PERFORCE change 64688 for review

[John Baldwin]

http://perforce.freebsd.org/chv.cgi?CH=64688

Change 64688 by jhb at jhb_slimer on 2004/11/09 14:42:52

	Revert most of the changes here to get a much simpler overall diff
	that uses the 386 versions with no membars for all UP kernels and
	only uses hard barriers for SMP kernels.

Affected files ...

.. //depot/projects/smpng/sys/i386/include/atomic.h#17 edit

Differences ...

==== //depot/projects/smpng/sys/i386/include/atomic.h#17 (text+ko) ====

@@ -69,7 +69,7 @@
 
 int atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src);
 
-#define	ATOMIC_STORE_LOAD(TYPE, LOP)			\
+#define	ATOMIC_STORE_LOAD(TYPE, LOP, SOP)			\
 u_##TYPE	atomic_load_acq_##TYPE(volatile u_##TYPE *p);	\
 void		atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
 
@@ -172,15 +172,16 @@
 
 #if defined(__GNUC__) || defined(__INTEL_COMPILER)
 
-#if defined(I386_CPU)
+#if !defined(SMP)
 
 /*
  * We assume that a = b will do atomic loads and stores.  However, on a
  * PentiumPro or higher, reads may pass writes, so for that case we have
- * to use a serializing instruction (i.e. with LOCK) to do the load.  For
- * the 386 case we can use a simple read since 386s don't support SMP.
+ * to use a serializing instruction (i.e. with LOCK) to do the load in
+ * SMP kernels.  For UP kernels, however, the cache of the single processor
+ * is always consistent, so we don't need any memory barriers.
  */
-#define ATOMIC_STORE_LOAD(TYPE, LOP)			\
+#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP)		\
 static __inline u_##TYPE				\
 atomic_load_acq_##TYPE(volatile u_##TYPE *p)		\
 {							\
@@ -195,9 +196,9 @@
 }							\
 struct __hack
 
-#else /* !defined(I386_CPU) */
+#else /* defined(SMP) */
 
-#define ATOMIC_STORE_LOAD(TYPE, LOP)			\
+#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP)		\
 static __inline u_##TYPE				\
 atomic_load_acq_##TYPE(volatile u_##TYPE *p)		\
 {							\
@@ -211,21 +212,26 @@
 	return (res);					\
 }							\
 							\
+/*							\
+ * The XCHG instruction asserts LOCK automagically.	\
+ */							\
 static __inline void					\
 atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)\
 {							\
-	__asm __volatile("" : : : "memory");		\
-	*p = v;						\
+	__asm __volatile(SOP				\
+	: "+m" (*p),			/* 0 */		\
+	  "+r" (v)			/* 1 */		\
+	: : "memory");				 	\
 }							\
 struct __hack
 
-#endif	/* defined(I386_CPU) */
+#endif	/* !defined(SMP) */
 
 #else /* !(defined(__GNUC__) || defined(__INTEL_COMPILER)) */
 
 extern int atomic_cmpset_int(volatile u_int *, u_int, u_int);
 
-#define ATOMIC_STORE_LOAD(TYPE, LOP)					\
+#define ATOMIC_STORE_LOAD(TYPE, LOP, SOP)				\
 extern u_##TYPE atomic_load_acq_##TYPE(volatile u_##TYPE *p);		\
 extern void atomic_store_rel_##TYPE(volatile u_##TYPE *p, u_##TYPE v)
 
@@ -253,10 +259,10 @@
 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");
-ATOMIC_STORE_LOAD(short,"cmpxchgw %w0,%1");
-ATOMIC_STORE_LOAD(int,	"cmpxchgl %0,%1");
-ATOMIC_STORE_LOAD(long,	"cmpxchgl %0,%1");
+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");
 
 #undef ATOMIC_ASM
 #undef ATOMIC_STORE_LOAD