linux-kernel - Re: [PATCH] PCI: Fix endianness issues in pci_bus_read

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <aIu7uO4NOanxMfAO@wunner.de>
Date: Thu, 31 Jul 2025 20:53:44 +0200
From: Lukas Wunner <lukas@...ner.de>
To: Gerd Bayer <gbayer@...ux.ibm.com>
Cc: 18255117159@....com, bhelgaas@...gle.com, helgaas@...nel.org,
	agordeev@...ux.ibm.com, borntraeger@...ux.ibm.com,
	ilpo.jarvinen@...ux.intel.com, jingoohan1@...il.com,
	kwilczynski@...nel.org, linux-kernel@...r.kernel.org,
	linux-s390@...r.kernel.org, linux-next@...r.kernel.org,
	linux-pci@...r.kernel.org, lpieralisi@...nel.org, mani@...nel.org,
	robh@...nel.org, schnelle@...ux.ibm.com
Subject: Re: [PATCH] PCI: Fix endianness issues in pci_bus_read_config()

On Thu, Jul 31, 2025 at 07:38:58PM +0200, Gerd Bayer wrote:
> Simple pointer-casts to map byte and word reads from PCI config space
> into dwords (i.e. u32) produce unintended results on big-endian systems.
> Add the necessary adjustments under compile-time switch
> CONFIG_CPU_BIG_ENDIAN.
> 
> pci_bus_read_config() was just introduced with
> https://lore.kernel.org/all/20250716161203.83823-2-18255117159@163.com/
> 
> Signed-off-by: Gerd Bayer <gbayer@...ux.ibm.com>
> ---
> Sorry to spill this endianness aware code into drivers/pci, feel free to
> suggest a cleaner approach. This has fixed the issues seen on s390 systems

PCI is little-endian.  On big-endian systems, the endianness conversion
of Config Space accesses happens transparently in the struct pci_ops
->read() and ->write() callbacks.  E.g. on s390, zpci_cfg_load() and
zpci_cfg_store() call le64_to_cpu() and cpu_to_le64(), respectively.

We do not want to mess with endianness in the PCI core, so this isn't
a proper fix IMO.

A viable approach might be to turn pci_bus_read_config() into a macro
in include/linux/pci.h which calls the byte/word/dword variant based
on sizeof(*val) or something like that.

But at this point, with the merge window already open, it's probably
better to drop the pci/capability-search topic branch from the pull
request and retry in the next cycle.

> Since this is still sitting in the a pull-request for upstream,
> I'm not sure if this warrants a Fixes: tag.

In cases like this, do include a Fixes tag but no stable designation.

Thanks,

Lukas