diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 1109 |
1 files changed, 901 insertions, 208 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 973073e07b2a..352ce16a0963 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -28,6 +28,7 @@ #include <linux/reboot.h> #include <linux/syscalls.h> #include <linux/pm_runtime.h> +#include <linux/list_sort.h> #include "amdgpu.h" #include "amdgpu_ras.h" @@ -38,6 +39,7 @@ #include "nbio_v7_9.h" #include "atom.h" #include "amdgpu_reset.h" +#include "amdgpu_psp.h" #ifdef CONFIG_X86_MCE_AMD #include <asm/mce.h> @@ -72,6 +74,8 @@ const char *ras_block_string[] = { "mca", "vcn", "jpeg", + "ih", + "mpio", }; const char *ras_mca_block_string[] = { @@ -93,7 +97,8 @@ const char *get_ras_block_str(struct ras_common_if *ras_block) if (!ras_block) return "NULL"; - if (ras_block->block >= AMDGPU_RAS_BLOCK_COUNT) + if (ras_block->block >= AMDGPU_RAS_BLOCK_COUNT || + ras_block->block >= ARRAY_SIZE(ras_block_string)) return "OUT OF RANGE"; if (ras_block->block == AMDGPU_RAS_BLOCK__MCA) @@ -115,6 +120,8 @@ const char *get_ras_block_str(struct ras_common_if *ras_block) /* typical ECC bad page rate is 1 bad page per 100MB VRAM */ #define RAS_BAD_PAGE_COVER (100 * 1024 * 1024ULL) +#define MAX_UMC_POISON_POLLING_TIME_ASYNC 100 //ms + enum amdgpu_ras_retire_page_reservation { AMDGPU_RAS_RETIRE_PAGE_RESERVED, AMDGPU_RAS_RETIRE_PAGE_PENDING, @@ -304,11 +311,13 @@ static int amdgpu_ras_debugfs_ctrl_parse_data(struct file *f, return -EINVAL; data->head.block = block_id; - /* only ue and ce errors are supported */ + /* only ue, ce and poison errors are supported */ if (!memcmp("ue", err, 2)) data->head.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; else if (!memcmp("ce", err, 2)) data->head.type = AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE; + else if (!memcmp("poison", err, 6)) + data->head.type = AMDGPU_RAS_ERROR__POISON; else return -EINVAL; @@ -430,9 +439,10 @@ static void amdgpu_ras_instance_mask_check(struct amdgpu_device *adev, * The block is one of: umc, sdma, gfx, etc. * see ras_block_string[] for details * - * The error type is one of: ue, ce, where, + * The error type is one of: ue, ce and poison where, * ue is multi-uncorrectable * ce is single-correctable + * poison is poison * * The sub-block is a the sub-block index, pass 0 if there is no sub-block. * The address and value are hexadecimal numbers, leading 0x is optional. @@ -624,8 +634,12 @@ static ssize_t amdgpu_ras_sysfs_read(struct device *dev, dev_warn(obj->adev->dev, "Failed to reset error counter and error status"); } - return sysfs_emit(buf, "%s: %lu\n%s: %lu\n", "ue", info.ue_count, - "ce", info.ce_count); + if (info.head.block == AMDGPU_RAS_BLOCK__UMC) + return sysfs_emit(buf, "%s: %lu\n%s: %lu\n%s: %lu\n", "ue", info.ue_count, + "ce", info.ce_count, "de", info.de_count); + else + return sysfs_emit(buf, "%s: %lu\n%s: %lu\n", "ue", info.ue_count, + "ce", info.ce_count); } /* obj begin */ @@ -635,8 +649,11 @@ static ssize_t amdgpu_ras_sysfs_read(struct device *dev, static inline void put_obj(struct ras_manager *obj) { - if (obj && (--obj->use == 0)) + if (obj && (--obj->use == 0)) { list_del(&obj->node); + amdgpu_ras_error_data_fini(&obj->err_data); + } + if (obj && (obj->use < 0)) DRM_ERROR("RAS ERROR: Unbalance obj(%s) use\n", get_ras_block_str(&obj->head)); } @@ -666,6 +683,9 @@ static struct ras_manager *amdgpu_ras_create_obj(struct amdgpu_device *adev, if (alive_obj(obj)) return NULL; + if (amdgpu_ras_error_data_init(&obj->err_data)) + return NULL; + obj->head = *head; obj->adev = adev; list_add(&obj->node, &con->head); @@ -1023,108 +1043,326 @@ 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_manager *ras_mgr, struct ras_err_data *err_data, - bool is_ue) + struct ras_query_context *qctx, + const char *blk_name, + bool is_ue, + bool is_de) { - 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; + u64 event_id = qctx->event_id; + + if (is_ue) { + for_each_ras_error(err_node, err_data) { + err_info = &err_node->err_info; + mcm_info = &err_info->mcm_info; + if (err_info->ue_count) { + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " + "%lld new uncorrectable hardware errors detected in %s block\n", + mcm_info->socket_id, + mcm_info->die_id, + err_info->ue_count, + blk_name); + } + } - 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, &ras_mgr->err_data) { + err_info = &err_node->err_info; + mcm_info = &err_info->mcm_info; + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " + "%lld uncorrectable hardware errors detected in total in %s block\n", + mcm_info->socket_id, mcm_info->die_id, err_info->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); + } else { + if (is_de) { + for_each_ras_error(err_node, err_data) { + err_info = &err_node->err_info; + mcm_info = &err_info->mcm_info; + if (err_info->de_count) { + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " + "%lld new deferred hardware errors detected in %s block\n", + mcm_info->socket_id, + mcm_info->die_id, + err_info->de_count, + blk_name); + } + } + + for_each_ras_error(err_node, &ras_mgr->err_data) { + err_info = &err_node->err_info; + mcm_info = &err_info->mcm_info; + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " + "%lld deferred hardware errors detected in total in %s block\n", + mcm_info->socket_id, mcm_info->die_id, + err_info->de_count, blk_name); + } + } else { + for_each_ras_error(err_node, err_data) { + err_info = &err_node->err_info; + mcm_info = &err_info->mcm_info; + if (err_info->ce_count) { + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " + "%lld new correctable hardware errors detected in %s block\n", + mcm_info->socket_id, + mcm_info->die_id, + err_info->ce_count, + blk_name); + } + } + + for_each_ras_error(err_node, &ras_mgr->err_data) { + err_info = &err_node->err_info; + mcm_info = &err_info->mcm_info; + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " + "%lld correctable hardware errors detected in total in %s block\n", + mcm_info->socket_id, mcm_info->die_id, + err_info->ce_count, blk_name); + } } } } +static inline bool err_data_has_source_info(struct ras_err_data *data) +{ + return !list_empty(&data->err_node_list); +} + 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_err_data *err_data, + struct ras_query_context *qctx) { struct ras_manager *ras_mgr = amdgpu_ras_find_obj(adev, &query_if->head); const char *blk_name = get_ras_block_str(&query_if->head); + u64 event_id = qctx->event_id; 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); + if (err_data_has_source_info(err_data)) { + amdgpu_ras_error_print_error_data(adev, ras_mgr, err_data, qctx, + blk_name, false, 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); + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d " + "%ld correctable 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.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); + RAS_EVENT_LOG(adev, event_id, "%ld correctable hardware errors " + "detected in %s block\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); + if (err_data_has_source_info(err_data)) { + amdgpu_ras_error_print_error_data(adev, ras_mgr, err_data, qctx, + blk_name, true, false); + } else if (!adev->aid_mask && + adev->smuio.funcs && + adev->smuio.funcs->get_socket_id && + adev->smuio.funcs->get_die_id) { + RAS_EVENT_LOG(adev, event_id, "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 { + RAS_EVENT_LOG(adev, event_id, "%ld uncorrectable hardware errors " + "detected in %s block\n", + ras_mgr->err_data.ue_count, + blk_name); + } + } + + if (err_data->de_count) { + if (err_data_has_source_info(err_data)) { + amdgpu_ras_error_print_error_data(adev, ras_mgr, err_data, qctx, + blk_name, false, 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); + RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d " + "%ld deferred 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.de_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); + RAS_EVENT_LOG(adev, event_id, "%ld deferred hardware errors " + "detected in %s block\n", + ras_mgr->err_data.de_count, + blk_name); } } +} +static void amdgpu_rasmgr_error_data_statistic_update(struct ras_manager *obj, struct ras_err_data *err_data) +{ + struct ras_err_node *err_node; + struct ras_err_info *err_info; + + if (err_data_has_source_info(err_data)) { + for_each_ras_error(err_node, err_data) { + err_info = &err_node->err_info; + amdgpu_ras_error_statistic_de_count(&obj->err_data, + &err_info->mcm_info, NULL, err_info->de_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 */ + obj->err_data.ue_count += err_data->ue_count; + obj->err_data.ce_count += err_data->ce_count; + obj->err_data.de_count += err_data->de_count; + } } -/* query/inject/cure begin */ -int amdgpu_ras_query_error_status(struct amdgpu_device *adev, - struct ras_query_if *info) +static struct ras_manager *get_ras_manager(struct amdgpu_device *adev, enum amdgpu_ras_block blk) +{ + struct ras_common_if head; + + memset(&head, 0, sizeof(head)); + head.block = blk; + + return amdgpu_ras_find_obj(adev, &head); +} + +int amdgpu_ras_bind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk, + const struct aca_info *aca_info, void *data) +{ + struct ras_manager *obj; + + obj = get_ras_manager(adev, blk); + if (!obj) + return -EINVAL; + + return amdgpu_aca_add_handle(adev, &obj->aca_handle, ras_block_str(blk), aca_info, data); +} + +int amdgpu_ras_unbind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk) +{ + struct ras_manager *obj; + + obj = get_ras_manager(adev, blk); + if (!obj) + return -EINVAL; + + amdgpu_aca_remove_handle(&obj->aca_handle); + + return 0; +} + +static int amdgpu_aca_log_ras_error_data(struct amdgpu_device *adev, enum amdgpu_ras_block blk, + enum aca_error_type type, struct ras_err_data *err_data, + struct ras_query_context *qctx) +{ + struct ras_manager *obj; + + obj = get_ras_manager(adev, blk); + if (!obj) + return -EINVAL; + + return amdgpu_aca_get_error_data(adev, &obj->aca_handle, type, err_data, qctx); +} + +ssize_t amdgpu_ras_aca_sysfs_read(struct device *dev, struct device_attribute *attr, + struct aca_handle *handle, char *buf, void *data) +{ + struct ras_manager *obj = container_of(handle, struct ras_manager, aca_handle); + struct ras_query_if info = { + .head = obj->head, + }; + + if (amdgpu_ras_query_error_status(obj->adev, &info)) + return -EINVAL; + + return sysfs_emit(buf, "%s: %lu\n%s: %lu\n%s: %lu\n", "ue", info.ue_count, + "ce", info.ce_count, "de", info.ue_count); +} + +static int amdgpu_ras_query_error_status_helper(struct amdgpu_device *adev, + struct ras_query_if *info, + struct ras_err_data *err_data, + struct ras_query_context *qctx, + unsigned int error_query_mode) { + enum amdgpu_ras_block blk = info ? info->head.block : AMDGPU_RAS_BLOCK_COUNT; struct amdgpu_ras_block_object *block_obj = NULL; + int ret; + + if (blk == AMDGPU_RAS_BLOCK_COUNT) + return -EINVAL; + + if (error_query_mode == AMDGPU_RAS_INVALID_ERROR_QUERY) + return -EINVAL; + + if (error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY) { + if (info->head.block == AMDGPU_RAS_BLOCK__UMC) { + amdgpu_ras_get_ecc_info(adev, err_data); + } else { + block_obj = amdgpu_ras_get_ras_block(adev, info->head.block, 0); + 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; + } + + if (block_obj->hw_ops->query_ras_error_count) + block_obj->hw_ops->query_ras_error_count(adev, err_data); + + if ((info->head.block == AMDGPU_RAS_BLOCK__SDMA) || + (info->head.block == AMDGPU_RAS_BLOCK__GFX) || + (info->head.block == AMDGPU_RAS_BLOCK__MMHUB)) { + if (block_obj->hw_ops->query_ras_error_status) + block_obj->hw_ops->query_ras_error_status(adev); + } + } + } else { + if (amdgpu_aca_is_enabled(adev)) { + ret = amdgpu_aca_log_ras_error_data(adev, blk, ACA_ERROR_TYPE_UE, err_data, qctx); + if (ret) + return ret; + + ret = amdgpu_aca_log_ras_error_data(adev, blk, ACA_ERROR_TYPE_CE, err_data, qctx); + if (ret) + return ret; + + ret = amdgpu_aca_log_ras_error_data(adev, blk, ACA_ERROR_TYPE_DEFERRED, err_data, qctx); + if (ret) + return ret; + } else { + /* FIXME: add code to check return value later */ + amdgpu_mca_smu_log_ras_error(adev, blk, AMDGPU_MCA_ERROR_TYPE_UE, err_data, qctx); + amdgpu_mca_smu_log_ras_error(adev, blk, AMDGPU_MCA_ERROR_TYPE_CE, err_data, qctx); + } + } + + return 0; +} + +/* query/inject/cure begin */ +int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_if *info) +{ struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); struct ras_err_data err_data; + struct ras_query_context qctx; + unsigned int error_query_mode; int ret; if (!obj) @@ -1134,35 +1372,26 @@ int amdgpu_ras_query_error_status(struct amdgpu_device *adev, if (ret) return ret; - if (info->head.block == AMDGPU_RAS_BLOCK__UMC) { - amdgpu_ras_get_ecc_info(adev, &err_data); - } else { - block_obj = amdgpu_ras_get_ras_block(adev, info->head.block, 0); - 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)); - ret = -EINVAL; - goto out_fini_err_data; - } - - if (block_obj->hw_ops->query_ras_error_count) - block_obj->hw_ops->query_ras_error_count(adev, &err_data); + if (!amdgpu_ras_get_error_query_mode(adev, &error_query_mode)) + return -EINVAL; - if ((info->head.block == AMDGPU_RAS_BLOCK__SDMA) || - (info->head.block == AMDGPU_RAS_BLOCK__GFX) || - (info->head.block == AMDGPU_RAS_BLOCK__MMHUB)) { - if (block_obj->hw_ops->query_ras_error_status) - block_obj->hw_ops->query_ras_error_status(adev); - } - } + memset(&qctx, 0, sizeof(qctx)); + qctx.event_id = amdgpu_ras_acquire_event_id(adev, amdgpu_ras_intr_triggered() ? + RAS_EVENT_TYPE_ISR : RAS_EVENT_TYPE_INVALID); + ret = amdgpu_ras_query_error_status_helper(adev, info, + &err_data, + &qctx, + error_query_mode); + if (ret) + goto out_fini_err_data; - obj->err_data.ue_count += err_data.ue_count; - obj->err_data.ce_count += err_data.ce_count; + amdgpu_rasmgr_error_data_statistic_update(obj, &err_data); info->ue_count = obj->err_data.ue_count; info->ce_count = obj->err_data.ce_count; + info->de_count = obj->err_data.de_count; - amdgpu_ras_error_generate_report(adev, info, &err_data); + amdgpu_ras_error_generate_report(adev, info, &err_data, &qctx); out_fini_err_data: amdgpu_ras_error_data_fini(&err_data); @@ -1170,23 +1399,53 @@ out_fini_err_data: return ret; } -int amdgpu_ras_reset_error_status(struct amdgpu_device *adev, +int amdgpu_ras_reset_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block block) { struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, block, 0); + struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs; + const struct aca_smu_funcs *smu_funcs = adev->aca.smu_funcs; + struct amdgpu_hive_info *hive; + int hive_ras_recovery = 0; if (!block_obj || !block_obj->hw_ops) { dev_dbg_once(adev->dev, "%s doesn't config RAS function\n", - ras_block_str(block)); - return 0; + ras_block_str(block)); + return -EOPNOTSUPP; } - if (!amdgpu_ras_is_supported(adev, block)) - return 0; + if (!amdgpu_ras_is_supported(adev, block) || + !amdgpu_ras_get_aca_debug_mode(adev)) + return -EOPNOTSUPP; + + hive = amdgpu_get_xgmi_hive(adev); + if (hive) { + hive_ras_recovery = atomic_read(&hive->ras_recovery); + amdgpu_put_xgmi_hive(hive); + } + + /* skip ras error reset in gpu reset */ + if ((amdgpu_in_reset(adev) || atomic_read(&ras->in_recovery) || + hive_ras_recovery) && + ((smu_funcs && smu_funcs->set_debug_mode) || + (mca_funcs && mca_funcs->mca_set_debug_mode))) + return -EOPNOTSUPP; if (block_obj->hw_ops->reset_ras_error_count) block_obj->hw_ops->reset_ras_error_count(adev); + return 0; +} + +int amdgpu_ras_reset_error_status(struct amdgpu_device *adev, + enum amdgpu_ras_block block) +{ + struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, block, 0); + + if (amdgpu_ras_reset_error_count(adev, block) == -EOPNOTSUPP) + return 0; + if ((block == AMDGPU_RAS_BLOCK__GFX) || (block == AMDGPU_RAS_BLOCK__MMHUB)) { if (block_obj->hw_ops->reset_ras_error_status) @@ -1463,7 +1722,8 @@ static void amdgpu_ras_sysfs_remove_bad_page_node(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); - sysfs_remove_file_from_group(&adev->dev->kobj, + if (adev->dev->kobj.sd) + sysfs_remove_file_from_group(&adev->dev->kobj, &con->badpages_attr.attr, RAS_FS_NAME); } @@ -1482,7 +1742,8 @@ static int amdgpu_ras_sysfs_remove_dev_attr_node(struct amdgpu_device *adev) .attrs = attrs, }; - sysfs_remove_group(&adev->dev->kobj, &group); + if (adev->dev->kobj.sd) + sysfs_remove_group(&adev->dev->kobj, &group); return 0; } @@ -1529,7 +1790,8 @@ int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev, if (!obj || !obj->attr_inuse) return -EINVAL; - sysfs_remove_file_from_group(&adev->dev->kobj, + if (adev->dev->kobj.sd) + sysfs_remove_file_from_group(&adev->dev->kobj, &obj->sysfs_attr.attr, RAS_FS_NAME); obj->attr_inuse = 0; @@ -1663,7 +1925,10 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) } } - amdgpu_mca_smu_debugfs_init(adev, dir); + if (amdgpu_aca_is_enabled(adev)) + amdgpu_aca_smu_debugfs_init(adev, dir); + else + amdgpu_mca_smu_debugfs_init(adev, dir); } /* debugfs end */ @@ -1791,7 +2056,7 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager * } } - amdgpu_umc_poison_handler(adev, false); + amdgpu_umc_poison_handler(adev, obj->head.block, 0); if (block_obj->hw_ops && block_obj->hw_ops->handle_poison_consumption) poison_stat = block_obj->hw_ops->handle_poison_consumption(adev); @@ -1810,7 +2075,7 @@ static void amdgpu_ras_interrupt_poison_creation_handler(struct ras_manager *obj struct amdgpu_iv_entry *entry) { dev_info(obj->adev->dev, - "Poison is created, no user action is needed.\n"); + "Poison is created\n"); } static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj, @@ -1842,6 +2107,7 @@ 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; + obj->err_data.de_count += err_data.de_count; } amdgpu_ras_error_data_fini(&err_data); @@ -2133,6 +2399,19 @@ out: return ret; } +static void amdgpu_ras_set_fed_all(struct amdgpu_device *adev, + struct amdgpu_hive_info *hive, bool status) +{ + struct amdgpu_device *tmp_adev; + + if (hive) { + list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) + amdgpu_ras_set_fed(tmp_adev, status); + } else { + amdgpu_ras_set_fed(adev, status); + } +} + static void amdgpu_ras_do_recovery(struct work_struct *work) { struct amdgpu_ras *ras = @@ -2140,9 +2419,24 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) struct amdgpu_device *remote_adev = NULL; struct amdgpu_device *adev = ras->adev; struct list_head device_list, *device_list_handle = NULL; + struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); + + if (hive) { + atomic_set(&hive->ras_recovery, 1); + /* If any device which is part of the hive received RAS fatal + * error interrupt, set fatal error status on all. This + * condition will need a recovery, and flag will be cleared + * as part of recovery. + */ + list_for_each_entry(remote_adev, &hive->device_list, + gmc.xgmi.head) + if (amdgpu_ras_get_fed_status(remote_adev)) { + amdgpu_ras_set_fed_all(adev, hive, true); + break; + } + } if (!ras->disable_ras_err_cnt_harvest) { - struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); /* Build list of devices to query RAS related errors */ if (hive && adev->gmc.xgmi.num_physical_nodes > 1) { @@ -2159,7 +2453,6 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) amdgpu_ras_log_on_err_counter(remote_adev); } - amdgpu_put_xgmi_hive(hive); } if (amdgpu_device_should_recover_gpu(ras->adev)) { @@ -2194,6 +2487,10 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) amdgpu_device_gpu_recover(ras->adev, NULL, &reset_context); } atomic_set(&ras->in_recovery, 0); + if (hive) { + atomic_set(&hive->ras_recovery, 0); + amdgpu_put_xgmi_hive(hive); + } } /* alloc/realloc bps array */ @@ -2406,6 +2703,32 @@ static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev, } } +static int amdgpu_ras_page_retirement_thread(void *param) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)param; + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + while (!kthread_should_stop()) { + + wait_event_interruptible(con->page_retirement_wq, + kthread_should_stop() || + atomic_read(&con->page_retirement_req_cnt)); + + if (kthread_should_stop()) + break; + + dev_info(adev->dev, "Start processing page retirement. request:%d\n", + atomic_read(&con->page_retirement_req_cnt)); + + atomic_dec(&con->page_retirement_req_cnt); + + amdgpu_umc_bad_page_polling_timeout(adev, + 0, MAX_UMC_POISON_POLLING_TIME_ASYNC); + } + + return 0; +} + int amdgpu_ras_recovery_init(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -2428,7 +2751,7 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) return 0; data = &con->eh_data; - *data = kmalloc(sizeof(**data), GFP_KERNEL | __GFP_ZERO); + *data = kzalloc(sizeof(**data), GFP_KERNEL); if (!*data) { ret = -ENOMEM; goto out; @@ -2469,6 +2792,16 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) } } + mutex_init(&con->page_retirement_lock); + init_waitqueue_head(&con->page_retirement_wq); + atomic_set(&con->page_retirement_req_cnt, 0); + con->page_retirement_thread = + kthread_run(amdgpu_ras_page_retirement_thread, adev, "umc_page_retirement"); + if (IS_ERR(con->page_retirement_thread)) { + con->page_retirement_thread = NULL; + dev_warn(adev->dev, "Failed to create umc_page_retirement thread!!!\n"); + } + #ifdef CONFIG_X86_MCE_AMD if ((adev->asic_type == CHIP_ALDEBARAN) && (adev->gmc.xgmi.connected_to_cpu)) @@ -2504,6 +2837,11 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) if (!data) return 0; + if (con->page_retirement_thread) + kthread_stop(con->page_retirement_thread); + + atomic_set(&con->page_retirement_req_cnt, 0); + cancel_work_sync(&con->recovery_work); mutex_lock(&con->recovery_lock); @@ -2565,6 +2903,87 @@ static void amdgpu_ras_get_quirks(struct amdgpu_device *adev) adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__GFX); } +/* Query ras capablity via atomfirmware interface */ +static void amdgpu_ras_query_ras_capablity_from_vbios(struct amdgpu_device *adev) +{ + /* mem_ecc cap */ + if (amdgpu_atomfirmware_mem_ecc_supported(adev)) { + dev_info(adev->dev, "MEM ECC is active.\n"); + adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__UMC | + 1 << AMDGPU_RAS_BLOCK__DF); + } else { + dev_info(adev->dev, "MEM ECC is not presented.\n"); + } + + /* sram_ecc cap */ + if (amdgpu_atomfirmware_sram_ecc_supported(adev)) { + dev_info(adev->dev, "SRAM ECC is active.\n"); + if (!amdgpu_sriov_vf(adev)) + adev->ras_hw_enabled |= ~(1 << AMDGPU_RAS_BLOCK__UMC | + 1 << AMDGPU_RAS_BLOCK__DF); + 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 (amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(2, 6, 0) || + amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 0) || + amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 3)) + 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); + + /* + * XGMI RAS is not supported if xgmi num physical nodes + * is zero + */ + if (!adev->gmc.xgmi.num_physical_nodes) + adev->ras_hw_enabled &= ~(1 << AMDGPU_RAS_BLOCK__XGMI_WAFL); + } else { + dev_info(adev->dev, "SRAM ECC is not presented.\n"); + } +} + +/* Query poison mode from umc/df IP callbacks */ +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 || + adev->gmc.is_app_apu) { + /* 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); + } +} + /* * check hardware's ras ability which will be saved in hw_supported. * if hardware does not support ras, we can skip some ras initializtion and @@ -2581,47 +3000,13 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev) if (!amdgpu_ras_asic_supported(adev)) return; - if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) { - if (amdgpu_atomfirmware_mem_ecc_supported(adev)) { - dev_info(adev->dev, "MEM ECC is active.\n"); - adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__UMC | - 1 << AMDGPU_RAS_BLOCK__DF); - } else { - dev_info(adev->dev, "MEM ECC is not presented.\n"); - } - - if (amdgpu_atomfirmware_sram_ecc_supported(adev)) { - dev_info(adev->dev, "SRAM ECC is active.\n"); - if (!amdgpu_sriov_vf(adev)) - adev->ras_hw_enabled |= ~(1 << AMDGPU_RAS_BLOCK__UMC | - 1 << AMDGPU_RAS_BLOCK__DF); - else - adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__PCIE_BIF | - 1 << AMDGPU_RAS_BLOCK__SDMA | - 1 << AMDGPU_RAS_BLOCK__GFX); + /* query ras capability from psp */ + if (amdgpu_psp_get_ras_capability(&adev->psp)) + goto init_ras_enabled_flag; - /* VCN/JPEG RAS can be supported on both bare metal and - * SRIOV environment - */ - if (amdgpu_ip_version(adev, VCN_HWIP, 0) == - IP_VERSION(2, 6, 0) || - amdgpu_ip_version(adev, 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); - - /* - * XGMI RAS is not supported if xgmi num physical nodes - * is zero - */ - if (!adev->gmc.xgmi.num_physical_nodes) - adev->ras_hw_enabled &= ~(1 << AMDGPU_RAS_BLOCK__XGMI_WAFL); - } else { - dev_info(adev->dev, "SRAM ECC is not presented.\n"); - } + /* query ras capablity from bios */ + if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) { + amdgpu_ras_query_ras_capablity_from_vbios(adev); } else { /* driver only manages a few IP blocks RAS feature * when GPU is connected cpu through XGMI */ @@ -2630,23 +3015,21 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev) 1 << AMDGPU_RAS_BLOCK__MMHUB); } + /* apply asic specific settings (vega20 only for now) */ amdgpu_ras_get_quirks(adev); + /* query poison mode from umc/df ip callback */ + amdgpu_ras_query_poison_mode(adev); + +init_ras_enabled_flag: /* hw_supported needs to be aligned with RAS block mask. */ adev->ras_hw_enabled &= AMDGPU_RAS_BLOCK_MASK; + adev->ras_enabled = amdgpu_ras_enable == 0 ? 0 : + adev->ras_hw_enabled & amdgpu_ras_mask; - /* - * Disable ras feature for aqua vanjaram - * by default on apu platform. - */ - if (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 6) && - adev->gmc.is_app_apu) - adev->ras_enabled = amdgpu_ras_enable != 1 ? 0 : - adev->ras_hw_enabled & amdgpu_ras_mask; - else - adev->ras_enabled = amdgpu_ras_enable == 0 ? 0 : - adev->ras_hw_enabled & amdgpu_ras_mask; + /* aca is disabled by default */ + adev->aca.is_enabled = false; } static void amdgpu_ras_counte_dw(struct work_struct *work) @@ -2674,44 +3057,41 @@ Out: pm_runtime_put_autosuspend(dev->dev); } -static void amdgpu_ras_query_poison_mode(struct amdgpu_device *adev) +static int amdgpu_get_ras_schema(struct amdgpu_device *adev) { - struct amdgpu_ras *con = amdgpu_ras_get_context(adev); - bool df_poison, umc_poison; + return amdgpu_ras_is_poison_mode_supported(adev) ? AMDGPU_RAS_ERROR__POISON : 0 | + AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE | + AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE | + AMDGPU_RAS_ERROR__PARITY; +} - /* poison setting is useless on SRIOV guest */ - if (amdgpu_sriov_vf(adev) || !con) +static void ras_event_mgr_init(struct ras_event_manager *mgr) +{ + int i; + + for (i = 0; i < ARRAY_SIZE(mgr->seqnos); i++) + atomic64_set(&mgr->seqnos[i], 0); +} + +static void amdgpu_ras_event_mgr_init(struct amdgpu_device *adev) +{ + struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + struct amdgpu_hive_info *hive; + + if (!ras) 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); + hive = amdgpu_get_xgmi_hive(adev); + ras->event_mgr = hive ? &hive->event_mgr : &ras->__event_mgr; - /* 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); + /* init event manager with node 0 on xgmi system */ + if (!amdgpu_in_reset(adev)) { + if (!hive || adev->gmc.xgmi.node_id == 0) + ras_event_mgr_init(ras->event_mgr); } -} -static int amdgpu_get_ras_schema(struct amdgpu_device *adev) -{ - return amdgpu_ras_is_poison_mode_supported(adev) ? AMDGPU_RAS_ERROR__POISON : 0 | - AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE | - AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE | - AMDGPU_RAS_ERROR__PARITY; + if (hive) + amdgpu_put_xgmi_hive(hive); } int amdgpu_ras_init(struct amdgpu_device *adev) @@ -2722,10 +3102,10 @@ int amdgpu_ras_init(struct amdgpu_device *adev) if (con) return 0; - con = kmalloc(sizeof(struct amdgpu_ras) + + con = kzalloc(sizeof(*con) + sizeof(struct ras_manager) * AMDGPU_RAS_BLOCK_COUNT + sizeof(struct ras_manager) * AMDGPU_RAS_MCA_BLOCK_COUNT, - GFP_KERNEL|__GFP_ZERO); + GFP_KERNEL); if (!con) return -ENOMEM; @@ -2810,7 +3190,11 @@ int amdgpu_ras_init(struct amdgpu_device *adev) goto release_con; } - amdgpu_ras_query_poison_mode(adev); + /* Packed socket_id to ras feature mask bits[31:29] */ + if (adev->smuio.funcs && + adev->smuio.funcs->get_socket_id) + con->features |= ((adev->smuio.funcs->get_socket_id(adev)) << + AMDGPU_RAS_FEATURES_SOCKETID_SHIFT); /* Get RAS schema for particular SOC */ con->schema = amdgpu_get_ras_schema(adev); @@ -3016,7 +3400,7 @@ void amdgpu_ras_suspend(struct amdgpu_device *adev) amdgpu_ras_disable_all_features(adev, 0); /* Make sure all ras objects are disabled. */ - if (con->features) + if (AMDGPU_RAS_GET_FEATURES(con->features)) amdgpu_ras_disable_all_features(adev, 1); } @@ -3030,13 +3414,31 @@ int amdgpu_ras_late_init(struct amdgpu_device *adev) if (amdgpu_sriov_vf(adev)) return 0; + amdgpu_ras_event_mgr_init(adev); + + if (amdgpu_aca_is_enabled(adev)) { + if (amdgpu_in_reset(adev)) + r = amdgpu_aca_reset(adev); + else + r = amdgpu_aca_init(adev); + if (r) + return r; + + amdgpu_ras_set_aca_debug_mode(adev, false); + } else { + amdgpu_ras_set_mca_debug_mode(adev, false); + } + list_for_each_entry_safe(node, tmp, &adev->ras_list, node) { - if (!node->ras_obj) { + obj = node->ras_obj; + if (!obj) { dev_warn(adev->dev, "Warning: abnormal ras list node.\n"); continue; } - obj = node->ras_obj; + if (!amdgpu_ras_is_supported(adev, obj->ras_comm.block)) + continue; + if (obj->ras_late_init) { r = obj->ras_late_init(adev, &obj->ras_comm); if (r) { @@ -3061,7 +3463,7 @@ int amdgpu_ras_pre_fini(struct amdgpu_device *adev) /* Need disable ras on all IPs here before ip [hw/sw]fini */ - if (con->features) + if (AMDGPU_RAS_GET_FEATURES(con->features)) amdgpu_ras_disable_all_features(adev, 0); amdgpu_ras_recovery_fini(adev); return 0; @@ -3094,10 +3496,13 @@ int amdgpu_ras_fini(struct amdgpu_device *adev) amdgpu_ras_fs_fini(adev); amdgpu_ras_interrupt_remove_all(adev); - WARN(con->features, "Feature mask is not cleared"); + if (amdgpu_aca_is_enabled(adev)) + amdgpu_aca_fini(adev); - if (con->features) - amdgpu_ras_disable_all_features(adev, 1); + WARN(AMDGPU_RAS_GET_FEATURES(con->features), "Feature mask is not cleared"); + + if (AMDGPU_RAS_GET_FEATURES(con->features)) + amdgpu_ras_disable_all_features(adev, 0); cancel_delayed_work_sync(&con->ras_counte_delay_work); @@ -3107,14 +3512,59 @@ int amdgpu_ras_fini(struct amdgpu_device *adev) return 0; } +bool amdgpu_ras_get_fed_status(struct amdgpu_device *adev) +{ + struct amdgpu_ras *ras; + + ras = amdgpu_ras_get_context(adev); + if (!ras) + return false; + + return atomic_read(&ras->fed); +} + +void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status) +{ + struct amdgpu_ras *ras; + + ras = amdgpu_ras_get_context(adev); + if (ras) + atomic_set(&ras->fed, !!status); +} + +bool amdgpu_ras_event_id_is_valid(struct amdgpu_device *adev, u64 id) +{ + return !(id & BIT_ULL(63)); +} + +u64 amdgpu_ras_acquire_event_id(struct amdgpu_device *adev, enum ras_event_type type) +{ + struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + u64 id; + + switch (type) { + case RAS_EVENT_TYPE_ISR: + id = (u64)atomic64_read(&ras->event_mgr->seqnos[type]); + break; + case RAS_EVENT_TYPE_INVALID: + default: + id = BIT_ULL(63) | 0ULL; + break; + } + + return id; +} + void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev) { if (atomic_cmpxchg(&amdgpu_ras_in_intr, 0, 1) == 0) { struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + u64 event_id = (u64)atomic64_inc_return(&ras->event_mgr->seqnos[RAS_EVENT_TYPE_ISR]); - dev_info(adev->dev, "uncorrectable hardware error" - "(ERREVENT_ATHUB_INTERRUPT) detected!\n"); + RAS_EVENT_LOG(adev, event_id, "uncorrectable hardware error" + "(ERREVENT_ATHUB_INTERRUPT) detected!\n"); + amdgpu_ras_set_fed(adev, true); ras->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE1_RESET; amdgpu_ras_reset_gpu(adev); } @@ -3287,6 +3737,7 @@ int amdgpu_ras_is_supported(struct amdgpu_device *adev, block == AMDGPU_RAS_BLOCK__SDMA || block == AMDGPU_RAS_BLOCK__VCN || block == AMDGPU_RAS_BLOCK__JPEG) && + (amdgpu_ras_mask & (1 << block)) && amdgpu_ras_is_poison_mode_supported(adev) && amdgpu_ras_get_ras_block(adev, block, 0)) ret = 1; @@ -3303,6 +3754,73 @@ int amdgpu_ras_reset_gpu(struct amdgpu_device *adev) return 0; } +int amdgpu_ras_set_mca_debug_mode(struct amdgpu_device *adev, bool enable) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + int ret = 0; + + if (con) { + ret = amdgpu_mca_smu_set_debug_mode(adev, enable); + if (!ret) + con->is_aca_debug_mode = enable; + } + + return ret; +} + +int amdgpu_ras_set_aca_debug_mode(struct amdgpu_device *adev, bool enable) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + int ret = 0; + + if (con) { + if (amdgpu_aca_is_enabled(adev)) + ret = amdgpu_aca_smu_set_debug_mode(adev, enable); + else + ret = amdgpu_mca_smu_set_debug_mode(adev, enable); + if (!ret) + con->is_aca_debug_mode = enable; + } + + return ret; +} + +bool amdgpu_ras_get_aca_debug_mode(struct amdgpu_device *adev) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + const struct aca_smu_funcs *smu_funcs = adev->aca.smu_funcs; + const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs; + + if (!con) + return false; + + if ((amdgpu_aca_is_enabled(adev) && smu_funcs && smu_funcs->set_debug_mode) || + (!amdgpu_aca_is_enabled(adev) && mca_funcs && mca_funcs->mca_set_debug_mode)) + return con->is_aca_debug_mode; + else + return true; +} + +bool amdgpu_ras_get_error_query_mode(struct amdgpu_device *adev, + unsigned int *error_query_mode) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs; + const struct aca_smu_funcs *smu_funcs = adev->aca.smu_funcs; + + if (!con) { + *error_query_mode = AMDGPU_RAS_INVALID_ERROR_QUERY; + return false; + } + + if ((smu_funcs && smu_funcs->set_debug_mode) || (mca_funcs && mca_funcs->mca_set_debug_mode)) + *error_query_mode = + (con->is_aca_debug_mode) ? AMDGPU_RAS_DIRECT_ERROR_QUERY : AMDGPU_RAS_FIRMWARE_ERROR_QUERY; + else + *error_query_mode = AMDGPU_RAS_DIRECT_ERROR_QUERY; + + return true; +} /* Register each ip ras block into amdgpu ras */ int amdgpu_ras_register_ras_block(struct amdgpu_device *adev, @@ -3485,10 +4003,8 @@ 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) { + 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, @@ -3502,11 +4018,10 @@ static struct ras_err_node *amdgpu_ras_error_find_node_by_id(struct ras_err_data 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; + if (mcm_info->socket_id == ref_id->socket_id && + mcm_info->die_id == ref_id->die_id) + return err_node; } return NULL; @@ -3525,8 +4040,23 @@ static struct ras_err_node *amdgpu_ras_error_node_new(void) return err_node; } +static int ras_err_info_cmp(void *priv, const struct list_head *a, const struct list_head *b) +{ + struct ras_err_node *nodea = container_of(a, struct ras_err_node, node); + struct ras_err_node *nodeb = container_of(b, struct ras_err_node, node); + struct amdgpu_smuio_mcm_config_info *infoa = &nodea->err_info.mcm_info; + struct amdgpu_smuio_mcm_config_info *infob = &nodeb->err_info.mcm_info; + + if (unlikely(infoa->socket_id != infob->socket_id)) + return infoa->socket_id - infob->socket_id; + else + return infoa->die_id - infob->die_id; + + return 0; +} + 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_node *err_node; @@ -3538,16 +4068,43 @@ static struct ras_err_info *amdgpu_ras_error_get_info(struct ras_err_data *err_d if (!err_node) return NULL; + INIT_LIST_HEAD(&err_node->err_info.err_addr_list); + 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); + list_sort(NULL, &err_data->err_node_list, ras_err_info_cmp); return &err_node->err_info; } +void amdgpu_ras_add_mca_err_addr(struct ras_err_info *err_info, struct ras_err_addr *err_addr) +{ + struct ras_err_addr *mca_err_addr; + + mca_err_addr = kzalloc(sizeof(*mca_err_addr), GFP_KERNEL); + if (!mca_err_addr) + return; + + INIT_LIST_HEAD(&mca_err_addr->node); + + mca_err_addr->err_status = err_addr->err_status; + mca_err_addr->err_ipid = err_addr->err_ipid; + mca_err_addr->err_addr = err_addr->err_addr; + + list_add_tail(&mca_err_addr->node, &err_info->err_addr_list); +} + +void amdgpu_ras_del_mca_err_addr(struct ras_err_info *err_info, struct ras_err_addr *mca_err_addr) +{ + list_del(&mca_err_addr->node); + kfree(mca_err_addr); +} + 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; @@ -3561,6 +4118,9 @@ int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data, if (!err_info) return -EINVAL; + if (err_addr && err_addr->err_status) + amdgpu_ras_add_mca_err_addr(err_info, err_addr); + err_info->ue_count += count; err_data->ue_count += count; @@ -3568,7 +4128,8 @@ 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; @@ -3587,3 +4148,135 @@ int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data, return 0; } + +int amdgpu_ras_error_statistic_de_count(struct ras_err_data *err_data, + struct amdgpu_smuio_mcm_config_info *mcm_info, + struct ras_err_addr *err_addr, 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; + + if (err_addr && err_addr->err_status) + amdgpu_ras_add_mca_err_addr(err_info, err_addr); + + err_info->de_count += count; + err_data->de_count += count; + + return 0; +} + +#define mmMP0_SMN_C2PMSG_92 0x1609C +#define mmMP0_SMN_C2PMSG_126 0x160BE +static void amdgpu_ras_boot_time_error_reporting(struct amdgpu_device *adev, + u32 instance, u32 boot_error) +{ + u32 socket_id, aid_id, hbm_id; + u32 reg_data; + u64 reg_addr; + + socket_id = AMDGPU_RAS_GPU_ERR_SOCKET_ID(boot_error); + aid_id = AMDGPU_RAS_GPU_ERR_AID_ID(boot_error); + hbm_id = AMDGPU_RAS_GPU_ERR_HBM_ID(boot_error); + + /* The pattern for smn addressing in other SOC could be different from + * the one for aqua_vanjaram. We should revisit the code if the pattern + * is changed. In such case, replace the aqua_vanjaram implementation + * with more common helper */ + reg_addr = (mmMP0_SMN_C2PMSG_92 << 2) + + aqua_vanjaram_encode_ext_smn_addressing(instance); + + reg_data = amdgpu_device_indirect_rreg_ext(adev, reg_addr); + dev_err(adev->dev, "socket: %d, aid: %d, firmware boot failed, fw status is 0x%x\n", + socket_id, aid_id, reg_data); + + if (AMDGPU_RAS_GPU_ERR_MEM_TRAINING(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, hbm: %d, memory training failed\n", + socket_id, aid_id, hbm_id); + + if (AMDGPU_RAS_GPU_ERR_FW_LOAD(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, firmware load failed at boot time\n", + socket_id, aid_id); + + if (AMDGPU_RAS_GPU_ERR_WAFL_LINK_TRAINING(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, wafl link training failed\n", + socket_id, aid_id); + + if (AMDGPU_RAS_GPU_ERR_XGMI_LINK_TRAINING(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, xgmi link training failed\n", + socket_id, aid_id); + + if (AMDGPU_RAS_GPU_ERR_USR_CP_LINK_TRAINING(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, usr cp link training failed\n", + socket_id, aid_id); + + if (AMDGPU_RAS_GPU_ERR_USR_DP_LINK_TRAINING(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, usr dp link training failed\n", + socket_id, aid_id); + + if (AMDGPU_RAS_GPU_ERR_HBM_MEM_TEST(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, hbm: %d, hbm memory test failed\n", + socket_id, aid_id, hbm_id); + + if (AMDGPU_RAS_GPU_ERR_HBM_BIST_TEST(boot_error)) + dev_info(adev->dev, "socket: %d, aid: %d, hbm: %d, hbm bist test failed\n", + socket_id, aid_id, hbm_id); +} + +static int amdgpu_ras_wait_for_boot_complete(struct amdgpu_device *adev, + u32 instance, u32 *boot_error) +{ + u32 reg_addr; + u32 reg_data; + int retry_loop; + + reg_addr = (mmMP0_SMN_C2PMSG_92 << 2) + + aqua_vanjaram_encode_ext_smn_addressing(instance); + + for (retry_loop = 0; retry_loop < AMDGPU_RAS_BOOT_STATUS_POLLING_LIMIT; retry_loop++) { + reg_data = amdgpu_device_indirect_rreg_ext(adev, reg_addr); + if ((reg_data & AMDGPU_RAS_BOOT_STATUS_MASK) == AMDGPU_RAS_BOOT_STEADY_STATUS) { + *boot_error = AMDGPU_RAS_BOOT_SUCEESS; + return 0; + } + msleep(1); + } + + /* The pattern for smn addressing in other SOC could be different from + * the one for aqua_vanjaram. We should revisit the code if the pattern + * is changed. In such case, replace the aqua_vanjaram implementation + * with more common helper */ + reg_addr = (mmMP0_SMN_C2PMSG_126 << 2) + + aqua_vanjaram_encode_ext_smn_addressing(instance); + + for (retry_loop = 0; retry_loop < AMDGPU_RAS_BOOT_STATUS_POLLING_LIMIT; retry_loop++) { + reg_data = amdgpu_device_indirect_rreg_ext(adev, reg_addr); + if (AMDGPU_RAS_GPU_ERR_BOOT_STATUS(reg_data)) { + *boot_error = reg_data; + return 0; + } + msleep(1); + } + + *boot_error = reg_data; + return -ETIME; +} + +void amdgpu_ras_query_boot_status(struct amdgpu_device *adev, u32 num_instances) +{ + u32 boot_error = 0; + u32 i; + + for (i = 0; i < num_instances; i++) { + if (amdgpu_ras_wait_for_boot_complete(adev, i, &boot_error)) + amdgpu_ras_boot_time_error_reporting(adev, i, boot_error); + } +} |