From add7e4fc2420e35f200e4aa13764708e62690e2e Mon Sep 17 00:00:00 2001 From: Changbin Du Date: Thu, 2 Nov 2017 13:33:42 +0800 Subject: drm/i915/gvt: Emulate PCI expansion ROM base address register Our vGPU doesn't have a device ROM, we need follow the PCI spec to report this info to drivers. Otherwise, we would see below errors. Inspecting possible rom at 0xfe049000 (vd=8086:1912 bdf=00:10.0) qemu-system-x86_64: vfio-pci: Cannot read device rom at 00000000-0000-0000-0000-000000000001 Device option ROM contents are probably invalid (check dmesg). Skip option ROM probe with rombar=0, or load from file with romfile=No option rom signature (got 4860) I will also send a improvement patch to PCI subsystem related to PCI ROM. But no idea to omit below error, since no pattern to detect vbios shadow without touch its content. 0000:00:10.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0x0000 Signed-off-by: Changbin Du Signed-off-by: Zhenyu Wang (cherry picked from commit c4270d122ccff963a021d1beb893d6192336af96) --- drivers/gpu/drm/i915/gvt/cfg_space.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) (limited to 'drivers') diff --git a/drivers/gpu/drm/i915/gvt/cfg_space.c b/drivers/gpu/drm/i915/gvt/cfg_space.c index ab19545d59a1..4ce2e6bd0680 100644 --- a/drivers/gpu/drm/i915/gvt/cfg_space.c +++ b/drivers/gpu/drm/i915/gvt/cfg_space.c @@ -208,6 +208,20 @@ static int emulate_pci_command_write(struct intel_vgpu *vgpu, return 0; } +static int emulate_pci_rom_bar_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 *pval = (u32 *)(vgpu_cfg_space(vgpu) + offset); + u32 new = *(u32 *)(p_data); + + if ((new & PCI_ROM_ADDRESS_MASK) == PCI_ROM_ADDRESS_MASK) + /* We don't have rom, return size of 0. */ + *pval = 0; + else + vgpu_pci_cfg_mem_write(vgpu, offset, p_data, bytes); + return 0; +} + static int emulate_pci_bar_write(struct intel_vgpu *vgpu, unsigned int offset, void *p_data, unsigned int bytes) { @@ -300,6 +314,11 @@ int intel_vgpu_emulate_cfg_write(struct intel_vgpu *vgpu, unsigned int offset, } switch (rounddown(offset, 4)) { + case PCI_ROM_ADDRESS: + if (WARN_ON(!IS_ALIGNED(offset, 4))) + return -EINVAL; + return emulate_pci_rom_bar_write(vgpu, offset, p_data, bytes); + case PCI_BASE_ADDRESS_0 ... PCI_BASE_ADDRESS_5: if (WARN_ON(!IS_ALIGNED(offset, 4))) return -EINVAL; @@ -375,6 +394,8 @@ void intel_vgpu_init_cfg_space(struct intel_vgpu *vgpu, pci_resource_len(gvt->dev_priv->drm.pdev, 0); vgpu->cfg_space.bar[INTEL_GVT_PCI_BAR_APERTURE].size = pci_resource_len(gvt->dev_priv->drm.pdev, 2); + + memset(vgpu_cfg_space(vgpu) + PCI_ROM_ADDRESS, 0, 4); } /** -- cgit From 365ad5df9caa1a7ebc73b8d70ab94bbf6a74268a Mon Sep 17 00:00:00 2001 From: Zhi Wang Date: Sat, 30 Sep 2017 17:42:20 +0800 Subject: drm/i915/gvt: Export intel_gvt_render_mmio_to_ring_id() Since many emulation logic needs to convert the offset of ring registers into ring id, we export it for other caller which might need it. Signed-off-by: Zhi Wang (cherry picked from commit 62a6a53786fc4b4e7543cc63b704dbb3f7df4c0f) --- drivers/gpu/drm/i915/gvt/handlers.c | 21 +++++++++++++++------ drivers/gpu/drm/i915/gvt/mmio.h | 2 ++ 2 files changed, 17 insertions(+), 6 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/i915/gvt/handlers.c b/drivers/gpu/drm/i915/gvt/handlers.c index 44cd5ff5e97d..55cbdb022924 100644 --- a/drivers/gpu/drm/i915/gvt/handlers.c +++ b/drivers/gpu/drm/i915/gvt/handlers.c @@ -137,17 +137,26 @@ static int new_mmio_info(struct intel_gvt *gvt, return 0; } -static int render_mmio_to_ring_id(struct intel_gvt *gvt, unsigned int reg) +/** + * intel_gvt_render_mmio_to_ring_id - convert a mmio offset into ring id + * @gvt: a GVT device + * @offset: register offset + * + * Returns: + * Ring ID on success, negative error code if failed. + */ +int intel_gvt_render_mmio_to_ring_id(struct intel_gvt *gvt, + unsigned int offset) { enum intel_engine_id id; struct intel_engine_cs *engine; - reg &= ~GENMASK(11, 0); + offset &= ~GENMASK(11, 0); for_each_engine(engine, gvt->dev_priv, id) { - if (engine->mmio_base == reg) + if (engine->mmio_base == offset) return id; } - return -1; + return -ENODEV; } #define offset_to_fence_num(offset) \ @@ -1409,7 +1418,7 @@ static int mmio_read_from_hw(struct intel_vgpu *vgpu, static int elsp_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, void *p_data, unsigned int bytes) { - int ring_id = render_mmio_to_ring_id(vgpu->gvt, offset); + int ring_id = intel_gvt_render_mmio_to_ring_id(vgpu->gvt, offset); struct intel_vgpu_execlist *execlist; u32 data = *(u32 *)p_data; int ret = 0; @@ -1436,7 +1445,7 @@ static int ring_mode_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, void *p_data, unsigned int bytes) { u32 data = *(u32 *)p_data; - int ring_id = render_mmio_to_ring_id(vgpu->gvt, offset); + int ring_id = intel_gvt_render_mmio_to_ring_id(vgpu->gvt, offset); bool enable_execlist; write_vreg(vgpu, offset, p_data, bytes); diff --git a/drivers/gpu/drm/i915/gvt/mmio.h b/drivers/gpu/drm/i915/gvt/mmio.h index 32cd64ddad26..dbc04ad2c7a1 100644 --- a/drivers/gpu/drm/i915/gvt/mmio.h +++ b/drivers/gpu/drm/i915/gvt/mmio.h @@ -65,6 +65,8 @@ struct intel_gvt_mmio_info { struct hlist_node node; }; +int intel_gvt_render_mmio_to_ring_id(struct intel_gvt *gvt, + unsigned int reg); unsigned long intel_gvt_get_device_type(struct intel_gvt *gvt); bool intel_gvt_match_device(struct intel_gvt *gvt, unsigned long device); -- cgit From 29f9e425973d3c7cdfc32725a4d7c4c2f537d5a9 Mon Sep 17 00:00:00 2001 From: Xiong Zhang Date: Tue, 7 Nov 2017 05:23:02 +0800 Subject: drm/i915/gvt: Limit read hw reg to active vgpu mmio_read_from_hw() let vgpu could read hw reg, if vgpu's workload is running on hw, things is good. Otherwise vgpu will get other vgpu's reg val, it is unsafe. This patch limit such hw access to active vgpu. If vgpu isn't running on hw, the reg read of this vgpu will get the last active val which saved at schedule_out. v2: ring timestamp is walking continuously even if the ring is idle. so read hw directly. (Zhenyu) Signed-off-by: Xiong Zhang Signed-off-by: Zhenyu Wang (cherry picked from commit 295764cd2ff41e2c1bc8af4050de77cec5e7a1c0) --- drivers/gpu/drm/i915/gvt/handlers.c | 26 ++++++++++++++++++++++---- drivers/gpu/drm/i915/gvt/scheduler.c | 15 +++++++++++++++ 2 files changed, 37 insertions(+), 4 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/i915/gvt/handlers.c b/drivers/gpu/drm/i915/gvt/handlers.c index 55cbdb022924..1f840f6b81bb 100644 --- a/drivers/gpu/drm/i915/gvt/handlers.c +++ b/drivers/gpu/drm/i915/gvt/handlers.c @@ -1407,11 +1407,29 @@ static int skl_lcpll_write(struct intel_vgpu *vgpu, unsigned int offset, static int mmio_read_from_hw(struct intel_vgpu *vgpu, unsigned int offset, void *p_data, unsigned int bytes) { - struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + struct intel_gvt *gvt = vgpu->gvt; + struct drm_i915_private *dev_priv = gvt->dev_priv; + int ring_id; + u32 ring_base; + + ring_id = intel_gvt_render_mmio_to_ring_id(gvt, offset); + /** + * Read HW reg in following case + * a. the offset isn't a ring mmio + * b. the offset's ring is running on hw. + * c. the offset is ring time stamp mmio + */ + if (ring_id >= 0) + ring_base = dev_priv->engine[ring_id]->mmio_base; + + if (ring_id < 0 || vgpu == gvt->scheduler.engine_owner[ring_id] || + offset == i915_mmio_reg_offset(RING_TIMESTAMP(ring_base)) || + offset == i915_mmio_reg_offset(RING_TIMESTAMP_UDW(ring_base))) { + mmio_hw_access_pre(dev_priv); + vgpu_vreg(vgpu, offset) = I915_READ(_MMIO(offset)); + mmio_hw_access_post(dev_priv); + } - mmio_hw_access_pre(dev_priv); - vgpu_vreg(vgpu, offset) = I915_READ(_MMIO(offset)); - mmio_hw_access_post(dev_priv); return intel_vgpu_default_mmio_read(vgpu, offset, p_data, bytes); } diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c index 3ac1dc97a7a0..a63787be1e6b 100644 --- a/drivers/gpu/drm/i915/gvt/scheduler.c +++ b/drivers/gpu/drm/i915/gvt/scheduler.c @@ -131,6 +131,20 @@ static inline bool is_gvt_request(struct drm_i915_gem_request *req) return i915_gem_context_force_single_submission(req->ctx); } +static void save_ring_hw_state(struct intel_vgpu *vgpu, int ring_id) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + u32 ring_base = dev_priv->engine[ring_id]->mmio_base; + i915_reg_t reg; + + reg = RING_INSTDONE(ring_base); + vgpu_vreg(vgpu, i915_mmio_reg_offset(reg)) = I915_READ_FW(reg); + reg = RING_ACTHD(ring_base); + vgpu_vreg(vgpu, i915_mmio_reg_offset(reg)) = I915_READ_FW(reg); + reg = RING_ACTHD_UDW(ring_base); + vgpu_vreg(vgpu, i915_mmio_reg_offset(reg)) = I915_READ_FW(reg); +} + static int shadow_context_status_change(struct notifier_block *nb, unsigned long action, void *data) { @@ -176,6 +190,7 @@ static int shadow_context_status_change(struct notifier_block *nb, break; case INTEL_CONTEXT_SCHEDULE_OUT: case INTEL_CONTEXT_SCHEDULE_PREEMPTED: + save_ring_hw_state(workload->vgpu, ring_id); atomic_set(&workload->shadow_ctx_active, 0); break; default: -- cgit From ac7688c0397442f1c9eef7069737089660322497 Mon Sep 17 00:00:00 2001 From: Zhenyu Wang Date: Fri, 1 Dec 2017 14:59:53 +0800 Subject: drm/i915/gvt: Don't mark vgpu context as inactive when preempted We shouldn't mark inactive for vGPU context if preempted, which would still be re-scheduled later. So keep active state. Fixes: d6c0511300dc ("drm/i915/execlists: Distinguish the incomplete context notifies") Signed-off-by: Zhenyu Wang (cherry picked from commit da5f99eaccc10e30bf82eb02b1be74703b878720) --- drivers/gpu/drm/i915/gvt/scheduler.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c index a63787be1e6b..ab3d041ee5f9 100644 --- a/drivers/gpu/drm/i915/gvt/scheduler.c +++ b/drivers/gpu/drm/i915/gvt/scheduler.c @@ -189,10 +189,12 @@ static int shadow_context_status_change(struct notifier_block *nb, atomic_set(&workload->shadow_ctx_active, 1); break; case INTEL_CONTEXT_SCHEDULE_OUT: - case INTEL_CONTEXT_SCHEDULE_PREEMPTED: save_ring_hw_state(workload->vgpu, ring_id); atomic_set(&workload->shadow_ctx_active, 0); break; + case INTEL_CONTEXT_SCHEDULE_PREEMPTED: + save_ring_hw_state(workload->vgpu, ring_id); + break; default: WARN_ON(1); return NOTIFY_OK; -- cgit From 11474e9091cf2002e948647fd9f63a7f027e488a Mon Sep 17 00:00:00 2001 From: Zhenyu Wang Date: Mon, 4 Dec 2017 10:42:58 +0800 Subject: drm/i915/gvt: set max priority for gvt context This is to workaround guest driver hang regression after preemption enable that gvt hasn't enabled handling of that for guest workload. So in effect this disables preemption for gvt context now. Signed-off-by: Zhenyu Wang (cherry picked from commit 1603660b3342269c95fcafee1945790342a8c28e) --- drivers/gpu/drm/i915/gvt/scheduler.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers') diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c index ab3d041ee5f9..69f8f0d155b9 100644 --- a/drivers/gpu/drm/i915/gvt/scheduler.c +++ b/drivers/gpu/drm/i915/gvt/scheduler.c @@ -757,6 +757,9 @@ int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu) if (IS_ERR(vgpu->shadow_ctx)) return PTR_ERR(vgpu->shadow_ctx); + if (INTEL_INFO(vgpu->gvt->dev_priv)->has_logical_ring_preemption) + vgpu->shadow_ctx->priority = INT_MAX; + vgpu->shadow_ctx->engine[RCS].initialised = true; bitmap_zero(vgpu->shadow_ctx_desc_updated, I915_NUM_ENGINES); -- cgit