Index: linux-2.6.16-rc4/mm/slab.c
===================================================================
--- linux-2.6.16-rc4.orig/mm/slab.c	2006-02-17 14:23:45.000000000 -0800
+++ linux-2.6.16-rc4/mm/slab.c	2006-02-21 11:20:28.000000000 -0800
@@ -105,6 +105,7 @@
 #include	<linux/nodemask.h>
 #include	<linux/mempolicy.h>
 #include	<linux/mutex.h>
+#include	<linux/perf.h>
 
 #include	<asm/uaccess.h>
 #include	<asm/cacheflush.h>
@@ -3472,25 +3473,33 @@ static void cache_reap(void *unused)
 {
 	struct list_head *walk;
 	struct kmem_list3 *l3;
+	PC_T count;
+
+	PC_START(count);
 
 	if (!mutex_trylock(&cache_chain_mutex)) {
 		/* Give up. Setup the next iteration. */
 		schedule_delayed_work(&__get_cpu_var(reap_work),
 				      REAPTIMEOUT_CPUC);
+		PC_STOP(count, "MutexLocked");
 		return;
 	}
 
+	PC_STOP(count, "cache_reap_mutex");
 	list_for_each(walk, &cache_chain) {
 		struct kmem_cache *searchp;
 		struct list_head *p;
 		int tofree;
 		struct slab *slabp;
+		PC_T loop;
 
 		searchp = list_entry(walk, struct kmem_cache, next);
 
 		if (searchp->flags & SLAB_NO_REAP)
 			goto next;
 
+		PC_START(loop);
+
 		check_irq_on();
 
 		l3 = searchp->nodelists[numa_node_id()];
@@ -3501,6 +3510,8 @@ static void cache_reap(void *unused)
 		drain_array_locked(searchp, cpu_cache_get(searchp), 0,
 				   numa_node_id());
 
+		PC_STOP(loop, "cache_reap_inner_cpucache");
+
 		if (time_after(l3->next_reap, jiffies))
 			goto next_unlock;
 
@@ -3515,6 +3526,7 @@ static void cache_reap(void *unused)
 			goto next_unlock;
 		}
 
+		PC_STOP(loop, "cache_reap_inner_drain");
 		tofree =
 		    (l3->free_limit + 5 * searchp->num -
 		     1) / (5 * searchp->num);
@@ -3541,13 +3553,16 @@ static void cache_reap(void *unused)
 	      next_unlock:
 		spin_unlock_irq(&l3->list_lock);
 	      next:
+		PC_STOP(loop, "cache_reap_inner_total");
 		cond_resched();
 	}
 	check_irq_on();
 	mutex_unlock(&cache_chain_mutex);
+	PC_STOP(count, "cache_reap_scan");
 	drain_remote_pages();
 	/* Setup the next iteration */
 	schedule_delayed_work(&__get_cpu_var(reap_work), REAPTIMEOUT_CPUC);
+	PC_STOP(count,"cache_reap_total");
 }
 
 #ifdef CONFIG_PROC_FS
Index: linux-2.6.16-rc4/mm/memory.c
===================================================================
--- linux-2.6.16-rc4.orig/mm/memory.c	2006-02-21 11:07:25.000000000 -0800
+++ linux-2.6.16-rc4/mm/memory.c	2006-02-21 11:29:47.000000000 -0800
@@ -48,6 +48,7 @@
 #include <linux/rmap.h>
 #include <linux/module.h>
 #include <linux/init.h>
+#include <linux/perf.h>
 
 #include <asm/pgalloc.h>
 #include <asm/uaccess.h>
@@ -1457,6 +1458,9 @@ static int do_wp_page(struct mm_struct *
 		int reuse = can_share_swap_page(old_page);
 		unlock_page(old_page);
 		if (reuse) {
+			PC_T c;
+			PC_START(c);
+
 			flush_cache_page(vma, address, pte_pfn(orig_pte));
 			entry = pte_mkyoung(orig_pte);
 			entry = maybe_mkwrite(pte_mkdirty(entry), vma);
@@ -1464,6 +1468,7 @@ static int do_wp_page(struct mm_struct *
 			update_mmu_cache(vma, address, entry);
 			lazy_mmu_prot_update(entry);
 			ret |= VM_FAULT_WRITE;
+			PC_STOP(c, "dirty_do_wp_page");
 			goto unlock;
 		}
 	}
@@ -2227,11 +2232,14 @@ static inline int handle_pte_fault(struc
 	if (unlikely(!pte_same(*pte, entry)))
 		goto unlock;
 	if (write_access) {
+		PC_T c;
 		if (!pte_write(entry))
 			return do_wp_page(mm, vma, address,
 					pte, pmd, ptl, entry);
+		PC_START(c);
 		entry = pte_mkdirty(entry);
 		current->thread_info->dirty_pte_count++;
+		PC_STOP(c,"fault_dirty");
 	}
 	entry = pte_mkyoung(entry);
 	if (!pte_same(old_entry, entry)) {