mm: fix for infinite churning of mlocked pages
authorMinchan Kim <minchan.kim@gmail.com>
Wed, 26 Aug 2009 21:29:23 +0000 (14:29 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 27 Aug 2009 03:06:52 +0000 (20:06 -0700)
An mlocked page might lose the isolatation race.  This causes the page to
clear PG_mlocked while it remains in a VM_LOCKED vma.  This means it can
be put onto the [in]active list.  We can rescue it by using try_to_unmap()
in shrink_page_list().

But now, As Wu Fengguang pointed out, vmscan has a bug.  If the page has
PG_referenced, it can't reach try_to_unmap() in shrink_page_list() but is
put into the active list.  If the page is referenced repeatedly, it can
remain on the [in]active list without being moving to the unevictable
list.

This patch fixes it.

Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <<kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/rmap.c
mm/vmscan.c

index 836c6c63e1f2dbee16a922f7195da2f3ac237d83..0895b5c7cbff35fc2e5eb8dbe4b2ee8eeacf27da 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -358,6 +358,7 @@ static int page_referenced_one(struct page *page,
         */
        if (vma->vm_flags & VM_LOCKED) {
                *mapcount = 1;  /* break early from loop */
+               *vm_flags |= VM_LOCKED;
                goto out_unmap;
        }
 
index dea7abd310980daea1fa6c5a0c850a972fe37a21..94e86dd6954c295830478011fd8e71465f1a9f2d 100644 (file)
@@ -630,9 +630,14 @@ static unsigned long shrink_page_list(struct list_head *page_list,
 
                referenced = page_referenced(page, 1,
                                                sc->mem_cgroup, &vm_flags);
-               /* In active use or really unfreeable?  Activate it. */
+               /*
+                * In active use or really unfreeable?  Activate it.
+                * If page which have PG_mlocked lost isoltation race,
+                * try_to_unmap moves it to unevictable list
+                */
                if (sc->order <= PAGE_ALLOC_COSTLY_ORDER &&
-                                       referenced && page_mapping_inuse(page))
+                                       referenced && page_mapping_inuse(page)
+                                       && !(vm_flags & VM_LOCKED))
                        goto activate_locked;
 
                /*