[REGRESSION] next/master: suspect endianness issue in common PCI capability search macro

[Gerd Bayer <gbayer@linux.ibm.com>]

On Wed, 2025-07-16 at 18:11 -0500, Bjorn Helgaas wrote:
> On Thu, Jul 17, 2025 at 12:11:56AM +0800, Hans Zhang wrote:

<--- snip --->

> > 
> > 
> > Hans Zhang (7):
> >   PCI: Introduce generic bus config read helper function
> >   PCI: Clean up __pci_find_next_cap_ttl() readability
> >   PCI: Refactor standard capability search into common macro
> >   PCI: Refactor extended capability search into common macro
> >   PCI: dwc: Use common PCI host bridge APIs for finding the capabilities
> >   PCI: cadence: Use common PCI host bridge APIs for finding the
> >     capabilities
> >   PCI: cadence: Use cdns_pcie_find_*capability to avoid hardcode
> > 
> >  drivers/pci/access.c                          | 15 ++++
> >  .../pci/controller/cadence/pcie-cadence-ep.c  | 38 ++++----
> >  drivers/pci/controller/cadence/pcie-cadence.c | 30 +++++++
> >  drivers/pci/controller/cadence/pcie-cadence.h | 18 ++--
> >  drivers/pci/controller/dwc/pcie-designware.c  | 83 ++++--------------
> >  drivers/pci/pci.c                             | 76 +++-------------
> >  drivers/pci/pci.h                             | 87 +++++++++++++++++++
> >  include/uapi/linux/pci_regs.h                 |  3 +
> >  8 files changed, 196 insertions(+), 154 deletions(-)
> > 
> > 
> > base-commit: 19272b37aa4f83ca52bdf9c16d5d81bdd1354494
> 
> Applied to pci/capability-search for v6.17, thanks for all this work!

Dear all,

with this series commit 2502a619108b ("PCI: Refactor extended
capability search into PCI_FIND_NEXT_EXT_CAP()")
has landed in linux-next.

This breaks PCI capability search on our s390 test systems - that
showed through mlx5_core's error while binding:

[   27.158991] mlx5_core a000:00:00.0: mlx5_load:1355:(pid 998): Failed
to alloc IRQs

apparently, due to struct pci_dev not showing that it is MSI-X capable.
With this commit reverted, mlx5_core binds successfully again.

I'm sending this as a heads-up while I'll continue to debug this
further - presumably an endianness issue in the macro
PCI_FIND_NEXT_EXT_CAP.

Thanks,
Gerd