drm/amdgpu: MCA supports recording umc address information

MCA supports recording umc address information.

V2:
  Move err_addr variable from struct ras_err_node to
struct ras_err_info.
Signed-off-by: YiPeng Chai <YiPeng.Chai@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -218,6 +218,7 @@ static void amdgpu_mca_smu_mca_bank_dump(struct amdgpu_device *adev, int idx, st
 int amdgpu_mca_smu_log_ras_error(struct amdgpu_device *adev, enum amdgpu_ras_block blk, enum amdgpu_mca_error_type type, struct ras_err_data *err_data)
 {
 	struct amdgpu_smuio_mcm_config_info mcm_info;
+	struct ras_err_addr err_addr = {0};
 	struct mca_bank_set mca_set;
 	struct mca_bank_node *node;
 	struct mca_bank_entry *entry;
@@ -246,10 +247,18 @@ int amdgpu_mca_smu_log_ras_error(struct amdgpu_device *adev, enum amdgpu_ras_blo
 		mcm_info.socket_id = entry->info.socket_id;
 		mcm_info.die_id = entry->info.aid;
 
+		if (blk == AMDGPU_RAS_BLOCK__UMC) {
+			err_addr.err_status = entry->regs[MCA_REG_IDX_STATUS];
+			err_addr.err_ipid = entry->regs[MCA_REG_IDX_IPID];
+			err_addr.err_addr = entry->regs[MCA_REG_IDX_ADDR];
+		}
+
 		if (type == AMDGPU_MCA_ERROR_TYPE_UE)
-			amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, (uint64_t)count);
+			amdgpu_ras_error_statistic_ue_count(err_data,
+				&mcm_info, &err_addr, (uint64_t)count);
 		else
-			amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, (uint64_t)count);
+			amdgpu_ras_error_statistic_ce_count(err_data,
+				&mcm_info, &err_addr, (uint64_t)count);
 	}
 
 out_mca_release:

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

@@ -1156,8 +1156,10 @@ static void amdgpu_rasmgr_error_data_statistic_update(struct ras_manager *obj, s
 		for_each_ras_error(err_node, err_data) {
 			err_info = &err_node->err_info;
 
-			amdgpu_ras_error_statistic_ce_count(&obj->err_data, &err_info->mcm_info, err_info->ce_count);
-			amdgpu_ras_error_statistic_ue_count(&obj->err_data, &err_info->mcm_info, err_info->ue_count);
+			amdgpu_ras_error_statistic_ce_count(&obj->err_data,
+					&err_info->mcm_info, NULL, err_info->ce_count);
+			amdgpu_ras_error_statistic_ue_count(&obj->err_data,
+					&err_info->mcm_info, NULL, err_info->ue_count);
 		}
 	} else {
 		/* for legacy asic path which doesn't has error source info */
@@ -3691,7 +3693,8 @@ static int ras_err_info_cmp(void *priv, const struct list_head *a, const struct
 }
 
 static struct ras_err_info *amdgpu_ras_error_get_info(struct ras_err_data *err_data,
-						      struct amdgpu_smuio_mcm_config_info *mcm_info)
+				struct amdgpu_smuio_mcm_config_info *mcm_info,
+				struct ras_err_addr *err_addr)
 {
 	struct ras_err_node *err_node;
 
@@ -3705,6 +3708,9 @@ static struct ras_err_info *amdgpu_ras_error_get_info(struct ras_err_data *err_d
 
 	memcpy(&err_node->err_info.mcm_info, mcm_info, sizeof(*mcm_info));
 
+	if (err_addr)
+		memcpy(&err_node->err_info.err_addr, err_addr, sizeof(*err_addr));
+
 	err_data->err_list_count++;
 	list_add_tail(&err_node->node, &err_data->err_node_list);
 	list_sort(NULL, &err_data->err_node_list, ras_err_info_cmp);
@@ -3713,7 +3719,8 @@ static struct ras_err_info *amdgpu_ras_error_get_info(struct ras_err_data *err_d
 }
 
 int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
-					struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count)
+		struct amdgpu_smuio_mcm_config_info *mcm_info,
+		struct ras_err_addr *err_addr, u64 count)
 {
 	struct ras_err_info *err_info;
 
@@ -3723,7 +3730,7 @@ int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
 	if (!count)
 		return 0;
 
-	err_info = amdgpu_ras_error_get_info(err_data, mcm_info);
+	err_info = amdgpu_ras_error_get_info(err_data, mcm_info, err_addr);
 	if (!err_info)
 		return -EINVAL;
 
@@ -3734,7 +3741,8 @@ int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
 }
 
 int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data,
-					struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count)
+		struct amdgpu_smuio_mcm_config_info *mcm_info,
+		struct ras_err_addr *err_addr, u64 count)
 {
 	struct ras_err_info *err_info;
 
@@ -3744,7 +3752,7 @@ int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data,
 	if (!count)
 		return 0;
 
-	err_info = amdgpu_ras_error_get_info(err_data, mcm_info);
+	err_info = amdgpu_ras_error_get_info(err_data, mcm_info, err_addr);
 	if (!err_info)
 		return -EINVAL;
 

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h

@@ -452,10 +452,17 @@ struct ras_fs_data {
 	char debugfs_name[32];
 };
 
+struct ras_err_addr {
+	uint64_t err_status;
+	uint64_t err_ipid;
+	uint64_t err_addr;
+};
+
 struct ras_err_info {
 	struct amdgpu_smuio_mcm_config_info mcm_info;
 	u64 ce_count;
 	u64 ue_count;
+	struct ras_err_addr err_addr;
 };
 
 struct ras_err_node {
@@ -806,8 +813,10 @@ void amdgpu_ras_inst_reset_ras_error_count(struct amdgpu_device *adev,
 int amdgpu_ras_error_data_init(struct ras_err_data *err_data);
 void amdgpu_ras_error_data_fini(struct ras_err_data *err_data);
 int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data,
-					struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count);
+		struct amdgpu_smuio_mcm_config_info *mcm_info,
+		struct ras_err_addr *err_addr, u64 count);
 int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
-					struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count);
+		struct amdgpu_smuio_mcm_config_info *mcm_info,
+		struct ras_err_addr *err_addr, u64 count);
 
 #endif

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

@@ -1313,10 +1313,10 @@ static void __xgmi_v6_4_0_query_error_count(struct amdgpu_device *adev, struct a
 
 	switch (xgmi_v6_4_0_pcs_mca_get_error_type(adev, status)) {
 	case AMDGPU_MCA_ERROR_TYPE_UE:
-		amdgpu_ras_error_statistic_ue_count(err_data, mcm_info, 1ULL);
+		amdgpu_ras_error_statistic_ue_count(err_data, mcm_info, NULL, 1ULL);
 		break;
 	case AMDGPU_MCA_ERROR_TYPE_CE:
-		amdgpu_ras_error_statistic_ce_count(err_data, mcm_info, 1ULL);
+		amdgpu_ras_error_statistic_ce_count(err_data, mcm_info, NULL, 1ULL);
 		break;
 	default:
 		break;

drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c

@@ -3828,8 +3828,8 @@ static void gfx_v9_4_3_inst_query_ras_err_count(struct amdgpu_device *adev,
 	/* the caller should make sure initialize value of
 	 * err_data->ue_count and err_data->ce_count
 	 */
-	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count);
-	amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, ce_count);
+	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, NULL, ue_count);
+	amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, NULL, ce_count);
 }
 
 static void gfx_v9_4_3_inst_reset_ras_err_count(struct amdgpu_device *adev,

drivers/gpu/drm/amd/amdgpu/mmhub_v1_8.c

@@ -652,8 +652,8 @@ static void mmhub_v1_8_inst_query_ras_error_count(struct amdgpu_device *adev,
 					AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
 					&ue_count);
 
-	amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, ce_count);
-	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count);
+	amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, NULL, ce_count);
+	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, NULL, ue_count);
 }
 
 static void mmhub_v1_8_query_ras_error_count(struct amdgpu_device *adev,

drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c

@@ -2156,7 +2156,7 @@ static void sdma_v4_4_2_inst_query_ras_error_count(struct amdgpu_device *adev,
 					AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
 					&ue_count);
 
-	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count);
+	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, NULL, ue_count);
 }
 
 static void sdma_v4_4_2_query_ras_error_count(struct amdgpu_device *adev,

drivers/gpu/drm/amd/amdgpu/umc_v12_0.c

@@ -166,8 +166,8 @@ static int umc_v12_0_query_error_count(struct amdgpu_device *adev,
 	umc_v12_0_query_correctable_error_count(adev, umc_reg_offset, &ce_count);
 	umc_v12_0_query_uncorrectable_error_count(adev, umc_reg_offset, &ue_count);
 
-	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, ue_count);
-	amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, ce_count);
+	amdgpu_ras_error_statistic_ue_count(err_data, &mcm_info, NULL, ue_count);
+	amdgpu_ras_error_statistic_ce_count(err_data, &mcm_info, NULL, ce_count);
 
 	return 0;
 }