RE: [PATCH] drm/amdgpu: Disable runtime PM for externally attached dGPUs

AMD General

> -----Original Message-----
> From: Alex Deucher <[email protected]>
> Sent: Wednesday, August 12, 2026 10:30 PM
> To: Wang, Yang(Kevin) <[email protected]>
> Cc: [email protected]; Deucher, Alexander
> <[email protected]>; Zhang, Hawking
> <[email protected]>; Feng, Kenneth <[email protected]>; Li,
> Candice <[email protected]>
> Subject: Re: [PATCH] drm/amdgpu: Disable runtime PM for externally
> attached dGPUs
>
> On Wed, Aug 12, 2026 at 4:45 AM Yang Wang <[email protected]>
> wrote:
> >
> > pci_is_thunderbolt_attached() requires an upstream PCI bridge with
> > is_thunderbolt set from an Intel Thunderbolt VSEC. This does not cover
> > the affected ASM4242 USB4 PCI hierarchy:
> >
> >   00:02.2
> >     \- 0f:00.0 [1b21:2421]
> >        +- 10:01.0 [1b21:2423] -> 45:00.0 -> 46:00.0
> >        |  -> 47:00.0 -> 48:00.0 -> 49:00.0 [1002:7590]
> >        \- 10:03.0 -> 76:00.0 [1b21:2425] USB4 Host Router
> >
> > The host router is outside the GPU upstream bridge chain, leaving no
> > ancestor with is_thunderbolt set. PCI core propagates DEVICE_REMOVABLE
> > below the external-facing PCIe tunnel. Disable Runtime PM when either
> > pci_is_thunderbolt_attached() or dev_is_removable() is true.
> >
>
> I could see BOCO being an issue, but shouldn't BACO work?  That should be
> self contained to the driver.

We reproduced the failure on the affected ASM4242 USB4 eGPU topology.
Before applying the workaround, the driver selected BACO:

amdgpu 0000:49:00.0: Using BACO for runtime pm
amdgpu 0000:49:00.0: Failed to exit BACO state!

The failure therefore occurs in the BACO runtime‑resume path, not in a validated BOCO or _PR3 path.

We do not have an actual test case to verify the BOCO case, but it should be disabled together for safety.
Note that the final target state for BOCO is D3cold, while D3hot is an intermediate state.
It could also be problematic if the device cannot fully transition into D3cold, (though this is only my guess and not yet confirmed)

So, should we keep pursuing this issue, or limit our workaround scope to BAMXCO only?

Best Regards,
Kevin


Best Regards,
Kevin
>
> Alex
>
> > Signed-off-by: Yang Wang <[email protected]>
> > ---
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > index 8ebe24d59ab6..a4b4f9645c84 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > @@ -609,6 +609,13 @@ void
> amdgpu_device_detect_runtime_pm_mode(struct amdgpu_device *adev)
> >         int bamaco_support;
> >
> >         adev->pm.rpm_mode = AMDGPU_RUNPM_NONE;
> > +       if (pci_is_thunderbolt_attached(adev->pdev) ||
> > +           dev_is_removable(&adev->pdev->dev)) {
> > +               dev_info(adev->dev,
> > +                        "Runtime PM disabled for externally attached device\n");
> > +               return;
> > +       }
> > +
> >         bamaco_support = amdgpu_device_supports_baco(adev);
> >
> >         switch (amdgpu_runtime_pm) {
> > --
> > 2.54.0
> >