drm/amdgpu: trigger mode1 reset for RAS RMA status

Check RMA status in bad page retirement flow.

v2: fix coding bugs in v1.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 29 insertions, 11 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 2dffce25e273..671bbb19995a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2068,8 +2068,9 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *
 	struct amdgpu_device *adev = obj->adev;
 	struct amdgpu_ras_block_object *block_obj =
 		amdgpu_ras_get_ras_block(adev, obj->head.block, 0);
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 
-	if (!block_obj)
+	if (!block_obj || !con)
 		return;
 
 	/* both query_poison_status and handle_poison_consumption are optional,
@@ -2092,14 +2093,17 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *
 	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 */
-	if (poison_stat) {
+	/* gpu reset is fallback for failed and default cases.
+	 * For RMA case, amdgpu_umc_poison_handler will handle gpu reset.
+	 */
+	if (poison_stat && !con->is_rma) {
 		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);
 	}
+
+	if (!poison_stat)
+		amdgpu_gfx_poison_consumption_handler(adev, entry);
 }
 
 static void amdgpu_ras_interrupt_poison_creation_handler(struct ras_manager *obj,
@@ -2815,6 +2819,7 @@ static void amdgpu_ras_do_page_retirement(struct work_struct *work)
 					      page_retirement_dwork.work);
 	struct amdgpu_device *adev = con->adev;
 	struct ras_err_data err_data;
+	unsigned long err_cnt;
 
 	if (amdgpu_in_reset(adev) || atomic_read(&con->in_recovery))
 		return;
@@ -2822,9 +2827,13 @@ static void amdgpu_ras_do_page_retirement(struct work_struct *work)
 	amdgpu_ras_error_data_init(&err_data);
 
 	amdgpu_umc_handle_bad_pages(adev, &err_data);
+	err_cnt = err_data.err_addr_cnt;
 
 	amdgpu_ras_error_data_fini(&err_data);
 
+	if (err_cnt && con->is_rma)
+		amdgpu_ras_reset_gpu(adev);
+
 	mutex_lock(&con->umc_ecc_log.lock);
 	if (radix_tree_tagged(&con->umc_ecc_log.de_page_tree,
 				UMC_ECC_NEW_DETECTED_TAG))
@@ -2881,7 +2890,8 @@ static int amdgpu_ras_poison_consumption_handler(struct amdgpu_device *adev,
 	if (poison_msg->pasid_fn)
 		poison_msg->pasid_fn(adev, pasid, poison_msg->data);
 
-	if (reset) {
+	/* for RMA, amdgpu_ras_poison_creation_handler will trigger gpu reset */
+	if (reset && !con->is_rma) {
 		flush_delayed_work(&con->page_retirement_dwork);
 
 		con->gpu_reset_flags |= reset;
@@ -4010,6 +4020,12 @@ int amdgpu_ras_reset_gpu(struct amdgpu_device *adev)
 {
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
 
+	/* mode1 is the only selection for RMA status */
+	if (ras->is_rma) {
+		ras->gpu_reset_flags = 0;
+		ras->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE1_RESET;
+	}
+
 	if (atomic_cmpxchg(&ras->in_recovery, 0, 1) == 0)
 		amdgpu_reset_domain_schedule(ras->adev->reset_domain, &ras->recovery_work);
 	return 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
index 540e0f066b26..20e0e522fb51 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
@@ -195,7 +195,8 @@ static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev,
 	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
 	amdgpu_umc_handle_bad_pages(adev, ras_error_status);
 
-	if (err_data->ue_count && reset) {
+	if ((err_data->ue_count || err_data->de_count) &&
+	    (reset || (con && con->is_rma))) {
 		con->gpu_reset_flags |= reset;
 		amdgpu_ras_reset_gpu(adev);
 	}
@@ -211,6 +212,7 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev,
 		.block = AMDGPU_RAS_BLOCK__UMC,
 	};
 	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	uint32_t timeout = timeout_ms;
 
 	memset(&err_data, 0, sizeof(err_data));
@@ -243,9 +245,7 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev,
 
 	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
 
-	if (reset) {
-		struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
-
+	if (reset || (err_data.err_addr_cnt && con && con->is_rma)) {
 		con->gpu_reset_flags |= reset;
 		amdgpu_ras_reset_gpu(adev);
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c
index 9e7ce1e6bc06..9cd221ed240c 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c
@@ -85,6 +85,7 @@ static int gfx_v11_0_3_poison_consumption_handler(struct amdgpu_device *adev,
 	if (entry && (entry->client_id == SOC21_IH_CLIENTID_GFX) &&
 	    (entry->src_id == GFX_11_0_0__SRCID__RLC_GC_FED_INTERRUPT) &&
 	     !entry->vmid && !entry->pasid) {
+		struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 		uint32_t rlc_status0 = 0;
 
 		rlc_status0 = RREG32_SOC15(GC, 0, regRLC_RLCS_FED_STATUS_0);
@@ -96,7 +97,8 @@ static int gfx_v11_0_3_poison_consumption_handler(struct amdgpu_device *adev,
 			ras->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE2_RESET;
 		}
 
-		amdgpu_ras_reset_gpu(adev);
+		if (con && !con->is_rma)
+			amdgpu_ras_reset_gpu(adev);
 	}
 
 	return 0;