[PATCH] drm/amdgpu: disable ASPM for legacy products that don't support ASPM

[Alex Deucher]

On Fri, Apr 8, 2022 at 12:15 PM Gong, Richard <richard.gong at amd.com> wrote:
>
> Hi Alex,
>
> On 4/8/2022 10:54 AM, Alex Deucher wrote:
> > On Fri, Apr 8, 2022 at 11:47 AM Limonciello, Mario
> > <Mario.Limonciello at amd.com> wrote:
> >> [Public]
> >>
> >>
> >>
> >>> -----Original Message-----
> >>> From: Gong, Richard <Richard.Gong at amd.com>
> >>> Sent: Friday, April 8, 2022 10:45
> >>> To: Deucher, Alexander <Alexander.Deucher at amd.com>; Koenig, Christian
> >>> <Christian.Koenig at amd.com>; Pan, Xinhui <Xinhui.Pan at amd.com>;
> >>> airlied at linux.ie; daniel at ffwll.ch
> >>> Cc: amd-gfx at lists.freedesktop.org; dri-devel at lists.freedesktop.org; linux-
> >>> kernel at vger.kernel.org; Limonciello, Mario <Mario.Limonciello at amd.com>;
> >>> Gong, Richard <Richard.Gong at amd.com>
> >>> Subject: [PATCH] drm/amdgpu: disable ASPM for legacy products that don't
> >>> support ASPM
> >>>
> >>> Active State Power Management (ASPM) feature is enabled since kernel
> >>> 5.14.
> >>> However there are some legacy products (WX3200 and RX640 are examples)
> >>> that
> >>> do not support ASPM. Use them as video/display output and system would
> >>> hang
> >>> during suspend/resume.
> >>>
> >>> Add extra check to disable ASPM for old products that don't have
> >>> ASPM support.
> > The patch description is incorrect.  ASPM works just fine on these
> > GPUs.  It's more of an issue with whether the underlying platform
> > supports ASPM or not.  Rather than disabling a chip family, I would
> > prefer to add a check for problematic platforms and disable ASPM on
> > those platforms.
>
> I thought that initially.
>
> But I found out that suspend/resume works just fine on the "problematic"
> platform (Dell Precision 3660, Intel ADL based) + AMD W6400 GFX card.
> With WX3200 or RX640, suspend/resume works only when ASPM was disabled.
> Both WX3200 and RX640 are from CHIP_POLARIS12 family.
>
> This is why I take chip family approach.

Sure, but you could put those polaris12 boards into a different system
and they work just fine with ASPM enabled.

Alex

>
> Regards,
>
> Richard
>
> > Alex
> >
> >>> Signed-off-by: Richard Gong <richard.gong at amd.com>
> >>> Link: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.freedesktop.org%2Fdrm%2Famd%2F-%2Fissues%2F1885&data=04%7C01%7CRichard.Gong%40amd.com%7C96f8f686f75f43abb5ed08da19780fab%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637850300760921285%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=xVKC0Q16ho5Y2GDuN%2Fnx68wm6NzOIyR5xJbiXPgqPpQ%3D&reserved=0
> >>> ---
> >>>   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 ++++
> >>>   1 file changed, 4 insertions(+)
> >>>
> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>> index bb1c025d9001..8987107f41ee 100644
> >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>> @@ -2012,6 +2012,10 @@ static int amdgpu_pci_probe(struct pci_dev
> >>> *pdev,
> >>>        if (amdgpu_aspm == -1 && !pcie_aspm_enabled(pdev))
> >>>                amdgpu_aspm = 0;
> >>>
> >>> +     /* disable ASPM for the legacy products that don't support ASPM */
> >>> +     if ((flags & AMD_ASIC_MASK) == CHIP_POLARIS12)
> >>> +             amdgpu_aspm = 0;
> >>> +
> >> I think it's problematic to disable it for the entire driver.  There might be multiple
> >> AMDGPUs in the system, and others may support ASPM.
> >>
> >> Can it be done just as part of probe for Polaris?
> >>
> >>>        if (amdgpu_virtual_display ||
> >>>            amdgpu_device_asic_has_dc_support(flags & AMD_ASIC_MASK))
> >>>                supports_atomic = true;
> >>> --
> >>> 2.25.1