diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c | 1650 |
1 files changed, 1335 insertions, 315 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c index 21e7b88401e1..a316797875a8 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c @@ -22,154 +22,247 @@ */ #include <linux/firmware.h> -#include <drm/drmP.h> + #include "amdgpu.h" +#include "amdgpu_cs.h" #include "amdgpu_vcn.h" +#include "amdgpu_pm.h" +#include "soc15.h" #include "soc15d.h" #include "soc15_common.h" -#include "vega10/soc15ip.h" -#include "raven1/VCN/vcn_1_0_offset.h" -#include "raven1/VCN/vcn_1_0_sh_mask.h" -#include "vega10/HDP/hdp_4_0_offset.h" -#include "raven1/MMHUB/mmhub_9_1_offset.h" -#include "raven1/MMHUB/mmhub_9_1_sh_mask.h" +#include "vcn/vcn_1_0_offset.h" +#include "vcn/vcn_1_0_sh_mask.h" +#include "mmhub/mmhub_9_1_offset.h" +#include "mmhub/mmhub_9_1_sh_mask.h" + +#include "ivsrcid/vcn/irqsrcs_vcn_1_0.h" +#include "jpeg_v1_0.h" +#include "vcn_v1_0.h" + +#define mmUVD_RBC_XX_IB_REG_CHECK_1_0 0x05ab +#define mmUVD_RBC_XX_IB_REG_CHECK_1_0_BASE_IDX 1 +#define mmUVD_REG_XX_MASK_1_0 0x05ac +#define mmUVD_REG_XX_MASK_1_0_BASE_IDX 1 + +static const struct amdgpu_hwip_reg_entry vcn_reg_list_1_0[] = { + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_POWER_STATUS), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_STATUS), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID2), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA0), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA1), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_CMD), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI2), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO2), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI3), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO3), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI4), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO4), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR2), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR2), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR3), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR3), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR4), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR4), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE2), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE3), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE4), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_CONFIG), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_STATUS), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_CTL), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_DATA), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_MASK), + SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_PAUSE) +}; -static int vcn_v1_0_start(struct amdgpu_device *adev); -static int vcn_v1_0_stop(struct amdgpu_device *adev); +static int vcn_v1_0_stop(struct amdgpu_vcn_inst *vinst); static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev); static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev); static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev); +static int vcn_v1_0_set_pg_state(struct amdgpu_vcn_inst *vinst, + enum amd_powergating_state state); +static int vcn_v1_0_pause_dpg_mode(struct amdgpu_vcn_inst *vinst, + struct dpg_pause_state *new_state); + +static void vcn_v1_0_idle_work_handler(struct work_struct *work); +static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring); /** - * vcn_v1_0_early_init - set function pointers + * vcn_v1_0_early_init - set function pointers and load microcode * - * @handle: amdgpu_device pointer + * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. * * Set ring and irq function pointers + * Load microcode from filesystem */ -static int vcn_v1_0_early_init(void *handle) +static int vcn_v1_0_early_init(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; - adev->vcn.num_enc_rings = 2; + adev->vcn.inst[0].num_enc_rings = 2; + adev->vcn.inst[0].set_pg_state = vcn_v1_0_set_pg_state; vcn_v1_0_set_dec_ring_funcs(adev); vcn_v1_0_set_enc_ring_funcs(adev); vcn_v1_0_set_irq_funcs(adev); - return 0; + jpeg_v1_0_early_init(ip_block); + + return amdgpu_vcn_early_init(adev, 0); } /** * vcn_v1_0_sw_init - sw init for VCN block * - * @handle: amdgpu_device pointer + * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. * * Load firmware and sw initialization */ -static int vcn_v1_0_sw_init(void *handle) +static int vcn_v1_0_sw_init(struct amdgpu_ip_block *ip_block) { struct amdgpu_ring *ring; int i, r; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_1_0); + uint32_t *ptr; + struct amdgpu_device *adev = ip_block->adev; /* VCN DEC TRAP */ - r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VCN, 124, &adev->vcn.irq); + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, + VCN_1_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst->irq); if (r) return r; /* VCN ENC TRAP */ - for (i = 0; i < adev->vcn.num_enc_rings; ++i) { - r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VCN, i + 119, - &adev->vcn.irq); + for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) { + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN, i + VCN_1_0__SRCID__UVD_ENC_GENERAL_PURPOSE, + &adev->vcn.inst->irq); if (r) return r; } - r = amdgpu_vcn_sw_init(adev); + r = amdgpu_vcn_sw_init(adev, 0); if (r) return r; - r = amdgpu_vcn_resume(adev); + /* Override the work func */ + adev->vcn.inst[0].idle_work.work.func = vcn_v1_0_idle_work_handler; + + amdgpu_vcn_setup_ucode(adev, 0); + + r = amdgpu_vcn_resume(adev, 0); if (r) return r; - ring = &adev->vcn.ring_dec; + ring = &adev->vcn.inst->ring_dec; + ring->vm_hub = AMDGPU_MMHUB0(0); sprintf(ring->name, "vcn_dec"); - r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.irq, 0); + r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, + AMDGPU_RING_PRIO_DEFAULT, NULL); if (r) return r; - for (i = 0; i < adev->vcn.num_enc_rings; ++i) { - ring = &adev->vcn.ring_enc[i]; + adev->vcn.inst[0].internal.scratch9 = adev->vcn.inst->external.scratch9 = + SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9); + adev->vcn.inst[0].internal.data0 = adev->vcn.inst->external.data0 = + SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0); + adev->vcn.inst[0].internal.data1 = adev->vcn.inst->external.data1 = + SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1); + adev->vcn.inst[0].internal.cmd = adev->vcn.inst->external.cmd = + SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD); + adev->vcn.inst[0].internal.nop = adev->vcn.inst->external.nop = + SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP); + + for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) { + enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i); + + ring = &adev->vcn.inst->ring_enc[i]; + ring->vm_hub = AMDGPU_MMHUB0(0); sprintf(ring->name, "vcn_enc%d", i); - r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.irq, 0); + r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0, + hw_prio, NULL); if (r) return r; } + adev->vcn.inst[0].pause_dpg_mode = vcn_v1_0_pause_dpg_mode; + + if (amdgpu_vcnfw_log) { + struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared.cpu_addr; + + fw_shared->present_flag_0 = 0; + amdgpu_vcn_fwlog_init(adev->vcn.inst); + } + + r = jpeg_v1_0_sw_init(ip_block); + + /* Allocate memory for VCN IP Dump buffer */ + ptr = kcalloc(adev->vcn.num_vcn_inst * reg_count, sizeof(uint32_t), GFP_KERNEL); + if (!ptr) { + DRM_ERROR("Failed to allocate memory for VCN IP Dump\n"); + adev->vcn.ip_dump = NULL; + } else { + adev->vcn.ip_dump = ptr; + } return r; } /** * vcn_v1_0_sw_fini - sw fini for VCN block * - * @handle: amdgpu_device pointer + * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. * * VCN suspend and free up sw allocation */ -static int vcn_v1_0_sw_fini(void *handle) +static int vcn_v1_0_sw_fini(struct amdgpu_ip_block *ip_block) { int r; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; - r = amdgpu_vcn_suspend(adev); + r = amdgpu_vcn_suspend(adev, 0); if (r) return r; - r = amdgpu_vcn_sw_fini(adev); + jpeg_v1_0_sw_fini(ip_block); - return r; + amdgpu_vcn_sw_fini(adev, 0); + + kfree(adev->vcn.ip_dump); + + return 0; } /** * vcn_v1_0_hw_init - start and test VCN block * - * @handle: amdgpu_device pointer + * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. * * Initialize the hardware, boot up the VCPU and do some testing */ -static int vcn_v1_0_hw_init(void *handle) +static int vcn_v1_0_hw_init(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct amdgpu_ring *ring = &adev->vcn.ring_dec; + struct amdgpu_device *adev = ip_block->adev; + struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; int i, r; - r = vcn_v1_0_start(adev); + r = amdgpu_ring_test_helper(ring); if (r) - goto done; - - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; - goto done; - } + return r; - for (i = 0; i < adev->vcn.num_enc_rings; ++i) { - ring = &adev->vcn.ring_enc[i]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; - goto done; - } + for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) { + ring = &adev->vcn.inst->ring_enc[i]; + r = amdgpu_ring_test_helper(ring); + if (r) + return r; } -done: - if (!r) - DRM_INFO("VCN decode and encode initialized successfully.\n"); + ring = adev->jpeg.inst->ring_dec; + r = amdgpu_ring_test_helper(ring); return r; } @@ -177,21 +270,22 @@ done: /** * vcn_v1_0_hw_fini - stop the hardware block * - * @handle: amdgpu_device pointer + * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. * * Stop the VCN block, mark ring as not ready any more */ -static int vcn_v1_0_hw_fini(void *handle) +static int vcn_v1_0_hw_fini(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - struct amdgpu_ring *ring = &adev->vcn.ring_dec; - int r; + struct amdgpu_device *adev = ip_block->adev; + struct amdgpu_vcn_inst *vinst = adev->vcn.inst; - r = vcn_v1_0_stop(adev); - if (r) - return r; + cancel_delayed_work_sync(&vinst->idle_work); - ring->ready = false; + if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) || + (vinst->cur_state != AMD_PG_STATE_GATE && + RREG32_SOC15(VCN, 0, mmUVD_STATUS))) { + vinst->set_pg_state(vinst, AMD_PG_STATE_GATE); + } return 0; } @@ -199,20 +293,27 @@ static int vcn_v1_0_hw_fini(void *handle) /** * vcn_v1_0_suspend - suspend VCN block * - * @handle: amdgpu_device pointer + * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. * * HW fini and suspend VCN block */ -static int vcn_v1_0_suspend(void *handle) +static int vcn_v1_0_suspend(struct amdgpu_ip_block *ip_block) { int r; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; + bool idle_work_unexecuted; + + idle_work_unexecuted = cancel_delayed_work_sync(&adev->vcn.inst[0].idle_work); + if (idle_work_unexecuted) { + if (adev->pm.dpm_enabled) + amdgpu_dpm_enable_vcn(adev, false, 0); + } - r = vcn_v1_0_hw_fini(adev); + r = vcn_v1_0_hw_fini(ip_block); if (r) return r; - r = amdgpu_vcn_suspend(adev); + r = amdgpu_vcn_suspend(adev, 0); return r; } @@ -220,57 +321,71 @@ static int vcn_v1_0_suspend(void *handle) /** * vcn_v1_0_resume - resume VCN block * - * @handle: amdgpu_device pointer + * @ip_block: Pointer to the amdgpu_ip_block for this hw instance. * * Resume firmware and hw init VCN block */ -static int vcn_v1_0_resume(void *handle) +static int vcn_v1_0_resume(struct amdgpu_ip_block *ip_block) { int r; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - r = amdgpu_vcn_resume(adev); + r = amdgpu_vcn_resume(ip_block->adev, 0); if (r) return r; - r = vcn_v1_0_hw_init(adev); + r = vcn_v1_0_hw_init(ip_block); return r; } /** - * vcn_v1_0_mc_resume - memory controller programming + * vcn_v1_0_mc_resume_spg_mode - memory controller programming * - * @adev: amdgpu_device pointer + * @vinst: VCN instance * * Let the VCN memory controller know it's offsets */ -static void vcn_v1_0_mc_resume(struct amdgpu_device *adev) +static void vcn_v1_0_mc_resume_spg_mode(struct amdgpu_vcn_inst *vinst) { - uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4); - - WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, - lower_32_bits(adev->vcn.gpu_addr)); - WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, - upper_32_bits(adev->vcn.gpu_addr)); - WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, - AMDGPU_UVD_FIRMWARE_OFFSET >> 3); + struct amdgpu_device *adev = vinst->adev; + uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.inst[0].fw->size + 4); + uint32_t offset; + + /* cache window 0: fw */ + if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { + WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo)); + WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi)); + WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0); + offset = 0; + } else { + WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, + lower_32_bits(adev->vcn.inst->gpu_addr)); + WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, + upper_32_bits(adev->vcn.inst->gpu_addr)); + offset = size; + WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, + AMDGPU_UVD_FIRMWARE_OFFSET >> 3); + } + WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size); + /* cache window 1: stack */ WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, - lower_32_bits(adev->vcn.gpu_addr + size)); + lower_32_bits(adev->vcn.inst->gpu_addr + offset)); WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, - upper_32_bits(adev->vcn.gpu_addr + size)); + upper_32_bits(adev->vcn.inst->gpu_addr + offset)); WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0); - WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_HEAP_SIZE); + WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE); + /* cache window 2: context */ WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, - lower_32_bits(adev->vcn.gpu_addr + size + AMDGPU_VCN_HEAP_SIZE)); + lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, - upper_32_bits(adev->vcn.gpu_addr + size + AMDGPU_VCN_HEAP_SIZE)); + upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE)); WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0); - WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, - AMDGPU_VCN_STACK_SIZE + (AMDGPU_VCN_SESSION_SIZE * 40)); + WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE); WREG32_SOC15(UVD, 0, mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config); @@ -278,24 +393,115 @@ static void vcn_v1_0_mc_resume(struct amdgpu_device *adev) adev->gfx.config.gb_addr_config); WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_MIF_CURR_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_MIF_CURR_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_MIF_RECON1_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_MIF_RECON1_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_MIF_REF_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_MIF_REF_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_JPEG_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); + WREG32_SOC15(UVD, 0, mmUVD_JPEG_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config); +} + +static void vcn_v1_0_mc_resume_dpg_mode(struct amdgpu_vcn_inst *vinst) +{ + struct amdgpu_device *adev = vinst->adev; + uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.inst[0].fw->size + 4); + uint32_t offset; + + /* cache window 0: fw */ + if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) { + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), + 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, + (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), + 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0, + 0xFFFFFFFF, 0); + offset = 0; + } else { + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, + lower_32_bits(adev->vcn.inst->gpu_addr), 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, + upper_32_bits(adev->vcn.inst->gpu_addr), 0xFFFFFFFF, 0); + offset = size; + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, + AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0xFFFFFFFF, 0); + } + + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size, 0xFFFFFFFF, 0); + + /* cache window 1: stack */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW, + lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH, + upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0, + 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE, + 0xFFFFFFFF, 0); + + /* cache window 2: context */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW, + lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), + 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH, + upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), + 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE, + 0xFFFFFFFF, 0); + + /* VCN global tiling registers */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DB_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_UDEC_DBW_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_CURR_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_CURR_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_RECON1_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_RECON1_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_REF_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MIF_REF_UV_ADDR_CONFIG, + adev->gfx.config.gb_addr_config, 0xFFFFFFFF, 0); } /** * vcn_v1_0_disable_clock_gating - disable VCN clock gating * - * @adev: amdgpu_device pointer - * @sw: enable SW clock gating + * @vinst: VCN instance * * Disable clock gating for VCN block */ -static void vcn_v1_0_disable_clock_gating(struct amdgpu_device *adev, bool sw) +static void vcn_v1_0_disable_clock_gating(struct amdgpu_vcn_inst *vinst) { + struct amdgpu_device *adev = vinst->adev; uint32_t data; /* JPEG disable CGC */ data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL); - if (sw) + if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; else data &= ~JPEG_CGC_CTRL__DYN_CLOCK_MODE_MASK; @@ -310,10 +516,10 @@ static void vcn_v1_0_disable_clock_gating(struct amdgpu_device *adev, bool sw) /* UVD disable CGC */ data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); - if (sw) + if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; else - data &= ~ UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; + data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; @@ -410,18 +616,18 @@ static void vcn_v1_0_disable_clock_gating(struct amdgpu_device *adev, bool sw) /** * vcn_v1_0_enable_clock_gating - enable VCN clock gating * - * @adev: amdgpu_device pointer - * @sw: enable SW clock gating + * @vinst: Pointer to the VCN instance structure * * Enable clock gating for VCN block */ -static void vcn_v1_0_enable_clock_gating(struct amdgpu_device *adev, bool sw) +static void vcn_v1_0_enable_clock_gating(struct amdgpu_vcn_inst *vinst) { + struct amdgpu_device *adev = vinst->adev; uint32_t data = 0; /* enable JPEG CGC */ data = RREG32_SOC15(VCN, 0, mmJPEG_CGC_CTRL); - if (sw) + if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; else data |= 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; @@ -435,7 +641,7 @@ static void vcn_v1_0_enable_clock_gating(struct amdgpu_device *adev, bool sw) /* enable UVD CGC */ data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL); - if (sw) + if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; else data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; @@ -480,16 +686,161 @@ static void vcn_v1_0_enable_clock_gating(struct amdgpu_device *adev, bool sw) WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data); } +static void vcn_v1_0_clock_gating_dpg_mode(struct amdgpu_vcn_inst *vinst, + uint8_t sram_sel) +{ + struct amdgpu_device *adev = vinst->adev; + uint32_t reg_data = 0; + + /* disable JPEG CGC */ + if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) + reg_data = 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + else + reg_data = 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + reg_data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; + reg_data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT; + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmJPEG_CGC_CTRL, reg_data, 0xFFFFFFFF, sram_sel); + + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmJPEG_CGC_GATE, 0, 0xFFFFFFFF, sram_sel); + + /* enable sw clock gating control */ + if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG) + reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + else + reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT; + reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT; + reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT; + reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | + UVD_CGC_CTRL__UDEC_CM_MODE_MASK | + UVD_CGC_CTRL__UDEC_IT_MODE_MASK | + UVD_CGC_CTRL__UDEC_DB_MODE_MASK | + UVD_CGC_CTRL__UDEC_MP_MODE_MASK | + UVD_CGC_CTRL__SYS_MODE_MASK | + UVD_CGC_CTRL__UDEC_MODE_MASK | + UVD_CGC_CTRL__MPEG2_MODE_MASK | + UVD_CGC_CTRL__REGS_MODE_MASK | + UVD_CGC_CTRL__RBC_MODE_MASK | + UVD_CGC_CTRL__LMI_MC_MODE_MASK | + UVD_CGC_CTRL__LMI_UMC_MODE_MASK | + UVD_CGC_CTRL__IDCT_MODE_MASK | + UVD_CGC_CTRL__MPRD_MODE_MASK | + UVD_CGC_CTRL__MPC_MODE_MASK | + UVD_CGC_CTRL__LBSI_MODE_MASK | + UVD_CGC_CTRL__LRBBM_MODE_MASK | + UVD_CGC_CTRL__WCB_MODE_MASK | + UVD_CGC_CTRL__VCPU_MODE_MASK | + UVD_CGC_CTRL__SCPU_MODE_MASK); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_CGC_CTRL, reg_data, 0xFFFFFFFF, sram_sel); + + /* turn off clock gating */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_CGC_GATE, 0, 0xFFFFFFFF, sram_sel); + + /* turn on SUVD clock gating */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SUVD_CGC_GATE, 1, 0xFFFFFFFF, sram_sel); + + /* turn on sw mode in UVD_SUVD_CGC_CTRL */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SUVD_CGC_CTRL, 0, 0xFFFFFFFF, sram_sel); +} + +static void vcn_1_0_disable_static_power_gating(struct amdgpu_vcn_inst *vinst) +{ + struct amdgpu_device *adev = vinst->adev; + uint32_t data = 0; + + if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { + data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); + + WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); + SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON, 0xFFFFFF); + } else { + data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT + | 1 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); + WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); + SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0, 0xFFFFFFFF); + } + + /* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS , UVDU_PWR_STATUS are 0 (power on) */ + + data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); + data &= ~0x103; + if (adev->pg_flags & AMD_PG_SUPPORT_VCN) + data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON | UVD_POWER_STATUS__UVD_PG_EN_MASK; + + WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); +} + +static void vcn_1_0_enable_static_power_gating(struct amdgpu_vcn_inst *vinst) +{ + struct amdgpu_device *adev = vinst->adev; + uint32_t data = 0; + + if (adev->pg_flags & AMD_PG_SUPPORT_VCN) { + /* Before power off, this indicator has to be turned on */ + data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS); + data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK; + data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF; + WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data); + + + data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT + | 2 << UVD_PGFSM_CONFIG__UVDW_PWR_CONFIG__SHIFT); + + WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data); + + data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT + | 2 << UVD_PGFSM_STATUS__UVDW_PWR_STATUS__SHIFT); + SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFFFFF); + } +} + /** - * vcn_v1_0_start - start VCN block + * vcn_v1_0_start_spg_mode - start VCN block * - * @adev: amdgpu_device pointer + * @vinst: VCN instance * * Setup and start the VCN block */ -static int vcn_v1_0_start(struct amdgpu_device *adev) +static int vcn_v1_0_start_spg_mode(struct amdgpu_vcn_inst *vinst) { - struct amdgpu_ring *ring = &adev->vcn.ring_dec; + struct amdgpu_device *adev = vinst->adev; + struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; uint32_t rb_bufsz, tmp; uint32_t lmi_swap_cntl; int i, j, r; @@ -497,41 +848,25 @@ static int vcn_v1_0_start(struct amdgpu_device *adev) /* disable byte swapping */ lmi_swap_cntl = 0; - vcn_v1_0_mc_resume(adev); + vcn_1_0_disable_static_power_gating(vinst); + + tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY; + WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp); /* disable clock gating */ - vcn_v1_0_disable_clock_gating(adev, true); + vcn_v1_0_disable_clock_gating(vinst); /* disable interupt */ WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0, ~UVD_MASTINT_EN__VCPU_EN_MASK); - /* stall UMC and register bus before resetting VCPU */ - WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), - UVD_LMI_CTRL2__STALL_ARB_UMC_MASK, - ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); - mdelay(1); - - /* put LMI, VCPU, RBC etc... into reset */ - WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, - UVD_SOFT_RESET__LMI_SOFT_RESET_MASK | - UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | - UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK | - UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | - UVD_SOFT_RESET__CSM_SOFT_RESET_MASK | - UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | - UVD_SOFT_RESET__TAP_SOFT_RESET_MASK | - UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK); - mdelay(5); - /* initialize VCN memory controller */ - WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, - (0x40 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | - UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | - UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | - UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | - UVD_LMI_CTRL__REQ_MODE_MASK | - 0x00100000L); + tmp = RREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL); + WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, tmp | + UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | + UVD_LMI_CTRL__MASK_MC_URGENT_MASK | + UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | + UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK); #ifdef __BIG_ENDIAN /* swap (8 in 32) RB and IB */ @@ -539,41 +874,61 @@ static int vcn_v1_0_start(struct amdgpu_device *adev) #endif WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl); - WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0, 0x40c2040); - WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA1, 0x0); - WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0, 0x40c2040); - WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB1, 0x0); - WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_ALU, 0); - WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX, 0x88); + tmp = RREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL); + tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK; + tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT; + WREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL, tmp); - /* take all subblocks out of reset, except VCPU */ - WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, - UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); - mdelay(5); + WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0, + ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | + (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | + (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT))); + + WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0, + ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | + (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | + (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT))); + + WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX, + ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | + (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT))); + + vcn_v1_0_mc_resume_spg_mode(vinst); + + WREG32_SOC15(UVD, 0, mmUVD_REG_XX_MASK_1_0, 0x10); + WREG32_SOC15(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK_1_0, + RREG32_SOC15(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK_1_0) | 0x3); /* enable VCPU clock */ - WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL, - UVD_VCPU_CNTL__CLK_EN_MASK); + WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL, UVD_VCPU_CNTL__CLK_EN_MASK); + + /* boot up the VCPU */ + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0, + ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); /* enable UMC */ WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0, ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); - /* boot up the VCPU */ - WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, 0); - mdelay(10); + tmp = RREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET); + tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK; + tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK; + WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, tmp); for (i = 0; i < 10; ++i) { uint32_t status; for (j = 0; j < 100; ++j) { status = RREG32_SOC15(UVD, 0, mmUVD_STATUS); - if (status & 2) + if (status & UVD_STATUS__IDLE) break; mdelay(10); } r = 0; - if (status & 2) + if (status & UVD_STATUS__IDLE) break; DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n"); @@ -593,19 +948,22 @@ static int vcn_v1_0_start(struct amdgpu_device *adev) } /* enable master interrupt */ WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), - (UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK), - ~(UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK)); + UVD_MASTINT_EN__VCPU_EN_MASK, ~UVD_MASTINT_EN__VCPU_EN_MASK); + + /* enable system interrupt for JRBC, TODO: move to set interrupt*/ + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SYS_INT_EN), + UVD_SYS_INT_EN__UVD_JRBC_EN_MASK, + ~UVD_SYS_INT_EN__UVD_JRBC_EN_MASK); - /* clear the bit 4 of VCN_STATUS */ - WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0, - ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT)); + /* clear the busy bit of UVD_STATUS */ + tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) & ~UVD_STATUS__UVD_BUSY; + WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp); /* force RBC into idle state */ rb_bufsz = order_base_2(ring->ring_size); tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); - tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0); tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp); @@ -617,7 +975,7 @@ static int vcn_v1_0_start(struct amdgpu_device *adev) WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2)); - /* programm the RB_BASE for ring buffer */ + /* program the RB_BASE for ring buffer */ WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, lower_32_bits(ring->gpu_addr)); WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, @@ -626,6 +984,8 @@ static int vcn_v1_0_start(struct amdgpu_device *adev) /* Initialize the ring buffer's read and write pointers */ WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); + WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0); + ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr)); @@ -633,63 +993,462 @@ static int vcn_v1_0_start(struct amdgpu_device *adev) WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); - ring = &adev->vcn.ring_enc[0]; + ring = &adev->vcn.inst->ring_enc[0]; WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); - ring = &adev->vcn.ring_enc[1]; + ring = &adev->vcn.inst->ring_enc[1]; WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); + jpeg_v1_0_start(adev, 0); + + /* Keeping one read-back to ensure all register writes are done, + * otherwise it may introduce race conditions. + */ + RREG32_SOC15(UVD, 0, mmUVD_STATUS); + return 0; } +static int vcn_v1_0_start_dpg_mode(struct amdgpu_vcn_inst *vinst) +{ + struct amdgpu_device *adev = vinst->adev; + struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec; + uint32_t rb_bufsz, tmp; + uint32_t lmi_swap_cntl; + + /* disable byte swapping */ + lmi_swap_cntl = 0; + + vcn_1_0_enable_static_power_gating(vinst); + + /* enable dynamic power gating mode */ + tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS); + tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK; + tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK; + WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp); + + /* enable clock gating */ + vcn_v1_0_clock_gating_dpg_mode(vinst, 0); + + /* enable VCPU clock */ + tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT); + tmp |= UVD_VCPU_CNTL__CLK_EN_MASK; + tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK; + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_VCPU_CNTL, tmp, 0xFFFFFFFF, 0); + + /* disable interupt */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MASTINT_EN, + 0, UVD_MASTINT_EN__VCPU_EN_MASK, 0); + + /* initialize VCN memory controller */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL, + (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | + UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | + UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | + UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | + UVD_LMI_CTRL__REQ_MODE_MASK | + UVD_LMI_CTRL__CRC_RESET_MASK | + UVD_LMI_CTRL__MASK_MC_URGENT_MASK | + 0x00100000L, 0xFFFFFFFF, 0); + +#ifdef __BIG_ENDIAN + /* swap (8 in 32) RB and IB */ + lmi_swap_cntl = 0xa; +#endif + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl, 0xFFFFFFFF, 0); + + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_CNTL, + 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0xFFFFFFFF, 0); + + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUXA0, + ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) | + (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) | + (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0xFFFFFFFF, 0); + + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUXB0, + ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) | + (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) | + (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0xFFFFFFFF, 0); + + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MPC_SET_MUX, + ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) | + (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) | + (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0xFFFFFFFF, 0); + + vcn_v1_0_mc_resume_dpg_mode(vinst); + + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_REG_XX_MASK, 0x10, 0xFFFFFFFF, 0); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK, 0x3, 0xFFFFFFFF, 0); + + /* boot up the VCPU */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SOFT_RESET, 0, 0xFFFFFFFF, 0); + + /* enable UMC */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL2, + 0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, + 0xFFFFFFFF, 0); + + /* enable master interrupt */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_MASTINT_EN, + UVD_MASTINT_EN__VCPU_EN_MASK, UVD_MASTINT_EN__VCPU_EN_MASK, 0); + + vcn_v1_0_clock_gating_dpg_mode(vinst, 1); + /* setup mmUVD_LMI_CTRL */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_LMI_CTRL, + (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) | + UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK | + UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK | + UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK | + UVD_LMI_CTRL__REQ_MODE_MASK | + UVD_LMI_CTRL__CRC_RESET_MASK | + UVD_LMI_CTRL__MASK_MC_URGENT_MASK | + 0x00100000L, 0xFFFFFFFF, 1); + + tmp = adev->gfx.config.gb_addr_config; + /* setup VCN global tiling registers */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_JPEG_ADDR_CONFIG, tmp, 0xFFFFFFFF, 1); + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_JPEG_UV_ADDR_CONFIG, tmp, 0xFFFFFFFF, 1); + + /* enable System Interrupt for JRBC */ + WREG32_SOC15_DPG_MODE_1_0(UVD, 0, mmUVD_SYS_INT_EN, + UVD_SYS_INT_EN__UVD_JRBC_EN_MASK, 0xFFFFFFFF, 1); + + /* force RBC into idle state */ + rb_bufsz = order_base_2(ring->ring_size); + tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); + tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp); + + /* set the write pointer delay */ + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0); + + /* set the wb address */ + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR, + (upper_32_bits(ring->gpu_addr) >> 2)); + + /* program the RB_BASE for ring buffer */ + WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, + lower_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, + upper_32_bits(ring->gpu_addr)); + + /* Initialize the ring buffer's read and write pointers */ + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0); + + WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0); + + ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR); + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, + lower_32_bits(ring->wptr)); + + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0, + ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); + + jpeg_v1_0_start(adev, 1); + + /* Keeping one read-back to ensure all register writes are done, + * otherwise it may introduce race conditions. + */ + RREG32_SOC15(UVD, 0, mmUVD_STATUS); + + return 0; +} + +static int vcn_v1_0_start(struct amdgpu_vcn_inst *vinst) +{ + struct amdgpu_device *adev = vinst->adev; + + return (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ? + vcn_v1_0_start_dpg_mode(vinst) : vcn_v1_0_start_spg_mode(vinst); +} + /** - * vcn_v1_0_stop - stop VCN block + * vcn_v1_0_stop_spg_mode - stop VCN block * - * @adev: amdgpu_device pointer + * @vinst: VCN instance * * stop the VCN block */ -static int vcn_v1_0_stop(struct amdgpu_device *adev) +static int vcn_v1_0_stop_spg_mode(struct amdgpu_vcn_inst *vinst) { - /* force RBC into idle state */ - WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, 0x11010101); + struct amdgpu_device *adev = vinst->adev; + int tmp; + + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7); + + tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK | + UVD_LMI_STATUS__READ_CLEAN_MASK | + UVD_LMI_STATUS__WRITE_CLEAN_MASK | + UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK; + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_LMI_STATUS, tmp, tmp); - /* Stall UMC and register bus before resetting VCPU */ + /* stall UMC channel */ WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), - UVD_LMI_CTRL2__STALL_ARB_UMC_MASK, - ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); - mdelay(1); + UVD_LMI_CTRL2__STALL_ARB_UMC_MASK, + ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); - /* put VCPU into reset */ - WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, - UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); - mdelay(5); + tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK | + UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK; + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_LMI_STATUS, tmp, tmp); /* disable VCPU clock */ - WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL, 0x0); + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0, + ~UVD_VCPU_CNTL__CLK_EN_MASK); - /* Unstall UMC and register bus */ - WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0, - ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK); + /* reset LMI UMC/LMI */ + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), + UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK, + ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK); - /* enable clock gating */ - vcn_v1_0_enable_clock_gating(adev, true); + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), + UVD_SOFT_RESET__LMI_SOFT_RESET_MASK, + ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK); + + /* put VCPU into reset */ + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), + UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, + ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); + + WREG32_SOC15(UVD, 0, mmUVD_STATUS, 0); + + vcn_v1_0_enable_clock_gating(vinst); + vcn_1_0_enable_static_power_gating(vinst); + + /* Keeping one read-back to ensure all register writes are done, + * otherwise it may introduce race conditions. + */ + RREG32_SOC15(UVD, 0, mmUVD_STATUS); return 0; } -static int vcn_v1_0_set_clockgating_state(void *handle, +static int vcn_v1_0_stop_dpg_mode(struct amdgpu_vcn_inst *vinst) +{ + struct amdgpu_device *adev = vinst->adev; + uint32_t tmp; + + /* Wait for power status to be UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF */ + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, + UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); + + /* wait for read ptr to be equal to write ptr */ + tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF); + + tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF); + + tmp = RREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_JRBC_RB_RPTR, tmp, 0xFFFFFFFF); + + tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF; + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF); + + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, + UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); + + /* disable dynamic power gating mode */ + WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0, + ~UVD_POWER_STATUS__UVD_PG_MODE_MASK); + + /* Keeping one read-back to ensure all register writes are done, + * otherwise it may introduce race conditions. + */ + RREG32_SOC15(UVD, 0, mmUVD_STATUS); + + return 0; +} + +static int vcn_v1_0_stop(struct amdgpu_vcn_inst *vinst) +{ + struct amdgpu_device *adev = vinst->adev; + int r; + + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) + r = vcn_v1_0_stop_dpg_mode(vinst); + else + r = vcn_v1_0_stop_spg_mode(vinst); + + return r; +} + +static int vcn_v1_0_pause_dpg_mode(struct amdgpu_vcn_inst *vinst, + struct dpg_pause_state *new_state) +{ + struct amdgpu_device *adev = vinst->adev; + int inst_idx = vinst->inst; + int ret_code; + uint32_t reg_data = 0; + uint32_t reg_data2 = 0; + struct amdgpu_ring *ring; + + /* pause/unpause if state is changed */ + if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) { + DRM_DEBUG("dpg pause state changed %d:%d -> %d:%d", + adev->vcn.inst[inst_idx].pause_state.fw_based, + adev->vcn.inst[inst_idx].pause_state.jpeg, + new_state->fw_based, new_state->jpeg); + + reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) & + (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); + + if (new_state->fw_based == VCN_DPG_STATE__PAUSE) { + ret_code = 0; + + if (!(reg_data & UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK)) + ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, + UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); + + if (!ret_code) { + /* pause DPG non-jpeg */ + reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; + WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE, + UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, + UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK); + + /* Restore */ + ring = &adev->vcn.inst->ring_enc[0]; + WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr); + WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4); + WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr)); + WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); + + ring = &adev->vcn.inst->ring_enc[1]; + WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr); + WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4); + WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr)); + WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr)); + + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, + RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, + UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); + } + } else { + /* unpause dpg non-jpeg, no need to wait */ + reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK; + WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); + } + adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based; + } + + /* pause/unpause if state is changed */ + if (adev->vcn.inst[inst_idx].pause_state.jpeg != new_state->jpeg) { + DRM_DEBUG("dpg pause state changed %d:%d -> %d:%d", + adev->vcn.inst[inst_idx].pause_state.fw_based, + adev->vcn.inst[inst_idx].pause_state.jpeg, + new_state->fw_based, new_state->jpeg); + + reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) & + (~UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK); + + if (new_state->jpeg == VCN_DPG_STATE__PAUSE) { + ret_code = 0; + + if (!(reg_data & UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK)) + ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, + UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); + + if (!ret_code) { + /* Make sure JPRG Snoop is disabled before sending the pause */ + reg_data2 = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS); + reg_data2 |= UVD_POWER_STATUS__JRBC_SNOOP_DIS_MASK; + WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, reg_data2); + + /* pause DPG jpeg */ + reg_data |= UVD_DPG_PAUSE__JPEG_PAUSE_DPG_REQ_MASK; + WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE, + UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK, + UVD_DPG_PAUSE__JPEG_PAUSE_DPG_ACK_MASK); + + /* Restore */ + ring = adev->jpeg.inst->ring_dec; + WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_VMID, 0); + WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_CNTL, + UVD_JRBC_RB_CNTL__RB_NO_FETCH_MASK | + UVD_JRBC_RB_CNTL__RB_RPTR_WR_EN_MASK); + WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_LOW, + lower_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_HIGH, + upper_32_bits(ring->gpu_addr)); + WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_RPTR, ring->wptr); + WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_WPTR, ring->wptr); + WREG32_SOC15(UVD, 0, mmUVD_JRBC_RB_CNTL, + UVD_JRBC_RB_CNTL__RB_RPTR_WR_EN_MASK); + + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, + RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF); + SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, + UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK); + } + } else { + /* unpause dpg jpeg, no need to wait */ + reg_data &= ~UVD_DPG_PAUSE__JPEG_PAUSE_DPG_REQ_MASK; + WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data); + } + adev->vcn.inst[inst_idx].pause_state.jpeg = new_state->jpeg; + } + + return 0; +} + +static bool vcn_v1_0_is_idle(struct amdgpu_ip_block *ip_block) +{ + struct amdgpu_device *adev = ip_block->adev; + + return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE); +} + +static int vcn_v1_0_wait_for_idle(struct amdgpu_ip_block *ip_block) +{ + struct amdgpu_device *adev = ip_block->adev; + int ret; + + ret = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, + UVD_STATUS__IDLE); + + return ret; +} + +static int vcn_v1_0_set_clockgating_state(struct amdgpu_ip_block *ip_block, enum amd_clockgating_state state) { - /* needed for driver unload*/ + struct amdgpu_device *adev = ip_block->adev; + struct amdgpu_vcn_inst *vinst = adev->vcn.inst; + bool enable = (state == AMD_CG_STATE_GATE); + + if (enable) { + /* wait for STATUS to clear */ + if (!vcn_v1_0_is_idle(ip_block)) + return -EBUSY; + vcn_v1_0_enable_clock_gating(vinst); + } else { + /* disable HW gating and enable Sw gating */ + vcn_v1_0_disable_clock_gating(vinst); + } return 0; } @@ -732,6 +1491,10 @@ static void vcn_v1_0_dec_ring_set_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) + WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, + lower_32_bits(ring->wptr) | 0x80000000); + WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr)); } @@ -744,6 +1507,8 @@ static void vcn_v1_0_dec_ring_set_wptr(struct amdgpu_ring *ring) */ static void vcn_v1_0_dec_ring_insert_start(struct amdgpu_ring *ring) { + struct amdgpu_device *adev = ring->adev; + amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); amdgpu_ring_write(ring, 0); @@ -761,6 +1526,8 @@ static void vcn_v1_0_dec_ring_insert_start(struct amdgpu_ring *ring) */ static void vcn_v1_0_dec_ring_insert_end(struct amdgpu_ring *ring) { + struct amdgpu_device *adev = ring->adev; + amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); amdgpu_ring_write(ring, VCN_DEC_CMD_PACKET_END << 1); @@ -770,13 +1537,17 @@ static void vcn_v1_0_dec_ring_insert_end(struct amdgpu_ring *ring) * vcn_v1_0_dec_ring_emit_fence - emit an fence & trap command * * @ring: amdgpu_ring pointer - * @fence: fence to emit + * @addr: address + * @seq: sequence number + * @flags: fence related flags * * Write a fence and a trap command to the ring. */ static void vcn_v1_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, unsigned flags) { + struct amdgpu_device *adev = ring->adev; + WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); amdgpu_ring_write(ring, @@ -804,33 +1575,26 @@ static void vcn_v1_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 } /** - * vcn_v1_0_dec_ring_hdp_invalidate - emit an hdp invalidate - * - * @ring: amdgpu_ring pointer - * - * Emits an hdp invalidate. - */ -static void vcn_v1_0_dec_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) -{ - amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(HDP, 0, mmHDP_DEBUG0), 0)); - amdgpu_ring_write(ring, 1); -} - -/** * vcn_v1_0_dec_ring_emit_ib - execute indirect buffer * * @ring: amdgpu_ring pointer + * @job: job to retrieve vmid from * @ib: indirect buffer to execute + * @flags: unused * * Write ring commands to execute the indirect buffer */ static void vcn_v1_0_dec_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, - unsigned vm_id, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + struct amdgpu_device *adev = ring->adev; + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_VMID), 0)); - amdgpu_ring_write(ring, vm_id); + amdgpu_ring_write(ring, vmid); amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_LOW), 0)); @@ -843,29 +1607,18 @@ static void vcn_v1_0_dec_ring_emit_ib(struct amdgpu_ring *ring, amdgpu_ring_write(ring, ib->length_dw); } -static void vcn_v1_0_dec_vm_reg_write(struct amdgpu_ring *ring, - uint32_t data0, uint32_t data1) +static void vcn_v1_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, + uint32_t reg, uint32_t val, + uint32_t mask) { - amdgpu_ring_write(ring, - PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); - amdgpu_ring_write(ring, data0); - amdgpu_ring_write(ring, - PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); - amdgpu_ring_write(ring, data1); - amdgpu_ring_write(ring, - PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); - amdgpu_ring_write(ring, VCN_DEC_CMD_WRITE_REG << 1); -} + struct amdgpu_device *adev = ring->adev; -static void vcn_v1_0_dec_vm_reg_wait(struct amdgpu_ring *ring, - uint32_t data0, uint32_t data1, uint32_t mask) -{ amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); - amdgpu_ring_write(ring, data0); + amdgpu_ring_write(ring, reg << 2); amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); - amdgpu_ring_write(ring, data1); + amdgpu_ring_write(ring, val); amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GP_SCRATCH8), 0)); amdgpu_ring_write(ring, mask); @@ -875,39 +1628,34 @@ static void vcn_v1_0_dec_vm_reg_wait(struct amdgpu_ring *ring, } static void vcn_v1_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring, - unsigned vm_id, uint64_t pd_addr) + unsigned vmid, uint64_t pd_addr) { - struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub]; - uint32_t req = ring->adev->gart.gart_funcs->get_invalidate_req(vm_id); + struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub]; uint32_t data0, data1, mask; - unsigned eng = ring->vm_inv_eng; - - pd_addr = amdgpu_gart_get_vm_pde(ring->adev, pd_addr); - pd_addr |= AMDGPU_PTE_VALID; - data0 = (hub->ctx0_ptb_addr_hi32 + vm_id * 2) << 2; - data1 = upper_32_bits(pd_addr); - vcn_v1_0_dec_vm_reg_write(ring, data0, data1); + pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); - data0 = (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2; - data1 = lower_32_bits(pd_addr); - vcn_v1_0_dec_vm_reg_write(ring, data0, data1); - - data0 = (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2; + /* wait for register write */ + data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance; data1 = lower_32_bits(pd_addr); mask = 0xffffffff; - vcn_v1_0_dec_vm_reg_wait(ring, data0, data1, mask); - - /* flush TLB */ - data0 = (hub->vm_inv_eng0_req + eng) << 2; - data1 = req; - vcn_v1_0_dec_vm_reg_write(ring, data0, data1); - - /* wait for flush */ - data0 = (hub->vm_inv_eng0_ack + eng) << 2; - data1 = 1 << vm_id; - mask = 1 << vm_id; - vcn_v1_0_dec_vm_reg_wait(ring, data0, data1, mask); + vcn_v1_0_dec_ring_emit_reg_wait(ring, data0, data1, mask); +} + +static void vcn_v1_0_dec_ring_emit_wreg(struct amdgpu_ring *ring, + uint32_t reg, uint32_t val) +{ + struct amdgpu_device *adev = ring->adev; + + amdgpu_ring_write(ring, + PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0)); + amdgpu_ring_write(ring, reg << 2); + amdgpu_ring_write(ring, + PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0)); + amdgpu_ring_write(ring, val); + amdgpu_ring_write(ring, + PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0)); + amdgpu_ring_write(ring, VCN_DEC_CMD_WRITE_REG << 1); } /** @@ -921,7 +1669,7 @@ static uint64_t vcn_v1_0_enc_ring_get_rptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - if (ring == &adev->vcn.ring_enc[0]) + if (ring == &adev->vcn.inst->ring_enc[0]) return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR); else return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2); @@ -938,7 +1686,7 @@ static uint64_t vcn_v1_0_enc_ring_get_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - if (ring == &adev->vcn.ring_enc[0]) + if (ring == &adev->vcn.inst->ring_enc[0]) return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR); else return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2); @@ -955,7 +1703,7 @@ static void vcn_v1_0_enc_ring_set_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - if (ring == &adev->vcn.ring_enc[0]) + if (ring == &adev->vcn.inst->ring_enc[0]) WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr)); else @@ -967,7 +1715,9 @@ static void vcn_v1_0_enc_ring_set_wptr(struct amdgpu_ring *ring) * vcn_v1_0_enc_ring_emit_fence - emit an enc fence & trap command * * @ring: amdgpu_ring pointer - * @fence: fence to emit + * @addr: address + * @seq: sequence number + * @flags: fence related flags * * Write enc a fence and a trap command to the ring. */ @@ -992,56 +1742,55 @@ static void vcn_v1_0_enc_ring_insert_end(struct amdgpu_ring *ring) * vcn_v1_0_enc_ring_emit_ib - enc execute indirect buffer * * @ring: amdgpu_ring pointer + * @job: job to retrive vmid from * @ib: indirect buffer to execute + * @flags: unused * * Write enc ring commands to execute the indirect buffer */ static void vcn_v1_0_enc_ring_emit_ib(struct amdgpu_ring *ring, - struct amdgpu_ib *ib, unsigned int vm_id, bool ctx_switch) + struct amdgpu_job *job, + struct amdgpu_ib *ib, + uint32_t flags) { + unsigned vmid = AMDGPU_JOB_GET_VMID(job); + amdgpu_ring_write(ring, VCN_ENC_CMD_IB); - amdgpu_ring_write(ring, vm_id); + amdgpu_ring_write(ring, vmid); amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); amdgpu_ring_write(ring, ib->length_dw); } -static void vcn_v1_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring, - unsigned int vm_id, uint64_t pd_addr) +static void vcn_v1_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, + uint32_t reg, uint32_t val, + uint32_t mask) { - struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub]; - uint32_t req = ring->adev->gart.gart_funcs->get_invalidate_req(vm_id); - unsigned eng = ring->vm_inv_eng; - - pd_addr = amdgpu_gart_get_vm_pde(ring->adev, pd_addr); - pd_addr |= AMDGPU_PTE_VALID; + amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT); + amdgpu_ring_write(ring, reg << 2); + amdgpu_ring_write(ring, mask); + amdgpu_ring_write(ring, val); +} - amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE); - amdgpu_ring_write(ring, - (hub->ctx0_ptb_addr_hi32 + vm_id * 2) << 2); - amdgpu_ring_write(ring, upper_32_bits(pd_addr)); +static void vcn_v1_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring, + unsigned int vmid, uint64_t pd_addr) +{ + struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub]; - amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE); - amdgpu_ring_write(ring, - (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2); - amdgpu_ring_write(ring, lower_32_bits(pd_addr)); + pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); - amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT); - amdgpu_ring_write(ring, - (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2); - amdgpu_ring_write(ring, 0xffffffff); - amdgpu_ring_write(ring, lower_32_bits(pd_addr)); + /* wait for reg writes */ + vcn_v1_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + + vmid * hub->ctx_addr_distance, + lower_32_bits(pd_addr), 0xffffffff); +} - /* flush TLB */ +static void vcn_v1_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, + uint32_t reg, uint32_t val) +{ amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE); - amdgpu_ring_write(ring, (hub->vm_inv_eng0_req + eng) << 2); - amdgpu_ring_write(ring, req); - - /* wait for flush */ - amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT); - amdgpu_ring_write(ring, (hub->vm_inv_eng0_ack + eng) << 2); - amdgpu_ring_write(ring, 1 << vm_id); - amdgpu_ring_write(ring, 1 << vm_id); + amdgpu_ring_write(ring, reg << 2); + amdgpu_ring_write(ring, val); } static int vcn_v1_0_set_interrupt_state(struct amdgpu_device *adev, @@ -1060,13 +1809,13 @@ static int vcn_v1_0_process_interrupt(struct amdgpu_device *adev, switch (entry->src_id) { case 124: - amdgpu_fence_process(&adev->vcn.ring_dec); + amdgpu_fence_process(&adev->vcn.inst->ring_dec); break; case 119: - amdgpu_fence_process(&adev->vcn.ring_enc[0]); + amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]); break; case 120: - amdgpu_fence_process(&adev->vcn.ring_enc[1]); + amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]); break; default: DRM_ERROR("Unhandled interrupt: %d %d\n", @@ -1077,53 +1826,323 @@ static int vcn_v1_0_process_interrupt(struct amdgpu_device *adev, return 0; } +static void vcn_v1_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) +{ + struct amdgpu_device *adev = ring->adev; + int i; + + WARN_ON(ring->wptr % 2 || count % 2); + + for (i = 0; i < count / 2; i++) { + amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP), 0)); + amdgpu_ring_write(ring, 0); + } +} + +static int vcn_v1_0_set_pg_state(struct amdgpu_vcn_inst *vinst, + enum amd_powergating_state state) +{ + /* This doesn't actually powergate the VCN block. + * That's done in the dpm code via the SMC. This + * just re-inits the block as necessary. The actual + * gating still happens in the dpm code. We should + * revisit this when there is a cleaner line between + * the smc and the hw blocks + */ + int ret; + + if (state == vinst->cur_state) + return 0; + + if (state == AMD_PG_STATE_GATE) + ret = vcn_v1_0_stop(vinst); + else + ret = vcn_v1_0_start(vinst); + + if (!ret) + vinst->cur_state = state; + + return ret; +} + +static void vcn_v1_0_idle_work_handler(struct work_struct *work) +{ + struct amdgpu_vcn_inst *vcn_inst = + container_of(work, struct amdgpu_vcn_inst, idle_work.work); + struct amdgpu_device *adev = vcn_inst->adev; + unsigned int fences = 0, i; + + for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) + fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_enc[i]); + + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { + struct dpg_pause_state new_state; + + if (fences) + new_state.fw_based = VCN_DPG_STATE__PAUSE; + else + new_state.fw_based = VCN_DPG_STATE__UNPAUSE; + + if (amdgpu_fence_count_emitted(adev->jpeg.inst->ring_dec)) + new_state.jpeg = VCN_DPG_STATE__PAUSE; + else + new_state.jpeg = VCN_DPG_STATE__UNPAUSE; + + adev->vcn.inst->pause_dpg_mode(vcn_inst, &new_state); + } + + fences += amdgpu_fence_count_emitted(adev->jpeg.inst->ring_dec); + fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_dec); + + if (fences == 0) { + amdgpu_gfx_off_ctrl(adev, true); + if (adev->pm.dpm_enabled) + amdgpu_dpm_enable_vcn(adev, false, 0); + else + amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, + AMD_PG_STATE_GATE); + } else { + schedule_delayed_work(&adev->vcn.inst[0].idle_work, VCN_IDLE_TIMEOUT); + } +} + +static void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring) +{ + struct amdgpu_device *adev = ring->adev; + bool set_clocks = !cancel_delayed_work_sync(&adev->vcn.inst[0].idle_work); + + mutex_lock(&adev->vcn.inst[0].vcn1_jpeg1_workaround); + + if (amdgpu_fence_wait_empty(ring->adev->jpeg.inst->ring_dec)) + DRM_ERROR("VCN dec: jpeg dec ring may not be empty\n"); + + vcn_v1_0_set_pg_for_begin_use(ring, set_clocks); + +} + +void vcn_v1_0_set_pg_for_begin_use(struct amdgpu_ring *ring, bool set_clocks) +{ + struct amdgpu_device *adev = ring->adev; + + if (set_clocks) { + amdgpu_gfx_off_ctrl(adev, false); + if (adev->pm.dpm_enabled) + amdgpu_dpm_enable_vcn(adev, true, 0); + else + amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN, + AMD_PG_STATE_UNGATE); + } + + if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) { + struct dpg_pause_state new_state; + unsigned int fences = 0, i; + + for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) + fences += amdgpu_fence_count_emitted(&adev->vcn.inst->ring_enc[i]); + + if (fences) + new_state.fw_based = VCN_DPG_STATE__PAUSE; + else + new_state.fw_based = VCN_DPG_STATE__UNPAUSE; + + if (amdgpu_fence_count_emitted(adev->jpeg.inst->ring_dec)) + new_state.jpeg = VCN_DPG_STATE__PAUSE; + else + new_state.jpeg = VCN_DPG_STATE__UNPAUSE; + + if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) + new_state.fw_based = VCN_DPG_STATE__PAUSE; + else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG) + new_state.jpeg = VCN_DPG_STATE__PAUSE; + + adev->vcn.inst->pause_dpg_mode(adev->vcn.inst, &new_state); + } +} + +void vcn_v1_0_ring_end_use(struct amdgpu_ring *ring) +{ + schedule_delayed_work(&ring->adev->vcn.inst[0].idle_work, VCN_IDLE_TIMEOUT); + mutex_unlock(&ring->adev->vcn.inst[0].vcn1_jpeg1_workaround); +} + +static void vcn_v1_0_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p) +{ + struct amdgpu_device *adev = ip_block->adev; + int i, j; + uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_1_0); + uint32_t inst_off, is_powered; + + if (!adev->vcn.ip_dump) + return; + + drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst); + for (i = 0; i < adev->vcn.num_vcn_inst; i++) { + if (adev->vcn.harvest_config & (1 << i)) { + drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i); + continue; + } + + inst_off = i * reg_count; + is_powered = (adev->vcn.ip_dump[inst_off] & + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1; + + if (is_powered) { + drm_printf(p, "\nActive Instance:VCN%d\n", i); + for (j = 0; j < reg_count; j++) + drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_1_0[j].reg_name, + adev->vcn.ip_dump[inst_off + j]); + } else { + drm_printf(p, "\nInactive Instance:VCN%d\n", i); + } + } +} + +static void vcn_v1_0_dump_ip_state(struct amdgpu_ip_block *ip_block) +{ + struct amdgpu_device *adev = ip_block->adev; + int i, j; + bool is_powered; + uint32_t inst_off; + uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_1_0); + + if (!adev->vcn.ip_dump) + return; + + for (i = 0; i < adev->vcn.num_vcn_inst; i++) { + if (adev->vcn.harvest_config & (1 << i)) + continue; + + inst_off = i * reg_count; + /* mmUVD_POWER_STATUS is always readable and is first element of the array */ + adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, i, mmUVD_POWER_STATUS); + is_powered = (adev->vcn.ip_dump[inst_off] & + UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1; + + if (is_powered) + for (j = 1; j < reg_count; j++) + adev->vcn.ip_dump[inst_off + j] = + RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_1_0[j], i)); + } +} + static const struct amd_ip_funcs vcn_v1_0_ip_funcs = { .name = "vcn_v1_0", .early_init = vcn_v1_0_early_init, - .late_init = NULL, .sw_init = vcn_v1_0_sw_init, .sw_fini = vcn_v1_0_sw_fini, .hw_init = vcn_v1_0_hw_init, .hw_fini = vcn_v1_0_hw_fini, .suspend = vcn_v1_0_suspend, .resume = vcn_v1_0_resume, - .is_idle = NULL /* vcn_v1_0_is_idle */, - .wait_for_idle = NULL /* vcn_v1_0_wait_for_idle */, - .check_soft_reset = NULL /* vcn_v1_0_check_soft_reset */, - .pre_soft_reset = NULL /* vcn_v1_0_pre_soft_reset */, - .soft_reset = NULL /* vcn_v1_0_soft_reset */, - .post_soft_reset = NULL /* vcn_v1_0_post_soft_reset */, + .is_idle = vcn_v1_0_is_idle, + .wait_for_idle = vcn_v1_0_wait_for_idle, .set_clockgating_state = vcn_v1_0_set_clockgating_state, - .set_powergating_state = NULL /* vcn_v1_0_set_powergating_state */, + .set_powergating_state = vcn_set_powergating_state, + .dump_ip_state = vcn_v1_0_dump_ip_state, + .print_ip_state = vcn_v1_0_print_ip_state, }; +/* + * It is a hardware issue that VCN can't handle a GTT TMZ buffer on + * CHIP_RAVEN series ASIC. Move such a GTT TMZ buffer to VRAM domain + * before command submission as a workaround. + */ +static int vcn_v1_0_validate_bo(struct amdgpu_cs_parser *parser, + struct amdgpu_job *job, + uint64_t addr) +{ + struct ttm_operation_ctx ctx = { false, false }; + struct amdgpu_fpriv *fpriv = parser->filp->driver_priv; + struct amdgpu_vm *vm = &fpriv->vm; + struct amdgpu_bo_va_mapping *mapping; + struct amdgpu_bo *bo; + int r; + + addr &= AMDGPU_GMC_HOLE_MASK; + if (addr & 0x7) { + DRM_ERROR("VCN messages must be 8 byte aligned!\n"); + return -EINVAL; + } + + mapping = amdgpu_vm_bo_lookup_mapping(vm, addr/AMDGPU_GPU_PAGE_SIZE); + if (!mapping || !mapping->bo_va || !mapping->bo_va->base.bo) + return -EINVAL; + + bo = mapping->bo_va->base.bo; + if (!(bo->flags & AMDGPU_GEM_CREATE_ENCRYPTED)) + return 0; + + amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_VRAM); + r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); + if (r) { + DRM_ERROR("Failed to validate the VCN message BO (%d)!\n", r); + return r; + } + + return r; +} + +static int vcn_v1_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p, + struct amdgpu_job *job, + struct amdgpu_ib *ib) +{ + uint32_t msg_lo = 0, msg_hi = 0; + int i, r; + + if (!(ib->flags & AMDGPU_IB_FLAGS_SECURE)) + return 0; + + for (i = 0; i < ib->length_dw; i += 2) { + uint32_t reg = amdgpu_ib_get_value(ib, i); + uint32_t val = amdgpu_ib_get_value(ib, i + 1); + + if (reg == PACKET0(p->adev->vcn.inst[0].internal.data0, 0)) { + msg_lo = val; + } else if (reg == PACKET0(p->adev->vcn.inst[0].internal.data1, 0)) { + msg_hi = val; + } else if (reg == PACKET0(p->adev->vcn.inst[0].internal.cmd, 0)) { + r = vcn_v1_0_validate_bo(p, job, + ((u64)msg_hi) << 32 | msg_lo); + if (r) + return r; + } + } + + return 0; +} + static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = { .type = AMDGPU_RING_TYPE_VCN_DEC, .align_mask = 0xf, - .nop = PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP), 0), .support_64bit_ptrs = false, - .vmhub = AMDGPU_MMHUB, + .no_user_fence = true, + .secure_submission_supported = true, .get_rptr = vcn_v1_0_dec_ring_get_rptr, .get_wptr = vcn_v1_0_dec_ring_get_wptr, .set_wptr = vcn_v1_0_dec_ring_set_wptr, + .patch_cs_in_place = vcn_v1_0_ring_patch_cs_in_place, .emit_frame_size = - 2 + /* vcn_v1_0_dec_ring_emit_hdp_invalidate */ - 34 + /* vcn_v1_0_dec_ring_emit_vm_flush */ + 6 + 6 + /* hdp invalidate / flush */ + SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 + + SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 + + 8 + /* vcn_v1_0_dec_ring_emit_vm_flush */ 14 + 14 + /* vcn_v1_0_dec_ring_emit_fence x2 vm fence */ 6, .emit_ib_size = 8, /* vcn_v1_0_dec_ring_emit_ib */ .emit_ib = vcn_v1_0_dec_ring_emit_ib, .emit_fence = vcn_v1_0_dec_ring_emit_fence, .emit_vm_flush = vcn_v1_0_dec_ring_emit_vm_flush, - .emit_hdp_invalidate = vcn_v1_0_dec_ring_emit_hdp_invalidate, .test_ring = amdgpu_vcn_dec_ring_test_ring, .test_ib = amdgpu_vcn_dec_ring_test_ib, - .insert_nop = amdgpu_ring_insert_nop, + .insert_nop = vcn_v1_0_dec_ring_insert_nop, .insert_start = vcn_v1_0_dec_ring_insert_start, .insert_end = vcn_v1_0_dec_ring_insert_end, .pad_ib = amdgpu_ring_generic_pad_ib, - .begin_use = amdgpu_vcn_ring_begin_use, - .end_use = amdgpu_vcn_ring_end_use, + .begin_use = vcn_v1_0_ring_begin_use, + .end_use = vcn_v1_0_ring_end_use, + .emit_wreg = vcn_v1_0_dec_ring_emit_wreg, + .emit_reg_wait = vcn_v1_0_dec_ring_emit_reg_wait, + .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, }; static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = { @@ -1131,12 +2150,14 @@ static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = { .align_mask = 0x3f, .nop = VCN_ENC_CMD_NO_OP, .support_64bit_ptrs = false, - .vmhub = AMDGPU_MMHUB, + .no_user_fence = true, .get_rptr = vcn_v1_0_enc_ring_get_rptr, .get_wptr = vcn_v1_0_enc_ring_get_wptr, .set_wptr = vcn_v1_0_enc_ring_set_wptr, .emit_frame_size = - 17 + /* vcn_v1_0_enc_ring_emit_vm_flush */ + SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 + + SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 + + 4 + /* vcn_v1_0_enc_ring_emit_vm_flush */ 5 + 5 + /* vcn_v1_0_enc_ring_emit_fence x2 vm fence */ 1, /* vcn_v1_0_enc_ring_insert_end */ .emit_ib_size = 5, /* vcn_v1_0_enc_ring_emit_ib */ @@ -1148,24 +2169,24 @@ static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = { .insert_nop = amdgpu_ring_insert_nop, .insert_end = vcn_v1_0_enc_ring_insert_end, .pad_ib = amdgpu_ring_generic_pad_ib, - .begin_use = amdgpu_vcn_ring_begin_use, - .end_use = amdgpu_vcn_ring_end_use, + .begin_use = vcn_v1_0_ring_begin_use, + .end_use = vcn_v1_0_ring_end_use, + .emit_wreg = vcn_v1_0_enc_ring_emit_wreg, + .emit_reg_wait = vcn_v1_0_enc_ring_emit_reg_wait, + .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper, }; static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev) { - adev->vcn.ring_dec.funcs = &vcn_v1_0_dec_ring_vm_funcs; - DRM_INFO("VCN decode is enabled in VM mode\n"); + adev->vcn.inst->ring_dec.funcs = &vcn_v1_0_dec_ring_vm_funcs; } static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev) { int i; - for (i = 0; i < adev->vcn.num_enc_rings; ++i) - adev->vcn.ring_enc[i].funcs = &vcn_v1_0_enc_ring_vm_funcs; - - DRM_INFO("VCN encode is enabled in VM mode\n"); + for (i = 0; i < adev->vcn.inst[0].num_enc_rings; ++i) + adev->vcn.inst->ring_enc[i].funcs = &vcn_v1_0_enc_ring_vm_funcs; } static const struct amdgpu_irq_src_funcs vcn_v1_0_irq_funcs = { @@ -1175,12 +2196,11 @@ static const struct amdgpu_irq_src_funcs vcn_v1_0_irq_funcs = { static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev) { - adev->uvd.irq.num_types = adev->vcn.num_enc_rings + 1; - adev->vcn.irq.funcs = &vcn_v1_0_irq_funcs; + adev->vcn.inst->irq.num_types = adev->vcn.inst[0].num_enc_rings + 2; + adev->vcn.inst->irq.funcs = &vcn_v1_0_irq_funcs; } -const struct amdgpu_ip_block_version vcn_v1_0_ip_block = -{ +const struct amdgpu_ip_block_version vcn_v1_0_ip_block = { .type = AMD_IP_BLOCK_TYPE_VCN, .major = 1, .minor = 0, |