[Intel-gfx] [PATCH] drm/i915: Stop discarding GTT cache-domain on unbind vma

[Daniel Vetter]

On Tue, Oct 06, 2015 at 01:46:25PM +0100, Chris Wilson wrote:
> On Tue, Oct 06, 2015 at 02:40:26PM +0200, Daniel Vetter wrote:
> > On Tue, Oct 06, 2015 at 01:02:22PM +0100, Chris Wilson wrote:
> > > Since
> > > 
> > > commit 43566dedde54f9729113f5f9fde77d53e75e61e9
> > > Author: Chris Wilson <chris at chris-wilson.co.uk>
> > > Date:   Fri Jan 2 16:29:29 2015 +0530
> > > 
> > >     drm/i915: Broaden application of set-domain(GTT)
> > > 
> > > we allowed objects to be in the GTT domain, but unbound. Therefore
> > > removing the GTT cache domain when removing the GGTT vma is no longer
> > > semantically correct.
> > > 
> > > An unfortunate side-effect is we lose the wondrously named
> > > i915_gem_object_finish_gtt(), not to be confused with
> > > i915_gem_gtt_finish_object()!
> > > 
> > > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > > Cc: Akash Goel <akash.goel at intel.com>
> > > Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> > > Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/i915_gem.c | 20 +++-----------------
> > >  1 file changed, 3 insertions(+), 17 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> > > index 8f498d4d874d..682af2ae3681 100644
> > > --- a/drivers/gpu/drm/i915/i915_gem.c
> > > +++ b/drivers/gpu/drm/i915/i915_gem.c
> > > @@ -3183,20 +3183,6 @@ i915_gem_object_sync(struct drm_i915_gem_object *obj,
> > >  	return 0;
> > >  }
> > >  
> > > -static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj)
> > > -{
> > > -	/* Force a pagefault for domain tracking on next user access */
> > > -	i915_gem_release_mmap(obj);
> > > -
> > > -	if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0)
> > > -		return;
> > > -
> > > -	obj->base.read_domains &= ~I915_GEM_DOMAIN_GTT;
> > > -	obj->base.write_domain &= ~I915_GEM_DOMAIN_GTT;
> > > -
> > > -	trace_i915_gem_object_change_domain(obj);
> > > -}
> > > -
> > >  int i915_vma_unbind(struct i915_vma *vma)
> > >  {
> > >  	struct drm_i915_gem_object *obj = vma->obj;
> > > @@ -3228,12 +3214,12 @@ int i915_vma_unbind(struct i915_vma *vma)
> > >  	 */
> > >  
> > >  	if (vma->is_ggtt && vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL) {
> > > -		i915_gem_object_finish_gtt(obj);
> > > -
> > > -		/* release the fence reg _after_ flushing */
> > >  		ret = i915_gem_object_put_fence(obj);
> > >  		if (ret)
> > >  			return ret;
> > > +
> > > +		/* Force a pagefault for domain tracking on next user access */
> > > +		i915_gem_release_mmap(obj);
> > 
> > Can't put_fence before release_mmap ... I guess we should have a testcase
> > for this somewhere? Hard to provoke probably ...
> 
> Why not? i915_gem_object_put_fence() has to release the mmaps itself if
> it has a fence register assigned for the object.

Oh right, forgotten that put_fence is robust. Looking at this simply
brought up bad memories since I fixed this kind of bug 5 years ago once
;-)

Cheers, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch