mm: memcg: move pc lookup point to commit_charge()
authorJohannes Weiner <hannes@cmpxchg.org>
Tue, 24 Apr 2012 18:22:33 +0000 (20:22 +0200)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 26 Apr 2012 04:22:35 +0000 (21:22 -0700)
None of the callsites actually need the page_cgroup descriptor
themselves, so just pass the page and do the look up in there.

We already had two bugs (6568d4a 'mm: memcg: update the correct soft
limit tree during migration' and 'memcg: fix Bad page state after
replace_page_cache') where the passed page and pc were not referring
to the same page frame.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c

index b868def..31ab9c3 100644 (file)
@@ -2476,10 +2476,10 @@ struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
 static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
                                       struct page *page,
                                       unsigned int nr_pages,
-                                      struct page_cgroup *pc,
                                       enum charge_type ctype,
                                       bool lrucare)
 {
+       struct page_cgroup *pc = lookup_page_cgroup(page);
        struct zone *uninitialized_var(zone);
        bool was_on_lru = false;
        bool anon;
@@ -2716,7 +2716,6 @@ static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
 {
        struct mem_cgroup *memcg = NULL;
        unsigned int nr_pages = 1;
-       struct page_cgroup *pc;
        bool oom = true;
        int ret;
 
@@ -2730,11 +2729,10 @@ static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
                oom = false;
        }
 
-       pc = lookup_page_cgroup(page);
        ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
        if (ret == -ENOMEM)
                return ret;
-       __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype, false);
+       __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
        return 0;
 }
 
@@ -2831,16 +2829,13 @@ static void
 __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
                                        enum charge_type ctype)
 {
-       struct page_cgroup *pc;
-
        if (mem_cgroup_disabled())
                return;
        if (!memcg)
                return;
        cgroup_exclude_rmdir(&memcg->css);
 
-       pc = lookup_page_cgroup(page);
-       __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype, true);
+       __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
        /*
         * Now swap is on-memory. This means this page may be
         * counted both as mem and swap....double count.
@@ -3298,14 +3293,13 @@ int mem_cgroup_prepare_migration(struct page *page,
         * page. In the case new page is migrated but not remapped, new page's
         * mapcount will be finally 0 and we call uncharge in end_migration().
         */
-       pc = lookup_page_cgroup(newpage);
        if (PageAnon(page))
                ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
        else if (page_is_file_cache(page))
                ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
        else
                ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
-       __mem_cgroup_commit_charge(memcg, newpage, 1, pc, ctype, false);
+       __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
        return ret;
 }
 
@@ -3392,8 +3386,7 @@ void mem_cgroup_replace_page_cache(struct page *oldpage,
         * the newpage may be on LRU(or pagevec for LRU) already. We lock
         * LRU while we overwrite pc->mem_cgroup.
         */
-       pc = lookup_page_cgroup(newpage);
-       __mem_cgroup_commit_charge(memcg, newpage, 1, pc, type, true);
+       __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
 }
 
 #ifdef CONFIG_DEBUG_VM