drm/amdgpu: add ras_err_info to identify RAS error source

introduced "ras_err_info" to better identify a RAS ERROR source.

NOTE:
For legacy chips, keep the original RAS error print format.

v1:
RAS errors may come from different dies during a RAS error query,
therefore, need a new data structure to identify the source of RAS ERROR.

v2:
- use new data structure 'amdgpu_smuio_mcm_config_info' instead of
  ras_err_id (in v1 patch)
- refine ras error dump function name
- refine ras error dump log format
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@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_ras.c

@@ -152,8 +152,9 @@ static bool amdgpu_ras_get_error_query_ready(struct amdgpu_device *adev)
 
 static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t address)
 {
-	struct ras_err_data err_data = {0, 0, 0, NULL};
+	struct ras_err_data err_data;
 	struct eeprom_table_record err_rec;
+	int ret;
 
 	if ((address >= adev->gmc.mc_vram_size) ||
 	    (address >= RAS_UMC_INJECT_ADDR_LIMIT)) {
@@ -170,6 +171,10 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre
 		return 0;
 	}
 
+	ret = amdgpu_ras_error_data_init(&err_data);
+	if (ret)
+		return ret;
+
 	memset(&err_rec, 0x0, sizeof(struct eeprom_table_record));
 	err_data.err_addr = &err_rec;
 	amdgpu_umc_fill_error_record(&err_data, address, address, 0, 0);
@@ -180,6 +185,8 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre
 		amdgpu_ras_save_bad_pages(adev, NULL);
 	}
 
+	amdgpu_ras_error_data_fini(&err_data);
+
 	dev_warn(adev->dev, "WARNING: THIS IS ONLY FOR TEST PURPOSES AND WILL CORRUPT RAS EEPROM\n");
 	dev_warn(adev->dev, "Clear EEPROM:\n");
 	dev_warn(adev->dev, "    echo 1 > /sys/kernel/debug/dri/0/ras/ras_eeprom_reset\n");
@@ -1015,17 +1022,118 @@ static void amdgpu_ras_get_ecc_info(struct amdgpu_device *adev, struct ras_err_d
 	}
 }
 
+static void amdgpu_ras_error_print_error_data(struct amdgpu_device *adev,
+					      struct ras_query_if *query_if,
+					      struct ras_err_data *err_data,
+					      bool is_ue)
+{
+	struct ras_manager *ras_mgr = amdgpu_ras_find_obj(adev, &query_if->head);
+	const char *blk_name = get_ras_block_str(&query_if->head);
+	struct amdgpu_smuio_mcm_config_info *mcm_info;
+	struct ras_err_node *err_node;
+	struct ras_err_info *err_info;
+
+	if (is_ue)
+		dev_info(adev->dev, "%ld uncorrectable hardware errors detected in %s block\n",
+			 ras_mgr->err_data.ue_count, blk_name);
+	else
+		dev_info(adev->dev, "%ld correctable hardware errors detected in %s block\n",
+			 ras_mgr->err_data.ue_count, blk_name);
+
+	for_each_ras_error(err_node, err_data) {
+		err_info = &err_node->err_info;
+		mcm_info = &err_info->mcm_info;
+		if (is_ue && err_info->ue_count) {
+			dev_info(adev->dev, "socket: %d, die: %d "
+				 "%lld uncorrectable hardware errors detected in %s block\n",
+				 mcm_info->socket_id,
+				 mcm_info->die_id,
+				 err_info->ue_count,
+				 blk_name);
+		} else if (!is_ue && err_info->ce_count) {
+			dev_info(adev->dev, "socket: %d, die: %d "
+				 "%lld correctable hardware errors detected in %s block\n",
+				 mcm_info->socket_id,
+				 mcm_info->die_id,
+				 err_info->ue_count,
+				 blk_name);
+		}
+	}
+}
+
+static void amdgpu_ras_error_generate_report(struct amdgpu_device *adev,
+					     struct ras_query_if *query_if,
+					     struct ras_err_data *err_data)
+{
+	struct ras_manager *ras_mgr = amdgpu_ras_find_obj(adev, &query_if->head);
+	const char *blk_name = get_ras_block_str(&query_if->head);
+
+	if (err_data->ce_count) {
+		if (!list_empty(&err_data->err_node_list)) {
+			amdgpu_ras_error_print_error_data(adev, query_if,
+							  err_data, false);
+		} else if (!adev->aid_mask &&
+			   adev->smuio.funcs &&
+			   adev->smuio.funcs->get_socket_id &&
+			   adev->smuio.funcs->get_die_id) {
+			dev_info(adev->dev, "socket: %d, die: %d "
+				 "%ld correctable hardware errors "
+				 "detected in %s block, no user "
+				 "action is needed.\n",
+				 adev->smuio.funcs->get_socket_id(adev),
+				 adev->smuio.funcs->get_die_id(adev),
+				 ras_mgr->err_data.ce_count,
+				 blk_name);
+		} else {
+			dev_info(adev->dev, "%ld correctable hardware errors "
+				 "detected in %s block, no user "
+				 "action is needed.\n",
+				 ras_mgr->err_data.ce_count,
+				 blk_name);
+		}
+	}
+
+	if (err_data->ue_count) {
+		if (!list_empty(&err_data->err_node_list)) {
+			amdgpu_ras_error_print_error_data(adev, query_if,
+							  err_data, true);
+		} else if (!adev->aid_mask &&
+			   adev->smuio.funcs &&
+			   adev->smuio.funcs->get_socket_id &&
+			   adev->smuio.funcs->get_die_id) {
+			dev_info(adev->dev, "socket: %d, die: %d "
+				 "%ld uncorrectable hardware errors "
+				 "detected in %s block\n",
+				 adev->smuio.funcs->get_socket_id(adev),
+				 adev->smuio.funcs->get_die_id(adev),
+				 ras_mgr->err_data.ue_count,
+				 blk_name);
+		} else {
+			dev_info(adev->dev, "%ld uncorrectable hardware errors "
+				 "detected in %s block\n",
+				 ras_mgr->err_data.ue_count,
+				 blk_name);
+		}
+	}
+
+}
+
 /* query/inject/cure begin */
 int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
 				  struct ras_query_if *info)
 {
 	struct amdgpu_ras_block_object *block_obj = NULL;
 	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head);
-	struct ras_err_data err_data = {0, 0, 0, NULL};
+	struct ras_err_data err_data;
+	int ret;
 
 	if (!obj)
 		return -EINVAL;
 
+	ret = amdgpu_ras_error_data_init(&err_data);
+	if (ret)
+		return ret;
+
 	if (info->head.block == AMDGPU_RAS_BLOCK__UMC) {
 		amdgpu_ras_get_ecc_info(adev, &err_data);
 	} else {
@@ -1033,7 +1141,8 @@ int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
 		if (!block_obj || !block_obj->hw_ops)   {
 			dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",
 				     get_ras_block_str(&info->head));
-			return -EINVAL;
+			ret = -EINVAL;
+			goto out_fini_err_data;
 		}
 
 		if (block_obj->hw_ops->query_ras_error_count)
@@ -1053,48 +1162,12 @@ int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
 	info->ue_count = obj->err_data.ue_count;
 	info->ce_count = obj->err_data.ce_count;
 
-	if (err_data.ce_count) {
-		if (!adev->aid_mask &&
-		    adev->smuio.funcs &&
-		    adev->smuio.funcs->get_socket_id &&
-		    adev->smuio.funcs->get_die_id) {
-			dev_info(adev->dev, "socket: %d, die: %d "
-					"%ld correctable hardware errors "
-					"detected in %s block, no user "
-					"action is needed.\n",
-					adev->smuio.funcs->get_socket_id(adev),
-					adev->smuio.funcs->get_die_id(adev),
-					obj->err_data.ce_count,
-					get_ras_block_str(&info->head));
-		} else {
-			dev_info(adev->dev, "%ld correctable hardware errors "
-					"detected in %s block, no user "
-					"action is needed.\n",
-					obj->err_data.ce_count,
-					get_ras_block_str(&info->head));
-		}
-	}
-	if (err_data.ue_count) {
-		if (!adev->aid_mask &&
-		    adev->smuio.funcs &&
-		    adev->smuio.funcs->get_socket_id &&
-		    adev->smuio.funcs->get_die_id) {
-			dev_info(adev->dev, "socket: %d, die: %d "
-					"%ld uncorrectable hardware errors "
-					"detected in %s block\n",
-					adev->smuio.funcs->get_socket_id(adev),
-					adev->smuio.funcs->get_die_id(adev),
-					obj->err_data.ue_count,
-					get_ras_block_str(&info->head));
-		} else {
-			dev_info(adev->dev, "%ld uncorrectable hardware errors "
-					"detected in %s block\n",
-					obj->err_data.ue_count,
-					get_ras_block_str(&info->head));
-		}
-	}
+	amdgpu_ras_error_generate_report(adev, info, &err_data);
 
-	return 0;
+out_fini_err_data:
+	amdgpu_ras_error_data_fini(&err_data);
+
+	return ret;
 }
 
 int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
@@ -1744,12 +1817,16 @@ static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj,
 				struct amdgpu_iv_entry *entry)
 {
 	struct ras_ih_data *data = &obj->ih_data;
-	struct ras_err_data err_data = {0, 0, 0, NULL};
+	struct ras_err_data err_data;
 	int ret;
 
 	if (!data->cb)
 		return;
 
+	ret = amdgpu_ras_error_data_init(&err_data);
+	if (ret)
+		return;
+
 	/* Let IP handle its data, maybe we need get the output
 	 * from the callback to update the error type/count, etc
 	 */
@@ -1766,6 +1843,8 @@ static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj,
 		obj->err_data.ue_count += err_data.ue_count;
 		obj->err_data.ce_count += err_data.ce_count;
 	}
+
+	amdgpu_ras_error_data_fini(&err_data);
 }
 
 static void amdgpu_ras_interrupt_handler(struct ras_manager *obj)
@@ -3383,3 +3462,128 @@ void amdgpu_ras_inst_reset_ras_error_count(struct amdgpu_device *adev,
 		WREG32(err_status_hi_offset, 0);
 	}
 }
+
+int amdgpu_ras_error_data_init(struct ras_err_data *err_data)
+{
+	memset(err_data, 0, sizeof(*err_data));
+
+	INIT_LIST_HEAD(&err_data->err_node_list);
+
+	return 0;
+}
+
+static void amdgpu_ras_error_node_release(struct ras_err_node *err_node)
+{
+	if (!err_node)
+		return;
+
+	list_del(&err_node->node);
+	kvfree(err_node);
+}
+
+void amdgpu_ras_error_data_fini(struct ras_err_data *err_data)
+{
+	struct ras_err_node *err_node, *tmp;
+
+	list_for_each_entry_safe(err_node, tmp, &err_data->err_node_list, node) {
+		amdgpu_ras_error_node_release(err_node);
+		list_del(&err_node->node);
+	}
+}
+
+static struct ras_err_node *amdgpu_ras_error_find_node_by_id(struct ras_err_data *err_data,
+							     struct amdgpu_smuio_mcm_config_info *mcm_info)
+{
+	struct ras_err_node *err_node;
+	struct amdgpu_smuio_mcm_config_info *ref_id;
+
+	if (!err_data || !mcm_info)
+		return NULL;
+
+	for_each_ras_error(err_node, err_data) {
+		ref_id = &err_node->err_info.mcm_info;
+		if ((mcm_info->socket_id >= 0 && mcm_info->socket_id != ref_id->socket_id) ||
+		    (mcm_info->die_id >= 0 && mcm_info->die_id != ref_id->die_id))
+			continue;
+
+		return err_node;
+	}
+
+	return NULL;
+}
+
+static struct ras_err_node *amdgpu_ras_error_node_new(void)
+{
+	struct ras_err_node *err_node;
+
+	err_node = kvzalloc(sizeof(*err_node), GFP_KERNEL);
+	if (!err_node)
+		return NULL;
+
+	INIT_LIST_HEAD(&err_node->node);
+
+	return err_node;
+}
+
+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 ras_err_node *err_node;
+
+	err_node = amdgpu_ras_error_find_node_by_id(err_data, mcm_info);
+	if (err_node)
+		return &err_node->err_info;
+
+	err_node = amdgpu_ras_error_node_new();
+	if (!err_node)
+		return NULL;
+
+	memcpy(&err_node->err_info.mcm_info, mcm_info, sizeof(*mcm_info));
+
+	err_data->err_list_count++;
+	list_add_tail(&err_node->node, &err_data->err_node_list);
+
+	return &err_node->err_info;
+}
+
+int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
+					struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count)
+{
+	struct ras_err_info *err_info;
+
+	if (!err_data || !mcm_info)
+		return -EINVAL;
+
+	if (!count)
+		return 0;
+
+	err_info = amdgpu_ras_error_get_info(err_data, mcm_info);
+	if (!err_info)
+		return -EINVAL;
+
+	err_info->ue_count += count;
+	err_data->ue_count += count;
+
+	return 0;
+}
+
+int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data,
+					struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count)
+{
+	struct ras_err_info *err_info;
+
+	if (!err_data || !mcm_info)
+		return -EINVAL;
+
+	if (!count)
+		return 0;
+
+	err_info = amdgpu_ras_error_get_info(err_data, mcm_info);
+	if (!err_info)
+		return -EINVAL;
+
+	err_info->ce_count += count;
+	err_data->ce_count += count;
+
+	return 0;
+}

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

@@ -28,6 +28,7 @@
 #include <linux/list.h>
 #include "ta_ras_if.h"
 #include "amdgpu_ras_eeprom.h"
+#include "amdgpu_smuio.h"
 
 struct amdgpu_iv_entry;
 
@@ -443,13 +444,29 @@ struct ras_fs_data {
 	char debugfs_name[32];
 };
 
+struct ras_err_info {
+	struct amdgpu_smuio_mcm_config_info mcm_info;
+	u64 ce_count;
+	u64 ue_count;
+};
+
+struct ras_err_node {
+	struct list_head node;
+	struct ras_err_info err_info;
+};
+
 struct ras_err_data {
 	unsigned long ue_count;
 	unsigned long ce_count;
 	unsigned long err_addr_cnt;
 	struct eeprom_table_record *err_addr;
+	u32 err_list_count;
+	struct list_head err_node_list;
 };
 
+#define for_each_ras_error(err_node, err_data) \
+	list_for_each_entry(err_node, &(err_data)->err_node_list, node)
+
 struct ras_err_handler_data {
 	/* point to bad page records array */
 	struct eeprom_table_record *bps;
@@ -773,4 +790,12 @@ void amdgpu_ras_inst_reset_ras_error_count(struct amdgpu_device *adev,
 					   const struct amdgpu_ras_err_status_reg_entry *reg_list,
 					   uint32_t reg_list_size,
 					   uint32_t instance);
+
+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);
+int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
+					struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count);
+
 #endif

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

@@ -30,6 +30,11 @@ enum amdgpu_pkg_type {
 	AMDGPU_PKG_TYPE_UNKNOWN,
 };
 
+struct amdgpu_smuio_mcm_config_info {
+	int socket_id;
+	int die_id;
+};
+
 struct amdgpu_smuio_funcs {
 	u32 (*get_rom_index_offset)(struct amdgpu_device *adev);
 	u32 (*get_rom_data_offset)(struct amdgpu_device *adev);

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

@@ -45,8 +45,12 @@ static int amdgpu_umc_convert_error_address(struct amdgpu_device *adev,
 int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
 			uint64_t err_addr, uint32_t ch_inst, uint32_t umc_inst)
 {
-	struct ras_err_data err_data = {0, 0, 0, NULL};
-	int ret = AMDGPU_RAS_FAIL;
+	struct ras_err_data err_data;
+	int ret;
+
+	ret = amdgpu_ras_error_data_init(&err_data);
+	if (ret)
+		return ret;
 
 	err_data.err_addr =
 		kcalloc(adev->umc.max_ras_err_cnt_per_query,
@@ -54,7 +58,8 @@ int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
 	if (!err_data.err_addr) {
 		dev_warn(adev->dev,
 			"Failed to alloc memory for umc error record in MCA notifier!\n");
-		return AMDGPU_RAS_FAIL;
+		ret = AMDGPU_RAS_FAIL;
+		goto out_fini_err_data;
 	}
 
 	/*
@@ -63,7 +68,7 @@ int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
 	ret = amdgpu_umc_convert_error_address(adev, &err_data, err_addr,
 					ch_inst, umc_inst);
 	if (ret)
-		goto out;
+		goto out_free_err_addr;
 
 	if (amdgpu_bad_page_threshold != 0) {
 		amdgpu_ras_add_bad_pages(adev, err_data.err_addr,
@@ -71,8 +76,12 @@ int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
 		amdgpu_ras_save_bad_pages(adev, NULL);
 	}
 
-out:
+out_free_err_addr:
 	kfree(err_data.err_addr);
+
+out_fini_err_data:
+	amdgpu_ras_error_data_fini(&err_data);
+
 	return ret;
 }
 
@@ -182,18 +191,24 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, bool reset)
 	}
 
 	if (!amdgpu_sriov_vf(adev)) {
-		struct ras_err_data err_data = {0, 0, 0, NULL};
+		struct ras_err_data err_data;
 		struct ras_common_if head = {
 			.block = AMDGPU_RAS_BLOCK__UMC,
 		};
 		struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
 
+		ret = amdgpu_ras_error_data_init(&err_data);
+		if (ret)
+			return ret;
+
 		ret = amdgpu_umc_do_page_retirement(adev, &err_data, 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;
 		}
+
+		amdgpu_ras_error_data_fini(&err_data);
 	} else {
 		if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
 			adev->virt.ops->ras_poison_handler(adev);

drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c

@@ -365,9 +365,12 @@ static void nbio_v7_4_handle_ras_controller_intr_no_bifring(struct amdgpu_device
 {
 	uint32_t bif_doorbell_intr_cntl;
 	struct ras_manager *obj = amdgpu_ras_find_obj(adev, adev->nbio.ras_if);
-	struct ras_err_data err_data = {0, 0, 0, NULL};
+	struct ras_err_data err_data;
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
 
+	if (amdgpu_ras_error_data_init(&err_data))
+		return;
+
 	if (adev->asic_type == CHIP_ALDEBARAN)
 		bif_doorbell_intr_cntl = RREG32_SOC15(NBIO, 0, mmBIF_DOORBELL_INT_CNTL_ALDE);
 	else
@@ -418,6 +421,8 @@ static void nbio_v7_4_handle_ras_controller_intr_no_bifring(struct amdgpu_device
 		 */
 		amdgpu_ras_reset_gpu(adev);
 	}
+
+	amdgpu_ras_error_data_fini(&err_data);
 }
 
 static void nbio_v7_4_handle_ras_err_event_athub_intr_no_bifring(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/nbio_v7_9.c

@@ -560,9 +560,12 @@ static void nbio_v7_9_handle_ras_controller_intr_no_bifring(struct amdgpu_device
 {
 	uint32_t bif_doorbell_intr_cntl;
 	struct ras_manager *obj = amdgpu_ras_find_obj(adev, adev->nbio.ras_if);
-	struct ras_err_data err_data = {0, 0, 0, NULL};
+	struct ras_err_data err_data;
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
 
+	if (amdgpu_ras_error_data_init(&err_data))
+		return;
+
 	bif_doorbell_intr_cntl = RREG32_SOC15(NBIO, 0, regBIF_BX0_BIF_DOORBELL_INT_CNTL);
 
 	if (REG_GET_FIELD(bif_doorbell_intr_cntl,
@@ -607,6 +610,8 @@ static void nbio_v7_9_handle_ras_controller_intr_no_bifring(struct amdgpu_device
 		 */
 		amdgpu_ras_reset_gpu(adev);
 	}
+
+	amdgpu_ras_error_data_fini(&err_data);
 }
 
 static void nbio_v7_9_handle_ras_err_event_athub_intr_no_bifring(struct amdgpu_device *adev)