Further improvements to LOCK_PROFILING:

 - Fix missing initialization in kern_rwlock.c causing bogus times to be collected
 - Move updates to the lock hash to after the lock is released for spin mutexes,
   sleep mutexes, and sx locks
 - Add new kernel build option LOCK_PROFILE_FAST - only update lock profiling
   statistics when an acquisition is contended. This reduces the overhead of
   LOCK_PROFILING to increasing system time by 20%-25% which on
   "make -j8 kernel-toolchain" on a dual woodcrest is unmeasurable in terms
   of wall-clock time. Contrast this to enabling lock profiling without
   LOCK_PROFILE_FAST and I see a 5x-6x slowdown in wall-clock time.

1 files changed, 20 insertions, 5 deletions

diff --git a/sys/kern/kern_sx.c b/sys/kern/kern_sx.c
index e0bff42..2381c06 100644
--- a/sys/kern/kern_sx.c
+++ b/sys/kern/kern_sx.c
@@ -228,7 +228,9 @@ _sx_try_xlock(struct sx *sx, const char *file, int line)
 void
 _sx_sunlock(struct sx *sx, const char *file, int line)
 {
-
+	struct lock_object lo;
+	int count = -1;
+	
 	_sx_assert(sx, SX_SLOCKED, file, line);
 	mtx_lock(sx->sx_lock);
 
@@ -238,8 +240,13 @@ _sx_sunlock(struct sx *sx, const char *file, int line)
 	/* Release. */
 	sx->sx_cnt--;
 
-       if (sx->sx_cnt == 0)
-	       lock_profile_release_lock(&sx->sx_object);
+#ifdef LOCK_PROFILING
+	if (sx->sx_cnt == 0) {
+		memcpy(&lo, &sx->sx_object, sizeof(lo));
+		sx->sx_object.lo_flags &= ~LO_CONTESTED;
+		count = 0;
+	}
+#endif
 	/*
 	 * If we just released the last shared lock, wake any waiters up, giving
 	 * exclusive lockers precedence.  In order to make sure that exclusive
@@ -255,12 +262,16 @@ _sx_sunlock(struct sx *sx, const char *file, int line)
 	LOCK_LOG_LOCK("SUNLOCK", &sx->sx_object, 0, 0, file, line);
 
 	mtx_unlock(sx->sx_lock);
+	if (count == 0)
+		lock_profile_release_lock(&lo);
+
 }
 
 void
 _sx_xunlock(struct sx *sx, const char *file, int line)
 {
-
+	struct lock_object lo;
+	
 	_sx_assert(sx, SX_XLOCKED, file, line);
 	mtx_lock(sx->sx_lock);
 	MPASS(sx->sx_cnt == -1);
@@ -272,7 +283,10 @@ _sx_xunlock(struct sx *sx, const char *file, int line)
 	sx->sx_cnt++;
 	sx->sx_xholder = NULL;
 
-	lock_profile_release_lock(&sx->sx_object);
+#ifdef LOCK_PROFILING
+	memcpy(&lo, &sx->sx_object, sizeof(lo));
+	sx->sx_object.lo_flags &= ~LO_CONTESTED;
+#endif
 	/*
 	 * Wake up waiters if there are any.  Give precedence to slock waiters.
 	 */
@@ -284,6 +298,7 @@ _sx_xunlock(struct sx *sx, const char *file, int line)
 	LOCK_LOG_LOCK("XUNLOCK", &sx->sx_object, 0, 0, file, line);
 
 	mtx_unlock(sx->sx_lock);
+	lock_profile_release_lock(&lo);
 }
 
 int