Re: [PATCH 0/3] PCI/PM: Keep hotplug ports that can't wake from D3hot out of D3

From: Lukas Wunner

Date: Mon Nov 16 2026 - 12:03:35 EST


On Mon, Nov 16, 2026 at 06:55:22PM +0200, Aviv Bakal wrote:
> On Sun, Nov 15, 2026 at 07:22:33PM +0100, Lukas Wunner wrote:
> > Intel Catlow Lake suffers from a similar issue:
> >
> > https://lore.kernel.org/all/20261103193136.2410148-1-sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx/
> >
> > Please try to follow the approach taken there.
>
> Thanks Lukas. In the Catlow Lake case the Root Ports still signal the
> hot-add with a PME interrupt, and only PME Status is unreliable. Our
> ports are switch downstream ports that advertise PME from D3hot but
> never send a PM_PME message, so the Root Port gets no interrupt and a
> pcie_pme_irq() hook would never run. Keeping these ports out of D3
> looks like the only option for the existing devices.

We cache the power states which support PME in struct pci_dev::pme_support.
Have you considered just modifying that in a DECLARE_PCI_FIXUP_FINAL()
callback? See e.g. pci_fixup_no_d0_pme(), pci_fixup_no_msi_no_pme(),
pci_fixup_amd_ehci_pme(), pci_fixup_amd_fch_xhci_pme().

> I sent a v2 that fixes a runtime PM imbalance in patch 1 when NO_D3 is
> set by a FINAL fixup.

Please wait one or two weeks before respinning to allow time for the
community to review your proposed changes and provide feedback.

Thanks,

Lukas