[RFC PATCH 0/3] Propose new struct drm_mem_region

[Daniel Vetter]

On Tue, Jul 30, 2019 at 4:30 PM Michel Dänzer <michel at daenzer.net> wrote:
> On 2019-07-30 12:45 p.m., Daniel Vetter wrote:
> > On Tue, Jul 30, 2019 at 12:24 PM Koenig, Christian
> > <Christian.Koenig at amd.com> wrote:
> >> Am 30.07.19 um 11:38 schrieb Daniel Vetter:
> >>> On Tue, Jul 30, 2019 at 08:45:57AM +0000, Koenig, Christian wrote:
> >
> >>>>> +#define DRM_MEM_SYSTEM     0
> >>>>> +#define DRM_MEM_STOLEN     1
> >>>> I think we need a better naming for that.
> >>>>
> >>>> STOLEN sounds way to much like stolen VRAM for integrated GPUs, but at
> >>>> least for TTM this is the system memory currently GPU accessible.
> >>> Yup this is wrong, for i915 we use this as stolen, for ttm it's the gpu
> >>> translation table window into system memory. Not the same thing at all.
> >>
> >> Thought so. The closest I have in mind is GTT, but everything else works
> >> as well.
> >
> > Would your GPU_MAPPED above work for TT? I think we'll also need
> > STOLEN, I'm even hearing noises that there's going to be stolen for
> > discrete vram for us ...
>
> Could i915 use DRM_MEM_PRIV for stolen? Or is there other hardware with
> something similar?

I don't think it matters much what we name it ... _PRIV sounds as good
as anything else. As long as we make it clear that userspace bo also
might end up in there I think it's all good.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch