[Intel-gfx] [PATCH 01/10] drm/i915: Mark up protected uses of 'i915_request_completed'

[Chris Wilson]

Quoting Andi Shyti (2021-01-14 03:01:15)
> Hi Chris,
> 
> > diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> > index 0b1a46a0d866..784c05ac5cca 100644
> > --- a/drivers/gpu/drm/i915/i915_request.c
> > +++ b/drivers/gpu/drm/i915/i915_request.c
> > @@ -276,7 +276,7 @@ static void remove_from_engine(struct i915_request *rq)
> >  
> >  bool i915_request_retire(struct i915_request *rq)
> >  {
> > -     if (!i915_request_completed(rq))
> > +     if (!__i915_request_is_complete(rq))
> 
> 
> >               return false;
> >  
> >       RQ_TRACE(rq, "\n");
> > @@ -342,8 +342,7 @@ void i915_request_retire_upto(struct i915_request *rq)
> >       struct i915_request *tmp;
> >  
> >       RQ_TRACE(rq, "\n");
> > -
> > -     GEM_BUG_ON(!i915_request_completed(rq));
> > +     GEM_BUG_ON(!__i915_request_is_complete(rq));
> 
> I might be a bit shallow, but where is the lock here?

This holds the request->context->timeline->mutex.

The locking is basically split between frontend/backend:

  construction/destruction - under the timeline->mutex

  execution - under the engine->active.lock (coordinates with
              timeline->mutex)

At all other times, RCU protected access (using the SLAB_TYPESAFE_BY_RCU
so approach with caution).
-Chris