[PATCH 01/10] Revert "drm/amd/amdgpu: add pipe1 hardware support"

[Alex Deucher]

On Tue, May 27, 2025 at 12:45 PM Rodrigo Siqueira <siqueira at igalia.com> wrote:
>
> On 05/22, Alex Deucher wrote:
> > This reverts commit b7a1a0ef12b81957584fef7b61e2d5ec049c7209.
> >
> > A user reported stuttering under heavy gfx load with this commit.
> > I suspect it's due to the fact that the gfx contexts are shared
> > between the pipes so if there is alot of load on one pipe, we could
> > end up stalling waiting for a context.
>
> Is there any mechanism to check the load per pipe? Debugfs? UMR?

We expose one queue per pipe, so you can look at the outstanding
fences on each gfx queue in debugfs.

>
> >
> > On top of that, disabling the second pipe may fix the reliability
> > of vmid resets.
> >
> > Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3519
>
> btw, since this patch addresses an issue with Gitlab, should it be sent
> outside of this series?

I don't know that we ultimately want to apply this patch or not.  I
was thinking it might be helpful to address the issues Christian saw
with his original series and I just kept to reduce the possible
factors involved in validating this.  Ultimately though it's a trade
off between parallel execution and sharing available contexts.

Alex

>
> Thanks
>
> > Reviewed-by: Christian König <christian.koenig at amd.com>
> > Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
> > ---
> >  drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> > index 75ea071744eb5..14cbd1f08eb5c 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> > @@ -4781,7 +4781,7 @@ static int gfx_v10_0_sw_init(struct amdgpu_ip_block *ip_block)
> >       case IP_VERSION(10, 3, 3):
> >       case IP_VERSION(10, 3, 7):
> >               adev->gfx.me.num_me = 1;
> > -             adev->gfx.me.num_pipe_per_me = 2;
> > +             adev->gfx.me.num_pipe_per_me = 1;
> >               adev->gfx.me.num_queue_per_pipe = 2;
> >               adev->gfx.mec.num_mec = 2;
> >               adev->gfx.mec.num_pipe_per_mec = 4;
> > --
> > 2.49.0
> >
>
> --
> Rodrigo Siqueira