PERFORCE change 58403 for review
John Baldwin
jhb at FreeBSD.org
Wed Jul 28 13:50:10 PDT 2004
http://perforce.freebsd.org/chv.cgi?CH=58403
Change 58403 by jhb at jhb_slimer on 2004/07/28 20:49:40
Prompted by some questions on IRC from gibbs@, scottl@, and others:
atomic_store_rel() on ia32 does not actually need a lock, just a
simple store via mov will do. Specifically, in section 7.2.2 of
Volume 3: System Programming Guide of the IA32 Intel Arch Manuals,
it states:
* Writes by a single processor are observed in the same order by
all processors.
Affected files ...
.. //depot/projects/smpng/sys/i386/include/atomic.h#13 edit
Differences ...
==== //depot/projects/smpng/sys/i386/include/atomic.h#13 (text+ko) ====
@@ -180,7 +180,7 @@
* XXX: This is _NOT_ safe on a P6 or higher because it does not guarantee
* memory ordering. These should only be used on a 386.
*/
-#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
+#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP, SCONS) \
static __inline u_##TYPE \
atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
{ \
@@ -197,7 +197,7 @@
#else /* !defined(I386_CPU) */
-#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
+#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP, SCONS) \
static __inline u_##TYPE \
atomic_load_acq_##TYPE(volatile u_##TYPE *p) \
{ \
@@ -211,16 +211,13 @@
return (res); \
} \
\
-/* \
- * The XCHG instruction asserts LOCK automagically. \
- */ \
static __inline void \
atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
{ \
__asm __volatile(SOP \
- : "+m" (*p), /* 0 */ \
- "+r" (v) /* 1 */ \
- : : "memory"); \
+ : "=m" (*p), /* 0 */ \
+ : SCONS (v) /* 1 */ \
+ : "memory"); \
} \
struct __hack
@@ -230,7 +227,7 @@
extern int atomic_cmpset_int(volatile u_int *, u_int, u_int);
-#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP) \
+#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP, SCONS) \
extern u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p); \
extern void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
@@ -258,10 +255,10 @@
ATOMIC_ASM(add, long, "addl %1,%0", "ir", v);
ATOMIC_ASM(subtract, long, "subl %1,%0", "ir", v);
-ATOMIC_STORE_LOAD(char, "cmpxchgb %b0,%1", "xchgb %b1,%0");
-ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "xchgw %w1,%0");
-ATOMIC_STORE_LOAD(int, "cmpxchgl %0,%1", "xchgl %1,%0");
-ATOMIC_STORE_LOAD(long, "cmpxchgl %0,%1", "xchgl %1,%0");
+ATOMIC_STORE_LOAD(char, "cmpxchgb %b0,%1", "movb %b1,%0", "iq");
+ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1", "movw %w1,%0", "ir");
+ATOMIC_STORE_LOAD(int, "cmpxchgl %0,%1", "movl %1,%0", "ir");
+ATOMIC_STORE_LOAD(long, "cmpxchgl %0,%1", "movl %1,%0", "ir");
#undef ATOMIC_ASM
#undef ATOMIC_STORE_LOAD
More information about the p4-projects
mailing list