Re: [PATCH 08/17] drm/i915: Always prefer CPU relocations with LLC

[Ben Widawsky <ben@bwidawsk.net>]

On Thu, Aug 29, 2013 at 09:11:16PM +0200, Daniel Vetter wrote:
> On Thu, Aug 29, 2013 at 10:20:06AM -0700, Ben Widawsky wrote:
> > On Mon, Aug 26, 2013 at 07:51:00PM -0300, Rodrigo Vivi wrote:
> > > From: Chris Wilson <chris@chris-wilson.co.uk>
> > > 
> > > A follow-on to the update of the LLC coherency logic is that we can rely
> > > on the LLC being coherent with the CS for rewriting batchbuffers
> > > irrespective of their cache domain. (This should have no effect
> > > currently as all the batch buffers are expected to be I915_CACHE_LLC and
> > > so using the cpu relocation path anyway.)
> > > 
> > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > > ---
> > >  drivers/gpu/drm/i915/i915_gem_execbuffer.c | 5 +++--
> > >  1 file changed, 3 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> > > index 792c52a..3b64b9f 100644
> > > --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> > > +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> > > @@ -166,7 +166,8 @@ eb_destroy(struct eb_objects *eb)
> > >  
> > >  static inline int use_cpu_reloc(struct drm_i915_gem_object *obj)
> > >  {
> > > -	return (obj->base.write_domain == I915_GEM_DOMAIN_CPU ||
> > > +	return (HAS_LLC(obj->base.dev) ||
> > > +		obj->base.write_domain == I915_GEM_DOMAIN_CPU ||
> > >  		!obj->map_and_fenceable ||
> > >  		obj->cache_level != I915_CACHE_NONE);
> > 
> > Assuming the commit message is factually correct... the obj->cache_level
> > shouldn't factor into the equation at all.
> 
> We stil need to take the cache level into account on non-llc machines ...
> -Daniel

I always forget I915_CACHE_LLC is overloaded to mean snoopable.

-- 
Ben Widawsky, Intel Open Source Technology Center