[Intel-gfx] [PATCH 2/2] drm/i915: Discard the unbound list when suspending

[Daniel Vetter]

On Wed, Feb 06, 2013 at 09:31:21AM +0000, Chris Wilson wrote:
> The unbound tracking of objects is an optimisation to avoid costly
> domain transfers whilst the system is rendering. Across suspend, the
> objects will be involuntarily moved out of the GTT domain and will
> require fixup upon resume. Rather than perform those clflushes for all
> objects immediately upon resume, just move them out of the unbound
> tracking list on suspend and lazily reload them from the CPU domain as
> and when they are used afterwards.
> 
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>

Similar comment about fast resume performance: I think we should only do
these when freezing for hibernation. Otherwise browser performance with
cool new stuff like the fast S1x suspend (or whatever it's called again)
will be a horrible experience. Every time you just read for a few
seconds and the system suspend in the background we'll drop all gem bo's
on the floor ...
-Daniel

> ---
>  drivers/gpu/drm/i915/i915_gem.c |   24 ++++++++++++++++++++++++
>  1 file changed, 24 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index f6ef53f..db14c73 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -3912,6 +3912,17 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
>  	i915_gem_object_free(obj);
>  }
>  
> +static void
> +i915_gem_discard_unbound(struct drm_i915_private *dev_priv)
> +{
> +	struct drm_i915_gem_object *obj, *next;
> +
> +	list_for_each_entry_safe(obj, next,
> +				 &dev_priv->mm.unbound_list,
> +				 gtt_list)
> +		(void)i915_gem_object_put_pages(obj);
> +}
> +
>  int
>  i915_gem_idle(struct drm_device *dev)
>  {
> @@ -3936,6 +3947,19 @@ i915_gem_idle(struct drm_device *dev)
>  	if (!drm_core_check_feature(dev, DRIVER_MODESET))
>  		i915_gem_evict_everything(dev);
>  
> +	/* For KMS we just want to discard the unbound list as it will
> +	 * change domains when thawing - and restoring its domain upon
> +	 * resume seems like a false optimisation. Also note that discarding
> +	 * the unbound list will have the useful side-effect of dropping
> +	 * purgeable objects which it seems pointless to restore as they are
> +	 * merely a userspace cache.
> +	 * However, we may still have some pinned objects (like dma-buf)
> +	 * that cannot simply be discarded and will require domain
> +	 * restoration upon resume.
> +	 */
> +	if (drm_core_check_feature(dev, DRIVER_MODESET))
> +		i915_gem_discard_unbound(dev_priv);
> +
>  	i915_gem_reset_fences(dev);
>  
>  	/* Hack!  Don't let anybody do execbuf while we don't control the chip.
> -- 
> 1.7.10.4
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch