Re: [PATCH 05/12] drm/amdgpu: don't call drm_sched_stop/start() in asic reset

[Pierre-Eric Pelloux-Prayer <pierre-eric@damsy.net>]

Le 05/02/2026 à 15:26, Alex Deucher a écrit :
> On Thu, Feb 5, 2026 at 9:22 AM Pierre-Eric Pelloux-Prayer
> <pierre-eric@damsy.net> wrote:
>>
>>
>>
>> Le 30/01/2026 à 18:30, Alex Deucher a écrit :
>>> We only want to stop the work queues, not mess with the
>>> fences, etc.
>>>
>>> v2: add the job back to the pending list.
>>> v3: return the proper job status so scheduler adds the
>>>       job back to the pending list
>>>
>>> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>>> ---
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++--
>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_job.c    | 6 ++----
>>>    2 files changed, 4 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> index e69ab8a923e31..a5b43d57c7b05 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> @@ -6313,7 +6313,7 @@ static void amdgpu_device_halt_activities(struct amdgpu_device *adev,
>>>                        if (!amdgpu_ring_sched_ready(ring))
>>>                                continue;
>>>
>>> -                     drm_sched_stop(&ring->sched, job ? &job->base : NULL);
>>> +                     drm_sched_wqueue_stop(&ring->sched);
>>>
>>>                        if (need_emergency_restart)
>>>                                amdgpu_job_stop_all_jobs_on_sched(&ring->sched);
>>> @@ -6397,7 +6397,7 @@ static int amdgpu_device_sched_resume(struct list_head *device_list,
>>>                        if (!amdgpu_ring_sched_ready(ring))
>>>                                continue;
>>>
>>> -                     drm_sched_start(&ring->sched, 0);
>>> +                     drm_sched_wqueue_start(&ring->sched);
>>>                }
>>>
>>>                if (!drm_drv_uses_atomic_modeset(adev_to_drm(tmp_adev)) && !job_signaled)
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
>>> index df06a271bdf99..cd0707737a29b 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
>>> @@ -92,7 +92,6 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
>>>        struct drm_wedge_task_info *info = NULL;
>>>        struct amdgpu_task_info *ti = NULL;
>>>        struct amdgpu_device *adev = ring->adev;
>>> -     enum drm_gpu_sched_stat status = DRM_GPU_SCHED_STAT_RESET;
>>>        int idx, r;
>>>
>>>        if (!drm_dev_enter(adev_to_drm(adev), &idx)) {
>>> @@ -147,8 +146,6 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
>>>                                ring->sched.name);
>>>                        drm_dev_wedged_event(adev_to_drm(adev),
>>>                                             DRM_WEDGE_RECOVERY_NONE, info);
>>> -                     /* This is needed to add the job back to the pending list */
>>> -                     status = DRM_GPU_SCHED_STAT_NO_HANG;
>>>                        goto exit;
>>>                }
>>>                dev_err(adev->dev, "Ring %s reset failed\n", ring->sched.name);
>>> @@ -184,7 +181,8 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
>>>    exit:
>>>        amdgpu_vm_put_task_info(ti);
>>>        drm_dev_exit(idx);
>>> -     return status;
>>> +     /* This is needed to add the job back to the pending list */
>>> +     return DRM_GPU_SCHED_STAT_NO_HANG;
>>
>> This part seems unrelated to the patch and is overwriting what was done
>> in patch 1/12.
> 
> Patch 1 fixes the pending list handling for per queue resets.  This
> patch reworks the adapter reset path to match the behavior of the per
> queue reset path.  After this patch they match so we can safely return
> DRM_GPU_SCHED_STAT_NO_HANG in both cases.  Previously the adapter
> reset path called drm_sched_wqueue_stop()/start() which handles
> re-adding the job to the pending list.  Since it no longer does, we
> need to return DRM_GPU_SCHED_STAT_NO_HANG for both cases.

I looked a bit more in the patchset adding DRM_GPU_SCHED_STAT_NO_HANG 
and your changes make sense. This patch is:

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>