[Intel-gfx] [PATCH 01/14] drm/i915: Keep a global seqno per-engine

[Chris Wilson]

On Tue, Feb 14, 2017 at 01:13:58PM +0200, Joonas Lahtinen wrote:
> On to, 2017-02-02 at 15:12 +0000, Chris Wilson wrote:
> > -static int reserve_global_seqno(struct drm_i915_private *i915)
> > +static int reserve_global_seqno(struct intel_engine_cs *engine)
> >  {
> > -	u32 active_requests = ++i915->gt.active_requests;
> > -	u32 seqno = atomic_read(&i915->gt.global_timeline.seqno);
> > +	u32 active = ++engine->timeline->active_seqno;
> > +	u32 seqno = engine->timeline->seqno;
> >  	int ret;
> >  
> >  	/* Reservation is fine until we need to wrap around */
> > -	if (likely(seqno + active_requests > seqno))
> > +	if (likely(seqno + active > seqno))
> >  		return 0;
> > 
> 
> Make a comment here that we don't currently track the semaphores
> waiting on specific engine, so we instead choose to wrap all of them.
>  
> > -	ret = i915_gem_init_global_seqno(i915, 0);
> > +	ret = i915_gem_init_global_seqno(engine->i915, 0);
> 
> Also, we don't have a global seqno anymore, so the function name needs
> updating. Maybe "init_{all_}global_seqnos" to make it clear.

However, i915_gem_init_global_seqno() is acting globally. It's debug code
but I think the implication is important to keep. reserve_global_seqno()
is now local and that feels appropriate to rename.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre