[Intel-gfx] [PATCH] drm/i915: Prevent use of global_seqno=0

[Chris Wilson]

Quoting Mika Kuoppala (2019-01-21 09:00:50)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > We are not allowed to assign rq->global_seqno=0 as it has a special
> > meaning of "inactive" (not executing on HW).
> >
> > Fixes: 6faf5916e6be ("drm/i915: Remove HW semaphores for gen7 inter-engine synchronisation")
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> > ---
> >  drivers/gpu/drm/i915/i915_request.c | 9 ++++++++-
> >  1 file changed, 8 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> > index 5403d4e2cee0..5e178f5ac18b 100644
> > --- a/drivers/gpu/drm/i915/i915_request.c
> > +++ b/drivers/gpu/drm/i915/i915_request.c
> > @@ -343,6 +343,13 @@ static void move_to_timeline(struct i915_request *request,
> >       spin_unlock(&request->timeline->lock);
> >  }
> >  
> > +static u32 next_global_seqno(struct i915_timeline *tl)
> > +{
> > +     if (!++tl->seqno)
> > +             ++tl->seqno;
> > +     return tl->seqno;
> > +}
> > +
> >  void __i915_request_submit(struct i915_request *request)
> >  {
> >       struct intel_engine_cs *engine = request->engine;
> > @@ -359,7 +366,7 @@ void __i915_request_submit(struct i915_request *request)
> >  
> >       GEM_BUG_ON(request->global_seqno);
> >  
> > -     seqno = timeline_get_seqno(&engine->timeline);
> > +     seqno = next_global_seqno(&engine->timeline);
> 
> Does it matter that we will allow dma fence to be
> with seqno zero?

Nope, that's just a plain old u32. (u64 if you believe some!)
 
> In other words, if we keep the global seqnos and timeline
> seqnos 'type similar' for readability reasons, should
> we enforce the 'not zero' in here too.

Nah, global_seqno is a temporary blip. This is to fix a recent regression
that will be removed in its entirely at the end of the series.
 
> Dma fence code seems to handle 32bits seqnos
> even tho the type is 64bit wide.

Right, 64b is a new extension unsuitable for our HW.
-Chris