sh: Fix up smp_mb__xxx() memory barriers for SH-4A SMP.

In the past these were simply wrapping to barrier() which was sufficient
on SH SMP platforms predating SH-4A. Unfortunately due to ll/sc semantics
an explicit synco is needed in these cases, which is sorted for us by
just switching these over to smp_mb(). smp_mb() also has the benefit of
being wrapped to barrier() in the UP and non-SH4A cases, so old behaviour
is maintained for those parts.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

arch/sh/include/asm/atomic.h

@@ -78,11 +78,10 @@ static inline int atomic_add_unless(atomic_t *v, int a, int u)
 #define atomic_xchg(v, new) (xchg(&((v)->counter), new))
 #define atomic_inc_not_zero(v) atomic_add_unless((v), 1, 0)
 
-/* Atomic operations are already serializing on SH */
-#define smp_mb__before_atomic_dec()	barrier()
-#define smp_mb__after_atomic_dec()	barrier()
-#define smp_mb__before_atomic_inc()	barrier()
-#define smp_mb__after_atomic_inc()	barrier()
+#define smp_mb__before_atomic_dec()	smp_mb()
+#define smp_mb__after_atomic_dec()	smp_mb()
+#define smp_mb__before_atomic_inc()	smp_mb()
+#define smp_mb__after_atomic_inc()	smp_mb()
 
 #include <asm-generic/atomic-long.h>
 #include <asm-generic/atomic64.h>

arch/sh/include/asm/bitops.h

@@ -26,8 +26,8 @@
 /*
  * clear_bit() doesn't provide any barrier for the compiler.
  */
-#define smp_mb__before_clear_bit()	barrier()
-#define smp_mb__after_clear_bit()	barrier()
+#define smp_mb__before_clear_bit()	smp_mb()
+#define smp_mb__after_clear_bit()	smp_mb()
 
 #ifdef CONFIG_SUPERH32
 static inline unsigned long ffz(unsigned long word)