diff -urN linux-2.6.0-test11.lockmeter.orig/arch/ia64/lib/dec_and_lock.c linux-2.6.0-test11.lockmeter/arch/ia64/lib/dec_and_lock.c
--- linux-2.6.0-test11.lockmeter.orig/arch/ia64/lib/dec_and_lock.c	2003-12-11 10:44:10.000000000 +0100
+++ linux-2.6.0-test11.lockmeter/arch/ia64/lib/dec_and_lock.c	2003-12-11 10:43:47.000000000 +0100
@@ -13,6 +13,7 @@
 #include <linux/spinlock.h>
 #include <asm/atomic.h>
 
+#ifndef ATOMIC_DEC_AND_LOCK
 int atomic_dec_and_lock(atomic_t *atomic, spinlock_t *lock)
 {
 	int counter;
@@ -36,3 +37,4 @@
 	spin_unlock(lock);
 	return 0;
 }
+#endif
diff -urN linux-2.6.0-test11.lockmeter.orig/include/asm-ia64/spinlock.h linux-2.6.0-test11.lockmeter/include/asm-ia64/spinlock.h
--- linux-2.6.0-test11.lockmeter.orig/include/asm-ia64/spinlock.h	2003-12-11 10:09:42.000000000 +0100
+++ linux-2.6.0-test11.lockmeter/include/asm-ia64/spinlock.h	2003-12-11 10:15:45.000000000 +0100
@@ -247,13 +247,27 @@
 extern void _metered_spin_unlock(spinlock_t *lock);
 
 /*
- *  Use a less efficient, and inline, atomic_dec_and_lock() if lockmetering
- *  so we can see the callerPC of who is actually doing the spin_lock().
- *  Otherwise, all we see is the generic rollup of all locks done by
- *  atomic_dec_and_lock().
+ *  Matches what is in arch/ia64/lib/dec_and_lock.c, except this one is
+ *  "static inline" so that the spin_lock(), if actually invoked, is charged
+ *  against the real caller, not against the catch-all atomic_dec_and_lock
  */
 static inline int atomic_dec_and_lock(atomic_t *atomic, spinlock_t *lock)
 {
+	int counter;
+	int newcount;
+
+repeat:
+	counter = atomic_read(atomic);
+	newcount = counter-1;
+
+	if (!newcount)
+		goto slow_path;
+
+	if(cmpxchg(&atomic->counter, counter, newcount) != counter)
+		goto repeat;
+	return 0;
+
+slow_path:
 	_metered_spin_lock(lock);
 	if (atomic_dec_and_test(atomic))
 		return 1;