[Intel-gfx] [PATCH] drm/i915: disable set/get_tiling ioctl on gen12+

[Daniel Vetter]

On Fri, Nov 8, 2019 at 12:07 AM Brian Welty <brian.welty at intel.com> wrote:
>
>
>
> On 8/28/2019 11:50 PM, Daniel Vetter wrote:
> > On Wed, Aug 28, 2019 at 08:31:27PM +0000, Souza, Jose wrote:
> >> On Wed, 2019-08-28 at 21:13 +0100, Chris Wilson wrote:
> >>> Quoting Souza, Jose (2019-08-28 21:11:53)
> >>>> Reviewed-by: José Roberto de Souza <jose.souza at intel.com>
> >>>
> >>> It's using a non-standard for i915 error code, and get_tiling is not
> >>
> >> Huum should it use ENOTSUPP then?!
> >
> > https://dri.freedesktop.org/docs/drm/gpu/drm-uapi.html#recommended-ioctl-return-values
> >
> > Per above "feature not supported" -> EOPNOTSUPP.
> >
> >>> affected, it will always return LINEAR. You cannot set tiling as
> >>
> >> Following this set_tiling() LINEAR should be allowed too.
> >> I prefer the current approach of returning error.
> >
> > I'm not seeing the value in keeping get_tiling supported. Either userspace
> > still uses the legacy backhannel and dri2, in which case it needs to be
> > fixed no matter what. Or it's using modifiers, in which case this is dead
> > code. Only other user I can think of is takeover for fastboot, but if you
> > get anything else than untiled it's also broken (we don't have an ioctl to
> > read out the modifiers, heck even all the planes, there's no getfb2).
> >
> > So really not seeing the point in keeping that working.
>
> Daniel,  I came across usage of GET_TILING in libdrm.
> Is used in drm_intel_bo_gem_create_from_name() and drm_intel_bo_gem_create_from_prime().
> Should these be updated so they don't fail when EOPNOTSUPP is returned on gen12+?
> Maybe libdrm should just set tiling_mode to 0 on EOPNOTSUPP error instead of those calls failing?

Yes, I guess that part of mesa hasn't seen much testing yet.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch