diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 1136 |
1 files changed, 890 insertions, 246 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 8ebab6f22e5a..1a1395c5fff1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -120,7 +120,11 @@ 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 +#define MAX_UMC_POISON_POLLING_TIME_ASYNC 300 //ms + +#define AMDGPU_RAS_RETIRE_PAGE_INTERVAL 100 //ms + +#define MAX_FLUSH_RETIRE_DWORK_TIMES 100 enum amdgpu_ras_retire_page_reservation { AMDGPU_RAS_RETIRE_PAGE_RESERVED, @@ -878,7 +882,7 @@ int amdgpu_ras_feature_enable_on_boot(struct amdgpu_device *adev, if (ret) return ret; - /* gfx block ras dsiable cmd must send to ras-ta */ + /* gfx block ras disable cmd must send to ras-ta */ if (head->block == AMDGPU_RAS_BLOCK__GFX) con->features |= BIT(head->block); @@ -1045,6 +1049,7 @@ 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_manager *ras_mgr, struct ras_err_data *err_data, + struct ras_query_context *qctx, const char *blk_name, bool is_ue, bool is_de) @@ -1052,27 +1057,28 @@ static void amdgpu_ras_error_print_error_data(struct amdgpu_device *adev, struct amdgpu_smuio_mcm_config_info *mcm_info; struct ras_err_node *err_node; struct ras_err_info *err_info; + u64 event_id = qctx->evid.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) { - dev_info(adev->dev, "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); + 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); } } for_each_ras_error(err_node, &ras_mgr->err_data) { err_info = &err_node->err_info; mcm_info = &err_info->mcm_info; - dev_info(adev->dev, "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); + 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); } } else { @@ -1081,44 +1087,44 @@ static void amdgpu_ras_error_print_error_data(struct amdgpu_device *adev, err_info = &err_node->err_info; mcm_info = &err_info->mcm_info; if (err_info->de_count) { - dev_info(adev->dev, "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); + 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; - dev_info(adev->dev, "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); + 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) { - dev_info(adev->dev, "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); + 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; - dev_info(adev->dev, "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); + 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); } } } @@ -1131,77 +1137,79 @@ static inline bool err_data_has_source_info(struct ras_err_data *data) 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->evid.event_id; if (err_data->ce_count) { if (err_data_has_source_info(err_data)) { - amdgpu_ras_error_print_error_data(adev, ras_mgr, 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\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\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 (err_data_has_source_info(err_data)) { - amdgpu_ras_error_print_error_data(adev, ras_mgr, 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) { - 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 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); + 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, + 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 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); + 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 deferred hardware errors " - "detected in %s block\n", - ras_mgr->err_data.de_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); } } } @@ -1215,11 +1223,11 @@ 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_de_count(&obj->err_data, - &err_info->mcm_info, NULL, err_info->de_count); + &err_info->mcm_info, err_info->de_count); amdgpu_ras_error_statistic_ce_count(&obj->err_data, - &err_info->mcm_info, NULL, err_info->ce_count); + &err_info->mcm_info, err_info->ce_count); amdgpu_ras_error_statistic_ue_count(&obj->err_data, - &err_info->mcm_info, NULL, err_info->ue_count); + &err_info->mcm_info, err_info->ue_count); } } else { /* for legacy asic path which doesn't has error source info */ @@ -1244,6 +1252,10 @@ int amdgpu_ras_bind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk, { struct ras_manager *obj; + /* in resume phase, no need to create aca fs node */ + if (adev->in_suspend || amdgpu_in_reset(adev)) + return 0; + obj = get_ras_manager(adev, blk); if (!obj) return -EINVAL; @@ -1265,7 +1277,8 @@ int amdgpu_ras_unbind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk) } 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) + enum aca_error_type type, struct ras_err_data *err_data, + struct ras_query_context *qctx) { struct ras_manager *obj; @@ -1273,7 +1286,7 @@ static int amdgpu_aca_log_ras_error_data(struct amdgpu_device *adev, enum amdgpu if (!obj) return -EINVAL; - return amdgpu_aca_get_error_data(adev, &obj->aca_handle, type, err_data); + 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, @@ -1284,16 +1297,20 @@ ssize_t amdgpu_ras_aca_sysfs_read(struct device *dev, struct device_attribute *a .head = obj->head, }; + if (!amdgpu_ras_get_error_query_ready(obj->adev)) + return sysfs_emit(buf, "Query currently inaccessible\n"); + if (amdgpu_ras_query_error_status(obj->adev, &info)) return -EINVAL; - return sysfs_emit(buf, "%s: %lu\n%s: %lu\n", "ue", info.ue_count, - "ce", info.ce_count); + 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); } 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; @@ -1329,17 +1346,21 @@ static int amdgpu_ras_query_error_status_helper(struct amdgpu_device *adev, } } else { if (amdgpu_aca_is_enabled(adev)) { - ret = amdgpu_aca_log_ras_error_data(adev, blk, ACA_ERROR_TYPE_UE, err_data); + 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_CE, err_data); + 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); - amdgpu_mca_smu_log_ras_error(adev, blk, AMDGPU_MCA_ERROR_TYPE_CE, err_data); + 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); } } @@ -1347,10 +1368,13 @@ static int amdgpu_ras_query_error_status_helper(struct amdgpu_device *adev, } /* query/inject/cure begin */ -int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_if *info) +static int amdgpu_ras_query_error_status_with_event(struct amdgpu_device *adev, + struct ras_query_if *info, + enum ras_event_type type) { 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; @@ -1364,9 +1388,20 @@ int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_i if (!amdgpu_ras_get_error_query_mode(adev, &error_query_mode)) return -EINVAL; + memset(&qctx, 0, sizeof(qctx)); + qctx.evid.type = type; + qctx.evid.event_id = amdgpu_ras_acquire_event_id(adev, type); + + if (!down_read_trylock(&adev->reset_domain->sem)) { + ret = -EIO; + goto out_fini_err_data; + } + ret = amdgpu_ras_query_error_status_helper(adev, info, &err_data, + &qctx, error_query_mode); + up_read(&adev->reset_domain->sem); if (ret) goto out_fini_err_data; @@ -1376,7 +1411,7 @@ int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_i 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); @@ -1384,15 +1419,17 @@ out_fini_err_data: return ret; } +int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_if *info) +{ + return amdgpu_ras_query_error_status_with_event(adev, info, RAS_EVENT_TYPE_INVALID); +} + 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", @@ -1404,15 +1441,8 @@ int amdgpu_ras_reset_error_count(struct amdgpu_device *adev, !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) && + if ((amdgpu_in_reset(adev) || amdgpu_ras_in_recovery(adev)) && ((smu_funcs && smu_funcs->set_debug_mode) || (mca_funcs && mca_funcs->mca_set_debug_mode))) return -EOPNOTSUPP; @@ -1703,6 +1733,39 @@ static ssize_t amdgpu_ras_sysfs_schema_show(struct device *dev, return sysfs_emit(buf, "schema: 0x%x\n", con->schema); } +static struct { + enum ras_event_type type; + const char *name; +} dump_event[] = { + {RAS_EVENT_TYPE_FATAL, "Fatal Error"}, + {RAS_EVENT_TYPE_POISON_CREATION, "Poison Creation"}, + {RAS_EVENT_TYPE_POISON_CONSUMPTION, "Poison Consumption"}, +}; + +static ssize_t amdgpu_ras_sysfs_event_state_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct amdgpu_ras *con = + container_of(attr, struct amdgpu_ras, event_state_attr); + struct ras_event_manager *event_mgr = con->event_mgr; + struct ras_event_state *event_state; + int i, size = 0; + + if (!event_mgr) + return -EINVAL; + + size += sysfs_emit_at(buf, size, "current seqno: %llu\n", atomic64_read(&event_mgr->seqno)); + for (i = 0; i < ARRAY_SIZE(dump_event); i++) { + event_state = &event_mgr->event_state[dump_event[i].type]; + size += sysfs_emit_at(buf, size, "%s: count:%llu, last_seqno:%llu\n", + dump_event[i].name, + atomic64_read(&event_state->count), + event_state->last_seqno); + } + + return (ssize_t)size; +} + static void amdgpu_ras_sysfs_remove_bad_page_node(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -1720,6 +1783,7 @@ static int amdgpu_ras_sysfs_remove_dev_attr_node(struct amdgpu_device *adev) &con->features_attr.attr, &con->version_attr.attr, &con->schema_attr.attr, + &con->event_state_attr.attr, NULL }; struct attribute_group group = { @@ -1738,6 +1802,9 @@ int amdgpu_ras_sysfs_create(struct amdgpu_device *adev, { struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); + if (amdgpu_aca_is_enabled(adev)) + return 0; + if (!obj || obj->attr_inuse) return -EINVAL; @@ -1772,6 +1839,9 @@ int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev, { struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); + if (amdgpu_aca_is_enabled(adev)) + return 0; + if (!obj || !obj->attr_inuse) return -EINVAL; @@ -1884,6 +1954,23 @@ static void amdgpu_ras_debugfs_create(struct amdgpu_device *adev, obj, &amdgpu_ras_debugfs_ops); } +static bool amdgpu_ras_aca_is_supported(struct amdgpu_device *adev) +{ + bool ret; + + switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) { + case IP_VERSION(13, 0, 6): + case IP_VERSION(13, 0, 14): + ret = true; + break; + default: + ret = false; + break; + } + + return ret; +} + void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -1910,10 +1997,12 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) } } - if (amdgpu_aca_is_enabled(adev)) - amdgpu_aca_smu_debugfs_init(adev, dir); - else - amdgpu_mca_smu_debugfs_init(adev, dir); + if (amdgpu_ras_aca_is_supported(adev)) { + if (amdgpu_aca_is_enabled(adev)) + amdgpu_aca_smu_debugfs_init(adev, dir); + else + amdgpu_mca_smu_debugfs_init(adev, dir); + } } /* debugfs end */ @@ -1927,6 +2016,8 @@ static DEVICE_ATTR(version, 0444, amdgpu_ras_sysfs_version_show, NULL); static DEVICE_ATTR(schema, 0444, amdgpu_ras_sysfs_schema_show, NULL); +static DEVICE_ATTR(event_state, 0444, + amdgpu_ras_sysfs_event_state_show, NULL); static int amdgpu_ras_fs_init(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -1937,6 +2028,7 @@ static int amdgpu_ras_fs_init(struct amdgpu_device *adev) &con->features_attr.attr, &con->version_attr.attr, &con->schema_attr.attr, + &con->event_state_attr.attr, NULL }; struct bin_attribute *bin_attrs[] = { @@ -1959,6 +2051,10 @@ static int amdgpu_ras_fs_init(struct amdgpu_device *adev) con->schema_attr = dev_attr_schema; sysfs_attr_init(attrs[2]); + /* add event_state entry */ + con->event_state_attr = dev_attr_event_state; + sysfs_attr_init(attrs[3]); + if (amdgpu_bad_page_threshold != 0) { /* add bad_page_features entry */ bin_attr_gpu_vram_bad_pages.private = NULL; @@ -2022,8 +2118,16 @@ 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); + enum ras_event_type type = RAS_EVENT_TYPE_POISON_CONSUMPTION; + u64 event_id; + int ret; - if (!block_obj) + if (!block_obj || !con) + return; + + ret = amdgpu_ras_mark_ras_event(adev, type); + if (ret) return; /* both query_poison_status and handle_poison_consumption are optional, @@ -2041,26 +2145,49 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager * } } - amdgpu_umc_poison_handler(adev, obj->head.block, 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); - /* gpu reset is fallback for failed and default cases */ - if (poison_stat) { - dev_info(adev->dev, "GPU reset for %s RAS poison consumption is issued!\n", - block_obj->ras_comm.name); + /* gpu reset is fallback for failed and default cases. + * For RMA case, amdgpu_umc_poison_handler will handle gpu reset. + */ + if (poison_stat && !amdgpu_ras_is_rma(adev)) { + event_id = amdgpu_ras_acquire_event_id(adev, type); + RAS_EVENT_LOG(adev, event_id, + "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, struct amdgpu_iv_entry *entry) { - dev_info(obj->adev->dev, - "Poison is created\n"); + struct amdgpu_device *adev = obj->adev; + enum ras_event_type type = RAS_EVENT_TYPE_POISON_CREATION; + u64 event_id; + int ret; + + ret = amdgpu_ras_mark_ras_event(adev, type); + if (ret) + return; + + event_id = amdgpu_ras_acquire_event_id(adev, type); + RAS_EVENT_LOG(adev, event_id, "Poison is created\n"); + + if (amdgpu_ip_version(obj->adev, UMC_HWIP, 0) >= IP_VERSION(12, 0, 0)) { + struct amdgpu_ras *con = amdgpu_ras_get_context(obj->adev); + + atomic_inc(&con->page_retirement_req_cnt); + atomic_inc(&con->poison_creation_count); + + wake_up(&con->page_retirement_wq); + } } static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj, @@ -2080,6 +2207,7 @@ static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj, /* Let IP handle its data, maybe we need get the output * from the callback to update the error type/count, etc */ + amdgpu_ras_set_fed(obj->adev, true); ret = data->cb(obj->adev, &err_data, entry); /* ue will trigger an interrupt, and in that case * we need do a reset to recovery the whole system. @@ -2140,12 +2268,15 @@ static void amdgpu_ras_interrupt_process_handler(struct work_struct *work) int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev, struct ras_dispatch_if *info) { - struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); - struct ras_ih_data *data = &obj->ih_data; + struct ras_manager *obj; + struct ras_ih_data *data; + obj = amdgpu_ras_find_obj(adev, &info->head); if (!obj) return -EINVAL; + data = &obj->ih_data; + if (data->inuse == 0) return 0; @@ -2242,7 +2373,7 @@ static int amdgpu_ras_interrupt_remove_all(struct amdgpu_device *adev) /* ih end */ /* traversal all IPs except NBIO to query error counter */ -static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev) +static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev, enum ras_event_type type) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); struct ras_manager *obj; @@ -2275,7 +2406,7 @@ static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev) IP_VERSION(13, 0, 2))) continue; - amdgpu_ras_query_error_status(adev, &info); + amdgpu_ras_query_error_status_with_event(adev, &info, type); if (amdgpu_ip_version(adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 2) && @@ -2371,7 +2502,7 @@ static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, .flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED, }; status = amdgpu_vram_mgr_query_page_status(&adev->mman.vram_mgr, - data->bps[i].retired_page); + data->bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT); if (status == -EBUSY) (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING; else if (status == -ENOENT) @@ -2384,6 +2515,44 @@ 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); + } +} + +bool amdgpu_ras_in_recovery(struct amdgpu_device *adev) +{ + struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); + struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + int hive_ras_recovery = 0; + + if (hive) { + hive_ras_recovery = atomic_read(&hive->ras_recovery); + amdgpu_put_xgmi_hive(hive); + } + + if (ras && (atomic_read(&ras->in_recovery) || hive_ras_recovery)) + return true; + + return false; +} + +static enum ras_event_type amdgpu_ras_get_fatal_error_event(struct amdgpu_device *adev) +{ + if (amdgpu_ras_intr_triggered()) + return RAS_EVENT_TYPE_FATAL; + else + return RAS_EVENT_TYPE_POISON_CONSUMPTION; +} + static void amdgpu_ras_do_recovery(struct work_struct *work) { struct amdgpu_ras *ras = @@ -2392,9 +2561,23 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) 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); + enum ras_event_type type; - if (hive) + 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) { /* Build list of devices to query RAS related errors */ @@ -2406,10 +2589,11 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) device_list_handle = &device_list; } + type = amdgpu_ras_get_fatal_error_event(adev); list_for_each_entry(remote_adev, device_list_handle, gmc.xgmi.head) { amdgpu_ras_query_err_status(remote_adev); - amdgpu_ras_log_on_err_counter(remote_adev); + amdgpu_ras_log_on_err_counter(remote_adev, type); } } @@ -2420,6 +2604,7 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) reset_context.method = AMD_RESET_METHOD_NONE; reset_context.reset_req_dev = adev; + reset_context.src = AMDGPU_RESET_SRC_RAS; /* Perform full reset in fatal error mode */ if (!amdgpu_ras_is_poison_mode_supported(ras->adev)) @@ -2439,18 +2624,6 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) ras->gpu_reset_flags &= ~AMDGPU_RAS_GPU_RESET_MODE1_RESET; set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); - /* For any RAS error that needs a full reset to - * recover, set the fatal error status - */ - if (hive) { - list_for_each_entry(remote_adev, - &hive->device_list, - gmc.xgmi.head) - amdgpu_ras_set_fed(remote_adev, - true); - } else { - amdgpu_ras_set_fed(adev, true); - } psp_fatal_error_recovery_quirk(&adev->psp); } } @@ -2516,9 +2689,7 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, goto out; } - amdgpu_vram_mgr_reserve_range(&adev->mman.vram_mgr, - bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT, - AMDGPU_GPU_PAGE_SIZE); + amdgpu_ras_reserve_page(adev, bps[i].retired_page); memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps)); data->count++; @@ -2674,10 +2845,236 @@ static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev, } } +int amdgpu_ras_put_poison_req(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) +{ + int ret = 0; + struct ras_poison_msg poison_msg; + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + memset(&poison_msg, 0, sizeof(poison_msg)); + poison_msg.block = block; + poison_msg.pasid = pasid; + poison_msg.reset = reset; + poison_msg.pasid_fn = pasid_fn; + poison_msg.data = data; + + ret = kfifo_put(&con->poison_fifo, poison_msg); + if (!ret) { + dev_err(adev->dev, "Poison message fifo is full!\n"); + return -ENOSPC; + } + + return 0; +} + +static int amdgpu_ras_get_poison_req(struct amdgpu_device *adev, + struct ras_poison_msg *poison_msg) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + return kfifo_get(&con->poison_fifo, poison_msg); +} + +static void amdgpu_ras_ecc_log_init(struct ras_ecc_log_info *ecc_log) +{ + mutex_init(&ecc_log->lock); + + INIT_RADIX_TREE(&ecc_log->de_page_tree, GFP_KERNEL); + ecc_log->de_queried_count = 0; + ecc_log->prev_de_queried_count = 0; +} + +static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) +{ + struct radix_tree_iter iter; + void __rcu **slot; + struct ras_ecc_err *ecc_err; + + mutex_lock(&ecc_log->lock); + radix_tree_for_each_slot(slot, &ecc_log->de_page_tree, &iter, 0) { + ecc_err = radix_tree_deref_slot(slot); + kfree(ecc_err->err_pages.pfn); + kfree(ecc_err); + radix_tree_iter_delete(&ecc_log->de_page_tree, &iter, slot); + } + mutex_unlock(&ecc_log->lock); + + mutex_destroy(&ecc_log->lock); + ecc_log->de_queried_count = 0; + ecc_log->prev_de_queried_count = 0; +} + +static bool amdgpu_ras_schedule_retirement_dwork(struct amdgpu_ras *con, + uint32_t delayed_ms) +{ + int ret; + + mutex_lock(&con->umc_ecc_log.lock); + ret = radix_tree_tagged(&con->umc_ecc_log.de_page_tree, + UMC_ECC_NEW_DETECTED_TAG); + mutex_unlock(&con->umc_ecc_log.lock); + + if (ret) + schedule_delayed_work(&con->page_retirement_dwork, + msecs_to_jiffies(delayed_ms)); + + return ret ? true : false; +} + +static void amdgpu_ras_do_page_retirement(struct work_struct *work) +{ + struct amdgpu_ras *con = container_of(work, struct amdgpu_ras, + page_retirement_dwork.work); + struct amdgpu_device *adev = con->adev; + struct ras_err_data err_data; + unsigned long err_cnt; + + /* If gpu reset is ongoing, delay retiring the bad pages */ + if (amdgpu_in_reset(adev) || amdgpu_ras_in_recovery(adev)) { + amdgpu_ras_schedule_retirement_dwork(con, + AMDGPU_RAS_RETIRE_PAGE_INTERVAL * 3); + return; + } + + 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 && amdgpu_ras_is_rma(adev)) + amdgpu_ras_reset_gpu(adev); + + amdgpu_ras_schedule_retirement_dwork(con, + AMDGPU_RAS_RETIRE_PAGE_INTERVAL); +} + +static int amdgpu_ras_poison_creation_handler(struct amdgpu_device *adev, + uint32_t poison_creation_count) +{ + int ret = 0; + struct ras_ecc_log_info *ecc_log; + struct ras_query_if info; + uint32_t timeout = 0; + struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + uint64_t de_queried_count; + uint32_t new_detect_count, total_detect_count; + uint32_t need_query_count = poison_creation_count; + bool query_data_timeout = false; + enum ras_event_type type = RAS_EVENT_TYPE_POISON_CREATION; + + memset(&info, 0, sizeof(info)); + info.head.block = AMDGPU_RAS_BLOCK__UMC; + + ecc_log = &ras->umc_ecc_log; + total_detect_count = 0; + do { + ret = amdgpu_ras_query_error_status_with_event(adev, &info, type); + if (ret) + return ret; + + de_queried_count = ecc_log->de_queried_count; + if (de_queried_count > ecc_log->prev_de_queried_count) { + new_detect_count = de_queried_count - ecc_log->prev_de_queried_count; + ecc_log->prev_de_queried_count = de_queried_count; + timeout = 0; + } else { + new_detect_count = 0; + } + + if (new_detect_count) { + total_detect_count += new_detect_count; + } else { + if (!timeout && need_query_count) + timeout = MAX_UMC_POISON_POLLING_TIME_ASYNC; + + if (timeout) { + if (!--timeout) { + query_data_timeout = true; + break; + } + msleep(1); + } + } + } while (total_detect_count < need_query_count); + + if (query_data_timeout) { + dev_warn(adev->dev, "Can't find deferred error! count: %u\n", + (need_query_count - total_detect_count)); + return -ENOENT; + } + + if (total_detect_count) + schedule_delayed_work(&ras->page_retirement_dwork, 0); + + return 0; +} + +static void amdgpu_ras_clear_poison_fifo(struct amdgpu_device *adev) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_poison_msg msg; + int ret; + + do { + ret = kfifo_get(&con->poison_fifo, &msg); + } while (ret); +} + +static int amdgpu_ras_poison_consumption_handler(struct amdgpu_device *adev, + uint32_t msg_count, uint32_t *gpu_reset) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + uint32_t reset_flags = 0, reset = 0; + struct ras_poison_msg msg; + int ret, i; + + kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); + + for (i = 0; i < msg_count; i++) { + ret = amdgpu_ras_get_poison_req(adev, &msg); + if (!ret) + continue; + + if (msg.pasid_fn) + msg.pasid_fn(adev, msg.pasid, msg.data); + + reset_flags |= msg.reset; + } + + /* for RMA, amdgpu_ras_poison_creation_handler will trigger gpu reset */ + if (reset_flags && !amdgpu_ras_is_rma(adev)) { + if (reset_flags & AMDGPU_RAS_GPU_RESET_MODE1_RESET) + reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET; + else if (reset_flags & AMDGPU_RAS_GPU_RESET_MODE2_RESET) + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET; + else + reset = reset_flags; + + flush_delayed_work(&con->page_retirement_dwork); + + con->gpu_reset_flags |= reset; + amdgpu_ras_reset_gpu(adev); + + *gpu_reset = reset; + + /* Wait for gpu recovery to complete */ + flush_work(&con->recovery_work); + } + + return 0; +} + 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); + uint32_t poison_creation_count, msg_count; + uint32_t gpu_reset; + int ret; while (!kthread_should_stop()) { @@ -2688,13 +3085,62 @@ static int amdgpu_ras_page_retirement_thread(void *param) if (kthread_should_stop()) break; - dev_info(adev->dev, "Start processing page retirement. request:%d\n", - atomic_read(&con->page_retirement_req_cnt)); + gpu_reset = 0; + + do { + poison_creation_count = atomic_read(&con->poison_creation_count); + ret = amdgpu_ras_poison_creation_handler(adev, poison_creation_count); + if (ret == -EIO) + break; + + if (poison_creation_count) { + atomic_sub(poison_creation_count, &con->poison_creation_count); + atomic_sub(poison_creation_count, &con->page_retirement_req_cnt); + } + } while (atomic_read(&con->poison_creation_count)); + + if (ret != -EIO) { + msg_count = kfifo_len(&con->poison_fifo); + if (msg_count) { + ret = amdgpu_ras_poison_consumption_handler(adev, + msg_count, &gpu_reset); + if ((ret != -EIO) && + (gpu_reset != AMDGPU_RAS_GPU_RESET_MODE1_RESET)) + atomic_sub(msg_count, &con->page_retirement_req_cnt); + } + } + + if ((ret == -EIO) || (gpu_reset == AMDGPU_RAS_GPU_RESET_MODE1_RESET)) { + /* gpu mode-1 reset is ongoing or just completed ras mode-1 reset */ + /* Clear poison creation request */ + atomic_set(&con->poison_creation_count, 0); + + /* Clear poison fifo */ + amdgpu_ras_clear_poison_fifo(adev); + + /* Clear all poison requests */ + atomic_set(&con->page_retirement_req_cnt, 0); - atomic_dec(&con->page_retirement_req_cnt); + if (ret == -EIO) { + /* Wait for mode-1 reset to complete */ + down_read(&adev->reset_domain->sem); + up_read(&adev->reset_domain->sem); + } + + /* Wake up work to save bad pages to eeprom */ + schedule_delayed_work(&con->page_retirement_dwork, 0); + } else if (gpu_reset) { + /* gpu just completed mode-2 reset or other reset */ + /* Clear poison consumption messages cached in fifo */ + msg_count = kfifo_len(&con->poison_fifo); + if (msg_count) { + amdgpu_ras_clear_poison_fifo(adev); + atomic_sub(msg_count, &con->page_retirement_req_cnt); + } - amdgpu_umc_bad_page_polling_timeout(adev, - false, MAX_UMC_POISON_POLLING_TIME_ASYNC); + /* Wake up work to save bad pages to eeprom */ + schedule_delayed_work(&con->page_retirement_dwork, 0); + } } return 0; @@ -2705,7 +3151,6 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) struct amdgpu_ras *con = amdgpu_ras_get_context(adev); struct ras_err_handler_data **data; u32 max_eeprom_records_count = 0; - bool exc_err_limit = false; int ret; if (!con || amdgpu_sriov_vf(adev)) @@ -2742,12 +3187,12 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) */ if (adev->gmc.xgmi.pending_reset) return 0; - ret = amdgpu_ras_eeprom_init(&con->eeprom_control, &exc_err_limit); + ret = amdgpu_ras_eeprom_init(&con->eeprom_control); /* - * This calling fails when exc_err_limit is true or + * This calling fails when is_rma is true or * ret != 0. */ - if (exc_err_limit || ret) + if (amdgpu_ras_is_rma(adev) || ret) goto free; if (con->eeprom_control.ras_num_recs) { @@ -2763,9 +3208,12 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) } } + mutex_init(&con->page_rsv_lock); + INIT_KFIFO(con->poison_fifo); mutex_init(&con->page_retirement_lock); init_waitqueue_head(&con->page_retirement_wq); atomic_set(&con->page_retirement_req_cnt, 0); + atomic_set(&con->poison_creation_count, 0); con->page_retirement_thread = kthread_run(amdgpu_ras_page_retirement_thread, adev, "umc_page_retirement"); if (IS_ERR(con->page_retirement_thread)) { @@ -2773,6 +3221,8 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) dev_warn(adev->dev, "Failed to create umc_page_retirement thread!!!\n"); } + INIT_DELAYED_WORK(&con->page_retirement_dwork, amdgpu_ras_do_page_retirement); + amdgpu_ras_ecc_log_init(&con->umc_ecc_log); #ifdef CONFIG_X86_MCE_AMD if ((adev->asic_type == CHIP_ALDEBARAN) && (adev->gmc.xgmi.connected_to_cpu)) @@ -2791,7 +3241,7 @@ out: * Except error threshold exceeding case, other failure cases in this * function would not fail amdgpu driver init. */ - if (!exc_err_limit) + if (!amdgpu_ras_is_rma(adev)) ret = 0; else ret = -EINVAL; @@ -2803,18 +3253,33 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); struct ras_err_handler_data *data = con->eh_data; + int max_flush_timeout = MAX_FLUSH_RETIRE_DWORK_TIMES; + bool ret; /* recovery_init failed to init it, fini is useless */ if (!data) return 0; + /* Save all cached bad pages to eeprom */ + do { + flush_delayed_work(&con->page_retirement_dwork); + ret = amdgpu_ras_schedule_retirement_dwork(con, 0); + } while (ret && max_flush_timeout--); + if (con->page_retirement_thread) kthread_stop(con->page_retirement_thread); atomic_set(&con->page_retirement_req_cnt, 0); + atomic_set(&con->poison_creation_count, 0); + + mutex_destroy(&con->page_rsv_lock); cancel_work_sync(&con->recovery_work); + cancel_delayed_work_sync(&con->page_retirement_dwork); + + amdgpu_ras_ecc_log_fini(&con->umc_ecc_log); + mutex_lock(&con->recovery_lock); con->eh_data = NULL; kfree(data->bps); @@ -2831,6 +3296,7 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev) switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) { case IP_VERSION(13, 0, 2): case IP_VERSION(13, 0, 6): + case IP_VERSION(13, 0, 14): return true; default: return false; @@ -2842,6 +3308,7 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev) case IP_VERSION(13, 0, 0): case IP_VERSION(13, 0, 6): case IP_VERSION(13, 0, 10): + case IP_VERSION(13, 0, 14): return true; default: return false; @@ -3001,6 +3468,11 @@ init_ras_enabled_flag: /* aca is disabled by default */ adev->aca.is_enabled = false; + + /* bad page feature is not applicable to specific app platform */ + if (adev->gmc.is_app_apu && + amdgpu_ip_version(adev, UMC_HWIP, 0) == IP_VERSION(12, 0, 0)) + amdgpu_bad_page_threshold = 0; } static void amdgpu_ras_counte_dw(struct work_struct *work) @@ -3036,6 +3508,60 @@ static int amdgpu_get_ras_schema(struct amdgpu_device *adev) AMDGPU_RAS_ERROR__PARITY; } +static void ras_event_mgr_init(struct ras_event_manager *mgr) +{ + struct ras_event_state *event_state; + int i; + + memset(mgr, 0, sizeof(*mgr)); + atomic64_set(&mgr->seqno, 0); + + for (i = 0; i < ARRAY_SIZE(mgr->event_state); i++) { + event_state = &mgr->event_state[i]; + event_state->last_seqno = RAS_EVENT_INVALID_ID; + atomic64_set(&event_state->count, 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; + + hive = amdgpu_get_xgmi_hive(adev); + ras->event_mgr = hive ? &hive->event_mgr : &ras->__event_mgr; + + /* 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); + } + + if (hive) + amdgpu_put_xgmi_hive(hive); +} + +static void amdgpu_ras_init_reserved_vram_size(struct amdgpu_device *adev) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + if (!con || (adev->flags & AMD_IS_APU)) + return; + + switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) { + case IP_VERSION(13, 0, 2): + case IP_VERSION(13, 0, 6): + case IP_VERSION(13, 0, 14): + con->reserved_pages_in_bytes = AMDGPU_RAS_RESERVED_VRAM_SIZE; + break; + default: + break; + } +} + int amdgpu_ras_init(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -3141,11 +3667,22 @@ int amdgpu_ras_init(struct amdgpu_device *adev) /* Get RAS schema for particular SOC */ con->schema = amdgpu_get_ras_schema(adev); + amdgpu_ras_init_reserved_vram_size(adev); + if (amdgpu_ras_fs_init(adev)) { r = -EINVAL; goto release_con; } + if (amdgpu_ras_aca_is_supported(adev)) { + if (amdgpu_aca_is_enabled(adev)) + r = amdgpu_aca_init(adev); + else + r = amdgpu_mca_init(adev); + if (r) + goto release_con; + } + dev_info(adev->dev, "RAS INFO: ras initialized successfully, " "hardware ability[%x] ras_mask[%x]\n", adev->ras_hw_enabled, adev->ras_enabled); @@ -3352,23 +3889,30 @@ int amdgpu_ras_late_init(struct amdgpu_device *adev) struct amdgpu_ras_block_object *obj; int r; - /* Guest side doesn't need init ras feature */ - 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; + if (amdgpu_ras_aca_is_supported(adev)) { + if (amdgpu_in_reset(adev)) { + if (amdgpu_aca_is_enabled(adev)) + r = amdgpu_aca_reset(adev); + else + r = amdgpu_mca_reset(adev); + if (r) + return r; + } - amdgpu_ras_set_aca_debug_mode(adev, false); - } else { - amdgpu_ras_set_mca_debug_mode(adev, false); + if (!amdgpu_sriov_vf(adev)) { + if (amdgpu_aca_is_enabled(adev)) + amdgpu_ras_set_aca_debug_mode(adev, false); + else + amdgpu_ras_set_mca_debug_mode(adev, false); + } } + /* Guest side doesn't need init ras feature */ + if (amdgpu_sriov_vf(adev)) + return 0; + list_for_each_entry_safe(node, tmp, &adev->ras_list, node) { obj = node->ras_obj; if (!obj) { @@ -3436,8 +3980,12 @@ int amdgpu_ras_fini(struct amdgpu_device *adev) amdgpu_ras_fs_fini(adev); amdgpu_ras_interrupt_remove_all(adev); - if (amdgpu_aca_is_enabled(adev)) - amdgpu_aca_fini(adev); + if (amdgpu_ras_aca_is_supported(adev)) { + if (amdgpu_aca_is_enabled(adev)) + amdgpu_aca_fini(adev); + else + amdgpu_mca_fini(adev); + } WARN(AMDGPU_RAS_GET_FEATURES(con->features), "Feature mask is not cleared"); @@ -3472,14 +4020,90 @@ void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status) atomic_set(&ras->fed, !!status); } +static struct ras_event_manager *__get_ras_event_mgr(struct amdgpu_device *adev) +{ + struct amdgpu_ras *ras; + + ras = amdgpu_ras_get_context(adev); + if (!ras) + return NULL; + + return ras->event_mgr; +} + +int amdgpu_ras_mark_ras_event_caller(struct amdgpu_device *adev, enum ras_event_type type, + const void *caller) +{ + struct ras_event_manager *event_mgr; + struct ras_event_state *event_state; + int ret = 0; + + if (type >= RAS_EVENT_TYPE_COUNT) { + ret = -EINVAL; + goto out; + } + + event_mgr = __get_ras_event_mgr(adev); + if (!event_mgr) { + ret = -EINVAL; + goto out; + } + + event_state = &event_mgr->event_state[type]; + event_state->last_seqno = atomic64_inc_return(&event_mgr->seqno); + atomic64_inc(&event_state->count); + +out: + if (ret && caller) + dev_warn(adev->dev, "failed mark ras event (%d) in %ps, ret:%d\n", + (int)type, caller, ret); + + return ret; +} + +u64 amdgpu_ras_acquire_event_id(struct amdgpu_device *adev, enum ras_event_type type) +{ + struct ras_event_manager *event_mgr; + u64 id; + + if (type >= RAS_EVENT_TYPE_COUNT) + return RAS_EVENT_INVALID_ID; + + switch (type) { + case RAS_EVENT_TYPE_FATAL: + case RAS_EVENT_TYPE_POISON_CREATION: + case RAS_EVENT_TYPE_POISON_CONSUMPTION: + event_mgr = __get_ras_event_mgr(adev); + if (!event_mgr) + return RAS_EVENT_INVALID_ID; + + id = event_mgr->event_state[type].last_seqno; + break; + case RAS_EVENT_TYPE_INVALID: + default: + id = RAS_EVENT_INVALID_ID; + 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); + enum ras_event_type type = RAS_EVENT_TYPE_FATAL; + u64 event_id; + + if (amdgpu_ras_mark_ras_event(adev, type)) + return; + + event_id = amdgpu_ras_acquire_event_id(adev, type); - 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); } @@ -3664,6 +4288,12 @@ 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 (amdgpu_ras_is_rma(adev)) { + 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; @@ -3983,8 +4613,6 @@ 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++; @@ -3994,32 +4622,9 @@ static struct ras_err_info *amdgpu_ras_error_get_info(struct ras_err_data *err_d 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, - struct ras_err_addr *err_addr, u64 count) + struct amdgpu_smuio_mcm_config_info *mcm_info, + u64 count) { struct ras_err_info *err_info; @@ -4033,9 +4638,6 @@ 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; @@ -4043,8 +4645,8 @@ 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, - struct ras_err_addr *err_addr, u64 count) + struct amdgpu_smuio_mcm_config_info *mcm_info, + u64 count) { struct ras_err_info *err_info; @@ -4065,8 +4667,8 @@ int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data, } 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 amdgpu_smuio_mcm_config_info *mcm_info, + u64 count) { struct ras_err_info *err_info; @@ -4080,9 +4682,6 @@ int amdgpu_ras_error_statistic_de_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->de_count += count; err_data->de_count += count; @@ -4092,64 +4691,84 @@ int amdgpu_ras_error_statistic_de_count(struct ras_err_data *err_data, #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 instance) { u32 socket_id, aid_id, hbm_id; - u32 reg_data; + u32 fw_status; + u32 boot_error; 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); + fw_status = amdgpu_device_indirect_rreg_ext(adev, reg_addr); - 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); + reg_addr = (mmMP0_SMN_C2PMSG_126 << 2) + + aqua_vanjaram_encode_ext_smn_addressing(instance); + boot_error = amdgpu_device_indirect_rreg_ext(adev, reg_addr); + + socket_id = AMDGPU_RAS_GPU_ERR_SOCKET_ID(boot_error); + aid_id = AMDGPU_RAS_GPU_ERR_AID_ID(boot_error); + hbm_id = ((1 == AMDGPU_RAS_GPU_ERR_HBM_ID(boot_error)) ? 0 : 1); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, hbm: %d, fw_status: 0x%x, memory training failed\n", + socket_id, aid_id, hbm_id, fw_status); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, firmware load failed at boot time\n", + socket_id, aid_id, fw_status); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, wafl link training failed\n", + socket_id, aid_id, fw_status); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, xgmi link training failed\n", + socket_id, aid_id, fw_status); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, usr cp link training failed\n", + socket_id, aid_id, fw_status); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, usr dp link training failed\n", + socket_id, aid_id, fw_status); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, hbm: %d, fw_status: 0x%x, hbm memory test failed\n", + socket_id, aid_id, hbm_id, fw_status); 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); + dev_info(adev->dev, + "socket: %d, aid: %d, hbm: %d, fw_status: 0x%x, hbm bist test failed\n", + socket_id, aid_id, hbm_id, fw_status); + + if (AMDGPU_RAS_GPU_ERR_DATA_ABORT(boot_error)) + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, data abort exception\n", + socket_id, aid_id, fw_status); + + if (AMDGPU_RAS_GPU_ERR_UNKNOWN(boot_error)) + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, unknown boot time errors\n", + socket_id, aid_id, fw_status); } -static int amdgpu_ras_wait_for_boot_complete(struct amdgpu_device *adev, - u32 instance, u32 *boot_error) +static bool amdgpu_ras_boot_error_detected(struct amdgpu_device *adev, + u32 instance) { - u32 reg_addr; + u64 reg_addr; u32 reg_data; int retry_loop; @@ -4158,40 +4777,65 @@ static int amdgpu_ras_wait_for_boot_complete(struct amdgpu_device *adev, 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); + if ((reg_data & AMDGPU_RAS_BOOT_STATUS_MASK) == AMDGPU_RAS_BOOT_STEADY_STATUS) + return false; + else + msleep(1); } - *boot_error = reg_data; - return -ETIME; + return true; } 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); + if (amdgpu_ras_boot_error_detected(adev, i)) + amdgpu_ras_boot_time_error_reporting(adev, i); } } + +int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr; + uint64_t start = pfn << AMDGPU_GPU_PAGE_SHIFT; + int ret = 0; + + mutex_lock(&con->page_rsv_lock); + ret = amdgpu_vram_mgr_query_page_status(mgr, start); + if (ret == -ENOENT) + ret = amdgpu_vram_mgr_reserve_range(mgr, start, AMDGPU_GPU_PAGE_SIZE); + mutex_unlock(&con->page_rsv_lock); + + return ret; +} + +void amdgpu_ras_event_log_print(struct amdgpu_device *adev, u64 event_id, + const char *fmt, ...) +{ + struct va_format vaf; + va_list args; + + va_start(args, fmt); + vaf.fmt = fmt; + vaf.va = &args; + + if (RAS_EVENT_ID_IS_VALID(event_id)) + dev_printk(KERN_INFO, adev->dev, "{%llu}%pV", event_id, &vaf); + else + dev_printk(KERN_INFO, adev->dev, "%pV", &vaf); + + va_end(args); +} + +bool amdgpu_ras_is_rma(struct amdgpu_device *adev) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + if (!con) + return false; + + return con->is_rma; +} |