[PATCH 1/3] drm/radeon: GPU virtual memory support v22

[Daniel Vetter]

On Mon, Jan 09, 2012 at 11:56:20AM -0500, Alex Deucher wrote:
> On Mon, Jan 9, 2012 at 11:49 AM, Daniel Vetter <daniel at ffwll.ch> wrote:
> > On Mon, Jan 09, 2012 at 11:07:44AM -0500, Alex Deucher wrote:
> >> We would have liked to as well, but with the holidays and internal
> >> review process and people being on vacation, and the kernel merge
> >> window timing it didn't work out this time.  Still as Jerome said, our
> >> VM setup is tightly coupled with the way our HW works.  I'm not sure
> >> how much code could actually be shared.
> >
> > Again to clarify, I'm not so much interesting in sharing code here - I
> > know it'll be nigh to impossible by having looked through nouveau already
> > and quickly glossed through your stuff. I see the value is just in having
> > a minimal set of common idioms that we can use to build cool stuff like
> > resource management.
> >
> > Think of gem, which is actually about next to nothing. There's a the flink
> > stuff and a common close ioctl and recently also a dummy create unusable
> > for anythine else than boot spalsh screens. Plus the shmem backing store
> > which seems to be not so much in favour with embedded chips, i.e doesn't
> > really count. Ridiculous compared to all the other stuff required to draw
> > triangles on a gpu, but good enough to make things like dri2 and wayland
> > possible.
> >
> 
> I generally agree here, but I don't think this patch prevents
> something like that in the future.  Also, I think we need to actually
> use GPU VM for a bit and see how everyone's implementations and plans
> shake out before we can decide which idioms are common.

Agreed. My sligh miff has only really ben that this got posted only
recently and there was no chance to have this quick discussion before it
got merged.

Cheers, Daniel
-- 
Daniel Vetter
Mail: daniel at ffwll.ch
Mobile: +41 (0)79 365 57 48