From c154a96b550b66bdf032216cab613269cb91e79e Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Fri, 25 Jul 2025 10:47:35 +0800 Subject: [PATCH] drm/amdgpu: load RAS bad page from PMFW in page retirement In legacy way, bad page is queried from MCA registers, switch to getting it from PMFW when PMFW manages eeprom data. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 8 +- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 137 ++++++++++++++---------- 2 files changed, 90 insertions(+), 55 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 055a9bbabbdbb..36a5393d6b74e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -3300,7 +3300,13 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev, mutex_lock(&con->recovery_lock); control = &con->eeprom_control; data = con->eh_data; - unit_num = data->count / adev->umc.retire_unit - 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); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 2e039fb778ea8..3eb252de343bb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -96,67 +96,96 @@ void amdgpu_umc_handle_bad_pages(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); + struct amdgpu_ras_eeprom_control *control = &con->eeprom_control; unsigned int error_query_mode; int ret = 0; unsigned long err_count; amdgpu_ras_get_error_query_mode(adev, &error_query_mode); + 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"); + else + err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query; + mutex_lock(&con->page_retirement_lock); - ret = amdgpu_dpm_get_ecc_info(adev, (void *)&(con->umc_ecc)); - if (ret == -EOPNOTSUPP && - error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY) { - if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && - adev->umc.ras->ras_block.hw_ops->query_ras_error_count) - adev->umc.ras->ras_block.hw_ops->query_ras_error_count(adev, ras_error_status); - - if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && - adev->umc.ras->ras_block.hw_ops->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"); - else - err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query; - - /* umc query_ras_error_address is also responsible for clearing - * error status - */ - adev->umc.ras->ras_block.hw_ops->query_ras_error_address(adev, ras_error_status); + if (!amdgpu_ras_smu_eeprom_supported(adev)) { + ret = amdgpu_dpm_get_ecc_info(adev, (void *)&(con->umc_ecc)); + if (ret == -EOPNOTSUPP && + error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY) { + if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && + adev->umc.ras->ras_block.hw_ops->query_ras_error_count) + adev->umc.ras->ras_block.hw_ops->query_ras_error_count(adev, + ras_error_status); + + if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && + adev->umc.ras->ras_block.hw_ops->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"); + else + err_data->err_addr_len = + adev->umc.max_ras_err_cnt_per_query; + + /* umc query_ras_error_address is also responsible for clearing + * error status + */ + adev->umc.ras->ras_block.hw_ops->query_ras_error_address(adev, + ras_error_status); + } + } else if (error_query_mode == AMDGPU_RAS_FIRMWARE_ERROR_QUERY || + (!ret && error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY)) { + if (adev->umc.ras && + adev->umc.ras->ecc_info_query_ras_error_count) + adev->umc.ras->ecc_info_query_ras_error_count(adev, + ras_error_status); + + if (adev->umc.ras && + adev->umc.ras->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"); + else + err_data->err_addr_len = + adev->umc.max_ras_err_cnt_per_query; + + /* umc query_ras_error_address is also responsible for clearing + * error status + */ + adev->umc.ras->ecc_info_query_ras_error_address(adev, + ras_error_status); + } } - } else if (error_query_mode == AMDGPU_RAS_FIRMWARE_ERROR_QUERY || - (!ret && error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY)) { - if (adev->umc.ras && - adev->umc.ras->ecc_info_query_ras_error_count) - adev->umc.ras->ecc_info_query_ras_error_count(adev, ras_error_status); - - if (adev->umc.ras && - adev->umc.ras->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"); - else - err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query; - - /* umc query_ras_error_address is also responsible for clearing - * error status - */ - adev->umc.ras->ecc_info_query_ras_error_address(adev, ras_error_status); + } else { + if (!amdgpu_ras_eeprom_update_record_num(control)) { + err_data->err_addr_cnt = err_data->de_count = + control->ras_num_recs - control->ras_num_recs_old; + amdgpu_ras_eeprom_read_idx(control, err_data->err_addr, + control->ras_num_recs_old, err_data->de_count); } } @@ -166,7 +195,7 @@ void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, if ((amdgpu_bad_page_threshold != 0) && err_data->err_addr_cnt) { amdgpu_ras_add_bad_pages(adev, err_data->err_addr, - err_data->err_addr_cnt, false); + err_data->err_addr_cnt, amdgpu_ras_smu_eeprom_supported(adev)); amdgpu_ras_save_bad_pages(adev, &err_count); amdgpu_dpm_send_hbm_bad_pages_num(adev, -- 2.47.3