Re: [PATCH v10 02/14] xen: introduce generic non-atomic test_*bit()

["Oleksii K." <oleksii.kurochko@xxxxxxxxx>]

On Thu, 2024-05-23 at 15:33 +0100, Julien Grall wrote:
> > > >     #include <asm/bitops.h>
> > > >     
> > > > +/**
> > > > + * generic__test_and_set_bit - Set a bit and return its old
> > > > value
> > > > + * @nr: Bit to set
> > > > + * @addr: Address to count from
> > > > + *
> > > > + * This operation is non-atomic and can be reordered.
> > > > + * If two examples of this operation race, one can appear to
> > > > succeed
> > > > + * but actually fail.  You must protect multiple accesses with
> > > > a
> > > > lock.
> > > > + */
> > > 
> > > Sorry for only mentioning this on v10. I think this comment
> > > should be
> > > duplicated (or moved to) on top of test_bit() because this is
> > > what
> > > everyone will use. This will avoid the developper to follow the
> > > function
> > > calls and only notice the x86 version which says "This function
> > > is
> > > atomic and may not be reordered." and would be wrong for all the
> > > other arch.
> > It makes sense to add this comment on top of test_bit(), but I am
> > curious if it is needed to mention that for x86 arch_test_bit() "is
> > atomic and may not be reordered":
> 
> I would say no because any developper modifying common code can't 
> relying it.
But won't then be confusion that if not generic implementation of
test_bit() is chosen then test_bit() can be " atomic and cannot be
reordered " ( as it is in case of x86 )?

~ Oleksii