPU_UCODE_ID_VCN:
+		*type = GFX_FW_TYPE_VCN;
+		break;
+	case AMDGPU_UCODE_ID_VCN1:
+		*type = GFX_FW_TYPE_VCN1;
+		break;
+	case AMDGPU_UCODE_ID_DMCU_ERAM:
+		*type = GFX_FW_TYPE_DMCU_ERAM;
+		break;
+	case AMDGPU_UCODE_ID_DMCU_INTV:
+		*type = GFX_FW_TYPE_DMCU_ISR;
+		break;
+	case AMDGPU_UCODE_ID_VCN0_RAM:
+		*type = GFX_FW_TYPE_VCN0_RAM;
+		break;
+	case AMDGPU_UCODE_ID_VCN1_RAM:
+		*type = GFX_FW_TYPE_VCN1_RAM;
+		break;
+	case AMDGPU_UCODE_ID_DMCUB:
+		*type = GFX_FW_TYPE_DMUB;
+		break;
+	case AMDGPU_UCODE_ID_SDMA_UCODE_TH0:
+	case AMDGPU_UCODE_ID_SDMA_RS64:
+		*type = GFX_FW_TYPE_SDMA_UCODE_TH0;
+		break;
+	case AMDGPU_UCODE_ID_SDMA_UCODE_TH1:
+		*type = GFX_FW_TYPE_SDMA_UCODE_TH1;
+		break;
+	case AMDGPU_UCODE_ID_IMU_I:
+		*type = GFX_FW_TYPE_IMU_I;
+		break;
+	case AMDGPU_UCODE_ID_IMU_D:
+		*type = GFX_FW_TYPE_IMU_D;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_PFP:
+		*type = GFX_FW_TYPE_RS64_PFP;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_ME:
+		*type = GFX_FW_TYPE_RS64_ME;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_MEC:
+		*type = GFX_FW_TYPE_RS64_MEC;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_PFP_P0_STACK:
+		*type = GFX_FW_TYPE_RS64_PFP_P0_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_PFP_P1_STACK:
+		*type = GFX_FW_TYPE_RS64_PFP_P1_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_ME_P0_STACK:
+		*type = GFX_FW_TYPE_RS64_ME_P0_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_ME_P1_STACK:
+		*type = GFX_FW_TYPE_RS64_ME_P1_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P0_STACK:
+		*type = GFX_FW_TYPE_RS64_MEC_P0_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P1_STACK:
+		*type = GFX_FW_TYPE_RS64_MEC_P1_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P2_STACK:
+		*type = GFX_FW_TYPE_RS64_MEC_P2_STACK;
+		break;
+	case AMDGPU_UCODE_ID_CP_RS64_MEC_P3_STACK:
+		*type = GFX_FW_TYPE_RS64_MEC_P3_STACK;
+		break;
+	case AMDGPU_UCODE_ID_VPE_CTX:
+		*type = GFX_FW_TYPE_VPEC_FW1;
+		break;
+	case AMDGPU_UCODE_ID_VPE_CTL:
+		*type = GFX_FW_TYPE_VPEC_FW2;
+		break;
+	case AMDGPU_UCODE_ID_VPE:
+		*type = GFX_FW_TYPE_VPE;
+		break;
+	case AMDGPU_UCODE_ID_UMSCH_MM_UCODE:
+		*type = GFX_FW_TYPE_UMSCH_UCODE;
+		break;
+	case AMDGPU_UCODE_ID_UMSCH_MM_DATA:
+		*type = GFX_FW_TYPE_UMSCH_DATA;
+		break;
+	case AMDGPU_UCODE_ID_UMSCH_MM_CMD_BUFFER:
+		*type = GFX_FW_TYPE_UMSCH_CMD_BUFFER;
+		break;
+	case AMDGPU_UCODE_ID_P2S_TABLE:
+		*type = GFX_FW_TYPE_P2S_TABLE;
+		break;
+	case AMDGPU_UCODE_ID_JPEG_RAM:
+		*type = GFX_FW_TYPE_JPEG_RAM;
+		break;
+	case AMDGPU_UCODE_ID_ISP:
+		*type = GFX_FW_TYPE_ISP;
+		break;
+	case AMDGPU_UCODE_ID_MAXIMUM:
+	default:
+		return -EINVAL;
+	}
 
 	return 0;
 }
 
-static int psp_np_fw_load(struct psp_context *psp)
+static void psp_print_fw_hdr(struct psp_context *psp,
+			     struct amdgpu_firmware_info *ucode)
+{
+	struct amdgpu_device *adev = psp->adev;
+	struct common_firmware_header *hdr;
+
+	switch (ucode->ucode_id) {
+	case AMDGPU_UCODE_ID_SDMA0:
+	case AMDGPU_UCODE_ID_SDMA1:
+	case AMDGPU_UCODE_ID_SDMA2:
+	case AMDGPU_UCODE_ID_SDMA3:
+	case AMDGPU_UCODE_ID_SDMA4:
+	case AMDGPU_UCODE_ID_SDMA5:
+	case AMDGPU_UCODE_ID_SDMA6:
+	case AMDGPU_UCODE_ID_SDMA7:
+		hdr = (struct common_firmware_header *)
+			adev->sdma.instance[ucode->ucode_id - AMDGPU_UCODE_ID_SDMA0].fw->data;
+		amdgpu_ucode_print_sdma_hdr(hdr);
+		break;
+	case AMDGPU_UCODE_ID_CP_CE:
+		hdr = (struct common_firmware_header *)adev->gfx.ce_fw->data;
+		amdgpu_ucode_print_gfx_hdr(hdr);
+		break;
+	case AMDGPU_UCODE_ID_CP_PFP:
+		hdr = (struct common_firmware_header *)adev->gfx.pfp_fw->data;
+		amdgpu_ucode_print_gfx_hdr(hdr);
+		break;
+	case AMDGPU_UCODE_ID_CP_ME:
+		hdr = (struct common_firmware_header *)adev->gfx.me_fw->data;
+		amdgpu_ucode_print_gfx_hdr(hdr);
+		break;
+	case AMDGPU_UCODE_ID_CP_MEC1:
+		hdr = (struct common_firmware_header *)adev->gfx.mec_fw->data;
+		amdgpu_ucode_print_gfx_hdr(hdr);
+		break;
+	case AMDGPU_UCODE_ID_RLC_G:
+		hdr = (struct common_firmware_header *)adev->gfx.rlc_fw->data;
+		amdgpu_ucode_print_rlc_hdr(hdr);
+		break;
+	case AMDGPU_UCODE_ID_SMC:
+		hdr = (struct common_firmware_header *)adev->pm.fw->data;
+		amdgpu_ucode_print_smc_hdr(hdr);
+		break;
+	default:
+		break;
+	}
+}
+
+static int psp_prep_load_ip_fw_cmd_buf(struct psp_context *psp,
+				       struct amdgpu_firmware_info *ucode,
+				       struct psp_gfx_cmd_resp *cmd)
+{
+	int ret;
+	uint64_t fw_mem_mc_addr = ucode->mc_addr;
+
+	cmd->cmd_id = GFX_CMD_ID_LOAD_IP_FW;
+	cmd->cmd.cmd_load_ip_fw.fw_phy_addr_lo = lower_32_bits(fw_mem_mc_addr);
+	cmd->cmd.cmd_load_ip_fw.fw_phy_addr_hi = upper_32_bits(fw_mem_mc_addr);
+	cmd->cmd.cmd_load_ip_fw.fw_size = ucode->ucode_size;
+
+	ret = psp_get_fw_type(ucode, &cmd->cmd.cmd_load_ip_fw.fw_type);
+	if (ret)
+		dev_err(psp->adev->dev, "Unknown firmware type\n");
+
+	return ret;
+}
+
+int psp_execute_ip_fw_load(struct psp_context *psp,
+			   struct amdgpu_firmware_info *ucode)
+{
+	int ret = 0;
+	struct psp_gfx_cmd_resp *cmd = acquire_psp_cmd_buf(psp);
+
+	ret = psp_prep_load_ip_fw_cmd_buf(psp, ucode, cmd);
+	if (!ret) {
+		ret = psp_cmd_submit_buf(psp, ucode, cmd,
+					 psp->fence_buf_mc_addr);
+	}
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+static int psp_load_p2s_table(struct psp_context *psp)
+{
+	int ret;
+	struct amdgpu_device *adev = psp->adev;
+	struct amdgpu_firmware_info *ucode =
+		&adev->firmware.ucode[AMDGPU_UCODE_ID_P2S_TABLE];
+
+	if (adev->in_runpm && ((adev->pm.rpm_mode == AMDGPU_RUNPM_BACO) ||
+				(adev->pm.rpm_mode == AMDGPU_RUNPM_BAMACO)))
+		return 0;
+
+	if (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 6) ||
+	    amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 14)) {
+		uint32_t supp_vers = adev->flags & AMD_IS_APU ? 0x0036013D :
+								0x0036003C;
+		if (psp->sos.fw_version < supp_vers)
+			return 0;
+	}
+
+	if (!ucode->fw || amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	ret = psp_execute_ip_fw_load(psp, ucode);
+
+	return ret;
+}
+
+static int psp_load_smu_fw(struct psp_context *psp)
+{
+	int ret;
+	struct amdgpu_device *adev = psp->adev;
+	struct amdgpu_firmware_info *ucode =
+			&adev->firmware.ucode[AMDGPU_UCODE_ID_SMC];
+	struct amdgpu_ras *ras = psp->ras_context.ras;
+
+	/*
+	 * Skip SMU FW reloading in case of using BACO for runpm only,
+	 * as SMU is always alive.
+	 */
+	if (adev->in_runpm && ((adev->pm.rpm_mode == AMDGPU_RUNPM_BACO) ||
+				(adev->pm.rpm_mode == AMDGPU_RUNPM_BAMACO)))
+		return 0;
+
+	if (!ucode->fw || amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if ((amdgpu_in_reset(adev) && ras && adev->ras_enabled &&
+	     (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(11, 0, 4) ||
+	      amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(11, 0, 2)))) {
+		ret = amdgpu_dpm_set_mp1_state(adev, PP_MP1_STATE_UNLOAD);
+		if (ret)
+			dev_err(adev->dev, "Failed to set MP1 state prepare for reload\n");
+	}
+
+	ret = psp_execute_ip_fw_load(psp, ucode);
+
+	if (ret)
+		dev_err(adev->dev, "PSP load smu failed!\n");
+
+	return ret;
+}
+
+static bool fw_load_skip_check(struct psp_context *psp,
+			       struct amdgpu_firmware_info *ucode)
+{
+	if (!ucode->fw || !ucode->ucode_size)
+		return true;
+
+	if (ucode->ucode_id == AMDGPU_UCODE_ID_P2S_TABLE)
+		return true;
+
+	if (ucode->ucode_id == AMDGPU_UCODE_ID_SMC &&
+	    (psp_smu_reload_quirk(psp) ||
+	     psp->autoload_supported ||
+	     psp->pmfw_centralized_cstate_management))
+		return true;
+
+	if (amdgpu_sriov_vf(psp->adev) &&
+	    amdgpu_virt_fw_load_skip_check(psp->adev, ucode->ucode_id))
+		return true;
+
+	if (psp->autoload_supported &&
+	    (ucode->ucode_id == AMDGPU_UCODE_ID_CP_MEC1_JT ||
+	     ucode->ucode_id == AMDGPU_UCODE_ID_CP_MEC2_JT))
+		/* skip mec JT when autoload is enabled */
+		return true;
+
+	return false;
+}
+
+int psp_load_fw_list(struct psp_context *psp,
+		     struct amdgpu_firmware_info **ucode_list, int ucode_count)
+{
+	int ret = 0, i;
+	struct amdgpu_firmware_info *ucode;
+
+	for (i = 0; i < ucode_count; ++i) {
+		ucode = ucode_list[i];
+		psp_print_fw_hdr(psp, ucode);
+		ret = psp_execute_ip_fw_load(psp, ucode);
+		if (ret)
+			return ret;
+	}
+	return ret;
+}
+
+static int psp_load_non_psp_fw(struct psp_context *psp)
 {
 	int i, ret;
 	struct amdgpu_firmware_info *ucode;
-	struct amdgpu_device* adev = psp->adev;
+	struct amdgpu_device *adev = psp->adev;
+
+	if (psp->autoload_supported &&
+	    !psp->pmfw_centralized_cstate_management) {
+		ret = psp_load_smu_fw(psp);
+		if (ret)
+			return ret;
+	}
+
+	/* Load P2S table first if it's available */
+	psp_load_p2s_table(psp);
 
 	for (i = 0; i < adev->firmware.max_ucodes; i++) {
 		ucode = &adev->firmware.ucode[i];
-		if (!ucode->fw)
-			continue;
 
 		if (ucode->ucode_id == AMDGPU_UCODE_ID_SMC &&
-		    psp_smu_reload_quirk(psp))
+		    !fw_load_skip_check(psp, ucode)) {
+			ret = psp_load_smu_fw(psp);
+			if (ret)
+				return ret;
 			continue;
-		if (amdgpu_sriov_vf(adev) &&
-		   (ucode->ucode_id == AMDGPU_UCODE_ID_SDMA0
-		    || ucode->ucode_id == AMDGPU_UCODE_ID_SDMA1
-		    || ucode->ucode_id == AMDGPU_UCODE_ID_RLC_G))
-			/*skip ucode loading in SRIOV VF */
+		}
+
+		if (fw_load_skip_check(psp, ucode))
 			continue;
 
-		ret = psp_prep_cmd_buf(ucode, psp->cmd);
-		if (ret)
-			return ret;
+		if (psp->autoload_supported &&
+		    (amdgpu_ip_version(adev, MP0_HWIP, 0) ==
+			     IP_VERSION(11, 0, 7) ||
+		     amdgpu_ip_version(adev, MP0_HWIP, 0) ==
+			     IP_VERSION(11, 0, 11) ||
+		     amdgpu_ip_version(adev, MP0_HWIP, 0) ==
+			     IP_VERSION(11, 0, 12)) &&
+		    (ucode->ucode_id == AMDGPU_UCODE_ID_SDMA1 ||
+		     ucode->ucode_id == AMDGPU_UCODE_ID_SDMA2 ||
+		     ucode->ucode_id == AMDGPU_UCODE_ID_SDMA3))
+			/* PSP only receive one SDMA fw for sienna_cichlid,
+			 * as all four sdma fw are same
+			 */
+			continue;
 
-		ret = psp_cmd_submit_buf(psp, ucode, psp->cmd,
-					 psp->fence_buf_mc_addr, i + 3);
+		psp_print_fw_hdr(psp, ucode);
+
+		ret = psp_execute_ip_fw_load(psp, ucode);
 		if (ret)
 			return ret;
 
-#if 0
-		/* check if firmware loaded sucessfully */
-		if (!amdgpu_psp_check_fw_loading_status(adev, i))
-			return -EINVAL;
-#endif
+		/* Start rlc autoload after psp received all the gfx firmware */
+		if (psp->autoload_supported && ucode->ucode_id == (amdgpu_sriov_vf(adev) ?
+		    adev->virt.autoload_ucode_id : AMDGPU_UCODE_ID_RLC_G)) {
+			ret = psp_rlc_autoload_start(psp);
+			if (ret) {
+				dev_err(adev->dev, "Failed to start rlc autoload\n");
+				return ret;
+			}
+		}
 	}
 
 	return 0;
@@ -319,88 +3110,107 @@ static int psp_load_fw(struct amdgpu_device *adev)
 {
 	int ret;
 	struct psp_context *psp = &adev->psp;
-	struct psp_gfx_cmd_resp *cmd;
 
-	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
-	if (!cmd)
-		return -ENOMEM;
+	if (amdgpu_sriov_vf(adev) && amdgpu_in_reset(adev)) {
+		/* should not destroy ring, only stop */
+		psp_ring_stop(psp, PSP_RING_TYPE__KM);
+	} else {
+		memset(psp->fence_buf, 0, PSP_FENCE_BUFFER_SIZE);
 
-	psp->cmd = cmd;
+		ret = psp_ring_init(psp, PSP_RING_TYPE__KM);
+		if (ret) {
+			dev_err(adev->dev, "PSP ring init failed!\n");
+			goto failed;
+		}
+	}
 
-	ret = amdgpu_bo_create_kernel(adev, PSP_1_MEG, PSP_1_MEG,
-				      AMDGPU_GEM_DOMAIN_GTT,
-				      &psp->fw_pri_bo,
-				      &psp->fw_pri_mc_addr,
-				      &psp->fw_pri_buf);
+	ret = psp_hw_start(psp);
 	if (ret)
 		goto failed;
 
-	ret = amdgpu_bo_create_kernel(adev, PSP_FENCE_BUFFER_SIZE, PAGE_SIZE,
-				      AMDGPU_GEM_DOMAIN_VRAM,
-				      &psp->fence_buf_bo,
-				      &psp->fence_buf_mc_addr,
-				      &psp->fence_buf);
+	ret = psp_load_non_psp_fw(psp);
 	if (ret)
-		goto failed_mem1;
+		goto failed1;
 
-	memset(psp->fence_buf, 0, PSP_FENCE_BUFFER_SIZE);
+	ret = psp_asd_initialize(psp);
+	if (ret) {
+		dev_err(adev->dev, "PSP load asd failed!\n");
+		goto failed1;
+	}
 
-	ret = psp_ring_init(psp, PSP_RING_TYPE__KM);
-	if (ret)
-		goto failed_mem1;
+	ret = psp_rl_load(adev);
+	if (ret) {
+		dev_err(adev->dev, "PSP load RL failed!\n");
+		goto failed1;
+	}
 
-	ret = psp_tmr_init(psp);
-	if (ret)
-		goto failed_mem;
+	if (amdgpu_sriov_vf(adev) && amdgpu_in_reset(adev)) {
+		if (adev->gmc.xgmi.num_physical_nodes > 1) {
+			ret = psp_xgmi_initialize(psp, false, true);
+			/* Warning the XGMI seesion initialize failure
+			 * Instead of stop driver initialization
+			 */
+			if (ret)
+				dev_err(psp->adev->dev,
+					"XGMI: Failed to initialize XGMI session\n");
+		}
+	}
 
-	ret = psp_asd_init(psp);
-	if (ret)
-		goto failed_mem;
+	if (psp->ta_fw) {
+		ret = psp_ras_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"RAS: Failed to initialize RAS\n");
 
-	ret = psp_hw_start(psp);
-	if (ret)
-		goto failed_mem;
+		ret = psp_hdcp_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"HDCP: Failed to initialize HDCP\n");
 
-	ret = psp_np_fw_load(psp);
-	if (ret)
-		goto failed_mem;
+		ret = psp_dtm_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"DTM: Failed to initialize DTM\n");
 
-	kfree(cmd);
+		ret = psp_rap_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"RAP: Failed to initialize RAP\n");
+
+		ret = psp_securedisplay_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"SECUREDISPLAY: Failed to initialize SECUREDISPLAY\n");
+	}
 
 	return 0;
 
-failed_mem:
-	amdgpu_bo_free_kernel(&psp->fence_buf_bo,
-			      &psp->fence_buf_mc_addr, &psp->fence_buf);
-failed_mem1:
-	amdgpu_bo_free_kernel(&psp->fw_pri_bo,
-			      &psp->fw_pri_mc_addr, &psp->fw_pri_buf);
+failed1:
+	psp_free_shared_bufs(psp);
 failed:
-	kfree(cmd);
+	/*
+	 * all cleanup jobs (xgmi terminate, ras terminate,
+	 * ring destroy, cmd/fence/fw buffers destory,
+	 * psp->cmd destory) are delayed to psp_hw_fini
+	 */
+	psp_ring_destroy(psp, PSP_RING_TYPE__KM);
 	return ret;
 }
 
-static int psp_hw_init(void *handle)
+static int psp_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	int ret;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-
-	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
-		return 0;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	mutex_lock(&adev->firmware.mutex);
-	/*
-	 * This sequence is just used on hw_init only once, no need on
-	 * resume.
-	 */
+
 	ret = amdgpu_ucode_init_bo(adev);
 	if (ret)
 		goto failed;
 
 	ret = psp_load_fw(adev);
 	if (ret) {
-		DRM_ERROR("PSP firmware loading failed\n");
+		dev_err(adev->dev, "PSP firmware loading failed\n");
 		goto failed;
 	}
 
@@ -413,132 +3223,1311 @@ failed:
 	return -EINVAL;
 }
 
-static int psp_hw_fini(void *handle)
+static int psp_hw_fini(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 	struct psp_context *psp = &adev->psp;
 
-	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
-		return 0;
-
-	amdgpu_ucode_fini_bo(adev);
-
-	psp_ring_destroy(psp, PSP_RING_TYPE__KM);
+	if (psp->ta_fw) {
+		psp_ras_terminate(psp);
+		psp_securedisplay_terminate(psp);
+		psp_rap_terminate(psp);
+		psp_dtm_terminate(psp);
+		psp_hdcp_terminate(psp);
 
-	if (psp->tmr_buf)
-		amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr, &psp->tmr_buf);
+		if (adev->gmc.xgmi.num_physical_nodes > 1)
+			psp_xgmi_terminate(psp);
+	}
 
-	if (psp->fw_pri_buf)
-		amdgpu_bo_free_kernel(&psp->fw_pri_bo,
-				      &psp->fw_pri_mc_addr, &psp->fw_pri_buf);
+	psp_asd_terminate(psp);
+	psp_tmr_terminate(psp);
 
-	if (psp->fence_buf_bo)
-		amdgpu_bo_free_kernel(&psp->fence_buf_bo,
-				      &psp->fence_buf_mc_addr, &psp->fence_buf);
+	psp_ring_destroy(psp, PSP_RING_TYPE__KM);
 
 	return 0;
 }
 
-static int psp_suspend(void *handle)
+static int psp_suspend(struct amdgpu_ip_block *ip_block)
 {
-	return 0;
+	int ret = 0;
+	struct amdgpu_device *adev = ip_block->adev;
+	struct psp_context *psp = &adev->psp;
+
+	if (adev->gmc.xgmi.num_physical_nodes > 1 &&
+	    psp->xgmi_context.context.initialized) {
+		ret = psp_xgmi_terminate(psp);
+		if (ret) {
+			dev_err(adev->dev, "Failed to terminate xgmi ta\n");
+			goto out;
+		}
+	}
+
+	if (psp->ta_fw) {
+		ret = psp_ras_terminate(psp);
+		if (ret) {
+			dev_err(adev->dev, "Failed to terminate ras ta\n");
+			goto out;
+		}
+		ret = psp_hdcp_terminate(psp);
+		if (ret) {
+			dev_err(adev->dev, "Failed to terminate hdcp ta\n");
+			goto out;
+		}
+		ret = psp_dtm_terminate(psp);
+		if (ret) {
+			dev_err(adev->dev, "Failed to terminate dtm ta\n");
+			goto out;
+		}
+		ret = psp_rap_terminate(psp);
+		if (ret) {
+			dev_err(adev->dev, "Failed to terminate rap ta\n");
+			goto out;
+		}
+		ret = psp_securedisplay_terminate(psp);
+		if (ret) {
+			dev_err(adev->dev, "Failed to terminate securedisplay ta\n");
+			goto out;
+		}
+	}
+
+	ret = psp_asd_terminate(psp);
+	if (ret) {
+		dev_err(adev->dev, "Failed to terminate asd\n");
+		goto out;
+	}
+
+	ret = psp_tmr_terminate(psp);
+	if (ret) {
+		dev_err(adev->dev, "Failed to terminate tmr\n");
+		goto out;
+	}
+
+	ret = psp_ring_stop(psp, PSP_RING_TYPE__KM);
+	if (ret)
+		dev_err(adev->dev, "PSP ring stop failed\n");
+
+out:
+	return ret;
 }
 
-static int psp_resume(void *handle)
+static int psp_resume(struct amdgpu_ip_block *ip_block)
 {
 	int ret;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 	struct psp_context *psp = &adev->psp;
 
-	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
-		return 0;
+	dev_info(adev->dev, "PSP is resuming...\n");
 
-	DRM_INFO("PSP is resuming...\n");
+	if (psp->mem_train_ctx.enable_mem_training) {
+		ret = psp_mem_training(psp, PSP_MEM_TRAIN_RESUME);
+		if (ret) {
+			dev_err(adev->dev, "Failed to process memory training!\n");
+			return ret;
+		}
+	}
 
 	mutex_lock(&adev->firmware.mutex);
 
+	ret = amdgpu_ucode_init_bo(adev);
+	if (ret)
+		goto failed;
+
 	ret = psp_hw_start(psp);
 	if (ret)
 		goto failed;
 
-	ret = psp_np_fw_load(psp);
+	ret = psp_load_non_psp_fw(psp);
 	if (ret)
 		goto failed;
 
+	ret = psp_asd_initialize(psp);
+	if (ret) {
+		dev_err(adev->dev, "PSP load asd failed!\n");
+		goto failed;
+	}
+
+	ret = psp_rl_load(adev);
+	if (ret) {
+		dev_err(adev->dev, "PSP load RL failed!\n");
+		goto failed;
+	}
+
+	if (adev->gmc.xgmi.num_physical_nodes > 1) {
+		ret = psp_xgmi_initialize(psp, false, true);
+		/* Warning the XGMI seesion initialize failure
+		 * Instead of stop driver initialization
+		 */
+		if (ret)
+			dev_err(psp->adev->dev,
+				"XGMI: Failed to initialize XGMI session\n");
+	}
+
+	if (psp->ta_fw) {
+		ret = psp_ras_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"RAS: Failed to initialize RAS\n");
+
+		ret = psp_hdcp_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"HDCP: Failed to initialize HDCP\n");
+
+		ret = psp_dtm_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"DTM: Failed to initialize DTM\n");
+
+		ret = psp_rap_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"RAP: Failed to initialize RAP\n");
+
+		ret = psp_securedisplay_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"SECUREDISPLAY: Failed to initialize SECUREDISPLAY\n");
+	}
+
 	mutex_unlock(&adev->firmware.mutex);
 
 	return 0;
 
 failed:
-	DRM_ERROR("PSP resume failed\n");
+	dev_err(adev->dev, "PSP resume failed\n");
 	mutex_unlock(&adev->firmware.mutex);
 	return ret;
 }
 
-static bool psp_check_fw_loading_status(struct amdgpu_device *adev,
-					enum AMDGPU_UCODE_ID ucode_type)
+int psp_gpu_reset(struct amdgpu_device *adev)
 {
-	struct amdgpu_firmware_info *ucode = NULL;
+	int ret;
 
-	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
-		DRM_INFO("firmware is not loaded by PSP\n");
-		return true;
+	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
+		return 0;
+
+	mutex_lock(&adev->psp.mutex);
+	ret = psp_mode1_reset(&adev->psp);
+	mutex_unlock(&adev->psp.mutex);
+
+	return ret;
+}
+
+int psp_rlc_autoload_start(struct psp_context *psp)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id = GFX_CMD_ID_AUTOLOAD_RLC;
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+				 psp->fence_buf_mc_addr);
+
+	release_psp_cmd_buf(psp);
+
+	return ret;
+}
+
+int psp_ring_cmd_submit(struct psp_context *psp,
+			uint64_t cmd_buf_mc_addr,
+			uint64_t fence_mc_addr,
+			int index)
+{
+	unsigned int psp_write_ptr_reg = 0;
+	struct psp_gfx_rb_frame *write_frame;
+	struct psp_ring *ring = &psp->km_ring;
+	struct psp_gfx_rb_frame *ring_buffer_start = ring->ring_mem;
+	struct psp_gfx_rb_frame *ring_buffer_end = ring_buffer_start +
+		ring->ring_size / sizeof(struct psp_gfx_rb_frame) - 1;
+	struct amdgpu_device *adev = psp->adev;
+	uint32_t ring_size_dw = ring->ring_size / 4;
+	uint32_t rb_frame_size_dw = sizeof(struct psp_gfx_rb_frame) / 4;
+
+	/* KM (GPCOM) prepare write pointer */
+	psp_write_ptr_reg = psp_ring_get_wptr(psp);
+
+	/* Update KM RB frame pointer to new frame */
+	/* write_frame ptr increments by size of rb_frame in bytes */
+	/* psp_write_ptr_reg increments by size of rb_frame in DWORDs */
+	if ((psp_write_ptr_reg % ring_size_dw) == 0)
+		write_frame = ring_buffer_start;
+	else
+		write_frame = ring_buffer_start + (psp_write_ptr_reg / rb_frame_size_dw);
+	/* Check invalid write_frame ptr address */
+	if ((write_frame < ring_buffer_start) || (ring_buffer_end < write_frame)) {
+		dev_err(adev->dev,
+			"ring_buffer_start = %p; ring_buffer_end = %p; write_frame = %p\n",
+			ring_buffer_start, ring_buffer_end, write_frame);
+		dev_err(adev->dev,
+			"write_frame is pointing to address out of bounds\n");
+		return -EINVAL;
 	}
 
-	if (!adev->firmware.fw_size)
-		return false;
+	/* Initialize KM RB frame */
+	memset(write_frame, 0, sizeof(struct psp_gfx_rb_frame));
 
-	ucode = &adev->firmware.ucode[ucode_type];
-	if (!ucode->fw || !ucode->ucode_size)
-		return false;
+	/* Update KM RB frame */
+	write_frame->cmd_buf_addr_hi = upper_32_bits(cmd_buf_mc_addr);
+	write_frame->cmd_buf_addr_lo = lower_32_bits(cmd_buf_mc_addr);
+	write_frame->fence_addr_hi = upper_32_bits(fence_mc_addr);
+	write_frame->fence_addr_lo = lower_32_bits(fence_mc_addr);
+	write_frame->fence_value = index;
+	amdgpu_device_flush_hdp(adev, NULL);
 
-	return psp_compare_sram_data(&adev->psp, ucode, ucode_type);
+	/* Update the write Pointer in DWORDs */
+	psp_write_ptr_reg = (psp_write_ptr_reg + rb_frame_size_dw) % ring_size_dw;
+	psp_ring_set_wptr(psp, psp_write_ptr_reg);
+	return 0;
 }
 
-static int psp_set_clockgating_state(void *handle,
-				     enum amd_clockgating_state state)
+int psp_init_asd_microcode(struct psp_context *psp, const char *chip_name)
 {
+	struct amdgpu_device *adev = psp->adev;
+	const struct psp_firmware_header_v1_0 *asd_hdr;
+	int err = 0;
+
+	err = amdgpu_ucode_request(adev, &adev->psp.asd_fw, AMDGPU_UCODE_REQUIRED,
+				   "amdgpu/%s_asd.bin", chip_name);
+	if (err)
+		goto out;
+
+	asd_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.asd_fw->data;
+	adev->psp.asd_context.bin_desc.fw_version = le32_to_cpu(asd_hdr->header.ucode_version);
+	adev->psp.asd_context.bin_desc.feature_version = le32_to_cpu(asd_hdr->sos.fw_version);
+	adev->psp.asd_context.bin_desc.size_bytes = le32_to_cpu(asd_hdr->header.ucode_size_bytes);
+	adev->psp.asd_context.bin_desc.start_addr = (uint8_t *)asd_hdr +
+				le32_to_cpu(asd_hdr->header.ucode_array_offset_bytes);
 	return 0;
+out:
+	amdgpu_ucode_release(&adev->psp.asd_fw);
+	return err;
 }
 
-static int psp_set_powergating_state(void *handle,
+int psp_init_toc_microcode(struct psp_context *psp, const char *chip_name)
+{
+	struct amdgpu_device *adev = psp->adev;
+	const struct psp_firmware_header_v1_0 *toc_hdr;
+	int err = 0;
+
+	err = amdgpu_ucode_request(adev, &adev->psp.toc_fw, AMDGPU_UCODE_REQUIRED,
+				   "amdgpu/%s_toc.bin", chip_name);
+	if (err)
+		goto out;
+
+	toc_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.toc_fw->data;
+	adev->psp.toc.fw_version = le32_to_cpu(toc_hdr->header.ucode_version);
+	adev->psp.toc.feature_version = le32_to_cpu(toc_hdr->sos.fw_version);
+	adev->psp.toc.size_bytes = le32_to_cpu(toc_hdr->header.ucode_size_bytes);
+	adev->psp.toc.start_addr = (uint8_t *)toc_hdr +
+				le32_to_cpu(toc_hdr->header.ucode_array_offset_bytes);
+	return 0;
+out:
+	amdgpu_ucode_release(&adev->psp.toc_fw);
+	return err;
+}
+
+static int parse_sos_bin_descriptor(struct psp_context *psp,
+				   const struct psp_fw_bin_desc *desc,
+				   const struct psp_firmware_header_v2_0 *sos_hdr)
+{
+	uint8_t *ucode_start_addr  = NULL;
+
+	if (!psp || !desc || !sos_hdr)
+		return -EINVAL;
+
+	ucode_start_addr  = (uint8_t *)sos_hdr +
+			    le32_to_cpu(desc->offset_bytes) +
+			    le32_to_cpu(sos_hdr->header.ucode_array_offset_bytes);
+
+	switch (desc->fw_type) {
+	case PSP_FW_TYPE_PSP_SOS:
+		psp->sos.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->sos.feature_version   = le32_to_cpu(desc->fw_version);
+		psp->sos.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->sos.start_addr	   = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_SYS_DRV:
+		psp->sys.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->sys.feature_version   = le32_to_cpu(desc->fw_version);
+		psp->sys.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->sys.start_addr        = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_KDB:
+		psp->kdb.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->kdb.feature_version   = le32_to_cpu(desc->fw_version);
+		psp->kdb.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->kdb.start_addr        = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_TOC:
+		psp->toc.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->toc.feature_version   = le32_to_cpu(desc->fw_version);
+		psp->toc.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->toc.start_addr        = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_SPL:
+		psp->spl.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->spl.feature_version   = le32_to_cpu(desc->fw_version);
+		psp->spl.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->spl.start_addr        = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_RL:
+		psp->rl.fw_version         = le32_to_cpu(desc->fw_version);
+		psp->rl.feature_version    = le32_to_cpu(desc->fw_version);
+		psp->rl.size_bytes         = le32_to_cpu(desc->size_bytes);
+		psp->rl.start_addr         = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_SOC_DRV:
+		psp->soc_drv.fw_version         = le32_to_cpu(desc->fw_version);
+		psp->soc_drv.feature_version    = le32_to_cpu(desc->fw_version);
+		psp->soc_drv.size_bytes         = le32_to_cpu(desc->size_bytes);
+		psp->soc_drv.start_addr         = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_INTF_DRV:
+		psp->intf_drv.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->intf_drv.feature_version   = le32_to_cpu(desc->fw_version);
+		psp->intf_drv.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->intf_drv.start_addr        = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_DBG_DRV:
+		psp->dbg_drv.fw_version         = le32_to_cpu(desc->fw_version);
+		psp->dbg_drv.feature_version    = le32_to_cpu(desc->fw_version);
+		psp->dbg_drv.size_bytes         = le32_to_cpu(desc->size_bytes);
+		psp->dbg_drv.start_addr         = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_RAS_DRV:
+		psp->ras_drv.fw_version         = le32_to_cpu(desc->fw_version);
+		psp->ras_drv.feature_version    = le32_to_cpu(desc->fw_version);
+		psp->ras_drv.size_bytes         = le32_to_cpu(desc->size_bytes);
+		psp->ras_drv.start_addr         = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_IPKEYMGR_DRV:
+		psp->ipkeymgr_drv.fw_version         = le32_to_cpu(desc->fw_version);
+		psp->ipkeymgr_drv.feature_version    = le32_to_cpu(desc->fw_version);
+		psp->ipkeymgr_drv.size_bytes         = le32_to_cpu(desc->size_bytes);
+		psp->ipkeymgr_drv.start_addr         = ucode_start_addr;
+		break;
+	case PSP_FW_TYPE_PSP_SPDM_DRV:
+		psp->spdm_drv.fw_version	= le32_to_cpu(desc->fw_version);
+		psp->spdm_drv.feature_version	= le32_to_cpu(desc->fw_version);
+		psp->spdm_drv.size_bytes	= le32_to_cpu(desc->size_bytes);
+		psp->spdm_drv.start_addr	= ucode_start_addr;
+		break;
+	default:
+		dev_warn(psp->adev->dev, "Unsupported PSP FW type: %d\n", desc->fw_type);
+		break;
+	}
+
+	return 0;
+}
+
+static int psp_init_sos_base_fw(struct amdgpu_device *adev)
+{
+	const struct psp_firmware_header_v1_0 *sos_hdr;
+	const struct psp_firmware_header_v1_3 *sos_hdr_v1_3;
+	uint8_t *ucode_array_start_addr;
+
+	sos_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.sos_fw->data;
+	ucode_array_start_addr = (uint8_t *)sos_hdr +
+		le32_to_cpu(sos_hdr->header.ucode_array_offset_bytes);
+
+	if (adev->gmc.xgmi.connected_to_cpu ||
+	    (amdgpu_ip_version(adev, MP0_HWIP, 0) != IP_VERSION(13, 0, 2))) {
+		adev->psp.sos.fw_version = le32_to_cpu(sos_hdr->header.ucode_version);
+		adev->psp.sos.feature_version = le32_to_cpu(sos_hdr->sos.fw_version);
+
+		adev->psp.sys.size_bytes = le32_to_cpu(sos_hdr->sos.offset_bytes);
+		adev->psp.sys.start_addr = ucode_array_start_addr;
+
+		adev->psp.sos.size_bytes = le32_to_cpu(sos_hdr->sos.size_bytes);
+		adev->psp.sos.start_addr = ucode_array_start_addr +
+				le32_to_cpu(sos_hdr->sos.offset_bytes);
+	} else {
+		/* Load alternate PSP SOS FW */
+		sos_hdr_v1_3 = (const struct psp_firmware_header_v1_3 *)adev->psp.sos_fw->data;
+
+		adev->psp.sos.fw_version = le32_to_cpu(sos_hdr_v1_3->sos_aux.fw_version);
+		adev->psp.sos.feature_version = le32_to_cpu(sos_hdr_v1_3->sos_aux.fw_version);
+
+		adev->psp.sys.size_bytes = le32_to_cpu(sos_hdr_v1_3->sys_drv_aux.size_bytes);
+		adev->psp.sys.start_addr = ucode_array_start_addr +
+			le32_to_cpu(sos_hdr_v1_3->sys_drv_aux.offset_bytes);
+
+		adev->psp.sos.size_bytes = le32_to_cpu(sos_hdr_v1_3->sos_aux.size_bytes);
+		adev->psp.sos.start_addr = ucode_array_start_addr +
+			le32_to_cpu(sos_hdr_v1_3->sos_aux.offset_bytes);
+	}
+
+	if ((adev->psp.sys.size_bytes == 0) || (adev->psp.sos.size_bytes == 0)) {
+		dev_warn(adev->dev, "PSP SOS FW not available");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+int psp_init_sos_microcode(struct psp_context *psp, const char *chip_name)
+{
+	struct amdgpu_device *adev = psp->adev;
+	const struct psp_firmware_header_v1_0 *sos_hdr;
+	const struct psp_firmware_header_v1_1 *sos_hdr_v1_1;
+	const struct psp_firmware_header_v1_2 *sos_hdr_v1_2;
+	const struct psp_firmware_header_v1_3 *sos_hdr_v1_3;
+	const struct psp_firmware_header_v2_0 *sos_hdr_v2_0;
+	const struct psp_firmware_header_v2_1 *sos_hdr_v2_1;
+	int fw_index, fw_bin_count, start_index = 0;
+	const struct psp_fw_bin_desc *fw_bin;
+	uint8_t *ucode_array_start_addr;
+	int err = 0;
+
+	if (amdgpu_is_kicker_fw(adev))
+		err = amdgpu_ucode_request(adev, &adev->psp.sos_fw, AMDGPU_UCODE_REQUIRED,
+					   "amdgpu/%s_sos_kicker.bin", chip_name);
+	else
+		err = amdgpu_ucode_request(adev, &adev->psp.sos_fw, AMDGPU_UCODE_REQUIRED,
+					   "amdgpu/%s_sos.bin", chip_name);
+	if (err)
+		goto out;
+
+	sos_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.sos_fw->data;
+	ucode_array_start_addr = (uint8_t *)sos_hdr +
+		le32_to_cpu(sos_hdr->header.ucode_array_offset_bytes);
+	amdgpu_ucode_print_psp_hdr(&sos_hdr->header);
+
+	switch (sos_hdr->header.header_version_major) {
+	case 1:
+		err = psp_init_sos_base_fw(adev);
+		if (err)
+			goto out;
+
+		if (sos_hdr->header.header_version_minor == 1) {
+			sos_hdr_v1_1 = (const struct psp_firmware_header_v1_1 *)adev->psp.sos_fw->data;
+			adev->psp.toc.size_bytes = le32_to_cpu(sos_hdr_v1_1->toc.size_bytes);
+			adev->psp.toc.start_addr = (uint8_t *)adev->psp.sys.start_addr +
+					le32_to_cpu(sos_hdr_v1_1->toc.offset_bytes);
+			adev->psp.kdb.size_bytes = le32_to_cpu(sos_hdr_v1_1->kdb.size_bytes);
+			adev->psp.kdb.start_addr = (uint8_t *)adev->psp.sys.start_addr +
+					le32_to_cpu(sos_hdr_v1_1->kdb.offset_bytes);
+		}
+		if (sos_hdr->header.header_version_minor == 2) {
+			sos_hdr_v1_2 = (const struct psp_firmware_header_v1_2 *)adev->psp.sos_fw->data;
+			adev->psp.kdb.size_bytes = le32_to_cpu(sos_hdr_v1_2->kdb.size_bytes);
+			adev->psp.kdb.start_addr = (uint8_t *)adev->psp.sys.start_addr +
+						    le32_to_cpu(sos_hdr_v1_2->kdb.offset_bytes);
+		}
+		if (sos_hdr->header.header_version_minor == 3) {
+			sos_hdr_v1_3 = (const struct psp_firmware_header_v1_3 *)adev->psp.sos_fw->data;
+			adev->psp.toc.size_bytes = le32_to_cpu(sos_hdr_v1_3->v1_1.toc.size_bytes);
+			adev->psp.toc.start_addr = ucode_array_start_addr +
+				le32_to_cpu(sos_hdr_v1_3->v1_1.toc.offset_bytes);
+			adev->psp.kdb.size_bytes = le32_to_cpu(sos_hdr_v1_3->v1_1.kdb.size_bytes);
+			adev->psp.kdb.start_addr = ucode_array_start_addr +
+				le32_to_cpu(sos_hdr_v1_3->v1_1.kdb.offset_bytes);
+			adev->psp.spl.size_bytes = le32_to_cpu(sos_hdr_v1_3->spl.size_bytes);
+			adev->psp.spl.start_addr = ucode_array_start_addr +
+				le32_to_cpu(sos_hdr_v1_3->spl.offset_bytes);
+			adev->psp.rl.size_bytes = le32_to_cpu(sos_hdr_v1_3->rl.size_bytes);
+			adev->psp.rl.start_addr = ucode_array_start_addr +
+				le32_to_cpu(sos_hdr_v1_3->rl.offset_bytes);
+		}
+		break;
+	case 2:
+		sos_hdr_v2_0 = (const struct psp_firmware_header_v2_0 *)adev->psp.sos_fw->data;
+
+		fw_bin_count = le32_to_cpu(sos_hdr_v2_0->psp_fw_bin_count);
+
+		if (fw_bin_count >= UCODE_MAX_PSP_PACKAGING) {
+			dev_err(adev->dev, "packed SOS count exceeds maximum limit\n");
+			err = -EINVAL;
+			goto out;
+		}
+
+		if (sos_hdr_v2_0->header.header_version_minor == 1) {
+			sos_hdr_v2_1 = (const struct psp_firmware_header_v2_1 *)adev->psp.sos_fw->data;
+
+			fw_bin = sos_hdr_v2_1->psp_fw_bin;
+
+			if (psp_is_aux_sos_load_required(psp))
+				start_index = le32_to_cpu(sos_hdr_v2_1->psp_aux_fw_bin_index);
+			else
+				fw_bin_count -= le32_to_cpu(sos_hdr_v2_1->psp_aux_fw_bin_index);
+
+		} else {
+			fw_bin = sos_hdr_v2_0->psp_fw_bin;
+		}
+
+		for (fw_index = start_index; fw_index < fw_bin_count; fw_index++) {
+			err = parse_sos_bin_descriptor(psp, fw_bin + fw_index,
+						       sos_hdr_v2_0);
+			if (err)
+				goto out;
+		}
+		break;
+	default:
+		dev_err(adev->dev,
+			"unsupported psp sos firmware\n");
+		err = -EINVAL;
+		goto out;
+	}
+
+	return 0;
+out:
+	amdgpu_ucode_release(&adev->psp.sos_fw);
+
+	return err;
+}
+
+static bool is_ta_fw_applicable(struct psp_context *psp,
+			     const struct psp_fw_bin_desc *desc)
+{
+	struct amdgpu_device *adev = psp->adev;
+	uint32_t fw_version;
+
+	switch (desc->fw_type) {
+	case TA_FW_TYPE_PSP_XGMI:
+	case TA_FW_TYPE_PSP_XGMI_AUX:
+		/* for now, AUX TA only exists on 13.0.6 ta bin,
+		 * from v20.00.0x.14
+		 */
+		if (amdgpu_ip_version(adev, MP0_HWIP, 0) ==
+		    IP_VERSION(13, 0, 6)) {
+			fw_version = le32_to_cpu(desc->fw_version);
+
+			if (adev->flags & AMD_IS_APU &&
+			    (fw_version & 0xff) >= 0x14)
+				return desc->fw_type == TA_FW_TYPE_PSP_XGMI_AUX;
+			else
+				return desc->fw_type == TA_FW_TYPE_PSP_XGMI;
+		}
+		break;
+	default:
+		break;
+	}
+
+	return true;
+}
+
+static int parse_ta_bin_descriptor(struct psp_context *psp,
+				   const struct psp_fw_bin_desc *desc,
+				   const struct ta_firmware_header_v2_0 *ta_hdr)
+{
+	uint8_t *ucode_start_addr  = NULL;
+
+	if (!psp || !desc || !ta_hdr)
+		return -EINVAL;
+
+	if (!is_ta_fw_applicable(psp, desc))
+		return 0;
+
+	ucode_start_addr  = (uint8_t *)ta_hdr +
+			    le32_to_cpu(desc->offset_bytes) +
+			    le32_to_cpu(ta_hdr->header.ucode_array_offset_bytes);
+
+	switch (desc->fw_type) {
+	case TA_FW_TYPE_PSP_ASD:
+		psp->asd_context.bin_desc.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->asd_context.bin_desc.feature_version   = le32_to_cpu(desc->fw_version);
+		psp->asd_context.bin_desc.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->asd_context.bin_desc.start_addr        = ucode_start_addr;
+		break;
+	case TA_FW_TYPE_PSP_XGMI:
+	case TA_FW_TYPE_PSP_XGMI_AUX:
+		psp->xgmi_context.context.bin_desc.fw_version       = le32_to_cpu(desc->fw_version);
+		psp->xgmi_context.context.bin_desc.size_bytes       = le32_to_cpu(desc->size_bytes);
+		psp->xgmi_context.context.bin_desc.start_addr       = ucode_start_addr;
+		break;
+	case TA_FW_TYPE_PSP_RAS:
+		psp->ras_context.context.bin_desc.fw_version        = le32_to_cpu(desc->fw_version);
+		psp->ras_context.context.bin_desc.size_bytes        = le32_to_cpu(desc->size_bytes);
+		psp->ras_context.context.bin_desc.start_addr        = ucode_start_addr;
+		break;
+	case TA_FW_TYPE_PSP_HDCP:
+		psp->hdcp_context.context.bin_desc.fw_version       = le32_to_cpu(desc->fw_version);
+		psp->hdcp_context.context.bin_desc.size_bytes       = le32_to_cpu(desc->size_bytes);
+		psp->hdcp_context.context.bin_desc.start_addr       = ucode_start_addr;
+		break;
+	case TA_FW_TYPE_PSP_DTM:
+		psp->dtm_context.context.bin_desc.fw_version       = le32_to_cpu(desc->fw_version);
+		psp->dtm_context.context.bin_desc.size_bytes       = le32_to_cpu(desc->size_bytes);
+		psp->dtm_context.context.bin_desc.start_addr       = ucode_start_addr;
+		break;
+	case TA_FW_TYPE_PSP_RAP:
+		psp->rap_context.context.bin_desc.fw_version       = le32_to_cpu(desc->fw_version);
+		psp->rap_context.context.bin_desc.size_bytes       = le32_to_cpu(desc->size_bytes);
+		psp->rap_context.context.bin_desc.start_addr       = ucode_start_addr;
+		break;
+	case TA_FW_TYPE_PSP_SECUREDISPLAY:
+		psp->securedisplay_context.context.bin_desc.fw_version =
+			le32_to_cpu(desc->fw_version);
+		psp->securedisplay_context.context.bin_desc.size_bytes =
+			le32_to_cpu(desc->size_bytes);
+		psp->securedisplay_context.context.bin_desc.start_addr =
+			ucode_start_addr;
+		break;
+	default:
+		dev_warn(psp->adev->dev, "Unsupported TA type: %d\n", desc->fw_type);
+		break;
+	}
+
+	return 0;
+}
+
+static int parse_ta_v1_microcode(struct psp_context *psp)
+{
+	const struct ta_firmware_header_v1_0 *ta_hdr;
+	struct amdgpu_device *adev = psp->adev;
+
+	ta_hdr = (const struct ta_firmware_header_v1_0 *) adev->psp.ta_fw->data;
+
+	if (le16_to_cpu(ta_hdr->header.header_version_major) != 1)
+		return -EINVAL;
+
+	adev->psp.xgmi_context.context.bin_desc.fw_version =
+		le32_to_cpu(ta_hdr->xgmi.fw_version);
+	adev->psp.xgmi_context.context.bin_desc.size_bytes =
+		le32_to_cpu(ta_hdr->xgmi.size_bytes);
+	adev->psp.xgmi_context.context.bin_desc.start_addr =
+		(uint8_t *)ta_hdr +
+		le32_to_cpu(ta_hdr->header.ucode_array_offset_bytes);
+
+	adev->psp.ras_context.context.bin_desc.fw_version =
+		le32_to_cpu(ta_hdr->ras.fw_version);
+	adev->psp.ras_context.context.bin_desc.size_bytes =
+		le32_to_cpu(ta_hdr->ras.size_bytes);
+	adev->psp.ras_context.context.bin_desc.start_addr =
+		(uint8_t *)adev->psp.xgmi_context.context.bin_desc.start_addr +
+		le32_to_cpu(ta_hdr->ras.offset_bytes);
+
+	adev->psp.hdcp_context.context.bin_desc.fw_version =
+		le32_to_cpu(ta_hdr->hdcp.fw_version);
+	adev->psp.hdcp_context.context.bin_desc.size_bytes =
+		le32_to_cpu(ta_hdr->hdcp.size_bytes);
+	adev->psp.hdcp_context.context.bin_desc.start_addr =
+		(uint8_t *)ta_hdr +
+		le32_to_cpu(ta_hdr->header.ucode_array_offset_bytes);
+
+	adev->psp.dtm_context.context.bin_desc.fw_version =
+		le32_to_cpu(ta_hdr->dtm.fw_version);
+	adev->psp.dtm_context.context.bin_desc.size_bytes =
+		le32_to_cpu(ta_hdr->dtm.size_bytes);
+	adev->psp.dtm_context.context.bin_desc.start_addr =
+		(uint8_t *)adev->psp.hdcp_context.context.bin_desc.start_addr +
+		le32_to_cpu(ta_hdr->dtm.offset_bytes);
+
+	adev->psp.securedisplay_context.context.bin_desc.fw_version =
+		le32_to_cpu(ta_hdr->securedisplay.fw_version);
+	adev->psp.securedisplay_context.context.bin_desc.size_bytes =
+		le32_to_cpu(ta_hdr->securedisplay.size_bytes);
+	adev->psp.securedisplay_context.context.bin_desc.start_addr =
+		(uint8_t *)adev->psp.hdcp_context.context.bin_desc.start_addr +
+		le32_to_cpu(ta_hdr->securedisplay.offset_bytes);
+
+	adev->psp.ta_fw_version = le32_to_cpu(ta_hdr->header.ucode_version);
+
+	return 0;
+}
+
+static int parse_ta_v2_microcode(struct psp_context *psp)
+{
+	const struct ta_firmware_header_v2_0 *ta_hdr;
+	struct amdgpu_device *adev = psp->adev;
+	int err = 0;
+	int ta_index = 0;
+
+	ta_hdr = (const struct ta_firmware_header_v2_0 *)adev->psp.ta_fw->data;
+
+	if (le16_to_cpu(ta_hdr->header.header_version_major) != 2)
+		return -EINVAL;
+
+	if (le32_to_cpu(ta_hdr->ta_fw_bin_count) >= UCODE_MAX_PSP_PACKAGING) {
+		dev_err(adev->dev, "packed TA count exceeds maximum limit\n");
+		return -EINVAL;
+	}
+
+	for (ta_index = 0; ta_index < le32_to_cpu(ta_hdr->ta_fw_bin_count); ta_index++) {
+		err = parse_ta_bin_descriptor(psp,
+					      &ta_hdr->ta_fw_bin[ta_index],
+					      ta_hdr);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+int psp_init_ta_microcode(struct psp_context *psp, const char *chip_name)
+{
+	const struct common_firmware_header *hdr;
+	struct amdgpu_device *adev = psp->adev;
+	int err;
+
+	if (amdgpu_is_kicker_fw(adev))
+		err = amdgpu_ucode_request(adev, &adev->psp.ta_fw, AMDGPU_UCODE_REQUIRED,
+					   "amdgpu/%s_ta_kicker.bin", chip_name);
+	else
+		err = amdgpu_ucode_request(adev, &adev->psp.ta_fw, AMDGPU_UCODE_REQUIRED,
+					   "amdgpu/%s_ta.bin", chip_name);
+	if (err)
+		return err;
+
+	hdr = (const struct common_firmware_header *)adev->psp.ta_fw->data;
+	switch (le16_to_cpu(hdr->header_version_major)) {
+	case 1:
+		err = parse_ta_v1_microcode(psp);
+		break;
+	case 2:
+		err = parse_ta_v2_microcode(psp);
+		break;
+	default:
+		dev_err(adev->dev, "unsupported TA header version\n");
+		err = -EINVAL;
+	}
+
+	if (err)
+		amdgpu_ucode_release(&adev->psp.ta_fw);
+
+	return err;
+}
+
+int psp_init_cap_microcode(struct psp_context *psp, const char *chip_name)
+{
+	struct amdgpu_device *adev = psp->adev;
+	const struct psp_firmware_header_v1_0 *cap_hdr_v1_0;
+	struct amdgpu_firmware_info *info = NULL;
+	int err = 0;
+
+	if (!amdgpu_sriov_vf(adev)) {
+		dev_err(adev->dev, "cap microcode should only be loaded under SRIOV\n");
+		return -EINVAL;
+	}
+
+	err = amdgpu_ucode_request(adev, &adev->psp.cap_fw, AMDGPU_UCODE_OPTIONAL,
+				   "amdgpu/%s_cap.bin", chip_name);
+	if (err) {
+		if (err == -ENODEV) {
+			dev_warn(adev->dev, "cap microcode does not exist, skip\n");
+			err = 0;
+		} else {
+			dev_err(adev->dev, "fail to initialize cap microcode\n");
+		}
+		goto out;
+	}
+
+	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CAP];
+	info->ucode_id = AMDGPU_UCODE_ID_CAP;
+	info->fw = adev->psp.cap_fw;
+	cap_hdr_v1_0 = (const struct psp_firmware_header_v1_0 *)
+		adev->psp.cap_fw->data;
+	adev->firmware.fw_size += ALIGN(
+			le32_to_cpu(cap_hdr_v1_0->header.ucode_size_bytes), PAGE_SIZE);
+	adev->psp.cap_fw_version = le32_to_cpu(cap_hdr_v1_0->header.ucode_version);
+	adev->psp.cap_feature_version = le32_to_cpu(cap_hdr_v1_0->sos.fw_version);
+	adev->psp.cap_ucode_size = le32_to_cpu(cap_hdr_v1_0->header.ucode_size_bytes);
+
+	return 0;
+
+out:
+	amdgpu_ucode_release(&adev->psp.cap_fw);
+	return err;
+}
+
+int psp_config_sq_perfmon(struct psp_context *psp,
+		uint32_t xcp_id, bool core_override_enable,
+		bool reg_override_enable, bool perfmon_override_enable)
+{
+	int ret;
+
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (xcp_id > MAX_XCP) {
+		dev_err(psp->adev->dev, "invalid xcp_id %d\n", xcp_id);
+		return -EINVAL;
+	}
+
+	if (amdgpu_ip_version(psp->adev, MP0_HWIP, 0) != IP_VERSION(13, 0, 6)) {
+		dev_err(psp->adev->dev, "Unsupported MP0 version 0x%x for CONFIG_SQ_PERFMON command\n",
+			amdgpu_ip_version(psp->adev, MP0_HWIP, 0));
+		return -EINVAL;
+	}
+	struct psp_gfx_cmd_resp *cmd = acquire_psp_cmd_buf(psp);
+
+	cmd->cmd_id	=	GFX_CMD_ID_CONFIG_SQ_PERFMON;
+	cmd->cmd.config_sq_perfmon.gfx_xcp_mask	=	BIT_MASK(xcp_id);
+	cmd->cmd.config_sq_perfmon.core_override	=	core_override_enable;
+	cmd->cmd.config_sq_perfmon.reg_override	=	reg_override_enable;
+	cmd->cmd.config_sq_perfmon.perfmon_override = perfmon_override_enable;
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+	if (ret)
+		dev_warn(psp->adev->dev, "PSP failed to config sq: xcp%d core%d reg%d perfmon%d\n",
+			xcp_id, core_override_enable, reg_override_enable, perfmon_override_enable);
+
+	release_psp_cmd_buf(psp);
+	return ret;
+}
+
+static int psp_set_clockgating_state(struct amdgpu_ip_block *ip_block,
+					enum amd_clockgating_state state)
+{
+	return 0;
+}
+
+static int psp_set_powergating_state(struct amdgpu_ip_block *ip_block,
 				     enum amd_powergating_state state)
 {
 	return 0;
 }
 
+static ssize_t psp_usbc_pd_fw_sysfs_read(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct amdgpu_ip_block *ip_block;
+	uint32_t fw_ver;
+	int ret;
+
+	ip_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_PSP);
+	if (!ip_block || !ip_block->status.late_initialized) {
+		dev_info(adev->dev, "PSP block is not ready yet\n.");
+		return -EBUSY;
+	}
+
+	mutex_lock(&adev->psp.mutex);
+	ret = psp_read_usbc_pd_fw(&adev->psp, &fw_ver);
+	mutex_unlock(&adev->psp.mutex);
+
+	if (ret) {
+		dev_err(adev->dev, "Failed to read USBC PD FW, err = %d\n", ret);
+		return ret;
+	}
+
+	return sysfs_emit(buf, "%x\n", fw_ver);
+}
+
+static ssize_t psp_usbc_pd_fw_sysfs_write(struct device *dev,
+						       struct device_attribute *attr,
+						       const char *buf,
+						       size_t count)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	int ret, idx;
+	const struct firmware *usbc_pd_fw;
+	struct amdgpu_bo *fw_buf_bo = NULL;
+	uint64_t fw_pri_mc_addr;
+	void *fw_pri_cpu_addr;
+	struct amdgpu_ip_block *ip_block;
+
+	ip_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_PSP);
+	if (!ip_block || !ip_block->status.late_initialized) {
+		dev_err(adev->dev, "PSP block is not ready yet.");
+		return -EBUSY;
+	}
+
+	if (!drm_dev_enter(ddev, &idx))
+		return -ENODEV;
+
+	ret = amdgpu_ucode_request(adev, &usbc_pd_fw, AMDGPU_UCODE_REQUIRED,
+				   "amdgpu/%s", buf);
+	if (ret)
+		goto fail;
+
+	/* LFB address which is aligned to 1MB boundary per PSP request */
+	ret = amdgpu_bo_create_kernel(adev, usbc_pd_fw->size, 0x100000,
+				      AMDGPU_GEM_DOMAIN_VRAM |
+				      AMDGPU_GEM_DOMAIN_GTT,
+				      &fw_buf_bo, &fw_pri_mc_addr,
+				      &fw_pri_cpu_addr);
+	if (ret)
+		goto rel_buf;
+
+	memcpy_toio(fw_pri_cpu_addr, usbc_pd_fw->data, usbc_pd_fw->size);
+
+	mutex_lock(&adev->psp.mutex);
+	ret = psp_load_usbc_pd_fw(&adev->psp, fw_pri_mc_addr);
+	mutex_unlock(&adev->psp.mutex);
+
+	amdgpu_bo_free_kernel(&fw_buf_bo, &fw_pri_mc_addr, &fw_pri_cpu_addr);
+
+rel_buf:
+	amdgpu_ucode_release(&usbc_pd_fw);
+fail:
+	if (ret) {
+		dev_err(adev->dev, "Failed to load USBC PD FW, err = %d", ret);
+		count = ret;
+	}
+
+	drm_dev_exit(idx);
+	return count;
+}
+
+void psp_copy_fw(struct psp_context *psp, uint8_t *start_addr, uint32_t bin_size)
+{
+	int idx;
+
+	if (!drm_dev_enter(adev_to_drm(psp->adev), &idx))
+		return;
+
+	memset(psp->fw_pri_buf, 0, PSP_1_MEG);
+	memcpy(psp->fw_pri_buf, start_addr, bin_size);
+
+	drm_dev_exit(idx);
+}
+
+/**
+ * DOC: usbc_pd_fw
+ * Reading from this file will retrieve the USB-C PD firmware version. Writing to
+ * this file will trigger the update process.
+ */
+static DEVICE_ATTR(usbc_pd_fw, 0644,
+		   psp_usbc_pd_fw_sysfs_read,
+		   psp_usbc_pd_fw_sysfs_write);
+
+int is_psp_fw_valid(struct psp_bin_desc bin)
+{
+	return bin.size_bytes;
+}
+
+static ssize_t amdgpu_psp_vbflash_write(struct file *filp, struct kobject *kobj,
+					const struct bin_attribute *bin_attr,
+					char *buffer, loff_t pos, size_t count)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	adev->psp.vbflash_done = false;
+
+	/* Safeguard against memory drain */
+	if (adev->psp.vbflash_image_size > AMD_VBIOS_FILE_MAX_SIZE_B) {
+		dev_err(adev->dev, "File size cannot exceed %u\n", AMD_VBIOS_FILE_MAX_SIZE_B);
+		kvfree(adev->psp.vbflash_tmp_buf);
+		adev->psp.vbflash_tmp_buf = NULL;
+		adev->psp.vbflash_image_size = 0;
+		return -ENOMEM;
+	}
+
+	/* TODO Just allocate max for now and optimize to realloc later if needed */
+	if (!adev->psp.vbflash_tmp_buf) {
+		adev->psp.vbflash_tmp_buf = kvmalloc(AMD_VBIOS_FILE_MAX_SIZE_B, GFP_KERNEL);
+		if (!adev->psp.vbflash_tmp_buf)
+			return -ENOMEM;
+	}
+
+	mutex_lock(&adev->psp.mutex);
+	memcpy(adev->psp.vbflash_tmp_buf + pos, buffer, count);
+	adev->psp.vbflash_image_size += count;
+	mutex_unlock(&adev->psp.mutex);
+
+	dev_dbg(adev->dev, "IFWI staged for update\n");
+
+	return count;
+}
+
+static ssize_t amdgpu_psp_vbflash_read(struct file *filp, struct kobject *kobj,
+				       const struct bin_attribute *bin_attr, char *buffer,
+				       loff_t pos, size_t count)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	struct amdgpu_bo *fw_buf_bo = NULL;
+	uint64_t fw_pri_mc_addr;
+	void *fw_pri_cpu_addr;
+	int ret;
+
+	if (adev->psp.vbflash_image_size == 0)
+		return -EINVAL;
+
+	dev_dbg(adev->dev, "PSP IFWI flash process initiated\n");
+
+	ret = amdgpu_bo_create_kernel(adev, adev->psp.vbflash_image_size,
+					AMDGPU_GPU_PAGE_SIZE,
+					AMDGPU_GEM_DOMAIN_VRAM,
+					&fw_buf_bo,
+					&fw_pri_mc_addr,
+					&fw_pri_cpu_addr);
+	if (ret)
+		goto rel_buf;
+
+	memcpy_toio(fw_pri_cpu_addr, adev->psp.vbflash_tmp_buf, adev->psp.vbflash_image_size);
+
+	mutex_lock(&adev->psp.mutex);
+	ret = psp_update_spirom(&adev->psp, fw_pri_mc_addr);
+	mutex_unlock(&adev->psp.mutex);
+
+	amdgpu_bo_free_kernel(&fw_buf_bo, &fw_pri_mc_addr, &fw_pri_cpu_addr);
+
+rel_buf:
+	kvfree(adev->psp.vbflash_tmp_buf);
+	adev->psp.vbflash_tmp_buf = NULL;
+	adev->psp.vbflash_image_size = 0;
+
+	if (ret) {
+		dev_err(adev->dev, "Failed to load IFWI, err = %d\n", ret);
+		return ret;
+	}
+
+	dev_dbg(adev->dev, "PSP IFWI flash process done\n");
+	return 0;
+}
+
+/**
+ * DOC: psp_vbflash
+ * Writing to this file will stage an IFWI for update. Reading from this file
+ * will trigger the update process.
+ */
+static const struct bin_attribute psp_vbflash_bin_attr = {
+	.attr = {.name = "psp_vbflash", .mode = 0660},
+	.size = 0,
+	.write = amdgpu_psp_vbflash_write,
+	.read = amdgpu_psp_vbflash_read,
+};
+
+/**
+ * DOC: psp_vbflash_status
+ * The status of the flash process.
+ * 0: IFWI flash not complete.
+ * 1: IFWI flash complete.
+ */
+static ssize_t amdgpu_psp_vbflash_status(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+	uint32_t vbflash_status;
+
+	vbflash_status = psp_vbflash_status(&adev->psp);
+	if (!adev->psp.vbflash_done)
+		vbflash_status = 0;
+	else if (adev->psp.vbflash_done && !(vbflash_status & 0x80000000))
+		vbflash_status = 1;
+
+	return sysfs_emit(buf, "0x%x\n", vbflash_status);
+}
+static DEVICE_ATTR(psp_vbflash_status, 0440, amdgpu_psp_vbflash_status, NULL);
+
+static const struct bin_attribute *const bin_flash_attrs[] = {
+	&psp_vbflash_bin_attr,
+	NULL
+};
+
+static struct attribute *flash_attrs[] = {
+	&dev_attr_psp_vbflash_status.attr,
+	&dev_attr_usbc_pd_fw.attr,
+	NULL
+};
+
+static umode_t amdgpu_flash_attr_is_visible(struct kobject *kobj, struct attribute *attr, int idx)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	if (attr == &dev_attr_usbc_pd_fw.attr)
+		return adev->psp.sup_pd_fw_up ? 0660 : 0;
+
+	return adev->psp.sup_ifwi_up ? 0440 : 0;
+}
+
+static umode_t amdgpu_bin_flash_attr_is_visible(struct kobject *kobj,
+						const struct bin_attribute *attr,
+						int idx)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = drm_to_adev(ddev);
+
+	return adev->psp.sup_ifwi_up ? 0660 : 0;
+}
+
+const struct attribute_group amdgpu_flash_attr_group = {
+	.attrs = flash_attrs,
+	.bin_attrs = bin_flash_attrs,
+	.is_bin_visible = amdgpu_bin_flash_attr_is_visible,
+	.is_visible = amdgpu_flash_attr_is_visible,
+};
+
+#if defined(CONFIG_DEBUG_FS)
+static int psp_read_spirom_debugfs_open(struct inode *inode, struct file *filp)
+{
+	struct amdgpu_device *adev = filp->f_inode->i_private;
+	struct spirom_bo *bo_triplet;
+	int ret;
+
+	/* serialize the open() file calling */
+	if (!mutex_trylock(&adev->psp.mutex))
+		return -EBUSY;
+
+	/*
+	 * make sure only one userpace process is alive for dumping so that
+	 * only one memory buffer of AMD_VBIOS_FILE_MAX_SIZE * 2 is consumed.
+	 * let's say the case where one process try opening the file while
+	 * another one has proceeded to read or release. In this way, eliminate
+	 * the use of mutex for read() or release() callback as well.
+	 */
+	if (adev->psp.spirom_dump_trip) {
+		mutex_unlock(&adev->psp.mutex);
+		return -EBUSY;
+	}
+
+	bo_triplet = kzalloc(sizeof(struct spirom_bo), GFP_KERNEL);
+	if (!bo_triplet) {
+		mutex_unlock(&adev->psp.mutex);
+		return -ENOMEM;
+	}
+
+	ret = amdgpu_bo_create_kernel(adev, AMD_VBIOS_FILE_MAX_SIZE_B * 2,
+				      AMDGPU_GPU_PAGE_SIZE,
+				      AMDGPU_GEM_DOMAIN_GTT,
+				      &bo_triplet->bo,
+				      &bo_triplet->mc_addr,
+				      &bo_triplet->cpu_addr);
+	if (ret)
+		goto rel_trip;
+
+	ret = psp_dump_spirom(&adev->psp, bo_triplet->mc_addr);
+	if (ret)
+		goto rel_bo;
+
+	adev->psp.spirom_dump_trip = bo_triplet;
+	mutex_unlock(&adev->psp.mutex);
+	return 0;
+rel_bo:
+	amdgpu_bo_free_kernel(&bo_triplet->bo, &bo_triplet->mc_addr,
+			      &bo_triplet->cpu_addr);
+rel_trip:
+	kfree(bo_triplet);
+	mutex_unlock(&adev->psp.mutex);
+	dev_err(adev->dev, "Trying IFWI dump fails, err = %d\n", ret);
+	return ret;
+}
+
+static ssize_t psp_read_spirom_debugfs_read(struct file *filp, char __user *buf, size_t size,
+					    loff_t *pos)
+{
+	struct amdgpu_device *adev = filp->f_inode->i_private;
+	struct spirom_bo *bo_triplet = adev->psp.spirom_dump_trip;
+
+	if (!bo_triplet)
+		return -EINVAL;
+
+	return simple_read_from_buffer(buf,
+				       size,
+				       pos, bo_triplet->cpu_addr,
+				       AMD_VBIOS_FILE_MAX_SIZE_B * 2);
+}
+
+static int psp_read_spirom_debugfs_release(struct inode *inode, struct file *filp)
+{
+	struct amdgpu_device *adev = filp->f_inode->i_private;
+	struct spirom_bo *bo_triplet = adev->psp.spirom_dump_trip;
+
+	if (bo_triplet) {
+		amdgpu_bo_free_kernel(&bo_triplet->bo, &bo_triplet->mc_addr,
+				      &bo_triplet->cpu_addr);
+		kfree(bo_triplet);
+	}
+
+	adev->psp.spirom_dump_trip = NULL;
+	return 0;
+}
+
+static const struct file_operations psp_dump_spirom_debugfs_ops = {
+	.owner = THIS_MODULE,
+	.open = psp_read_spirom_debugfs_open,
+	.read = psp_read_spirom_debugfs_read,
+	.release = psp_read_spirom_debugfs_release,
+	.llseek = default_llseek,
+};
+#endif
+
+void amdgpu_psp_debugfs_init(struct amdgpu_device *adev)
+{
+#if defined(CONFIG_DEBUG_FS)
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+
+	debugfs_create_file_size("psp_spirom_dump", 0444, minor->debugfs_root,
+				 adev, &psp_dump_spirom_debugfs_ops, AMD_VBIOS_FILE_MAX_SIZE_B * 2);
+#endif
+}
+
 const struct amd_ip_funcs psp_ip_funcs = {
 	.name = "psp",
 	.early_init = psp_early_init,
-	.late_init = NULL,
 	.sw_init = psp_sw_init,
 	.sw_fini = psp_sw_fini,
 	.hw_init = psp_hw_init,
 	.hw_fini = psp_hw_fini,
 	.suspend = psp_suspend,
 	.resume = psp_resume,
-	.is_idle = NULL,
-	.wait_for_idle = NULL,
-	.soft_reset = NULL,
 	.set_clockgating_state = psp_set_clockgating_state,
 	.set_powergating_state = psp_set_powergating_state,
 };
 
-static const struct amdgpu_psp_funcs psp_funcs = {
-	.check_fw_loading_status = psp_check_fw_loading_status,
+const struct amdgpu_ip_block_version psp_v3_1_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_PSP,
+	.major = 3,
+	.minor = 1,
+	.rev = 0,
+	.funcs = &psp_ip_funcs,
 };
 
-static void psp_set_funcs(struct amdgpu_device *adev)
-{
-	if (NULL == adev->firmware.funcs)
-		adev->firmware.funcs = &psp_funcs;
-}
+const struct amdgpu_ip_block_version psp_v10_0_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_PSP,
+	.major = 10,
+	.minor = 0,
+	.rev = 0,
+	.funcs = &psp_ip_funcs,
+};
 
-const struct amdgpu_ip_block_version psp_v3_1_ip_block =
-{
+const struct amdgpu_ip_block_version psp_v11_0_ip_block = {
 	.type = AMD_IP_BLOCK_TYPE_PSP,
-	.major = 3,
-	.minor = 1,
+	.major = 11,
+	.minor = 0,
+	.rev = 0,
+	.funcs = &psp_ip_funcs,
+};
+
+const struct amdgpu_ip_block_version psp_v11_0_8_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_PSP,
+	.major = 11,
+	.minor = 0,
+	.rev = 8,
+	.funcs = &psp_ip_funcs,
+};
+
+const struct amdgpu_ip_block_version psp_v12_0_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_PSP,
+	.major = 12,
+	.minor = 0,
+	.rev = 0,
+	.funcs = &psp_ip_funcs,
+};
+
+const struct amdgpu_ip_block_version psp_v13_0_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_PSP,
+	.major = 13,
+	.minor = 0,
+	.rev = 0,
+	.funcs = &psp_ip_funcs,
+};
+
+const struct amdgpu_ip_block_version psp_v13_0_4_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_PSP,
+	.major = 13,
+	.minor = 0,
+	.rev = 4,
+	.funcs = &psp_ip_funcs,
+};
+
+const struct amdgpu_ip_block_version psp_v14_0_ip_block = {
+	.type = AMD_IP_BLOCK_TYPE_PSP,
+	.major = 14,
+	.minor = 0,
 	.rev = 0,
 	.funcs = &psp_ip_funcs,
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
index 0301e4e0b297..237b624aa51c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
@@ -27,14 +27,86 @@
 
 #include "amdgpu.h"
 #include "psp_gfx_if.h"
+#include "ta_xgmi_if.h"
+#include "ta_ras_if.h"
+#include "ta_rap_if.h"
+#include "ta_secureDisplay_if.h"
 
 #define PSP_FENCE_BUFFER_SIZE	0x1000
 #define PSP_CMD_BUFFER_SIZE	0x1000
-#define PSP_ASD_SHARED_MEM_SIZE	0x4000
 #define PSP_1_MEG		0x100000
+#define PSP_TMR_SIZE(adev)	((adev)->asic_type == CHIP_ALDEBARAN ? 0x800000 : 0x400000)
+#define PSP_TMR_ALIGNMENT	0x100000
+#define PSP_FW_NAME_LEN		0x24
 
-enum psp_ring_type
-{
+/* VBIOS gfl defines */
+#define MBOX_READY_MASK 0x80000000
+#define MBOX_STATUS_MASK 0x0000FFFF
+#define MBOX_COMMAND_MASK 0x00FF0000
+#define MBOX_READY_FLAG 0x80000000
+#define C2PMSG_CMD_SPI_UPDATE_ROM_IMAGE_ADDR_LO 0x2
+#define C2PMSG_CMD_SPI_UPDATE_ROM_IMAGE_ADDR_HI 0x3
+#define C2PMSG_CMD_SPI_UPDATE_FLASH_IMAGE 0x4
+#define C2PMSG_CMD_SPI_GET_ROM_IMAGE_ADDR_LO 0xf
+#define C2PMSG_CMD_SPI_GET_ROM_IMAGE_ADDR_HI 0x10
+#define C2PMSG_CMD_SPI_GET_FLASH_IMAGE 0x11
+
+/* Command register bit 31 set to indicate readiness */
+#define MBOX_TOS_READY_FLAG (GFX_FLAG_RESPONSE)
+#define MBOX_TOS_READY_MASK (GFX_CMD_RESPONSE_MASK | GFX_CMD_STATUS_MASK)
+
+/* Values to check for a successful GFX_CMD response wait. Check against
+ * both status bits and response state - helps to detect a command failure
+ * or other unexpected cases like a device drop reading all 0xFFs
+ */
+#define MBOX_TOS_RESP_FLAG (GFX_FLAG_RESPONSE)
+#define MBOX_TOS_RESP_MASK (GFX_CMD_RESPONSE_MASK | GFX_CMD_STATUS_MASK)
+
+extern const struct attribute_group amdgpu_flash_attr_group;
+
+enum psp_shared_mem_size {
+	PSP_ASD_SHARED_MEM_SIZE				= 0x0,
+	PSP_XGMI_SHARED_MEM_SIZE			= 0x4000,
+	PSP_RAS_SHARED_MEM_SIZE				= 0x4000,
+	PSP_HDCP_SHARED_MEM_SIZE			= 0x4000,
+	PSP_DTM_SHARED_MEM_SIZE				= 0x4000,
+	PSP_RAP_SHARED_MEM_SIZE				= 0x4000,
+	PSP_SECUREDISPLAY_SHARED_MEM_SIZE	= 0x4000,
+};
+
+enum ta_type_id {
+	TA_TYPE_XGMI = 1,
+	TA_TYPE_RAS,
+	TA_TYPE_HDCP,
+	TA_TYPE_DTM,
+	TA_TYPE_RAP,
+	TA_TYPE_SECUREDISPLAY,
+
+	TA_TYPE_MAX_INDEX,
+};
+
+struct psp_context;
+struct psp_xgmi_node_info;
+struct psp_xgmi_topology_info;
+struct psp_bin_desc;
+
+enum psp_bootloader_cmd {
+	PSP_BL__LOAD_SYSDRV		= 0x10000,
+	PSP_BL__LOAD_SOSDRV		= 0x20000,
+	PSP_BL__LOAD_KEY_DATABASE	= 0x80000,
+	PSP_BL__LOAD_SOCDRV             = 0xB0000,
+	PSP_BL__LOAD_DBGDRV             = 0xC0000,
+	PSP_BL__LOAD_HADDRV		= PSP_BL__LOAD_DBGDRV,
+	PSP_BL__LOAD_INTFDRV		= 0xD0000,
+	PSP_BL__LOAD_RASDRV		= 0xE0000,
+	PSP_BL__LOAD_IPKEYMGRDRV	= 0xF0000,
+	PSP_BL__DRAM_LONG_TRAIN		= 0x100000,
+	PSP_BL__DRAM_SHORT_TRAIN	= 0x200000,
+	PSP_BL__LOAD_TOS_SPL_TABLE	= 0x10000000,
+	PSP_BL__LOAD_SPDMDRV		= 0x20000000,
+};
+
+enum psp_ring_type {
 	PSP_RING_TYPE__INVALID = 0,
 	/*
 	 * These values map to the way the PSP kernel identifies the
@@ -44,70 +116,337 @@ enum psp_ring_type
 	PSP_RING_TYPE__KM = 2  /* Kernel mode ring (formerly called GPCOM) */
 };
 
-struct psp_ring
-{
+struct psp_ring {
 	enum psp_ring_type		ring_type;
 	struct psp_gfx_rb_frame		*ring_mem;
 	uint64_t			ring_mem_mc_addr;
 	void				*ring_mem_handle;
 	uint32_t			ring_size;
+	uint32_t			ring_wptr;
 };
 
-struct psp_context
-{
-	struct amdgpu_device            *adev;
-	struct psp_ring                 km_ring;
-	struct psp_gfx_cmd_resp		*cmd;
+/* More registers may will be supported */
+enum psp_reg_prog_id {
+	PSP_REG_IH_RB_CNTL        = 0,  /* register IH_RB_CNTL */
+	PSP_REG_IH_RB_CNTL_RING1  = 1,  /* register IH_RB_CNTL_RING1 */
+	PSP_REG_IH_RB_CNTL_RING2  = 2,  /* register IH_RB_CNTL_RING2 */
+	PSP_REG_MMHUB_L1_TLB_CNTL = 25,
+	PSP_REG_LAST
+};
+
+#define PSP_WAITREG_CHANGED BIT(0) /* check if the value has changed */
+#define PSP_WAITREG_NOVERBOSE BIT(1) /* No error verbose */
 
+struct psp_funcs {
 	int (*init_microcode)(struct psp_context *psp);
+	int (*wait_for_bootloader)(struct psp_context *psp);
+	int (*bootloader_load_kdb)(struct psp_context *psp);
+	int (*bootloader_load_spl)(struct psp_context *psp);
 	int (*bootloader_load_sysdrv)(struct psp_context *psp);
+	int (*bootloader_load_soc_drv)(struct psp_context *psp);
+	int (*bootloader_load_intf_drv)(struct psp_context *psp);
+	int (*bootloader_load_dbg_drv)(struct psp_context *psp);
+	int (*bootloader_load_ras_drv)(struct psp_context *psp);
+	int (*bootloader_load_ipkeymgr_drv)(struct psp_context *psp);
+	int (*bootloader_load_spdm_drv)(struct psp_context *psp);
 	int (*bootloader_load_sos)(struct psp_context *psp);
-	int (*prep_cmd_buf)(struct amdgpu_firmware_info *ucode,
-			    struct psp_gfx_cmd_resp *cmd);
-	int (*ring_init)(struct psp_context *psp, enum psp_ring_type ring_type);
-	int (*ring_create)(struct psp_context *psp, enum psp_ring_type ring_type);
+	int (*ring_create)(struct psp_context *psp,
+			   enum psp_ring_type ring_type);
+	int (*ring_stop)(struct psp_context *psp,
+			    enum psp_ring_type ring_type);
 	int (*ring_destroy)(struct psp_context *psp,
 			    enum psp_ring_type ring_type);
-	int (*cmd_submit)(struct psp_context *psp, struct amdgpu_firmware_info *ucode,
-			  uint64_t cmd_buf_mc_addr, uint64_t fence_mc_addr, int index);
-	bool (*compare_sram_data)(struct psp_context *psp,
-				  struct amdgpu_firmware_info *ucode,
-				  enum AMDGPU_UCODE_ID ucode_type);
 	bool (*smu_reload_quirk)(struct psp_context *psp);
+	int (*mode1_reset)(struct psp_context *psp);
+	int (*mem_training)(struct psp_context *psp, uint32_t ops);
+	uint32_t (*ring_get_wptr)(struct psp_context *psp);
+	void (*ring_set_wptr)(struct psp_context *psp, uint32_t value);
+	int (*load_usbc_pd_fw)(struct psp_context *psp, uint64_t fw_pri_mc_addr);
+	int (*read_usbc_pd_fw)(struct psp_context *psp, uint32_t *fw_ver);
+	int (*update_spirom)(struct psp_context *psp, uint64_t fw_pri_mc_addr);
+	int (*dump_spirom)(struct psp_context *psp, uint64_t fw_pri_mc_addr);
+	int (*vbflash_stat)(struct psp_context *psp);
+	int (*fatal_error_recovery_quirk)(struct psp_context *psp);
+	bool (*get_ras_capability)(struct psp_context *psp);
+	bool (*is_aux_sos_load_required)(struct psp_context *psp);
+	bool (*is_reload_needed)(struct psp_context *psp);
+	int (*reg_program_no_ring)(struct psp_context *psp, uint32_t val,
+				   enum psp_reg_prog_id id);
+};
 
-	/* fence buffer */
-	struct amdgpu_bo 		*fw_pri_bo;
-	uint64_t 			fw_pri_mc_addr;
+struct ta_funcs {
+	int (*fn_ta_initialize)(struct psp_context *psp);
+	int (*fn_ta_invoke)(struct psp_context *psp, uint32_t ta_cmd_id);
+	int (*fn_ta_terminate)(struct psp_context *psp);
+};
+
+#define AMDGPU_XGMI_MAX_CONNECTED_NODES		64
+struct psp_xgmi_node_info {
+	uint64_t				node_id;
+	uint8_t					num_hops;
+	uint8_t					is_sharing_enabled;
+	enum ta_xgmi_assigned_sdma_engine	sdma_engine;
+	uint8_t					num_links;
+	struct xgmi_connected_port_num		port_num[TA_XGMI__MAX_PORT_NUM];
+};
+
+struct psp_xgmi_topology_info {
+	uint32_t			num_nodes;
+	struct psp_xgmi_node_info	nodes[AMDGPU_XGMI_MAX_CONNECTED_NODES];
+};
+
+struct psp_bin_desc {
+	uint32_t fw_version;
+	uint32_t feature_version;
+	uint32_t size_bytes;
+	uint8_t *start_addr;
+};
+
+struct ta_mem_context {
+	struct amdgpu_bo		*shared_bo;
+	uint64_t		shared_mc_addr;
+	void			*shared_buf;
+	enum psp_shared_mem_size	shared_mem_size;
+};
+
+struct ta_context {
+	bool			initialized;
+	uint32_t		session_id;
+	uint32_t		resp_status;
+	struct ta_mem_context	mem_context;
+	struct psp_bin_desc		bin_desc;
+	enum psp_gfx_cmd_id		ta_load_type;
+	enum ta_type_id		ta_type;
+};
+
+struct ta_cp_context {
+	struct ta_context		context;
+	struct mutex			mutex;
+};
+
+struct psp_xgmi_context {
+	struct ta_context		context;
+	struct psp_xgmi_topology_info	top_info;
+	bool				supports_extended_data;
+	uint8_t				xgmi_ta_caps;
+};
+
+struct psp_ras_context {
+	struct ta_context		context;
+	struct amdgpu_ras		*ras;
+	struct mutex			mutex;
+};
+
+#define MEM_TRAIN_SYSTEM_SIGNATURE		0x54534942
+#define GDDR6_MEM_TRAINING_DATA_SIZE_IN_BYTES	0x1000
+#define GDDR6_MEM_TRAINING_OFFSET		0x8000
+/*Define the VRAM size that will be encroached by BIST training.*/
+#define BIST_MEM_TRAINING_ENCROACHED_SIZE	0x2000000
+
+enum psp_memory_training_init_flag {
+	PSP_MEM_TRAIN_NOT_SUPPORT	= 0x0,
+	PSP_MEM_TRAIN_SUPPORT		= 0x1,
+	PSP_MEM_TRAIN_INIT_FAILED	= 0x2,
+	PSP_MEM_TRAIN_RESERVE_SUCCESS	= 0x4,
+	PSP_MEM_TRAIN_INIT_SUCCESS	= 0x8,
+};
+
+enum psp_memory_training_ops {
+	PSP_MEM_TRAIN_SEND_LONG_MSG	= 0x1,
+	PSP_MEM_TRAIN_SAVE		= 0x2,
+	PSP_MEM_TRAIN_RESTORE		= 0x4,
+	PSP_MEM_TRAIN_SEND_SHORT_MSG	= 0x8,
+	PSP_MEM_TRAIN_COLD_BOOT		= PSP_MEM_TRAIN_SEND_LONG_MSG,
+	PSP_MEM_TRAIN_RESUME		= PSP_MEM_TRAIN_SEND_SHORT_MSG,
+};
+
+struct psp_memory_training_context {
+	/*training data size*/
+	u64 train_data_size;
+	/*
+	 * sys_cache
+	 * cpu virtual address
+	 * system memory buffer that used to store the training data.
+	 */
+	void *sys_cache;
+
+	/*vram offset of the p2c training data*/
+	u64 p2c_train_data_offset;
+
+	/*vram offset of the c2p training data*/
+	u64 c2p_train_data_offset;
+	struct amdgpu_bo *c2p_bo;
+
+	enum psp_memory_training_init_flag init;
+	u32 training_cnt;
+	bool enable_mem_training;
+};
+
+/** PSP runtime DB **/
+#define PSP_RUNTIME_DB_SIZE_IN_BYTES		0x10000
+#define PSP_RUNTIME_DB_OFFSET			0x100000
+#define PSP_RUNTIME_DB_COOKIE_ID		0x0ed5
+#define PSP_RUNTIME_DB_VER_1			0x0100
+#define PSP_RUNTIME_DB_DIAG_ENTRY_MAX_COUNT	0x40
+
+enum psp_runtime_entry_type {
+	PSP_RUNTIME_ENTRY_TYPE_INVALID		= 0x0,
+	PSP_RUNTIME_ENTRY_TYPE_TEST		= 0x1,
+	PSP_RUNTIME_ENTRY_TYPE_MGPU_COMMON	= 0x2,  /* Common mGPU runtime data */
+	PSP_RUNTIME_ENTRY_TYPE_MGPU_WAFL	= 0x3,  /* WAFL runtime data */
+	PSP_RUNTIME_ENTRY_TYPE_MGPU_XGMI	= 0x4,  /* XGMI runtime data */
+	PSP_RUNTIME_ENTRY_TYPE_BOOT_CONFIG	= 0x5,  /* Boot Config runtime data */
+	PSP_RUNTIME_ENTRY_TYPE_PPTABLE_ERR_STATUS = 0x6, /* SCPM validation data */
+};
+
+/* PSP runtime DB header */
+struct psp_runtime_data_header {
+	/* determine the existence of runtime db */
+	uint16_t cookie;
+	/* version of runtime db */
+	uint16_t version;
+};
+
+/* PSP runtime DB entry */
+struct psp_runtime_entry {
+	/* type of runtime db entry */
+	uint32_t entry_type;
+	/* offset of entry in bytes */
+	uint16_t offset;
+	/* size of entry in bytes */
+	uint16_t size;
+};
+
+/* PSP runtime DB directory */
+struct psp_runtime_data_directory {
+	/* number of valid entries */
+	uint16_t			entry_count;
+	/* db entries*/
+	struct psp_runtime_entry	entry_list[PSP_RUNTIME_DB_DIAG_ENTRY_MAX_COUNT];
+};
+
+/* PSP runtime DB boot config feature bitmask */
+enum psp_runtime_boot_cfg_feature {
+	BOOT_CFG_FEATURE_GECC                       = 0x1,
+	BOOT_CFG_FEATURE_TWO_STAGE_DRAM_TRAINING    = 0x2,
+};
+
+/* PSP run time DB SCPM authentication defines */
+enum psp_runtime_scpm_authentication {
+	SCPM_DISABLE                     = 0x0,
+	SCPM_ENABLE                      = 0x1,
+	SCPM_ENABLE_WITH_SCPM_ERR        = 0x2,
+};
+
+/* PSP runtime DB boot config entry */
+struct psp_runtime_boot_cfg_entry {
+	uint32_t boot_cfg_bitmask;
+	uint32_t reserved;
+};
+
+/* PSP runtime DB SCPM entry */
+struct psp_runtime_scpm_entry {
+	enum psp_runtime_scpm_authentication scpm_status;
+};
+
+#if defined(CONFIG_DEBUG_FS)
+struct spirom_bo {
+	struct amdgpu_bo *bo;
+	uint64_t mc_addr;
+	void *cpu_addr;
+};
+#endif
+
+struct psp_context {
+	struct amdgpu_device		*adev;
+	struct psp_ring			km_ring;
+	struct psp_gfx_cmd_resp		*cmd;
+
+	const struct psp_funcs		*funcs;
+	const struct ta_funcs		*ta_funcs;
+
+	/* firmware buffer */
+	struct amdgpu_bo		*fw_pri_bo;
+	uint64_t			fw_pri_mc_addr;
 	void				*fw_pri_buf;
 
 	/* sos firmware */
 	const struct firmware		*sos_fw;
-	uint32_t			sos_fw_version;
-	uint32_t			sos_feature_version;
-	uint32_t			sys_bin_size;
-	uint32_t			sos_bin_size;
-	uint8_t				*sys_start_addr;
-	uint8_t				*sos_start_addr;
+	struct psp_bin_desc		sys;
+	struct psp_bin_desc		sos;
+	struct psp_bin_desc		toc;
+	struct psp_bin_desc		kdb;
+	struct psp_bin_desc		spl;
+	struct psp_bin_desc		rl;
+	struct psp_bin_desc		soc_drv;
+	struct psp_bin_desc		intf_drv;
+	struct psp_bin_desc		dbg_drv;
+	struct psp_bin_desc		ras_drv;
+	struct psp_bin_desc		ipkeymgr_drv;
+	struct psp_bin_desc		spdm_drv;
 
 	/* tmr buffer */
-	struct amdgpu_bo 		*tmr_bo;
-	uint64_t 			tmr_mc_addr;
-	void				*tmr_buf;
+	struct amdgpu_bo		*tmr_bo;
+	uint64_t			tmr_mc_addr;
 
-	/* asd firmware and buffer */
+	/* asd firmware */
 	const struct firmware		*asd_fw;
-	uint32_t			asd_fw_version;
-	uint32_t			asd_feature_version;
-	uint32_t			asd_ucode_size;
-	uint8_t				*asd_start_addr;
-	struct amdgpu_bo 		*asd_shared_bo;
-	uint64_t 			asd_shared_mc_addr;
-	void				*asd_shared_buf;
+
+	/* toc firmware */
+	const struct firmware		*toc_fw;
+
+	/* cap firmware */
+	const struct firmware		*cap_fw;
 
 	/* fence buffer */
-	struct amdgpu_bo 		*fence_buf_bo;
-	uint64_t 			fence_buf_mc_addr;
+	struct amdgpu_bo		*fence_buf_bo;
+	uint64_t			fence_buf_mc_addr;
 	void				*fence_buf;
+
+	/* cmd buffer */
+	struct amdgpu_bo		*cmd_buf_bo;
+	uint64_t			cmd_buf_mc_addr;
+	struct psp_gfx_cmd_resp		*cmd_buf_mem;
+
+	/* fence value associated with cmd buffer */
+	atomic_t			fence_value;
+	/* flag to mark whether gfx fw autoload is supported or not */
+	bool				autoload_supported;
+	/* flag to mark whether psp use runtime TMR or boottime TMR */
+	bool				boot_time_tmr;
+	/* flag to mark whether df cstate management centralized to PMFW */
+	bool				pmfw_centralized_cstate_management;
+
+	/* xgmi ta firmware and buffer */
+	const struct firmware		*ta_fw;
+	uint32_t			ta_fw_version;
+
+	uint32_t			cap_fw_version;
+	uint32_t			cap_feature_version;
+	uint32_t			cap_ucode_size;
+
+	struct ta_context		asd_context;
+	struct psp_xgmi_context		xgmi_context;
+	struct psp_ras_context		ras_context;
+	struct ta_cp_context		hdcp_context;
+	struct ta_cp_context		dtm_context;
+	struct ta_cp_context		rap_context;
+	struct ta_cp_context		securedisplay_context;
+	struct mutex			mutex;
+	struct psp_memory_training_context mem_train_ctx;
+
+	uint32_t			boot_cfg_bitmask;
+
+	/* firmware upgrades supported */
+	bool				sup_pd_fw_up;
+	bool				sup_ifwi_up;
+
+	char				*vbflash_tmp_buf;
+	size_t				vbflash_image_size;
+	bool				vbflash_done;
+#if defined(CONFIG_DEBUG_FS)
+	struct spirom_bo *spirom_dump_trip;
+#endif
 };
 
 struct amdgpu_psp_funcs {
@@ -115,27 +454,173 @@ struct amdgpu_psp_funcs {
 					enum AMDGPU_UCODE_ID);
 };
 
-#define psp_prep_cmd_buf(ucode, type) (psp)->prep_cmd_buf((ucode), (type))
-#define psp_ring_init(psp, type) (psp)->ring_init((psp), (type))
-#define psp_ring_create(psp, type) (psp)->ring_create((psp), (type))
-#define psp_ring_destroy(psp, type) ((psp)->ring_destroy((psp), (type)))
-#define psp_cmd_submit(psp, ucode, cmd_mc, fence_mc, index) \
-		(psp)->cmd_submit((psp), (ucode), (cmd_mc), (fence_mc), (index))
-#define psp_compare_sram_data(psp, ucode, type) \
-		(psp)->compare_sram_data((psp), (ucode), (type))
+
+#define psp_ring_create(psp, type) (psp)->funcs->ring_create((psp), (type))
+#define psp_ring_stop(psp, type) (psp)->funcs->ring_stop((psp), (type))
+#define psp_ring_destroy(psp, type) ((psp)->funcs->ring_destroy((psp), (type)))
 #define psp_init_microcode(psp) \
-		((psp)->init_microcode ? (psp)->init_microcode((psp)) : 0)
+		((psp)->funcs->init_microcode ? (psp)->funcs->init_microcode((psp)) : 0)
+#define psp_bootloader_load_kdb(psp) \
+		((psp)->funcs->bootloader_load_kdb ? (psp)->funcs->bootloader_load_kdb((psp)) : 0)
+#define psp_bootloader_load_spl(psp) \
+		((psp)->funcs->bootloader_load_spl ? (psp)->funcs->bootloader_load_spl((psp)) : 0)
 #define psp_bootloader_load_sysdrv(psp) \
-		((psp)->bootloader_load_sysdrv ? (psp)->bootloader_load_sysdrv((psp)) : 0)
+		((psp)->funcs->bootloader_load_sysdrv ? (psp)->funcs->bootloader_load_sysdrv((psp)) : 0)
+#define psp_bootloader_load_soc_drv(psp) \
+		((psp)->funcs->bootloader_load_soc_drv ? (psp)->funcs->bootloader_load_soc_drv((psp)) : 0)
+#define psp_bootloader_load_intf_drv(psp) \
+		((psp)->funcs->bootloader_load_intf_drv ? (psp)->funcs->bootloader_load_intf_drv((psp)) : 0)
+#define psp_bootloader_load_dbg_drv(psp) \
+		((psp)->funcs->bootloader_load_dbg_drv ? (psp)->funcs->bootloader_load_dbg_drv((psp)) : 0)
+#define psp_bootloader_load_ras_drv(psp) \
+		((psp)->funcs->bootloader_load_ras_drv ? \
+		(psp)->funcs->bootloader_load_ras_drv((psp)) : 0)
+#define psp_bootloader_load_ipkeymgr_drv(psp) \
+		((psp)->funcs->bootloader_load_ipkeymgr_drv ? \
+		 (psp)->funcs->bootloader_load_ipkeymgr_drv((psp)) : 0)
+#define psp_bootloader_load_spdm_drv(psp) \
+		((psp)->funcs->bootloader_load_spdm_drv ? \
+		 (psp)->funcs->bootloader_load_spdm_drv((psp)) : 0)
 #define psp_bootloader_load_sos(psp) \
-		((psp)->bootloader_load_sos ? (psp)->bootloader_load_sos((psp)) : 0)
+		((psp)->funcs->bootloader_load_sos ? (psp)->funcs->bootloader_load_sos((psp)) : 0)
 #define psp_smu_reload_quirk(psp) \
-		((psp)->smu_reload_quirk ? (psp)->smu_reload_quirk((psp)) : false)
+		((psp)->funcs->smu_reload_quirk ? (psp)->funcs->smu_reload_quirk((psp)) : false)
+#define psp_mode1_reset(psp) \
+		((psp)->funcs->mode1_reset ? (psp)->funcs->mode1_reset((psp)) : false)
+#define psp_mem_training(psp, ops) \
+	((psp)->funcs->mem_training ? (psp)->funcs->mem_training((psp), (ops)) : 0)
+
+#define psp_ring_get_wptr(psp) (psp)->funcs->ring_get_wptr((psp))
+#define psp_ring_set_wptr(psp, value) (psp)->funcs->ring_set_wptr((psp), (value))
+
+#define psp_load_usbc_pd_fw(psp, fw_pri_mc_addr) \
+	((psp)->funcs->load_usbc_pd_fw ? \
+	(psp)->funcs->load_usbc_pd_fw((psp), (fw_pri_mc_addr)) : -EINVAL)
+
+#define psp_read_usbc_pd_fw(psp, fw_ver) \
+	((psp)->funcs->read_usbc_pd_fw ? \
+	(psp)->funcs->read_usbc_pd_fw((psp), fw_ver) : -EINVAL)
+
+#define psp_update_spirom(psp, fw_pri_mc_addr) \
+	((psp)->funcs->update_spirom ? \
+	(psp)->funcs->update_spirom((psp), fw_pri_mc_addr) : -EINVAL)
+
+#define psp_dump_spirom(psp, fw_pri_mc_addr) \
+	((psp)->funcs->dump_spirom ? \
+	(psp)->funcs->dump_spirom((psp), fw_pri_mc_addr) : -EINVAL)
+
+#define psp_vbflash_status(psp) \
+	((psp)->funcs->vbflash_stat ? \
+	(psp)->funcs->vbflash_stat((psp)) : -EINVAL)
+
+#define psp_fatal_error_recovery_quirk(psp) \
+	((psp)->funcs->fatal_error_recovery_quirk ? \
+	(psp)->funcs->fatal_error_recovery_quirk((psp)) : 0)
+
+#define psp_is_aux_sos_load_required(psp) \
+	((psp)->funcs->is_aux_sos_load_required ? (psp)->funcs->is_aux_sos_load_required((psp)) : 0)
+
+#define psp_reg_program_no_ring(psp, val, id) \
+	((psp)->funcs->reg_program_no_ring ? \
+	(psp)->funcs->reg_program_no_ring((psp), val, id) : -EINVAL)
 
 extern const struct amd_ip_funcs psp_ip_funcs;
 
 extern const struct amdgpu_ip_block_version psp_v3_1_ip_block;
-extern int psp_wait_for(struct psp_context *psp, uint32_t reg_index,
-			uint32_t field_val, uint32_t mask, bool check_changed);
+extern const struct amdgpu_ip_block_version psp_v10_0_ip_block;
+extern const struct amdgpu_ip_block_version psp_v11_0_ip_block;
+extern const struct amdgpu_ip_block_version psp_v11_0_8_ip_block;
+extern const struct amdgpu_ip_block_version psp_v12_0_ip_block;
+extern const struct amdgpu_ip_block_version psp_v13_0_ip_block;
+extern const struct amdgpu_ip_block_version psp_v13_0_4_ip_block;
+extern const struct amdgpu_ip_block_version psp_v14_0_ip_block;
+
+int psp_wait_for(struct psp_context *psp, uint32_t reg_index,
+		 uint32_t field_val, uint32_t mask, uint32_t flags);
+extern int psp_wait_for_spirom_update(struct psp_context *psp, uint32_t reg_index,
+			uint32_t field_val, uint32_t mask, uint32_t msec_timeout);
+
+int psp_execute_ip_fw_load(struct psp_context *psp,
+			   struct amdgpu_firmware_info *ucode);
+
+int psp_gpu_reset(struct amdgpu_device *adev);
+
+int psp_ta_init_shared_buf(struct psp_context *psp,
+				  struct ta_mem_context *mem_ctx);
+void psp_ta_free_shared_buf(struct ta_mem_context *mem_ctx);
+int psp_ta_unload(struct psp_context *psp, struct ta_context *context);
+int psp_ta_load(struct psp_context *psp, struct ta_context *context);
+int psp_ta_invoke(struct psp_context *psp,
+			uint32_t ta_cmd_id,
+			struct ta_context *context);
+
+int psp_xgmi_initialize(struct psp_context *psp, bool set_extended_data, bool load_ta);
+int psp_xgmi_terminate(struct psp_context *psp);
+int psp_xgmi_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+int psp_xgmi_get_hive_id(struct psp_context *psp, uint64_t *hive_id);
+int psp_xgmi_get_node_id(struct psp_context *psp, uint64_t *node_id);
+int psp_xgmi_get_topology_info(struct psp_context *psp,
+			       int number_devices,
+			       struct psp_xgmi_topology_info *topology,
+			       bool get_extended_data);
+int psp_xgmi_set_topology_info(struct psp_context *psp,
+			       int number_devices,
+			       struct psp_xgmi_topology_info *topology);
+int psp_ras_initialize(struct psp_context *psp);
+int psp_ras_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+int psp_ras_enable_features(struct psp_context *psp,
+		union ta_ras_cmd_input *info, bool enable);
+int psp_ras_trigger_error(struct psp_context *psp,
+			  struct ta_ras_trigger_error_input *info, uint32_t instance_mask);
+int psp_ras_terminate(struct psp_context *psp);
+int psp_ras_query_address(struct psp_context *psp,
+			  struct ta_ras_query_address_input *addr_in,
+			  struct ta_ras_query_address_output *addr_out);
+
+int psp_hdcp_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+int psp_dtm_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id, enum ta_rap_status *status);
+int psp_securedisplay_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+
+int psp_rlc_autoload_start(struct psp_context *psp);
+
+int psp_reg_program(struct psp_context *psp, enum psp_reg_prog_id reg,
+		uint32_t value);
+int psp_ring_cmd_submit(struct psp_context *psp,
+			uint64_t cmd_buf_mc_addr,
+			uint64_t fence_mc_addr,
+			int index);
+int psp_init_asd_microcode(struct psp_context *psp,
+			   const char *chip_name);
+int psp_init_toc_microcode(struct psp_context *psp,
+			   const char *chip_name);
+int psp_init_sos_microcode(struct psp_context *psp,
+			   const char *chip_name);
+int psp_init_ta_microcode(struct psp_context *psp,
+			  const char *chip_name);
+int psp_init_cap_microcode(struct psp_context *psp,
+			  const char *chip_name);
+int psp_get_fw_attestation_records_addr(struct psp_context *psp,
+					uint64_t *output_ptr);
+int psp_update_fw_reservation(struct psp_context *psp);
+int psp_load_fw_list(struct psp_context *psp,
+		     struct amdgpu_firmware_info **ucode_list, int ucode_count);
+void psp_copy_fw(struct psp_context *psp, uint8_t *start_addr, uint32_t bin_size);
+
+int psp_spatial_partition(struct psp_context *psp, int mode);
+int psp_memory_partition(struct psp_context *psp, int mode);
+
+int is_psp_fw_valid(struct psp_bin_desc bin);
+
+int amdgpu_psp_wait_for_bootloader(struct amdgpu_device *adev);
+bool amdgpu_psp_get_ras_capability(struct psp_context *psp);
+
+int psp_config_sq_perfmon(struct psp_context *psp, uint32_t xcp_id,
+	bool core_override_enable, bool reg_override_enable, bool perfmon_override_enable);
+bool amdgpu_psp_tos_reload_needed(struct amdgpu_device *adev);
+int amdgpu_psp_reg_program_no_ring(struct psp_context *psp, uint32_t val,
+				   enum psp_reg_prog_id id);
+void amdgpu_psp_debugfs_init(struct amdgpu_device *adev);
+
 
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp_ta.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp_ta.c
new file mode 100644
index 000000000000..6e8aad91bcd3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp_ta.c
@@ -0,0 +1,392 @@
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "amdgpu.h"
+#include "amdgpu_psp_ta.h"
+
+#if defined(CONFIG_DEBUG_FS)
+
+static ssize_t ta_if_load_debugfs_write(struct file *fp, const char *buf,
+					    size_t len, loff_t *off);
+static ssize_t ta_if_unload_debugfs_write(struct file *fp, const char *buf,
+					    size_t len, loff_t *off);
+static ssize_t ta_if_invoke_debugfs_write(struct file *fp, const char *buf,
+					    size_t len, loff_t *off);
+
+static uint32_t get_bin_version(const uint8_t *bin)
+{
+	const struct common_firmware_header *hdr =
+			     (const struct common_firmware_header *)bin;
+
+	return hdr->ucode_version;
+}
+
+static int prep_ta_mem_context(struct ta_mem_context *mem_context,
+					     uint8_t *shared_buf,
+					     uint32_t shared_buf_len)
+{
+	if (mem_context->shared_mem_size < shared_buf_len)
+		return -EINVAL;
+	memset(mem_context->shared_buf, 0, mem_context->shared_mem_size);
+	memcpy((void *)mem_context->shared_buf, shared_buf, shared_buf_len);
+
+	return 0;
+}
+
+static bool is_ta_type_valid(enum ta_type_id ta_type)
+{
+	switch (ta_type) {
+	case TA_TYPE_RAS:
+		return true;
+	default:
+		return false;
+	}
+}
+
+static const struct ta_funcs ras_ta_funcs = {
+	.fn_ta_initialize = psp_ras_initialize,
+	.fn_ta_invoke    = psp_ras_invoke,
+	.fn_ta_terminate = psp_ras_terminate
+};
+
+static void set_ta_context_funcs(struct psp_context *psp,
+						      enum ta_type_id ta_type,
+						      struct ta_context **pcontext)
+{
+	switch (ta_type) {
+	case TA_TYPE_RAS:
+		*pcontext = &psp->ras_context.context;
+		psp->ta_funcs = &ras_ta_funcs;
+		break;
+	default:
+		break;
+	}
+}
+
+static const struct file_operations ta_load_debugfs_fops = {
+	.write  = ta_if_load_debugfs_write,
+	.llseek = default_llseek,
+	.owner  = THIS_MODULE
+};
+
+static const struct file_operations ta_unload_debugfs_fops = {
+	.write  = ta_if_unload_debugfs_write,
+	.llseek = default_llseek,
+	.owner  = THIS_MODULE
+};
+
+static const struct file_operations ta_invoke_debugfs_fops = {
+	.write  = ta_if_invoke_debugfs_write,
+	.llseek = default_llseek,
+	.owner  = THIS_MODULE
+};
+
+/*
+ * DOC: AMDGPU TA debugfs interfaces
+ *
+ * Three debugfs interfaces can be opened by a program to
+ * load/invoke/unload TA,
+ *
+ * - /sys/kernel/debug/dri/<N>/ta_if/ta_load
+ * - /sys/kernel/debug/dri/<N>/ta_if/ta_invoke
+ * - /sys/kernel/debug/dri/<N>/ta_if/ta_unload
+ *
+ * How to use the interfaces in a program?
+ *
+ * A program needs to provide transmit buffer to the interfaces
+ * and will receive buffer from the interfaces below,
+ *
+ * - For TA load debugfs interface:
+ *   Transmit buffer:
+ *    - TA type (4bytes)
+ *    - TA bin length (4bytes)
+ *    - TA bin
+ *   Receive buffer:
+ *    - TA ID (4bytes)
+ *
+ * - For TA invoke debugfs interface:
+ *   Transmit buffer:
+ *    - TA type (4bytes)
+ *    - TA ID (4bytes)
+ *    - TA CMD ID (4bytes)
+ *    - TA shard buf length
+ *      (4bytes, value not beyond TA shared memory size)
+ *    - TA shared buf
+ *   Receive buffer:
+ *    - TA shared buf
+ *
+ * - For TA unload debugfs interface:
+ *   Transmit buffer:
+ *    - TA type (4bytes)
+ *    - TA ID (4bytes)
+ */
+
+static ssize_t ta_if_load_debugfs_write(struct file *fp, const char *buf, size_t len, loff_t *off)
+{
+	uint32_t ta_type    = 0;
+	uint32_t ta_bin_len = 0;
+	uint8_t  *ta_bin    = NULL;
+	uint32_t copy_pos   = 0;
+	int      ret        = 0;
+
+	struct amdgpu_device *adev    = (struct amdgpu_device *)file_inode(fp)->i_private;
+	struct psp_context   *psp     = &adev->psp;
+	struct ta_context    *context = NULL;
+
+	if (!buf)
+		return -EINVAL;
+
+	ret = copy_from_user((void *)&ta_type, &buf[copy_pos], sizeof(uint32_t));
+	if (ret || (!is_ta_type_valid(ta_type)))
+		return -EFAULT;
+
+	copy_pos += sizeof(uint32_t);
+
+	ret = copy_from_user((void *)&ta_bin_len, &buf[copy_pos], sizeof(uint32_t));
+	if (ret)
+		return -EFAULT;
+
+	if (ta_bin_len > PSP_1_MEG)
+		return -EINVAL;
+
+	copy_pos += sizeof(uint32_t);
+
+	ta_bin = memdup_user(&buf[copy_pos], ta_bin_len);
+	if (IS_ERR(ta_bin))
+		return PTR_ERR(ta_bin);
+
+	/* Set TA context and functions */
+	set_ta_context_funcs(psp, ta_type, &context);
+
+	if (!psp->ta_funcs || !psp->ta_funcs->fn_ta_terminate) {
+		dev_err(adev->dev, "Unsupported function to terminate TA\n");
+		ret = -EOPNOTSUPP;
+		goto err_free_bin;
+	}
+
+	/*
+	 * Allocate TA shared buf in case shared buf was freed
+	 * due to loading TA failed before.
+	 */
+	if (!context->mem_context.shared_buf) {
+		ret = psp_ta_init_shared_buf(psp, &context->mem_context);
+		if (ret) {
+			ret = -ENOMEM;
+			goto err_free_bin;
+		}
+	}
+
+	ret = psp_fn_ta_terminate(psp);
+	if (ret || context->resp_status) {
+		dev_err(adev->dev,
+			"Failed to unload embedded TA (%d) and status (0x%X)\n",
+			ret, context->resp_status);
+		if (!ret)
+			ret = -EINVAL;
+		goto err_free_ta_shared_buf;
+	}
+
+	/* Prepare TA context for TA initialization */
+	context->ta_type                     = ta_type;
+	context->bin_desc.fw_version         = get_bin_version(ta_bin);
+	context->bin_desc.size_bytes         = ta_bin_len;
+	context->bin_desc.start_addr         = ta_bin;
+
+	if (!psp->ta_funcs->fn_ta_initialize) {
+		dev_err(adev->dev, "Unsupported function to initialize TA\n");
+		ret = -EOPNOTSUPP;
+		goto err_free_ta_shared_buf;
+	}
+
+	ret = psp_fn_ta_initialize(psp);
+	if (ret || context->resp_status) {
+		dev_err(adev->dev, "Failed to load TA via debugfs (%d) and status (0x%X)\n",
+			ret, context->resp_status);
+		if (!ret)
+			ret = -EINVAL;
+		goto err_free_ta_shared_buf;
+	}
+
+	if (copy_to_user((char *)buf, (void *)&context->session_id, sizeof(uint32_t)))
+		ret = -EFAULT;
+
+err_free_ta_shared_buf:
+	/* Only free TA shared buf when returns error code */
+	if (ret && context->mem_context.shared_buf)
+		psp_ta_free_shared_buf(&context->mem_context);
+err_free_bin:
+	kfree(ta_bin);
+
+	return ret;
+}
+
+static ssize_t ta_if_unload_debugfs_write(struct file *fp, const char *buf, size_t len, loff_t *off)
+{
+	uint32_t ta_type    = 0;
+	uint32_t ta_id      = 0;
+	uint32_t copy_pos   = 0;
+	int      ret        = 0;
+
+	struct amdgpu_device *adev    = (struct amdgpu_device *)file_inode(fp)->i_private;
+	struct psp_context   *psp     = &adev->psp;
+	struct ta_context    *context = NULL;
+
+	if (!buf)
+		return -EINVAL;
+
+	ret = copy_from_user((void *)&ta_type, &buf[copy_pos], sizeof(uint32_t));
+	if (ret || (!is_ta_type_valid(ta_type)))
+		return -EFAULT;
+
+	copy_pos += sizeof(uint32_t);
+
+	ret = copy_from_user((void *)&ta_id, &buf[copy_pos], sizeof(uint32_t));
+	if (ret)
+		return -EFAULT;
+
+	set_ta_context_funcs(psp, ta_type, &context);
+	context->session_id = ta_id;
+
+	if (!psp->ta_funcs || !psp->ta_funcs->fn_ta_terminate) {
+		dev_err(adev->dev, "Unsupported function to terminate TA\n");
+		return -EOPNOTSUPP;
+	}
+
+	ret = psp_fn_ta_terminate(psp);
+	if (ret || context->resp_status) {
+		dev_err(adev->dev, "Failed to unload TA via debugfs (%d) and status (0x%X)\n",
+			ret, context->resp_status);
+		if (!ret)
+			ret = -EINVAL;
+	}
+
+	if (context->mem_context.shared_buf)
+		psp_ta_free_shared_buf(&context->mem_context);
+
+	return ret;
+}
+
+static ssize_t ta_if_invoke_debugfs_write(struct file *fp, const char *buf, size_t len, loff_t *off)
+{
+	uint32_t ta_type        = 0;
+	uint32_t ta_id          = 0;
+	uint32_t cmd_id         = 0;
+	uint32_t shared_buf_len = 0;
+	uint8_t *shared_buf     = NULL;
+	uint32_t copy_pos       = 0;
+	int      ret            = 0;
+
+	struct amdgpu_device *adev    = (struct amdgpu_device *)file_inode(fp)->i_private;
+	struct psp_context   *psp     = &adev->psp;
+	struct ta_context    *context = NULL;
+
+	if (!buf)
+		return -EINVAL;
+
+	ret = copy_from_user((void *)&ta_type, &buf[copy_pos], sizeof(uint32_t));
+	if (ret)
+		return -EFAULT;
+	copy_pos += sizeof(uint32_t);
+
+	ret = copy_from_user((void *)&ta_id, &buf[copy_pos], sizeof(uint32_t));
+	if (ret)
+		return -EFAULT;
+	copy_pos += sizeof(uint32_t);
+
+	ret = copy_from_user((void *)&cmd_id, &buf[copy_pos], sizeof(uint32_t));
+	if (ret)
+		return -EFAULT;
+	copy_pos += sizeof(uint32_t);
+
+	ret = copy_from_user((void *)&shared_buf_len, &buf[copy_pos], sizeof(uint32_t));
+	if (ret)
+		return -EFAULT;
+	copy_pos += sizeof(uint32_t);
+
+	shared_buf = memdup_user(&buf[copy_pos], shared_buf_len);
+	if (IS_ERR(shared_buf))
+		return PTR_ERR(shared_buf);
+
+	set_ta_context_funcs(psp, ta_type, &context);
+
+	if (!context || !context->initialized) {
+		dev_err(adev->dev, "TA is not initialized\n");
+		ret = -EINVAL;
+		goto err_free_shared_buf;
+	}
+
+	if (!psp->ta_funcs || !psp->ta_funcs->fn_ta_invoke) {
+		dev_err(adev->dev, "Unsupported function to invoke TA\n");
+		ret = -EOPNOTSUPP;
+		goto err_free_shared_buf;
+	}
+
+	context->session_id = ta_id;
+
+	mutex_lock(&psp->ras_context.mutex);
+	ret = prep_ta_mem_context(&context->mem_context, shared_buf, shared_buf_len);
+	if (ret)
+		goto err_free_shared_buf;
+
+	ret = psp_fn_ta_invoke(psp, cmd_id);
+	if (ret || context->resp_status) {
+		dev_err(adev->dev, "Failed to invoke TA via debugfs (%d) and status (0x%X)\n",
+			ret, context->resp_status);
+		if (!ret) {
+			ret = -EINVAL;
+			goto err_free_shared_buf;
+		}
+	}
+
+	if (copy_to_user((char *)&buf[copy_pos], context->mem_context.shared_buf, shared_buf_len))
+		ret = -EFAULT;
+
+err_free_shared_buf:
+	mutex_unlock(&psp->ras_context.mutex);
+	kfree(shared_buf);
+
+	return ret;
+}
+
+void amdgpu_ta_if_debugfs_init(struct amdgpu_device *adev)
+{
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+
+	struct dentry *dir = debugfs_create_dir("ta_if", minor->debugfs_root);
+
+	debugfs_create_file("ta_load", 0200, dir, adev,
+				     &ta_load_debugfs_fops);
+
+	debugfs_create_file("ta_unload", 0200, dir,
+				     adev, &ta_unload_debugfs_fops);
+
+	debugfs_create_file("ta_invoke", 0200, dir,
+				     adev, &ta_invoke_debugfs_fops);
+}
+
+#else
+void amdgpu_ta_if_debugfs_init(struct amdgpu_device *adev)
+{
+
+}
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp_ta.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp_ta.h
new file mode 100644
index 000000000000..14cd1c81c3e6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp_ta.h
@@ -0,0 +1,34 @@
+/*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __AMDGPU_PSP_TA_H__
+#define __AMDGPU_PSP_TA_H__
+
+/* Calling set_ta_context_funcs is required before using the following macros */
+#define psp_fn_ta_initialize(psp) ((psp)->ta_funcs->fn_ta_initialize((psp)))
+#define psp_fn_ta_invoke(psp, ta_cmd_id) ((psp)->ta_funcs->fn_ta_invoke((psp), (ta_cmd_id)))
+#define psp_fn_ta_terminate(psp) ((psp)->ta_funcs->fn_ta_terminate((psp)))
+
+void amdgpu_ta_if_debugfs_init(struct amdgpu_device *adev);
+
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_rap.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_rap.c
new file mode 100644
index 000000000000..bacf888735db
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_rap.c
@@ -0,0 +1,125 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ *
+ */
+#include <linux/debugfs.h>
+#include <linux/pm_runtime.h>
+
+#include "amdgpu.h"
+#include "amdgpu_rap.h"
+
+/**
+ * DOC: AMDGPU RAP debugfs test interface
+ *
+ * how to use?
+ * echo opcode > <debugfs_dir>/dri/xxx/rap_test
+ *
+ * opcode:
+ * currently, only 2 is supported by Linux host driver,
+ * opcode 2 stands for TA_CMD_RAP__VALIDATE_L0, used to
+ * trigger L0 policy validation, you can refer more detail
+ * from header file ta_rap_if.h
+ *
+ */
+static ssize_t amdgpu_rap_debugfs_write(struct file *f, const char __user *buf,
+		size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
+	struct ta_rap_shared_memory *rap_shared_mem;
+	struct ta_rap_cmd_output_data *rap_cmd_output;
+	struct drm_device *dev = adev_to_drm(adev);
+	uint32_t op;
+	enum ta_rap_status status;
+	int ret;
+
+	if (*pos || size != 2)
+		return -EINVAL;
+
+	ret = kstrtouint_from_user(buf, size, *pos, &op);
+	if (ret)
+		return ret;
+
+	ret = pm_runtime_get_sync(dev->dev);
+	if (ret < 0) {
+		pm_runtime_put_autosuspend(dev->dev);
+		return ret;
+	}
+
+	/* make sure gfx core is on, RAP TA cann't handle
+	 * GFX OFF case currently.
+	 */
+	amdgpu_gfx_off_ctrl(adev, false);
+
+	switch (op) {
+	case 2:
+		ret = psp_rap_invoke(&adev->psp, op, &status);
+		if (!ret && status == TA_RAP_STATUS__SUCCESS) {
+			dev_info(adev->dev, "RAP L0 validate test success.\n");
+		} else {
+			rap_shared_mem = (struct ta_rap_shared_memory *)
+					 adev->psp.rap_context.context.mem_context.shared_buf;
+			rap_cmd_output = &(rap_shared_mem->rap_out_message.output);
+
+			dev_info(adev->dev, "RAP test failed, the output is:\n");
+			dev_info(adev->dev, "\tlast_subsection: 0x%08x.\n",
+				 rap_cmd_output->last_subsection);
+			dev_info(adev->dev, "\tnum_total_validate: 0x%08x.\n",
+				 rap_cmd_output->num_total_validate);
+			dev_info(adev->dev, "\tnum_valid: 0x%08x.\n",
+				 rap_cmd_output->num_valid);
+			dev_info(adev->dev, "\tlast_validate_addr: 0x%08x.\n",
+				 rap_cmd_output->last_validate_addr);
+			dev_info(adev->dev, "\tlast_validate_val: 0x%08x.\n",
+				 rap_cmd_output->last_validate_val);
+			dev_info(adev->dev, "\tlast_validate_val_exptd: 0x%08x.\n",
+				 rap_cmd_output->last_validate_val_exptd);
+		}
+		break;
+	default:
+		dev_info(adev->dev, "Unsupported op id: %d, ", op);
+		dev_info(adev->dev, "Only support op 2(L0 validate test).\n");
+		break;
+	}
+
+	amdgpu_gfx_off_ctrl(adev, true);
+	pm_runtime_put_autosuspend(dev->dev);
+
+	return size;
+}
+
+static const struct file_operations amdgpu_rap_debugfs_ops = {
+	.owner = THIS_MODULE,
+	.read = NULL,
+	.write = amdgpu_rap_debugfs_write,
+	.llseek = default_llseek
+};
+
+void amdgpu_rap_debugfs_init(struct amdgpu_device *adev)
+{
+	struct drm_minor *minor = adev_to_drm(adev)->primary;
+
+	if (!adev->psp.rap_context.context.initialized)
+		return;
+
+	debugfs_create_file("rap_test", S_IWUSR, minor->debugfs_root,
+				adev, &amdgpu_rap_debugfs_ops);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_rap.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_rap.h
new file mode 100644
index 000000000000..ec6d7632d3a0
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_rap.h
@@ -0,0 +1,30 @@
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ *
+ */
+#ifndef _AMDGPU_RAP_H
+#define _AMDGPU_RAP_H
+
+#include "amdgpu.h"
+
+void amdgpu_rap_debugfs_init(struct amdgpu_device *adev);
+#endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
new file mode 100644
index 000000000000..2a6cf7963dde
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -0,0 +1,5748 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ *
+ */
+#include <linux/debugfs.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/uaccess.h>
+#include <linux/reboot.h>
+#include <linux/syscalls.h>
+#include <linux/pm_runtime.h>
+#include <linux/list_sort.h>
+
+#include "amdgpu.h"
+#include "amdgpu_ras.h"
+#include "amdgpu_atomfirmware.h"
+#include "amdgpu_xgmi.h"
+#include "ivsrcid/nbio/irqsrcs_nbif_7_4.h"
+#include "nbio_v4_3.h"
+#include "nbif_v6_3_1.h"
+#include "nbio_v7_9.h"
+#include "atom.h"
+#include "amdgpu_reset.h"
+#include "amdgpu_psp.h"
+#include "amdgpu_ras_mgr.h"
+
+#ifdef CONFIG_X86_MCE_AMD
+#include <asm/mce.h>
+
+static bool notifier_registered;
+#endif
+static const char *RAS_FS_NAME = "ras";
+
+const char *ras_error_string[] = {
+	"none",
+	"parity",
+	"single_correctable",
+	"multi_uncorrectable",
+	"poison",
+};
+
+const char *ras_block_string[] = {
+	"umc",
+	"sdma",
+	"gfx",
+	"mmhub",
+	"athub",
+	"pcie_bif",
+	"hdp",
+	"xgmi_wafl",
+	"df",
+	"smn",
+	"sem",
+	"mp0",
+	"mp1",
+	"fuse",
+	"mca",
+	"vcn",
+	"jpeg",
+	"ih",
+	"mpio",
+	"mmsch",
+};
+
+const char *ras_mca_block_string[] = {
+	"mca_mp0",
+	"mca_mp1",
+	"mca_mpio",
+	"mca_iohc",
+};
+
+struct amdgpu_ras_block_list {
+	/* ras block link */
+	struct list_head node;
+
+	struct amdgpu_ras_block_object *ras_obj;
+};
+
+const char *get_ras_block_str(struct ras_common_if *ras_block)
+{
+	if (!ras_block)
+		return "NULL";
+
+	if (ras_block->block >= AMDGPU_RAS_BLOCK_COUNT ||
+	    ras_block->block >= ARRAY_SIZE(ras_block_string))
+		return "OUT OF RANGE";
+
+	if (ras_block->block == AMDGPU_RAS_BLOCK__MCA)
+		return ras_mca_block_string[ras_block->sub_block_index];
+
+	return ras_block_string[ras_block->block];
+}
+
+#define ras_block_str(_BLOCK_) \
+	(((_BLOCK_) < ARRAY_SIZE(ras_block_string)) ? ras_block_string[_BLOCK_] : "Out Of Range")
+
+#define ras_err_str(i) (ras_error_string[ffs(i)])
+
+#define RAS_DEFAULT_FLAGS (AMDGPU_RAS_FLAG_INIT_BY_VBIOS)
+
+/* inject address is 52 bits */
+#define	RAS_UMC_INJECT_ADDR_LIMIT	(0x1ULL << 52)
+
+/* typical ECC bad page rate is 1 bad page per 100MB VRAM */
+#define RAS_BAD_PAGE_COVER              (100 * 1024 * 1024ULL)
+
+#define MAX_UMC_POISON_POLLING_TIME_ASYNC  10
+
+#define AMDGPU_RAS_RETIRE_PAGE_INTERVAL 100  //ms
+
+#define MAX_FLUSH_RETIRE_DWORK_TIMES  100
+
+#define BYPASS_ALLOCATED_ADDRESS        0x0
+#define BYPASS_INITIALIZATION_ADDRESS   0x1
+
+enum amdgpu_ras_retire_page_reservation {
+	AMDGPU_RAS_RETIRE_PAGE_RESERVED,
+	AMDGPU_RAS_RETIRE_PAGE_PENDING,
+	AMDGPU_RAS_RETIRE_PAGE_FAULT,
+};
+
+atomic_t amdgpu_ras_in_intr = ATOMIC_INIT(0);
+
+static int amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con,
+				uint64_t addr);
+static int amdgpu_ras_check_bad_page(struct amdgpu_device *adev,
+				uint64_t addr);
+
+static void amdgpu_ras_critical_region_init(struct amdgpu_device *adev);
+static void amdgpu_ras_critical_region_fini(struct amdgpu_device *adev);
+
+#ifdef CONFIG_X86_MCE_AMD
+static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev);
+static void
+amdgpu_unregister_bad_pages_mca_notifier(struct amdgpu_device *adev);
+struct mce_notifier_adev_list {
+	struct amdgpu_device *devs[MAX_GPU_INSTANCE];
+	int num_gpu;
+};
+static struct mce_notifier_adev_list mce_adev_list;
+#endif
+
+void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready)
+{
+	if (adev && amdgpu_ras_get_context(adev))
+		amdgpu_ras_get_context(adev)->error_query_ready = ready;
+}
+
+static bool amdgpu_ras_get_error_query_ready(struct amdgpu_device *adev)
+{
+	if (adev && amdgpu_ras_get_context(adev))
+		return amdgpu_ras_get_context(adev)->error_query_ready;
+
+	return false;
+}
+
+static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t address)
+{
+	struct ras_err_data err_data;
+	struct eeprom_table_record err_rec;
+	int ret;
+
+	ret = amdgpu_ras_check_bad_page(adev, address);
+	if (ret == -EINVAL) {
+		dev_warn(adev->dev,
+			"RAS WARN: input address 0x%llx is invalid.\n",
+			address);
+		return -EINVAL;
+	} else if (ret == 1) {
+		dev_warn(adev->dev,
+			"RAS WARN: 0x%llx has already been marked as bad page!\n",
+			address);
+		return 0;
+	}
+
+	ret = amdgpu_ras_error_data_init(&err_data);
+	if (ret)
+		return ret;
+
+	memset(&err_rec, 0x0, sizeof(struct eeprom_table_record));
+	err_data.err_addr = &err_rec;
+	amdgpu_umc_fill_error_record(&err_data, address, address, 0, 0);
+
+	if (amdgpu_bad_page_threshold != 0) {
+		amdgpu_ras_add_bad_pages(adev, err_data.err_addr,
+					 err_data.err_addr_cnt, false);
+		amdgpu_ras_save_bad_pages(adev, NULL);
+	}
+
+	amdgpu_ras_error_data_fini(&err_data);
+
+	dev_warn(adev->dev, "WARNING: THIS IS ONLY FOR TEST PURPOSES AND WILL CORRUPT RAS EEPROM\n");
+	dev_warn(adev->dev, "Clear EEPROM:\n");
+	dev_warn(adev->dev, "    echo 1 > /sys/kernel/debug/dri/0/ras/ras_eeprom_reset\n");
+
+	return 0;
+}
+
+static int amdgpu_check_address_validity(struct amdgpu_device *adev,
+			uint64_t address, uint64_t flags)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct amdgpu_vram_block_info blk_info;
+	uint64_t page_pfns[32] = {0};
+	int i, ret, count;
+	bool hit = false;
+
+	if (amdgpu_ip_version(adev, UMC_HWIP, 0) < IP_VERSION(12, 0, 0))
+		return 0;
+
+	if (amdgpu_sriov_vf(adev)) {
+		if (amdgpu_virt_check_vf_critical_region(adev, address, &hit))
+			return -EPERM;
+		return hit ? -EACCES : 0;
+	}
+
+	if ((address >= adev->gmc.mc_vram_size) ||
+	    (address >= RAS_UMC_INJECT_ADDR_LIMIT))
+		return -EFAULT;
+
+	count = amdgpu_umc_lookup_bad_pages_in_a_row(adev,
+				address, page_pfns, ARRAY_SIZE(page_pfns));
+	if (count <= 0)
+		return -EPERM;
+
+	for (i = 0; i < count; i++) {
+		memset(&blk_info, 0, sizeof(blk_info));
+		ret = amdgpu_vram_mgr_query_address_block_info(&adev->mman.vram_mgr,
+					page_pfns[i] << AMDGPU_GPU_PAGE_SHIFT, &blk_info);
+		if (!ret) {
+			/* The input address that needs to be checked is allocated by
+			 * current calling process, so it is necessary to exclude
+			 * the calling process.
+			 */
+			if ((flags == BYPASS_ALLOCATED_ADDRESS) &&
+			    ((blk_info.task.pid != task_pid_nr(current)) ||
+				strncmp(blk_info.task.comm, current->comm, TASK_COMM_LEN)))
+				return -EACCES;
+			else if ((flags == BYPASS_INITIALIZATION_ADDRESS) &&
+				(blk_info.task.pid == con->init_task_pid) &&
+				!strncmp(blk_info.task.comm, con->init_task_comm, TASK_COMM_LEN))
+				return -EACCES;
+		}
+	}
+
+	return 0;
+}
+
+static ssize_t amdgpu_ras_debugfs_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct ras_manager *obj = (struct ras_manager *)file_inode(f)->i_private;
+	struct ras_query_if info = {
+		.head = obj->head,
+	};
+	ssize_t s;
+	char val[128];
+
+	if (amdgpu_ras_query_error_status(obj->adev, &info))
+		return -EINVAL;
+
+	/* Hardware counter will be reset automatically after the query on Vega20 and Arcturus */
+	if (amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 2) &&
+	    amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 4)) {
+		if (amdgpu_ras_reset_error_status(obj->adev, info.head.block))
+			dev_warn(obj->adev->dev, "Failed to reset error counter and error status");
+	}
+
+	s = snprintf(val, sizeof(val), "%s: %lu\n%s: %lu\n",
+			"ue", info.ue_count,
+			"ce", info.ce_count);
+	if (*pos >= s)
+		return 0;
+
+	s -= *pos;
+	s = min_t(u64, s, size);
+
+
+	if (copy_to_user(buf, &val[*pos], s))
+		return -EINVAL;
+
+	*pos += s;
+
+	return s;
+}
+
+static const struct file_operations amdgpu_ras_debugfs_ops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_ras_debugfs_read,
+	.write = NULL,
+	.llseek = default_llseek
+};
+
+static int amdgpu_ras_find_block_id_by_name(const char *name, int *block_id)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(ras_block_string); i++) {
+		*block_id = i;
+		if (strcmp(name, ras_block_string[i]) == 0)
+			return 0;
+	}
+	return -EINVAL;
+}
+
+static int amdgpu_ras_debugfs_ctrl_parse_data(struct file *f,
+		const char __user *buf, size_t size,
+		loff_t *pos, struct ras_debug_if *data)
+{
+	ssize_t s = min_t(u64, 64, size);
+	char str[65];
+	char block_name[33];
+	char err[9] = "ue";
+	int op = -1;
+	int block_id;
+	uint32_t sub_block;
+	u64 address, value;
+	/* default value is 0 if the mask is not set by user */
+	u32 instance_mask = 0;
+
+	if (*pos)
+		return -EINVAL;
+	*pos = size;
+
+	memset(str, 0, sizeof(str));
+	memset(data, 0, sizeof(*data));
+
+	if (copy_from_user(str, buf, s))
+		return -EINVAL;
+
+	if (sscanf(str, "disable %32s", block_name) == 1)
+		op = 0;
+	else if (sscanf(str, "enable %32s %8s", block_name, err) == 2)
+		op = 1;
+	else if (sscanf(str, "inject %32s %8s", block_name, err) == 2)
+		op = 2;
+	else if (strstr(str, "retire_page") != NULL)
+		op = 3;
+	else if (strstr(str, "check_address") != NULL)
+		op = 4;
+	else if (str[0] && str[1] && str[2] && str[3])
+		/* ascii string, but commands are not matched. */
+		return -EINVAL;
+
+	if (op != -1) {
+		if (op == 3) {
+			if (sscanf(str, "%*s 0x%llx", &address) != 1 &&
+			    sscanf(str, "%*s %llu", &address) != 1)
+				return -EINVAL;
+
+			data->op = op;
+			data->inject.address = address;
+
+			return 0;
+		} else if (op == 4) {
+			if (sscanf(str, "%*s 0x%llx 0x%llx", &address, &value) != 2 &&
+			    sscanf(str, "%*s %llu %llu", &address, &value) != 2)
+				return -EINVAL;
+
+			data->op = op;
+			data->inject.address = address;
+			data->inject.value = value;
+			return 0;
+		}
+
+		if (amdgpu_ras_find_block_id_by_name(block_name, &block_id))
+			return -EINVAL;
+
+		data->head.block = block_id;
+		/* only ue, ce and poison errors are supported */
+		if (!memcmp("ue", err, 2))
+			data->head.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+		else if (!memcmp("ce", err, 2))
+			data->head.type = AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE;
+		else if (!memcmp("poison", err, 6))
+			data->head.type = AMDGPU_RAS_ERROR__POISON;
+		else
+			return -EINVAL;
+
+		data->op = op;
+
+		if (op == 2) {
+			if (sscanf(str, "%*s %*s %*s 0x%x 0x%llx 0x%llx 0x%x",
+				   &sub_block, &address, &value, &instance_mask) != 4 &&
+			    sscanf(str, "%*s %*s %*s %u %llu %llu %u",
+				   &sub_block, &address, &value, &instance_mask) != 4 &&
+				sscanf(str, "%*s %*s %*s 0x%x 0x%llx 0x%llx",
+				   &sub_block, &address, &value) != 3 &&
+			    sscanf(str, "%*s %*s %*s %u %llu %llu",
+				   &sub_block, &address, &value) != 3)
+				return -EINVAL;
+			data->head.sub_block_index = sub_block;
+			data->inject.address = address;
+			data->inject.value = value;
+			data->inject.instance_mask = instance_mask;
+		}
+	} else {
+		if (size < sizeof(*data))
+			return -EINVAL;
+
+		if (copy_from_user(data, buf, sizeof(*data)))
+			return -EINVAL;
+	}
+
+	return 0;
+}
+
+static void amdgpu_ras_instance_mask_check(struct amdgpu_device *adev,
+				struct ras_debug_if *data)
+{
+	int num_xcc = adev->gfx.xcc_mask ? NUM_XCC(adev->gfx.xcc_mask) : 1;
+	uint32_t mask, inst_mask = data->inject.instance_mask;
+
+	/* no need to set instance mask if there is only one instance */
+	if (num_xcc <= 1 && inst_mask) {
+		data->inject.instance_mask = 0;
+		dev_dbg(adev->dev,
+			"RAS inject mask(0x%x) isn't supported and force it to 0.\n",
+			inst_mask);
+
+		return;
+	}
+
+	switch (data->head.block) {
+	case AMDGPU_RAS_BLOCK__GFX:
+		mask = GENMASK(num_xcc - 1, 0);
+		break;
+	case AMDGPU_RAS_BLOCK__SDMA:
+		mask = GENMASK(adev->sdma.num_instances - 1, 0);
+		break;
+	case AMDGPU_RAS_BLOCK__VCN:
+	case AMDGPU_RAS_BLOCK__JPEG:
+		mask = GENMASK(adev->vcn.num_vcn_inst - 1, 0);
+		break;
+	default:
+		mask = inst_mask;
+		break;
+	}
+
+	/* remove invalid bits in instance mask */
+	data->inject.instance_mask &= mask;
+	if (inst_mask != data->inject.instance_mask)
+		dev_dbg(adev->dev,
+			"Adjust RAS inject mask 0x%x to 0x%x\n",
+			inst_mask, data->inject.instance_mask);
+}
+
+/**
+ * DOC: AMDGPU RAS debugfs control interface
+ *
+ * The control interface accepts struct ras_debug_if which has two members.
+ *
+ * First member: ras_debug_if::head or ras_debug_if::inject.
+ *
+ * head is used to indicate which IP block will be under control.
+ *
+ * head has four members, they are block, type, sub_block_index, name.
+ * block: which IP will be under control.
+ * type: what kind of error will be enabled/disabled/injected.
+ * sub_block_index: some IPs have subcomponets. say, GFX, sDMA.
+ * name: the name of IP.
+ *
+ * inject has three more members than head, they are address, value and mask.
+ * As their names indicate, inject operation will write the
+ * value to the address.
+ *
+ * The second member: struct ras_debug_if::op.
+ * It has three kinds of operations.
+ *
+ * - 0: disable RAS on the block. Take ::head as its data.
+ * - 1: enable RAS on the block. Take ::head as its data.
+ * - 2: inject errors on the block. Take ::inject as its data.
+ *
+ * How to use the interface?
+ *
+ * In a program
+ *
+ * Copy the struct ras_debug_if in your code and initialize it.
+ * Write the struct to the control interface.
+ *
+ * From shell
+ *
+ * .. code-block:: bash
+ *
+ *	echo "disable <block>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl
+ *	echo "enable  <block> <error>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl
+ *	echo "inject  <block> <error> <sub-block> <address> <value> <mask>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl
+ *
+ * Where N, is the card which you want to affect.
+ *
+ * "disable" requires only the block.
+ * "enable" requires the block and error type.
+ * "inject" requires the block, error type, address, and value.
+ *
+ * The block is one of: umc, sdma, gfx, etc.
+ *	see ras_block_string[] for details
+ *
+ * The error type is one of: ue, ce and poison where,
+ *	ue is multi-uncorrectable
+ *	ce is single-correctable
+ *	poison is poison
+ *
+ * The sub-block is a the sub-block index, pass 0 if there is no sub-block.
+ * The address and value are hexadecimal numbers, leading 0x is optional.
+ * The mask means instance mask, is optional, default value is 0x1.
+ *
+ * For instance,
+ *
+ * .. code-block:: bash
+ *
+ *	echo inject umc ue 0x0 0x0 0x0 > /sys/kernel/debug/dri/0/ras/ras_ctrl
+ *	echo inject umc ce 0 0 0 3 > /sys/kernel/debug/dri/0/ras/ras_ctrl
+ *	echo disable umc > /sys/kernel/debug/dri/0/ras/ras_ctrl
+ *
+ * How to check the result of the operation?
+ *
+ * To check disable/enable, see "ras" features at,
+ * /sys/class/drm/card[0/1/2...]/device/ras/features
+ *
+ * To check inject, see the corresponding error count at,
+ * /sys/class/drm/card[0/1/2...]/device/ras/[gfx|sdma|umc|...]_err_count
+ *
+ * .. note::
+ *	Operations are only allowed on blocks which are supported.
+ *	Check the "ras" mask at /sys/module/amdgpu/parameters/ras_mask
+ *	to see which blocks support RAS on a particular asic.
+ *
+ */
+static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f,
+					     const char __user *buf,
+					     size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
+	struct ras_debug_if data;
+	int ret = 0;
+
+	if (!amdgpu_ras_get_error_query_ready(adev)) {
+		dev_warn(adev->dev, "RAS WARN: error injection "
+				"currently inaccessible\n");
+		return size;
+	}
+
+	ret = amdgpu_ras_debugfs_ctrl_parse_data(f, buf, size, pos, &data);
+	if (ret)
+		return ret;
+
+	if (data.op == 3) {
+		ret = amdgpu_reserve_page_direct(adev, data.inject.address);
+		if (!ret)
+			return size;
+		else
+			return ret;
+	} else if (data.op == 4) {
+		ret = amdgpu_check_address_validity(adev, data.inject.address, data.inject.value);
+		return ret ? ret : size;
+	}
+
+	if (!amdgpu_ras_is_supported(adev, data.head.block))
+		return -EINVAL;
+
+	switch (data.op) {
+	case 0:
+		ret = amdgpu_ras_feature_enable(adev, &data.head, 0);
+		break;
+	case 1:
+		ret = amdgpu_ras_feature_enable(adev, &data.head, 1);
+		break;
+	case 2:
+		/* umc ce/ue error injection for a bad page is not allowed */
+		if (data.head.block == AMDGPU_RAS_BLOCK__UMC)
+			ret = amdgpu_ras_check_bad_page(adev, data.inject.address);
+		if (ret == -EINVAL) {
+			dev_warn(adev->dev, "RAS WARN: input address 0x%llx is invalid.",
+					data.inject.address);
+			break;
+		} else if (ret == 1) {
+			dev_warn(adev->dev, "RAS WARN: inject: 0x%llx has already been marked as bad!\n",
+					data.inject.address);
+			break;
+		}
+
+		amdgpu_ras_instance_mask_check(adev, &data);
+
+		/* data.inject.address is offset instead of absolute gpu address */
+		ret = amdgpu_ras_error_inject(adev, &data.inject);
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+	if (ret)
+		return ret;
+
+	return size;
+}
+
+static int amdgpu_uniras_clear_badpages_info(struct amdgpu_device *adev);
+
+/**
+ * DOC: AMDGPU RAS debugfs EEPROM table reset interface
+ *
+ * Some boards contain an EEPROM which is used to persistently store a list of
+ * bad pages which experiences ECC errors in vram.  This interface provides
+ * a way to reset the EEPROM, e.g., after testing error injection.
+ *
+ * Usage:
+ *
+ * .. code-block:: bash
+ *
+ *	echo 1 > ../ras/ras_eeprom_reset
+ *
+ * will reset EEPROM table to 0 entries.
+ *
+ */
+static ssize_t amdgpu_ras_debugfs_eeprom_write(struct file *f,
+					       const char __user *buf,
+					       size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev =
+		(struct amdgpu_device *)file_inode(f)->i_private;
+	int ret;
+
+	if (amdgpu_uniras_enabled(adev)) {
+		ret = amdgpu_uniras_clear_badpages_info(adev);
+		return ret ? ret : size;
+	}
+
+	ret = amdgpu_ras_eeprom_reset_table(
+		&(amdgpu_ras_get_context(adev)->eeprom_control));
+
+	if (!ret) {
+		/* Something was written to EEPROM.
+		 */
+		amdgpu_ras_get_context(adev)->flags = RAS_DEFAULT_FLAGS;
+		return size;
+	} else {
+		return ret;
+	}
+}
+
+static const struct file_operations amdgpu_ras_debugfs_ctrl_ops = {
+	.owner = THIS_MODULE,
+	.read = NULL,
+	.write = amdgpu_ras_debugfs_ctrl_write,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_ras_debugfs_eeprom_ops = {
+	.owner = THIS_MODULE,
+	.read = NULL,
+	.write = amdgpu_ras_debugfs_eeprom_write,
+	.llseek = default_llseek
+};
+
+/**
+ * DOC: AMDGPU RAS sysfs Error Count Interface
+ *
+ * It allows the user to read the error count for each IP block on the gpu through
+ * /sys/class/drm/card[0/1/2...]/device/ras/[gfx/sdma/...]_err_count
+ *
+ * It outputs the multiple lines which report the uncorrected (ue) and corrected
+ * (ce) error counts.
+ *
+ * The format of one line is below,
+ *
+ * [ce|ue]: count
+ *
+ * Example:
+ *
+ * .. code-block:: bash
+ *
+ *	ue: 0
+ *	ce: 1
+ *
+ */
+static ssize_t amdgpu_ras_sysfs_read(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct ras_manager *obj = container_of(attr, struct ras_manager, sysfs_attr);
+	struct ras_query_if info = {
+		.head = obj->head,
+	};
+
+	if (!amdgpu_ras_get_error_query_ready(obj->adev))
+		return sysfs_emit(buf, "Query currently inaccessible\n");
+
+	if (amdgpu_ras_query_error_status(obj->adev, &info))
+		return -EINVAL;
+
+	if (amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 2) &&
+	    amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 4)) {
+		if (amdgpu_ras_reset_error_status(obj->adev, info.head.block))
+			dev_warn(obj->adev->dev, "Failed to reset error counter and error status");
+	}
+
+	if (info.head.block == AMDGPU_RAS_BLOCK__UMC)
+		return sysfs_emit(buf, "%s: %lu\n%s: %lu\n%s: %lu\n", "ue", info.ue_count,
+				"ce", info.ce_count, "de", info.de_count);
+	else
+		return sysfs_emit(buf, "%s: %lu\n%s: %lu\n", "ue", info.ue_count,
+				"ce", info.ce_count);
+}
+
+/* obj begin */
+
+#define get_obj(obj) do { (obj)->use++; } while (0)
+#define alive_obj(obj) ((obj)->use)
+
+static inline void put_obj(struct ras_manager *obj)
+{
+	if (obj && (--obj->use == 0)) {
+		list_del(&obj->node);
+		amdgpu_ras_error_data_fini(&obj->err_data);
+	}
+
+	if (obj && (obj->use < 0))
+		DRM_ERROR("RAS ERROR: Unbalance obj(%s) use\n", get_ras_block_str(&obj->head));
+}
+
+/* make one obj and return it. */
+static struct ras_manager *amdgpu_ras_create_obj(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj;
+
+	if (!adev->ras_enabled || !con)
+		return NULL;
+
+	if (head->block >= AMDGPU_RAS_BLOCK_COUNT)
+		return NULL;
+
+	if (head->block == AMDGPU_RAS_BLOCK__MCA) {
+		if (head->sub_block_index >= AMDGPU_RAS_MCA_BLOCK__LAST)
+			return NULL;
+
+		obj = &con->objs[AMDGPU_RAS_BLOCK__LAST + head->sub_block_index];
+	} else
+		obj = &con->objs[head->block];
+
+	/* already exist. return obj? */
+	if (alive_obj(obj))
+		return NULL;
+
+	if (amdgpu_ras_error_data_init(&obj->err_data))
+		return NULL;
+
+	obj->head = *head;
+	obj->adev = adev;
+	list_add(&obj->node, &con->head);
+	get_obj(obj);
+
+	return obj;
+}
+
+/* return an obj equal to head, or the first when head is NULL */
+struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj;
+	int i;
+
+	if (!adev->ras_enabled || !con)
+		return NULL;
+
+	if (head) {
+		if (head->block >= AMDGPU_RAS_BLOCK_COUNT)
+			return NULL;
+
+		if (head->block == AMDGPU_RAS_BLOCK__MCA) {
+			if (head->sub_block_index >= AMDGPU_RAS_MCA_BLOCK__LAST)
+				return NULL;
+
+			obj = &con->objs[AMDGPU_RAS_BLOCK__LAST + head->sub_block_index];
+		} else
+			obj = &con->objs[head->block];
+
+		if (alive_obj(obj))
+			return obj;
+	} else {
+		for (i = 0; i < AMDGPU_RAS_BLOCK_COUNT + AMDGPU_RAS_MCA_BLOCK_COUNT; i++) {
+			obj = &con->objs[i];
+			if (alive_obj(obj))
+				return obj;
+		}
+	}
+
+	return NULL;
+}
+/* obj end */
+
+/* feature ctl begin */
+static int amdgpu_ras_is_feature_allowed(struct amdgpu_device *adev,
+					 struct ras_common_if *head)
+{
+	return adev->ras_hw_enabled & BIT(head->block);
+}
+
+static int amdgpu_ras_is_feature_enabled(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+
+	return con->features & BIT(head->block);
+}
+
+/*
+ * if obj is not created, then create one.
+ * set feature enable flag.
+ */
+static int __amdgpu_ras_feature_enable(struct amdgpu_device *adev,
+		struct ras_common_if *head, int enable)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, head);
+
+	/* If hardware does not support ras, then do not create obj.
+	 * But if hardware support ras, we can create the obj.
+	 * Ras framework checks con->hw_supported to see if it need do
+	 * corresponding initialization.
+	 * IP checks con->support to see if it need disable ras.
+	 */
+	if (!amdgpu_ras_is_feature_allowed(adev, head))
+		return 0;
+
+	if (enable) {
+		if (!obj) {
+			obj = amdgpu_ras_create_obj(adev, head);
+			if (!obj)
+				return -EINVAL;
+		} else {
+			/* In case we create obj somewhere else */
+			get_obj(obj);
+		}
+		con->features |= BIT(head->block);
+	} else {
+		if (obj && amdgpu_ras_is_feature_enabled(adev, head)) {
+			con->features &= ~BIT(head->block);
+			put_obj(obj);
+		}
+	}
+
+	return 0;
+}
+
+/* wrapper of psp_ras_enable_features */
+int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
+		struct ras_common_if *head, bool enable)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	union ta_ras_cmd_input *info;
+	int ret;
+
+	if (!con)
+		return -EINVAL;
+
+	/* For non-gfx ip, do not enable ras feature if it is not allowed */
+	/* For gfx ip, regardless of feature support status, */
+	/* Force issue enable or disable ras feature commands */
+	if (head->block != AMDGPU_RAS_BLOCK__GFX &&
+	    !amdgpu_ras_is_feature_allowed(adev, head))
+		return 0;
+
+	/* Only enable gfx ras feature from host side */
+	if (head->block == AMDGPU_RAS_BLOCK__GFX &&
+	    !amdgpu_sriov_vf(adev) &&
+	    !amdgpu_ras_intr_triggered()) {
+		info = kzalloc(sizeof(union ta_ras_cmd_input), GFP_KERNEL);
+		if (!info)
+			return -ENOMEM;
+
+		if (!enable) {
+			info->disable_features = (struct ta_ras_disable_features_input) {
+				.block_id =  amdgpu_ras_block_to_ta(head->block),
+				.error_type = amdgpu_ras_error_to_ta(head->type),
+			};
+		} else {
+			info->enable_features = (struct ta_ras_enable_features_input) {
+				.block_id =  amdgpu_ras_block_to_ta(head->block),
+				.error_type = amdgpu_ras_error_to_ta(head->type),