Merge tag 'drm-fixes-2019-11-22' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Two sets of fixes in here, one for amdgpu, and one for i915.

  The amdgpu ones are pretty small, i915's CI system seems to have a few
  problems in the last week or so, there is one major regression fix for
  fb_mmap, but there are a bunch of other issues fixed in there as well,
  oops, screen flashes and rcu related.

  amdgpu:
   - Remove experimental flag for navi14
   - Fix confusing power message failures on older VI parts
   - Hang fix for gfxoff when using the read register interface
   - Two stability regression fixes for Raven

  i915:
   - Fix kernel oops on dumb_create ioctl on no crtc situation
   - Fix bad ugly colored flash on VLV/CHV related to gamma LUT update
   - Fix unity of the frequencies reported on PMU
   - Fix kernel oops on set_page_dirty using better locks around it
   - Protect the request pointer with RCU to prevent it being freed
     while we might need still
   - Make pool objects read-only
   - Restore physical addresses for fb_map to avoid corrupted page
     table"

* tag 'drm-fixes-2019-11-22' of git://anongit.freedesktop.org/drm/drm:
  drm/i915/fbdev: Restore physical addresses for fb_mmap()
  Revert "drm/amd/display: enable S/G for RAVEN chip"
  drm/amdgpu: disable gfxoff on original raven
  drm/amdgpu: disable gfxoff when using register read interface
  drm/amd/powerplay: correct fine grained dpm force level setting
  drm/amd/powerplay: issue no PPSMC_MSG_GetCurrPkgPwr on unsupported ASICs
  drm/amdgpu: remove experimental flag for Navi14
  drm/i915: make pool objects read-only
  drm/i915: Protect request peeking with RCU
  drm/i915/userptr: Try to acquire the page lock around set_page_dirty()
  drm/i915/pmu: "Frequency" is reported as accumulated cycles
  drm/i915: Preload LUTs if the hw isn't currently using them
  drm/i915: Don't oops in dumb_create ioctl if we have no crtcs

16 files changed, 183 insertions, 32 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
index 1d4aaa9580f4..82efc1e22e61 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
@@ -511,7 +511,7 @@ uint32_t amdgpu_display_supported_domains(struct amdgpu_device *adev,
 	 * Also, don't allow GTT domain if the BO doens't have USWC falg set.
 	 */
 	if (adev->asic_type >= CHIP_CARRIZO &&
-	    adev->asic_type <= CHIP_RAVEN &&
+	    adev->asic_type < CHIP_RAVEN &&
 	    (adev->flags & AMD_IS_APU) &&
 	    (bo_flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) &&
 	    amdgpu_bo_support_uswc(bo_flags) &&
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index e1c15721611a..b19157b19fa0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -1013,10 +1013,10 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x731B, 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|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x7341, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x7347, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14|AMD_EXP_HW_SUPPORT},
-	{0x1002, 0x734F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14|AMD_EXP_HW_SUPPORT},
+	{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, 0x1636, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RENOIR|AMD_IS_APU|AMD_EXP_HW_SUPPORT},
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index a042ef471fbd..a73206784cba 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -649,15 +649,19 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			return -ENOMEM;
 		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);
+				amdgpu_gfx_off_ctrl(adev, true);
 				return -EFAULT;
 			}
+		}
+		amdgpu_gfx_off_ctrl(adev, true);
 		n = copy_to_user(out, regs, min(size, alloc_size));
 		kfree(regs);
 		return n ? -EFAULT : 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index dfca83a2de47..97cf0b536873 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -1038,8 +1038,13 @@ static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev)
 	case CHIP_VEGA20:
 		break;
 	case CHIP_RAVEN:
-		if (!(adev->rev_id >= 0x8 || adev->pdev->device == 0x15d8)
-			&&((adev->gfx.rlc_fw_version != 106 &&
+		/* Disable GFXOFF on original raven.  There are combinations
+		 * of sbios and platforms that are not stable.
+		 */
+		if (!(adev->rev_id >= 0x8 || adev->pdev->device == 0x15d8))
+			adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
+		else if (!(adev->rev_id >= 0x8 || adev->pdev->device == 0x15d8)
+			 &&((adev->gfx.rlc_fw_version != 106 &&
 			     adev->gfx.rlc_fw_version < 531) ||
 			    (adev->gfx.rlc_fw_version == 53815) ||
 			    (adev->gfx.rlc_feature_version < 1) ||
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index a52f0b13a2c8..4139f129eafb 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -688,7 +688,7 @@ static int amdgpu_dm_init(struct amdgpu_device *adev)
 	 */
 	if (adev->flags & AMD_IS_APU &&
 	    adev->asic_type >= CHIP_CARRIZO &&
-	    adev->asic_type <= CHIP_RAVEN)
+	    adev->asic_type < CHIP_RAVEN)
 		init_data.flags.gpu_vm_support = true;
 
 	if (amdgpu_dc_feature_mask & DC_FBC_MASK)
diff --git a/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_hwmgr.c b/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_hwmgr.c
index 34f95e0e3ea4..203ce4b1028f 100644
--- a/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_hwmgr.c
+++ b/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_hwmgr.c
@@ -3478,18 +3478,31 @@ static int smu7_get_pp_table_entry(struct pp_hwmgr *hwmgr,
 
 static int smu7_get_gpu_power(struct pp_hwmgr *hwmgr, u32 *query)
 {
+	struct amdgpu_device *adev = hwmgr->adev;
 	int i;
 	u32 tmp = 0;
 
 	if (!query)
 		return -EINVAL;
 
-	smum_send_msg_to_smc_with_parameter(hwmgr, PPSMC_MSG_GetCurrPkgPwr, 0);
-	tmp = cgs_read_register(hwmgr->device, mmSMC_MSG_ARG_0);
-	*query = tmp;
+	/*
+	 * PPSMC_MSG_GetCurrPkgPwr is not supported on:
+	 *  - Hawaii
+	 *  - Bonaire
+	 *  - Fiji
+	 *  - Tonga
+	 */
+	if ((adev->asic_type != CHIP_HAWAII) &&
+	    (adev->asic_type != CHIP_BONAIRE) &&
+	    (adev->asic_type != CHIP_FIJI) &&
+	    (adev->asic_type != CHIP_TONGA)) {
+		smum_send_msg_to_smc_with_parameter(hwmgr, PPSMC_MSG_GetCurrPkgPwr, 0);
+		tmp = cgs_read_register(hwmgr->device, mmSMC_MSG_ARG_0);
+		*query = tmp;
 
-	if (tmp != 0)
-		return 0;
+		if (tmp != 0)
+			return 0;
+	}
 
 	smum_send_msg_to_smc(hwmgr, PPSMC_MSG_PmStatusLogStart);
 	cgs_write_ind_register(hwmgr->device, CGS_IND_REG__SMC,
diff --git a/drivers/gpu/drm/amd/powerplay/navi10_ppt.c b/drivers/gpu/drm/amd/powerplay/navi10_ppt.c
index 3ec5a10a7c4d..328e258a6895 100644
--- a/drivers/gpu/drm/amd/powerplay/navi10_ppt.c
+++ b/drivers/gpu/drm/amd/powerplay/navi10_ppt.c
@@ -759,6 +759,12 @@ static int navi10_force_clk_levels(struct smu_context *smu,
 	case SMU_UCLK:
 	case SMU_DCEFCLK:
 	case SMU_FCLK:
+		/* There is only 2 levels for fine grained DPM */
+		if (navi10_is_support_fine_grained_dpm(smu, clk_type)) {
+			soft_max_level = (soft_max_level >= 1 ? 1 : 0);
+			soft_min_level = (soft_min_level >= 1 ? 1 : 0);
+		}
+
 		ret = smu_get_dpm_freq_by_index(smu, clk_type, soft_min_level, &min_freq);
 		if (ret)
 			return size;
diff --git a/drivers/gpu/drm/i915/display/intel_atomic.c b/drivers/gpu/drm/i915/display/intel_atomic.c
index d3fb75bb9eb1..7cb2257bbb93 100644
--- a/drivers/gpu/drm/i915/display/intel_atomic.c
+++ b/drivers/gpu/drm/i915/display/intel_atomic.c
@@ -201,6 +201,7 @@ intel_crtc_duplicate_state(struct drm_crtc *crtc)
 	crtc_state->update_wm_post = false;
 	crtc_state->fb_changed = false;
 	crtc_state->fifo_changed = false;
+	crtc_state->preload_luts = false;
 	crtc_state->wm.need_postvbl_update = false;
 	crtc_state->fb_bits = 0;
 	crtc_state->update_planes = 0;
diff --git a/drivers/gpu/drm/i915/display/intel_color.c b/drivers/gpu/drm/i915/display/intel_color.c
index 71a0201437a9..aa1e2c670bc4 100644
--- a/drivers/gpu/drm/i915/display/intel_color.c
+++ b/drivers/gpu/drm/i915/display/intel_color.c
@@ -990,6 +990,55 @@ void intel_color_commit(const struct intel_crtc_state *crtc_state)
 	dev_priv->display.color_commit(crtc_state);
 }
 
+static bool intel_can_preload_luts(const struct intel_crtc_state *new_crtc_state)
+{
+	struct intel_crtc *crtc = to_intel_crtc(new_crtc_state->base.crtc);
+	struct intel_atomic_state *state =
+		to_intel_atomic_state(new_crtc_state->base.state);
+	const struct intel_crtc_state *old_crtc_state =
+		intel_atomic_get_old_crtc_state(state, crtc);
+
+	return !old_crtc_state->base.gamma_lut &&
+		!old_crtc_state->base.degamma_lut;
+}
+
+static bool chv_can_preload_luts(const struct intel_crtc_state *new_crtc_state)
+{
+	struct intel_crtc *crtc = to_intel_crtc(new_crtc_state->base.crtc);
+	struct intel_atomic_state *state =
+		to_intel_atomic_state(new_crtc_state->base.state);
+	const struct intel_crtc_state *old_crtc_state =
+		intel_atomic_get_old_crtc_state(state, crtc);
+
+	/*
+	 * CGM_PIPE_MODE is itself single buffered. We'd have to
+	 * somehow split it out from chv_load_luts() if we wanted
+	 * the ability to preload the CGM LUTs/CSC without tearing.
+	 */
+	if (old_crtc_state->cgm_mode || new_crtc_state->cgm_mode)
+		return false;
+
+	return !old_crtc_state->base.gamma_lut;
+}
+
+static bool glk_can_preload_luts(const struct intel_crtc_state *new_crtc_state)
+{
+	struct intel_crtc *crtc = to_intel_crtc(new_crtc_state->base.crtc);
+	struct intel_atomic_state *state =
+		to_intel_atomic_state(new_crtc_state->base.state);
+	const struct intel_crtc_state *old_crtc_state =
+		intel_atomic_get_old_crtc_state(state, crtc);
+
+	/*
+	 * The hardware degamma is active whenever the pipe
+	 * CSC is active. Thus even if the old state has no
+	 * software degamma we need to avoid clobbering the
+	 * linear hardware degamma mid scanout.
+	 */
+	return !old_crtc_state->csc_enable &&
+		!old_crtc_state->base.gamma_lut;
+}
+
 int intel_color_check(struct intel_crtc_state *crtc_state)
 {
 	struct drm_i915_private *dev_priv = to_i915(crtc_state->base.crtc->dev);
@@ -1133,6 +1182,8 @@ static int i9xx_color_check(struct intel_crtc_state *crtc_state)
 	if (ret)
 		return ret;
 
+	crtc_state->preload_luts = intel_can_preload_luts(crtc_state);
+
 	return 0;
 }
 
@@ -1185,6 +1236,8 @@ static int chv_color_check(struct intel_crtc_state *crtc_state)
 	if (ret)
 		return ret;
 
+	crtc_state->preload_luts = chv_can_preload_luts(crtc_state);
+
 	return 0;
 }
 
@@ -1224,6 +1277,8 @@ static int ilk_color_check(struct intel_crtc_state *crtc_state)
 	if (ret)
 		return ret;
 
+	crtc_state->preload_luts = intel_can_preload_luts(crtc_state);
+
 	return 0;
 }
 
@@ -1281,6 +1336,8 @@ static int ivb_color_check(struct intel_crtc_state *crtc_state)
 	if (ret)
 		return ret;
 
+	crtc_state->preload_luts = intel_can_preload_luts(crtc_state);
+
 	return 0;
 }
 
@@ -1319,6 +1376,8 @@ static int glk_color_check(struct intel_crtc_state *crtc_state)
 	if (ret)
 		return ret;
 
+	crtc_state->preload_luts = glk_can_preload_luts(crtc_state);
+
 	return 0;
 }
 
@@ -1368,6 +1427,8 @@ static int icl_color_check(struct intel_crtc_state *crtc_state)
 
 	crtc_state->csc_mode = icl_csc_mode(crtc_state);
 
+	crtc_state->preload_luts = intel_can_preload_luts(crtc_state);
+
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
index dfff6f4357b8..af50f05f4e9d 100644
--- a/drivers/gpu/drm/i915/display/intel_display.c
+++ b/drivers/gpu/drm/i915/display/intel_display.c
@@ -2504,6 +2504,9 @@ u32 intel_plane_fb_max_stride(struct drm_i915_private *dev_priv,
 	 * the highest stride limits of them all.
 	 */
 	crtc = intel_get_crtc_for_pipe(dev_priv, PIPE_A);
+	if (!crtc)
+		return 0;
+
 	plane = to_intel_plane(crtc->base.primary);
 
 	return plane->max_stride(plane, pixel_format, modifier,
@@ -13740,6 +13743,11 @@ static void intel_update_crtc(struct intel_crtc *crtc,
 		/* vblanks work again, re-enable pipe CRC. */
 		intel_crtc_enable_pipe_crc(crtc);
 	} else {
+		if (new_crtc_state->preload_luts &&
+		    (new_crtc_state->base.color_mgmt_changed ||
+		     new_crtc_state->update_pipe))
+			intel_color_load_luts(new_crtc_state);
+
 		intel_pre_plane_update(old_crtc_state, new_crtc_state);
 
 		if (new_crtc_state->update_pipe)
@@ -14034,6 +14042,7 @@ static void intel_atomic_commit_tail(struct intel_atomic_state *state)
 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
 		if (new_crtc_state->base.active &&
 		    !needs_modeset(new_crtc_state) &&
+		    !new_crtc_state->preload_luts &&
 		    (new_crtc_state->base.color_mgmt_changed ||
 		     new_crtc_state->update_pipe))
 			intel_color_load_luts(new_crtc_state);
diff --git a/drivers/gpu/drm/i915/display/intel_display_types.h b/drivers/gpu/drm/i915/display/intel_display_types.h
index 449abaea619f..4075b0387c87 100644
--- a/drivers/gpu/drm/i915/display/intel_display_types.h
+++ b/drivers/gpu/drm/i915/display/intel_display_types.h
@@ -761,6 +761,7 @@ struct intel_crtc_state {
 	bool update_wm_pre, update_wm_post; /* watermarks are updated */
 	bool fb_changed; /* fb on any of the planes is changed */
 	bool fifo_changed; /* FIFO split is changed */
+	bool preload_luts;
 
 	/* Pipe source size (ie. panel fitter input size)
 	 * All planes will be positioned inside this space,
diff --git a/drivers/gpu/drm/i915/display/intel_fbdev.c b/drivers/gpu/drm/i915/display/intel_fbdev.c
index d59eee5c5d9c..b5c588e511dd 100644
--- a/drivers/gpu/drm/i915/display/intel_fbdev.c
+++ b/drivers/gpu/drm/i915/display/intel_fbdev.c
@@ -235,6 +235,11 @@ static int intelfb_create(struct drm_fb_helper *helper,
 	info->apertures->ranges[0].base = ggtt->gmadr.start;
 	info->apertures->ranges[0].size = ggtt->mappable_end;
 
+	/* Our framebuffer is the entirety of fbdev's system memory */
+	info->fix.smem_start =
+		(unsigned long)(ggtt->gmadr.start + vma->node.start);
+	info->fix.smem_len = vma->node.size;
+
 	vaddr = i915_vma_pin_iomap(vma);
 	if (IS_ERR(vaddr)) {
 		DRM_ERROR("Failed to remap framebuffer into virtual memory\n");
@@ -244,10 +249,6 @@ static int intelfb_create(struct drm_fb_helper *helper,
 	info->screen_base = vaddr;
 	info->screen_size = vma->node.size;
 
-	/* Our framebuffer is the entirety of fbdev's system memory */
-	info->fix.smem_start = (unsigned long)info->screen_base;
-	info->fix.smem_len = info->screen_size;
-
 	drm_fb_helper_fill_info(info, &ifbdev->helper, sizes);
 
 	/* If the object is shmemfs backed, it will have given us zeroed pages.
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
index 6b3b50f0f6d9..abfbac49b8e8 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
@@ -671,8 +671,28 @@ i915_gem_userptr_put_pages(struct drm_i915_gem_object *obj,
 		obj->mm.dirty = false;
 
 	for_each_sgt_page(page, sgt_iter, pages) {
-		if (obj->mm.dirty)
+		if (obj->mm.dirty && trylock_page(page)) {
+			/*
+			 * As this may not be anonymous memory (e.g. shmem)
+			 * but exist on a real mapping, we have to lock
+			 * the page in order to dirty it -- holding
+			 * the page reference is not sufficient to
+			 * prevent the inode from being truncated.
+			 * Play safe and take the lock.
+			 *
+			 * However...!
+			 *
+			 * The mmu-notifier can be invalidated for a
+			 * migrate_page, that is alreadying holding the lock
+			 * on the page. Such a try_to_unmap() will result
+			 * in us calling put_pages() and so recursively try
+			 * to lock the page. We avoid that deadlock with
+			 * a trylock_page() and in exchange we risk missing
+			 * some page dirtying.
+			 */
 			set_page_dirty(page);
+			unlock_page(page);
+		}
 
 		mark_page_accessed(page);
 		put_page(page);
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pool.c b/drivers/gpu/drm/i915/gt/intel_engine_pool.c
index 4cd54c569911..379a91780bd4 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_pool.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_pool.c
@@ -103,6 +103,8 @@ node_create(struct intel_engine_pool *pool, size_t sz)
 		return ERR_CAST(obj);
 	}
 
+	i915_gem_object_set_readonly(obj);
+
 	node->obj = obj;
 	return node;
 }
diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
index 8e251e719390..212acaef581e 100644
--- a/drivers/gpu/drm/i915/i915_pmu.c
+++ b/drivers/gpu/drm/i915/i915_pmu.c
@@ -843,8 +843,8 @@ create_event_attributes(struct i915_pmu *pmu)
 		const char *name;
 		const char *unit;
 	} events[] = {
-		__event(I915_PMU_ACTUAL_FREQUENCY, "actual-frequency", "MHz"),
-		__event(I915_PMU_REQUESTED_FREQUENCY, "requested-frequency", "MHz"),
+		__event(I915_PMU_ACTUAL_FREQUENCY, "actual-frequency", "M"),
+		__event(I915_PMU_REQUESTED_FREQUENCY, "requested-frequency", "M"),
 		__event(I915_PMU_INTERRUPTS, "interrupts", NULL),
 		__event(I915_PMU_RC6_RESIDENCY, "rc6-residency", "ns"),
 	};
diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c
index 7b84ebca2901..3eba8a2b39c2 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -177,9 +177,37 @@ static inline int rq_prio(const struct i915_request *rq)
 	return rq->sched.attr.priority | __NO_PREEMPTION;
 }
 
-static void kick_submission(struct intel_engine_cs *engine, int prio)
+static inline bool need_preempt(int prio, int active)
 {
-	const struct i915_request *inflight = *engine->execlists.active;
+	/*
+	 * Allow preemption of low -> normal -> high, but we do
+	 * not allow low priority tasks to preempt other low priority
+	 * tasks under the impression that latency for low priority
+	 * tasks does not matter (as much as background throughput),
+	 * so kiss.
+	 */
+	return prio >= max(I915_PRIORITY_NORMAL, active);
+}
+
+static void kick_submission(struct intel_engine_cs *engine,
+			    const struct i915_request *rq,
+			    int prio)
+{
+	const struct i915_request *inflight;
+
+	/*
+	 * We only need to kick the tasklet once for the high priority
+	 * new context we add into the queue.
+	 */
+	if (prio <= engine->execlists.queue_priority_hint)
+		return;
+
+	rcu_read_lock();
+
+	/* Nothing currently active? We're overdue for a submission! */
+	inflight = execlists_active(&engine->execlists);
+	if (!inflight)
+		goto unlock;
 
 	/*
 	 * If we are already the currently executing context, don't
@@ -188,10 +216,15 @@ static void kick_submission(struct intel_engine_cs *engine, int prio)
 	 * tasklet, i.e. we have not change the priority queue
 	 * sufficiently to oust the running context.
 	 */
-	if (!inflight || !i915_scheduler_need_preempt(prio, rq_prio(inflight)))
-		return;
+	if (inflight->hw_context == rq->hw_context)
+		goto unlock;
 
-	tasklet_hi_schedule(&engine->execlists.tasklet);
+	engine->execlists.queue_priority_hint = prio;
+	if (need_preempt(prio, rq_prio(inflight)))
+		tasklet_hi_schedule(&engine->execlists.tasklet);
+
+unlock:
+	rcu_read_unlock();
 }
 
 static void __i915_schedule(struct i915_sched_node *node,
@@ -317,13 +350,8 @@ static void __i915_schedule(struct i915_sched_node *node,
 			list_move_tail(&node->link, cache.priolist);
 		}
 
-		if (prio <= engine->execlists.queue_priority_hint)
-			continue;
-
-		engine->execlists.queue_priority_hint = prio;
-
 		/* Defer (tasklet) submission until after all of our updates. */
-		kick_submission(engine, prio);
+		kick_submission(engine, node_to_request(node), prio);
 	}
 
 	spin_unlock(&engine->active.lock);