Re: [PATCH 5/6] amdgpu: fix invalid memory access in amdgpu_fence_driver_sw_fini()

[Gerry Liu <gerry@xxxxxxxxxxxxxxxxx>]

2025年1月3日 07:09，Chen, Xiaogang <xiaogang.chen@xxxxxxx> 写道：

On 1/1/2025 11:36 PM, Jiang Liu wrote:

Function detects initialization status by checking sched->ops, so set
sched->ops to non-NULL just before return in function drm_sched_init()
to avoid possible invalid memory access on error recover path.

Signed-off-by: Jiang Liu <gerry@xxxxxxxxxxxxxxxxx>
---
 drivers/gpu/drm/scheduler/sched_main.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
index 5adab4b3386c..ce2afdfc0158 100644
--- a/drivers/gpu/drm/scheduler/sched_main.c
+++ b/drivers/gpu/drm/scheduler/sched_main.c
@@ -1331,6 +1331,8 @@ int drm_sched_init(struct drm_gpu_scheduler *sched,
  if (sched->own_submit_wq)
  destroy_workqueue(sched->submit_wq);
  drm_err(sched, "%s: Failed to setup GPU scheduler--out of memory\n", __func__);
+ // amdgpu_fence_driver_sw_fini() checks sched->ops for status.
+ sched->ops = NULL;

Instead change drm/scheduler how about change amdgpu driver? At amdgpu_device_init_schedulers if drm_sched_init fail for a ring scheduler set

ring->sched->op = NULL;

It is not drm issue. If drm's scheduler init fail its operation has no meaning, we can set it to null at amdgpu driver.

Good suggestion, I will fix it in this way.

Regards

Xiaogang

  return -ENOMEM;
 }
 EXPORT_SYMBOL(drm_sched_init);
@@ -1375,6 +1377,7 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
  sched->ready = false;
  kfree(sched->sched_rq);
  sched->sched_rq = NULL;
+ sched->ops = NULL;
 }
 EXPORT_SYMBOL(drm_sched_fini);