[v3] drm/amdgpu: reset asic after system-wide suspend aborted (v3)

[Alex Deucher]

On Wed, Dec 1, 2021 at 1:46 PM Limonciello, Mario
<mario.limonciello at amd.com> wrote:
>
> On 11/24/2021 23:48, Prike Liang wrote:
> > Do ASIC reset at the moment Sx suspend aborted behind of amdgpu suspend
> > to keep AMDGPU in a clean reset state and that can avoid re-initialize
> > device improperly error. Currently,we just always do asic reset in the
> > amdgpu resume until sort out the PM abort case.
> >
> > v2: Remove incomplete PM abort flag and add GPU hive case check for
> > GPU reset.
> >
> > v3: Some dGPU reset method not support at the early resume time and
> > temprorary skip the dGPU case.
> >
> > Signed-off-by: Prike Liang <Prike.Liang at amd.com>
> > ---
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 ++++++++
> >   1 file changed, 8 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > index 7d4115d..f6e1a6a 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > @@ -3983,6 +3983,14 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon)
> >       if (adev->in_s0ix)
> >               amdgpu_gfx_state_change_set(adev, sGpuChangeState_D0Entry);
> >
> > +     /*TODO: In order to not let all-always asic reset affect resume latency
> > +      * need sort out the case which really need asic reset in the resume process.
> > +      * As to the known issue on the system suspend abort behind the AMDGPU suspend,
> > +      * may can sort this case by checking struct suspend_stats which need exported
> > +      * firstly.
> > +      */
> > +     if (adev->flags & AMD_IS_APU)
> > +             amdgpu_asic_reset(adev);
>
> Ideally you only want this to happen on S3 right?  So shouldn't there be
> an extra check for `amdgpu_acpi_is_s0ix_active`?

Shouldn't matter on the resume side.  Only the suspend side.  If we
reset in suspend, we'd end up disabling gfxoff.  On the resume side,
it should safe, but the resume paths for various IPs probably are not
adequate to deal with a reset for S0i3 since they don't re-init as
much hardware.  So it's probably better to skip this for S0i3.

Alex


>
> >       /* post card */
> >       if (amdgpu_device_need_post(adev)) {
> >               r = amdgpu_device_asic_init(adev);
> >
>