[PATCH] drm/scheduler: Fix sched hang when killing app with dependent jobs

[Philipp Stanner]

On Thu, 2025-07-10 at 08:33 +0000, cao, lin wrote:
> 
> [AMD Official Use Only - AMD Internal Distribution Only]
> 
> 
> 
> Hi Christian,
> 
> 
> Thanks for your suggestion, I modified the patch as:

Looks promising. You'll send a v2 I guess :)

P.

> 
> 
> diff --git a/drivers/gpu/drm/scheduler/sched_entity.c
> b/drivers/gpu/drm/scheduler/sched_entity.c
> index e671aa241720..66f2a43c58fd 100644
> --- a/drivers/gpu/drm/scheduler/sched_entity.c
> +++ b/drivers/gpu/drm/scheduler/sched_entity.c
> @@ -177,6 +177,7 @@ static void
> drm_sched_entity_kill_jobs_work(struct work_struct *wrk)
>     struct drm_sched_job *job = container_of(wrk, typeof(*job),
> work);
>  
>     drm_sched_fence_scheduled(job->s_fence, NULL);
> +   drm_sched_wakeup(job->sched);
>     drm_sched_fence_finished(job->s_fence, -ESRCH);
>     WARN_ON(job->s_fence->parent);
>     job->sched->ops->free_job(job);
> --
> 
> 
> Thanks,
> Lin
> 
> 
> 
> 
> 
> From: Koenig, Christian <Christian.Koenig at amd.com>
> Sent: Thursday, July 10, 2025 15:52
> To: cao, lin <lin.cao at amd.com>; dri-devel at lists.freedesktop.org
> <dri-devel at lists.freedesktop.org>; amd-gfx at lists.freedesktop.org
> <amd-gfx at lists.freedesktop.org>
> Cc: Yin, ZhenGuo (Chris) <ZhenGuo.Yin at amd.com>; Deng, Emily
> <Emily.Deng at amd.com>; Matthew Brost <matthew.brost at intel.com>; Danilo
> Krummrich <dakr at kernel.org>; Philipp Stanner <phasta at kernel.org>
> Subject: Re: [PATCH] drm/scheduler: Fix sched hang when killing app
> with dependent jobs
> 
> 
> First of all you need to CC the scheduler maintainers, try to use the
> get_maintainer.pl script. Adding them on CC.
> 
> On 10.07.25 08:36, Lin.Cao wrote:
> > When Application A submits jobs (a1, a2, a3) and application B
> > submits
> > job b1 with a dependency on a2's scheduler fence, killing
> > application A
> > before run_job(a1) causes drm_sched_entity_kill_jobs_work() to
> > force
> > signal all jobs sequentially. However, due to missing work_run_job
> > or
> > work_free_job in entity_kill_job_work(), the scheduler enters sleep
> > state, causing application B hang.
> 
> Ah! Because of optimizing the dependency when submitting to the same
> scheduler in drm_sched_entity_add_dependency_cb().
> 
> Yeah that suddenly starts to make sense.
> 
> > Add drm_sched_wakeup() when entity_kill_job_work() to preventing
> > scheduler sleep and subsequent application hangs.
> > 
> > Signed-off-by: Lin.Cao <lincao12 at amd.com>
> > ---
> >   drivers/gpu/drm/scheduler/sched_entity.c | 1 +
> >   1 file changed, 1 insertion(+)
> > 
> > diff --git a/drivers/gpu/drm/scheduler/sched_entity.c
> > b/drivers/gpu/drm/scheduler/sched_entity.c
> > index e671aa241720..a22b0f65558a 100644
> > --- a/drivers/gpu/drm/scheduler/sched_entity.c
> > +++ b/drivers/gpu/drm/scheduler/sched_entity.c
> > @@ -180,6 +180,7 @@ static void
> > drm_sched_entity_kill_jobs_work(struct work_struct *wrk)
> >         drm_sched_fence_finished(job->s_fence, -ESRCH);
> >         WARN_ON(job->s_fence->parent);
> >         job->sched->ops->free_job(job);
> > +     drm_sched_wakeup(job->sched);
> 
> That should probably be after drm_sched_fence_scheduled().
> 
> Alternatively we could also drop the optimization in
> drm_sched_entity_add_dependency_cb(), scheduling the work item again
> has only minimal overhead.
> 
> Apart from that looks good to me.
> 
> Regards,
> Christian.
> 
> >   }
> >  
> >   /* Signal the scheduler finished fence when the entity in
> > question is killed. */
>