From: Kees Cook <kees@kernel.org> Sent: Friday, February 28, 2025 9:46 AM

On Thu, Feb 27, 2025 at 03:15:35PM +0100, Christophe Leroy wrote:

quoted

Le 27/02/2025 à 15:05, Michael Kelley a écrit :

quoted

From: Christophe Leroy <redacted> Sent: Thursday, February 27, 2025 2:43 AM

quoted

Le 27/02/2025 à 02:38, Stephen Rothwell a écrit :

quoted

Hi Venkat,

CC Kees Cook for advice.  This is a result of the tests added in commit

    bbeb38b8487a ("string.h: Validate memtostr*()/strtomem*() arguments more carefully")

from the kspp tree.

I note that the comment about memtostr() says "Copy a possibly
non-NUL-term string".

Can you tell more about your config and your environment ?

I just tested with ppc64_defconfig and ppc64le_defconfig, with gcc 12.4,
gcc 13.2 and gcc 14.2 and didn't get that build error.

Christophe

FWIW, I see the same build failures related to __must_be_noncstr()
when building natively on x86 and on arm64. In both cases, it's an
Ubuntu 20.04 installation with gcc 9.4.0 and binutils 2.34.

Looks like I get that problem only with GCC 8.5 and GCC 9.5.

Okay, I've figured this out, and sent an updated patch:
https://lore.kernel.org/lkml/20250228174130.it.875-kees@kernel.org/ (local) 

This matches what you found, namely:

quoted

I don't get it with gcc 10.3 nor 11.3 nor 12.4 nor 13.2 nor 14.2

These have both nonstring and __builtin_has_attribute()

quoted

I don't get it either with gcc 5.5 or 7.5

These have neither.

The problem was in the span of time when nonstring got introduced, but
__builtin_has_attribute() hadn't been yet (GCC 8 and 9). I had accounted
for having neither, but not for missing one. :|

The new patch solves the problem I was seeing with gcc 9.4.0 on
x86 and arm64.

Thanks!

Michael