diff --git a/sys/arch/sh3/include/mutex.h b/sys/arch/sh3/include/mutex.h
index eddb6001fd2e..a683f051615b 100644
--- a/sys/arch/sh3/include/mutex.h
+++ b/sys/arch/sh3/include/mutex.h
@@ -1,4 +1,4 @@
-/*	$NetBSD: mutex.h,v 1.7 2007/11/21 10:19:08 yamt Exp $	*/
+/*	$NetBSD: mutex.h,v 1.8 2007/11/29 17:33:09 ad Exp $	*/
 
 /*-
  * Copyright (c) 2002, 2007 The NetBSD Foundation, Inc.
@@ -80,9 +80,9 @@ struct kmutex {
  */
 #define	MUTEX_GIVE(mtx)			/* nothing */
 
-int	_lock_cas(volatile uintptr_t *, uintptr_t, uintptr_t);
+uintptr_t	_lock_cas(volatile uintptr_t *, uintptr_t, uintptr_t);
 
-#define	MUTEX_CAS(p, o, n)		_lock_cas((p), (o), (n))
+#define	MUTEX_CAS(p, o, n)		(_lock_cas((p), (o), (n)) == (o))
 
 #endif	/* __MUTEX_PRIVATE */
 
diff --git a/sys/arch/sh3/include/rwlock.h b/sys/arch/sh3/include/rwlock.h
index b9f73c86ba6d..e35298ade536 100644
--- a/sys/arch/sh3/include/rwlock.h
+++ b/sys/arch/sh3/include/rwlock.h
@@ -1,4 +1,4 @@
-/*	$NetBSD: rwlock.h,v 1.3 2007/11/21 10:19:08 yamt Exp $	*/
+/*	$NetBSD: rwlock.h,v 1.4 2007/11/29 17:33:09 ad Exp $	*/
 
 /*-
  * Copyright (c) 2002, 2007 The NetBSD Foundation, Inc.
@@ -50,9 +50,9 @@ struct krwlock {
 #define	RW_RECEIVE(rw)			/* nothing */
 #define	RW_GIVE(rw)			/* nothing */
 
-#define	RW_CAS(p, o, n)			_lock_cas((p), (o), (n))
+#define	RW_CAS(p, o, n)			(_lock_cas((p), (o), (n)) == (o))
 
-int	_lock_cas(volatile uintptr_t *, uintptr_t, uintptr_t);
+uintptr_t	_lock_cas(volatile uintptr_t *, uintptr_t, uintptr_t);
 
 #endif	/* __RWLOCK_PRIVATE */
 
diff --git a/sys/arch/sh3/sh3/lock_stubs.S b/sys/arch/sh3/sh3/lock_stubs.S
index e6c8d91307e5..00f3a430eecf 100644
--- a/sys/arch/sh3/sh3/lock_stubs.S
+++ b/sys/arch/sh3/sh3/lock_stubs.S
@@ -1,4 +1,4 @@
-/*	$NetBSD: lock_stubs.S,v 1.10 2007/10/17 19:57:08 garbled Exp $	*/
+/*	$NetBSD: lock_stubs.S,v 1.11 2007/11/29 17:33:09 ad Exp $	*/
 
 /*
  * Copyright (c) 2007 Valeriy E. Ushakov
@@ -40,7 +40,7 @@
 
 
 /*
- * LINTSTUB: Func: int _lock_cas(volatile uintptr_t *ptr, uintptr_t old, uintptr_t new);
+ * LINTSTUB: Func: uintptr_t _lock_cas(volatile uintptr_t *ptr, uintptr_t old, uintptr_t new);
  *
  *	Atomic compare-and-swap for kernel use.  SuperH machines are
  *	uniprocessor, so we need to be atomic only w.r.t. interrupts.
@@ -54,15 +54,23 @@
  */
 NENTRY(_lock_cas)
 ALTENTRY(_lock_cas_ras_start)
-	mov.l	@r4, r1
-	cmp/eq	r1, r5		! T = (*ptr == old)
+	mov.l	@r4, r0
+	cmp/eq	r0, r5		! T = (*ptr == old)
 	bf	1f
 	mov.l	r6, @r4		! *ptr = new
 ALTENTRY(_lock_cas_ras_end)
 1:	rts
-	 movt	r0		! retval = T
+	 nop			! retval = *ptr at time of compare
 	SET_ENTRY_SIZE(_lock_cas)
 
+STRONG_ALIAS(_atomic_cas_ulong,_lock_cas)
+STRONG_ALIAS(atomic_cas_ulong,_lock_cas)
+STRONG_ALIAS(_atomic_cas_32,_lock_cas)
+STRONG_ALIAS(atomic_cas_32,_lock_cas)
+STRONG_ALIAS(_atomic_cas_uint,_lock_cas)
+STRONG_ALIAS(atomic_cas_uint,_lock_cas)
+STRONG_ALIAS(_atomic_cas_ptr,_lock_cas)
+STRONG_ALIAS(atomic_cas_ptr,_lock_cas)
 
 #if !defined(LOCKDEBUG)