Re: [PATCH] x86, kasan: add KASAN checks to atomic operations

[Dmitry Vyukov]

On Tue, Mar 14, 2017 at 4:44 PM, Mark Rutland <mark.rutland@xxxxxxx> wrote:
>> > -static __always_inline int atomic_read(const atomic_t *v)
>> > +static __always_inline int arch_atomic_read(const atomic_t *v)
>> >  {
>> > -   return READ_ONCE((v)->counter);
>> > +   return READ_ONCE_NOCHECK((v)->counter);
>>
>> Should NOCHEKC come with a comment, because i've no idea why this is so.
>
> I suspect the idea is that given the wrapper will have done the KASAN
> check, duplicating it here is either sub-optimal, or results in
> duplicate splats. READ_ONCE() has an implicit KASAN check,
> READ_ONCE_NOCHECK() does not.
>
> If this is to solve duplicate splats, it'd be worth having a
> WRITE_ONCE_NOCHECK() for arch_atomic_set().
>
> Agreed on the comment, regardless.


Reverted xchg changes.
Added comments re READ_ONCE_NOCHECK() and WRITE_ONCE().
Added file comment.
Split into 3 patches and mailed.

Thanks!