diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2047 |
1 files changed, 1646 insertions, 401 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 1adc81a55734..2a6cf7963dde 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -36,10 +36,12 @@ #include "amdgpu_xgmi.h" #include "ivsrcid/nbio/irqsrcs_nbif_7_4.h" #include "nbio_v4_3.h" +#include "nbif_v6_3_1.h" #include "nbio_v7_9.h" #include "atom.h" #include "amdgpu_reset.h" #include "amdgpu_psp.h" +#include "amdgpu_ras_mgr.h" #ifdef CONFIG_X86_MCE_AMD #include <asm/mce.h> @@ -76,6 +78,7 @@ const char *ras_block_string[] = { "jpeg", "ih", "mpio", + "mmsch", }; const char *ras_mca_block_string[] = { @@ -120,10 +123,15 @@ 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 10 #define AMDGPU_RAS_RETIRE_PAGE_INTERVAL 100 //ms +#define MAX_FLUSH_RETIRE_DWORK_TIMES 100 + +#define BYPASS_ALLOCATED_ADDRESS 0x0 +#define BYPASS_INITIALIZATION_ADDRESS 0x1 + enum amdgpu_ras_retire_page_reservation { AMDGPU_RAS_RETIRE_PAGE_RESERVED, AMDGPU_RAS_RETIRE_PAGE_PENDING, @@ -132,12 +140,18 @@ enum amdgpu_ras_retire_page_reservation { atomic_t amdgpu_ras_in_intr = ATOMIC_INIT(0); -static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, +static int amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, uint64_t addr); -static bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev, +static int amdgpu_ras_check_bad_page(struct amdgpu_device *adev, uint64_t addr); + +static void amdgpu_ras_critical_region_init(struct amdgpu_device *adev); +static void amdgpu_ras_critical_region_fini(struct amdgpu_device *adev); + #ifdef CONFIG_X86_MCE_AMD static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev); +static void +amdgpu_unregister_bad_pages_mca_notifier(struct amdgpu_device *adev); struct mce_notifier_adev_list { struct amdgpu_device *devs[MAX_GPU_INSTANCE]; int num_gpu; @@ -165,18 +179,16 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre struct eeprom_table_record err_rec; int ret; - if ((address >= adev->gmc.mc_vram_size) || - (address >= RAS_UMC_INJECT_ADDR_LIMIT)) { + ret = amdgpu_ras_check_bad_page(adev, address); + if (ret == -EINVAL) { dev_warn(adev->dev, - "RAS WARN: input address 0x%llx is invalid.\n", - address); + "RAS WARN: input address 0x%llx is invalid.\n", + address); return -EINVAL; - } - - if (amdgpu_ras_check_bad_page(adev, address)) { + } else if (ret == 1) { dev_warn(adev->dev, - "RAS WARN: 0x%llx has already been marked as bad page!\n", - address); + "RAS WARN: 0x%llx has already been marked as bad page!\n", + address); return 0; } @@ -190,7 +202,7 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre if (amdgpu_bad_page_threshold != 0) { amdgpu_ras_add_bad_pages(adev, err_data.err_addr, - err_data.err_addr_cnt); + err_data.err_addr_cnt, false); amdgpu_ras_save_bad_pages(adev, NULL); } @@ -203,6 +215,56 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre return 0; } +static int amdgpu_check_address_validity(struct amdgpu_device *adev, + uint64_t address, uint64_t flags) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct amdgpu_vram_block_info blk_info; + uint64_t page_pfns[32] = {0}; + int i, ret, count; + bool hit = false; + + if (amdgpu_ip_version(adev, UMC_HWIP, 0) < IP_VERSION(12, 0, 0)) + return 0; + + if (amdgpu_sriov_vf(adev)) { + if (amdgpu_virt_check_vf_critical_region(adev, address, &hit)) + return -EPERM; + return hit ? -EACCES : 0; + } + + if ((address >= adev->gmc.mc_vram_size) || + (address >= RAS_UMC_INJECT_ADDR_LIMIT)) + return -EFAULT; + + count = amdgpu_umc_lookup_bad_pages_in_a_row(adev, + address, page_pfns, ARRAY_SIZE(page_pfns)); + if (count <= 0) + return -EPERM; + + for (i = 0; i < count; i++) { + memset(&blk_info, 0, sizeof(blk_info)); + ret = amdgpu_vram_mgr_query_address_block_info(&adev->mman.vram_mgr, + page_pfns[i] << AMDGPU_GPU_PAGE_SHIFT, &blk_info); + if (!ret) { + /* The input address that needs to be checked is allocated by + * current calling process, so it is necessary to exclude + * the calling process. + */ + if ((flags == BYPASS_ALLOCATED_ADDRESS) && + ((blk_info.task.pid != task_pid_nr(current)) || + strncmp(blk_info.task.comm, current->comm, TASK_COMM_LEN))) + return -EACCES; + else if ((flags == BYPASS_INITIALIZATION_ADDRESS) && + (blk_info.task.pid == con->init_task_pid) && + !strncmp(blk_info.task.comm, con->init_task_comm, TASK_COMM_LEN)) + return -EACCES; + } + } + + return 0; +} + static ssize_t amdgpu_ras_debugfs_read(struct file *f, char __user *buf, size_t size, loff_t *pos) { @@ -293,6 +355,8 @@ static int amdgpu_ras_debugfs_ctrl_parse_data(struct file *f, op = 2; else if (strstr(str, "retire_page") != NULL) op = 3; + else if (strstr(str, "check_address") != NULL) + op = 4; else if (str[0] && str[1] && str[2] && str[3]) /* ascii string, but commands are not matched. */ return -EINVAL; @@ -307,6 +371,15 @@ static int amdgpu_ras_debugfs_ctrl_parse_data(struct file *f, data->inject.address = address; return 0; + } else if (op == 4) { + if (sscanf(str, "%*s 0x%llx 0x%llx", &address, &value) != 2 && + sscanf(str, "%*s %llu %llu", &address, &value) != 2) + return -EINVAL; + + data->op = op; + data->inject.address = address; + data->inject.value = value; + return 0; } if (amdgpu_ras_find_block_id_by_name(block_name, &block_id)) @@ -496,6 +569,9 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, return size; else return ret; + } else if (data.op == 4) { + ret = amdgpu_check_address_validity(adev, data.inject.address, data.inject.value); + return ret ? ret : size; } if (!amdgpu_ras_is_supported(adev, data.head.block)) @@ -509,22 +585,16 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, ret = amdgpu_ras_feature_enable(adev, &data.head, 1); break; case 2: - if ((data.inject.address >= adev->gmc.mc_vram_size && - adev->gmc.mc_vram_size) || - (data.inject.address >= RAS_UMC_INJECT_ADDR_LIMIT)) { - dev_warn(adev->dev, "RAS WARN: input address " - "0x%llx is invalid.", + /* umc ce/ue error injection for a bad page is not allowed */ + if (data.head.block == AMDGPU_RAS_BLOCK__UMC) + ret = amdgpu_ras_check_bad_page(adev, data.inject.address); + if (ret == -EINVAL) { + dev_warn(adev->dev, "RAS WARN: input address 0x%llx is invalid.", data.inject.address); - ret = -EINVAL; break; - } - - /* umc ce/ue error injection for a bad page is not allowed */ - if ((data.head.block == AMDGPU_RAS_BLOCK__UMC) && - amdgpu_ras_check_bad_page(adev, data.inject.address)) { - dev_warn(adev->dev, "RAS WARN: inject: 0x%llx has " - "already been marked as bad!\n", - data.inject.address); + } else if (ret == 1) { + dev_warn(adev->dev, "RAS WARN: inject: 0x%llx has already been marked as bad!\n", + data.inject.address); break; } @@ -544,6 +614,8 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, return size; } +static int amdgpu_uniras_clear_badpages_info(struct amdgpu_device *adev); + /** * DOC: AMDGPU RAS debugfs EEPROM table reset interface * @@ -568,6 +640,11 @@ static ssize_t amdgpu_ras_debugfs_eeprom_write(struct file *f, (struct amdgpu_device *)file_inode(f)->i_private; int ret; + if (amdgpu_uniras_enabled(adev)) { + ret = amdgpu_uniras_clear_badpages_info(adev); + return ret ? ret : size; + } + ret = amdgpu_ras_eeprom_reset_table( &(amdgpu_ras_get_context(adev)->eeprom_control)); @@ -880,7 +957,7 @@ 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); @@ -1055,7 +1132,7 @@ 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->event_id; + u64 event_id = qctx->evid.event_id; if (is_ue) { for_each_ras_error(err_node, err_data) { @@ -1103,6 +1180,9 @@ static void amdgpu_ras_error_print_error_data(struct amdgpu_device *adev, err_info->de_count, blk_name); } } else { + if (adev->debug_disable_ce_logs) + return; + for_each_ras_error(err_node, err_data) { err_info = &err_node->err_info; mcm_info = &err_info->mcm_info; @@ -1140,7 +1220,7 @@ static void amdgpu_ras_error_generate_report(struct amdgpu_device *adev, { 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; + u64 event_id = qctx->evid.event_id; if (err_data->ce_count) { if (err_data_has_source_info(err_data)) { @@ -1212,6 +1292,42 @@ static void amdgpu_ras_error_generate_report(struct amdgpu_device *adev, } } +static void amdgpu_ras_virt_error_generate_report(struct amdgpu_device *adev, + struct ras_query_if *query_if, + struct ras_err_data *err_data, + struct ras_query_context *qctx) +{ + unsigned long new_ue, new_ce, new_de; + struct ras_manager *obj = 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; + + new_ce = err_data->ce_count - obj->err_data.ce_count; + new_ue = err_data->ue_count - obj->err_data.ue_count; + new_de = err_data->de_count - obj->err_data.de_count; + + if (new_ce) { + RAS_EVENT_LOG(adev, event_id, "%lu correctable hardware errors " + "detected in %s block\n", + new_ce, + blk_name); + } + + if (new_ue) { + RAS_EVENT_LOG(adev, event_id, "%lu uncorrectable hardware errors " + "detected in %s block\n", + new_ue, + blk_name); + } + + if (new_de) { + RAS_EVENT_LOG(adev, event_id, "%lu deferred hardware errors " + "detected in %s block\n", + new_de, + 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; @@ -1221,11 +1337,11 @@ 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 */ @@ -1235,6 +1351,15 @@ static void amdgpu_rasmgr_error_data_statistic_update(struct ras_manager *obj, s } } +static void amdgpu_ras_mgr_virt_error_data_statistics_update(struct ras_manager *obj, + struct ras_err_data *err_data) +{ + /* Host reports absolute counts */ + 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; +} + static struct ras_manager *get_ras_manager(struct amdgpu_device *adev, enum amdgpu_ras_block blk) { struct ras_common_if head; @@ -1251,7 +1376,7 @@ 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)) + if (adev->in_suspend || amdgpu_reset_in_recovery(adev)) return 0; obj = get_ras_manager(adev, blk); @@ -1295,11 +1420,14 @@ 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%s: %lu\n", "ue", info.ue_count, - "ce", info.ce_count, "de", info.ue_count); + "ce", info.ce_count, "de", info.de_count); } static int amdgpu_ras_query_error_status_helper(struct amdgpu_device *adev, @@ -1318,7 +1446,9 @@ static int amdgpu_ras_query_error_status_helper(struct amdgpu_device *adev, if (error_query_mode == AMDGPU_RAS_INVALID_ERROR_QUERY) return -EINVAL; - if (error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY) { + if (error_query_mode == AMDGPU_RAS_VIRT_ERROR_COUNT_QUERY) { + return amdgpu_virt_req_ras_err_count(adev, blk, err_data); + } else 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 { @@ -1363,7 +1493,9 @@ 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; @@ -1382,38 +1514,97 @@ int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_i return -EINVAL; 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); + 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; - amdgpu_rasmgr_error_data_statistic_update(obj, &err_data); + if (error_query_mode != AMDGPU_RAS_VIRT_ERROR_COUNT_QUERY) { + amdgpu_rasmgr_error_data_statistic_update(obj, &err_data); + amdgpu_ras_error_generate_report(adev, info, &err_data, &qctx); + } else { + /* Host provides absolute error counts. First generate the report + * using the previous VF internal count against new host count. + * Then Update VF internal count. + */ + amdgpu_ras_virt_error_generate_report(adev, info, &err_data, &qctx); + amdgpu_ras_mgr_virt_error_data_statistics_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, &qctx); - out_fini_err_data: amdgpu_ras_error_data_fini(&err_data); return ret; } +static int amdgpu_uniras_clear_badpages_info(struct amdgpu_device *adev) +{ + struct ras_cmd_dev_handle req = {0}; + int ret; + + ret = amdgpu_ras_mgr_handle_ras_cmd(adev, RAS_CMD__CLEAR_BAD_PAGE_INFO, + &req, sizeof(req), NULL, 0); + if (ret) { + dev_err(adev->dev, "Failed to clear bad pages info, ret: %d\n", ret); + return ret; + } + + return 0; +} + +static int amdgpu_uniras_query_block_ecc(struct amdgpu_device *adev, + struct ras_query_if *info) +{ + struct ras_cmd_block_ecc_info_req req = {0}; + struct ras_cmd_block_ecc_info_rsp rsp = {0}; + int ret; + + if (!info) + return -EINVAL; + + req.block_id = info->head.block; + req.subblock_id = info->head.sub_block_index; + + ret = amdgpu_ras_mgr_handle_ras_cmd(adev, RAS_CMD__GET_BLOCK_ECC_STATUS, + &req, sizeof(req), &rsp, sizeof(rsp)); + if (!ret) { + info->ce_count = rsp.ce_count; + info->ue_count = rsp.ue_count; + info->de_count = rsp.de_count; + } + + return ret; +} + +int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_if *info) +{ + if (amdgpu_uniras_enabled(adev)) + return amdgpu_uniras_query_block_ecc(adev, info); + else + 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", @@ -1425,15 +1616,11 @@ 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); - } + if (amdgpu_sriov_vf(adev)) + return -EOPNOTSUPP; /* 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; @@ -1461,6 +1648,27 @@ int amdgpu_ras_reset_error_status(struct amdgpu_device *adev, return 0; } +static int amdgpu_uniras_error_inject(struct amdgpu_device *adev, + struct ras_inject_if *info) +{ + struct ras_cmd_inject_error_req inject_req; + struct ras_cmd_inject_error_rsp rsp; + + if (!info) + return -EINVAL; + + memset(&inject_req, 0, sizeof(inject_req)); + inject_req.block_id = info->head.block; + inject_req.subblock_id = info->head.sub_block_index; + inject_req.address = info->address; + inject_req.error_type = info->head.type; + inject_req.instance_mask = info->instance_mask; + inject_req.method = info->value; + + return amdgpu_ras_mgr_handle_ras_cmd(adev, RAS_CMD__INJECT_ERROR, + &inject_req, sizeof(inject_req), &rsp, sizeof(rsp)); +} + /* wrapper of psp_ras_trigger_error */ int amdgpu_ras_error_inject(struct amdgpu_device *adev, struct ras_inject_if *info) @@ -1478,6 +1686,9 @@ int amdgpu_ras_error_inject(struct amdgpu_device *adev, info->head.block, info->head.sub_block_index); + if (amdgpu_uniras_enabled(adev)) + return amdgpu_uniras_error_inject(adev, info); + /* inject on guest isn't allowed, return success directly */ if (amdgpu_sriov_vf(adev)) return 0; @@ -1622,7 +1833,9 @@ int amdgpu_ras_query_error_count(struct amdgpu_device *adev, /* sysfs begin */ static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, - struct ras_badpage **bps, unsigned int *count); + struct ras_badpage *bps, uint32_t count, uint32_t start); +static int amdgpu_uniras_badpages_read(struct amdgpu_device *adev, + struct ras_badpage *bps, uint32_t count, uint32_t start); static char *amdgpu_ras_badpage_flags_str(unsigned int flags) { @@ -1668,7 +1881,7 @@ static char *amdgpu_ras_badpage_flags_str(unsigned int flags) */ static ssize_t amdgpu_ras_sysfs_badpages_read(struct file *f, - struct kobject *kobj, struct bin_attribute *attr, + struct kobject *kobj, const struct bin_attribute *attr, char *buf, loff_t ppos, size_t count) { struct amdgpu_ras *con = @@ -1680,19 +1893,50 @@ static ssize_t amdgpu_ras_sysfs_badpages_read(struct file *f, unsigned int end = div64_ul(ppos + count - 1, element_size); ssize_t s = 0; struct ras_badpage *bps = NULL; - unsigned int bps_count = 0; + int bps_count = 0, i, status; + uint64_t address; memset(buf, 0, count); - if (amdgpu_ras_badpages_read(adev, &bps, &bps_count)) + bps_count = end - start; + bps = kmalloc_array(bps_count, sizeof(*bps), GFP_KERNEL); + if (!bps) + return 0; + + memset(bps, 0, sizeof(*bps) * bps_count); + + if (amdgpu_uniras_enabled(adev)) + bps_count = amdgpu_uniras_badpages_read(adev, bps, bps_count, start); + else + bps_count = amdgpu_ras_badpages_read(adev, bps, bps_count, start); + + if (bps_count <= 0) { + kfree(bps); return 0; + } + + for (i = 0; i < bps_count; i++) { + address = ((uint64_t)bps[i].bp) << AMDGPU_GPU_PAGE_SHIFT; + if (amdgpu_ras_check_critical_address(adev, address)) + continue; + + bps[i].size = AMDGPU_GPU_PAGE_SIZE; + + status = amdgpu_vram_mgr_query_page_status(&adev->mman.vram_mgr, + address); + if (status == -EBUSY) + bps[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING; + else if (status == -ENOENT) + bps[i].flags = AMDGPU_RAS_RETIRE_PAGE_FAULT; + else + bps[i].flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED; - for (; start < end && start < bps_count; start++) s += scnprintf(&buf[s], element_size + 1, "0x%08x : 0x%08x : %1s\n", - bps[start].bp, - bps[start].size, - amdgpu_ras_badpage_flags_str(bps[start].flags)); + bps[i].bp, + bps[i].size, + amdgpu_ras_badpage_flags_str(bps[i].flags)); + } kfree(bps); @@ -1708,12 +1952,42 @@ static ssize_t amdgpu_ras_sysfs_features_read(struct device *dev, return sysfs_emit(buf, "feature mask: 0x%x\n", con->features); } +static bool amdgpu_ras_get_version_info(struct amdgpu_device *adev, u32 *major, + u32 *minor, u32 *rev) +{ + int i; + + if (!adev || !major || !minor || !rev || !amdgpu_uniras_enabled(adev)) + return false; + + for (i = 0; i < adev->num_ip_blocks; i++) { + if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_RAS) { + *major = adev->ip_blocks[i].version->major; + *minor = adev->ip_blocks[i].version->minor; + *rev = adev->ip_blocks[i].version->rev; + return true; + } + } + + return false; +} + static ssize_t amdgpu_ras_sysfs_version_show(struct device *dev, struct device_attribute *attr, char *buf) { struct amdgpu_ras *con = container_of(attr, struct amdgpu_ras, version_attr); - return sysfs_emit(buf, "table version: 0x%x\n", con->eeprom_control.tbl_hdr.version); + u32 major, minor, rev; + ssize_t size = 0; + + size += sysfs_emit_at(buf, size, "table version: 0x%x\n", + con->eeprom_control.tbl_hdr.version); + + if (amdgpu_ras_get_version_info(con->adev, &major, &minor, &rev)) + size += sysfs_emit_at(buf, size, "ras version: %u.%u.%u\n", + major, minor, rev); + + return size; } static ssize_t amdgpu_ras_sysfs_schema_show(struct device *dev, @@ -1724,6 +1998,39 @@ 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); @@ -1741,6 +2048,7 @@ 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 = { @@ -1759,9 +2067,15 @@ 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; + if (amdgpu_sriov_vf(adev) && !amdgpu_virt_ras_telemetry_block_en(adev, head->block)) + return 0; + get_obj(obj); snprintf(obj->fs_data.sysfs_name, sizeof(obj->fs_data.sysfs_name), @@ -1793,6 +2107,9 @@ 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; @@ -1905,6 +2222,24 @@ 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, 12): + 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); @@ -1931,23 +2266,27 @@ 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 */ /* ras fs */ -static BIN_ATTR(gpu_vram_bad_pages, S_IRUGO, - amdgpu_ras_sysfs_badpages_read, NULL, 0); +static const BIN_ATTR(gpu_vram_bad_pages, S_IRUGO, + amdgpu_ras_sysfs_badpages_read, NULL, 0); static DEVICE_ATTR(features, S_IRUGO, amdgpu_ras_sysfs_features_read, NULL); 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); @@ -1958,9 +2297,10 @@ 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[] = { + const struct bin_attribute *bin_attrs[] = { NULL, NULL, }; @@ -1980,13 +2320,16 @@ 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; con->badpages_attr = bin_attr_gpu_vram_bad_pages; + sysfs_bin_attr_init(&con->badpages_attr); bin_attrs[0] = &con->badpages_attr; group.bin_attrs = bin_attrs; - sysfs_bin_attr_init(bin_attrs[0]); } r = sysfs_create_group(&adev->dev->kobj, &group); @@ -2026,6 +2369,21 @@ void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev) /* Fatal error events are handled on host side */ if (amdgpu_sriov_vf(adev)) return; + /* + * If the current interrupt is caused by a non-fatal RAS error, skip + * check for fatal error. For fatal errors, FED status of all devices + * in XGMI hive gets set when the first device gets fatal error + * interrupt. The error gets propagated to other devices as well, so + * make sure to ack the interrupt regardless of FED status. + */ + if (!amdgpu_ras_get_fed_status(adev) && + amdgpu_ras_is_err_state(adev, AMDGPU_RAS_BLOCK__ANY)) + return; + + if (amdgpu_uniras_enabled(adev)) { + amdgpu_ras_mgr_handle_fatal_interrupt(adev, NULL); + return; + } if (adev->nbio.ras && adev->nbio.ras->handle_ras_controller_intr_no_bifring) @@ -2043,10 +2401,19 @@ 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; + amdgpu_ras_set_err_poison(adev, block_obj->ras_comm.block); /* both query_poison_status and handle_poison_consumption are optional, * but at least one of them should be implemented if we need poison * consumption handler @@ -2067,29 +2434,41 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager * if (block_obj->hw_ops && block_obj->hw_ops->handle_poison_consumption) poison_stat = block_obj->hw_ops->handle_poison_consumption(adev); - /* gpu reset is fallback for failed and default cases */ - if (poison_stat) { - 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); - amdgpu_ras_put_poison_req(obj->adev, - AMDGPU_RAS_BLOCK__UMC, 0, NULL, NULL, false); - atomic_inc(&con->page_retirement_req_cnt); + atomic_inc(&con->poison_creation_count); wake_up(&con->page_retirement_wq); } @@ -2112,6 +2491,7 @@ 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. @@ -2172,12 +2552,25 @@ 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; + + if (amdgpu_uniras_enabled(adev)) { + struct ras_ih_info ih_info; + memset(&ih_info, 0, sizeof(ih_info)); + ih_info.block = info->head.block; + memcpy(&ih_info.iv_entry, info->entry, sizeof(struct amdgpu_iv_entry)); + + return amdgpu_ras_mgr_handle_controller_interrupt(adev, &ih_info); + } + + obj = amdgpu_ras_find_obj(adev, &info->head); if (!obj) return -EINVAL; + data = &obj->ih_data; + if (data->inuse == 0) return 0; @@ -2274,7 +2667,7 @@ 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; @@ -2307,7 +2700,7 @@ 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) && @@ -2366,54 +2759,83 @@ static void amdgpu_ras_query_err_status(struct amdgpu_device *adev) } } -/* recovery begin */ - -/* return 0 on success. - * caller need free bps. - */ static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, - struct ras_badpage **bps, unsigned int *count) + struct ras_badpage *bps, uint32_t count, uint32_t start) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); struct ras_err_handler_data *data; - int i = 0; - int ret = 0, status; + int r = 0; + uint32_t i; if (!con || !con->eh_data || !bps || !count) return -EINVAL; mutex_lock(&con->recovery_lock); data = con->eh_data; - if (!data || data->count == 0) { - *bps = NULL; - ret = -EINVAL; - goto out; + if (start < data->count) { + for (i = start; i < data->count; i++) { + if (!data->bps[i].ts) + continue; + + bps[r].bp = data->bps[i].retired_page; + r++; + if (r >= count) + break; + } } + mutex_unlock(&con->recovery_lock); - *bps = kmalloc(sizeof(struct ras_badpage) * data->count, GFP_KERNEL); - if (!*bps) { - ret = -ENOMEM; - goto out; - } + return r; +} - for (; i < data->count; i++) { - (*bps)[i] = (struct ras_badpage){ - .bp = data->bps[i].retired_page, - .size = AMDGPU_GPU_PAGE_SIZE, - .flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED, - }; - status = amdgpu_vram_mgr_query_page_status(&adev->mman.vram_mgr, - data->bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT); - if (status == -EBUSY) - (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING; - else if (status == -ENOENT) - (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_FAULT; +static int amdgpu_uniras_badpages_read(struct amdgpu_device *adev, + struct ras_badpage *bps, uint32_t count, uint32_t start) +{ + struct ras_cmd_bad_pages_info_req cmd_input; + struct ras_cmd_bad_pages_info_rsp *output; + uint32_t group, start_group, end_group; + uint32_t pos, pos_in_group; + int r = 0, i; + + if (!bps || !count) + return -EINVAL; + + output = kmalloc(sizeof(*output), GFP_KERNEL); + if (!output) + return -ENOMEM; + + memset(&cmd_input, 0, sizeof(cmd_input)); + + start_group = start / RAS_CMD_MAX_BAD_PAGES_PER_GROUP; + end_group = (start + count + RAS_CMD_MAX_BAD_PAGES_PER_GROUP - 1) / + RAS_CMD_MAX_BAD_PAGES_PER_GROUP; + + pos = start; + for (group = start_group; group < end_group; group++) { + memset(output, 0, sizeof(*output)); + cmd_input.group_index = group; + if (amdgpu_ras_mgr_handle_ras_cmd(adev, RAS_CMD__GET_BAD_PAGES, + &cmd_input, sizeof(cmd_input), output, sizeof(*output))) + goto out; + + if (pos >= output->bp_total_cnt) + goto out; + + pos_in_group = pos - group * RAS_CMD_MAX_BAD_PAGES_PER_GROUP; + for (i = pos_in_group; i < output->bp_in_group; i++, pos++) { + if (!output->records[i].ts) + continue; + + bps[r].bp = output->records[i].retired_page; + r++; + if (r >= count) + goto out; + } } - *count = data->count; out: - mutex_unlock(&con->recovery_lock); - return ret; + kfree(output); + return r; } static void amdgpu_ras_set_fed_all(struct amdgpu_device *adev, @@ -2429,6 +2851,31 @@ static void amdgpu_ras_set_fed_all(struct amdgpu_device *adev, } } +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 = @@ -2437,6 +2884,8 @@ 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); + unsigned int error_query_mode; + enum ras_event_type type; if (hive) { atomic_set(&hive->ras_recovery, 1); @@ -2464,10 +2913,22 @@ static void amdgpu_ras_do_recovery(struct work_struct *work) device_list_handle = &device_list; } + if (amdgpu_ras_get_error_query_mode(adev, &error_query_mode)) { + if (error_query_mode == AMDGPU_RAS_FIRMWARE_ERROR_QUERY) { + /* wait 500ms to ensure pmfw polling mca bank info done */ + msleep(500); + } + } + + 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); + if (amdgpu_uniras_enabled(remote_adev)) { + amdgpu_ras_mgr_update_ras_ecc(remote_adev); + } else { + amdgpu_ras_query_err_status(remote_adev); + amdgpu_ras_log_on_err_counter(remote_adev, type); + } } } @@ -2478,6 +2939,8 @@ 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; + set_bit(AMDGPU_SKIP_COREDUMP, &reset_context.flags); /* Perform full reset in fatal error mode */ if (!amdgpu_ras_is_poison_mode_supported(ras->adev)) @@ -2517,7 +2980,7 @@ static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev, unsigned int old_space = data->count + data->space_left; unsigned int new_space = old_space + pages; unsigned int align_space = ALIGN(new_space, 512); - void *bps = kmalloc(align_space * sizeof(*data->bps), GFP_KERNEL); + void *bps = kmalloc_array(align_space, sizeof(*data->bps), GFP_KERNEL); if (!bps) { return -ENOMEM; @@ -2534,41 +2997,283 @@ static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev, return 0; } -/* it deal with vram only. */ -int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, - struct eeprom_table_record *bps, int pages) +static int amdgpu_ras_mca2pa_by_idx(struct amdgpu_device *adev, + struct eeprom_table_record *bps, + struct ras_err_data *err_data) { - struct amdgpu_ras *con = amdgpu_ras_get_context(adev); - struct ras_err_handler_data *data; + struct ta_ras_query_address_input addr_in; + uint32_t socket = 0; int ret = 0; - uint32_t i; - if (!con || !con->eh_data || !bps || pages <= 0) - return 0; + if (adev->smuio.funcs && adev->smuio.funcs->get_socket_id) + socket = adev->smuio.funcs->get_socket_id(adev); - mutex_lock(&con->recovery_lock); - data = con->eh_data; - if (!data) - goto out; + /* reinit err_data */ + err_data->err_addr_cnt = 0; + err_data->err_addr_len = adev->umc.retire_unit; + + memset(&addr_in, 0, sizeof(addr_in)); + addr_in.ma.err_addr = bps->address; + addr_in.ma.socket_id = socket; + addr_in.ma.ch_inst = bps->mem_channel; + if (!amdgpu_ras_smu_eeprom_supported(adev)) { + /* tell RAS TA the node instance is not used */ + addr_in.ma.node_inst = TA_RAS_INV_NODE; + } else { + addr_in.ma.umc_inst = bps->mcumc_id; + addr_in.ma.node_inst = bps->cu; + } + + if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr) + ret = adev->umc.ras->convert_ras_err_addr(adev, err_data, + &addr_in, NULL, false); + + return ret; +} + +static int amdgpu_ras_mca2pa(struct amdgpu_device *adev, + struct eeprom_table_record *bps, + struct ras_err_data *err_data) +{ + struct ta_ras_query_address_input addr_in; + uint32_t die_id, socket = 0; - for (i = 0; i < pages; i++) { + if (adev->smuio.funcs && adev->smuio.funcs->get_socket_id) + socket = adev->smuio.funcs->get_socket_id(adev); + + /* although die id is gotten from PA in nps1 mode, the id is + * fitable for any nps mode + */ + if (adev->umc.ras && adev->umc.ras->get_die_id_from_pa) + die_id = adev->umc.ras->get_die_id_from_pa(adev, bps->address, + bps->retired_page << AMDGPU_GPU_PAGE_SHIFT); + else + return -EINVAL; + + /* reinit err_data */ + err_data->err_addr_cnt = 0; + err_data->err_addr_len = adev->umc.retire_unit; + + memset(&addr_in, 0, sizeof(addr_in)); + addr_in.ma.err_addr = bps->address; + addr_in.ma.ch_inst = bps->mem_channel; + addr_in.ma.umc_inst = bps->mcumc_id; + addr_in.ma.node_inst = die_id; + addr_in.ma.socket_id = socket; + + if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr) + return adev->umc.ras->convert_ras_err_addr(adev, err_data, + &addr_in, NULL, false); + else + return -EINVAL; +} + +static int __amdgpu_ras_restore_bad_pages(struct amdgpu_device *adev, + struct eeprom_table_record *bps, int count) +{ + int j; + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_err_handler_data *data = con->eh_data; + + for (j = 0; j < count; j++) { if (amdgpu_ras_check_bad_page_unlock(con, - bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT)) + bps[j].retired_page << AMDGPU_GPU_PAGE_SHIFT)) { + data->count++; + data->space_left--; continue; + } if (!data->space_left && - amdgpu_ras_realloc_eh_data_space(adev, data, 256)) { - ret = -ENOMEM; - goto out; + amdgpu_ras_realloc_eh_data_space(adev, data, 256)) { + return -ENOMEM; } - amdgpu_ras_reserve_page(adev, bps[i].retired_page); + amdgpu_ras_reserve_page(adev, bps[j].retired_page); - memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps)); + memcpy(&data->bps[data->count], &(bps[j]), + sizeof(struct eeprom_table_record)); data->count++; data->space_left--; + con->bad_page_num++; + } + + return 0; +} + +static int __amdgpu_ras_convert_rec_array_from_rom(struct amdgpu_device *adev, + struct eeprom_table_record *bps, struct ras_err_data *err_data, + enum amdgpu_memory_partition nps) +{ + int i = 0; + enum amdgpu_memory_partition save_nps; + + save_nps = (bps[0].retired_page >> UMC_NPS_SHIFT) & UMC_NPS_MASK; + + /*old asics just have pa in eeprom*/ + if (IP_VERSION_MAJ(amdgpu_ip_version(adev, UMC_HWIP, 0)) < 12) { + memcpy(err_data->err_addr, bps, + sizeof(struct eeprom_table_record) * adev->umc.retire_unit); + goto out; + } + + for (i = 0; i < adev->umc.retire_unit; i++) + bps[i].retired_page &= ~(UMC_NPS_MASK << UMC_NPS_SHIFT); + + if (save_nps) { + if (save_nps == nps) { + if (amdgpu_umc_pages_in_a_row(adev, err_data, + bps[0].retired_page << AMDGPU_GPU_PAGE_SHIFT)) + return -EINVAL; + for (i = 0; i < adev->umc.retire_unit; i++) { + err_data->err_addr[i].address = bps[0].address; + err_data->err_addr[i].mem_channel = bps[0].mem_channel; + err_data->err_addr[i].bank = bps[0].bank; + err_data->err_addr[i].err_type = bps[0].err_type; + err_data->err_addr[i].mcumc_id = bps[0].mcumc_id; + } + } else { + if (amdgpu_ras_mca2pa_by_idx(adev, &bps[0], err_data)) + return -EINVAL; + } + } else { + if (bps[0].address == 0) { + /* for specific old eeprom data, mca address is not stored, + * calc it from pa + */ + if (amdgpu_umc_pa2mca(adev, bps[0].retired_page << AMDGPU_GPU_PAGE_SHIFT, + &(bps[0].address), AMDGPU_NPS1_PARTITION_MODE)) + return -EINVAL; + } + + if (amdgpu_ras_mca2pa(adev, &bps[0], err_data)) { + if (nps == AMDGPU_NPS1_PARTITION_MODE) + memcpy(err_data->err_addr, bps, + sizeof(struct eeprom_table_record) * adev->umc.retire_unit); + else + return -EOPNOTSUPP; + } } + out: + return __amdgpu_ras_restore_bad_pages(adev, err_data->err_addr, adev->umc.retire_unit); +} + +static int __amdgpu_ras_convert_rec_from_rom(struct amdgpu_device *adev, + struct eeprom_table_record *bps, struct ras_err_data *err_data, + enum amdgpu_memory_partition nps) +{ + int i = 0; + enum amdgpu_memory_partition save_nps; + + if (!amdgpu_ras_smu_eeprom_supported(adev)) { + save_nps = (bps->retired_page >> UMC_NPS_SHIFT) & UMC_NPS_MASK; + bps->retired_page &= ~(UMC_NPS_MASK << UMC_NPS_SHIFT); + } else { + /* if pmfw manages eeprom, save_nps is not stored on eeprom, + * we should always convert mca address into physical address, + * make save_nps different from nps + */ + save_nps = nps + 1; + } + + if (save_nps == nps) { + if (amdgpu_umc_pages_in_a_row(adev, err_data, + bps->retired_page << AMDGPU_GPU_PAGE_SHIFT)) + return -EINVAL; + for (i = 0; i < adev->umc.retire_unit; i++) { + err_data->err_addr[i].address = bps->address; + err_data->err_addr[i].mem_channel = bps->mem_channel; + err_data->err_addr[i].bank = bps->bank; + err_data->err_addr[i].err_type = bps->err_type; + err_data->err_addr[i].mcumc_id = bps->mcumc_id; + } + } else { + if (bps->address) { + if (amdgpu_ras_mca2pa_by_idx(adev, bps, err_data)) + return -EINVAL; + } else { + /* for specific old eeprom data, mca address is not stored, + * calc it from pa + */ + if (amdgpu_umc_pa2mca(adev, bps->retired_page << AMDGPU_GPU_PAGE_SHIFT, + &(bps->address), AMDGPU_NPS1_PARTITION_MODE)) + return -EINVAL; + + if (amdgpu_ras_mca2pa(adev, bps, err_data)) + return -EOPNOTSUPP; + } + } + + return __amdgpu_ras_restore_bad_pages(adev, err_data->err_addr, + adev->umc.retire_unit); +} + +/* it deal with vram only. */ +int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, + struct eeprom_table_record *bps, int pages, bool from_rom) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_err_data err_data; + struct amdgpu_ras_eeprom_control *control = + &adev->psp.ras_context.ras->eeprom_control; + enum amdgpu_memory_partition nps = AMDGPU_NPS1_PARTITION_MODE; + int ret = 0; + uint32_t i = 0; + + if (!con || !con->eh_data || !bps || pages <= 0) + return 0; + + if (from_rom) { + err_data.err_addr = + kcalloc(adev->umc.retire_unit, + sizeof(struct eeprom_table_record), GFP_KERNEL); + if (!err_data.err_addr) { + dev_warn(adev->dev, "Failed to alloc UMC error address record in mca2pa conversion!\n"); + return -ENOMEM; + } + + if (adev->gmc.gmc_funcs->query_mem_partition_mode) + nps = adev->gmc.gmc_funcs->query_mem_partition_mode(adev); + } + + mutex_lock(&con->recovery_lock); + + if (from_rom) { + /* there is no pa recs in V3, so skip pa recs processing */ + if ((control->tbl_hdr.version < RAS_TABLE_VER_V3) && + !amdgpu_ras_smu_eeprom_supported(adev)) { + for (i = 0; i < pages; i++) { + if (control->ras_num_recs - i >= adev->umc.retire_unit) { + if ((bps[i].address == bps[i + 1].address) && + (bps[i].mem_channel == bps[i + 1].mem_channel)) { + /* deal with retire_unit records a time */ + ret = __amdgpu_ras_convert_rec_array_from_rom(adev, + &bps[i], &err_data, nps); + if (ret) + con->bad_page_num -= adev->umc.retire_unit; + i += (adev->umc.retire_unit - 1); + } else { + break; + } + } else { + break; + } + } + } + for (; i < pages; i++) { + ret = __amdgpu_ras_convert_rec_from_rom(adev, + &bps[i], &err_data, nps); + if (ret) + con->bad_page_num -= adev->umc.retire_unit; + } + + con->eh_data->count_saved = con->eh_data->count; + } else { + ret = __amdgpu_ras_restore_bad_pages(adev, bps, pages); + } + + if (from_rom) + kfree(err_data.err_addr); mutex_unlock(&con->recovery_lock); return ret; @@ -2585,7 +3290,7 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev, struct amdgpu_ras *con = amdgpu_ras_get_context(adev); struct ras_err_handler_data *data; struct amdgpu_ras_eeprom_control *control; - int save_count; + int save_count, unit_num, i; if (!con || !con->eh_data) { if (new_cnt) @@ -2594,25 +3299,53 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev, return 0; } + if (!con->eeprom_control.is_eeprom_valid) { + dev_warn(adev->dev, + "Failed to save EEPROM table data because of EEPROM data corruption!"); + if (new_cnt) + *new_cnt = 0; + + return 0; + } + mutex_lock(&con->recovery_lock); control = &con->eeprom_control; data = con->eh_data; - save_count = data->count - control->ras_num_recs; + if (amdgpu_ras_smu_eeprom_supported(adev)) + unit_num = control->ras_num_recs - + control->ras_num_recs_old; + else + unit_num = data->count / adev->umc.retire_unit - + control->ras_num_recs; + + save_count = con->bad_page_num - control->ras_num_bad_pages; mutex_unlock(&con->recovery_lock); if (new_cnt) - *new_cnt = save_count / adev->umc.retire_unit; + *new_cnt = unit_num; /* only new entries are saved */ - if (save_count > 0) { - if (amdgpu_ras_eeprom_append(control, - &data->bps[control->ras_num_recs], - save_count)) { - dev_err(adev->dev, "Failed to save EEPROM table data!"); - return -EIO; + if (unit_num && save_count) { + /*old asics only save pa to eeprom like before*/ + if (IP_VERSION_MAJ(amdgpu_ip_version(adev, UMC_HWIP, 0)) < 12) { + if (amdgpu_ras_eeprom_append(control, + &data->bps[data->count_saved], unit_num)) { + dev_err(adev->dev, "Failed to save EEPROM table data!"); + return -EIO; + } + } else { + for (i = 0; i < unit_num; i++) { + if (amdgpu_ras_eeprom_append(control, + &data->bps[data->count_saved + + i * adev->umc.retire_unit], 1)) { + dev_err(adev->dev, "Failed to save EEPROM table data!"); + return -EIO; + } + } } dev_info(adev->dev, "Saved %d pages to EEPROM table.\n", save_count); + data->count_saved = data->count; } return 0; @@ -2627,7 +3360,7 @@ static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev) struct amdgpu_ras_eeprom_control *control = &adev->psp.ras_context.ras->eeprom_control; struct eeprom_table_record *bps; - int ret; + int ret, i = 0; /* no bad page record, skip eeprom access */ if (control->ras_num_recs == 0 || amdgpu_bad_page_threshold == 0) @@ -2638,27 +3371,72 @@ static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev) return -ENOMEM; ret = amdgpu_ras_eeprom_read(control, bps, control->ras_num_recs); - if (ret) + if (ret) { dev_err(adev->dev, "Failed to load EEPROM table records!"); - else - ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs); + } else { + if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr) { + /*In V3, there is no pa recs, and some cases(when address==0) may be parsed + as pa recs, so add verion check to avoid it. + */ + if ((control->tbl_hdr.version < RAS_TABLE_VER_V3) && + !amdgpu_ras_smu_eeprom_supported(adev)) { + for (i = 0; i < control->ras_num_recs; i++) { + if ((control->ras_num_recs - i) >= adev->umc.retire_unit) { + if ((bps[i].address == bps[i + 1].address) && + (bps[i].mem_channel == bps[i + 1].mem_channel)) { + control->ras_num_pa_recs += adev->umc.retire_unit; + i += (adev->umc.retire_unit - 1); + } else { + control->ras_num_mca_recs += + (control->ras_num_recs - i); + break; + } + } else { + control->ras_num_mca_recs += (control->ras_num_recs - i); + break; + } + } + } else { + control->ras_num_mca_recs = control->ras_num_recs; + } + } + + ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs, true); + if (ret) + goto out; + + ret = amdgpu_ras_eeprom_check(control); + if (ret) + goto out; + /* HW not usable */ + if (amdgpu_ras_is_rma(adev)) + ret = -EHWPOISON; + } + +out: kfree(bps); return ret; } -static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, +static int amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, uint64_t addr) { struct ras_err_handler_data *data = con->eh_data; + struct amdgpu_device *adev = con->adev; int i; + if ((addr >= adev->gmc.mc_vram_size && + adev->gmc.mc_vram_size) || + (addr >= RAS_UMC_INJECT_ADDR_LIMIT)) + return -EINVAL; + addr >>= AMDGPU_GPU_PAGE_SHIFT; for (i = 0; i < data->count; i++) if (addr == data->bps[i].retired_page) - return true; + return 1; - return false; + return 0; } /* @@ -2666,11 +3444,11 @@ static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, * * Note: this check is only for umc block */ -static bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev, +static int amdgpu_ras_check_bad_page(struct amdgpu_device *adev, uint64_t addr) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); - bool ret = false; + int ret = 0; if (!con || !con->eh_data) return ret; @@ -2687,31 +3465,29 @@ static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev, struct amdgpu_ras *con = amdgpu_ras_get_context(adev); /* - * Justification of value bad_page_cnt_threshold in ras structure - * - * Generally, 0 <= amdgpu_bad_page_threshold <= max record length - * in eeprom or amdgpu_bad_page_threshold == -2, introduce two - * scenarios accordingly. - * - * Bad page retirement enablement: - * - If amdgpu_bad_page_threshold = -2, - * bad_page_cnt_threshold = typical value by formula. - * - * - When the value from user is 0 < amdgpu_bad_page_threshold < - * max record length in eeprom, use it directly. - * - * Bad page retirement disablement: - * - If amdgpu_bad_page_threshold = 0, bad page retirement - * functionality is disabled, and bad_page_cnt_threshold will - * take no effect. + * amdgpu_bad_page_threshold is used to config + * the threshold for the number of bad pages. + * -1: Threshold is set to default value + * Driver will issue a warning message when threshold is reached + * and continue runtime services. + * 0: Disable bad page retirement + * Driver will not retire bad pages + * which is intended for debugging purpose. + * -2: Threshold is determined by a formula + * that assumes 1 bad page per 100M of local memory. + * Driver will continue runtime services when threhold is reached. + * 0 < threshold < max number of bad page records in EEPROM, + * A user-defined threshold is set + * Driver will halt runtime services when this custom threshold is reached. */ - - if (amdgpu_bad_page_threshold < 0) { + if (amdgpu_bad_page_threshold == -2) { u64 val = adev->gmc.mc_vram_size; do_div(val, RAS_BAD_PAGE_COVER); con->bad_page_cnt_threshold = min(lower_32_bits(val), max_count); + } else if (amdgpu_bad_page_threshold == -1) { + con->bad_page_cnt_threshold = ((con->reserved_pages_in_bytes) >> 21) << 4; } else { con->bad_page_cnt_threshold = min_t(int, max_count, amdgpu_bad_page_threshold); @@ -2754,11 +3530,9 @@ static void amdgpu_ras_ecc_log_init(struct ras_ecc_log_info *ecc_log) { mutex_init(&ecc_log->lock); - /* Set any value as siphash key */ - memset(&ecc_log->ecc_key, 0xad, sizeof(ecc_log->ecc_key)); - INIT_RADIX_TREE(&ecc_log->de_page_tree, GFP_KERNEL); - ecc_log->de_updated = false; + ecc_log->de_queried_count = 0; + ecc_log->consumption_q_count = 0; } static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) @@ -2777,7 +3551,25 @@ static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) mutex_unlock(&ecc_log->lock); mutex_destroy(&ecc_log->lock); - ecc_log->de_updated = false; + ecc_log->de_queried_count = 0; + ecc_log->consumption_q_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) @@ -2787,8 +3579,12 @@ static void amdgpu_ras_do_page_retirement(struct work_struct *work) struct amdgpu_device *adev = con->adev; struct ras_err_data err_data; - if (amdgpu_in_reset(adev) || atomic_read(&con->in_recovery)) + /* 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); @@ -2796,77 +3592,106 @@ static void amdgpu_ras_do_page_retirement(struct work_struct *work) amdgpu_ras_error_data_fini(&err_data); - mutex_lock(&con->umc_ecc_log.lock); - if (radix_tree_tagged(&con->umc_ecc_log.de_page_tree, - UMC_ECC_NEW_DETECTED_TAG)) - schedule_delayed_work(&con->page_retirement_dwork, - msecs_to_jiffies(AMDGPU_RAS_RETIRE_PAGE_INTERVAL)); - mutex_unlock(&con->umc_ecc_log.lock); + amdgpu_ras_schedule_retirement_dwork(con, + AMDGPU_RAS_RETIRE_PAGE_INTERVAL); } -static int amdgpu_ras_query_ecc_status(struct amdgpu_device *adev, - enum amdgpu_ras_block ras_block, uint32_t timeout_ms) +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 = timeout_ms; + u32 timeout = MAX_UMC_POISON_POLLING_TIME_ASYNC; struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + u64 de_queried_count; + u64 consumption_q_count; + enum ras_event_type type = RAS_EVENT_TYPE_POISON_CREATION; memset(&info, 0, sizeof(info)); - info.head.block = ras_block; + info.head.block = AMDGPU_RAS_BLOCK__UMC; ecc_log = &ras->umc_ecc_log; - ecc_log->de_updated = false; + ecc_log->de_queried_count = 0; + ecc_log->consumption_q_count = 0; + do { - ret = amdgpu_ras_query_error_status(adev, &info); - if (ret) { - dev_err(adev->dev, "Failed to query ras error! ret:%d\n", ret); + ret = amdgpu_ras_query_error_status_with_event(adev, &info, type); + if (ret) return ret; - } - if (timeout && !ecc_log->de_updated) { - msleep(1); - timeout--; - } - } while (timeout && !ecc_log->de_updated); + de_queried_count = ecc_log->de_queried_count; + consumption_q_count = ecc_log->consumption_q_count; - if (timeout_ms && !timeout) { - dev_warn(adev->dev, "Can't find deferred error\n"); - return -ETIMEDOUT; - } + if (de_queried_count && consumption_q_count) + break; + + msleep(100); + } while (--timeout); + + if (de_queried_count) + schedule_delayed_work(&ras->page_retirement_dwork, 0); + + if (amdgpu_ras_is_rma(adev) && atomic_cmpxchg(&ras->rma_in_recovery, 0, 1) == 0) + amdgpu_ras_reset_gpu(adev); return 0; } -static void amdgpu_ras_poison_creation_handler(struct amdgpu_device *adev, - uint32_t timeout) +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; - ret = amdgpu_ras_query_ecc_status(adev, AMDGPU_RAS_BLOCK__UMC, timeout); - if (!ret) - schedule_delayed_work(&con->page_retirement_dwork, 0); + do { + ret = kfifo_get(&con->poison_fifo, &msg); + } while (ret); } static int amdgpu_ras_poison_consumption_handler(struct amdgpu_device *adev, - struct ras_poison_msg *poison_msg) + uint32_t msg_count, uint32_t *gpu_reset) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); - uint32_t reset = poison_msg->reset; - uint16_t pasid = poison_msg->pasid; + uint32_t reset_flags = 0, reset = 0; + struct ras_poison_msg msg; + int ret, i; kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); - if (poison_msg->pasid_fn) - poison_msg->pasid_fn(adev, pasid, poison_msg->data); + 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); - if (reset) { - flush_delayed_work(&con->page_retirement_dwork); + reset_flags |= msg.reset; + } + + /* + * Try to ensure poison creation handler is completed first + * to set rma if bad page exceed threshold. + */ + flush_delayed_work(&con->page_retirement_dwork); + + /* 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; 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; @@ -2876,9 +3701,9 @@ 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); - struct ras_poison_msg poison_msg; - enum amdgpu_ras_block ras_block; - bool poison_creation_is_handled = false; + uint32_t poison_creation_count, msg_count; + uint32_t gpu_reset; + int ret; while (!kthread_should_stop()) { @@ -2889,45 +3714,130 @@ static int amdgpu_ras_page_retirement_thread(void *param) if (kthread_should_stop()) break; - atomic_dec(&con->page_retirement_req_cnt); + mutex_lock(&con->poison_lock); + gpu_reset = 0; - if (!amdgpu_ras_get_poison_req(adev, &poison_msg)) - continue; + do { + poison_creation_count = atomic_read(&con->poison_creation_count); + ret = amdgpu_ras_poison_creation_handler(adev, poison_creation_count); + if (ret == -EIO) + break; - ras_block = poison_msg.block; + 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) && + !atomic_read(&con->poison_consumption_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); + } + } - dev_info(adev->dev, "Start processing ras block %s(%d)\n", - ras_block_str(ras_block), ras_block); + 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); + atomic_set(&con->poison_consumption_count, 0); - if (ras_block == AMDGPU_RAS_BLOCK__UMC) { - amdgpu_ras_poison_creation_handler(adev, - MAX_UMC_POISON_POLLING_TIME_ASYNC); - poison_creation_is_handled = true; - } else { - /* poison_creation_is_handled: - * false: no poison creation interrupt, but it has poison - * consumption interrupt. - * true: It has poison creation interrupt at the beginning, - * but it has no poison creation interrupt later. - */ - amdgpu_ras_poison_creation_handler(adev, - poison_creation_is_handled ? - 0 : MAX_UMC_POISON_POLLING_TIME_ASYNC); + /* Clear poison fifo */ + amdgpu_ras_clear_poison_fifo(adev); - amdgpu_ras_poison_consumption_handler(adev, &poison_msg); - poison_creation_is_handled = false; + /* Clear all poison requests */ + atomic_set(&con->page_retirement_req_cnt, 0); + + 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); + } + + atomic_set(&con->poison_consumption_count, 0); + + /* Wake up work to save bad pages to eeprom */ + schedule_delayed_work(&con->page_retirement_dwork, 0); } + mutex_unlock(&con->poison_lock); } return 0; } -int amdgpu_ras_recovery_init(struct amdgpu_device *adev) +int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct amdgpu_ras_eeprom_control *control; + int ret; + + if (!con || amdgpu_sriov_vf(adev)) + return 0; + + if (amdgpu_uniras_enabled(adev)) + return 0; + + control = &con->eeprom_control; + con->ras_smu_drv = amdgpu_dpm_get_ras_smu_driver(adev); + + ret = amdgpu_ras_eeprom_init(control); + control->is_eeprom_valid = !ret; + + if (!adev->umc.ras || !adev->umc.ras->convert_ras_err_addr) + control->ras_num_pa_recs = control->ras_num_recs; + + if (adev->umc.ras && + adev->umc.ras->get_retire_flip_bits) + adev->umc.ras->get_retire_flip_bits(adev); + + if (control->ras_num_recs && control->is_eeprom_valid) { + ret = amdgpu_ras_load_bad_pages(adev); + if (ret) { + control->is_eeprom_valid = false; + return 0; + } + + amdgpu_dpm_send_hbm_bad_pages_num( + adev, control->ras_num_bad_pages); + + if (con->update_channel_flag == true) { + amdgpu_dpm_send_hbm_bad_channel_flag( + adev, control->bad_channel_bitmap); + con->update_channel_flag = false; + } + + /* The format action is only applied to new ASICs */ + if (IP_VERSION_MAJ(amdgpu_ip_version(adev, UMC_HWIP, 0)) >= 12 && + control->tbl_hdr.version < RAS_TABLE_VER_V3) + if (!amdgpu_ras_eeprom_reset_table(control)) + if (amdgpu_ras_save_bad_pages(adev, NULL)) + dev_warn(adev->dev, "Failed to format RAS EEPROM data in V3 version!\n"); + } + + return 0; +} + +int amdgpu_ras_recovery_init(struct amdgpu_device *adev, bool init_bp_info) { 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)) @@ -2951,38 +3861,19 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) } mutex_init(&con->recovery_lock); + mutex_init(&con->poison_lock); INIT_WORK(&con->recovery_work, amdgpu_ras_do_recovery); atomic_set(&con->in_recovery, 0); + atomic_set(&con->rma_in_recovery, 0); con->eeprom_control.bad_channel_bitmap = 0; max_eeprom_records_count = amdgpu_ras_eeprom_max_record_count(&con->eeprom_control); amdgpu_ras_validate_threshold(adev, max_eeprom_records_count); - /* Todo: During test the SMU might fail to read the eeprom through I2C - * when the GPU is pending on XGMI reset during probe time - * (Mostly after second bus reset), skip it now - */ - if (adev->gmc.xgmi.pending_reset) - return 0; - ret = amdgpu_ras_eeprom_init(&con->eeprom_control, &exc_err_limit); - /* - * This calling fails when exc_err_limit is true or - * ret != 0. - */ - if (exc_err_limit || ret) - goto free; - - if (con->eeprom_control.ras_num_recs) { - ret = amdgpu_ras_load_bad_pages(adev); + if (init_bp_info) { + ret = amdgpu_ras_init_badpage_info(adev); if (ret) goto free; - - amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs); - - if (con->update_channel_flag == true) { - amdgpu_dpm_send_hbm_bad_channel_flag(adev, con->eeprom_control.bad_channel_bitmap); - con->update_channel_flag = false; - } } mutex_init(&con->page_rsv_lock); @@ -2990,6 +3881,8 @@ 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); + atomic_set(&con->poison_creation_count, 0); + atomic_set(&con->poison_consumption_count, 0); con->page_retirement_thread = kthread_run(amdgpu_ras_page_retirement_thread, adev, "umc_page_retirement"); if (IS_ERR(con->page_retirement_thread)) { @@ -3017,7 +3910,7 @@ 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; @@ -3029,15 +3922,24 @@ 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); @@ -3053,6 +3955,10 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) kfree(data); mutex_unlock(&con->recovery_lock); + amdgpu_ras_critical_region_init(adev); +#ifdef CONFIG_X86_MCE_AMD + amdgpu_unregister_bad_pages_mca_notifier(adev); +#endif return 0; } /* recovery end */ @@ -3063,6 +3969,8 @@ 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, 12): + case IP_VERSION(13, 0, 14): return true; default: return false; @@ -3074,6 +3982,9 @@ 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, 12): + case IP_VERSION(13, 0, 14): + case IP_VERSION(14, 0, 3): return true; default: return false; @@ -3135,7 +4046,8 @@ static void amdgpu_ras_query_ras_capablity_from_vbios(struct amdgpu_device *adev */ 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)) + amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 3) || + amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(5, 0, 1)) adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__VCN | 1 << AMDGPU_RAS_BLOCK__JPEG); else @@ -3203,6 +4115,11 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev) if (!amdgpu_ras_asic_supported(adev)) return; + if (amdgpu_sriov_vf(adev)) { + if (amdgpu_virt_get_ras_capability(adev)) + goto init_ras_enabled_flag; + } + /* query ras capability from psp */ if (amdgpu_psp_get_ras_capability(&adev->psp)) goto init_ras_enabled_flag; @@ -3231,8 +4148,18 @@ init_ras_enabled_flag: 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; + /* aca is disabled by default except for psp v13_0_6/v13_0_12/v13_0_14 */ + if (!amdgpu_sriov_vf(adev)) { + adev->aca.is_enabled = + (amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 6) || + amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 12) || + amdgpu_ip_version(adev, MP0_HWIP, 0) == IP_VERSION(13, 0, 14)); + } + + /* 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) @@ -3255,7 +4182,6 @@ static void amdgpu_ras_counte_dw(struct work_struct *work) atomic_set(&con->ras_ue_count, ue_count); } - pm_runtime_mark_last_busy(dev->dev); Out: pm_runtime_put_autosuspend(dev->dev); } @@ -3270,10 +4196,17 @@ static int amdgpu_get_ras_schema(struct amdgpu_device *adev) static void ras_event_mgr_init(struct ras_event_manager *mgr) { + struct ras_event_state *event_state; int i; - for (i = 0; i < ARRAY_SIZE(mgr->seqnos); i++) - atomic64_set(&mgr->seqnos[i], 0); + 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) @@ -3288,7 +4221,7 @@ static void amdgpu_ras_event_mgr_init(struct amdgpu_device *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 (!amdgpu_reset_in_recovery(adev)) { if (!hive || adev->gmc.xgmi.node_id == 0) ras_event_mgr_init(ras->event_mgr); } @@ -3297,6 +4230,27 @@ static void amdgpu_ras_event_mgr_init(struct amdgpu_device *adev) 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, 12): + con->reserved_pages_in_bytes = AMDGPU_RAS_RESERVED_VRAM_SIZE_DEFAULT; + break; + case IP_VERSION(13, 0, 14): + con->reserved_pages_in_bytes = (AMDGPU_RAS_RESERVED_VRAM_SIZE_DEFAULT << 1); + break; + default: + break; + } +} + int amdgpu_ras_init(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -3364,7 +4318,19 @@ int amdgpu_ras_init(struct amdgpu_device *adev) * check DF RAS */ adev->nbio.ras = &nbio_v4_3_ras; break; + case IP_VERSION(6, 3, 1): + if (adev->ras_hw_enabled & (1 << AMDGPU_RAS_BLOCK__DF)) + /* unlike other generation of nbio ras, + * nbif v6_3_1 only support fatal error interrupt + * to inform software that DF is freezed due to + * system fatal error event. driver should not + * enable nbio ras in such case. Instead, + * check DF RAS + */ + adev->nbio.ras = &nbif_v6_3_1_ras; + break; case IP_VERSION(7, 9, 0): + case IP_VERSION(7, 9, 1): if (!adev->gmc.is_app_apu) adev->nbio.ras = &nbio_v7_9_ras; break; @@ -3402,11 +4368,28 @@ 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; + } + + con->init_task_pid = task_pid_nr(current); + get_task_comm(con->init_task_comm, current); + + mutex_init(&con->critical_region_lock); + INIT_LIST_HEAD(&con->critical_region_head); + dev_info(adev->dev, "RAS INFO: ras initialized successfully, " "hardware ability[%x] ras_mask[%x]\n", adev->ras_hw_enabled, adev->ras_enabled); @@ -3474,7 +4457,7 @@ int amdgpu_ras_block_late_init(struct amdgpu_device *adev, r = amdgpu_ras_feature_enable_on_boot(adev, ras_block, 1); if (r) { - if (adev->in_suspend || amdgpu_in_reset(adev)) { + if (adev->in_suspend || amdgpu_reset_in_recovery(adev)) { /* in resume phase, if fail to enable ras, * clean up all ras fs nodes, and disable ras */ goto cleanup; @@ -3486,7 +4469,7 @@ int amdgpu_ras_block_late_init(struct amdgpu_device *adev, amdgpu_persistent_edc_harvesting(adev, ras_block); /* in resume phase, no need to create ras fs node */ - if (adev->in_suspend || amdgpu_in_reset(adev)) + if (adev->in_suspend || amdgpu_reset_in_recovery(adev)) return 0; ras_obj = container_of(ras_block, struct amdgpu_ras_block_object, ras_comm); @@ -3613,25 +4596,30 @@ 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_reset_in_recovery(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) && !amdgpu_sriov_ras_telemetry_en(adev)) + return 0; + list_for_each_entry_safe(node, tmp, &adev->ras_list, node) { obj = node->ras_obj; if (!obj) { @@ -3681,6 +4669,9 @@ int amdgpu_ras_fini(struct amdgpu_device *adev) if (!adev->ras_enabled || !con) return 0; + amdgpu_ras_critical_region_fini(adev); + mutex_destroy(&con->critical_region_lock); + list_for_each_entry_safe(ras_node, tmp, &adev->ras_list, node) { if (ras_node->ras_obj) { obj = ras_node->ras_obj; @@ -3699,8 +4690,12 @@ 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"); @@ -3723,7 +4718,7 @@ bool amdgpu_ras_get_fed_status(struct amdgpu_device *adev) if (!ras) return false; - return atomic_read(&ras->fed); + return test_bit(AMDGPU_RAS_BLOCK__LAST, &ras->ras_err_state); } void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status) @@ -3731,38 +4726,135 @@ void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status) struct amdgpu_ras *ras; ras = amdgpu_ras_get_context(adev); + if (ras) { + if (status) + set_bit(AMDGPU_RAS_BLOCK__LAST, &ras->ras_err_state); + else + clear_bit(AMDGPU_RAS_BLOCK__LAST, &ras->ras_err_state); + } +} + +void amdgpu_ras_clear_err_state(struct amdgpu_device *adev) +{ + struct amdgpu_ras *ras; + + ras = amdgpu_ras_get_context(adev); + if (ras) { + ras->ras_err_state = 0; + ras->gpu_reset_flags = 0; + } +} + +void amdgpu_ras_set_err_poison(struct amdgpu_device *adev, + enum amdgpu_ras_block block) +{ + struct amdgpu_ras *ras; + + ras = amdgpu_ras_get_context(adev); if (ras) - atomic_set(&ras->fed, !!status); + set_bit(block, &ras->ras_err_state); +} + +bool amdgpu_ras_is_err_state(struct amdgpu_device *adev, int block) +{ + struct amdgpu_ras *ras; + + ras = amdgpu_ras_get_context(adev); + if (ras) { + if (block == AMDGPU_RAS_BLOCK__ANY) + return (ras->ras_err_state != 0); + else + return test_bit(block, &ras->ras_err_state) || + test_bit(AMDGPU_RAS_BLOCK__LAST, + &ras->ras_err_state); + } + + return false; +} + +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; } -bool amdgpu_ras_event_id_is_valid(struct amdgpu_device *adev, u64 id) +int amdgpu_ras_mark_ras_event_caller(struct amdgpu_device *adev, enum ras_event_type type, + const void *caller) { - return !(id & BIT_ULL(63)); + struct ras_event_manager *event_mgr; + struct ras_event_state *event_state; + int ret = 0; + + if (amdgpu_uniras_enabled(adev)) + return 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 amdgpu_ras *ras = amdgpu_ras_get_context(adev); + 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_ISR: - id = (u64)atomic64_read(&ras->event_mgr->seqnos[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 = BIT_ULL(63) | 0ULL; + id = RAS_EVENT_INVALID_ID; break; } return id; } -void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev) +int 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]); + enum ras_event_type type = RAS_EVENT_TYPE_FATAL; + u64 event_id = RAS_EVENT_INVALID_ID; + + if (amdgpu_uniras_enabled(adev)) + return 0; + + if (!amdgpu_ras_mark_ras_event(adev, type)) + event_id = amdgpu_ras_acquire_event_id(adev, type); RAS_EVENT_LOG(adev, event_id, "uncorrectable hardware error" "(ERREVENT_ATHUB_INTERRUPT) detected!\n"); @@ -3771,6 +4863,8 @@ void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev) ras->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE1_RESET; amdgpu_ras_reset_gpu(adev); } + + return -EBUSY; } bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev) @@ -3898,6 +4992,28 @@ static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev) notifier_registered = true; } } +static void amdgpu_unregister_bad_pages_mca_notifier(struct amdgpu_device *adev) +{ + int i, j; + + if (!notifier_registered && !mce_adev_list.num_gpu) + return; + for (i = 0, j = 0; i < mce_adev_list.num_gpu; i++) { + if (mce_adev_list.devs[i] == adev) + mce_adev_list.devs[i] = NULL; + if (!mce_adev_list.devs[i]) + ++j; + } + + if (j == mce_adev_list.num_gpu) { + mce_adev_list.num_gpu = 0; + /* Unregister x86 notifier with MCE subsystem. */ + if (notifier_registered) { + mce_unregister_decode_chain(&amdgpu_bad_page_nb); + notifier_registered = false; + } + } +} #endif struct amdgpu_ras *amdgpu_ras_get_context(struct amdgpu_device *adev) @@ -3952,8 +5068,33 @@ int amdgpu_ras_reset_gpu(struct amdgpu_device *adev) { struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); - if (atomic_cmpxchg(&ras->in_recovery, 0, 1) == 0) + /* 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) { + struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); + int hive_ras_recovery = 0; + + if (hive) { + hive_ras_recovery = atomic_read(&hive->ras_recovery); + amdgpu_put_xgmi_hive(hive); + } + /* In the case of multiple GPUs, after a GPU has started + * resetting all GPUs on hive, other GPUs do not need to + * trigger GPU reset again. + */ + if (!hive_ras_recovery) + amdgpu_reset_domain_schedule(ras->adev->reset_domain, &ras->recovery_work); + else + atomic_set(&ras->in_recovery, 0); + } else { + flush_work(&ras->recovery_work); amdgpu_reset_domain_schedule(ras->adev->reset_domain, &ras->recovery_work); + } + return 0; } @@ -4016,11 +5157,14 @@ bool amdgpu_ras_get_error_query_mode(struct amdgpu_device *adev, return false; } - if ((smu_funcs && smu_funcs->set_debug_mode) || (mca_funcs && mca_funcs->mca_set_debug_mode)) + if (amdgpu_sriov_vf(adev)) { + *error_query_mode = AMDGPU_RAS_VIRT_ERROR_COUNT_QUERY; + } else 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 + } else { *error_query_mode = AMDGPU_RAS_DIRECT_ERROR_QUERY; + } return true; } @@ -4271,8 +5415,6 @@ 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++; @@ -4282,34 +5424,9 @@ 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; - - /* This function will be retired. */ - return; - 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; @@ -4323,9 +5440,6 @@ 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; @@ -4333,8 +5447,8 @@ 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; @@ -4355,8 +5469,8 @@ 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; @@ -4370,9 +5484,6 @@ 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; @@ -4382,64 +5493,84 @@ 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_addr = (mmMP0_SMN_C2PMSG_126 << 2) + + aqua_vanjaram_encode_ext_smn_addressing(instance); + boot_error = 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); + 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_GENERIC(boot_error)) + dev_info(adev->dev, + "socket: %d, aid: %d, fw_status: 0x%x, Boot Controller Generic Error\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; @@ -4448,41 +5579,22 @@ 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); } } @@ -4493,6 +5605,9 @@ int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn) uint64_t start = pfn << AMDGPU_GPU_PAGE_SHIFT; int ret = 0; + if (amdgpu_ras_check_critical_address(adev, start)) + return 0; + mutex_lock(&con->page_rsv_lock); ret = amdgpu_vram_mgr_query_page_status(mgr, start); if (ret == -ENOENT) @@ -4501,3 +5616,133 @@ int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn) 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 (amdgpu_uniras_enabled(adev)) + return amdgpu_ras_mgr_is_rma(adev); + + if (!con) + return false; + + return con->is_rma; +} + +int amdgpu_ras_add_critical_region(struct amdgpu_device *adev, + struct amdgpu_bo *bo) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct amdgpu_vram_mgr_resource *vres; + struct ras_critical_region *region; + struct drm_buddy_block *block; + int ret = 0; + + if (!bo || !bo->tbo.resource) + return -EINVAL; + + vres = to_amdgpu_vram_mgr_resource(bo->tbo.resource); + + mutex_lock(&con->critical_region_lock); + + /* Check if the bo had been recorded */ + list_for_each_entry(region, &con->critical_region_head, node) + if (region->bo == bo) + goto out; + + /* Record new critical amdgpu bo */ + list_for_each_entry(block, &vres->blocks, link) { + region = kzalloc(sizeof(*region), GFP_KERNEL); + if (!region) { + ret = -ENOMEM; + goto out; + } + region->bo = bo; + region->start = amdgpu_vram_mgr_block_start(block); + region->size = amdgpu_vram_mgr_block_size(block); + list_add_tail(®ion->node, &con->critical_region_head); + } + +out: + mutex_unlock(&con->critical_region_lock); + + return ret; +} + +static void amdgpu_ras_critical_region_init(struct amdgpu_device *adev) +{ + amdgpu_ras_add_critical_region(adev, adev->mman.fw_reserved_memory); +} + +static void amdgpu_ras_critical_region_fini(struct amdgpu_device *adev) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_critical_region *region, *tmp; + + mutex_lock(&con->critical_region_lock); + list_for_each_entry_safe(region, tmp, &con->critical_region_head, node) { + list_del(®ion->node); + kfree(region); + } + mutex_unlock(&con->critical_region_lock); +} + +bool amdgpu_ras_check_critical_address(struct amdgpu_device *adev, uint64_t addr) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_critical_region *region; + bool ret = false; + + mutex_lock(&con->critical_region_lock); + list_for_each_entry(region, &con->critical_region_head, node) { + if ((region->start <= addr) && + (addr < (region->start + region->size))) { + ret = true; + break; + } + } + mutex_unlock(&con->critical_region_lock); + + return ret; +} + +void amdgpu_ras_pre_reset(struct amdgpu_device *adev, + struct list_head *device_list) +{ + struct amdgpu_device *tmp_adev = NULL; + + list_for_each_entry(tmp_adev, device_list, reset_list) { + if (amdgpu_uniras_enabled(tmp_adev)) + amdgpu_ras_mgr_pre_reset(tmp_adev); + } +} + +void amdgpu_ras_post_reset(struct amdgpu_device *adev, + struct list_head *device_list) +{ + struct amdgpu_device *tmp_adev = NULL; + + list_for_each_entry(tmp_adev, device_list, reset_list) { + if (amdgpu_uniras_enabled(tmp_adev)) + amdgpu_ras_mgr_post_reset(tmp_adev); + } +} |