drm/amdgpu: add RAS poison handling for MCA

For MCA poison, if unmap queue fails, only gpu reset should be
triggered without page retirement handling, MCA notifier will do it.

v2: handle MCA poison consumption in umc_poison_handler directly.
Signed-off-by: NTao Zhou <tao.zhou1@amd.com>
Reviewed-by: NHawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c

@@ -169,19 +169,28 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
 		void *ras_error_status,
 		bool reset)
 {
-	int ret;
-	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
-	struct ras_common_if head = {
-		.block = AMDGPU_RAS_BLOCK__UMC,
-	};
-	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
+	int ret = AMDGPU_RAS_SUCCESS;
 
-	ret =
-		amdgpu_umc_do_page_retirement(adev, ras_error_status, NULL, reset);
+	if (!adev->gmc.xgmi.connected_to_cpu) {
+		struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
+		struct ras_common_if head = {
+			.block = AMDGPU_RAS_BLOCK__UMC,
+		};
+		struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
 
-	if (ret == AMDGPU_RAS_SUCCESS && obj) {
-		obj->err_data.ue_count += err_data->ue_count;
-		obj->err_data.ce_count += err_data->ce_count;
+		ret =
+			amdgpu_umc_do_page_retirement(adev, ras_error_status, NULL, reset);
+
+		if (ret == AMDGPU_RAS_SUCCESS && obj) {
+			obj->err_data.ue_count += err_data->ue_count;
+			obj->err_data.ce_count += err_data->ce_count;
+		}
+	} else if (reset) {
+		/* MCA poison handler is only responsible for GPU reset,
+		 * let MCA notifier do page retirement.
+		 */
+		kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
+		amdgpu_ras_reset_gpu(adev);
 	}
 
 	return ret;