memcg: fix split_huge_page_refcounts()
authorHugh Dickins <hughd@google.com>
Fri, 13 Jan 2012 01:19:52 +0000 (17:19 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 13 Jan 2012 04:13:09 +0000 (20:13 -0800)
This patch started off as a cleanup: __split_huge_page_refcounts() has to
cope with two scenarios, when the hugepage being split is already on LRU,
and when it is not; but why does it have to split that accounting across
three different sites?  Consolidate it in lru_add_page_tail(), handling
evictable and unevictable alike, and use standard add_page_to_lru_list()
when accounting is needed (when the head is not yet on LRU).

But a recent regression in -next, I guess the removal of PageCgroupAcctLRU
test from mem_cgroup_split_huge_fixup(), makes this now a necessary fix:
under load, the MEM_CGROUP_ZSTAT count was wrapping to a huge number,
messing up reclaim calculations and causing a freeze at rmdir of cgroup.

Add a VM_BUG_ON to mem_cgroup_lru_del_list() when we're about to wrap that
count - this has not been the only such incident.  Document that
lru_add_page_tail() is for Transparent HugePages by #ifdef around it.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/huge_memory.c
mm/memcontrol.c
mm/swap.c

index 76cc3f7..b3ffc21 100644 (file)
@@ -1229,7 +1229,6 @@ static void __split_huge_page_refcount(struct page *page)
 {
        int i;
        struct zone *zone = page_zone(page);
-       int zonestat;
        int tail_count = 0;
 
        /* prevent PageLRU to go away from under us, and freeze lru stats */
@@ -1317,15 +1316,6 @@ static void __split_huge_page_refcount(struct page *page)
        __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
        __mod_zone_page_state(zone, NR_ANON_PAGES, HPAGE_PMD_NR);
 
-       /*
-        * A hugepage counts for HPAGE_PMD_NR pages on the LRU statistics,
-        * so adjust those appropriately if this page is on the LRU.
-        */
-       if (PageLRU(page)) {
-               zonestat = NR_LRU_BASE + page_lru(page);
-               __mod_zone_page_state(zone, zonestat, -(HPAGE_PMD_NR-1));
-       }
-
        ClearPageCompound(page);
        compound_unlock(page);
        spin_unlock_irq(&zone->lru_lock);
index 609c49f..9f2f646 100644 (file)
@@ -1071,6 +1071,7 @@ void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru)
        VM_BUG_ON(!memcg);
        mz = page_cgroup_zoneinfo(memcg, page);
        /* huge page split is done under lru_lock. so, we have no races. */
+       VM_BUG_ON(MEM_CGROUP_ZSTAT(mz, lru) < (1 << compound_order(page)));
        MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
 }
 
@@ -2465,9 +2466,7 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
 void mem_cgroup_split_huge_fixup(struct page *head)
 {
        struct page_cgroup *head_pc = lookup_page_cgroup(head);
-       struct mem_cgroup_per_zone *mz;
        struct page_cgroup *pc;
-       enum lru_list lru;
        int i;
 
        if (mem_cgroup_disabled())
@@ -2478,15 +2477,8 @@ void mem_cgroup_split_huge_fixup(struct page *head)
                smp_wmb();/* see __commit_charge() */
                pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
        }
-       /*
-        * Tail pages will be added to LRU.
-        * We hold lru_lock,then,reduce counter directly.
-        */
-       lru = page_lru(head);
-       mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head);
-       MEM_CGROUP_ZSTAT(mz, lru) -= HPAGE_PMD_NR - 1;
 }
-#endif
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
 /**
  * mem_cgroup_move_account - move account of the page
index ddccf8e..db6defa 100644 (file)
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -650,6 +650,7 @@ void __pagevec_release(struct pagevec *pvec)
 
 EXPORT_SYMBOL(__pagevec_release);
 
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 /* used by __split_huge_page_refcount() */
 void lru_add_page_tail(struct zone* zone,
                       struct page *page, struct page *page_tail)
@@ -666,8 +667,6 @@ void lru_add_page_tail(struct zone* zone,
        SetPageLRU(page_tail);
 
        if (page_evictable(page_tail, NULL)) {
-               struct lruvec *lruvec;
-
                if (PageActive(page)) {
                        SetPageActive(page_tail);
                        active = 1;
@@ -677,18 +676,28 @@ void lru_add_page_tail(struct zone* zone,
                        lru = LRU_INACTIVE_ANON;
                }
                update_page_reclaim_stat(zone, page_tail, file, active);
-               lruvec = mem_cgroup_lru_add_list(zone, page_tail, lru);
-               if (likely(PageLRU(page)))
-                       list_add(&page_tail->lru, page->lru.prev);
-               else
-                       list_add(&page_tail->lru, lruvec->lists[lru].prev);
-               __mod_zone_page_state(zone, NR_LRU_BASE + lru,
-                                     hpage_nr_pages(page_tail));
        } else {
                SetPageUnevictable(page_tail);
-               add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE);
+               lru = LRU_UNEVICTABLE;
+       }
+
+       if (likely(PageLRU(page)))
+               list_add_tail(&page_tail->lru, &page->lru);
+       else {
+               struct list_head *list_head;
+               /*
+                * Head page has not yet been counted, as an hpage,
+                * so we must account for each subpage individually.
+                *
+                * Use the standard add function to put page_tail on the list,
+                * but then correct its position so they all end up in order.
+                */
+               add_page_to_lru_list(zone, page_tail, lru);
+               list_head = page_tail->lru.prev;
+               list_move_tail(&page_tail->lru, list_head);
        }
 }
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
 static void ____pagevec_lru_add_fn(struct page *page, void *arg)
 {