From fdcb279d5b798d13b4365bdcf5548855f6c562a1 Mon Sep 17 00:00:00 2001 From: "Stanley.Yang" Date: Thu, 18 Nov 2021 16:30:43 +0800 Subject: drm/amdgpu: query umc error info from ecc_table v2 if smu support ECCTABLE, driver can message smu to get ecc_table then query umc error info from ECCTABLE v2: optimize source code makes logical more reasonable Signed-off-by: Stanley.Yang Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 72 +++++++++++++++++++++++---------- 1 file changed, 50 insertions(+), 22 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index a90029ee9733..6e4bea012ea4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -94,30 +94,58 @@ int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, { struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + int ret = 0; kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); - if (adev->umc.ras_funcs && - adev->umc.ras_funcs->query_ras_error_count) - adev->umc.ras_funcs->query_ras_error_count(adev, ras_error_status); - - if (adev->umc.ras_funcs && - adev->umc.ras_funcs->query_ras_error_address && - adev->umc.max_ras_err_cnt_per_query) { - err_data->err_addr = - kcalloc(adev->umc.max_ras_err_cnt_per_query, - sizeof(struct eeprom_table_record), GFP_KERNEL); - - /* still call query_ras_error_address to clear error status - * even NOMEM error is encountered - */ - if(!err_data->err_addr) - dev_warn(adev->dev, "Failed to alloc memory for " - "umc error address record!\n"); - - /* umc query_ras_error_address is also responsible for clearing - * error status - */ - adev->umc.ras_funcs->query_ras_error_address(adev, ras_error_status); + ret = smu_get_ecc_info(&adev->smu, (void *)&(con->umc_ecc)); + if (ret == -EOPNOTSUPP) { + if (adev->umc.ras_funcs && + adev->umc.ras_funcs->query_ras_error_count) + adev->umc.ras_funcs->query_ras_error_count(adev, ras_error_status); + + if (adev->umc.ras_funcs && + adev->umc.ras_funcs->query_ras_error_address && + adev->umc.max_ras_err_cnt_per_query) { + err_data->err_addr = + kcalloc(adev->umc.max_ras_err_cnt_per_query, + sizeof(struct eeprom_table_record), GFP_KERNEL); + + /* still call query_ras_error_address to clear error status + * even NOMEM error is encountered + */ + if(!err_data->err_addr) + dev_warn(adev->dev, "Failed to alloc memory for " + "umc error address record!\n"); + + /* umc query_ras_error_address is also responsible for clearing + * error status + */ + adev->umc.ras_funcs->query_ras_error_address(adev, ras_error_status); + } + } else if (!ret) { + if (adev->umc.ras_funcs && + adev->umc.ras_funcs->ecc_info_query_ras_error_count) + adev->umc.ras_funcs->ecc_info_query_ras_error_count(adev, ras_error_status); + + if (adev->umc.ras_funcs && + adev->umc.ras_funcs->ecc_info_query_ras_error_address && + adev->umc.max_ras_err_cnt_per_query) { + err_data->err_addr = + kcalloc(adev->umc.max_ras_err_cnt_per_query, + sizeof(struct eeprom_table_record), GFP_KERNEL); + + /* still call query_ras_error_address to clear error status + * even NOMEM error is encountered + */ + if(!err_data->err_addr) + dev_warn(adev->dev, "Failed to alloc memory for " + "umc error address record!\n"); + + /* umc query_ras_error_address is also responsible for clearing + * error status + */ + adev->umc.ras_funcs->ecc_info_query_ras_error_address(adev, ras_error_status); + } } /* only uncorrectable error needs gpu reset */ -- cgit From f4409ee84658ff95d92daace9f094fa6bf80b0e3 Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Fri, 10 Dec 2021 17:01:15 +0800 Subject: drm/amdgpu: add gpu reset control for umc page retirement Add a reset parameter for umc page retirement, let user decide whether call gpu reset in umc page retirement. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Acked-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 6e4bea012ea4..0c33f367a4e5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -23,6 +23,13 @@ #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; @@ -88,9 +95,10 @@ void amdgpu_umc_ras_fini(struct amdgpu_device *adev) } } -int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, +int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, void *ras_error_status, - struct amdgpu_iv_entry *entry) + struct amdgpu_iv_entry *entry, + bool reset) { struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -164,7 +172,8 @@ int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, adev->smu.ppt_funcs->send_hbm_bad_pages_num(&adev->smu, con->eeprom_control.ras_num_recs); } - amdgpu_ras_reset_gpu(adev); + if (reset) + amdgpu_ras_reset_gpu(adev); } kfree(err_data->err_addr); -- cgit 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_umc.c | 168 ++++++++++++++++++-------------- 1 file changed, 95 insertions(+), 73 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c') 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) -- cgit