Re: [PATCH] atomic: skip alignment check for try_cmpxchg() old arg

[Finn Thain]

On Mon, 6 Oct 2025, Arnd Bergmann wrote:

> From: Arnd Bergmann <arnd@xxxxxxxx>
> 
> The 'old' argument in atomic_try_cmpxchg() and related functions is a
> pointer to a normal non-atomic integer number, which does not require
> to be naturally aligned, unlike the atomic_t/atomic64_t types themselves.
> 
> In order to add an alignment check with CONFIG_DEBUG_ATOMIC into the
> normal instrument_atomic_read_write() helper, change this check to use
> the non-atomic instrument_read_write(), the same way that was done
> earlier for try_cmpxchg() in commit ec570320b09f ("locking/atomic:
> Correct (cmp)xchg() instrumentation").
> 
> This prevents warnings on m68k calling the 32-bit atomic_try_cmpxchg()
> with 16-bit aligned arguments as well as several more architectures
> including x86-32 when calling atomic64_try_cmpxchg() with 32-bit
> aligned u64 arguments.
> 
> Reported-by: Finn Thain <fthain@xxxxxxxxxxxxxx>

Tested-by: Finn Thain <fthain@xxxxxxxxxxxxxx>

> Link: https://lore.kernel.org/all/cover.1757810729.git.fthain@xxxxxxxxxxxxxx/
> Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> ---
>  include/linux/atomic/atomic-instrumented.h | 26 +++++++++++-----------
>  scripts/atomic/gen-atomic-instrumented.sh  | 11 +++++----
>  2 files changed, 20 insertions(+), 17 deletions(-)
>