On Mon, 2025-09-01 at 15:14 +0200, Pierre-Eric Pelloux-Prayer wrote:
> 
> 
> Le 25/08/2025 à 15:13, Philipp Stanner a écrit :
> > On Fri, 2025-08-22 at 15:43 +0200, Pierre-Eric Pelloux-Prayer wrote:
> > > Currently, the scheduler score is incremented when a job is pushed to an
> > > entity and when an entity is attached to the scheduler.
> > 
> > It's indeed awkward why attaching is treated equivalently to job
> > submission.
> > 
> > Can you expand the documentation for drm_sched_init_args a bit so that
> > it gets clearer what the score is supposed to do?
> 
> 
> drm_sched_init_args.score is the feature allowing multiple schedulers to 
> share a 
> score so I suppose you meant drm_gpu_scheduler.score?

I mean both.

struct drm_sched_init_args has a @score which is passed by the driver
during init and will then be stored in drm_gpu_scheduler.score

The docu should be improved for both.

> 
> The doc currently says "score to help loadbalancer pick a idle sched" which 
> is a 
> bit vague. It could be modified to become:
> 
>    @score: holds the number of yet-to-be-completed jobs pushed to each 
> scheduler.
>            It's used when load balancing between different schedulers.

"to each scheduler" reads a bit as if each scheduler has an individual
score. Is it used that way, though? I think it's a pointer because all
schedulers share this atomic. Thus, it "holds the total number of yet-
to-be-completed jobs for all schedulers sharing this atomic", right?

> 
> What do you think?
> 
> > 
> > > 
> > > This leads to some bad scheduling decision where the score value is
> > > largely made of idle entities.
> > > 
> > > For instance, a scenario with 2 schedulers and where 10 entities submit
> > > a single job, then do nothing, each scheduler will probably end up with
> > > a score of 5.
> > > Now, 5 userspace apps exit, so their entities will be dropped.
> > > 
> > 
> > "entities will be dropped" == "drm_sched_entity_kill() gets called",
> > right?
> 
> Yes.

then nit: better say "their entities will be killed" instead of
dropped.

> 
> > > In
> > > the worst case, these apps' entities where all attached to the same
> > 
> > s/where/were
> > 
> > or better yet: "could be"
> 
> Will fix, thanks.
> 
> > 
> > > scheduler and we end up with score=5 (the 5 remaining entities) and
> > > score=0, despite the 2 schedulers being idle.
> > 
> > Sounds indeed like a (small) problem to me.
> > 
> > 
> > > When new entities show up, they will all select the second scheduler
> > > based on its low score value, instead of alternating between the 2.
> > > 
> > > Some amdgpu rings depended on this feature, but the previous commit
> > > implemented the same thing in amdgpu directly so it can be safely
> > > removed from drm/sched.
> > 
> > Can we be that sure that other drivers don't depend on it, though? I
> > suspect it's likely that it's just amdgpu, but…
> > 
> 
> Aside from the new "rocket" as pointed out by Tvrtko, amdgpu is the only 
> driver 
> passing more than one schedulers to entities so they're the only ones that 
> could 
> be affected.
> 
> I verified amdgpu and Tvrtko pinged the rocket maintainers in the other 
> thread.

Very good! Then let's give those guys a few days to jump into the
discussion.

> 
> > 
> > 
> > BTW, since you're cleaning up related stuff currently: I saw that it
> > seems that the only driver that sets &struct drm_sched_init_args.score
> > is amdgpu. Would be cool if you can take a look whether that's still
> > needed.
> 
> It cannot really be removed yet as it's useful when a single hardware block 
> is 
> exposed through different schedulers (so pushing jobs to one of the 
> schedulers 
> should increase the load of the underlying hw).

OK.


Thx
P.

> 
> Thanks,
> Pierre-Eric
> 
> > 
> > 
> > P.
> > 
> > > 
> > > Signed-off-by: Pierre-Eric Pelloux-Prayer 
> > > <[email protected]>
> > > ---
> > >   drivers/gpu/drm/scheduler/sched_main.c | 2 --
> > >   1 file changed, 2 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/scheduler/sched_main.c 
> > > b/drivers/gpu/drm/scheduler/sched_main.c
> > > index 5a550fd76bf0..e6d232a8ec58 100644
> > > --- a/drivers/gpu/drm/scheduler/sched_main.c
> > > +++ b/drivers/gpu/drm/scheduler/sched_main.c
> > > @@ -206,7 +206,6 @@ void drm_sched_rq_add_entity(struct drm_sched_rq *rq,
> > >           if (!list_empty(&entity->list))
> > >                   return;
> > >   
> > > - atomic_inc(rq->sched->score);
> > >           list_add_tail(&entity->list, &rq->entities);
> > >   }
> > >   
> > > @@ -228,7 +227,6 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq 
> > > *rq,
> > >   
> > >           spin_lock(&rq->lock);
> > >   
> > > - atomic_dec(rq->sched->score);
> > >           list_del_init(&entity->list);
> > >   
> > >           if (rq->current_entity == entity)

Reply via email to