[-mm] [PATCH 0/4] memcg : radix-tree page_cgroup v2

[-mm] [PATCH 0/4] memcg : radix-tree page_cgroup v2

[KAMEZAWA Hiroyuki]

Hi, this is updated version, but still against 2.6.25-rc5-mm1.

Now, a memory controller needs an extra member in 'struct page' to manage 
'struct page_cgroup'. This patch tries to remove it by radix-tree.

Before this patch, page and page_cgroup relationship is

    pfn <-> struct page <-> struct page_cgroup
    lock for page_cgroup is in struct page.

After this patch, 
    struct page <-> pfn <-> struct page_cgroup.
    lock for page_cgroup is in struct page_cgroup itself.

Pros.
   - we can remove an extra pointer in struct page.
   - lock for page_cgroup is moved to page_cgroup inself from struct page.
Cons.
   - For avoiding too much access to radix-tree, some kind of workaround is
     necessary. On this patch set, page_cgroup is managed by chunk of an order.
   - you'll see small performance regression.

I measured UnixBench execl test (this is very sensitive to page_cgroup).
and adjusted codes to get better performance.

Below is a test result on x86-64/2cpu/1core machine.
(All below result is under memory controller with unlimited limit.
 kmem allocator is SLAB.
 All tests are done right after boot.)

(1) is a result before this set, (4) is after.

         TEST                                BASELINE     RESULT      INDEX
(1)      Execl Throughput                        43.0     2868.8      667.2
(2)      Execl Throughput                        43.0     2810.3      653.6
(3)      Execl Throughput                        43.0     2836.9      659.7
(4)      Execl Throughput                        43.0     2846.0      661.9
(5)      Execl Throughput                        43.0     2862.0      665.6
(6)      Execl Throughput                        43.0     3110.0      723.3

(1) .... rc5-mm1 + memory controller
(2) .... patch 1/4 is applied.      (use radix-tree always.)
(3) .... patch [1-3]/4 are applied. (caching by percpu)
(4) .... patch [1-4]/4 are applied. (uses prefetch)
(5) .... adjust sizeof(struct page) to be 64 bytes by padding.
(6) .... rc5-mm1 *without* memory controller

If you notice something good to be tried, please teach me :)

ia64 8cpu/NUMA shows similar results.

Changelog:
 - folded into a big patch to make bisect easier. The number of patches is 4, now.
 - updated to comments on version 1.
 - small performance improvemnts against version 1.

Thanks,
-Kame




--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[-mm] [PATCH 1/4] memcg : radix-tree page_cgroup

[KAMEZAWA Hiroyuki]

This patch implements radixt-tree based page cgroup.

This patch does
 * add radix-tree based page_cgroup look up subsystem.
 * remove bit_spin_lock used by page_cgroup.

changes:

Before patch
 * struct page had pointer to page_cgroup. Then, relationship between objects
   was pfn <-> struct page <-> struct page_cgroup
 * (spin) lock for page_cgroup was in struct page.
 * page_cgroup->refcnt is incremented before charge is done.
 * page migration does complicated page_cgroup migration under locks.

After patch
 * struct page has no pointer to page_cgroup. Relationship between objects
   is struct page <-> pfn <-> struct page_cgroup -> struct page,
 * page_cgroup has its own spin lock.
 * page_cgroup->refcnt is incremented after charge is done.
 * page migration accounts a new page before migration. By this, we can
   avoid complicated locks. 

tested on ia64/NUMA, x86_64/SMP.

Changelog v1 -> v2:
 * create a folded patch. maybe good for bysect.
 * removed special handling codes for new pages under migration
   Added PG_LRU check to force_empty.
 * reflected comments.
 * Added comments in the head of page_cgroup.c
 * order of page_cgroup is automatically calculated.
 * fixed handling of root_node[] entries in page_cgroup_init().
 * rewrite init_page_cgroup_head() to do minimum work.
 * fixed N_NORMAL_MEMORY handling.
 
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>


 include/linux/memcontrol.h  |   17 --
 include/linux/mm_types.h    |    3 
 include/linux/page_cgroup.h |   56 +++++++
 mm/Makefile                 |    2 
 mm/memcontrol.c             |  332 ++++++++++++++++----------------------------
 mm/migrate.c                |   22 +-
 mm/page_alloc.c             |    8 -
 mm/page_cgroup.c            |  260 ++++++++++++++++++++++++++++++++++
 8 files changed, 463 insertions(+), 237 deletions(-)

Index: linux-2.6.25-rc5-mm1-k/include/linux/page_cgroup.h
===================================================================
--- /dev/null
+++ linux-2.6.25-rc5-mm1-k/include/linux/page_cgroup.h
@@ -0,0 +1,56 @@
+#ifndef __LINUX_PAGE_CGROUP_H
+#define __LINUX_PAGE_CGROUP_H
+
+#ifdef CONFIG_CGROUP_MEM_RES_CTLR
+/*
+ * page_cgroup is yet another mem_map structure for accounting  usage.
+ * but, unlike mem_map, allocated on demand for accounted pages.
+ * see also memcontrol.h
+ * In nature, this consumes much amount of memory.
+ */
+
+struct mem_cgroup;
+
+struct page_cgroup {
+	spinlock_t		lock;        /* lock for all members */
+	int  			refcnt;      /* reference count */
+	struct mem_cgroup 	*mem_cgroup; /* current cgroup subsys */
+	struct list_head 	lru;         /* for per cgroup LRU */
+	int    			flags;	     /* See below */
+	struct page 		*page;       /* the page this accounts for*/
+};
+
+/* flags */
+#define PAGE_CGROUP_FLAG_CACHE	(0x1)	/* charged as cache. */
+#define PAGE_CGROUP_FLAG_ACTIVE (0x2)	/* is on active list */
+
+/*
+ * look up page_cgroup. returns NULL if not exists.
+ */
+extern struct page_cgroup *get_page_cgroup(struct page *page);
+
+
+/*
+ * look up page_cgroup, allocate new one if it doesn't exist.
+ * Return value is
+ *   1. page_cgroup, at success.
+ *   2. -EXXXXX, at failure.
+ *   3. NULL, at boot.
+ */
+extern struct page_cgroup *
+get_alloc_page_cgroup(struct page *page, gfp_t gfpmask);
+
+#else
+
+static inline struct page_cgroup *get_page_cgroup(struct page *page)
+{
+	return NULL;
+}
+
+static inline struct page_cgroup *
+get_alloc_page_cgroup(struct page *page, gfp_t gfpmask)
+{
+	return NULL;
+}
+#endif
+#endif
Index: linux-2.6.25-rc5-mm1-k/mm/memcontrol.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/memcontrol.c
+++ linux-2.6.25-rc5-mm1-k/mm/memcontrol.c
@@ -30,6 +30,7 @@
 #include <linux/spinlock.h>
 #include <linux/fs.h>
 #include <linux/seq_file.h>
+#include <linux/page_cgroup.h>
 
 #include <asm/uaccess.h>
 
@@ -92,7 +93,7 @@ struct mem_cgroup_per_zone {
 	/*
 	 * spin_lock to protect the per cgroup LRU
 	 */
-	spinlock_t		lru_lock;
+	spinlock_t		lru_lock;	/* irq should be off. */
 	struct list_head	active_list;
 	struct list_head	inactive_list;
 	unsigned long count[NR_MEM_CGROUP_ZSTAT];
@@ -139,33 +140,6 @@ struct mem_cgroup {
 };
 static struct mem_cgroup init_mem_cgroup;
 
-/*
- * We use the lower bit of the page->page_cgroup pointer as a bit spin
- * lock.  We need to ensure that page->page_cgroup is at least two
- * byte aligned (based on comments from Nick Piggin).  But since
- * bit_spin_lock doesn't actually set that lock bit in a non-debug
- * uniprocessor kernel, we should avoid setting it here too.
- */
-#define PAGE_CGROUP_LOCK_BIT 	0x0
-#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
-#define PAGE_CGROUP_LOCK 	(1 << PAGE_CGROUP_LOCK_BIT)
-#else
-#define PAGE_CGROUP_LOCK	0x0
-#endif
-
-/*
- * A page_cgroup page is associated with every page descriptor. The
- * page_cgroup helps us identify information about the cgroup
- */
-struct page_cgroup {
-	struct list_head lru;		/* per cgroup LRU list */
-	struct page *page;
-	struct mem_cgroup *mem_cgroup;
-	int ref_cnt;			/* cached, mapped, migrating */
-	int flags;
-};
-#define PAGE_CGROUP_FLAG_CACHE	(0x1)	/* charged as cache */
-#define PAGE_CGROUP_FLAG_ACTIVE (0x2)	/* page is active in this cgroup */
 
 static int page_cgroup_nid(struct page_cgroup *pc)
 {
@@ -256,37 +230,6 @@ void mm_free_cgroup(struct mm_struct *mm
 	css_put(&mm->mem_cgroup->css);
 }
 
-static inline int page_cgroup_locked(struct page *page)
-{
-	return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
-}
-
-static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
-{
-	VM_BUG_ON(!page_cgroup_locked(page));
-	page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
-}
-
-struct page_cgroup *page_get_page_cgroup(struct page *page)
-{
-	return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
-}
-
-static void lock_page_cgroup(struct page *page)
-{
-	bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
-}
-
-static int try_lock_page_cgroup(struct page *page)
-{
-	return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
-}
-
-static void unlock_page_cgroup(struct page *page)
-{
-	bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
-}
-
 static void __mem_cgroup_remove_list(struct page_cgroup *pc)
 {
 	int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
@@ -356,6 +299,10 @@ void mem_cgroup_move_lists(struct page *
 	struct mem_cgroup_per_zone *mz;
 	unsigned long flags;
 
+	/* This GFP will be ignored..*/
+	pc = get_page_cgroup(page);
+	if (!pc)
+		return;
 	/*
 	 * We cannot lock_page_cgroup while holding zone's lru_lock,
 	 * because other holders of lock_page_cgroup can be interrupted
@@ -363,17 +310,15 @@ void mem_cgroup_move_lists(struct page *
 	 * safely get to page_cgroup without it, so just try_lock it:
 	 * mem_cgroup_isolate_pages allows for page left on wrong list.
 	 */
-	if (!try_lock_page_cgroup(page))
+	if (!spin_trylock_irqsave(&pc->lock, flags))
 		return;
-
-	pc = page_get_page_cgroup(page);
-	if (pc) {
+	if (pc->refcnt) {
 		mz = page_cgroup_zoneinfo(pc);
-		spin_lock_irqsave(&mz->lru_lock, flags);
+		spin_lock(&mz->lru_lock);
 		__mem_cgroup_move_lists(pc, active);
-		spin_unlock_irqrestore(&mz->lru_lock, flags);
+		spin_unlock(&mz->lru_lock);
 	}
-	unlock_page_cgroup(page);
+	spin_unlock_irqrestore(&pc->lock, flags);
 }
 
 /*
@@ -525,7 +470,8 @@ unsigned long mem_cgroup_isolate_pages(u
  * < 0 if the cgroup is over its limit
  */
 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
-				gfp_t gfp_mask, enum charge_type ctype)
+				gfp_t gfp_mask, enum charge_type ctype,
+				struct mem_cgroup *memcg)
 {
 	struct mem_cgroup *mem;
 	struct page_cgroup *pc;
@@ -536,33 +482,23 @@ static int mem_cgroup_charge_common(stru
 	if (mem_cgroup_subsys.disabled)
 		return 0;
 
+	pc = get_alloc_page_cgroup(page, gfp_mask);
+	/* Before kamalloc initialization, get_page_cgroup can retrun NULL */
+	if (unlikely(!pc || IS_ERR(pc)))
+		return PTR_ERR(pc);
+
+	spin_lock_irqsave(&pc->lock, flags);
 	/*
-	 * Should page_cgroup's go to their own slab?
-	 * One could optimize the performance of the charging routine
-	 * by saving a bit in the page_flags and using it as a lock
-	 * to see if the cgroup page already has a page_cgroup associated
-	 * with it
-	 */
-retry:
-	lock_page_cgroup(page);
-	pc = page_get_page_cgroup(page);
-	/*
-	 * The page_cgroup exists and
-	 * the page has already been accounted.
+	 * Has the page already been accounted ?
 	 */
-	if (pc) {
-		VM_BUG_ON(pc->page != page);
-		VM_BUG_ON(pc->ref_cnt <= 0);
-
-		pc->ref_cnt++;
-		unlock_page_cgroup(page);
-		goto done;
+	if (pc->refcnt > 0) {
+		pc->refcnt++;
+		spin_unlock_irqrestore(&pc->lock, flags);
+		goto success;
 	}
-	unlock_page_cgroup(page);
+	spin_unlock_irqrestore(&pc->lock, flags);
 
-	pc = kzalloc(sizeof(struct page_cgroup), gfp_mask);
-	if (pc == NULL)
-		goto err;
+	/* Note: *new* pc's refcnt is still 0 here. */
 
 	/*
 	 * We always charge the cgroup the mm_struct belongs to.
@@ -570,20 +506,24 @@ retry:
 	 * thread group leader migrates. It's possible that mm is not
 	 * set, if so charge the init_mm (happens for pagecache usage).
 	 */
-	if (!mm)
-		mm = &init_mm;
-
-	rcu_read_lock();
-	mem = rcu_dereference(mm->mem_cgroup);
-	/*
-	 * For every charge from the cgroup, increment reference count
-	 */
-	css_get(&mem->css);
-	rcu_read_unlock();
+	if (memcg) {
+		mem = memcg;
+		css_get(&mem->css);
+	} else {
+		if (!mm)
+			mm = &init_mm;
+		rcu_read_lock();
+		mem = rcu_dereference(mm->mem_cgroup);
+		/*
+		 * For every charge from the cgroup, increment reference count
+		 */
+		css_get(&mem->css);
+		rcu_read_unlock();
+	}
 
 	while (res_counter_charge(&mem->res, PAGE_SIZE)) {
 		if (!(gfp_mask & __GFP_WAIT))
-			goto out;
+			goto nomem;
 
 		if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
 			continue;
@@ -600,52 +540,51 @@ retry:
 
 		if (!nr_retries--) {
 			mem_cgroup_out_of_memory(mem, gfp_mask);
-			goto out;
+			goto nomem;
 		}
 		congestion_wait(WRITE, HZ/10);
 	}
-
-	pc->ref_cnt = 1;
-	pc->mem_cgroup = mem;
-	pc->page = page;
-	pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
-	if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
-		pc->flags |= PAGE_CGROUP_FLAG_CACHE;
-
-	lock_page_cgroup(page);
-	if (page_get_page_cgroup(page)) {
-		unlock_page_cgroup(page);
-		/*
-		 * Another charge has been added to this page already.
-		 * We take lock_page_cgroup(page) again and read
-		 * page->cgroup, increment refcnt.... just retry is OK.
-		 */
+	/*
+	 * We have to acquire 2 spinlocks.
+	 */
+	spin_lock_irqsave(&pc->lock, flags);
+	/* Is anyone charged ? */
+	if (unlikely(pc->refcnt)) {
+		/* Someone charged this page while we released the lock */
+		pc->refcnt++;
+		spin_unlock_irqrestore(&pc->lock, flags);
 		res_counter_uncharge(&mem->res, PAGE_SIZE);
 		css_put(&mem->css);
-		kfree(pc);
-		goto retry;
+		goto success;
 	}
-	page_assign_page_cgroup(page, pc);
+	/* Anyone doesn't touch this. */
+	VM_BUG_ON(pc->mem_cgroup);
+
+	if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
+		pc->flags = PAGE_CGROUP_FLAG_ACTIVE | PAGE_CGROUP_FLAG_CACHE;
+	else
+		pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
+	pc->refcnt = 1;
+	pc->mem_cgroup = mem;
 
 	mz = page_cgroup_zoneinfo(pc);
-	spin_lock_irqsave(&mz->lru_lock, flags);
+
+	spin_lock(&mz->lru_lock);
 	__mem_cgroup_add_list(pc);
-	spin_unlock_irqrestore(&mz->lru_lock, flags);
+	spin_unlock(&mz->lru_lock);
+	spin_unlock_irqrestore(&pc->lock, flags);
 
-	unlock_page_cgroup(page);
-done:
+success:
 	return 0;
-out:
+nomem:
 	css_put(&mem->css);
-	kfree(pc);
-err:
 	return -ENOMEM;
 }
 
 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
 {
 	return mem_cgroup_charge_common(page, mm, gfp_mask,
-				MEM_CGROUP_CHARGE_TYPE_MAPPED);
+				MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
 }
 
 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
@@ -654,7 +593,7 @@ int mem_cgroup_cache_charge(struct page 
 	if (!mm)
 		mm = &init_mm;
 	return mem_cgroup_charge_common(page, mm, gfp_mask,
-				MEM_CGROUP_CHARGE_TYPE_CACHE);
+				MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
 }
 
 /*
@@ -664,105 +603,87 @@ int mem_cgroup_cache_charge(struct page 
 void mem_cgroup_uncharge_page(struct page *page)
 {
 	struct page_cgroup *pc;
-	struct mem_cgroup *mem;
-	struct mem_cgroup_per_zone *mz;
-	unsigned long flags;
 
 	if (mem_cgroup_subsys.disabled)
 		return;
-
 	/*
 	 * Check if our page_cgroup is valid
 	 */
-	lock_page_cgroup(page);
-	pc = page_get_page_cgroup(page);
-	if (!pc)
-		goto unlock;
-
-	VM_BUG_ON(pc->page != page);
-	VM_BUG_ON(pc->ref_cnt <= 0);
+	pc = get_page_cgroup(page);
+	if (likely(pc)) {
+		unsigned long flags;
+		struct mem_cgroup *mem;
+		struct mem_cgroup_per_zone *mz;
 
-	if (--(pc->ref_cnt) == 0) {
+		spin_lock_irqsave(&pc->lock, flags);
+		if (!pc->refcnt || --pc->refcnt > 0) {
+			spin_unlock_irqrestore(&pc->lock, flags);
+			return;
+		}
+		mem = pc->mem_cgroup;
 		mz = page_cgroup_zoneinfo(pc);
-		spin_lock_irqsave(&mz->lru_lock, flags);
+		spin_lock(&mz->lru_lock);
 		__mem_cgroup_remove_list(pc);
-		spin_unlock_irqrestore(&mz->lru_lock, flags);
-
-		page_assign_page_cgroup(page, NULL);
-		unlock_page_cgroup(page);
+		spin_unlock(&mz->lru_lock);
+		pc->flags = 0;
+		pc->mem_cgroup = 0;
+		spin_unlock_irqrestore(&pc->lock, flags);
 
-		mem = pc->mem_cgroup;
 		res_counter_uncharge(&mem->res, PAGE_SIZE);
 		css_put(&mem->css);
-
-		kfree(pc);
-		return;
 	}
-
-unlock:
-	unlock_page_cgroup(page);
 }
 
 /*
- * Returns non-zero if a page (under migration) has valid page_cgroup member.
- * Refcnt of page_cgroup is incremented.
+ * Pre-charge against newpage while moving a page.
+ * This function is called before taking page locks.
  */
-int mem_cgroup_prepare_migration(struct page *page)
+int mem_cgroup_prepare_migration(struct page *page, struct page *newpage)
 {
 	struct page_cgroup *pc;
+	struct mem_cgroup *mem = NULL;
+	int ret = 0;
+	enum charge_type type = 0;
+	unsigned long flags;
 
 	if (mem_cgroup_subsys.disabled)
-		return 0;
+		return ret;
 
-	lock_page_cgroup(page);
-	pc = page_get_page_cgroup(page);
-	if (pc)
-		pc->ref_cnt++;
-	unlock_page_cgroup(page);
-	return pc != NULL;
-}
+	/* check newpage isn't under memory resource control */
+	pc = get_page_cgroup(newpage);
+	VM_BUG_ON(pc && pc->refcnt);
 
-void mem_cgroup_end_migration(struct page *page)
-{
-	mem_cgroup_uncharge_page(page);
-}
+	pc = get_page_cgroup(page);
 
+	if (pc) {
+		spin_lock_irqsave(&pc->lock, flags);
+		if (pc->refcnt) {
+			mem = pc->mem_cgroup;
+			css_get(&mem->css);
+			if (pc->flags & PAGE_CGROUP_FLAG_CACHE)
+				type = MEM_CGROUP_CHARGE_TYPE_CACHE;
+			else
+				type = MEM_CGROUP_CHARGE_TYPE_MAPPED;
+		}
+		spin_unlock_irqrestore(&pc->lock, flags);
+		if (mem) {
+			ret = mem_cgroup_charge_common(newpage, NULL,
+				GFP_KERNEL, type, mem);
+			css_put(&mem->css);
+		}
+	}
+	return ret;
+}
 /*
- * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
- * And no race with uncharge() routines because page_cgroup for *page*
- * has extra one reference by mem_cgroup_prepare_migration.
+ * At the end of migration, we'll push newpage to LRU and
+ * drop one refcnt which added at prepare_migration.
  */
-void mem_cgroup_page_migration(struct page *page, struct page *newpage)
+void mem_cgroup_end_migration(struct page *newpage)
 {
-	struct page_cgroup *pc;
-	struct mem_cgroup_per_zone *mz;
-	unsigned long flags;
-
-	lock_page_cgroup(page);
-	pc = page_get_page_cgroup(page);
-	if (!pc) {
-		unlock_page_cgroup(page);
+	if (mem_cgroup_subsys.disabled)
 		return;
-	}
 
-	mz = page_cgroup_zoneinfo(pc);
-	spin_lock_irqsave(&mz->lru_lock, flags);
-	__mem_cgroup_remove_list(pc);
-	spin_unlock_irqrestore(&mz->lru_lock, flags);
-
-	page_assign_page_cgroup(page, NULL);
-	unlock_page_cgroup(page);
-
-	pc->page = newpage;
-	lock_page_cgroup(newpage);
-	page_assign_page_cgroup(newpage, pc);
-
-	mz = page_cgroup_zoneinfo(pc);
-	spin_lock_irqsave(&mz->lru_lock, flags);
-	__mem_cgroup_add_list(pc);
-	spin_unlock_irqrestore(&mz->lru_lock, flags);
-
-	unlock_page_cgroup(newpage);
+	mem_cgroup_uncharge_page(newpage);
 }
 
 /*
@@ -790,10 +711,13 @@ static void mem_cgroup_force_empty_list(
 	while (!list_empty(list)) {
 		pc = list_entry(list->prev, struct page_cgroup, lru);
 		page = pc->page;
-		get_page(page);
-		spin_unlock_irqrestore(&mz->lru_lock, flags);
-		mem_cgroup_uncharge_page(page);
-		put_page(page);
+		if (PageLRU(page)) {
+			get_page(page);
+			spin_unlock_irqrestore(&mz->lru_lock, flags);
+			mem_cgroup_uncharge_page(page);
+			put_page(page);
+		} else
+			count = 0;
 		if (--count <= 0) {
 			count = FORCE_UNCHARGE_BATCH;
 			cond_resched();
Index: linux-2.6.25-rc5-mm1-k/include/linux/memcontrol.h
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/include/linux/memcontrol.h
+++ linux-2.6.25-rc5-mm1-k/include/linux/memcontrol.h
@@ -19,6 +19,7 @@
 
 #ifndef _LINUX_MEMCONTROL_H
 #define _LINUX_MEMCONTROL_H
+#include <linux/page_cgroup.h>
 
 struct mem_cgroup;
 struct page_cgroup;
@@ -30,9 +31,6 @@ struct mm_struct;
 extern void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p);
 extern void mm_free_cgroup(struct mm_struct *mm);
 
-#define page_reset_bad_cgroup(page)	((page)->page_cgroup = 0)
-
-extern struct page_cgroup *page_get_page_cgroup(struct page *page);
 extern int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
 				gfp_t gfp_mask);
 extern int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
@@ -51,9 +49,8 @@ int task_in_mem_cgroup(struct task_struc
 #define mm_match_cgroup(mm, cgroup)	\
 	((cgroup) == rcu_dereference((mm)->mem_cgroup))
 
-extern int mem_cgroup_prepare_migration(struct page *page);
-extern void mem_cgroup_end_migration(struct page *page);
-extern void mem_cgroup_page_migration(struct page *page, struct page *newpage);
+extern int mem_cgroup_prepare_migration(struct page *, struct page *);
+extern void mem_cgroup_end_migration(struct page *);
 
 /*
  * For memory reclaim.
@@ -82,14 +79,6 @@ static inline void mm_free_cgroup(struct
 {
 }
 
-static inline void page_reset_bad_cgroup(struct page *page)
-{
-}
-
-static inline struct page_cgroup *page_get_page_cgroup(struct page *page)
-{
-	return NULL;
-}
 
 static inline int mem_cgroup_charge(struct page *page,
 					struct mm_struct *mm, gfp_t gfp_mask)
@@ -122,7 +111,8 @@ static inline int task_in_mem_cgroup(str
 	return 1;
 }
 
-static inline int mem_cgroup_prepare_migration(struct page *page)
+static inline int
+mem_cgroup_prepare_migration(struct page *page , struct page *newpage)
 {
 	return 0;
 }
Index: linux-2.6.25-rc5-mm1-k/mm/page_alloc.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/page_alloc.c
+++ linux-2.6.25-rc5-mm1-k/mm/page_alloc.c
@@ -222,17 +222,11 @@ static inline int bad_range(struct zone 
 
 static void bad_page(struct page *page)
 {
-	void *pc = page_get_page_cgroup(page);
-
 	printk(KERN_EMERG "Bad page state in process '%s'\n" KERN_EMERG
 		"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
 		current->comm, page, (int)(2*sizeof(unsigned long)),
 		(unsigned long)page->flags, page->mapping,
 		page_mapcount(page), page_count(page));
-	if (pc) {
-		printk(KERN_EMERG "cgroup:%p\n", pc);
-		page_reset_bad_cgroup(page);
-	}
 	printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
 		KERN_EMERG "Backtrace:\n");
 	dump_stack();
@@ -478,7 +472,6 @@ static inline int free_pages_check(struc
 {
 	if (unlikely(page_mapcount(page) |
 		(page->mapping != NULL)  |
-		(page_get_page_cgroup(page) != NULL) |
 		(page_count(page) != 0)  |
 		(page->flags & (
 			1 << PG_lru	|
@@ -628,7 +621,6 @@ static int prep_new_page(struct page *pa
 {
 	if (unlikely(page_mapcount(page) |
 		(page->mapping != NULL)  |
-		(page_get_page_cgroup(page) != NULL) |
 		(page_count(page) != 0)  |
 		(page->flags & (
 			1 << PG_lru	|
Index: linux-2.6.25-rc5-mm1-k/include/linux/mm_types.h
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/include/linux/mm_types.h
+++ linux-2.6.25-rc5-mm1-k/include/linux/mm_types.h
@@ -88,9 +88,6 @@ struct page {
 	void *virtual;			/* Kernel virtual address (NULL if
 					   not kmapped, ie. highmem) */
 #endif /* WANT_PAGE_VIRTUAL */
-#ifdef CONFIG_CGROUP_MEM_RES_CTLR
-	unsigned long page_cgroup;
-#endif
 #ifdef CONFIG_PAGE_OWNER
 	int order;
 	unsigned int gfp_mask;
Index: linux-2.6.25-rc5-mm1-k/mm/migrate.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/migrate.c
+++ linux-2.6.25-rc5-mm1-k/mm/migrate.c
@@ -358,6 +358,12 @@ static int migrate_page_move_mapping(str
 
 	write_unlock_irq(&mapping->tree_lock);
 
+	/* by mem_cgroup_prepare_migration, newpage is already
+	   assigned to valid cgroup. and current->mm and GFP_ATOMIC
+	   will not be used...*/
+	mem_cgroup_uncharge_page(page);
+	mem_cgroup_cache_charge(newpage, current->mm, GFP_ATOMIC);
+
 	return 0;
 }
 
@@ -603,7 +609,6 @@ static int move_to_new_page(struct page 
 		rc = fallback_migrate_page(mapping, newpage, page);
 
 	if (!rc) {
-		mem_cgroup_page_migration(page, newpage);
 		remove_migration_ptes(page, newpage);
 	} else
 		newpage->mapping = NULL;
@@ -633,6 +638,12 @@ static int unmap_and_move(new_page_t get
 		/* page was freed from under us. So we are done. */
 		goto move_newpage;
 
+	charge = mem_cgroup_prepare_migration(page, newpage);
+	if (charge == -ENOMEM) {
+		rc = -ENOMEM;
+		goto move_newpage;
+	}
+
 	rc = -EAGAIN;
 	if (TestSetPageLocked(page)) {
 		if (!force)
@@ -684,19 +695,14 @@ static int unmap_and_move(new_page_t get
 		goto rcu_unlock;
 	}
 
-	charge = mem_cgroup_prepare_migration(page);
 	/* Establish migration ptes or remove ptes */
 	try_to_unmap(page, 1);
 
 	if (!page_mapped(page))
 		rc = move_to_new_page(newpage, page);
 
-	if (rc) {
+	if (rc)
 		remove_migration_ptes(page, page);
-		if (charge)
-			mem_cgroup_end_migration(page);
-	} else if (charge)
- 		mem_cgroup_end_migration(newpage);
 rcu_unlock:
 	if (rcu_locked)
 		rcu_read_unlock();
@@ -717,6 +723,8 @@ unlock:
 	}
 
 move_newpage:
+	if (!charge)
+		mem_cgroup_end_migration(newpage);
 	/*
 	 * Move the new page to the LRU. If migration was not successful
 	 * then this will free the page.
Index: linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
===================================================================
--- /dev/null
+++ linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
@@ -0,0 +1,260 @@
+/*
+ * per-page accounting subsystem infrastructure. - linux/mm/page_cgroup.c
+ *
+ * (C) 2008 FUJITSU, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * page_cgroup is yet another mem_map under memory resoruce controller.
+ * It containes information which cannot be stored in usual mem_map.
+ * This allows us to keep 'struct page' small when a user doesn't activate
+ * memory resource controller.
+ *
+ * We can translate : struct page <-> pfn -> page_cgroup -> struct page.
+ *
+ */
+
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/radix-tree.h>
+#include <linux/memcontrol.h>
+#include <linux/err.h>
+
+static int page_cgroup_order __read_mostly;
+static int page_cgroup_head_size __read_mostly;
+
+#define PCGRP_SHIFT	(page_cgroup_order)
+#define PCGRP_SIZE	(1 << PCGRP_SHIFT)
+#define PCGRP_MASK	(PCGRP_SIZE - 1)
+
+struct page_cgroup_head {
+	struct page_cgroup pc[0];
+};
+
+struct page_cgroup_root {
+	spinlock_t	       tree_lock;
+	struct radix_tree_root root_node;
+};
+
+/*
+ * Calculate page_cgroup order to be not larger than order-2 page allocation.
+ */
+static void calc_page_cgroup_order(void)
+{
+	int order = pageblock_order;
+	unsigned long size = sizeof(struct page_cgroup) << order;
+
+	while (size > PAGE_SIZE * 2) {
+		order -= 1;
+		size = sizeof(struct page_cgroup) << order;
+	}
+
+	page_cgroup_order = order;
+	page_cgroup_head_size = sizeof(struct page_cgroup_head) +
+				(sizeof(struct page_cgroup) << order);
+}
+
+static struct page_cgroup_root __initdata *tmp_root_dir[MAX_NUMNODES];
+static struct page_cgroup_root *root_node[MAX_NUMNODES] __read_mostly;
+
+static void
+init_page_cgroup_head(struct page_cgroup_head *head, unsigned long pfn)
+{
+	struct page *page;
+	struct page_cgroup *pc;
+	int i;
+
+	for (i = 0, page = pfn_to_page(pfn), pc = &head->pc[0];
+		i < PCGRP_SIZE; i++, page++, pc++) {
+		pc->refcnt = 0;
+		pc->page = page;
+		spin_lock_init(&pc->lock);
+	}
+}
+
+
+struct kmem_cache *page_cgroup_cachep;
+
+static struct page_cgroup_head *
+alloc_page_cgroup_head(unsigned long pfn, int nid, gfp_t mask)
+{
+	struct page_cgroup_head *head;
+
+	if (!node_state(nid, N_NORMAL_MEMORY))
+		nid = -1;
+	head = kmem_cache_alloc_node(page_cgroup_cachep, mask, nid);
+	if (head)
+		init_page_cgroup_head(head, pfn);
+
+	return head;
+}
+
+void free_page_cgroup(struct page_cgroup_head *head)
+{
+	kmem_cache_free(page_cgroup_cachep, head);
+}
+
+static struct page_cgroup_root *pcgroup_get_root(struct page *page)
+{
+	int nid;
+
+	VM_BUG_ON(!page);
+
+	nid = page_to_nid(page);
+
+	return root_node[nid];
+}
+
+/**
+ * get_page_cgroup - look up a page_cgroup for a page
+ * @page: the page whose page_cgroup is looked up.
+ *
+ * This just does lookup.
+ */
+struct page_cgroup *get_page_cgroup(struct page *page)
+{
+	struct page_cgroup_head *head;
+	struct page_cgroup_root *root;
+	struct page_cgroup *ret = NULL;
+	unsigned long pfn, idx;
+
+	/*
+	 * NULL can be returned before initialization
+	 */
+	root = pcgroup_get_root(page);
+	if (unlikely(!root))
+		return ret;
+
+	pfn = page_to_pfn(page);
+	idx = pfn >> PCGRP_SHIFT;
+	/*
+	 * We don't need lock here because no one deletes this head.
+	 * (Freeing routtine will be added later.)
+	 */
+	rcu_read_lock();
+	head = radix_tree_lookup(&root->root_node, idx);
+	rcu_read_unlock();
+
+	if (likely(head))
+		ret = &head->pc[pfn & PCGRP_MASK];
+
+	return ret;
+}
+
+/**
+ * get_alloc_page_cgroup - look up or allocate a page_cgroup for a page
+ * @page: the page whose page_cgroup is looked up.
+ * @gfpmask: the gfpmask which will be used for page allocatiopn.
+ *
+ * look up and allocate if not found.
+ */
+
+struct page_cgroup *
+get_alloc_page_cgroup(struct page *page, gfp_t gfpmask)
+{
+	struct page_cgroup_root *root;
+	struct page_cgroup_head *head;
+	struct page_cgroup *pc;
+	unsigned long pfn, idx;
+	int nid;
+	unsigned long base_pfn, flags;
+	int error = 0;
+
+	might_sleep_if(gfpmask & __GFP_WAIT);
+
+retry:
+	pc = get_page_cgroup(page);
+	if (pc)
+		return pc;
+	/*
+	 * NULL can be returned before initialization.
+	 */
+	root = pcgroup_get_root(page);
+	if (unlikely(!root))
+		return NULL;
+
+	pfn = page_to_pfn(page);
+	idx = pfn >> PCGRP_SHIFT;
+	nid = page_to_nid(page);
+	base_pfn = idx << PCGRP_SHIFT;
+
+	gfpmask = gfpmask & ~(__GFP_HIGHMEM | __GFP_MOVABLE);
+
+	head = alloc_page_cgroup_head(base_pfn, nid, gfpmask);
+	if (!head)
+		return ERR_PTR(-ENOMEM);
+
+	pc = &head->pc[pfn & PCGRP_MASK];
+
+	error = radix_tree_preload(gfpmask);
+	if (error)
+		goto out;
+	spin_lock_irqsave(&root->tree_lock, flags);
+	error = radix_tree_insert(&root->root_node, idx, head);
+	spin_unlock_irqrestore(&root->tree_lock, flags);
+	radix_tree_preload_end();
+out:
+	if (error) {
+		free_page_cgroup(head);
+		if (error == -EEXIST)
+			goto retry;
+		pc = ERR_PTR(error);
+	}
+	return pc;
+}
+
+__init int page_cgroup_init(void)
+{
+	int tmp, nid;
+	struct page_cgroup_root *root;
+
+	calc_page_cgroup_order();
+
+	page_cgroup_cachep = kmem_cache_create("page_cgroup",
+				page_cgroup_head_size, 0,
+				SLAB_PANIC | SLAB_DESTROY_BY_RCU, NULL);
+
+	if (!page_cgroup_cachep) {
+		printk(KERN_ERR "page accouning setup failure\n");
+		printk(KERN_ERR "can't initialize slab memory\n");
+		return -ENOMEM;
+	}
+
+	for_each_node(nid) {
+		tmp = nid;
+		if (!node_state(nid, N_NORMAL_MEMORY))
+			tmp = -1;
+		root = kmalloc_node(sizeof(struct page_cgroup_root),
+					GFP_KERNEL, tmp);
+		if (!root)
+			goto unroll;
+		INIT_RADIX_TREE(&root->root_node, GFP_ATOMIC);
+		spin_lock_init(&root->tree_lock);
+		tmp_root_dir[nid] = root;
+	}
+	/*
+	 * By filling node_root[], this tree turns to be visible.
+	 * Because we have to finish initialization of the tree before
+	 * we make it visible, memory barrier is necessary.
+	 */
+	smp_wmb();
+	for_each_node(nid)
+		root_node[nid] = tmp_root_dir[nid];
+
+	printk(KERN_INFO "Page Accouintg is activated\n");
+	return 0;
+unroll:
+	for_each_node(nid)
+		kfree(tmp_root_dir[nid]);
+
+	return -ENOMEM;
+}
+late_initcall(page_cgroup_init);
Index: linux-2.6.25-rc5-mm1-k/mm/Makefile
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/Makefile
+++ linux-2.6.25-rc5-mm1-k/mm/Makefile
@@ -32,5 +32,5 @@ obj-$(CONFIG_FS_XIP) += filemap_xip.o
 obj-$(CONFIG_MIGRATION) += migrate.o
 obj-$(CONFIG_SMP) += allocpercpu.o
 obj-$(CONFIG_QUICKLIST) += quicklist.o
-obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o
+obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o
 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [-mm] [PATCH 1/4] memcg : radix-tree page_cgroup

[Li Zefan]

KAMEZAWA Hiroyuki wrote:
> This patch implements radixt-tree based page cgroup.
> 
> This patch does
>  * add radix-tree based page_cgroup look up subsystem.page_cgroup_init
>  * remove bit_spin_lock used by page_cgroup.
> 
> changes:
> 
> Before patch
>  * struct page had pointer to page_cgroup. Then, relationship between objects
>    was pfn <-> struct page <-> struct page_cgroup
>  * (spin) lock for page_cgroup was in struct page.
>  * page_cgroup->refcnt is incremented before charge is done.
>  * page migration does complicated page_cgroup migration under locks.
> 
> After patch
>  * struct page has no pointer to page_cgroup. Relationship between objects
>    is struct page <-> pfn <-> struct page_cgroup -> struct page,
>  * page_cgroup has its own spin lock.
>  * page_cgroup->refcnt is incremented after charge is done.
>  * page migration accounts a new page before migration. By this, we can
>    avoid complicated locks. 
> 
> tested on ia64/NUMA, x86_64/SMP.
> 
> Changelog v1 -> v2:
>  * create a folded patch. maybe good for bysect.
>  * removed special handling codes for new pages under migration
>    Added PG_LRU check to force_empty.
>  * reflected comments.
>  * Added comments in the head of page_cgroup.c
>  * order of page_cgroup is automatically calculated.
>  * fixed handling of root_node[] entries in page_cgroup_init().
>  * rewrite init_page_cgroup_head() to do minimum work.
>  * fixed N_NORMAL_MEMORY handling.
>  
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> 

At a first glance, some typos and small things, see below ;)

> 
>  include/linux/memcontrol.h  |   17 --
>  include/linux/mm_types.h    |    3 
>  include/linux/page_cgroup.h |   56 +++++++
>  mm/Makefile                 |    2 
>  mm/memcontrol.c             |  332 ++++++++++++++++----------------------------
>  mm/migrate.c                |   22 +-
>  mm/page_alloc.c             |    8 -
>  mm/page_cgroup.c            |  260 ++++++++++++++++++++++++++++++++++
>  8 files changed, 463 insertions(+), 237 deletions(-)
> 
> Index: linux-2.6.25-rc5-mm1-k/include/linux/page_cgroup.h
> ===================================================================
> --- /dev/null
> +++ linux-2.6.25-rc5-mm1-k/include/linux/page_cgroup.h
> @@ -0,0 +1,56 @@
> +#ifndef __LINUX_PAGE_CGROUP_H
> +#define __LINUX_PAGE_CGROUP_H
> +
> +#ifdef CONFIG_CGROUP_MEM_RES_CTLR
> +/*
> + * page_cgroup is yet another mem_map structure for accounting  usage.

Not fix yes:
						     extra space ^^

> + * but, unlike mem_map, allocated on demand for accounted pages.
> + * see also memcontrol.h
> + * In nature, this consumes much amount of memory.
> + */
> +
> +struct mem_cgroup;
> +
> +struct page_cgroup {
> +	spinlock_t		lock;        /* lock for all members */
> +	int  			refcnt;      /* reference count */
> +	struct mem_cgroup 	*mem_cgroup; /* current cgroup subsys */
> +	struct list_head 	lru;         /* for per cgroup LRU */
> +	int    			flags;	     /* See below */
> +	struct page 		*page;       /* the page this accounts for*/
> +};
> +
> +/* flags */
> +#define PAGE_CGROUP_FLAG_CACHE	(0x1)	/* charged as cache. */
> +#define PAGE_CGROUP_FLAG_ACTIVE (0x2)	/* is on active list */
> +
> +/*
> + * look up page_cgroup. returns NULL if not exists.
> + */
> +extern struct page_cgroup *get_page_cgroup(struct page *page);
> +
> +
> +/*
> + * look up page_cgroup, allocate new one if it doesn't exist.
> + * Return value is
> + *   1. page_cgroup, at success.
> + *   2. -EXXXXX, at failure.
> + *   3. NULL, at boot.
> + */
> +extern struct page_cgroup *
> +get_alloc_page_cgroup(struct page *page, gfp_t gfpmask);
> +
> +#else
> +
> +static inline struct page_cgroup *get_page_cgroup(struct page *page)
> +{
> +	return NULL;
> +}
> +
> +static inline struct page_cgroup *
> +get_alloc_page_cgroup(struct page *page, gfp_t gfpmask)
> +{
> +	return NULL;
> +}
> +#endif
> +#endif
> Index: linux-2.6.25-rc5-mm1-k/mm/memcontrol.c
> ===================================================================
> --- linux-2.6.25-rc5-mm1-k.orig/mm/memcontrol.c
> +++ linux-2.6.25-rc5-mm1-k/mm/memcontrol.c
> @@ -30,6 +30,7 @@
>  #include <linux/spinlock.h>
>  #include <linux/fs.h>
>  #include <linux/seq_file.h>
> +#include <linux/page_cgroup.h>
>  
>  #include <asm/uaccess.h>
>  
> @@ -92,7 +93,7 @@ struct mem_cgroup_per_zone {
>  	/*
>  	 * spin_lock to protect the per cgroup LRU
>  	 */
> -	spinlock_t		lru_lock;
> +	spinlock_t		lru_lock;	/* irq should be off. */
>  	struct list_head	active_list;
>  	struct list_head	inactive_list;
>  	unsigned long count[NR_MEM_CGROUP_ZSTAT];
> @@ -139,33 +140,6 @@ struct mem_cgroup {
>  };
>  static struct mem_cgroup init_mem_cgroup;
>  
> -/*
> - * We use the lower bit of the page->page_cgroup pointer as a bit spin
> - * lock.  We need to ensure that page->page_cgroup is at least two
> - * byte aligned (based on comments from Nick Piggin).  But since
> - * bit_spin_lock doesn't actually set that lock bit in a non-debug
> - * uniprocessor kernel, we should avoid setting it here too.
> - */
> -#define PAGE_CGROUP_LOCK_BIT 	0x0
> -#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
> -#define PAGE_CGROUP_LOCK 	(1 << PAGE_CGROUP_LOCK_BIT)
> -#else
> -#define PAGE_CGROUP_LOCK	0x0
> -#endif
> -
> -/*
> - * A page_cgroup page is associated with every page descriptor. The
> - * page_cgroup helps us identify information about the cgroup
> - */
> -struct page_cgroup {
> -	struct list_head lru;		/* per cgroup LRU list */
> -	struct page *page;
> -	struct mem_cgroup *mem_cgroup;
> -	int ref_cnt;			/* cached, mapped, migrating */
> -	int flags;
> -};
> -#define PAGE_CGROUP_FLAG_CACHE	(0x1)	/* charged as cache */
> -#define PAGE_CGROUP_FLAG_ACTIVE (0x2)	/* page is active in this cgroup */
>  
>  static int page_cgroup_nid(struct page_cgroup *pc)
>  {
> @@ -256,37 +230,6 @@ void mm_free_cgroup(struct mm_struct *mm
>  	css_put(&mm->mem_cgroup->css);
>  }
>  
> -static inline int page_cgroup_locked(struct page *page)
> -{
> -	return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
> -}
> -
> -static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
> -{
> -	VM_BUG_ON(!page_cgroup_locked(page));
> -	page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
> -}
> -
> -struct page_cgroup *page_get_page_cgroup(struct page *page)
> -{
> -	return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
> -}
> -
> -static void lock_page_cgroup(struct page *page)
> -{
> -	bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
> -}
> -
> -static int try_lock_page_cgroup(struct page *page)
> -{
> -	return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
> -}
> -
> -static void unlock_page_cgroup(struct page *page)
> -{
> -	bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
> -}
> -
>  static void __mem_cgroup_remove_list(struct page_cgroup *pc)
>  {
>  	int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
> @@ -356,6 +299,10 @@ void mem_cgroup_move_lists(struct page *
>  	struct mem_cgroup_per_zone *mz;
>  	unsigned long flags;
>  
> +	/* This GFP will be ignored..*/
> +	pc = get_page_cgroup(page);
> +	if (!pc)
> +		return;
>  	/*
>  	 * We cannot lock_page_cgroup while holding zone's lru_lock,
>  	 * because other holders of lock_page_cgroup can be interrupted
> @@ -363,17 +310,15 @@ void mem_cgroup_move_lists(struct page *
>  	 * safely get to page_cgroup without it, so just try_lock it:
>  	 * mem_cgroup_isolate_pages allows for page left on wrong list.
>  	 */
> -	if (!try_lock_page_cgroup(page))
> +	if (!spin_trylock_irqsave(&pc->lock, flags))
>  		return;
> -
> -	pc = page_get_page_cgroup(page);
> -	if (pc) {
> +	if (pc->refcnt) {
>  		mz = page_cgroup_zoneinfo(pc);
> -		spin_lock_irqsave(&mz->lru_lock, flags);
> +		spin_lock(&mz->lru_lock);
>  		__mem_cgroup_move_lists(pc, active);
> -		spin_unlock_irqrestore(&mz->lru_lock, flags);
> +		spin_unlock(&mz->lru_lock);
>  	}
> -	unlock_page_cgroup(page);
> +	spin_unlock_irqrestore(&pc->lock, flags);
>  }
>  
>  /*
> @@ -525,7 +470,8 @@ unsigned long mem_cgroup_isolate_pages(u
>   * < 0 if the cgroup is over its limit
>   */
>  static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
> -				gfp_t gfp_mask, enum charge_type ctype)
> +				gfp_t gfp_mask, enum charge_type ctype,
> +				struct mem_cgroup *memcg)
>  {
>  	struct mem_cgroup *mem;
>  	struct page_cgroup *pc;
> @@ -536,33 +482,23 @@ static int mem_cgroup_charge_common(stru
>  	if (mem_cgroup_subsys.disabled)
>  		return 0;
>  
> +	pc = get_alloc_page_cgroup(page, gfp_mask);
> +	/* Before kamalloc initialization, get_page_cgroup can retrun NULL */

		  kmalloc				       return

> +	if (unlikely(!pc || IS_ERR(pc)))
> +		return PTR_ERR(pc);
> +
> +	spin_lock_irqsave(&pc->lock, flags);
>  	/*
> -	 * Should page_cgroup's go to their own slab?
> -	 * One could optimize the performance of the charging routine
> -	 * by saving a bit in the page_flags and using it as a lock
> -	 * to see if the cgroup page already has a page_cgroup associated
> -	 * with it
> -	 */
> -retry:
> -	lock_page_cgroup(page);
> -	pc = page_get_page_cgroup(page);
> -	/*
> -	 * The page_cgroup exists and
> -	 * the page has already been accounted.
> +	 * Has the page already been accounted ?
>  	 */
> -	if (pc) {
> -		VM_BUG_ON(pc->page != page);
> -		VM_BUG_ON(pc->ref_cnt <= 0);
> -
> -		pc->ref_cnt++;
> -		unlock_page_cgroup(page);
> -		goto done;
> +	if (pc->refcnt > 0) {
> +		pc->refcnt++;
> +		spin_unlock_irqrestore(&pc->lock, flags);
> +		goto success;
>  	}
> -	unlock_page_cgroup(page);
> +	spin_unlock_irqrestore(&pc->lock, flags);
>  
> -	pc = kzalloc(sizeof(struct page_cgroup), gfp_mask);
> -	if (pc == NULL)
> -		goto err;
> +	/* Note: *new* pc's refcnt is still 0 here. */
>  
>  	/*
>  	 * We always charge the cgroup the mm_struct belongs to.
> @@ -570,20 +506,24 @@ retry:
>  	 * thread group leader migrates. It's possible that mm is not
>  	 * set, if so charge the init_mm (happens for pagecache usage).
>  	 */
> -	if (!mm)
> -		mm = &init_mm;
> -
> -	rcu_read_lock();
> -	mem = rcu_dereference(mm->mem_cgroup);
> -	/*
> -	 * For every charge from the cgroup, increment reference count
> -	 */
> -	css_get(&mem->css);
> -	rcu_read_unlock();
> +	if (memcg) {
> +		mem = memcg;
> +		css_get(&mem->css);
> +	} else {
> +		if (!mm)
> +			mm = &init_mm;
> +		rcu_read_lock();
> +		mem = rcu_dereference(mm->mem_cgroup);
> +		/*
> +		 * For every charge from the cgroup, increment reference count
> +		 */
> +		css_get(&mem->css);
> +		rcu_read_unlock();
> +	}
>  
>  	while (res_counter_charge(&mem->res, PAGE_SIZE)) {
>  		if (!(gfp_mask & __GFP_WAIT))
> -			goto out;
> +			goto nomem;
>  
>  		if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
>  			continue;
> @@ -600,52 +540,51 @@ retry:
>  
>  		if (!nr_retries--) {
>  			mem_cgroup_out_of_memory(mem, gfp_mask);
> -			goto out;
> +			goto nomem;
>  		}
>  		congestion_wait(WRITE, HZ/10);
>  	}
> -
> -	pc->ref_cnt = 1;
> -	pc->mem_cgroup = mem;
> -	pc->page = page;
> -	pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
> -	if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
> -		pc->flags |= PAGE_CGROUP_FLAG_CACHE;
> -
> -	lock_page_cgroup(page);
> -	if (page_get_page_cgroup(page)) {
> -		unlock_page_cgroup(page);
> -		/*
> -		 * Another charge has been added to this page already.
> -		 * We take lock_page_cgroup(page) again and read
> -		 * page->cgroup, increment refcnt.... just retry is OK.
> -		 */
> +	/*
> +	 * We have to acquire 2 spinlocks.
> +	 */
> +	spin_lock_irqsave(&pc->lock, flags);
> +	/* Is anyone charged ? */
> +	if (unlikely(pc->refcnt)) {
> +		/* Someone charged this page while we released the lock */
> +		pc->refcnt++;
> +		spin_unlock_irqrestore(&pc->lock, flags);
>  		res_counter_uncharge(&mem->res, PAGE_SIZE);
>  		css_put(&mem->css);
> -		kfree(pc);
> -		goto retry;
> +		goto success;
>  	}
> -	page_assign_page_cgroup(page, pc);
> +	/* Anyone doesn't touch this. */
> +	VM_BUG_ON(pc->mem_cgroup);
> +
> +	if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
> +		pc->flags = PAGE_CGROUP_FLAG_ACTIVE | PAGE_CGROUP_FLAG_CACHE;
> +	else
> +		pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
> +	pc->refcnt = 1;
> +	pc->mem_cgroup = mem;
>  
>  	mz = page_cgroup_zoneinfo(pc);
> -	spin_lock_irqsave(&mz->lru_lock, flags);
> +
> +	spin_lock(&mz->lru_lock);
>  	__mem_cgroup_add_list(pc);
> -	spin_unlock_irqrestore(&mz->lru_lock, flags);
> +	spin_unlock(&mz->lru_lock);
> +	spin_unlock_irqrestore(&pc->lock, flags);
>  
> -	unlock_page_cgroup(page);
> -done:
> +success:
>  	return 0;
> -out:
> +nomem:
>  	css_put(&mem->css);
> -	kfree(pc);
> -err:
>  	return -ENOMEM;
>  }
>  
>  int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
>  {
>  	return mem_cgroup_charge_common(page, mm, gfp_mask,
> -				MEM_CGROUP_CHARGE_TYPE_MAPPED);
> +				MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
>  }
>  
>  int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
> @@ -654,7 +593,7 @@ int mem_cgroup_cache_charge(struct page 
>  	if (!mm)
>  		mm = &init_mm;
>  	return mem_cgroup_charge_common(page, mm, gfp_mask,
> -				MEM_CGROUP_CHARGE_TYPE_CACHE);
> +				MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
>  }
>  
>  /*
> @@ -664,105 +603,87 @@ int mem_cgroup_cache_charge(struct page 
>  void mem_cgroup_uncharge_page(struct page *page)
>  {
>  	struct page_cgroup *pc;
> -	struct mem_cgroup *mem;
> -	struct mem_cgroup_per_zone *mz;
> -	unsigned long flags;
>  
>  	if (mem_cgroup_subsys.disabled)
>  		return;
> -
>  	/*
>  	 * Check if our page_cgroup is valid
>  	 */
> -	lock_page_cgroup(page);
> -	pc = page_get_page_cgroup(page);
> -	if (!pc)
> -		goto unlock;
> -
> -	VM_BUG_ON(pc->page != page);
> -	VM_BUG_ON(pc->ref_cnt <= 0);
> +	pc = get_page_cgroup(page);
> +	if (likely(pc)) {
> +		unsigned long flags;
> +		struct mem_cgroup *mem;
> +		struct mem_cgroup_per_zone *mz;
>  
> -	if (--(pc->ref_cnt) == 0) {
> +		spin_lock_irqsave(&pc->lock, flags);
> +		if (!pc->refcnt || --pc->refcnt > 0) {
> +			spin_unlock_irqrestore(&pc->lock, flags);
> +			return;
> +		}
> +		mem = pc->mem_cgroup;
>  		mz = page_cgroup_zoneinfo(pc);
> -		spin_lock_irqsave(&mz->lru_lock, flags);
> +		spin_lock(&mz->lru_lock);
>  		__mem_cgroup_remove_list(pc);
> -		spin_unlock_irqrestore(&mz->lru_lock, flags);
> -
> -		page_assign_page_cgroup(page, NULL);
> -		unlock_page_cgroup(page);
> +		spin_unlock(&mz->lru_lock);
> +		pc->flags = 0;
> +		pc->mem_cgroup = 0;
> +		spin_unlock_irqrestore(&pc->lock, flags);
>  
> -		mem = pc->mem_cgroup;
>  		res_counter_uncharge(&mem->res, PAGE_SIZE);
>  		css_put(&mem->css);
> -
> -		kfree(pc);
> -		return;
>  	}
> -
> -unlock:
> -	unlock_page_cgroup(page);
>  }
>  
>  /*
> - * Returns non-zero if a page (under migration) has valid page_cgroup member.
> - * Refcnt of page_cgroup is incremented.
> + * Pre-charge against newpage while moving a page.
> + * This function is called before taking page locks.
>   */
> -int mem_cgroup_prepare_migration(struct page *page)
> +int mem_cgroup_prepare_migration(struct page *page, struct page *newpage)
>  {
>  	struct page_cgroup *pc;
> +	struct mem_cgroup *mem = NULL;
> +	int ret = 0;
> +	enum charge_type type = 0;

don't initialize an enum variable with integer value.

> +	unsigned long flags;
>  
>  	if (mem_cgroup_subsys.disabled)
> -		return 0;
> +		return ret;
>  
> -	lock_page_cgroup(page);
> -	pc = page_get_page_cgroup(page);
> -	if (pc)
> -		pc->ref_cnt++;
> -	unlock_page_cgroup(page);
> -	return pc != NULL;
> -}
> +	/* check newpage isn't under memory resource control */
> +	pc = get_page_cgroup(newpage);
> +	VM_BUG_ON(pc && pc->refcnt);
>  
> -void mem_cgroup_end_migration(struct page *page)
> -{
> -	mem_cgroup_uncharge_page(page);
> -}
> +	pc = get_page_cgroup(page);
>  
> +	if (pc) {
> +		spin_lock_irqsave(&pc->lock, flags);
> +		if (pc->refcnt) {
> +			mem = pc->mem_cgroup;
> +			css_get(&mem->css);
> +			if (pc->flags & PAGE_CGROUP_FLAG_CACHE)
> +				type = MEM_CGROUP_CHARGE_TYPE_CACHE;
> +			else
> +				type = MEM_CGROUP_CHARGE_TYPE_MAPPED;
> +		}
> +		spin_unlock_irqrestore(&pc->lock, flags);
> +		if (mem) {
> +			ret = mem_cgroup_charge_common(newpage, NULL,
> +				GFP_KERNEL, type, mem);
> +			css_put(&mem->css);
> +		}
> +	}
> +	return ret;
> +}
>  /*
> - * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
> - * And no race with uncharge() routines because page_cgroup for *page*
> - * has extra one reference by mem_cgroup_prepare_migration.
> + * At the end of migration, we'll push newpage to LRU and
> + * drop one refcnt which added at prepare_migration.
>   */
> -void mem_cgroup_page_migration(struct page *page, struct page *newpage)
> +void mem_cgroup_end_migration(struct page *newpage)
>  {
> -	struct page_cgroup *pc;
> -	struct mem_cgroup_per_zone *mz;
> -	unsigned long flags;
> -
> -	lock_page_cgroup(page);
> -	pc = page_get_page_cgroup(page);
> -	if (!pc) {
> -		unlock_page_cgroup(page);
> +	if (mem_cgroup_subsys.disabled)
>  		return;
> -	}
>  
> -	mz = page_cgroup_zoneinfo(pc);
> -	spin_lock_irqsave(&mz->lru_lock, flags);
> -	__mem_cgroup_remove_list(pc);
> -	spin_unlock_irqrestore(&mz->lru_lock, flags);
> -
> -	page_assign_page_cgroup(page, NULL);
> -	unlock_page_cgroup(page);
> -
> -	pc->page = newpage;
> -	lock_page_cgroup(newpage);
> -	page_assign_page_cgroup(newpage, pc);
> -
> -	mz = page_cgroup_zoneinfo(pc);
> -	spin_lock_irqsave(&mz->lru_lock, flags);
> -	__mem_cgroup_add_list(pc);
> -	spin_unlock_irqrestore(&mz->lru_lock, flags);
> -
> -	unlock_page_cgroup(newpage);
> +	mem_cgroup_uncharge_page(newpage);
>  }
>  
>  /*
> @@ -790,10 +711,13 @@ static void mem_cgroup_force_empty_list(
>  	while (!list_empty(list)) {
>  		pc = list_entry(list->prev, struct page_cgroup, lru);
>  		page = pc->page;
> -		get_page(page);
> -		spin_unlock_irqrestore(&mz->lru_lock, flags);
> -		mem_cgroup_uncharge_page(page);
> -		put_page(page);
> +		if (PageLRU(page)) {
> +			get_page(page);
> +			spin_unlock_irqrestore(&mz->lru_lock, flags);
> +			mem_cgroup_uncharge_page(page);
> +			put_page(page);
> +		} else
> +			count = 0;
>  		if (--count <= 0) {
>  			count = FORCE_UNCHARGE_BATCH;
>  			cond_resched();
> Index: linux-2.6.25-rc5-mm1-k/include/linux/memcontrol.h
> ===================================================================
> --- linux-2.6.25-rc5-mm1-k.orig/include/linux/memcontrol.h
> +++ linux-2.6.25-rc5-mm1-k/include/linux/memcontrol.h
> @@ -19,6 +19,7 @@
>  
>  #ifndef _LINUX_MEMCONTROL_H
>  #define _LINUX_MEMCONTROL_H
> +#include <linux/page_cgroup.h>
>  
>  struct mem_cgroup;
>  struct page_cgroup;
> @@ -30,9 +31,6 @@ struct mm_struct;
>  extern void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p);
>  extern void mm_free_cgroup(struct mm_struct *mm);
>  
> -#define page_reset_bad_cgroup(page)	((page)->page_cgroup = 0)
> -
> -extern struct page_cgroup *page_get_page_cgroup(struct page *page);
>  extern int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
>  				gfp_t gfp_mask);
>  extern int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
> @@ -51,9 +49,8 @@ int task_in_mem_cgroup(struct task_struc
>  #define mm_match_cgroup(mm, cgroup)	\
>  	((cgroup) == rcu_dereference((mm)->mem_cgroup))
>  
> -extern int mem_cgroup_prepare_migration(struct page *page);
> -extern void mem_cgroup_end_migration(struct page *page);
> -extern void mem_cgroup_page_migration(struct page *page, struct page *newpage);
> +extern int mem_cgroup_prepare_migration(struct page *, struct page *);
> +extern void mem_cgroup_end_migration(struct page *);
>  
>  /*
>   * For memory reclaim.
> @@ -82,14 +79,6 @@ static inline void mm_free_cgroup(struct
>  {
>  }
>  
> -static inline void page_reset_bad_cgroup(struct page *page)
> -{
> -}
> -
> -static inline struct page_cgroup *page_get_page_cgroup(struct page *page)
> -{
> -	return NULL;
> -}
>  
>  static inline int mem_cgroup_charge(struct page *page,
>  					struct mm_struct *mm, gfp_t gfp_mask)
> @@ -122,7 +111,8 @@ static inline int task_in_mem_cgroup(str
>  	return 1;
>  }
>  
> -static inline int mem_cgroup_prepare_migration(struct page *page)
> +static inline int
> +mem_cgroup_prepare_migration(struct page *page , struct page *newpage)
>  {
>  	return 0;
>  }
> Index: linux-2.6.25-rc5-mm1-k/mm/page_alloc.c
> ===================================================================
> --- linux-2.6.25-rc5-mm1-k.orig/mm/page_alloc.c
> +++ linux-2.6.25-rc5-mm1-k/mm/page_alloc.c
> @@ -222,17 +222,11 @@ static inline int bad_range(struct zone 
>  
>  static void bad_page(struct page *page)
>  {
> -	void *pc = page_get_page_cgroup(page);
> -
>  	printk(KERN_EMERG "Bad page state in process '%s'\n" KERN_EMERG
>  		"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
>  		current->comm, page, (int)(2*sizeof(unsigned long)),
>  		(unsigned long)page->flags, page->mapping,
>  		page_mapcount(page), page_count(page));
> -	if (pc) {
> -		printk(KERN_EMERG "cgroup:%p\n", pc);
> -		page_reset_bad_cgroup(page);
> -	}
>  	printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
>  		KERN_EMERG "Backtrace:\n");
>  	dump_stack();
> @@ -478,7 +472,6 @@ static inline int free_pages_check(struc
>  {
>  	if (unlikely(page_mapcount(page) |
>  		(page->mapping != NULL)  |
> -		(page_get_page_cgroup(page) != NULL) |
>  		(page_count(page) != 0)  |
>  		(page->flags & (
>  			1 << PG_lru	|
> @@ -628,7 +621,6 @@ static int prep_new_page(struct page *pa
>  {
>  	if (unlikely(page_mapcount(page) |
>  		(page->mapping != NULL)  |
> -		(page_get_page_cgroup(page) != NULL) |
>  		(page_count(page) != 0)  |
>  		(page->flags & (
>  			1 << PG_lru	|
> Index: linux-2.6.25-rc5-mm1-k/include/linux/mm_types.h
> ===================================================================
> --- linux-2.6.25-rc5-mm1-k.orig/include/linux/mm_types.h
> +++ linux-2.6.25-rc5-mm1-k/include/linux/mm_types.h
> @@ -88,9 +88,6 @@ struct page {
>  	void *virtual;			/* Kernel virtual address (NULL if
>  					   not kmapped, ie. highmem) */
>  #endif /* WANT_PAGE_VIRTUAL */
> -#ifdef CONFIG_CGROUP_MEM_RES_CTLR
> -	unsigned long page_cgroup;
> -#endif
>  #ifdef CONFIG_PAGE_OWNER
>  	int order;
>  	unsigned int gfp_mask;
> Index: linux-2.6.25-rc5-mm1-k/mm/migrate.c
> ===================================================================
> --- linux-2.6.25-rc5-mm1-k.orig/mm/migrate.c
> +++ linux-2.6.25-rc5-mm1-k/mm/migrate.c
> @@ -358,6 +358,12 @@ static int migrate_page_move_mapping(str
>  
>  	write_unlock_irq(&mapping->tree_lock);
>  
> +	/* by mem_cgroup_prepare_migration, newpage is already
> +	   assigned to valid cgroup. and current->mm and GFP_ATOMIC
> +	   will not be used...*/
> +	mem_cgroup_uncharge_page(page);
> +	mem_cgroup_cache_charge(newpage, current->mm, GFP_ATOMIC);
> +
>  	return 0;
>  }
>  
> @@ -603,7 +609,6 @@ static int move_to_new_page(struct page 
>  		rc = fallback_migrate_page(mapping, newpage, page);
>  
>  	if (!rc) {
> -		mem_cgroup_page_migration(page, newpage);
>  		remove_migration_ptes(page, newpage);
>  	} else
>  		newpage->mapping = NULL;
> @@ -633,6 +638,12 @@ static int unmap_and_move(new_page_t get
>  		/* page was freed from under us. So we are done. */
>  		goto move_newpage;
>  
> +	charge = mem_cgroup_prepare_migration(page, newpage);
> +	if (charge == -ENOMEM) {
> +		rc = -ENOMEM;
> +		goto move_newpage;
> +	}
> +
>  	rc = -EAGAIN;
>  	if (TestSetPageLocked(page)) {
>  		if (!force)
> @@ -684,19 +695,14 @@ static int unmap_and_move(new_page_t get
>  		goto rcu_unlock;
>  	}
>  
> -	charge = mem_cgroup_prepare_migration(page);
>  	/* Establish migration ptes or remove ptes */
>  	try_to_unmap(page, 1);
>  
>  	if (!page_mapped(page))
>  		rc = move_to_new_page(newpage, page);
>  
> -	if (rc) {
> +	if (rc)
>  		remove_migration_ptes(page, page);
> -		if (charge)
> -			mem_cgroup_end_migration(page);
> -	} else if (charge)
> - 		mem_cgroup_end_migration(newpage);
>  rcu_unlock:
>  	if (rcu_locked)
>  		rcu_read_unlock();
> @@ -717,6 +723,8 @@ unlock:
>  	}
>  
>  move_newpage:
> +	if (!charge)
> +		mem_cgroup_end_migration(newpage);
>  	/*
>  	 * Move the new page to the LRU. If migration was not successful
>  	 * then this will free the page.
> Index: linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
> ===================================================================
> --- /dev/null
> +++ linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
> @@ -0,0 +1,260 @@
> +/*
> + * per-page accounting subsystem infrastructure. - linux/mm/page_cgroup.c
> + *
> + * (C) 2008 FUJITSU, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 of the License, or
> + * (at your option) any later version.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + *
> + * page_cgroup is yet another mem_map under memory resoruce controller.
> + * It containes information which cannot be stored in usual mem_map.
> + * This allows us to keep 'struct page' small when a user doesn't activate
> + * memory resource controller.
> + *
> + * We can translate : struct page <-> pfn -> page_cgroup -> struct page.
> + *
> + */
> +
> +#include <linux/mm.h>
> +#include <linux/slab.h>
> +#include <linux/radix-tree.h>
> +#include <linux/memcontrol.h>
> +#include <linux/err.h>
> +
> +static int page_cgroup_order __read_mostly;
> +static int page_cgroup_head_size __read_mostly;
> +
> +#define PCGRP_SHIFT	(page_cgroup_order)
> +#define PCGRP_SIZE	(1 << PCGRP_SHIFT)
> +#define PCGRP_MASK	(PCGRP_SIZE - 1)
> +
> +struct page_cgroup_head {
> +	struct page_cgroup pc[0];
> +};
> +
> +struct page_cgroup_root {
> +	spinlock_t	       tree_lock;
> +	struct radix_tree_root root_node;
> +};
> +
> +/*
> + * Calculate page_cgroup order to be not larger than order-2 page allocation.
> + */
> +static void calc_page_cgroup_order(void)
> +{
> +	int order = pageblock_order;
> +	unsigned long size = sizeof(struct page_cgroup) << order;
> +
> +	while (size > PAGE_SIZE * 2) {
> +		order -= 1;
> +		size = sizeof(struct page_cgroup) << order;
> +	}
> +
> +	page_cgroup_order = order;
> +	page_cgroup_head_size = sizeof(struct page_cgroup_head) +
> +				(sizeof(struct page_cgroup) << order);
> +}
> +
> +static struct page_cgroup_root __initdata *tmp_root_dir[MAX_NUMNODES];
> +static struct page_cgroup_root *root_node[MAX_NUMNODES] __read_mostly;
> +
> +static void
> +init_page_cgroup_head(struct page_cgroup_head *head, unsigned long pfn)
> +{
> +	struct page *page;
> +	struct page_cgroup *pc;
> +	int i;
> +
> +	for (i = 0, page = pfn_to_page(pfn), pc = &head->pc[0];
> +		i < PCGRP_SIZE; i++, page++, pc++) {

indentation:
	     i < PCGRP_SIZE; i++, page++, pc++) {

> +		pc->refcnt = 0;
> +		pc->page = page;
> +		spin_lock_init(&pc->lock);
> +	}
> +}
> +
> +
> +struct kmem_cache *page_cgroup_cachep;
> +
> +static struct page_cgroup_head *
> +alloc_page_cgroup_head(unsigned long pfn, int nid, gfp_t mask)
> +{
> +	struct page_cgroup_head *head;
> +
> +	if (!node_state(nid, N_NORMAL_MEMORY))
> +		nid = -1;
> +	head = kmem_cache_alloc_node(page_cgroup_cachep, mask, nid);
> +	if (head)
> +		init_page_cgroup_head(head, pfn);
> +
> +	return head;
> +}
> +
> +void free_page_cgroup(struct page_cgroup_head *head)
> +{
> +	kmem_cache_free(page_cgroup_cachep, head);
> +}
> +
> +static struct page_cgroup_root *pcgroup_get_root(struct page *page)
> +{
> +	int nid;
> +
> +	VM_BUG_ON(!page);
> +
> +	nid = page_to_nid(page);
> +
> +	return root_node[nid];
> +}
> +
> +/**
> + * get_page_cgroup - look up a page_cgroup for a page
> + * @page: the page whose page_cgroup is looked up.
> + *
> + * This just does lookup.
> + */
> +struct page_cgroup *get_page_cgroup(struct page *page)
> +{
> +	struct page_cgroup_head *head;
> +	struct page_cgroup_root *root;
> +	struct page_cgroup *ret = NULL;
> +	unsigned long pfn, idx;
> +
> +	/*
> +	 * NULL can be returned before initialization
> +	 */
> +	root = pcgroup_get_root(page);
> +	if (unlikely(!root))
> +		return ret;
> +
> +	pfn = page_to_pfn(page);
> +	idx = pfn >> PCGRP_SHIFT;
> +	/*
> +	 * We don't need lock here because no one deletes this head.
> +	 * (Freeing routtine will be added later.)
> +	 */
> +	rcu_read_lock();
> +	head = radix_tree_lookup(&root->root_node, idx);
> +	rcu_read_unlock();
> +
> +	if (likely(head))
> +		ret = &head->pc[pfn & PCGRP_MASK];
> +
> +	return ret;
> +}
> +
> +/**
> + * get_alloc_page_cgroup - look up or allocate a page_cgroup for a page
> + * @page: the page whose page_cgroup is looked up.
> + * @gfpmask: the gfpmask which will be used for page allocatiopn.
> + *
> + * look up and allocate if not found.
> + */
> +
> +struct page_cgroup *
> +get_alloc_page_cgroup(struct page *page, gfp_t gfpmask)
> +{
> +	struct page_cgroup_root *root;
> +	struct page_cgroup_head *head;
> +	struct page_cgroup *pc;
> +	unsigned long pfn, idx;
> +	int nid;
> +	unsigned long base_pfn, flags;
> +	int error = 0;
> +
> +	might_sleep_if(gfpmask & __GFP_WAIT);
> +
> +retry:
> +	pc = get_page_cgroup(page);
> +	if (pc)
> +		return pc;
> +	/*
> +	 * NULL can be returned before initialization.
> +	 */
> +	root = pcgroup_get_root(page);
> +	if (unlikely(!root))
> +		return NULL;
> +
> +	pfn = page_to_pfn(page);
> +	idx = pfn >> PCGRP_SHIFT;
> +	nid = page_to_nid(page);
> +	base_pfn = idx << PCGRP_SHIFT;
> +
> +	gfpmask = gfpmask & ~(__GFP_HIGHMEM | __GFP_MOVABLE);
> +
> +	head = alloc_page_cgroup_head(base_pfn, nid, gfpmask);
> +	if (!head)
> +		return ERR_PTR(-ENOMEM);
> +
> +	pc = &head->pc[pfn & PCGRP_MASK];
> +
> +	error = radix_tree_preload(gfpmask);
> +	if (error)
> +		goto out;
> +	spin_lock_irqsave(&root->tree_lock, flags);
> +	error = radix_tree_insert(&root->root_node, idx, head);
> +	spin_unlock_irqrestore(&root->tree_lock, flags);
> +	radix_tree_preload_end();
> +out:
> +	if (error) {
> +		free_page_cgroup(head);
> +		if (error == -EEXIST)
> +			goto retry;
> +		pc = ERR_PTR(error);
> +	}
> +	return pc;
> +}
> +
> +__init int page_cgroup_init(void)

static int __init

> +{
> +	int tmp, nid;
> +	struct page_cgroup_root *root;
> +
> +	calc_page_cgroup_order();
> +
> +	page_cgroup_cachep = kmem_cache_create("page_cgroup",
> +				page_cgroup_head_size, 0,
> +				SLAB_PANIC | SLAB_DESTROY_BY_RCU, NULL);
> +
> +	if (!page_cgroup_cachep) {
> +		printk(KERN_ERR "page accouning setup failure\n");
> +		printk(KERN_ERR "can't initialize slab memory\n");

why not:

printk(KERN_ERR "page accouning setup failure, can't initialize slab memory\n");

otherwise the user may think 2 bad things happened

> +		return -ENOMEM;
> +	}
> +
> +	for_each_node(nid) {
> +		tmp = nid;
> +		if (!node_state(nid, N_NORMAL_MEMORY))
> +			tmp = -1;
> +		root = kmalloc_node(sizeof(struct page_cgroup_root),
> +					GFP_KERNEL, tmp);
> +		if (!root)
> +			goto unroll;
> +		INIT_RADIX_TREE(&root->root_node, GFP_ATOMIC);
> +		spin_lock_init(&root->tree_lock);
> +		tmp_root_dir[nid] = root;
> +	}
> +	/*
> +	 * By filling node_root[], this tree turns to be visible.
> +	 * Because we have to finish initialization of the tree before
> +	 * we make it visible, memory barrier is necessary.
> +	 */
> +	smp_wmb();
> +	for_each_node(nid)
> +		root_node[nid] = tmp_root_dir[nid];
> +
> +	printk(KERN_INFO "Page Accouintg is activated\n");

			       Accounting

> +	return 0;
> +unroll:
> +	for_each_node(nid)
> +		kfree(tmp_root_dir[nid]);
> +
> +	return -ENOMEM;
> +}
> +late_initcall(page_cgroup_init);
> Index: linux-2.6.25-rc5-mm1-k/mm/Makefile
> ===================================================================
> --- linux-2.6.25-rc5-mm1-k.orig/mm/Makefile
> +++ linux-2.6.25-rc5-mm1-k/mm/Makefile
> @@ -32,5 +32,5 @@ obj-$(CONFIG_FS_XIP) += filemap_xip.o
>  obj-$(CONFIG_MIGRATION) += migrate.o
>  obj-$(CONFIG_SMP) += allocpercpu.o
>  obj-$(CONFIG_QUICKLIST) += quicklist.o
> -obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o
> +obj-$(CONFIG_CGROUP_MEM_RES_CTLR) += memcontrol.o page_cgroup.o
>  
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [-mm] [PATCH 1/4] memcg : radix-tree page_cgroup

[KAMEZAWA Hiroyuki]

On Fri, 28 Mar 2008 08:55:01 +0800
Li Zefan <lizf@cn.fujitsu.com> wrote:

> > Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > 
> 
> At a first glance, some typos and small things, see below ;)
>
ok, thank you for review.

Regards,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[-mm] [PATCH 2/4] memcg: boost by percpu

[KAMEZAWA Hiroyuki]

This patch adds per-cpu look up cache for get_page_cgroup().
Works well when nearby pages are accessed continuously.

Changelog v1 -> v2:
  * avoid inlining by adding function to page_cgroup.h
  * set to be cacheline-aligned.
  * added hashfunc() macro.
  * changed what should be recorded in cache.
    This version rememveres pre-caluculated base address, not "head".


Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

 mm/page_cgroup.c |   53 ++++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 50 insertions(+), 3 deletions(-)

Index: linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/page_cgroup.c
+++ linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
@@ -26,6 +26,7 @@
 #include <linux/slab.h>
 #include <linux/radix-tree.h>
 #include <linux/memcontrol.h>
+#include <linux/interrupt.h>
 #include <linux/err.h>
 
 static int page_cgroup_order __read_mostly;
@@ -80,6 +81,47 @@ init_page_cgroup_head(struct page_cgroup
 	}
 }
 
+#define PCGRP_CACHE_SIZE	(0x10)
+#define PCGRP_CACHE_MASK	(PCGRP_CACHE_SIZE - 1)
+struct page_cgroup_cache {
+	struct {
+		unsigned long idx;
+		struct page_cgroup *base;
+	} ents[PCGRP_CACHE_SIZE];
+};
+DEFINE_PER_CPU(struct page_cgroup_cache, pcpu_pcgroup_cache) ____cacheline_aligned;
+
+#define hashfunc(idx)	((idx) & PCGRP_CACHE_MASK)
+
+static struct page_cgroup *pcp_lookup(unsigned long pfn, unsigned long idx)
+{
+	struct page_cgroup *ret = NULL;
+	struct page_cgroup_cache *pcp;
+	int hnum = hashfunc(idx);
+
+	pcp = &get_cpu_var(pcpu_pcgroup_cache);
+	if (pcp->ents[hnum].idx == idx && pcp->ents[hnum].base)
+		ret = pcp->ents[hnum].base + pfn;
+	put_cpu_var(pcpu_pcgroup_cache);
+	return ret;
+}
+
+static void cache_result(unsigned long idx, struct page_cgroup_head *head)
+{
+	struct page_cgroup_cache *pcp;
+	int hnum = hashfunc(idx);
+
+	/*
+	 * Because look up is done under preempt_disable, don't modifies
+	 * an entry in interrupt.
+	 */
+	if (in_interrupt())
+		return;
+	pcp = &get_cpu_var(pcpu_pcgroup_cache);
+	pcp->ents[hnum].idx = idx;
+	pcp->ents[hnum].base = &head->pc[0] - (idx << PCGRP_SHIFT);
+	put_cpu_var(pcpu_pcgroup_cache);
+}
 
 struct kmem_cache *page_cgroup_cachep;
 
@@ -126,6 +168,11 @@ struct page_cgroup *get_page_cgroup(stru
 	struct page_cgroup *ret = NULL;
 	unsigned long pfn, idx;
 
+	pfn = page_to_pfn(page);
+	idx = pfn >> PCGRP_SHIFT;
+	ret = pcp_lookup(pfn, idx);
+	if (ret)
+		return ret;
 	/*
 	 * NULL can be returned before initialization
 	 */
@@ -133,8 +180,6 @@ struct page_cgroup *get_page_cgroup(stru
 	if (unlikely(!root))
 		return ret;
 
-	pfn = page_to_pfn(page);
-	idx = pfn >> PCGRP_SHIFT;
 	/*
 	 * We don't need lock here because no one deletes this head.
 	 * (Freeing routtine will be added later.)
@@ -143,8 +188,10 @@ struct page_cgroup *get_page_cgroup(stru
 	head = radix_tree_lookup(&root->root_node, idx);
 	rcu_read_unlock();
 
-	if (likely(head))
+	if (likely(head)) {
+		cache_result(idx, head);
 		ret = &head->pc[pfn & PCGRP_MASK];
+	}
 
 	return ret;
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[-mm] [PATCH 3/4] memcg : shirink page cgroup

[KAMEZAWA Hiroyuki]

This patch is for freeing page_cgroup if a chunk of pages are freed.

How this works 
 * when the order of free page reaches PCGRP_SHRINK_ORDER, pcgrp is freed.
   This will be done by RCU.

Changelog v1 -> v2:
 * shrink_order is automatically calculated.
 * added comments.
 * added cpu mask to page_cgroup_head. (this will be help at clearing.)
 * moved a routine for flushing percpu idx after deletion in radix-tree.
 * removed memory barrier. (I noticed that clearing can be done after
   lock/unlock.)
 * add a sanity check not to access a page_cgroup for a page whose refcnt is 0.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsuc.com>


 include/linux/page_cgroup.h |   13 ++++++++
 mm/page_alloc.c             |    2 +
 mm/page_cgroup.c            |   70 ++++++++++++++++++++++++++++++++++++++++++--
 3 files changed, 83 insertions(+), 2 deletions(-)

Index: linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/page_cgroup.c
+++ linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
@@ -31,12 +31,14 @@
 
 static int page_cgroup_order __read_mostly;
 static int page_cgroup_head_size __read_mostly;
+int pcgroup_shrink_order __read_mostly;
 
 #define PCGRP_SHIFT	(page_cgroup_order)
 #define PCGRP_SIZE	(1 << PCGRP_SHIFT)
 #define PCGRP_MASK	(PCGRP_SIZE - 1)
 
 struct page_cgroup_head {
+	cpumask_t mask;
 	struct page_cgroup pc[0];
 };
 
@@ -61,6 +63,10 @@ static void calc_page_cgroup_order(void)
 	page_cgroup_order = order;
 	page_cgroup_head_size = sizeof(struct page_cgroup_head) +
 				(sizeof(struct page_cgroup) << order);
+	if (order + 1 < MAX_ORDER)
+		pcgroup_shrink_order = order + 1;
+	else
+		pcgroup_shrink_order = MAX_ORDER - 1;
 }
 
 static struct page_cgroup_root __initdata *tmp_root_dir[MAX_NUMNODES];
@@ -72,7 +78,7 @@ init_page_cgroup_head(struct page_cgroup
 	struct page *page;
 	struct page_cgroup *pc;
 	int i;
-
+	cpus_clear(head->mask);
 	for (i = 0, page = pfn_to_page(pfn), pc = &head->pc[0];
 		i < PCGRP_SIZE; i++, page++, pc++) {
 		pc->refcnt = 0;
@@ -120,6 +126,7 @@ static void cache_result(unsigned long i
 	pcp = &get_cpu_var(pcpu_pcgroup_cache);
 	pcp->ents[hnum].idx = idx;
 	pcp->ents[hnum].base = &head->pc[0] - (idx << PCGRP_SHIFT);
+	cpu_set(smp_processor_id(), head->mask);
 	put_cpu_var(pcpu_pcgroup_cache);
 }
 
@@ -150,6 +157,8 @@ static struct page_cgroup_root *pcgroup_
 
 	VM_BUG_ON(!page);
 
+	VM_BUG_ON(!page_count(page));
+
 	nid = page_to_nid(page);
 
 	return root_node[nid];
@@ -257,6 +266,62 @@ out:
 	}
 	return pc;
 }
+/*
+ * This will be called from deep place in free_pages()
+ * Because this is called by page allocator, we can assume that
+ * 1. zone->lock is held.
+ * 2. no one touches pages in [page...page + (1 << order))
+ * 3. Because of 2, page_cgroups against [page...page + (1 << order)]
+ *    is not touched and will not be touched while we hold zone->lock.
+ */
+void __shrink_page_cgroup(struct page *page, int order)
+{
+	struct page_cgroup_root *root;
+	unsigned long pfn = page_to_pfn(page);
+	unsigned long end_pfn;
+	int cpu;
+
+	root = pcgroup_get_root(page);
+	if (in_interrupt() || !root || (order < PCGRP_SHIFT))
+		return;
+
+	pfn = page_to_pfn(page);
+	end_pfn = pfn + (1 << order);
+
+	while (pfn != end_pfn) {
+		if (spin_trylock(&root->tree_lock)) {
+			struct page_cgroup_cache *pcp;
+			struct page_cgroup_head *head = NULL;
+			int idx = pfn >> PCGRP_SHIFT;
+			/*
+			 * Because [pfn, end_pfn) are free pages, we can assume
+			 * no lookup in this range will occur. So this has no
+			 * race. For rafix-tree, we have to take a lock.
+			 * Radix tree is freed by RCU. so they will not call
+			 * free_pages() directly under this.
+			 */
+			head = radix_tree_delete(&root->root_node, idx);
+			spin_unlock(&root->tree_lock);
+
+			/*
+			 * It's guaranteed that no one has access to this pfn
+			 * because there isn't and won't be access to the page
+			 * and page_cgroup.
+			 */
+			if (head) {
+				int hnum = hashfunc(idx);
+				for_each_cpu_mask(cpu, head->mask) {
+					pcp = &per_cpu(pcpu_pcgroup_cache, cpu);
+					if (pcp->ents[hnum].idx == idx)
+						pcp->ents[hnum].base = NULL;
+				}
+				/* SLAB for head is SLAB_DESTROY_BY_RCU. */
+				free_page_cgroup(head);
+			}
+		}
+		pfn += PCGRP_SIZE;
+	}
+}
 
 __init int page_cgroup_init(void)
 {
@@ -296,7 +361,8 @@ __init int page_cgroup_init(void)
 	for_each_node(nid)
 		root_node[nid] = tmp_root_dir[nid];
 
-	printk(KERN_INFO "Page Accouintg is activated\n");
+	printk(KERN_INFO "Page Accouintg is activated, its order is %d\n",
+			page_cgroup_order);
 	return 0;
 unroll:
 	for_each_node(nid)
Index: linux-2.6.25-rc5-mm1-k/mm/page_alloc.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/page_alloc.c
+++ linux-2.6.25-rc5-mm1-k/mm/page_alloc.c
@@ -45,6 +45,7 @@
 #include <linux/fault-inject.h>
 #include <linux/page-isolation.h>
 #include <linux/memcontrol.h>
+#include <linux/page_cgroup.h>
 
 #include <asm/tlbflush.h>
 #include <asm/div64.h>
@@ -463,6 +464,7 @@ static inline void __free_one_page(struc
 		order++;
 	}
 	set_page_order(page, order);
+	shrink_page_cgroup(page, order);
 	list_add(&page->lru,
 		&zone->free_area[order].free_list[migratetype]);
 	zone->free_area[order].nr_free++;
Index: linux-2.6.25-rc5-mm1-k/include/linux/page_cgroup.h
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/include/linux/page_cgroup.h
+++ linux-2.6.25-rc5-mm1-k/include/linux/page_cgroup.h
@@ -40,6 +40,15 @@ extern struct page_cgroup *get_page_cgro
 extern struct page_cgroup *
 get_alloc_page_cgroup(struct page *page, gfp_t gfpmask);
 
+extern int pcgroup_shrink_order;
+extern void __shrink_page_cgroup(struct page *page, int order);
+
+static inline void shrink_page_cgroup(struct page *page, int order)
+{
+	if (unlikely(order >= pcgroup_shrink_order))
+		__shrink_page_cgroup(page, order);
+}
+
 #else
 
 static inline struct page_cgroup *get_page_cgroup(struct page *page)
@@ -52,5 +61,9 @@ get_alloc_page_cgroup(struct page *page,
 {
 	return NULL;
 }
+static inline void
+shrink_page_cgroup(struct page *page, int order)
+{
+}
 #endif
 #endif

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[-mm] [PATCH 4/4] memcg : radix-tree page_cgroup v2

[KAMEZAWA Hiroyuki]

This 2 prefetches works well under my test.
(*)Tested on x86_64 and ia64 by unixbench/execl.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 mm/memcontrol.c  |    5 ++++-
 mm/page_cgroup.c |    4 +++-
 2 files changed, 7 insertions(+), 2 deletions(-)

Index: linux-2.6.25-rc5-mm1-k/mm/memcontrol.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/memcontrol.c
+++ linux-2.6.25-rc5-mm1-k/mm/memcontrol.c
@@ -182,10 +182,13 @@ static struct mem_cgroup_per_zone *
 page_cgroup_zoneinfo(struct page_cgroup *pc)
 {
 	struct mem_cgroup *mem = pc->mem_cgroup;
+	struct mem_cgroup_per_zone *mz;
 	int nid = page_cgroup_nid(pc);
 	int zid = page_cgroup_zid(pc);
 
-	return mem_cgroup_zoneinfo(mem, nid, zid);
+	mz = mem_cgroup_zoneinfo(mem, nid, zid);
+	prefetch(mz);
+	return mz;
 }
 
 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
Index: linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
===================================================================
--- linux-2.6.25-rc5-mm1-k.orig/mm/page_cgroup.c
+++ linux-2.6.25-rc5-mm1-k/mm/page_cgroup.c
@@ -106,8 +106,10 @@ static struct page_cgroup *pcp_lookup(un
 	int hnum = hashfunc(idx);
 
 	pcp = &get_cpu_var(pcpu_pcgroup_cache);
-	if (pcp->ents[hnum].idx == idx && pcp->ents[hnum].base)
+	if (pcp->ents[hnum].idx == idx && pcp->ents[hnum].base) {
 		ret = pcp->ents[hnum].base + pfn;
+		prefetch(ret);
+	}
 	put_cpu_var(pcpu_pcgroup_cache);
 	return ret;
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [-mm] [PATCH 4/4] memcg : radix-tree page_cgroup v2

[KAMEZAWA Hiroyuki]

Sorry, Bad title...

[PATCH 4/4] memcg: add prefetch.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [-mm] [PATCH 0/4] memcg : radix-tree page_cgroup v2

[KOSAKI Motohiro]

Hi

>          TEST                                BASELINE     RESULT      INDEX
> (1)      Execl Throughput                        43.0     2868.8      667.2
> (2)      Execl Throughput                        43.0     2810.3      653.6
> (3)      Execl Throughput                        43.0     2836.9      659.7
> (4)      Execl Throughput                        43.0     2846.0      661.9
> (5)      Execl Throughput                        43.0     2862.0      665.6
> (6)      Execl Throughput                        43.0     3110.0      723.3
> 
> (1) .... rc5-mm1 + memory controller
> (2) .... patch 1/4 is applied.      (use radix-tree always.)
> (3) .... patch [1-3]/4 are applied. (caching by percpu)
> (4) .... patch [1-4]/4 are applied. (uses prefetch)
> (5) .... adjust sizeof(struct page) to be 64 bytes by padding.
> (6) .... rc5-mm1 *without* memory controller

I am very surprised this result. 
723.3 -> 667.2 seems large performance impact.

Why do you need count resource usage when unlimited limit.
Could you separate unlimited group to resource usage counting and no counting.
I hope default cgroup keep no counting and no decrease performance.



--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [-mm] [PATCH 0/4] memcg : radix-tree page_cgroup v2

[KAMEZAWA Hiroyuki]

On Thu, 27 Mar 2008 18:12:42 +0900
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:

> Hi
> 
> >          TEST                                BASELINE     RESULT      INDEX
> > (1)      Execl Throughput                        43.0     2868.8      667.2
> > (2)      Execl Throughput                        43.0     2810.3      653.6
> > (3)      Execl Throughput                        43.0     2836.9      659.7
> > (4)      Execl Throughput                        43.0     2846.0      661.9
> > (5)      Execl Throughput                        43.0     2862.0      665.6
> > (6)      Execl Throughput                        43.0     3110.0      723.3
> > 
> > (1) .... rc5-mm1 + memory controller
> > (2) .... patch 1/4 is applied.      (use radix-tree always.)
> > (3) .... patch [1-3]/4 are applied. (caching by percpu)
> > (4) .... patch [1-4]/4 are applied. (uses prefetch)
> > (5) .... adjust sizeof(struct page) to be 64 bytes by padding.
> > (6) .... rc5-mm1 *without* memory controller
> 
> I am very surprised this result. 
> 723.3 -> 667.2 seems large performance impact.
> 
> Why do you need count resource usage when unlimited limit.
> Could you separate unlimited group to resource usage counting and no counting.
> I hope default cgroup keep no counting and no decrease performance.

At first, I'd like to reduce this overhead even under memory resource
controller's accounting ;)
We have boot-time-disable option now. But it doesn't seem what you want.

Considering workaround....
In current system, *unlimited* doesn't mean *no account*.
So, I think we have an option to add "no account" flag per cgroup.

Hmm..some interface to do
- allow "no account" -> "account"
- disallow "account" -> "no account"

Balbir-san, how do you think ?


Thanks,
-Kame





--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [-mm] [PATCH 0/4] memcg : radix-tree page_cgroup v2

[Balbir Singh]

KAMEZAWA Hiroyuki wrote:
> On Thu, 27 Mar 2008 18:12:42 +0900
> KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
> 
>> Hi
>>
>>>          TEST                                BASELINE     RESULT      INDEX
>>> (1)      Execl Throughput                        43.0     2868.8      667.2
>>> (2)      Execl Throughput                        43.0     2810.3      653.6
>>> (3)      Execl Throughput                        43.0     2836.9      659.7
>>> (4)      Execl Throughput                        43.0     2846.0      661.9
>>> (5)      Execl Throughput                        43.0     2862.0      665.6
>>> (6)      Execl Throughput                        43.0     3110.0      723.3
>>>
>>> (1) .... rc5-mm1 + memory controller
>>> (2) .... patch 1/4 is applied.      (use radix-tree always.)
>>> (3) .... patch [1-3]/4 are applied. (caching by percpu)
>>> (4) .... patch [1-4]/4 are applied. (uses prefetch)
>>> (5) .... adjust sizeof(struct page) to be 64 bytes by padding.
>>> (6) .... rc5-mm1 *without* memory controller
>> I am very surprised this result. 
>> 723.3 -> 667.2 seems large performance impact.
>>
>> Why do you need count resource usage when unlimited limit.
>> Could you separate unlimited group to resource usage counting and no counting.
>> I hope default cgroup keep no counting and no decrease performance.
> 
> At first, I'd like to reduce this overhead even under memory resource
> controller's accounting ;)
> We have boot-time-disable option now. But it doesn't seem what you want.
> 
> Considering workaround....
> In current system, *unlimited* doesn't mean *no account*.
> So, I think we have an option to add "no account" flag per cgroup.
> 
> Hmm..some interface to do
> - allow "no account" -> "account"
> - disallow "account" -> "no account"
> 
> Balbir-san, how do you think ?

The reason we do accounting for default group is to allow reporting of
usage/statistics and in the future when we do hierarchial accounting and
control, it will be much more useful.

I like the interface idea, but I'd like to do two things

1. Keeping accounting on by default or have an option to do so
2. Reduce the memory controller overhead

-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [-mm] [PATCH 0/4] memcg : radix-tree page_cgroup v2

[KAMEZAWA Hiroyuki]

On Thu, 27 Mar 2008 15:23:57 +0530
Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> >>>          TEST                                BASELINE     RESULT      INDEX
> >>> (1)      Execl Throughput                        43.0     2868.8      667.2
> >>> (2)      Execl Throughput                        43.0     2810.3      653.6
> >>> (3)      Execl Throughput                        43.0     2836.9      659.7
> >>> (4)      Execl Throughput                        43.0     2846.0      661.9
> >>> (5)      Execl Throughput                        43.0     2862.0      665.6
> >>> (6)      Execl Throughput                        43.0     3110.0      723.3
> >>>
> >>> (1) .... rc5-mm1 + memory controller
> >>> (2) .... patch 1/4 is applied.      (use radix-tree always.)
> >>> (3) .... patch [1-3]/4 are applied. (caching by percpu)
> >>> (4) .... patch [1-4]/4 are applied. (uses prefetch)
> >>> (5) .... adjust sizeof(struct page) to be 64 bytes by padding.
> >>> (6) .... rc5-mm1 *without* memory controller
> >> I am very surprised this result. 
> >> 723.3 -> 667.2 seems large performance impact.
> >>
> >> Why do you need count resource usage when unlimited limit.
> >> Could you separate unlimited group to resource usage counting and no counting.
> >> I hope default cgroup keep no counting and no decrease performance.
> > 
> > At first, I'd like to reduce this overhead even under memory resource
> > controller's accounting ;)
> > We have boot-time-disable option now. But it doesn't seem what you want.
> > 
> > Considering workaround....
> > In current system, *unlimited* doesn't mean *no account*.
> > So, I think we have an option to add "no account" flag per cgroup.
> > 
> > Hmm..some interface to do
> > - allow "no account" -> "account"
> > - disallow "account" -> "no account"
> > 
> > Balbir-san, how do you think ?
> 
> The reason we do accounting for default group is to allow reporting of
> usage/statistics and in the future when we do hierarchial accounting and
> control, it will be much more useful.
> 
I see.

> I like the interface idea, but I'd like to do two things
> 
> 1. Keeping accounting on by default or have an option to do so
> 2. Reduce the memory controller overhead
> 
I'll do something against "1" if I can think of.
I'd like to try "2" after this patches. Maybe reducing lock-bouncing can be
good help...

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>