[Intel-gfx] [PATCH 24/66] drm/i915: Move aliasing_ppgtt

[Daniel Vetter]

On Mon, Jul 1, 2013 at 9:48 PM, Ben Widawsky <ben at bwidawsk.net> wrote:
>> Full ppgtt otoh need the address space of course.
>>
>> The above structure layout would allow that. It means that aliasing
>> ppgtt will stick out a bit like a sore thumb but imo that's ok since
>> it really _is_ a bit an odd thing. I've just noticed that you've
>> partially disabled the has_aliasing_ppgtt_mapping logic in your
>> branch, so I wonder a bit how aliasing ppgtt works there.
>
> It doesn't. The intent was to completely eclipse aliasing PPGTT. If we
> keep the pin interface (as discussed in the other thread), I am not
> certain that is still true, though I personally have no issue forcing a
> performance penalty on users that wish to use that interface.

Ah, that explains my confusion ;-)

I guess we simply have to be careful to keep this working properly
when merging the patches. Like I've said I'm pretty sure that we can
keep all the major concepts unchanged and just bolt aliasing ppgtt
onto the side like it's done today.

For i915_ppgtt vs. i915_hw_ppgtt I think we can revisite that we the
in-depth review of the relevant patches is on the table.
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch