Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230329095933.1203559-1-kai.heng.feng@canonical.com> <CADnq5_MMg4XD3QK72ZfLYvq02hYrX2aApNT2JUEAvQGvBP+NeQ@mail.gmail.com>
In-Reply-To: <CADnq5_MMg4XD3QK72ZfLYvq02hYrX2aApNT2JUEAvQGvBP+NeQ@mail.gmail.com>
From:   Kai-Heng Feng <kai.heng.feng@canonical.com>
Date:   Thu, 30 Mar 2023 08:49:02 +0800
Message-ID: <CAAd53p5A4Rxs0FHXzkgP9sXe=5_mWjfkDeROBT2k71PYKVKTwQ@mail.gmail.com>
Subject: Re: [PATCH 1/2] drm/amdgpu: Reset GPU on S0ix when device supports BOCO
To:     Alex Deucher <alexdeucher@gmail.com>
Cc:     alexander.deucher@amd.com, christian.koenig@amd.com,
        Xinhui.Pan@amd.com, Jingyu Wang <jingyuwang_vip@163.com>,
        Andrey Grodzovsky <andrey.grodzovsky@amd.com>,
        Lijo Lazar <lijo.lazar@amd.com>,
        dri-devel@lists.freedesktop.org,
        =?UTF-8?Q?Michel_D=C3=A4nzer?= <mdaenzer@redhat.com>,
        YiPeng Chai <YiPeng.Chai@amd.com>,
        Mario Limonciello <mario.limonciello@amd.com>,
        Guchun Chen <guchun.chen@amd.com>,
        "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>,
        amd-gfx@lists.freedesktop.org,
        Jiansong Chen <Jiansong.Chen@amd.com>,
        Kenneth Feng <kenneth.feng@amd.com>,
        Tim Huang <tim.huang@amd.com>,
        Bokun Zhang <Bokun.Zhang@amd.com>,
        Hans de Goede <hdegoede@redhat.com>,
        Maxime Ripard <maxime@cerno.tech>,
        Evan Quan <evan.quan@amd.com>,
        Somalapuram Amaranath <Amaranath.Somalapuram@amd.com>,
        linux-kernel@vger.kernel.org, Hawking Zhang <Hawking.Zhang@amd.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, Mar 29, 2023 at 9:21=E2=80=AFPM Alex Deucher <alexdeucher@gmail.com=
> wrote:
>
> On Wed, Mar 29, 2023 at 6:00=E2=80=AFAM Kai-Heng Feng
> <kai.heng.feng@canonical.com> wrote:
> >
> > When the power is lost due to ACPI power resources being turned off, th=
e
> > driver should reset the GPU so it can work anew.
> >
> > First, _PR3 support of the hierarchy needs to be found correctly. Since
> > the GPU on some discrete GFX cards is behind a PCIe switch, checking th=
e
> > _PR3 on downstream port alone is not enough, as the _PR3 can associate
> > to the root port above the PCIe switch.
> >
> > Once the _PR3 is found and BOCO support is correctly marked, use that
> > information to inform the GPU should be reset. This solves an issue tha=
t
> > system freeze on a Intel ADL desktop that uses S0ix for sleep and D3col=
d
> > is supported for the GFX slot.
>
> I don't think we need to reset the GPU.  If the power is turned off, a
> reset shouldn't be necessary. The reset is only necessary when the
> power is not turned off to put the GPU into a known good state.  It
> should be in that state already if the power is turn off.  It sounds
> like the device is not actually getting powered off.

I had the impression that the GPU gets reset because S3 turned the
power rail off.

So the actual intention for GPU reset is because S3 doesn't guarantee
the power is being turned off?

Kai-Heng

>
> Alex
>
> >
> > Fixes: 0064b0ce85bb ("drm/amd/pm: enable ASPM by default")
> > Link: https://gitlab.freedesktop.org/drm/amd/-/issues/1885
> > Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2458
> > Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
> > ---
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c   |  3 +++
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  7 ++++++-
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c    | 12 +++++-------
> >  3 files changed, 14 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c b/drivers/gpu/drm=
/amd/amdgpu/amdgpu_acpi.c
> > index 60b1857f469e..407456ac0e84 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c
> > @@ -987,6 +987,9 @@ bool amdgpu_acpi_should_gpu_reset(struct amdgpu_dev=
ice *adev)
> >         if (amdgpu_sriov_vf(adev))
> >                 return false;
> >
> > +       if (amdgpu_device_supports_boco(adev_to_drm(adev)))
> > +               return true;
> > +
> >  #if IS_ENABLED(CONFIG_SUSPEND)
> >         return pm_suspend_target_state !=3D PM_SUSPEND_TO_IDLE;
> >  #else
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/d=
rm/amd/amdgpu/amdgpu_device.c
> > index f5658359ff5c..d56b7a2bafa6 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > @@ -2181,7 +2181,12 @@ static int amdgpu_device_ip_early_init(struct am=
dgpu_device *adev)
> >
> >         if (!(adev->flags & AMD_IS_APU)) {
> >                 parent =3D pci_upstream_bridge(adev->pdev);
> > -               adev->has_pr3 =3D parent ? pci_pr3_present(parent) : fa=
lse;
> > +               do {
> > +                       if (pci_pr3_present(parent)) {
> > +                               adev->has_pr3 =3D true;
> > +                               break;
> > +                       }
> > +               } while ((parent =3D pci_upstream_bridge(parent)));
> >         }
> >
> >         amdgpu_amdkfd_device_probe(adev);
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/=
amd/amdgpu/amdgpu_drv.c
> > index ba5def374368..5d81fcac4b0a 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> > @@ -2415,10 +2415,11 @@ static int amdgpu_pmops_suspend(struct device *=
dev)
> >         struct drm_device *drm_dev =3D dev_get_drvdata(dev);
> >         struct amdgpu_device *adev =3D drm_to_adev(drm_dev);
> >
> > -       if (amdgpu_acpi_is_s0ix_active(adev))
> > -               adev->in_s0ix =3D true;
> > -       else if (amdgpu_acpi_is_s3_active(adev))
> > +       if (amdgpu_acpi_is_s3_active(adev) ||
> > +           amdgpu_device_supports_boco(drm_dev))
> >                 adev->in_s3 =3D true;
> > +       else if (amdgpu_acpi_is_s0ix_active(adev))
> > +               adev->in_s0ix =3D true;
> >         if (!adev->in_s0ix && !adev->in_s3)
> >                 return 0;
> >         return amdgpu_device_suspend(drm_dev, true);
> > @@ -2449,10 +2450,7 @@ static int amdgpu_pmops_resume(struct device *de=
v)
> >                 adev->no_hw_access =3D true;
> >
> >         r =3D amdgpu_device_resume(drm_dev, true);
> > -       if (amdgpu_acpi_is_s0ix_active(adev))
> > -               adev->in_s0ix =3D false;
> > -       else
> > -               adev->in_s3 =3D false;
> > +       adev->in_s0ix =3D adev->in_s3 =3D false;
> >         return r;
> >  }
> >
> > --
> > 2.34.1
> >