[PATCH] panfrost: Fix job timeout handling

[Boris Brezillon]

Oops, the prefix should be "drm/panfrost", will fix that in v2.

On Thu,  1 Oct 2020 16:01:43 +0200
Boris Brezillon <boris.brezillon at collabora.com> wrote:

> If more than two or more jobs end up timeout-ing concurrently, only one
> of them (the one attached to the scheduler acquiring the lock) is fully
> handled. The other one remains in a dangling state where it's no longer
> part of the scheduling queue, but still blocks something in scheduler
> thus leading to repetitive timeouts when new jobs are queued.
> 
> Let's make sure all bad jobs are properly handled by the thread acquiring
> the lock.
> 
> Signed-off-by: Boris Brezillon <boris.brezillon at collabora.com>
> Fixes: f3ba91228e8e ("drm/panfrost: Add initial panfrost driver")
> Cc: <stable at vger.kernel.org>
> ---
>  drivers/gpu/drm/panfrost/panfrost_job.c | 18 ++++++++++++++----
>  1 file changed, 14 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
> index 30e7b7196dab..e87edca51d84 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_job.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_job.c
> @@ -25,7 +25,7 @@
>  
>  struct panfrost_queue_state {
>  	struct drm_gpu_scheduler sched;
> -
> +	struct drm_sched_job *bad;
>  	u64 fence_context;
>  	u64 emit_seqno;
>  };
> @@ -392,19 +392,29 @@ static void panfrost_job_timedout(struct drm_sched_job *sched_job)
>  		job_read(pfdev, JS_TAIL_LO(js)),
>  		sched_job);
>  
> +	/*
> +	 * Collect the bad job here so it can be processed by the thread
> +	 * acquiring the reset lock.
> +	 */
> +	pfdev->js->queue[js].bad = sched_job;
> +
>  	if (!mutex_trylock(&pfdev->reset_lock))
>  		return;
>  
>  	for (i = 0; i < NUM_JOB_SLOTS; i++) {
>  		struct drm_gpu_scheduler *sched = &pfdev->js->queue[i].sched;
>  
> -		drm_sched_stop(sched, sched_job);
>  		if (js != i)
>  			/* Ensure any timeouts on other slots have finished */
>  			cancel_delayed_work_sync(&sched->work_tdr);
> -	}
>  
> -	drm_sched_increase_karma(sched_job);
> +		drm_sched_stop(sched, pfdev->js->queue[i].bad);
> +
> +		if (pfdev->js->queue[i].bad)
> +			drm_sched_increase_karma(pfdev->js->queue[i].bad);
> +
> +		pfdev->js->queue[i].bad = NULL;
> +	}
>  
>  	spin_lock_irqsave(&pfdev->js->job_lock, flags);
>  	for (i = 0; i < NUM_JOB_SLOTS; i++) {