[PATCH 1/3] Revert "drm/amdgpu: change the reference clock for raven/raven2"

[Alex Deucher]

On Mon, Jun 26, 2023 at 9:58 AM Christian König
<ckoenig.leichtzumerken at gmail.com> wrote:
>
> Hi guys,
>
> Vitaly and Jasber have been recently working on disabling the IGT tests
> for the TSC query on RV/RV2 (which I'm not very keen on).
>
> And additional to that we have this RADV merge request here:
> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23481
>
> What exactly is going on here? That sounds like we are working around a
> FW or more specific GFXOFF bug in userspace.

The clock source used by the gfx firmware on Raven is part of the gfx
domain so it gets powered off when gfxoff turns off the graphics
block.  There is a clock source in an always on domain, but the gfx
block doesn't use it.  I don't know off hand if the clock source used
by gfx can be changed or not on raven (IIRC, I don't think it can,
which is why this was never fixed on RV).  Since they are different
clock sources, the GPU timestamp doesn't match the GPU timestamp.  On
renoir and newer APUs, gfx uses the always on clock source so it's
always consistent between CPU and GPU.

Alex

>
> Regards,
> Christian.
>
> Am 05.06.23 um 10:57 schrieb Michel Dänzer:
> > On 6/2/23 20:43, Alex Deucher wrote:
> >> This reverts commit fbc24293ca16b3b9ef891fe32ccd04735a6f8dc1.
> >>
> >> This results in inconsistent timing reported via asynchronous
> >> GPU queries.
> >>
> >> Link: https://lists.freedesktop.org/archives/amd-gfx/2023-May/093731.html
> >> Cc: Jesse.Zhang at amd.com
> >> Cc: michel at daenzer.net
> >> Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
> > The series is
> >
> > Reviewed-by: Michel Dänzer <mdaenzer at redhat.com>
> >
> > Thanks!
> >
> >
>