[ltt-dev] [PATCH 3/4] Add native ARM port for armv7l

[paulmck@linux.vnet.ibm.com (Paul E. McKenney)]

On Wed, Jun 16, 2010 at 05:23:15PM -0400, Mathieu Desnoyers wrote:
> * Paul E. McKenney (paulmck at linux.vnet.ibm.com) wrote:
> > Add native support for armv7l.  Other variants of ARM will likely require
> > separate ports.
> > 
> > Signed-off-by: Paul E. McKenney <paulmck at linux.vnet.ibm.com>
> > ---
> >  configure.ac               |    4 +++
> >  urcu/arch_armv7l.h         |   59 ++++++++++++++++++++++++++++++++++++++++++++
> >  urcu/uatomic_arch_armv7l.h |   48 +++++++++++++++++++++++++++++++++++

[ . . . ]

> > +#ifdef __cplusplus
> > +extern "C" {
> > +#endif 
> > +
> > +/* xchg */
> > +#define uatomic_xchg(addr, v) __sync_lock_test_and_set(addr, v)
> > +
> > +/* cmpxchg */
> > +#define uatomic_cmpxchg(addr, old, _new) \
> > +	__sync_val_compare_and_swap(addr, old, _new)
> > +
> > +/* uatomic_add_return */
> > +#define uatomic_add_return(addr, v) __sync_add_and_fetch(addr, v)
> 
> So, do we end up trusting that gcc got the memory barriers right in the ARM
> __sync_() primitives ? That sounds unlikely.
> 
> I'd vote for surrounding these primitives with smp_mb().

On ARM, my current belief is that the primitives other than
__sync_synchronize() and __sync_lock_release() are set up correctly.

However, I must defer to Paolo and Uli on this.

							Thanx, Paul

> Thanks,
> 
> Mathieu
> 
> > +
> > +#ifdef __cplusplus 
> > +}
> > +#endif
> > +
> > +#include <urcu/uatomic_generic.h>
> > +
> > +#endif /* _URCU_ARCH_UATOMIC_ARMV7L_H */
> > -- 
> > 1.7.0.6
> > 
> 
> -- 
> Mathieu Desnoyers
> Operating System Efficiency R&D Consultant
> EfficiOS Inc.
> http://www.efficios.com