---
 include/linux/perf.h     |   22 ++++++++++++----------
 include/linux/slub_def.h |   14 ++++++++++++--
 kernel/Makefile          |    1 +
 mm/slub.c                |    1 +
 4 files changed, 26 insertions(+), 12 deletions(-)

Index: linux-2.6.23-rc1/include/linux/perf.h
===================================================================
--- linux-2.6.23-rc1.orig/include/linux/perf.h	2007-07-26 23:20:43.000000000 -0700
+++ linux-2.6.23-rc1/include/linux/perf.h	2007-07-26 23:20:46.000000000 -0700
@@ -20,6 +20,10 @@ enum pc_item {
 	PC_DIRECT_RECLAIM_RECLAIMED,
 	PC_RECLAIM_SCANNED,
 	PC_RECLAIM_RECLAIMED,
+	PC_SLAB_ALLOC_FAST,
+	PC_SLAB_ALLOC_SLOW,
+	PC_SLAB_FREE_FAST,
+	PC_SLAB_FREE_SLOW,
 	NR_PC_ITEMS
 };
 
@@ -29,30 +33,28 @@ enum pc_item {
 struct pc {
 	unsigned long time;
 	int processor;
-	enum pc_item item;
 };
 
-#define pc_stop(__pc) pc_bytes(__pc, 0)
+#define pc_stop(__pc, __nr) pc_bytes(__pc, 0, __nr)
 
 #ifdef CONFIG_PERFCOUNT
 
-#define INIT_PC(__var, __item) struct pc __var = \
-		{ get_cycles(), smp_processor_id(), __item }
+#define INIT_PC(__var) struct pc __var = \
+		{ get_cycles(), smp_processor_id() }
 
-static inline void pc_start(struct pc *pc, enum pc_item nr)
+static inline void pc_start(struct pc *pc)
 {
-	pc->item = nr;
 	pc->processor = smp_processor_id();
 	pc->time = get_cycles();
 }
 
-void pc_bytes(struct pc *pc, unsigned long bytes);
+void pc_bytes(struct pc *pc, unsigned long bytes, enum pc_item nr);
 
 #else
 
-#define INIT_PC(__var, __item) do { } while(0)
-static inline void pc_start(struct pc *pc, enum pc_item nr) {}
-static inline void pc_bytes(struct pc *pc, unsigned long bytes) {}
+#define INIT_PC(__var) do { } while(0)
+static inline void pc_start(struct pc *pc) {}
+static inline void pc_bytes(struct pc *pc, unsigned long bytes, enum pc_item nr) {}
 
 #endif
 
Index: linux-2.6.23-rc1/kernel/Makefile
===================================================================
--- linux-2.6.23-rc1.orig/kernel/Makefile	2007-07-26 23:19:55.000000000 -0700
+++ linux-2.6.23-rc1/kernel/Makefile	2007-07-26 23:20:46.000000000 -0700
@@ -51,6 +51,7 @@ obj-$(CONFIG_RELAY) += relay.o
 obj-$(CONFIG_SYSCTL) += utsname_sysctl.o
 obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o
 obj-$(CONFIG_TASKSTATS) += taskstats.o tsacct.o
+obj-$(CONFIG_PERFCOUNT) += perf.o
 
 ifneq ($(CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER),y)
 # According to Alan Modra <alan@linuxcare.com.au>, the -fno-omit-frame-pointer is
Index: linux-2.6.23-rc1/mm/slub.c
===================================================================
--- linux-2.6.23-rc1.orig/mm/slub.c	2007-07-26 23:19:55.000000000 -0700
+++ linux-2.6.23-rc1/mm/slub.c	2007-07-26 23:37:12.000000000 -0700
@@ -20,6 +20,7 @@
 #include <linux/mempolicy.h>
 #include <linux/ctype.h>
 #include <linux/kallsyms.h>
+#include <linux/perf.h>
 
 /*
  * Lock order:
Index: linux-2.6.23-rc1/include/linux/slub_def.h
===================================================================
--- linux-2.6.23-rc1.orig/include/linux/slub_def.h	2007-07-26 23:37:12.000000000 -0700
+++ linux-2.6.23-rc1/include/linux/slub_def.h	2007-07-26 23:39:12.000000000 -0700
@@ -12,6 +12,7 @@
 #include <linux/kobject.h>
 #include <linux/mm_types.h>
 #include <linux/page-flags.h>
+#include <linux/perf.h>
 
 struct kmem_cache_node {
 	spinlock_t list_lock;	/* Protect partial list and nr_partial */
@@ -217,6 +218,8 @@ void *__slab_alloc(struct kmem_cache *s,
 void __slab_free(struct kmem_cache *s,
 			struct page *page, void *x, void *addr);
 
+#include <asm/slub_def.h>
+
 #ifndef ARCH_HAS_SLAB_ALLOC
 /*
  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
@@ -234,6 +237,7 @@ static void __always_inline *slab_alloc(
 	struct page *page;
 	void **object;
 	unsigned long flags;
+	INIT_PC(x);
 
 	local_irq_save(flags);
 	page = s->cpu_slab[smp_processor_id()];
@@ -253,13 +257,16 @@ static void __always_inline *slab_alloc(
 	page->lockless_freelist = object[page->offset];
 	local_irq_restore(flags);
 
-	if (unlikely((gfpflags & __GFP_ZERO) && object))
+	if (unlikely((gfpflags & __GFP_ZERO)))
 		memset(object, 0, s->objsize);
 
+	pc_bytes(&x, s->objsize, PC_SLAB_ALLOC_FAST);
 	return object;
 slow:
 	local_irq_restore(flags);
-	return __slab_alloc(s, gfpflags, node, addr);
+	object = __slab_alloc(s, gfpflags, node, addr);
+	pc_bytes(&x, s->objsize, PC_SLAB_ALLOC_SLOW);
+	return object;
 }
 #endif
 
@@ -280,6 +287,7 @@ static void __always_inline slab_free(st
 {
 	void **object = (void *)x;
 	unsigned long flags;
+	INIT_PC(xx);
 
 	local_irq_save(flags);
 	if (unlikely(page != s->cpu_slab[smp_processor_id()]))
@@ -291,11 +299,13 @@ static void __always_inline slab_free(st
 	object[page->offset] = page->lockless_freelist;
 	page->lockless_freelist = object;
 	local_irq_restore(flags);
+	pc_bytes(&xx, s->objsize, PC_SLAB_FREE_FAST);
 	return;
 
 slow:
 	local_irq_restore(flags);
 	__slab_free(s, page, x, addr);
+	pc_bytes(&xx, s->objsize, PC_SLAB_FREE_SLOW);
 }
 #endif