diff --git a/sys/kern/kern_sx.c b/sys/kern/kern_sx.c
index 183726c8e558..4cce16d301b8 100644
--- a/sys/kern/kern_sx.c
+++ b/sys/kern/kern_sx.c
@@ -310,6 +310,7 @@ sx_try_slock_(struct sx *sx, const char *file, int line)
 int
 _sx_xlock(struct sx *sx, int opts, const char *file, int line)
 {
+	uintptr_t tid, x;
 	int error = 0;
 
 	if (SCHEDULER_STOPPED())
@@ -321,7 +322,13 @@ _sx_xlock(struct sx *sx, int opts, const char *file, int line)
 	    ("sx_xlock() of destroyed sx @ %s:%d", file, line));
 	WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file,
 	    line, NULL);
-	error = __sx_xlock(sx, curthread, opts, file, line);
+	tid = (uintptr_t)curthread;
+	x = SX_LOCK_UNLOCKED;
+	if (!atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid))
+		error = _sx_xlock_hard(sx, x, tid, opts, file, line);
+	else
+		LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx,
+		    0, 0, file, line, LOCKSTAT_WRITER);
 	if (!error) {
 		LOCK_LOG_LOCK("XLOCK", &sx->lock_object, 0, sx->sx_recurse,
 		    file, line);
@@ -379,7 +386,7 @@ _sx_xunlock(struct sx *sx, const char *file, int line)
 	WITNESS_UNLOCK(&sx->lock_object, LOP_EXCLUSIVE, file, line);
 	LOCK_LOG_LOCK("XUNLOCK", &sx->lock_object, 0, sx->sx_recurse, file,
 	    line);
-	__sx_xunlock(sx, curthread, file, line);
+	_sx_xunlock_hard(sx, (uintptr_t)curthread, file, line);
 	TD_LOCKS_DEC(curthread);
 }
 
@@ -757,8 +764,13 @@ _sx_xunlock_hard(struct sx *sx, uintptr_t tid, const char *file, int line)
 
 	MPASS(!(sx->sx_lock & SX_LOCK_SHARED));
 
-	/* If the lock is recursed, then unrecurse one level. */
-	if (sx_xlocked(sx) && sx_recursed(sx)) {
+	if (!sx_recursed(sx)) {
+		LOCKSTAT_PROFILE_RELEASE_RWLOCK(sx__release, sx,
+		    LOCKSTAT_WRITER);
+		if (atomic_cmpset_rel_ptr(&sx->sx_lock, tid, SX_LOCK_UNLOCKED))
+			return;
+	} else {
+		/* The lock is recursed, unrecurse one level. */
 		if ((--sx->sx_recurse) == 0)
 			atomic_clear_ptr(&sx->sx_lock, SX_LOCK_RECURSED);
 		if (LOCK_LOG_TEST(&sx->lock_object, 0))
diff --git a/sys/sys/sx.h b/sys/sys/sx.h
index a676d2a3deb8..50b0a2415f28 100644
--- a/sys/sys/sx.h
+++ b/sys/sys/sx.h
@@ -145,21 +145,19 @@ struct sx_args {
  * deferred to 'tougher' functions.
  */
 
+#if	(LOCK_DEBUG == 0) && !defined(SX_NOINLINE)
 /* Acquire an exclusive lock. */
 static __inline int
 __sx_xlock(struct sx *sx, struct thread *td, int opts, const char *file,
     int line)
 {
 	uintptr_t tid = (uintptr_t)td;
-	uintptr_t v;
+	uintptr_t v = SX_LOCK_UNLOCKED;
 	int error = 0;
 
-	v = SX_LOCK_UNLOCKED;
-	if (!atomic_fcmpset_acq_ptr(&sx->sx_lock, &v, tid))
+	if (__predict_false(LOCKSTAT_PROFILE_ENABLED(sx__acquire) ||
+	    !atomic_fcmpset_acq_ptr(&sx->sx_lock, &v, tid)))
 		error = _sx_xlock_hard(sx, v, tid, opts, file, line);
-	else 
-		LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx,
-		    0, 0, file, line, LOCKSTAT_WRITER);
 
 	return (error);
 }
@@ -170,12 +168,11 @@ __sx_xunlock(struct sx *sx, struct thread *td, const char *file, int line)
 {
 	uintptr_t tid = (uintptr_t)td;
 
-	if (sx->sx_recurse == 0)
-		LOCKSTAT_PROFILE_RELEASE_RWLOCK(sx__release, sx,
-		    LOCKSTAT_WRITER);
-	if (!atomic_cmpset_rel_ptr(&sx->sx_lock, tid, SX_LOCK_UNLOCKED))
+	if (__predict_false(LOCKSTAT_PROFILE_ENABLED(sx__release) ||
+	    !atomic_cmpset_rel_ptr(&sx->sx_lock, tid, SX_LOCK_UNLOCKED)))
 		_sx_xunlock_hard(sx, tid, file, line);
 }
+#endif
 
 /*
  * Public interface for lock operations.