[slub p2 0/4] SLUB: [RFC] Per cpu partial lists V2

[slub p2 0/4] SLUB: [RFC] Per cpu partial lists V2

[Christoph Lameter]

The following patchset applied on top of the lockless patchset V7. It
introduces per cpu partial lists which allow a performance increase of
around ~15 during contention for the nodelock (can be tested using
hackbench).

These lists help to avoid per nodelocking overhead. Allocator latency
could be further reduced by making these operations work without
disabling interrupts (like the fastpath and the free slowpath) as well as
implementing better ways of handling ther cpu array with partial pages.

I am still not satisfied with the cleanliness of the code after these
changes. Some review with suggestions as to how to restructure the
code given these changes in operations would be appreciated.

It is interesting to note that BSD has gone to a scheme with partial
pages only per cpu (source: Adrian). Transfer of cpu ownerships is
done using IPIs. Probably too much overhead for our taste. The use
of a few per cpu partial pages looks to be beneficial though.

Note that there is no performance gain when there is no contention.

Performance:

				Before		After
./hackbench 100 process 200000
				Time: 2299.072	1742.454
./hackbench 100 process 20000
				Time: 224.654	182.393
./hackbench 100 process 20000
				Time: 227.126	182.780
./hackbench 100 process 20000
				Time: 219.608	182.899
./hackbench 10 process 20000
				Time: 21.769	18.756
./hackbench 10 process 20000
				Time: 21.657	18.938
./hackbench 10 process 20000
				Time: 23.193	19.537
./hackbench 1 process 20000
				Time: 2.337	2.263
./hackbench 1 process 20000
				Time: 2.223	2.271
./hackbench 1 process 20000
				Time: 2.269	2.301

[slub p2 1/4] slub: Prepare inuse field in new_slab()

[Christoph Lameter]

[-- Attachment #1: new_slab --]
[-- Type: text/plain, Size: 1180 bytes --]

inuse will always be set to page->objects. There is no point in
initializing the field to zero in new_slab() and then overwriting
the value in __slab_alloc().

Signed-off-by: Christoph Lameter <cl@linux.com>


---
 mm/slub.c |    5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c	2011-06-14 11:43:54.864072250 -0500
+++ linux-2.6/mm/slub.c	2011-06-14 12:33:15.887854446 -0500
@@ -1341,7 +1341,7 @@ static struct page *new_slab(struct kmem
 	set_freepointer(s, last, NULL);
 
 	page->freelist = start;
-	page->inuse = 0;
+	page->inuse = page->objects;
 	page->frozen = 1;
 out:
 	return page;
@@ -2036,7 +2036,6 @@ new_slab:
 		 */
 		object = page->freelist;
 		page->freelist = NULL;
-		page->inuse = page->objects;
 
 		stat(s, ALLOC_SLAB);
 		c->node = page_to_nid(page);
@@ -2574,7 +2573,7 @@ static void early_kmem_cache_node_alloc(
 	n = page->freelist;
 	BUG_ON(!n);
 	page->freelist = get_freepointer(kmem_cache_node, n);
-	page->inuse++;
+	page->inuse = 1;
 	page->frozen = 0;
 	kmem_cache_node->node[node] = n;
 #ifdef CONFIG_SLUB_DEBUG

[slub p2 2/4] slub: pass kmem_cache_cpu pointer to get_partial()

[Christoph Lameter]

[-- Attachment #1: push_c_into_get_partial --]
[-- Type: text/plain, Size: 3456 bytes --]

Pass the kmem_cache_cpu pointer to get_partial(). That way
we can avoid the this_cpu_write() statements.

Signed-off-by: Christoph Lameter <cl@linux.com>


---
 mm/slub.c |   30 +++++++++++++++---------------
 1 file changed, 15 insertions(+), 15 deletions(-)

Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c	2011-06-14 12:33:15.887854446 -0500
+++ linux-2.6/mm/slub.c	2011-06-14 12:33:20.237854416 -0500
@@ -1451,7 +1451,8 @@ static inline void remove_partial(struct
  * Must hold list_lock.
  */
 static inline int acquire_slab(struct kmem_cache *s,
-		struct kmem_cache_node *n, struct page *page)
+		struct kmem_cache_node *n, struct page *page,
+		struct kmem_cache_cpu *c)
 {
 	void *freelist;
 	unsigned long counters;
@@ -1480,9 +1481,9 @@ static inline int acquire_slab(struct km
 
 	if (freelist) {
 		/* Populate the per cpu freelist */
-		this_cpu_write(s->cpu_slab->freelist, freelist);
-		this_cpu_write(s->cpu_slab->page, page);
-		this_cpu_write(s->cpu_slab->node, page_to_nid(page));
+		c->freelist = freelist;
+		c->page = page;
+		c->node = page_to_nid(page);
 		return 1;
 	} else {
 		/*
@@ -1500,7 +1501,7 @@ static inline int acquire_slab(struct km
  * Try to allocate a partial slab from a specific node.
  */
 static struct page *get_partial_node(struct kmem_cache *s,
-					struct kmem_cache_node *n)
+		struct kmem_cache_node *n, struct kmem_cache_cpu *c)
 {
 	struct page *page;
 
@@ -1515,7 +1516,7 @@ static struct page *get_partial_node(str
 
 	spin_lock(&n->list_lock);
 	list_for_each_entry(page, &n->partial, lru)
-		if (acquire_slab(s, n, page))
+		if (acquire_slab(s, n, page, c))
 			goto out;
 	page = NULL;
 out:
@@ -1526,7 +1527,8 @@ out:
 /*
  * Get a page from somewhere. Search in increasing NUMA distances.
  */
-static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
+static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags,
+		struct kmem_cache_cpu *c)
 {
 #ifdef CONFIG_NUMA
 	struct zonelist *zonelist;
@@ -1566,7 +1568,7 @@ static struct page *get_any_partial(stru
 
 		if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
 				n->nr_partial > s->min_partial) {
-			page = get_partial_node(s, n);
+			page = get_partial_node(s, n, c);
 			if (page) {
 				put_mems_allowed();
 				return page;
@@ -1581,16 +1583,17 @@ static struct page *get_any_partial(stru
 /*
  * Get a partial page, lock it and return it.
  */
-static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
+static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node,
+		struct kmem_cache_cpu *c)
 {
 	struct page *page;
 	int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
 
-	page = get_partial_node(s, get_node(s, searchnode));
+	page = get_partial_node(s, get_node(s, searchnode), c);
 	if (page || node != NUMA_NO_NODE)
 		return page;
 
-	return get_any_partial(s, flags);
+	return get_any_partial(s, flags, c);
 }
 
 #ifdef CONFIG_PREEMPT
@@ -1659,9 +1662,6 @@ void init_kmem_cache_cpus(struct kmem_ca
 	for_each_possible_cpu(cpu)
 		per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
 }
-/*
- * Remove the cpu slab
- */
 
 /*
  * Remove the cpu slab
@@ -2013,7 +2013,7 @@ load_freelist:
 	return object;
 
 new_slab:
-	page = get_partial(s, gfpflags, node);
+	page = get_partial(s, gfpflags, node, c);
 	if (page) {
 		stat(s, ALLOC_FROM_PARTIAL);
 		object = c->freelist;

[slub p2 3/4] slub: return object pointer from get_partial() / new_slab().

[Christoph Lameter]

[-- Attachment #1: object_instead_of_page_return --]
[-- Type: text/plain, Size: 7367 bytes --]

There is no need anymore to return the pointer to a slab page from get_partial()
since it can be assigned to the kmem_cache_cpu structures "page" field.

Return an object pointer instead.

That in turn allows a simplification of the spaghetti code in __slab_alloc().

Signed-off-by: Christoph Lameter <cl@linux.com>


---
 mm/slub.c |  130 ++++++++++++++++++++++++++++++++++----------------------------
 1 file changed, 73 insertions(+), 57 deletions(-)

Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c	2011-05-31 10:20:23.322974925 -0500
+++ linux-2.6/mm/slub.c	2011-05-31 10:20:26.822974896 -0500
@@ -1448,9 +1448,11 @@ static inline void remove_partial(struct
  * Lock slab, remove from the partial list and put the object into the
  * per cpu freelist.
  *
+ * Returns a list of objects or NULL if it fails.
+ *
  * Must hold list_lock.
  */
-static inline int acquire_slab(struct kmem_cache *s,
+static inline void *acquire_slab(struct kmem_cache *s,
 		struct kmem_cache_node *n, struct page *page,
 		struct kmem_cache_cpu *c)
 {
@@ -1481,10 +1483,11 @@ static inline int acquire_slab(struct km
 
 	if (freelist) {
 		/* Populate the per cpu freelist */
-		c->freelist = freelist;
 		c->page = page;
 		c->node = page_to_nid(page);
-		return 1;
+		stat(s, ALLOC_FROM_PARTIAL);
+
+		return freelist;
 	} else {
 		/*
 		 * Slab page came from the wrong list. No object to allocate
@@ -1493,17 +1496,18 @@ static inline int acquire_slab(struct km
 		 */
 		printk(KERN_ERR "SLUB: %s : Page without available objects on"
 			" partial list\n", s->name);
-		return 0;
+		return NULL;
 	}
 }
 
 /*
  * Try to allocate a partial slab from a specific node.
  */
-static struct page *get_partial_node(struct kmem_cache *s,
+static void *get_partial_node(struct kmem_cache *s,
 		struct kmem_cache_node *n, struct kmem_cache_cpu *c)
 {
 	struct page *page;
+	void *object;
 
 	/*
 	 * Racy check. If we mistakenly see no partial slabs then we
@@ -1515,13 +1519,15 @@ static struct page *get_partial_node(str
 		return NULL;
 
 	spin_lock(&n->list_lock);
-	list_for_each_entry(page, &n->partial, lru)
-		if (acquire_slab(s, n, page, c))
+	list_for_each_entry(page, &n->partial, lru) {
+		object = acquire_slab(s, n, page, c);
+		if (object)
 			goto out;
-	page = NULL;
+	}
+	object = NULL;
 out:
 	spin_unlock(&n->list_lock);
-	return page;
+	return object;
 }
 
 /*
@@ -1535,7 +1541,7 @@ static struct page *get_any_partial(stru
 	struct zoneref *z;
 	struct zone *zone;
 	enum zone_type high_zoneidx = gfp_zone(flags);
-	struct page *page;
+	void *object;
 
 	/*
 	 * The defrag ratio allows a configuration of the tradeoffs between
@@ -1568,10 +1574,10 @@ static struct page *get_any_partial(stru
 
 		if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
 				n->nr_partial > s->min_partial) {
-			page = get_partial_node(s, n, c);
-			if (page) {
+			object = get_partial_node(s, n, c);
+			if (object) {
 				put_mems_allowed();
-				return page;
+				return object;
 			}
 		}
 	}
@@ -1583,15 +1589,15 @@ static struct page *get_any_partial(stru
 /*
  * Get a partial page, lock it and return it.
  */
-static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node,
+static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
 		struct kmem_cache_cpu *c)
 {
-	struct page *page;
+	void *object;
 	int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
 
-	page = get_partial_node(s, get_node(s, searchnode), c);
-	if (page || node != NUMA_NO_NODE)
-		return page;
+	object = get_partial_node(s, get_node(s, searchnode), c);
+	if (object || node != NUMA_NO_NODE)
+		return object;
 
 	return get_any_partial(s, flags, c);
 }
@@ -1921,6 +1927,35 @@ slab_out_of_memory(struct kmem_cache *s,
 	}
 }
 
+static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
+			int node, struct kmem_cache_cpu **pc)
+{
+	void *object;
+	struct kmem_cache_cpu *c;
+	struct page *page = new_slab(s, flags, node);
+
+	if (page) {
+		c = __this_cpu_ptr(s->cpu_slab);
+		if (c->page)
+			flush_slab(s, c);
+
+		/*
+		 * No other reference to the page yet so we can
+		 * muck around with it freely without cmpxchg
+		 */
+		object = page->freelist;
+		page->freelist = NULL;
+
+		stat(s, ALLOC_SLAB);
+		c->node = page_to_nid(page);
+		c->page = page;
+		*pc = c;
+	} else
+		object = NULL;
+
+	return object;
+}
+
 /*
  * Slow path. The lockless freelist is empty or we need to perform
  * debugging duties.
@@ -1943,7 +1978,6 @@ static void *__slab_alloc(struct kmem_ca
 			  unsigned long addr, struct kmem_cache_cpu *c)
 {
 	void **object;
-	struct page *page;
 	unsigned long flags;
 	struct page new;
 	unsigned long counters;
@@ -1961,8 +1995,7 @@ static void *__slab_alloc(struct kmem_ca
 	/* We handle __GFP_ZERO in the caller */
 	gfpflags &= ~__GFP_ZERO;
 
-	page = c->page;
-	if (!page)
+	if (!c->page)
 		goto new_slab;
 
 	if (unlikely(!node_match(c, node))) {
@@ -1974,8 +2007,8 @@ static void *__slab_alloc(struct kmem_ca
 	stat(s, ALLOC_SLOWPATH);
 
 	do {
-		object = page->freelist;
-		counters = page->counters;
+		object = c->page->freelist;
+		counters = c->page->counters;
 		new.counters = counters;
 		VM_BUG_ON(!new.frozen);
 
@@ -1987,12 +2020,12 @@ static void *__slab_alloc(struct kmem_ca
 		 *
 		 * If there are objects left then we retrieve them
 		 * and use them to refill the per cpu queue.
-		*/
+		 */
 
-		new.inuse = page->objects;
+		new.inuse = c->page->objects;
 		new.frozen = object != NULL;
 
-	} while (!cmpxchg_double_slab(s, page,
+	} while (!cmpxchg_double_slab(s, c->page,
 			object, counters,
 			NULL, new.counters,
 			"__slab_alloc"));
@@ -2006,50 +2039,33 @@ static void *__slab_alloc(struct kmem_ca
 	stat(s, ALLOC_REFILL);
 
 load_freelist:
-	VM_BUG_ON(!page->frozen);
 	c->freelist = get_freepointer(s, object);
 	c->tid = next_tid(c->tid);
 	local_irq_restore(flags);
 	return object;
 
 new_slab:
-	page = get_partial(s, gfpflags, node, c);
-	if (page) {
-		stat(s, ALLOC_FROM_PARTIAL);
-		object = c->freelist;
+	object = get_partial(s, gfpflags, node, c);
 
-		if (kmem_cache_debug(s))
-			goto debug;
-		goto load_freelist;
-	}
+	if (unlikely(!object)) {
 
-	page = new_slab(s, gfpflags, node);
+		object = new_slab_objects(s, gfpflags, node, &c);
 
-	if (page) {
-		c = __this_cpu_ptr(s->cpu_slab);
-		if (c->page)
-			flush_slab(s, c);
+		if (unlikely(!object)) {
+			if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
+				slab_out_of_memory(s, gfpflags, node);
 
-		/*
-		 * No other reference to the page yet so we can
-		 * muck around with it freely without cmpxchg
-		 */
-		object = page->freelist;
-		page->freelist = NULL;
+			local_irq_restore(flags);
+			return NULL;
+		}
+	}
 
-		stat(s, ALLOC_SLAB);
-		c->node = page_to_nid(page);
-		c->page = page;
+	if (likely(!kmem_cache_debug(s)))
 		goto load_freelist;
-	}
-	if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
-		slab_out_of_memory(s, gfpflags, node);
-	local_irq_restore(flags);
-	return NULL;
 
-debug:
-	if (!object || !alloc_debug_processing(s, page, object, addr))
-		goto new_slab;
+	/* Only entered in the debug case */
+	if (!alloc_debug_processing(s, c->page, object, addr))
+		goto new_slab;	/* Slab failed checks. Next slab needed */
 
 	c->freelist = get_freepointer(s, object);
 	deactivate_slab(s, c);

[slub p2 4/4] slub: [RFC] per cpu cache for partial pages

[Christoph Lameter]

[-- Attachment #1: per_cpu_partial --]
[-- Type: text/plain, Size: 15183 bytes --]

Allow filling out the rest of the kmem_cache_cpu cacheline with pointers to
partial pages. The partial page list is used in slab_free() to avoid
per node lock taking.

In __slab_alloc() we can then take multiple partial pages off the per
node partial list in one go reducing node lock pressure.
We can also use the per cpu partial list in slab_alloc() to avoid scanning
partial lists for pages with free objects.


The main effect of a per cpu partial list is that the per node list_lock
is taken for batches of partial pages instead of individual ones.

This is only a first stab at this. There are some limitations:

1. We have to scan through an percpu array of page pointers. That is fast
   since we stick to a cacheline size.

2. The pickup in __slab_alloc() could consider NUMA locality instead of
   blindly picking the first partial block.

3. The "unfreeze()" function should have common code with deactivate_slab().
   Maybe those can be unified.

Future enhancements:

1. The pickup from the partial list could be perhaps be done without disabling
   interrupts with some work. The free path already puts the page into the
   per cpu partial list without disabling interrupts.

2. Configure the size of the per cpu partial blocks dynamically like the other
   aspects of slab operations.

3. The __slab_free() likely has some code path that are unnecessary now or
   where code is duplicated.

4. We dump all partials if the per cpu array overflows. There must be some other
   better algorithm.

5. We could reduce list_lock overhead further by allocation a set of partial
   lists in slab alloc instead of only one.

Performance:

				Before		After
./hackbench 100 process 200000
				Time: 2299.072	1742.454
./hackbench 100 process 20000
				Time: 224.654	182.393
./hackbench 100 process 20000
				Time: 227.126	182.780
./hackbench 100 process 20000
				Time: 219.608	182.899
./hackbench 10 process 20000
				Time: 21.769	18.756
./hackbench 10 process 20000
				Time: 21.657	18.938
./hackbench 10 process 20000
				Time: 23.193	19.537
./hackbench 1 process 20000
				Time: 2.337	2.263
./hackbench 1 process 20000
				Time: 2.223	2.271
./hackbench 1 process 20000
				Time: 2.269	2.301

Signed-off-by: Christoph Lameter <cl@linux.com>


---
 include/linux/slub_def.h |    4 
 mm/slub.c                |  286 +++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 243 insertions(+), 47 deletions(-)

Index: linux-2.6/include/linux/slub_def.h
===================================================================
--- linux-2.6.orig/include/linux/slub_def.h	2011-06-20 10:02:19.927683344 -0500
+++ linux-2.6/include/linux/slub_def.h	2011-06-20 10:03:40.787682827 -0500
@@ -36,6 +36,8 @@ enum stat_item {
 	ORDER_FALLBACK,		/* Number of times fallback was necessary */
 	CMPXCHG_DOUBLE_CPU_FAIL,/* Failure of this_cpu_cmpxchg_double */
 	CMPXCHG_DOUBLE_FAIL,	/* Number of times that cmpxchg double did not match */
+	CPU_PARTIAL_ALLOC,	/* Used cpu partial on alloc */
+	CPU_PARTIAL_FREE,	/* USed cpu partial on free */
 	NR_SLUB_STAT_ITEMS };
 
 struct kmem_cache_cpu {
@@ -46,6 +48,7 @@ struct kmem_cache_cpu {
 #ifdef CONFIG_SLUB_STATS
 	unsigned stat[NR_SLUB_STAT_ITEMS];
 #endif
+	struct page *partial[];	/* Partially allocated frozen slabs */
 };
 
 struct kmem_cache_node {
@@ -79,6 +82,7 @@ struct kmem_cache {
 	int size;		/* The size of an object including meta data */
 	int objsize;		/* The size of an object without meta data */
 	int offset;		/* Free pointer offset. */
+	int cpu_partial;	/* Number of per cpu partial pages to keep around */
 	struct kmem_cache_order_objects oo;
 
 	/* Allocation and freeing of slabs */
Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c	2011-06-20 10:02:32.987683261 -0500
+++ linux-2.6/mm/slub.c	2011-06-20 10:16:42.027677825 -0500
@@ -1454,7 +1454,7 @@ static inline void remove_partial(struct
  */
 static inline void *acquire_slab(struct kmem_cache *s,
 		struct kmem_cache_node *n, struct page *page,
-		struct kmem_cache_cpu *c)
+		int mode)
 {
 	void *freelist;
 	unsigned long counters;
@@ -1469,7 +1469,8 @@ static inline void *acquire_slab(struct
 		freelist = page->freelist;
 		counters = page->counters;
 		new.counters = counters;
-		new.inuse = page->objects;
+		if (mode)
+			new.inuse = page->objects;
 
 		VM_BUG_ON(new.frozen);
 		new.frozen = 1;
@@ -1480,24 +1481,7 @@ static inline void *acquire_slab(struct
 			"lock and freeze"));
 
 	remove_partial(n, page);
-
-	if (freelist) {
-		/* Populate the per cpu freelist */
-		c->page = page;
-		c->node = page_to_nid(page);
-		stat(s, ALLOC_FROM_PARTIAL);
-
-		return freelist;
-	} else {
-		/*
-		 * Slab page came from the wrong list. No object to allocate
-		 * from. Put it onto the correct list and continue partial
-		 * scan.
-		 */
-		printk(KERN_ERR "SLUB: %s : Page without available objects on"
-			" partial list\n", s->name);
-		return NULL;
-	}
+	return freelist;
 }
 
 /*
@@ -1506,8 +1490,9 @@ static inline void *acquire_slab(struct
 static void *get_partial_node(struct kmem_cache *s,
 		struct kmem_cache_node *n, struct kmem_cache_cpu *c)
 {
-	struct page *page;
-	void *object;
+	struct page *page, *page2;
+	void *object = NULL;
+	int count = 0;
 
 	/*
 	 * Racy check. If we mistakenly see no partial slabs then we
@@ -1519,13 +1504,26 @@ static void *get_partial_node(struct kme
 		return NULL;
 
 	spin_lock(&n->list_lock);
-	list_for_each_entry(page, &n->partial, lru) {
-		object = acquire_slab(s, n, page, c);
-		if (object)
-			goto out;
+	list_for_each_entry_safe(page, page2, &n->partial, lru) {
+		void *t = acquire_slab(s, n, page, count == 0);
+
+		if (!t)
+			break;
+
+		if (!count) {
+			c->page = page;
+			c->node = page_to_nid(page);
+			stat(s, ALLOC_FROM_PARTIAL);
+			count++;
+			object = t;
+		} else {
+			c->partial[count++] = page;
+			page->freelist = t;
+		}
+
+ 		if (count > s->cpu_partial / 2)
+			break;
 	}
-	object = NULL;
-out:
 	spin_unlock(&n->list_lock);
 	return object;
 }
@@ -1820,6 +1818,104 @@ redo:
 	}
 }
 
+/*
+ * Unfreeze a page. Page cannot be full. May be empty. If n is passed then the list lock on that
+ * node was taken. The functions return the pointer to the list_lock that was eventually taken in
+ * this function.
+ *
+ * Races are limited to __slab_free. Meaning that the number of free objects may increase but not
+ * decrease.
+ */
+struct kmem_cache_node *unfreeze(struct kmem_cache *s, struct page *page, struct kmem_cache_node *n)
+{
+	enum slab_modes { M_PARTIAL, M_FREE };
+	enum slab_modes l = M_FREE, m = M_FREE;
+	struct page new;
+	struct page old;
+
+	do {
+
+		old.freelist = page->freelist;
+		old.counters = page->counters;
+		VM_BUG_ON(!old.frozen);
+
+		new.counters = old.counters;
+		new.freelist = old.freelist;
+
+		new.frozen = 0;
+
+		if (!new.inuse && (!n || n->nr_partial < s->min_partial))
+			m = M_FREE;
+		else {
+			struct kmem_cache_node *n2 = get_node(s, page_to_nid(page));
+
+			m = M_PARTIAL;
+			if (n != n2) {
+				if (n)
+					spin_unlock(&n->list_lock);
+
+				n = n2;
+				spin_lock(&n->list_lock);
+			}
+		}
+
+		if (l != m) {
+			if (l == M_PARTIAL)
+				remove_partial(n, page);
+			else
+				add_partial(n, page, 1);
+
+			l = m;
+		}
+
+	} while (!cmpxchg_double_slab(s, page,
+				old.freelist, old.counters,
+				new.freelist, new.counters,
+				"unfreezing slab"));
+
+	if (m == M_FREE) {
+		stat(s, DEACTIVATE_EMPTY);
+		discard_slab(s, page);
+		stat(s, FREE_SLAB);
+	}
+	return n;
+}
+
+/* Batch free the partial pages */
+static void __unfreeze_partials(struct kmem_cache *s, struct page *page)
+{
+	int i;
+	struct kmem_cache_node *n = NULL;
+
+	if (page)
+		n = unfreeze(s, page, NULL);
+
+	for (i = 0; i < s->cpu_partial; i++) {
+		page = this_cpu_read(s->cpu_slab->partial[i]);
+
+		if (page) {
+			this_cpu_write(s->cpu_slab->partial[i], NULL);
+			n = unfreeze(s, page, n);
+		}
+
+	}
+
+	if (n)
+		spin_unlock(&n->list_lock);
+}
+
+static void unfreeze_partials(struct kmem_cache *s, struct page *page)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	__unfreeze_partials(s, page);
+
+	local_irq_restore(flags);
+}
+
+
 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
 {
 	stat(s, CPUSLAB_FLUSH);
@@ -1835,8 +1931,12 @@ static inline void __flush_cpu_slab(stru
 {
 	struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
 
-	if (likely(c && c->page))
-		flush_slab(s, c);
+	if (likely(c)) {
+		if (c->page)
+			flush_slab(s, c);
+
+		__unfreeze_partials(s, NULL);
+	}
 }
 
 static void flush_cpu_slab(void *d)
@@ -1981,6 +2081,7 @@ static void *__slab_alloc(struct kmem_ca
 	unsigned long flags;
 	struct page new;
 	unsigned long counters;
+	int i;
 
 	local_irq_save(flags);
 #ifdef CONFIG_PREEMPT
@@ -1997,7 +2098,7 @@ static void *__slab_alloc(struct kmem_ca
 
 	if (!c->page)
 		goto new_slab;
-
+redo:
 	if (unlikely(!node_match(c, node))) {
 		stat(s, ALLOC_NODE_MISMATCH);
 		deactivate_slab(s, c);
@@ -2045,6 +2146,18 @@ load_freelist:
 	return object;
 
 new_slab:
+	/* First try our cache of partially allocated pages */
+	for (i = 0; i < s->cpu_partial; i++)
+		if (c->partial[i]) {
+			c->page = c->partial[i];
+			c->freelist = NULL;
+			c->partial[i] = NULL;
+			c->node = page_to_nid(c->page);
+			stat(s, CPU_PARTIAL_ALLOC);
+			goto redo;
+		}
+
+	/* Then do expensive stuff like retrieving pages from the partial lists */
 	object = get_partial(s, gfpflags, node, c);
 
 	if (unlikely(!object)) {
@@ -2239,16 +2352,29 @@ static void __slab_free(struct kmem_cach
 		was_frozen = new.frozen;
 		new.inuse--;
 		if ((!new.inuse || !prior) && !was_frozen && !n) {
-                        n = get_node(s, page_to_nid(page));
-			/*
-			 * Speculatively acquire the list_lock.
-			 * If the cmpxchg does not succeed then we may
-			 * drop the list_lock without any processing.
-			 *
-			 * Otherwise the list_lock will synchronize with
-			 * other processors updating the list of slabs.
-			 */
-                        spin_lock_irqsave(&n->list_lock, flags);
+
+			if (!kmem_cache_debug(s) && !prior)
+
+				/*
+				 * Slab was on no list before and will be partially empty
+				 * We can defer the list move and freeze it easily.
+				 */
+				new.frozen = 1;
+
+			else { /* Needs to be taken off a list */
+
+	                        n = get_node(s, page_to_nid(page));
+				/*
+				 * Speculatively acquire the list_lock.
+				 * If the cmpxchg does not succeed then we may
+				 * drop the list_lock without any processing.
+				 *
+				 * Otherwise the list_lock will synchronize with
+				 * other processors updating the list of slabs.
+				 */
+				spin_lock_irqsave(&n->list_lock, flags);
+
+			}
 		}
 		inuse = new.inuse;
 
@@ -2258,7 +2384,23 @@ static void __slab_free(struct kmem_cach
 		"__slab_free"));
 
 	if (likely(!n)) {
-                /*
+       		if (new.frozen && !was_frozen) {
+			int i;
+
+			for (i = 0; i < s->cpu_partial; i++)
+				if (this_cpu_cmpxchg(s->cpu_slab->partial[i], NULL, page) == NULL) {
+					stat(s, CPU_PARTIAL_FREE);
+					return;
+				}
+
+			/*
+			 * partial array is overflowing. Drop them all as well as the one we just
+			 * froze.
+			 */
+			unfreeze_partials(s, page);
+		}
+
+         	/*
 		 * The list lock was not taken therefore no list
 		 * activity can be necessary.
 		 */
@@ -2325,7 +2467,6 @@ static __always_inline void slab_free(st
 	slab_free_hook(s, x);
 
 redo:
-
 	/*
 	 * Determine the currently cpus per cpu slab.
 	 * The cpu may change afterward. However that does not matter since
@@ -2540,6 +2681,9 @@ init_kmem_cache_node(struct kmem_cache_n
 
 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
 {
+	int size = sizeof(struct kmem_cache_cpu) + s->cpu_partial * sizeof(void *);
+	int align = 2 * sizeof(void *);
+
 	BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
 			SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
 
@@ -2547,9 +2691,7 @@ static inline int alloc_kmem_cache_cpus(
 	 * Must align to double word boundary for the double cmpxchg
 	 * instructions to work; see __pcpu_double_call_return_bool().
 	 */
-	s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
-				     2 * sizeof(void *));
-
+	s->cpu_slab = __alloc_percpu(size, align);
 	if (!s->cpu_slab)
 		return 0;
 
@@ -2815,7 +2957,10 @@ static int kmem_cache_open(struct kmem_c
 	 * The larger the object size is, the more pages we want on the partial
 	 * list to avoid pounding the page allocator excessively.
 	 */
-	set_min_partial(s, ilog2(s->size));
+	set_min_partial(s, ilog2(s->size) / 2);
+	s->cpu_partial = min_t(int, (cache_line_size() -
+			sizeof(struct kmem_cache_cpu)) / sizeof(void),
+				s->min_partial / 2);
 	s->refcount = 1;
 #ifdef CONFIG_NUMA
 	s->remote_node_defrag_ratio = 1000;
@@ -4353,6 +4498,12 @@ static ssize_t min_partial_store(struct
 }
 SLAB_ATTR(min_partial);
 
+static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
+{
+	return sprintf(buf, "%u\n", s->cpu_partial);
+}
+SLAB_ATTR_RO(cpu_partial);
+
 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
 {
 	if (!s->ctor)
@@ -4391,6 +4542,41 @@ static ssize_t objects_partial_show(stru
 }
 SLAB_ATTR_RO(objects_partial);
 
+static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
+{
+	unsigned long sum  = 0;
+	int cpu;
+	int len;
+	int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
+
+	if (!data)
+		return -ENOMEM;
+
+	for_each_online_cpu(cpu) {
+		unsigned x = 0;
+		int i;
+
+		for (i = 0; i < s->cpu_partial; i++)
+			if (per_cpu_ptr(s->cpu_slab, cpu)->partial[i])
+				x++;
+
+		data[cpu] = x;
+		sum += x;
+	}
+
+	len = sprintf(buf, "%lu", sum);
+
+#ifdef CONFIG_SMP
+	for_each_online_cpu(cpu) {
+		if (data[cpu] && len < PAGE_SIZE - 20)
+			len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
+	}
+#endif
+	kfree(data);
+	return len + sprintf(buf + len, "\n");
+}
+SLAB_ATTR_RO(slabs_cpu_partial);
+
 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
 {
 	return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
@@ -4713,6 +4899,8 @@ STAT_ATTR(DEACTIVATE_BYPASS, deactivate_
 STAT_ATTR(ORDER_FALLBACK, order_fallback);
 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
+STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
+STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
 #endif
 
 static struct attribute *slab_attrs[] = {
@@ -4721,6 +4909,7 @@ static struct attribute *slab_attrs[] =
 	&objs_per_slab_attr.attr,
 	&order_attr.attr,
 	&min_partial_attr.attr,
+	&cpu_partial_attr.attr,
 	&objects_attr.attr,
 	&objects_partial_attr.attr,
 	&partial_attr.attr,
@@ -4733,6 +4922,7 @@ static struct attribute *slab_attrs[] =
 	&destroy_by_rcu_attr.attr,
 	&shrink_attr.attr,
 	&reserved_attr.attr,
+	&slabs_cpu_partial_attr.attr,
 #ifdef CONFIG_SLUB_DEBUG
 	&total_objects_attr.attr,
 	&slabs_attr.attr,
@@ -4774,6 +4964,8 @@ static struct attribute *slab_attrs[] =
 	&order_fallback_attr.attr,
 	&cmpxchg_double_fail_attr.attr,
 	&cmpxchg_double_cpu_fail_attr.attr,
+	&cpu_partial_alloc_attr.attr,
+	&cpu_partial_free_attr.attr,
 #endif
 #ifdef CONFIG_FAILSLAB
 	&failslab_attr.attr,

Re: [slub p2 0/4] SLUB: [RFC] Per cpu partial lists V2

[Andi Kleen]

On Mon, Jun 20, 2011 at 10:32:44AM -0500, Christoph Lameter wrote:
> The following patchset applied on top of the lockless patchset V7. It
> introduces per cpu partial lists which allow a performance increase of
> around ~15 during contention for the nodelock (can be tested using
> hackbench).

What size system did you test it on?
> 
> These lists help to avoid per nodelocking overhead. Allocator latency
> could be further reduced by making these operations work without
> disabling interrupts (like the fastpath and the free slowpath) as well as
> implementing better ways of handling ther cpu array with partial pages.

I think we really need better batching for the transfers.

-andi
-- 
ak@linux.intel.com -- Speaking for myself only.

Re: [slub p2 0/4] SLUB: [RFC] Per cpu partial lists V2

[Christoph Lameter]

On Mon, 20 Jun 2011, Andi Kleen wrote:

> On Mon, Jun 20, 2011 at 10:32:44AM -0500, Christoph Lameter wrote:
> > The following patchset applied on top of the lockless patchset V7. It
> > introduces per cpu partial lists which allow a performance increase of
> > around ~15 during contention for the nodelock (can be tested using
> > hackbench).
>
> What size system did you test it on?

Sandybridge 4 core (8 cpus) 8G Ram.

> > These lists help to avoid per nodelocking overhead. Allocator latency
> > could be further reduced by making these operations work without
> > disabling interrupts (like the fastpath and the free slowpath) as well as
> > implementing better ways of handling ther cpu array with partial pages.
>
> I think we really need better batching for the transfers.

Could you elaborate on that cryptic remark?

Re: [slub p2 0/4] SLUB: [RFC] Per cpu partial lists V2

[Pekka Enberg]

On Mon, 2011-06-20 at 10:32 -0500, Christoph Lameter wrote:
> The following patchset applied on top of the lockless patchset V7. It
> introduces per cpu partial lists which allow a performance increase of
> around ~15 during contention for the nodelock (can be tested using
> hackbench).
> 
> These lists help to avoid per nodelocking overhead. Allocator latency
> could be further reduced by making these operations work without
> disabling interrupts (like the fastpath and the free slowpath) as well as
> implementing better ways of handling ther cpu array with partial pages.
> 
> I am still not satisfied with the cleanliness of the code after these
> changes. Some review with suggestions as to how to restructure the
> code given these changes in operations would be appreciated.
> 
> It is interesting to note that BSD has gone to a scheme with partial
> pages only per cpu (source: Adrian). Transfer of cpu ownerships is
> done using IPIs. Probably too much overhead for our taste. The use
> of a few per cpu partial pages looks to be beneficial though.
> 
> Note that there is no performance gain when there is no contention.
> 
> Performance:
> 
> 				Before		After
> ./hackbench 100 process 200000
> 				Time: 2299.072	1742.454
> ./hackbench 100 process 20000
> 				Time: 224.654	182.393
> ./hackbench 100 process 20000
> 				Time: 227.126	182.780
> ./hackbench 100 process 20000
> 				Time: 219.608	182.899
> ./hackbench 10 process 20000
> 				Time: 21.769	18.756
> ./hackbench 10 process 20000
> 				Time: 21.657	18.938
> ./hackbench 10 process 20000
> 				Time: 23.193	19.537
> ./hackbench 1 process 20000
> 				Time: 2.337	2.263
> ./hackbench 1 process 20000
> 				Time: 2.223	2.271
> ./hackbench 1 process 20000
> 				Time: 2.269	2.301

Impressive numbers! David, comments on the series?

			Pekka