[Intel-gfx] [PATCH 01/13] drm/i915: Flush the context object from the CPU caches upon creation

[Daniel Vetter]

On Sat, Jul 14, 2012 at 2:48 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> On Sat, 14 Jul 2012 13:58:58 +0200, Daniel Vetter <daniel at ffwll.ch> wrote:
>> So afact this first patch here seems to add another instance of the very
>> bug this patch series tries squash ... Additionally I'm still hunting for
>> that other failure case, which can't be fixed by adding the flush in
>> execbuffer if ring->gpu_caches_dirty is set.
>>
>> /me is still lost
>
> Now all that you are missing is that we only flush GPU_DOMAINS when on
> the gpu_write_list.

Hm, I guess we've seen that one before ;-) But that would require that
we add an object with a non-gpu write domain to the active list, but
both with or without this patch I don't see how that should happen ...

I guess a WARN in move_to_active won't hurt (and we should have done
that when fixing the GTT_DOMAIN relocation hole in execbuf), but
besides that I'm still as dense as ever, it seems.
-Daniel
-- 
Daniel Vetter
daniel.vetter at ffwll.ch - +41 (0) 79 364 57 48 - http://blog.ffwll.ch