mm: memcontrol: track move_lock state internally
authorJohannes Weiner <hannes@cmpxchg.org>
Wed, 11 Feb 2015 23:25:01 +0000 (15:25 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 12 Feb 2015 01:06:00 +0000 (17:06 -0800)
The complexity of memcg page stat synchronization is currently leaking
into the callsites, forcing them to keep track of the move_lock state and
the IRQ flags.  Simplify the API by tracking it in the memcg.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/memcontrol.h
mm/memcontrol.c
mm/page-writeback.c
mm/rmap.c

index fb212e1d700d29b42036e448023bc70b2c95227d..76b4084b8d082d440f454e5b18eebe96830474ac 100644 (file)
@@ -138,12 +138,10 @@ static inline bool mem_cgroup_disabled(void)
        return false;
 }
 
-struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page, bool *locked,
-                                             unsigned long *flags);
-void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool *locked,
-                             unsigned long *flags);
+struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
 void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
                                 enum mem_cgroup_stat_index idx, int val);
+void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
 
 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
                                            enum mem_cgroup_stat_index idx)
@@ -285,14 +283,12 @@ mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 {
 }
 
-static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page,
-                                       bool *locked, unsigned long *flags)
+static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
 {
        return NULL;
 }
 
-static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg,
-                                       bool *locked, unsigned long *flags)
+static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
 {
 }
 
index f3f8a4f52a0ca31f0731663f0240e856651c5cc4..028d07c79104e8f138f8b6b045166a62d4ad8cf0 100644 (file)
@@ -325,9 +325,11 @@ struct mem_cgroup {
        /*
         * set > 0 if pages under this cgroup are moving to other cgroup.
         */
-       atomic_t        moving_account;
+       atomic_t                moving_account;
        /* taken only while moving_account > 0 */
-       spinlock_t      move_lock;
+       spinlock_t              move_lock;
+       struct task_struct      *move_lock_task;
+       unsigned long           move_lock_flags;
        /*
         * percpu counter.
         */
@@ -1977,34 +1979,33 @@ cleanup:
 /**
  * mem_cgroup_begin_page_stat - begin a page state statistics transaction
  * @page: page that is going to change accounted state
- * @locked: &memcg->move_lock slowpath was taken
- * @flags: IRQ-state flags for &memcg->move_lock
  *
  * This function must mark the beginning of an accounted page state
  * change to prevent double accounting when the page is concurrently
  * being moved to another memcg:
  *
- *   memcg = mem_cgroup_begin_page_stat(page, &locked, &flags);
+ *   memcg = mem_cgroup_begin_page_stat(page);
  *   if (TestClearPageState(page))
  *     mem_cgroup_update_page_stat(memcg, state, -1);
- *   mem_cgroup_end_page_stat(memcg, locked, flags);
- *
- * The RCU lock is held throughout the transaction.  The fast path can
- * get away without acquiring the memcg->move_lock (@locked is false)
- * because page moving starts with an RCU grace period.
- *
- * The RCU lock also protects the memcg from being freed when the page
- * state that is going to change is the only thing preventing the page
- * from being uncharged.  E.g. end-writeback clearing PageWriteback(),
- * which allows migration to go ahead and uncharge the page before the
- * account transaction might be complete.
+ *   mem_cgroup_end_page_stat(memcg);
  */
-struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page,
-                                             bool *locked,
-                                             unsigned long *flags)
+struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
 {
        struct mem_cgroup *memcg;
+       unsigned long flags;
 
+       /*
+        * The RCU lock is held throughout the transaction.  The fast
+        * path can get away without acquiring the memcg->move_lock
+        * because page moving starts with an RCU grace period.
+        *
+        * The RCU lock also protects the memcg from being freed when
+        * the page state that is going to change is the only thing
+        * preventing the page from being uncharged.
+        * E.g. end-writeback clearing PageWriteback(), which allows
+        * migration to go ahead and uncharge the page before the
+        * account transaction might be complete.
+        */
        rcu_read_lock();
 
        if (mem_cgroup_disabled())
@@ -2014,16 +2015,22 @@ again:
        if (unlikely(!memcg))
                return NULL;
 
-       *locked = false;
        if (atomic_read(&memcg->moving_account) <= 0)
                return memcg;
 
-       spin_lock_irqsave(&memcg->move_lock, *flags);
+       spin_lock_irqsave(&memcg->move_lock, flags);
        if (memcg != page->mem_cgroup) {
-               spin_unlock_irqrestore(&memcg->move_lock, *flags);
+               spin_unlock_irqrestore(&memcg->move_lock, flags);
                goto again;
        }
-       *locked = true;
+
+       /*
+        * When charge migration first begins, we can have locked and
+        * unlocked page stat updates happening concurrently.  Track
+        * the task who has the lock for mem_cgroup_end_page_stat().
+        */
+       memcg->move_lock_task = current;
+       memcg->move_lock_flags = flags;
 
        return memcg;
 }
@@ -2031,14 +2038,17 @@ again:
 /**
  * mem_cgroup_end_page_stat - finish a page state statistics transaction
  * @memcg: the memcg that was accounted against
- * @locked: value received from mem_cgroup_begin_page_stat()
- * @flags: value received from mem_cgroup_begin_page_stat()
  */
-void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool *locked,
-                             unsigned long *flags)
+void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
 {
-       if (memcg && *locked)
-               spin_unlock_irqrestore(&memcg->move_lock, *flags);
+       if (memcg && memcg->move_lock_task == current) {
+               unsigned long flags = memcg->move_lock_flags;
+
+               memcg->move_lock_task = NULL;
+               memcg->move_lock_flags = 0;
+
+               spin_unlock_irqrestore(&memcg->move_lock, flags);
+       }
 
        rcu_read_unlock();
 }
index 6f4335238e33311de251a647fe725d06d5897060..fb71e9deca856903def056230fa46f687f207a3f 100644 (file)
@@ -2308,12 +2308,10 @@ EXPORT_SYMBOL(clear_page_dirty_for_io);
 int test_clear_page_writeback(struct page *page)
 {
        struct address_space *mapping = page_mapping(page);
-       unsigned long memcg_flags;
        struct mem_cgroup *memcg;
-       bool locked;
        int ret;
 
-       memcg = mem_cgroup_begin_page_stat(page, &locked, &memcg_flags);
+       memcg = mem_cgroup_begin_page_stat(page);
        if (mapping) {
                struct backing_dev_info *bdi = mapping->backing_dev_info;
                unsigned long flags;
@@ -2338,19 +2336,17 @@ int test_clear_page_writeback(struct page *page)
                dec_zone_page_state(page, NR_WRITEBACK);
                inc_zone_page_state(page, NR_WRITTEN);
        }
-       mem_cgroup_end_page_stat(memcg, &locked, &memcg_flags);
+       mem_cgroup_end_page_stat(memcg);
        return ret;
 }
 
 int __test_set_page_writeback(struct page *page, bool keep_write)
 {
        struct address_space *mapping = page_mapping(page);
-       unsigned long memcg_flags;
        struct mem_cgroup *memcg;
-       bool locked;
        int ret;
 
-       memcg = mem_cgroup_begin_page_stat(page, &locked, &memcg_flags);
+       memcg = mem_cgroup_begin_page_stat(page);
        if (mapping) {
                struct backing_dev_info *bdi = mapping->backing_dev_info;
                unsigned long flags;
@@ -2380,7 +2376,7 @@ int __test_set_page_writeback(struct page *page, bool keep_write)
                mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
                inc_zone_page_state(page, NR_WRITEBACK);
        }
-       mem_cgroup_end_page_stat(memcg, &locked, &memcg_flags);
+       mem_cgroup_end_page_stat(memcg);
        return ret;
 
 }
index 70b32498d4f24ec9ec414926148d1a67999707e4..5e3e09081164b83814683513c174445e67669a4b 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1085,24 +1085,20 @@ void page_add_new_anon_rmap(struct page *page,
 void page_add_file_rmap(struct page *page)
 {
        struct mem_cgroup *memcg;
-       unsigned long flags;
-       bool locked;
 
-       memcg = mem_cgroup_begin_page_stat(page, &locked, &flags);
+       memcg = mem_cgroup_begin_page_stat(page);
        if (atomic_inc_and_test(&page->_mapcount)) {
                __inc_zone_page_state(page, NR_FILE_MAPPED);
                mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
        }
-       mem_cgroup_end_page_stat(memcg, &locked, &flags);
+       mem_cgroup_end_page_stat(memcg);
 }
 
 static void page_remove_file_rmap(struct page *page)
 {
        struct mem_cgroup *memcg;
-       unsigned long flags;
-       bool locked;
 
-       memcg = mem_cgroup_begin_page_stat(page, &locked, &flags);
+       memcg = mem_cgroup_begin_page_stat(page);
 
        /* page still mapped by someone else? */
        if (!atomic_add_negative(-1, &page->_mapcount))
@@ -1123,7 +1119,7 @@ static void page_remove_file_rmap(struct page *page)
        if (unlikely(PageMlocked(page)))
                clear_page_mlock(page);
 out:
-       mem_cgroup_end_page_stat(memcg, &locked, &flags);
+       mem_cgroup_end_page_stat(memcg);
 }
 
 /**