diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 239 |
1 files changed, 230 insertions, 9 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 352ce16a0963..1adc81a55734 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -122,6 +122,8 @@ const char *get_ras_block_str(struct ras_common_if *ras_block) #define MAX_UMC_POISON_POLLING_TIME_ASYNC 100 //ms +#define AMDGPU_RAS_RETIRE_PAGE_INTERVAL 100 //ms + enum amdgpu_ras_retire_page_reservation { AMDGPU_RAS_RETIRE_PAGE_RESERVED, AMDGPU_RAS_RETIRE_PAGE_PENDING, @@ -1248,6 +1250,10 @@ int amdgpu_ras_bind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk, { struct ras_manager *obj; + /* in resume phase, no need to create aca fs node */ + if (adev->in_suspend || amdgpu_in_reset(adev)) + return 0; + obj = get_ras_manager(adev, blk); if (!obj) return -EINVAL; @@ -2076,6 +2082,17 @@ static void amdgpu_ras_interrupt_poison_creation_handler(struct ras_manager *obj { dev_info(obj->adev->dev, "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); + + wake_up(&con->page_retirement_wq); + } } static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj, @@ -2386,7 +2403,7 @@ static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, .flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED, }; status = amdgpu_vram_mgr_query_page_status(&adev->mman.vram_mgr, - data->bps[i].retired_page); + data->bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT); if (status == -EBUSY) (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING; else if (status == -ENOENT) @@ -2545,9 +2562,7 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, goto out; } - amdgpu_vram_mgr_reserve_range(&adev->mman.vram_mgr, - bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT, - AMDGPU_GPU_PAGE_SIZE); + amdgpu_ras_reserve_page(adev, bps[i].retired_page); memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps)); data->count++; @@ -2703,10 +2718,167 @@ static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev, } } +int amdgpu_ras_put_poison_req(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) +{ + int ret = 0; + struct ras_poison_msg poison_msg; + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + memset(&poison_msg, 0, sizeof(poison_msg)); + poison_msg.block = block; + poison_msg.pasid = pasid; + poison_msg.reset = reset; + poison_msg.pasid_fn = pasid_fn; + poison_msg.data = data; + + ret = kfifo_put(&con->poison_fifo, poison_msg); + if (!ret) { + dev_err(adev->dev, "Poison message fifo is full!\n"); + return -ENOSPC; + } + + return 0; +} + +static int amdgpu_ras_get_poison_req(struct amdgpu_device *adev, + struct ras_poison_msg *poison_msg) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + + return kfifo_get(&con->poison_fifo, poison_msg); +} + +static void amdgpu_ras_ecc_log_init(struct ras_ecc_log_info *ecc_log) +{ + mutex_init(&ecc_log->lock); + + /* 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; +} + +static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) +{ + struct radix_tree_iter iter; + void __rcu **slot; + struct ras_ecc_err *ecc_err; + + mutex_lock(&ecc_log->lock); + radix_tree_for_each_slot(slot, &ecc_log->de_page_tree, &iter, 0) { + ecc_err = radix_tree_deref_slot(slot); + kfree(ecc_err->err_pages.pfn); + kfree(ecc_err); + radix_tree_iter_delete(&ecc_log->de_page_tree, &iter, slot); + } + mutex_unlock(&ecc_log->lock); + + mutex_destroy(&ecc_log->lock); + ecc_log->de_updated = false; +} + +static void amdgpu_ras_do_page_retirement(struct work_struct *work) +{ + struct amdgpu_ras *con = container_of(work, struct amdgpu_ras, + page_retirement_dwork.work); + struct amdgpu_device *adev = con->adev; + struct ras_err_data err_data; + + if (amdgpu_in_reset(adev) || atomic_read(&con->in_recovery)) + return; + + amdgpu_ras_error_data_init(&err_data); + + amdgpu_umc_handle_bad_pages(adev, &err_data); + + 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); +} + +static int amdgpu_ras_query_ecc_status(struct amdgpu_device *adev, + enum amdgpu_ras_block ras_block, uint32_t timeout_ms) +{ + int ret = 0; + struct ras_ecc_log_info *ecc_log; + struct ras_query_if info; + uint32_t timeout = timeout_ms; + struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); + + memset(&info, 0, sizeof(info)); + info.head.block = ras_block; + + ecc_log = &ras->umc_ecc_log; + ecc_log->de_updated = false; + do { + ret = amdgpu_ras_query_error_status(adev, &info); + if (ret) { + dev_err(adev->dev, "Failed to query ras error! ret:%d\n", ret); + return ret; + } + + if (timeout && !ecc_log->de_updated) { + msleep(1); + timeout--; + } + } while (timeout && !ecc_log->de_updated); + + if (timeout_ms && !timeout) { + dev_warn(adev->dev, "Can't find deferred error\n"); + return -ETIMEDOUT; + } + + return 0; +} + +static void amdgpu_ras_poison_creation_handler(struct amdgpu_device *adev, + uint32_t timeout) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + int ret; + + ret = amdgpu_ras_query_ecc_status(adev, AMDGPU_RAS_BLOCK__UMC, timeout); + if (!ret) + schedule_delayed_work(&con->page_retirement_dwork, 0); +} + +static int amdgpu_ras_poison_consumption_handler(struct amdgpu_device *adev, + struct ras_poison_msg *poison_msg) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + uint32_t reset = poison_msg->reset; + uint16_t pasid = poison_msg->pasid; + + kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); + + if (poison_msg->pasid_fn) + poison_msg->pasid_fn(adev, pasid, poison_msg->data); + + if (reset) { + flush_delayed_work(&con->page_retirement_dwork); + + con->gpu_reset_flags |= reset; + amdgpu_ras_reset_gpu(adev); + } + + return 0; +} + static int amdgpu_ras_page_retirement_thread(void *param) { struct amdgpu_device *adev = (struct amdgpu_device *)param; struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct ras_poison_msg poison_msg; + enum amdgpu_ras_block ras_block; + bool poison_creation_is_handled = false; while (!kthread_should_stop()) { @@ -2717,13 +2889,34 @@ static int amdgpu_ras_page_retirement_thread(void *param) if (kthread_should_stop()) break; - dev_info(adev->dev, "Start processing page retirement. request:%d\n", - atomic_read(&con->page_retirement_req_cnt)); - atomic_dec(&con->page_retirement_req_cnt); - amdgpu_umc_bad_page_polling_timeout(adev, - 0, MAX_UMC_POISON_POLLING_TIME_ASYNC); + if (!amdgpu_ras_get_poison_req(adev, &poison_msg)) + continue; + + ras_block = poison_msg.block; + + dev_info(adev->dev, "Start processing ras block %s(%d)\n", + ras_block_str(ras_block), ras_block); + + 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); + + amdgpu_ras_poison_consumption_handler(adev, &poison_msg); + poison_creation_is_handled = false; + } } return 0; @@ -2792,6 +2985,8 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) } } + mutex_init(&con->page_rsv_lock); + INIT_KFIFO(con->poison_fifo); mutex_init(&con->page_retirement_lock); init_waitqueue_head(&con->page_retirement_wq); atomic_set(&con->page_retirement_req_cnt, 0); @@ -2802,6 +2997,8 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) dev_warn(adev->dev, "Failed to create umc_page_retirement thread!!!\n"); } + INIT_DELAYED_WORK(&con->page_retirement_dwork, amdgpu_ras_do_page_retirement); + amdgpu_ras_ecc_log_init(&con->umc_ecc_log); #ifdef CONFIG_X86_MCE_AMD if ((adev->asic_type == CHIP_ALDEBARAN) && (adev->gmc.xgmi.connected_to_cpu)) @@ -2842,8 +3039,14 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) atomic_set(&con->page_retirement_req_cnt, 0); + mutex_destroy(&con->page_rsv_lock); + cancel_work_sync(&con->recovery_work); + cancel_delayed_work_sync(&con->page_retirement_dwork); + + amdgpu_ras_ecc_log_fini(&con->umc_ecc_log); + mutex_lock(&con->recovery_lock); con->eh_data = NULL; kfree(data->bps); @@ -4083,6 +4286,8 @@ void amdgpu_ras_add_mca_err_addr(struct ras_err_info *err_info, struct ras_err_a { 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; @@ -4280,3 +4485,19 @@ void amdgpu_ras_query_boot_status(struct amdgpu_device *adev, u32 num_instances) amdgpu_ras_boot_time_error_reporting(adev, i, boot_error); } } + +int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn) +{ + struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr; + uint64_t start = pfn << AMDGPU_GPU_PAGE_SHIFT; + int ret = 0; + + mutex_lock(&con->page_rsv_lock); + ret = amdgpu_vram_mgr_query_page_status(mgr, start); + if (ret == -ENOENT) + ret = amdgpu_vram_mgr_reserve_range(mgr, start, AMDGPU_GPU_PAGE_SIZE); + mutex_unlock(&con->page_rsv_lock); + + return ret; +} |