Re: [PATCH] drm/scheduler: fix inconsistent locking of job_list_lock

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Alex,

since you seem to be picking up scheduler patches, can I ask you to
take this one through your tree?

Regards,
Lucas

On Mo, 2020-01-20 at 11:59 +0100, Christian König wrote:
> Am 20.01.20 um 11:51 schrieb Lucas Stach:
> > 1db8c142b6c5 (drm/scheduler: Add drm_sched_suspend/resume_timeout()) made
> > the job_list_lock IRQ safe in as the suspend/resume calls were expected to
> > be called from IRQ context. This usage never materialized in upstream.
> > Instead amdgpu started locking the job_list_lock in an IRQ unsafe way in
> > amdgpu_ib_preempt_mark_partial_job() and amdgpu_ib_preempt_job_recovery(),
> > which leads to potential deadlock if one would actually start to call the
> > drm_sched_suspend/resume_timeout functions from IRQ context.
> > 
> > As no current user needs the locking to be IRQ safe, the local IRQ
> > disable/enable is pure overhead. Fix the inconsistent locking by changing
> > all uses of job_list_lock to use the IRQ unsafe locking primitives.
> > 
> > Signed-off-by: Lucas Stach <l.stach@xxxxxxxxxxxxxx>
> 
> Reviewed-by: Christian König <christian.koenig@xxxxxxx>, do you want to 
> merge through drm-misc-next or should I push it to amd-staging-drm-next?
> 
> Regards,
> Christian.
> 
> > ---
> >   drivers/gpu/drm/scheduler/sched_main.c | 38 ++++++++++----------------
> >   1 file changed, 15 insertions(+), 23 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
> > index 3c57e84222ca..d1b24444dd99 100644
> > --- a/drivers/gpu/drm/scheduler/sched_main.c
> > +++ b/drivers/gpu/drm/scheduler/sched_main.c
> > @@ -220,8 +220,7 @@ EXPORT_SYMBOL(drm_sched_fault);
> >    *
> >    * Suspend the delayed work timeout for the scheduler. This is done by
> >    * modifying the delayed work timeout to an arbitrary large value,
> > - * MAX_SCHEDULE_TIMEOUT in this case. Note that this function can be
> > - * called from an IRQ context.
> > + * MAX_SCHEDULE_TIMEOUT in this case.
> >    *
> >    * Returns the timeout remaining
> >    *
> > @@ -250,43 +249,39 @@ EXPORT_SYMBOL(drm_sched_suspend_timeout);
> >    * @sched: scheduler instance for which to resume the timeout
> >    * @remaining: remaining timeout
> >    *
> > - * Resume the delayed work timeout for the scheduler. Note that
> > - * this function can be called from an IRQ context.
> > + * Resume the delayed work timeout for the scheduler.
> >    */
> >   void drm_sched_resume_timeout(struct drm_gpu_scheduler *sched,
> >   		unsigned long remaining)
> >   {
> > -	unsigned long flags;
> > -
> > -	spin_lock_irqsave(&sched->job_list_lock, flags);
> > +	spin_lock(&sched->job_list_lock);
> >   
> >   	if (list_empty(&sched->ring_mirror_list))
> >   		cancel_delayed_work(&sched->work_tdr);
> >   	else
> >   		mod_delayed_work(system_wq, &sched->work_tdr, remaining);
> >   
> > -	spin_unlock_irqrestore(&sched->job_list_lock, flags);
> > +	spin_unlock(&sched->job_list_lock);
> >   }
> >   EXPORT_SYMBOL(drm_sched_resume_timeout);
> >   
> >   static void drm_sched_job_begin(struct drm_sched_job *s_job)
> >   {
> >   	struct drm_gpu_scheduler *sched = s_job->sched;
> > -	unsigned long flags;
> >   
> > -	spin_lock_irqsave(&sched->job_list_lock, flags);
> > +	spin_lock(&sched->job_list_lock);
> >   	list_add_tail(&s_job->node, &sched->ring_mirror_list);
> >   	drm_sched_start_timeout(sched);
> > -	spin_unlock_irqrestore(&sched->job_list_lock, flags);
> > +	spin_unlock(&sched->job_list_lock);
> >   }
> >   
> >   static void drm_sched_job_timedout(struct work_struct *work)
> >   {
> >   	struct drm_gpu_scheduler *sched;
> >   	struct drm_sched_job *job;
> > -	unsigned long flags;
> >   
> >   	sched = container_of(work, struct drm_gpu_scheduler, work_tdr.work);
> > +
> >   	job = list_first_entry_or_null(&sched->ring_mirror_list,
> >   				       struct drm_sched_job, node);
> >   
> > @@ -303,9 +298,9 @@ static void drm_sched_job_timedout(struct work_struct *work)
> >   		}
> >   	}
> >   
> > -	spin_lock_irqsave(&sched->job_list_lock, flags);
> > +	spin_lock(&sched->job_list_lock);
> >   	drm_sched_start_timeout(sched);
> > -	spin_unlock_irqrestore(&sched->job_list_lock, flags);
> > +	spin_unlock(&sched->job_list_lock);
> >   }
> >   
> >    /**
> > @@ -368,7 +363,6 @@ EXPORT_SYMBOL(drm_sched_increase_karma);
> >   void drm_sched_stop(struct drm_gpu_scheduler *sched, struct drm_sched_job *bad)
> >   {
> >   	struct drm_sched_job *s_job, *tmp;
> > -	unsigned long flags;
> >   
> >   	kthread_park(sched->thread);
> >   
> > @@ -388,9 +382,9 @@ void drm_sched_stop(struct drm_gpu_scheduler *sched, struct drm_sched_job *bad)
> >   			 * remove job from ring_mirror_list.
> >   			 * Locking here is for concurrent resume timeout
> >   			 */
> > -			spin_lock_irqsave(&sched->job_list_lock, flags);
> > +			spin_lock(&sched->job_list_lock);
> >   			list_del_init(&s_job->node);
> > -			spin_unlock_irqrestore(&sched->job_list_lock, flags);
> > +			spin_unlock(&sched->job_list_lock);
> >   
> >   			/*
> >   			 * Wait for job's HW fence callback to finish using s_job
> > @@ -433,7 +427,6 @@ EXPORT_SYMBOL(drm_sched_stop);
> >   void drm_sched_start(struct drm_gpu_scheduler *sched, bool full_recovery)
> >   {
> >   	struct drm_sched_job *s_job, *tmp;
> > -	unsigned long flags;
> >   	int r;
> >   
> >   	/*
> > @@ -462,9 +455,9 @@ void drm_sched_start(struct drm_gpu_scheduler *sched, bool full_recovery)
> >   	}
> >   
> >   	if (full_recovery) {
> > -		spin_lock_irqsave(&sched->job_list_lock, flags);
> > +		spin_lock(&sched->job_list_lock);
> >   		drm_sched_start_timeout(sched);
> > -		spin_unlock_irqrestore(&sched->job_list_lock, flags);
> > +		spin_unlock(&sched->job_list_lock);
> >   	}
> >   
> >   	kthread_unpark(sched->thread);
> > @@ -648,7 +641,6 @@ static struct drm_sched_job *
> >   drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
> >   {
> >   	struct drm_sched_job *job;
> > -	unsigned long flags;
> >   
> >   	/*
> >   	 * Don't destroy jobs while the timeout worker is running  OR thread
> > @@ -659,7 +651,7 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
> >   	    __kthread_should_park(sched->thread))
> >   		return NULL;
> >   
> > -	spin_lock_irqsave(&sched->job_list_lock, flags);
> > +	spin_lock(&sched->job_list_lock);
> >   
> >   	job = list_first_entry_or_null(&sched->ring_mirror_list,
> >   				       struct drm_sched_job, node);
> > @@ -673,7 +665,7 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched)
> >   		drm_sched_start_timeout(sched);
> >   	}
> >   
> > -	spin_unlock_irqrestore(&sched->job_list_lock, flags);
> > +	spin_unlock(&sched->job_list_lock);
> >   
> >   	return job;
> >   }

_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux