From b6485bed40d7859735bdbfedbd55dcc8366a88a7 Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Mon, 6 Dec 2021 15:54:54 +0800 Subject: drm/amdkfd: reset queue which consumes RAS poison (v2) CP supports unmap queue with reset mode which only destroys specific queue without affecting others. Replacing whole gpu reset with reset queue mode for RAS poison consumption saves much time, and we can also fallback to gpu reset solution if reset queue fails. v2: Return directly if process is NULL; Reset queue solution is not applicable to SDMA, fallback to legacy way; Call kfd_unref_process after lookup process. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Acked-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 46cf48b3904a..0bf09a94d944 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -721,13 +721,13 @@ bool amdgpu_amdkfd_have_atomics_support(struct amdgpu_device *adev) return adev->have_atomics_support; } -void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev) +void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, bool reset) { struct ras_err_data err_data = {0, 0, 0, NULL}; /* CPU MCA will handle page retirement if connected_to_cpu is 1 */ if (!adev->gmc.xgmi.connected_to_cpu) - amdgpu_umc_process_ras_data_cb(adev, &err_data, NULL); - else + amdgpu_umc_do_page_retirement(adev, &err_data, NULL, reset); + else if (reset) amdgpu_amdkfd_gpu_reset(adev); } -- cgit v1.2.3 From fec8c5244fc07b1f6a3249a8714489f594ff5c4f Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Mon, 20 Dec 2021 16:36:54 +0800 Subject: drm/amdgpu: save error count in RAS poison handler Otherwise the RAS error count couldn't be queried from sysfs. Signed-off-by: Tao Zhou Reviewed-by: Stanley.Yang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 168 ++++++++++++++++------------- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 3 +- 3 files changed, 97 insertions(+), 76 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 0bf09a94d944..776a947b45df 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -727,7 +727,7 @@ void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, bo /* CPU MCA will handle page retirement if connected_to_cpu is 1 */ if (!adev->gmc.xgmi.connected_to_cpu) - amdgpu_umc_do_page_retirement(adev, &err_data, NULL, reset); + amdgpu_umc_poison_handler(adev, &err_data, reset); else if (reset) amdgpu_amdkfd_gpu_reset(adev); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 0c33f367a4e5..46264a4002f7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -23,79 +23,7 @@ #include "amdgpu_ras.h" -static int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, - void *ras_error_status, - struct amdgpu_iv_entry *entry) -{ - return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry, true); -} - -int amdgpu_umc_ras_late_init(struct amdgpu_device *adev) -{ - int r; - struct ras_fs_if fs_info = { - .sysfs_name = "umc_err_count", - }; - struct ras_ih_if ih_info = { - .cb = amdgpu_umc_process_ras_data_cb, - }; - - if (!adev->umc.ras_if) { - adev->umc.ras_if = - kmalloc(sizeof(struct ras_common_if), GFP_KERNEL); - if (!adev->umc.ras_if) - return -ENOMEM; - adev->umc.ras_if->block = AMDGPU_RAS_BLOCK__UMC; - adev->umc.ras_if->type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; - adev->umc.ras_if->sub_block_index = 0; - } - ih_info.head = fs_info.head = *adev->umc.ras_if; - - r = amdgpu_ras_late_init(adev, adev->umc.ras_if, - &fs_info, &ih_info); - if (r) - goto free; - - if (amdgpu_ras_is_supported(adev, adev->umc.ras_if->block)) { - r = amdgpu_irq_get(adev, &adev->gmc.ecc_irq, 0); - if (r) - goto late_fini; - } else { - r = 0; - goto free; - } - - /* ras init of specific umc version */ - if (adev->umc.ras_funcs && - adev->umc.ras_funcs->err_cnt_init) - adev->umc.ras_funcs->err_cnt_init(adev); - - return 0; - -late_fini: - amdgpu_ras_late_fini(adev, adev->umc.ras_if, &ih_info); -free: - kfree(adev->umc.ras_if); - adev->umc.ras_if = NULL; - return r; -} - -void amdgpu_umc_ras_fini(struct amdgpu_device *adev) -{ - if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC) && - adev->umc.ras_if) { - struct ras_common_if *ras_if = adev->umc.ras_if; - struct ras_ih_if ih_info = { - .head = *ras_if, - .cb = amdgpu_umc_process_ras_data_cb, - }; - - amdgpu_ras_late_fini(adev, ras_if, &ih_info); - kfree(ras_if); - } -} - -int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, +static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, void *ras_error_status, struct amdgpu_iv_entry *entry, bool reset) @@ -180,6 +108,100 @@ int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, return AMDGPU_RAS_SUCCESS; } +int amdgpu_umc_poison_handler(struct amdgpu_device *adev, + void *ras_error_status, + bool reset) +{ + int ret; + struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; + struct ras_common_if head = { + .block = AMDGPU_RAS_BLOCK__UMC, + }; + struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head); + + ret = + amdgpu_umc_do_page_retirement(adev, ras_error_status, NULL, reset); + + if (ret == AMDGPU_RAS_SUCCESS && obj) { + obj->err_data.ue_count += err_data->ue_count; + obj->err_data.ce_count += err_data->ce_count; + } + + return ret; +} + +static int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, + void *ras_error_status, + struct amdgpu_iv_entry *entry) +{ + return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry, true); +} + +int amdgpu_umc_ras_late_init(struct amdgpu_device *adev) +{ + int r; + struct ras_fs_if fs_info = { + .sysfs_name = "umc_err_count", + }; + struct ras_ih_if ih_info = { + .cb = amdgpu_umc_process_ras_data_cb, + }; + + if (!adev->umc.ras_if) { + adev->umc.ras_if = + kmalloc(sizeof(struct ras_common_if), GFP_KERNEL); + if (!adev->umc.ras_if) + return -ENOMEM; + adev->umc.ras_if->block = AMDGPU_RAS_BLOCK__UMC; + adev->umc.ras_if->type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; + adev->umc.ras_if->sub_block_index = 0; + } + ih_info.head = fs_info.head = *adev->umc.ras_if; + + r = amdgpu_ras_late_init(adev, adev->umc.ras_if, + &fs_info, &ih_info); + if (r) + goto free; + + if (amdgpu_ras_is_supported(adev, adev->umc.ras_if->block)) { + r = amdgpu_irq_get(adev, &adev->gmc.ecc_irq, 0); + if (r) + goto late_fini; + } else { + r = 0; + goto free; + } + + /* ras init of specific umc version */ + if (adev->umc.ras_funcs && + adev->umc.ras_funcs->err_cnt_init) + adev->umc.ras_funcs->err_cnt_init(adev); + + return 0; + +late_fini: + amdgpu_ras_late_fini(adev, adev->umc.ras_if, &ih_info); +free: + kfree(adev->umc.ras_if); + adev->umc.ras_if = NULL; + return r; +} + +void amdgpu_umc_ras_fini(struct amdgpu_device *adev) +{ + if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC) && + adev->umc.ras_if) { + struct ras_common_if *ras_if = adev->umc.ras_if; + struct ras_ih_if ih_info = { + .head = *ras_if, + .cb = amdgpu_umc_process_ras_data_cb, + }; + + amdgpu_ras_late_fini(adev, ras_if, &ih_info); + kfree(ras_if); + } +} + int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 8d18d5121f66..b72194e8bfe5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -78,9 +78,8 @@ struct amdgpu_umc { int amdgpu_umc_ras_late_init(struct amdgpu_device *adev); void amdgpu_umc_ras_fini(struct amdgpu_device *adev); -int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, +int amdgpu_umc_poison_handler(struct amdgpu_device *adev, void *ras_error_status, - struct amdgpu_iv_entry *entry, bool reset); int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, -- cgit v1.2.3