Re: [PATCH] PCI: Encode corrected PCIe port type with FIELD_PREP()

From: Bjorn Helgaas

Date: Wed Nov 18 2026 - 14:15:19 EST


On Wed, Nov 18, 2026 at 10:25:47AM -0600, Bjorn Helgaas wrote:
> On Wed, Nov 18, 2026 at 03:35:05PM +0000, Gokulkrishna Sekar wrote:
> > set_pcie_port_type() validates PCIe port topology and corrects
> > invalid upstream/downstream port type information in the cached
> > pcie_flags_reg value.
> >
> > When performing these corrections, the code clears the
> > PCI_EXP_FLAGS_TYPE field but writes PCI_EXP_TYPE_UPSTREAM or
> > PCI_EXP_TYPE_DOWNSTREAM directly into pcie_flags_reg. The
> > PCI_EXP_TYPE_* values are decoded port type numbers, not values
> > encoded for the PCI_EXP_FLAGS_TYPE bitfield, so the corrected type
> > is stored in the wrong bit position.
> >
> > As a result, the cached port type may not match the corrected
> > topology and subsequent users of pcie_flags_reg can observe an
> > incorrect PCIe device type.
>
> Looks like subsequent users would think the device was a PCIe Endpoint
> (PCI_EXP_TYPE_ENDPOINT==0x0) instead of an Upstream or Downstream
> Port, but only for devices that claim to be the wrong type in the
> first place, so it probably doesn't actually affect many systems.
>
> But it seems like a candidate for stable anyway.
>
> > Use FIELD_PREP(PCI_EXP_FLAGS_TYPE, ...) when updating the cached
> > field so the corrected port type is encoded consistently with the
> > format used in the PCI Express Capabilities Register.
> >
> > Fixes: ca78410403dd ("PCI: Get rid of dev->has_secondary_link flag")
> > Signed-off-by: Gokulkrishna Sekar <Gokulkrishna.Sekar@xxxxxxxxxx>
> > ---
> > drivers/pci/probe.c | 6 ++++--
> > 1 file changed, 4 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> > index 9a178e0e5f7f..7b09a4463097 100644
> > --- a/drivers/pci/probe.c
> > +++ b/drivers/pci/probe.c
> > @@ -1674,7 +1674,8 @@ void set_pcie_port_type(struct pci_dev *pdev)
> > if (pcie_downstream_port(parent)) {
> > pci_info(pdev, "claims to be downstream port but is acting as upstream port, correcting type\n");
> > pdev->pcie_flags_reg &= ~PCI_EXP_FLAGS_TYPE;
> > - pdev->pcie_flags_reg |= PCI_EXP_TYPE_UPSTREAM;
> > + pdev->pcie_flags_reg |= FIELD_PREP(PCI_EXP_FLAGS_TYPE,
> > + PCI_EXP_TYPE_UPSTREAM);

Oh, I forgot: I think this is a candidate for FIELD_MODIFY() instead
of the "&= ~PCI_EXP_FLAGS_TYPE, FIELD_PREP(PCI_EXP_FLAGS_TYPE)"
sequence so we don't have to repeat PCI_EXP_FLAGS_TYPE.

> > }
> > } else if (type == PCI_EXP_TYPE_UPSTREAM) {
> > /*
> > @@ -1685,7 +1686,8 @@ void set_pcie_port_type(struct pci_dev *pdev)
> > if (pci_pcie_type(parent) == PCI_EXP_TYPE_UPSTREAM) {
> > pci_info(pdev, "claims to be upstream port but is acting as downstream port, correcting type\n");
> > pdev->pcie_flags_reg &= ~PCI_EXP_FLAGS_TYPE;
> > - pdev->pcie_flags_reg |= PCI_EXP_TYPE_DOWNSTREAM;
> > + pdev->pcie_flags_reg |= FIELD_PREP(PCI_EXP_FLAGS_TYPE,
> > + PCI_EXP_TYPE_DOWNSTREAM);
> > }
> > }
> > }
> > --
> > 2.56.1
> >