[Intel-gfx] [PATCH 1/2] drm/i915: Only shrink the unbound objects during freeze

[Joonas Lahtinen]

On ma, 2016-09-19 at 09:38 +0100, Chris Wilson wrote:
> On Mon, Sep 19, 2016 at 11:31:37AM +0300, Joonas Lahtinen wrote:
> > 
> > On pe, 2016-09-16 at 20:23 +0100, Chris Wilson wrote:
> > > 
> > >  int i915_gem_freeze_late(struct drm_i915_private *dev_priv)
> > >  {
> > > > 
> > > > > > > >  	struct drm_i915_gem_object *obj;
> > > @@ -4692,7 +4705,8 @@ int i915_gem_freeze_late(struct drm_i915_private *dev_priv)
> > > > 
> > > > > > > >  	 * the objects as well.
> > > > > > > >  	 */
> > >  
> > > > 
> > > > > > > > -	i915_gem_shrink_all(dev_priv);
> > > > > > +	mutex_lock(&dev_priv->drm.struct_mutex);
> > 
> > Previously the shrinking here was without a mutex. And the new
> > shrinking during _freeze is also mutex protected, any specific reason?
> 
> Silencing lockdep. We know we are single threaded here, but lockdep
> doesn't. This version of i915_gem_shrink() doesn't automagically take
> the struct_mutex (later versions do).

Throw the info to commit message. Then all good.

Regards, Joonas

> -Chris
> 
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation