Merge tag 'drm-next-2023-02-23' of git://anongit.freedesktop.org/drm/drm

Pull drm updates from Dave Airlie:
 "There are a bunch of changes all over in the usual places.

  Highlights:

   - habanalabs moves from misc to accel

   - first accel driver for Intel VPU (Versatile Processing Unit)
     inference engine

   - dropped all the ancient legacy DRI1 drivers. I think it's been at
     least 10 years since anyone has heard about these.

   - Intel DG2 updates and prelim Meteorlake enablement

   - etnaviv adds support for Versilicon NPU device (a GPU like engine
     with inference accelerators)

  Detailed summary:

  Removals:
   - remove legacy dri1 drivers: i810, mga, r128, savage, sis, tdfx, via

  New driver:
   - intel VPU accelerator driver
   - habanalabs comes via drm tree now

  drm/core:
   - use drm_dbg_ helpers in several places
   - Document defaults for CRTC backgrounds
   - Document use of drm_minor

  edid:
   - improve mode parsing and refactoring

  connector:
   - support analog TV mode property

  media:
   - add some common formats

  udmabuf:
   - add vmap/vunmap methods

  fourcc:
   - add XRGB1555 and RGB565 formats
   - document open source user waiver

  firmware:
   - fix color-format selection for system framebuffer

  format-helper:
   - Add conversion from XRGB8888 to various sysfb formats
   - Make XRGB8888 the only driver-emulated legacy format
   - Add conversion from XRGB8888 to XBGR8888 and ABGR8888

  fb-helper:
   - fix preferred depth and bpp values across drivers
   - Avoid blank consoles from selecting an incorrect color format

  probe-helper:
   - Enable/disable HPD on connectors

  scheduler:
   - Fix lockup in drm_sched_entity_kill()
   - Deprecate drm_sched_resubmit_jobs()

  bridge:
   - remove unused functions
   - implement i2c probe_new in various drivers
   - ite-it6505: Locking fixes, Cache EDID data
   - ite-it66121: Support IT6610 chip
   - lontium-tl9611: Fix HDMI on DragonBoard 845c
   - parade-ps8640: Use atomic bridge functions
   - Support i.MX93 LDB plus DT bindings

  debugfs:
   - add per device helpers and convert drivers

  displayport:
   - mst fixes
   - add DP adaptive sync DPCD definitions

  fbdev:
   - always pick 32bpp as default
   - remove some unused code

  simpledrm:
   - support system memory framebuffers

  panel:
   - add orientation quirks for Lenovo Yoga Tab 3 X90F and DynaBook K50
   - Use ktime_get_boottime() to measure power-down delay
   - Fix auto-suspend delay
   - Visionox VTDR6130 AMOLED DSI
   - Support Himax HX8394
   - Convert many drivers to common generic DSI write-sequence helper
   - AUO A030JTN01

  ttm:
   - drop bo wait wrapper
   - fix MIPS build

  habanalabs:
   - moved driver to accel subsystem
   - gaudi2 decoder error improvement
   - more trace events
   - Gaudi2 abrupt reset by firmware support
   - add uAPI to flush memory transactions
   - add uAPI to pass through userspace reqs to fw
   - remove dma-buf export by handle

  amdgpu:
   - add new INFO queries for peak and min sclk/mclk for profile modes
   - Add PCIe info to the INFO IOCTL
   - secure display support for multiple displays
   - DML optimizations
   - DCN 3.2 updates
   - PSR updates
   - DP 2.1 updates
   - SR-IOV RAS updates
   - VCN RAS support
   - SMU 13.x updates
   - Switch 1 element arrays to flexible arrays
   - Add RAS support for DF 4.3
   - Stack size improvements
   - S0ix rework
   - Allow 0 as a vram limit on APUs
   - Handle profiling modes for SMU13.x
   - Fix possible segfault in failure case
   - Rework FW requests to happen in early_init for all IPs so that we
     don't lose the sbios console if FW is missing
   - Fix power reporting on certain firmwares for CZN/RN
   - Allow S0ix without BIOS support
   - Enable freesync over PCon
   - Re-enable the AGP aperture on GMC 11.x

  amdkfd:
   - Error handling fixes
   - PASID fixes
   - Fix for cleared VRAM BOs
   - Fix cleanup if GPUVM creation fails
   - Memory accounting fix
   - Use resource_size rather than open codeing it
   - GC11 mGPU fix

  radeon:
   - Switch 1 element arrays to flexible arrays
   - Fix memory leak on shutdown
   - move to new logging

  i915:
   - Meteorlake display/OA/GSC fw/workarounds enabling
   - DP MST DSC support
   - Gamma/degamma readout support for the state checker
   - Enable SDP split support for DP 2.0
   - Add probe blocking support to i915.force_probe parameter
   - Enable Xe HP 4tile support
   - Avoid display direct calls to uncore
   - Fix HuC delayed load memory leaks
   - Add DG2 workarounds Wa_18018764978 and Wa_18019271663
   - Improve suspend / resume times with VT-d scanout workaround active
   - Fix DG2 visual corruption on small BAR systems by not forgetting to
     copy CCS aux state
   - Fix TLB invalidation for Gen12.50 video and compute engines
   - Enable HF-EEODB by switching HDMI, DP and LVDS to use struct
     drm_edid
   - Start using unversioned DMC firmware paths for new platforms
   - ELD refactor: Stop using hardware buffer, precompute ELD
   - lots of display code refactoring

  nouveau:
   - drop legacy ioctl support
   - replace 0-sized array

  msm:
   - dpu/dsi/mdss: Support for SM8350, SM8450 SM8550 and SC8280XP platform
   - Added bindings for SM8150
   - dpu: Partial support for DSC on SM8150 and SM8250
   - dpu: Fixed color transformation matrix being lost on suspend/resume
   - dp: Support SDM845 and SC8280XP platforms
   - dp: Support for limiting DP link rate via DT property
   - dsi: Validate display modes according to the DSI OPP table
   - dsi: DSI PHY support for the SM6375 platform
   - Add MSM_SUBMIT_BO_NO_IMPLICI
   - a2xx: Support to load legacy firmware
   - a6xx: GPU devcore dump updates for a650/a660
   - GPU devfreq tuning and fixes
   - Turn 8960 HDMI PHY into clock provider,
   - Make 8960 HDMI PHY use PXO clock from DT

  etnaviv:
   - experimental versilicon NPU support
   - report GPU load via fdinfo format
   - MMU fault message improvements

  tegra:
   - rework syncpoint interrupt

  mediatek:
   - DSI timing fix
   - fix config deps

  ast:
   - various fixes

  exynos:
   - restore bridge chain order fixes

  gud:
   - convert to shadow plane buffers
   - perform flushing synchronously during atomic update
   - Use new debugfs helpers

  arm/hdlcd:
   - Use new debugfs helper

  ili9486:
   - Support 16-bit pixel data

  imx:
   - Split off IPUv3 driver

  mipi-dbi:
   - convert to DRM shadow-plane helpers
   - rsp driver changes
   - Support separate I/O-voltage supply

  mxsfb:
   - Depend on ARCH_MXS or ARCH_MXC

  sun4i:
   - convert to new TV mode property

  vc4:
   - convert to new TV mode property
   - kunit tests
   - Support RGB565 and RGB666 formats
   - convert dsi driver to bridge
   - Various HVS an CRTC fixes

  v3d:
   - Do not opencode drm_gem_object_lookup()

  virtio:
   - improve tracing

  vkms:
   - support small cursors in IGT tests
   - Fix SEGFAULT from incorrect GEM-buffer mapping

  rcar-du:
   - fixes and improvements"

* tag 'drm-next-2023-02-23' of git://anongit.freedesktop.org/drm/drm: (1455 commits)
  msm/fbdev: fix unused variable warning with clang.
  drm/fb-helper: Remove drm_fb_helper_unprepare() from drm_fb_helper_fini()
  dma-buf: make kobj_type structure constant
  drm/shmem-helper: Fix locking for drm_gem_shmem_get_pages_sgt()
  drm/amd/display: disable SubVP + DRR to prevent underflow
  drm/amd/display: Fail atomic_check early on normalize_zpos error
  drm/amd/pm: avoid unaligned access warnings
  drm/amd/display: avoid unaligned access warnings
  drm/amd/display: Remove duplicate/repeating expressions
  drm/amd/display: Remove duplicate/repeating expression
  drm/amd/display: Make variables declaration inside ifdef guard
  drm/amd/display: Fix excess arguments on kernel-doc
  drm/amd/display: Add previously missing includes
  drm/amd/amdgpu: Add function prototypes to headers
  drm/amd/display: Add function prototypes to headers
  drm/amd/display: Turn global functions into static
  drm/amd/display: remove unused _calculate_degamma_curve function
  drm/amd/display: remove unused func declaration from resource headers
  drm/amd/display: unset initial value for tf since it's never used
  drm/amd/display: camel case cleanup in color_gamma file
  ...

1 files changed, 153 insertions, 62 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index ad490c1e2f57..6e543558386d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -706,13 +706,23 @@ static int __amdgpu_ras_feature_enable(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int amdgpu_ras_check_feature_allowed(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	if (amdgpu_ras_is_feature_allowed(adev, head) ||
+		amdgpu_ras_is_poison_mode_supported(adev))
+		return 1;
+	else
+		return 0;
+}
+
 /* wrapper of psp_ras_enable_features */
 int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
 		struct ras_common_if *head, bool enable)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	union ta_ras_cmd_input *info;
-	int ret;
+	int ret = 0;
 
 	if (!con)
 		return -EINVAL;
@@ -736,7 +746,8 @@ int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
 	}
 
 	/* Do not enable if it is not allowed. */
-	WARN_ON(enable && !amdgpu_ras_is_feature_allowed(adev, head));
+	if (enable && !amdgpu_ras_check_feature_allowed(adev, head))
+		goto out;
 
 	/* Only enable ras feature operation handle on host side */
 	if (head->block == AMDGPU_RAS_BLOCK__GFX &&
@@ -754,7 +765,6 @@ int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
 
 	/* setup the obj */
 	__amdgpu_ras_feature_enable(adev, head, enable);
-	ret = 0;
 out:
 	if (head->block == AMDGPU_RAS_BLOCK__GFX)
 		kfree(info);
@@ -910,9 +920,6 @@ static struct amdgpu_ras_block_object *amdgpu_ras_get_ras_block(struct amdgpu_de
 	if (block >= AMDGPU_RAS_BLOCK__LAST)
 		return NULL;
 
-	if (!amdgpu_ras_is_supported(adev, block))
-		return NULL;
-
 	list_for_each_entry_safe(node, tmp, &adev->ras_list, node) {
 		if (!node->ras_obj) {
 			dev_warn(adev->dev, "Warning: abnormal ras list node.\n");
@@ -1087,6 +1094,10 @@ int amdgpu_ras_error_inject(struct amdgpu_device *adev,
 							info->head.block,
 							info->head.sub_block_index);
 
+	/* inject on guest isn't allowed, return success directly */
+	if (amdgpu_sriov_vf(adev))
+		return 0;
+
 	if (!obj)
 		return -EINVAL;
 
@@ -1122,11 +1133,54 @@ int amdgpu_ras_error_inject(struct amdgpu_device *adev,
 }
 
 /**
- * amdgpu_ras_query_error_count -- Get error counts of all IPs
+ * amdgpu_ras_query_error_count_helper -- Get error counter for specific IP
+ * @adev: pointer to AMD GPU device
+ * @ce_count: pointer to an integer to be set to the count of correctible errors.
+ * @ue_count: pointer to an integer to be set to the count of uncorrectible errors.
+ * @query_info: pointer to ras_query_if
+ *
+ * Return 0 for query success or do nothing, otherwise return an error
+ * on failures
+ */
+static int amdgpu_ras_query_error_count_helper(struct amdgpu_device *adev,
+					       unsigned long *ce_count,
+					       unsigned long *ue_count,
+					       struct ras_query_if *query_info)
+{
+	int ret;
+
+	if (!query_info)
+		/* do nothing if query_info is not specified */
+		return 0;
+
+	ret = amdgpu_ras_query_error_status(adev, query_info);
+	if (ret)
+		return ret;
+
+	*ce_count += query_info->ce_count;
+	*ue_count += query_info->ue_count;
+
+	/* some hardware/IP supports read to clear
+	 * no need to explictly reset the err status after the query call */
+	if (adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 2) &&
+	    adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 4)) {
+		if (amdgpu_ras_reset_error_status(adev, query_info->head.block))
+			dev_warn(adev->dev,
+				 "Failed to reset error counter and error status\n");
+	}
+
+	return 0;
+}
+
+/**
+ * amdgpu_ras_query_error_count -- Get error counts of all IPs or specific IP
  * @adev: pointer to AMD GPU device
  * @ce_count: pointer to an integer to be set to the count of correctible errors.
  * @ue_count: pointer to an integer to be set to the count of uncorrectible
  * errors.
+ * @query_info: pointer to ras_query_if if the query request is only for
+ * specific ip block; if info is NULL, then the qurey request is for
+ * all the ip blocks that support query ras error counters/status
  *
  * If set, @ce_count or @ue_count, count and return the corresponding
  * error counts in those integer pointers. Return 0 if the device
@@ -1134,11 +1188,13 @@ int amdgpu_ras_error_inject(struct amdgpu_device *adev,
  */
 int amdgpu_ras_query_error_count(struct amdgpu_device *adev,
 				 unsigned long *ce_count,
-				 unsigned long *ue_count)
+				 unsigned long *ue_count,
+				 struct ras_query_if *query_info)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_manager *obj;
 	unsigned long ce, ue;
+	int ret;
 
 	if (!adev->ras_enabled || !con)
 		return -EOPNOTSUPP;
@@ -1150,26 +1206,23 @@ int amdgpu_ras_query_error_count(struct amdgpu_device *adev,
 
 	ce = 0;
 	ue = 0;
-	list_for_each_entry(obj, &con->head, node) {
-		struct ras_query_if info = {
-			.head = obj->head,
-		};
-		int res;
-
-		res = amdgpu_ras_query_error_status(adev, &info);
-		if (res)
-			return res;
+	if (!query_info) {
+		/* query all the ip blocks that support ras query interface */
+		list_for_each_entry(obj, &con->head, node) {
+			struct ras_query_if info = {
+				.head = obj->head,
+			};
 
-		if (adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 2) &&
-		    adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 4)) {
-			if (amdgpu_ras_reset_error_status(adev, info.head.block))
-				dev_warn(adev->dev, "Failed to reset error counter and error status");
+			ret = amdgpu_ras_query_error_count_helper(adev, &ce, &ue, &info);
 		}
-
-		ce += info.ce_count;
-		ue += info.ue_count;
+	} else {
+		/* query specific ip block */
+		ret = amdgpu_ras_query_error_count_helper(adev, &ce, &ue, query_info);
 	}
 
+	if (ret)
+		return ret;
+
 	if (ce_count)
 		*ce_count = ce;
 
@@ -1564,14 +1617,14 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *
 	struct amdgpu_ras_block_object *block_obj =
 		amdgpu_ras_get_ras_block(adev, obj->head.block, 0);
 
-	if (!block_obj || !block_obj->hw_ops)
+	if (!block_obj)
 		return;
 
 	/* both query_poison_status and handle_poison_consumption are optional,
 	 * but at least one of them should be implemented if we need poison
 	 * consumption handler
 	 */
-	if (block_obj->hw_ops->query_poison_status) {
+	if (block_obj->hw_ops && block_obj->hw_ops->query_poison_status) {
 		poison_stat = block_obj->hw_ops->query_poison_status(adev);
 		if (!poison_stat) {
 			/* Not poison consumption interrupt, no need to handle it */
@@ -1585,7 +1638,7 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *
 	if (!adev->gmc.xgmi.connected_to_cpu)
 		amdgpu_umc_poison_handler(adev, false);
 
-	if (block_obj->hw_ops->handle_poison_consumption)
+	if (block_obj->hw_ops && block_obj->hw_ops->handle_poison_consumption)
 		poison_stat = block_obj->hw_ops->handle_poison_consumption(adev);
 
 	/* gpu reset is fallback for failed and default cases */
@@ -1593,6 +1646,8 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *
 		dev_info(adev->dev, "GPU reset for %s RAS poison consumption is issued!\n",
 				block_obj->ras_comm.name);
 		amdgpu_ras_reset_gpu(adev);
+	} else {
+		amdgpu_gfx_poison_consumption_handler(adev, entry);
 	}
 }
 
@@ -2344,22 +2399,24 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev)
 
 		if (amdgpu_atomfirmware_sram_ecc_supported(adev)) {
 			dev_info(adev->dev, "SRAM ECC is active.\n");
-			if (!amdgpu_sriov_vf(adev)) {
+			if (!amdgpu_sriov_vf(adev))
 				adev->ras_hw_enabled |= ~(1 << AMDGPU_RAS_BLOCK__UMC |
 							    1 << AMDGPU_RAS_BLOCK__DF);
-
-				if (adev->ip_versions[VCN_HWIP][0] == IP_VERSION(2, 6, 0) ||
-				    adev->ip_versions[VCN_HWIP][0] == IP_VERSION(4, 0, 0))
-					adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__VCN |
-							1 << AMDGPU_RAS_BLOCK__JPEG);
-				else
-					adev->ras_hw_enabled &= ~(1 << AMDGPU_RAS_BLOCK__VCN |
-							1 << AMDGPU_RAS_BLOCK__JPEG);
-			} else {
+			else
 				adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__PCIE_BIF |
 								1 << AMDGPU_RAS_BLOCK__SDMA |
 								1 << AMDGPU_RAS_BLOCK__GFX);
-			}
+
+			/* VCN/JPEG RAS can be supported on both bare metal and
+			 * SRIOV environment
+			 */
+			if (adev->ip_versions[VCN_HWIP][0] == IP_VERSION(2, 6, 0) ||
+			    adev->ip_versions[VCN_HWIP][0] == IP_VERSION(4, 0, 0))
+				adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__VCN |
+							1 << AMDGPU_RAS_BLOCK__JPEG);
+			else
+				adev->ras_hw_enabled &= ~(1 << AMDGPU_RAS_BLOCK__VCN |
+							1 << AMDGPU_RAS_BLOCK__JPEG);
 		} else {
 			dev_info(adev->dev, "SRAM ECC is not presented.\n");
 		}
@@ -2395,7 +2452,7 @@ static void amdgpu_ras_counte_dw(struct work_struct *work)
 
 	/* Cache new values.
 	 */
-	if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count) == 0) {
+	if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count, NULL) == 0) {
 		atomic_set(&con->ras_ce_count, ce_count);
 		atomic_set(&con->ras_ue_count, ue_count);
 	}
@@ -2405,11 +2462,42 @@ Out:
 	pm_runtime_put_autosuspend(dev->dev);
 }
 
+static void amdgpu_ras_query_poison_mode(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	bool df_poison, umc_poison;
+
+	/* poison setting is useless on SRIOV guest */
+	if (amdgpu_sriov_vf(adev) || !con)
+		return;
+
+	/* Init poison supported flag, the default value is false */
+	if (adev->gmc.xgmi.connected_to_cpu) {
+		/* enabled by default when GPU is connected to CPU */
+		con->poison_supported = true;
+	} else if (adev->df.funcs &&
+	    adev->df.funcs->query_ras_poison_mode &&
+	    adev->umc.ras &&
+	    adev->umc.ras->query_ras_poison_mode) {
+		df_poison =
+			adev->df.funcs->query_ras_poison_mode(adev);
+		umc_poison =
+			adev->umc.ras->query_ras_poison_mode(adev);
+
+		/* Only poison is set in both DF and UMC, we can support it */
+		if (df_poison && umc_poison)
+			con->poison_supported = true;
+		else if (df_poison != umc_poison)
+			dev_warn(adev->dev,
+				"Poison setting is inconsistent in DF/UMC(%d:%d)!\n",
+				df_poison, umc_poison);
+	}
+}
+
 int amdgpu_ras_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	int r;
-	bool df_poison, umc_poison;
 
 	if (con)
 		return 0;
@@ -2484,26 +2572,7 @@ int amdgpu_ras_init(struct amdgpu_device *adev)
 			goto release_con;
 	}
 
-	/* Init poison supported flag, the default value is false */
-	if (adev->gmc.xgmi.connected_to_cpu) {
-		/* enabled by default when GPU is connected to CPU */
-		con->poison_supported = true;
-	}
-	else if (adev->df.funcs &&
-	    adev->df.funcs->query_ras_poison_mode &&
-	    adev->umc.ras &&
-	    adev->umc.ras->query_ras_poison_mode) {
-		df_poison =
-			adev->df.funcs->query_ras_poison_mode(adev);
-		umc_poison =
-			adev->umc.ras->query_ras_poison_mode(adev);
-		/* Only poison is set in both DF and UMC, we can support it */
-		if (df_poison && umc_poison)
-			con->poison_supported = true;
-		else if (df_poison != umc_poison)
-			dev_warn(adev->dev, "Poison setting is inconsistent in DF/UMC(%d:%d)!\n",
-					df_poison, umc_poison);
-	}
+	amdgpu_ras_query_poison_mode(adev);
 
 	if (amdgpu_ras_fs_init(adev)) {
 		r = -EINVAL;
@@ -2564,6 +2633,7 @@ int amdgpu_ras_block_late_init(struct amdgpu_device *adev,
 {
 	struct amdgpu_ras_block_object *ras_obj = NULL;
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_query_if *query_info;
 	unsigned long ue_count, ce_count;
 	int r;
 
@@ -2605,11 +2675,17 @@ int amdgpu_ras_block_late_init(struct amdgpu_device *adev,
 
 	/* Those are the cached values at init.
 	 */
-	if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count) == 0) {
+	query_info = kzalloc(sizeof(struct ras_query_if), GFP_KERNEL);
+	if (!query_info)
+		return -ENOMEM;
+	memcpy(&query_info->head, ras_block, sizeof(struct ras_common_if));
+
+	if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count, query_info) == 0) {
 		atomic_set(&con->ras_ce_count, ce_count);
 		atomic_set(&con->ras_ue_count, ue_count);
 	}
 
+	kfree(query_info);
 	return 0;
 
 interrupt:
@@ -2946,11 +3022,26 @@ int amdgpu_ras_set_context(struct amdgpu_device *adev, struct amdgpu_ras *ras_co
 int amdgpu_ras_is_supported(struct amdgpu_device *adev,
 		unsigned int block)
 {
+	int ret = 0;
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
 
 	if (block >= AMDGPU_RAS_BLOCK_COUNT)
 		return 0;
-	return ras && (adev->ras_enabled & (1 << block));
+
+	ret = ras && (adev->ras_enabled & (1 << block));
+
+	/* For the special asic with mem ecc enabled but sram ecc
+	 * not enabled, even if the ras block is not supported on
+	 * .ras_enabled, if the asic supports poison mode and the
+	 * ras block has ras configuration, it can be considered
+	 * that the ras block supports ras function.
+	 */
+	if (!ret &&
+	    amdgpu_ras_is_poison_mode_supported(adev) &&
+	    amdgpu_ras_get_ras_block(adev, block, 0))
+		ret = 1;
+
+	return ret;
 }
 
 int amdgpu_ras_reset_gpu(struct amdgpu_device *adev)