Re: [PATCH 1/5] atomics: add acquire/release/relaxed variants of some atomic operations

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

On Tue, Jul 14, 2015 at 12:24:20PM +0100, Peter Zijlstra wrote:
> On Tue, Jul 14, 2015 at 12:08:11PM +0100, Will Deacon wrote:
> > On Tue, Jul 14, 2015 at 11:58:37AM +0100, Peter Zijlstra wrote:
> 
> 
> > > #ifndef atomic_add_return_relaxed
> > > 
> > > #define atomic_add_return_relaxed	atomic_add_return
> > > /*
> > >  * If one cannot define a more relaxed version,
> > >  * acquire/release are out the window too.
> > >  */
> > > #define  atomic_add_return_acquire	atomic_add_return
> > > #define  atomic_add_return_release	atomic_add_return
> > > 
> > > #else /* relaxed */
> > > 
> > > #ifndef atomic_add_return_acquire
> > > #define  atomic_add_return_acquire(args...)	\
> > > do {						\
> > > 	atomic_add_return_relaxed(args);	\
> > > 	smp_mb__after_atomic();			\
> > > } while (0)
> > > #endif
> > > 
> > > #ifndef atomic_add_return_release
> > > #define  atomic_add_return_release(args...)	\
> > > do {						\
> > > 	smp_mb__before_atomic();		\
> > > 	atomic_add_return_relaxed(args);	\
> > > } while (0)
> > > #endif
> 
> One could even take it one step further and go:
> 
> #ifndef atomic_add_return
> #define atomic_add_return(args...)		\
> do {						\
> 	smp_mb__before_atomic();		\
> 	atomid_add_return_relaxed(args);	\
> 	smp_mb__after_atomic();			\
> } while (0)

...and

#ifndef atomic_add
#define atomic_add(args...)	(void)atomic_add_return_relaxed(args);

It would mean a new architecture only has to define a barrier instruction
and a handful of relaxed atomics for a bare-minimum atomic.h avoiding
spinlocks.

Will