[PATCH] drm/amdgpu: avoid NULL pointer dereference

Guchun Chen guchun.chen at amd.com
Wed Dec 22 14:19:56 UTC 2021


amdgpu_umc_poison_handler for UMC RAS consumption gets
called in KFD queue reset, but it needs to return early when
RAS context is NULL. This can guarantee lower access to
RAS context like in amdgpu_umc_do_page_retirement. Also
improve coding style in amdgpu_umc_poison_handler.

Signed-off-by: Guchun Chen <guchun.chen at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
index 46264a4002f7..b455fc7d1546 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
@@ -112,16 +112,20 @@ 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);
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj;
+	int ret;
+
+	if (!con)
+		return 0;
 
-	ret =
-		amdgpu_umc_do_page_retirement(adev, ras_error_status, NULL, reset);
+	ret = amdgpu_umc_do_page_retirement(adev, ras_error_status, NULL, reset);
 
+	obj = amdgpu_ras_find_obj(adev, &head);
 	if (ret == AMDGPU_RAS_SUCCESS && obj) {
 		obj->err_data.ue_count += err_data->ue_count;
 		obj->err_data.ce_count += err_data->ce_count;
-- 
2.17.1



More information about the amd-gfx mailing list