[PATCH] PCI/VGA: Don't assume only VGA device found is the boot VGA device

[Alex Deucher]

On Tue, Oct 22, 2024 at 9:27 PM Kai-Heng Feng <kaihengf at nvidia.com> wrote:
>
>
>
> On 2024/10/22 9:04 PM, Alex Deucher wrote:
> > External email: Use caution opening links or attachments
> >
> >
> > On Tue, Oct 22, 2024 at 2:31 AM Kai-Heng Feng <kaihengf at nvidia.com> wrote:
> >>
> >> Hi Luke,
> >>
> >> On 2024/10/15 4:04 PM, Luke Jones wrote:
> >>> On Mon, 14 Oct 2024, at 5:25 PM, Mario Limonciello wrote:
> >>>> From: Mario Limonciello <mario.limonciello at amd.com>
> >>>>
> >>>> The ASUS GA605W has a NVIDIA PCI VGA device and an AMD PCI display device.
> >>>>
> >>>> ```
> >>>> 65:00.0 VGA compatible controller: NVIDIA Corporation AD106M [GeForce
> >>>> RTX 4070 Max-Q / Mobile] (rev a1)
> >>>> 66:00.0 Display controller: Advanced Micro Devices, Inc. [AMD/ATI]
> >>>> Strix [Radeon 880M / 890M] (rev c1)
> >>>> ```
> >>>>
> >>>> The fallback logic in vga_is_boot_device() flags the NVIDIA dGPU as the
> >>>> boot VGA device, but really the eDP is connected to the AMD PCI display
> >>>> device.
> >>>>
> >>>> Drop this case to avoid marking the NVIDIA dGPU as the boot VGA device.
> >>>>
> >>>> Suggested-by: Alex Deucher <alexander.deucher at amd.com>
> >>>> Reported-by: Luke D. Jones <luke at ljones.dev>
> >>>> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3673
> >>>> Signed-off-by: Mario Limonciello <mario.limonciello at amd.com>
> >>>> ---
> >>>>    drivers/pci/vgaarb.c | 7 -------
> >>>>    1 file changed, 7 deletions(-)
> >>>>
> >>>> diff --git a/drivers/pci/vgaarb.c b/drivers/pci/vgaarb.c
> >>>> index 78748e8d2dba..05ac2b672d4b 100644
> >>>> --- a/drivers/pci/vgaarb.c
> >>>> +++ b/drivers/pci/vgaarb.c
> >>>> @@ -675,13 +675,6 @@ static bool vga_is_boot_device(struct vga_device *vgadev)
> >>>>               return true;
> >>>>       }
> >>>>
> >>>> -    /*
> >>>> -     * Vgadev has neither IO nor MEM enabled.  If we haven't found any
> >>>> -     * other VGA devices, it is the best candidate so far.
> >>>> -     */
> >>>> -    if (!boot_vga)
> >>>> -            return true;
> >>>> -
> >>>>       return false;
> >>>>    }
> >>>>
> >>>> --
> >>>> 2.43.0
> >>>
> >>> Hi Mario,
> >>>
> >>> I can verify that this does leave the `boot_vga` attribute set as 0 for the NVIDIA device.
> >>
> >> Does the following diff work for you?
> >> This variant should be less risky for most systems.
> >>
> >> diff --git a/drivers/pci/vgaarb.c b/drivers/pci/vgaarb.c
> >> index 78748e8d2dba..3fb734cb9c1b 100644
> >> --- a/drivers/pci/vgaarb.c
> >> +++ b/drivers/pci/vgaarb.c
> >> @@ -675,6 +675,9 @@ static bool vga_is_boot_device(struct vga_device *vgadev)
> >>                   return true;
> >>           }
> >>
> >> +       if (vga_arb_integrated_gpu(&pdev->dev))
> >> +               return true;
> >> +
> >
> > The problem is that the integrated graphics does not support VGA.
>
> Right, so the check has to be used much earlier.
>
> I wonder does the integrated GFX have _DOD/_DOS while the discrete one doesn't?
> If that's the case, vga_arb_integrated_gpu() can be used to differentiate which
> one is the boot GFX.

I think the problem is that the boot GPU is being conflated with vga
arb.  In this case the iGPU has no VGA so has no reason to be involved
in vga arb.  Trying to mess with any vga related resources on it could
be problematic.  Do higher levels of the stack look at vga arb to
determine the "primary" GPU?

Alex

>
> Kai-Heng
>
> >
> > Alex
> >
> >>           /*
> >>            * Vgadev has neither IO nor MEM enabled.  If we haven't found any
> >>            * other VGA devices, it is the best candidate so far.
> >>
> >>
> >> Kai-Heng
> >>
> >>>
> >>> Tested-by: Luke D. Jones <luke at ljones.dev>
> >>
>