Re: [PATCH 1/1] drm/xe: Don't short circuit TDR on jobs not started

[Matthew Brost <matthew.brost@intel.com>]

On Wed, Oct 23, 2024 at 10:47:05AM -0600, Zanoni, Paulo R wrote:
> On Tue, 2024-10-22 at 16:27 -0700, Matthew Brost wrote:
> > Short circuiting TDR on jobs not started is an optimization which is not
> > required. On LNL we are facing an issue where jobs do not get scheduled
> > by the GuC for an unknown reason. Removing this optimization allows jobs
> > to get scheduled after TDR fire once which is a big improvement. Remove
> > this optimization for now while root causing job scheduling issue on
> > LNL.
> 
> I just tested it and it seems to do what it promises. Thanks! Having a
> 5 second hiccup is still horribly bad, but it is - checks math notes -
> infinitely better than waiting forever for a syncobj that will never be
> signaled.
> 
> This patch will *tremendously* help Mesa CI, since we can reproduce
> this bug all the time with Vulkan CTS tests.
> 
> Suggestions:
> 
> - Can we get a message on dmesg every time this hiccup happens? We're
> not sure if it's happening on real workloads on people's machines, so
> maybe having some sort of indication "oops, we just unstuck the batch
> you submitted 300 frames ago!" would help.
> 

We will add 'notice' level message if this occurs.

> - Since we don't know how long until the real fix, can this be tagged
> for stable? If it turns out this requires special GuC, it would be even
> more valuable to have this in stable since those tend to take more to
> propagate to people's machines.

I don't see any reason why this can't be backported, will include required tags.

Matt

> 
> Thanks a lot!
> 
> > 
> > Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
> > Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_guc_submit.c | 4 ----
> >  1 file changed, 4 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
> > index 0b81972ff651..25ab675e9c7d 100644
> > --- a/drivers/gpu/drm/xe/xe_guc_submit.c
> > +++ b/drivers/gpu/drm/xe/xe_guc_submit.c
> > @@ -1052,10 +1052,6 @@ guc_exec_queue_timedout_job(struct drm_sched_job *drm_job)
> >  		exec_queue_killed_or_banned_or_wedged(q) ||
> >  		exec_queue_destroyed(q);
> >  
> > -	/* Job hasn't started, can't be timed out */
> > -	if (!skip_timeout_check && !xe_sched_job_started(job))
> > -		goto rearm;
> > -
> >  	/*
> >  	 * If devcoredump not captured and GuC capture for the job is not ready
> >  	 * do manual capture first and decide later if we need to use it
>