Re: [gcv v3 27/35] arm: Replace __get_cpu_var uses

[Will Deacon <will.deacon@arm.com>]

On Wed, Sep 04, 2013 at 03:54:04PM +0100, Christoph Lameter wrote:
> On Wed, 4 Sep 2013, Will Deacon wrote:
> 
> > Hmm, why can't you get interrupted during atomic64_xchg? On ARM, we have the
> > following sequence:
> 
> AFAICT atomic means one uninterruptible action.

I think it's more subtle than that, but this is all moot for ARM.

> >   static inline u64 atomic64_xchg(atomic64_t *ptr, u64 new)
> >   {
> >   	u64 result;
> >   	unsigned long tmp;
> >
> >   	smp_mb();
> >
> >   	__asm__ __volatile__("@ atomic64_xchg\n"
> >   "1:	ldrexd	%0, %H0, [%3]\n"
> >   "	strexd	%1, %4, %H4, [%3]\n"
> >   "	teq	%1, #0\n"
> >   "	bne	1b"
> >   	: "=&r" (result), "=&r" (tmp), "+Qo" (ptr->counter)
> >   	: "r" (&ptr->counter), "r" (new)
> >   	: "cc");
> >
> >   	smp_mb();
> >
> >   	return result;
> >   }
> >
> > which relies on interrupts clearing the exclusive monitor to force us back
> > around the loop in the inline asm. I could imagine other architectures doing
> > similar, but only detecting the other writer if it used the same
> > instructions.
> 
> Well I have never done ARM asm but this looks vaguely like a cmpxchg loop?
> That would either perform an atomic change or fail and retry?

Correct! The strexd instruction can fail if another access clears the
exclusive monitor.

> If so it still fits the definition of atomic. The change or fail operation
> is atomic.

On ARM, yes. I'm worried that there may be an architecture where the change-
or-fail operation would only fail if the access from the interrupt handler
*also* used that change-or-fail instruction, which isn't the case with
this_cpu_inc.

I have no idea if such an architecture exists :)

Will