On Tue, May 16, 2023 at 5:50 PM Limonciello, Mario <mlimonci@xxxxxxx> wrote:
>
>
> On 5/16/2023 4:39 PM, Alex Deucher wrote:
> > On Tue, May 16, 2023 at 2:15 PM Mario Limonciello
> > <mario.limonciello@xxxxxxx> wrote:
> >> On GFX11 if RLC is stopped when not in GFXOFF the system will hang.
> >> Prevent this case from ever happening.
> >>
> >> Tested-by: Juan Martinez <Juan.Martinez@xxxxxxx>
> >> Signed-off-by: Mario Limonciello <mario.limonciello@xxxxxxx>
> >> ---
> >> drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 4 ++++
> >> 1 file changed, 4 insertions(+)
> >>
> >> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
> >> index dcbdb2641086..f1f879d9ed8d 100644
> >> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
> >> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
> >> @@ -1766,6 +1766,10 @@ static void gfx_v11_0_rlc_stop(struct amdgpu_device *adev)
> >> {
> >> u32 tmp = RREG32_SOC15(GC, 0, regRLC_CNTL);
> >>
> >> + if (!adev->gfx.gfx_off_state) {
> >> + dev_err(adev->dev, "GFX is not in GFXOFF\n");
> >> + return;
> >> + }
> > This should move up before the RREG above? Also, I think it would be
> > cleaner to just not mess with the RLC in S0i3. Can we just return
> > early in smu_disable_dpms() for the APU case? All of the DPM features
> > are controlled by the SMU so that function is mostly a nop of APUs
> > anyway.
> >
> > Alex
> That was what the original attempt did when we first identified this issue.
> Unfortunately though just skipping RLC (without patches 1 and 2) means
> that GFXOFF still either doesn't get toggled at suspend entry or isn't fully
>
> off at suspend entry.
>
> This leads to the graphics core behaving erratically upon resume.
>
> So if you're OK with patches 1 and 2, I'll adjust patch 3 to also skip
> RLC for
> APU.
Sure. I wonder if we need something similar as patch 2 for other APUs?
Thinking out loud here, I wonder if we shouldn't just return early in
the top level suspend/resume functions for S0i3.
Alex
