Re: [RFC v2 2/3] atomic: Specify alignment for atomic_t and atomic64_t

[Geert Uytterhoeven]

Hi Finn,

On Tue, 30 Sept 2025 at 04:18, Finn Thain <fthain@xxxxxxxxxxxxxx> wrote:
> On Tue, 23 Sep 2025, I wrote:
> > ... there's still some kmem cache or other allocator somewhere that has
> > produced some misaligned path and dentry structures. So we get
> > misaligned atomics somewhere in the VFS and TTY layers. I was unable to
> > find those allocations.
>
> It turned out that the problem wasn't dynamic allocations, it was a local
> variable in the core locking code (kernel/locking/rwsem.c): a misaligned
> long used with an atomic operation (cmpxchg). To get natural alignment for
> 64-bit quantities, I had to align other local variables as well, such as
> the one in ktime_get_real_ts64_mg() that's used with
> atomic64_try_cmpxchg(). The atomic_t branch in my github repo has the
> patches I wrote for that.

So cmpxchg() and friends should not take a volatile void *, but (new)
properly-aligned types, using the new _Generic()?

> To silence the misalignment WARN from CONFIG_DEBUG_ATOMIC, for 64-bit
> atomic operations, for my small m68k .config, it was also necesary to
> increase ARCH_SLAB_MINALIGN to 8. However, I'm not advocating a

Probably ARCH_SLAB_MINALIGN should be 4 on m68k.  Somehow I thought
that was already the case, but it is __alignof__(unsigned long long) = 2.

> ARCH_SLAB_MINALIGN increase, as that wastes memory. I think it might be
> more useful to limit the alignment test for CONFIG_DEBUG_ATOMIC, as
> follows.

Did you check what would be the actual impact of increasing it to 4 or 8?

> --- a/include/linux/instrumented.h
> +++ b/include/linux/instrumented.h
> @@ -68,7 +68,7 @@ static __always_inline void instrument_atomic_read(const volatile void *v, size_
>  {
>         kasan_check_read(v, size);
>         kcsan_check_atomic_read(v, size);
> -       WARN_ON_ONCE(IS_ENABLED(CONFIG_DEBUG_ATOMIC) && ((unsigned long)v & (size - 1)));
> +       WARN_ON_ONCE(IS_ENABLED(CONFIG_DEBUG_ATOMIC) && ((unsigned long)v & (size - 1) & 3));

I'd make that an arch-overridable define instead of hardcoded 3.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds