linux-kernel - Re: [v11 2/2] misc: pci_endpoint_test: Fix overflow of bar

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: <a5a6677f-f2dd-4ff6-ab46-3a28f1d1487c@app.fastmail.com>
Date: Wed, 22 Jan 2025 11:26:46 +0100
From: "Arnd Bergmann" <arnd@...db.de>
To: "Niklas Cassel" <cassel@...nel.org>
Cc: "Hans Zhang" <18255117159@....com>,
 "Manivannan Sadhasivam" <manivannan.sadhasivam@...aro.org>,
 Krzysztof Wilczyński <kw@...ux.com>,
 "Kishon Vijay Abraham I" <kishon@...nel.org>,
 "Greg Kroah-Hartman" <gregkh@...uxfoundation.org>, linux-pci@...r.kernel.org,
 linux-kernel@...r.kernel.org, rockswang7@...il.com,
 "linux-tegra@...r.kernel.org" <linux-tegra@...r.kernel.org>,
 "Jon Hunter" <jonathanh@...dia.com>
Subject: Re: [v11 2/2] misc: pci_endpoint_test: Fix overflow of bar_size

On Wed, Jan 22, 2025, at 11:07, Niklas Cassel wrote:
> On Tue, Jan 21, 2025 at 05:46:43PM +0000, Jon Hunter wrote:
>> > diff --git a/drivers/misc/pci_endpoint_test.c b/drivers/misc/>> 
>> 
>> This change breaks building the kernel with GCC v7 and I see ...
>> 
>> ERROR: modpost: "__aeabi_uldivmod" [drivers/misc/pci_endpoint_test.ko]
>> undefined!
>> ERROR: modpost: "__aeabi_ldivmod" [drivers/misc/pci_endpoint_test.ko]
>> undefined!
>> 
>> I know that this is an old GCC version, but this is a farm builder and the
>> kernel still indicates that GCC v5.1 is still supported [0].
>
> do you have any idea what is going on here?
>
> I'm a bit puzzled, since looking at other reports of this error,
> e.g.:
> https://lore.kernel.org/all/20241018151016.3496613-1-arnd@kernel.org/
>
> using div_u64() is usually the solution for this problem (for things that
> are not performance critical), not the cause of it... any ideas?

I have tried to look at the email thread, but not tried to reproduce
it yet. I have two ideas about what might be happening:

a) something causes a /different/ division to call into
   __aeabi_uldivmod(), not the one from div_u64().

b) the compiler notices one of the arguments to div_u64() being
   constant in some cases and splits the calling function into
   two special cases, for both the constant and non-constant
   cases. This sometimes confuses the __builtin_constant_p()
   in do_div() that decides to fall back to a 32-bit division.

Try looking at the .s file when you run 
'make drivers/misc/pci_endpoint_test.s' to see where exactly it
calls the two division functions, maybe you see the problem then,
otherwise I can try to reproduce it here.

    Arnd