[PATCH 1/4] drm/amdgpu: add RAS page retirement functions for MCA

Tao Zhou tao.zhou1 at amd.com
Wed Oct 19 08:11:47 UTC 2022


Define page retirement functions for MCA platform.

v2: remove page retirement handling from MCA poison handler,
    let MCA notifier do page retirement.

Signed-off-by: Tao Zhou <tao.zhou1 at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 67 +++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h |  2 +
 2 files changed, 69 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
index aad3c8b4c810..9494fa14db9a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
@@ -22,6 +22,73 @@
  */
 
 #include "amdgpu.h"
+#include "umc_v6_7.h"
+
+static int amdgpu_umc_convert_error_address(struct amdgpu_device *adev,
+				    struct ras_err_data *err_data, uint64_t err_addr,
+				    uint32_t ch_inst, uint32_t umc_inst)
+{
+	switch (adev->ip_versions[UMC_HWIP][0]) {
+	case IP_VERSION(6, 7, 0):
+		umc_v6_7_convert_error_address(adev,
+				err_data, err_addr, ch_inst, umc_inst);
+		break;
+	default:
+		dev_warn(adev->dev,
+			 "UMC address to Physical address translation is not supported\n");
+		return AMDGPU_RAS_FAIL;
+	}
+
+	return AMDGPU_RAS_SUCCESS;
+}
+
+int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
+			uint64_t err_addr, uint32_t ch_inst, uint32_t umc_inst)
+{
+	struct ras_err_data err_data = {0, 0, 0, NULL};
+	int ret = AMDGPU_RAS_FAIL;
+
+	err_data.err_addr =
+		kcalloc(adev->umc.max_ras_err_cnt_per_query,
+			sizeof(struct eeprom_table_record), GFP_KERNEL);
+	if (!err_data.err_addr) {
+		dev_warn(adev->dev,
+			"Failed to alloc memory for umc error record in MCA notifier!\n");
+		return AMDGPU_RAS_FAIL;
+	}
+
+	/*
+	 * Translate UMC channel address to Physical address
+	 */
+	ret = amdgpu_umc_convert_error_address(adev, &err_data, err_addr,
+					ch_inst, umc_inst);
+	if (ret)
+		goto out;
+
+	if (amdgpu_bad_page_threshold != 0) {
+		amdgpu_ras_add_bad_pages(adev, err_data.err_addr,
+						err_data.err_addr_cnt);
+		amdgpu_ras_save_bad_pages(adev);
+	}
+
+out:
+	kfree(err_data.err_addr);
+	return ret;
+}
+
+static int amdgpu_umc_poison_handler_mca(struct amdgpu_device *adev,
+		struct ras_err_data *err_data, bool reset)
+{
+	/* MCA poison handler is only responsible for GPU reset,
+	 * let MCA notifier do page retirement.
+	 */
+	if (reset) {
+		kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
+		amdgpu_ras_reset_gpu(adev);
+	}
+
+	return AMDGPU_RAS_SUCCESS;
+}
 
 static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev,
 		void *ras_error_status,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
index 3629d8f292ef..659a10de29c9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
@@ -98,4 +98,6 @@ void amdgpu_umc_fill_error_record(struct ras_err_data *err_data,
 int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev,
 		void *ras_error_status,
 		struct amdgpu_iv_entry *entry);
+int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
+			uint64_t err_addr, uint32_t ch_inst, uint32_t umc_inst);
 #endif
-- 
2.35.1



More information about the amd-gfx mailing list