Use lwsync to provide memory barriers on systems that support it instead

of sync (lwsync is an alternate encoding of sync on systems that do not
support it, providing graceful fallback). This provides more than an order
of magnitude reduction in the time required to acquire or release a mutex.

MFC after:	2 months
This commit is contained in:
Nathan Whitehorn 2012-04-22 19:00:51 +00:00
parent a1f8f44820
commit 6f26a88999
2 changed files with 20 additions and 22 deletions

View File

@ -139,7 +139,7 @@ make_new_leaf(uint64_t esid, uint64_t slbv, struct slbtnode *parent)
* that a lockless searcher always sees a valid path through
* the tree.
*/
powerpc_sync();
mb();
idx = esid2idx(esid, parent->ua_level);
parent->u.ua_child[idx] = child;
@ -187,7 +187,7 @@ make_intermediate(uint64_t esid, struct slbtnode *parent)
idx = esid2idx(child->ua_base, inter->ua_level);
inter->u.ua_child[idx] = child;
setbit(&inter->ua_alloc, idx);
powerpc_sync();
mb();
/* Set up parent to point to intermediate node ... */
idx = esid2idx(inter->ua_base, parent->ua_level);

View File

@ -36,12 +36,10 @@
#error this file needs sys/cdefs.h as a prerequisite
#endif
#define __ATOMIC_BARRIER \
__asm __volatile("sync" : : : "memory")
#define mb() __ATOMIC_BARRIER
#define wmb() mb()
#define rmb() mb()
/* NOTE: lwsync is equivalent to sync on systems without lwsync */
#define mb() __asm __volatile("lwsync" : : : "memory")
#define wmb() __asm __volatile("lwsync" : : : "memory")
#define rmb() __asm __volatile("lwsync" : : : "memory")
/*
* atomic_add(p, v)
@ -94,13 +92,13 @@
atomic_add_acq_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__atomic_add_##type(p, v, t); \
__ATOMIC_BARRIER; \
rmb(); \
} \
\
static __inline void \
atomic_add_rel_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__ATOMIC_BARRIER; \
wmb(); \
__atomic_add_##type(p, v, t); \
} \
/* _ATOMIC_ADD */
@ -180,13 +178,13 @@ _ATOMIC_ADD(long)
atomic_clear_acq_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__atomic_clear_##type(p, v, t); \
__ATOMIC_BARRIER; \
rmb(); \
} \
\
static __inline void \
atomic_clear_rel_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__ATOMIC_BARRIER; \
wmb(); \
__atomic_clear_##type(p, v, t); \
} \
/* _ATOMIC_CLEAR */
@ -282,13 +280,13 @@ _ATOMIC_CLEAR(long)
atomic_set_acq_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__atomic_set_##type(p, v, t); \
__ATOMIC_BARRIER; \
rmb(); \
} \
\
static __inline void \
atomic_set_rel_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__ATOMIC_BARRIER; \
wmb(); \
__atomic_set_##type(p, v, t); \
} \
/* _ATOMIC_SET */
@ -368,13 +366,13 @@ _ATOMIC_SET(long)
atomic_subtract_acq_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__atomic_subtract_##type(p, v, t); \
__ATOMIC_BARRIER; \
rmb(); \
} \
\
static __inline void \
atomic_subtract_rel_##type(volatile u_##type *p, u_##type v) { \
u_##type t; \
__ATOMIC_BARRIER; \
wmb(); \
__atomic_subtract_##type(p, v, t); \
} \
/* _ATOMIC_SUBTRACT */
@ -481,14 +479,14 @@ atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
u_##TYPE v; \
\
v = *p; \
__ATOMIC_BARRIER; \
rmb(); \
return (v); \
} \
\
static __inline void \
atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v) \
{ \
__ATOMIC_BARRIER; \
wmb(); \
*p = v; \
}
@ -598,14 +596,14 @@ atomic_cmpset_acq_int(volatile u_int *p, u_int cmpval, u_int newval)
int retval;
retval = atomic_cmpset_int(p, cmpval, newval);
__ATOMIC_BARRIER;
rmb();
return (retval);
}
static __inline int
atomic_cmpset_rel_int(volatile u_int *p, u_int cmpval, u_int newval)
{
__ATOMIC_BARRIER;
wmb();
return (atomic_cmpset_int(p, cmpval, newval));
}
@ -615,14 +613,14 @@ atomic_cmpset_acq_long(volatile u_long *p, u_long cmpval, u_long newval)
u_long retval;
retval = atomic_cmpset_long(p, cmpval, newval);
__ATOMIC_BARRIER;
rmb();
return (retval);
}
static __inline int
atomic_cmpset_rel_long(volatile u_long *p, u_long cmpval, u_long newval)
{
__ATOMIC_BARRIER;
wmb();
return (atomic_cmpset_long(p, cmpval, newval));
}