[Intel-gfx] [PATCH 2/6] drm/i915: Tidy batch pool logic

[Chris Wilson]

On Tue, Mar 17, 2015 at 05:42:12PM +0000, Tvrtko Ursulin wrote:
> Lets see if I got it right: By making the returned object pinned and
> having the caller unpin when done with it, it now can always be in
> "dontneed" state, and will be a shrinker candidate only (or as soon)
> when it is unpinned, correct?
> 
> I just don't get at the moment, if previously there was some race or
> bad interaction with less pinning and that signaling using
> _EXEC_OBJECT_PURGEABLE?

The code was using overzealous pinning (too much too early), did too
much work manipulating tricky state and introduced an unneeded flag into
a user API field. It is very clumsy.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre