[Intel-gfx] [PATCH 09/33] drm/i915: Mark unmappable GGTT entries as PIN_HIGH

[Chris Wilson]

On Tue, Aug 09, 2016 at 12:13:32PM +0100, Chris Wilson wrote:
> On Tue, Aug 09, 2016 at 12:05:30PM +0100, Tvrtko Ursulin wrote:
> > 
> > On 07/08/16 15:45, Chris Wilson wrote:
> > >We allocate a few objects into the GGTT that we never need to access via
> > >the mappable aperture (such as contexts, status pages). We can request
> > >that these are bound high in the VM to increase the amount of mappable
> > >aperture available. However, anything that may be frequently pinned
> > >(such as logical contexts) we want to use the fast search & insert.
> > 
> > Is the last bit still true after you merged:
> > 
> >   commit 202b52b7fbf70858609ec20829c7d69a13ffa351
> >   Author: Chris Wilson <chris at chris-wilson.co.uk>
> >   Date:   Wed Aug 3 16:04:09 2016 +0100
> > 
> >       drm: Track drm_mm nodes with an interval tree
> > 
> > Or we could PIN_HIGH the LRCs as well now?
> 
> It is still true. I've another set of patches for drm_mm to speed up the
> actual PIN_HIGH searching, but that is still slower than using the
> eviction stack. It's a tradeoff of causing fragmentation (and maybe
> hitting a slow cleanup path) or faster allocation when creating
> contexts. I think the tradeoff is still towards causing fragmentation.

I should also mention that the DRM_MM_SEARCH_HIGH is broken right now as
it just allocates from the bottom of the eviction stack, not the first
hole available whilst scanning top-down.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre