mm: memcg: only check swap cache pages for repeated charging
Johannes Weiner [Tue, 31 Jul 2012 23:45:47 +0000 (16:45 -0700)]
Only anon and shmem pages in the swap cache are attempted to be charged
multiple times, from every swap pte fault or from shmem_unuse().  No other
pages require checking PageCgroupUsed().

Charging pages in the swap cache is also serialized by the page lock, and
since both the try_charge and commit_charge are called under the same page
lock section, the PageCgroupUsed() check might as well happen before the
counter charging, let alone reclaim.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Wanpeng Li <liwp.linux@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c

index 4cc962d..1a2020d 100644 (file)
@@ -2538,11 +2538,7 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
        bool anon;
 
        lock_page_cgroup(pc);
-       if (unlikely(PageCgroupUsed(pc))) {
-               unlock_page_cgroup(pc);
-               __mem_cgroup_cancel_charge(memcg, nr_pages);
-               return;
-       }
+       VM_BUG_ON(PageCgroupUsed(pc));
        /*
         * we don't need page_cgroup_lock about tail pages, becase they are not
         * accessed by any other context at this point.
@@ -2807,8 +2803,19 @@ static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
                                          struct mem_cgroup **memcgp)
 {
        struct mem_cgroup *memcg;
+       struct page_cgroup *pc;
        int ret;
 
+       pc = lookup_page_cgroup(page);
+       /*
+        * Every swap fault against a single page tries to charge the
+        * page, bail as early as possible.  shmem_unuse() encounters
+        * already charged pages, too.  The USED bit is protected by
+        * the page lock, which serializes swap cache removal, which
+        * in turn serializes uncharging.
+        */
+       if (PageCgroupUsed(pc))
+               return 0;
        if (!do_swap_account)
                goto charge_cur_mm;
        /*