On Wed, Jun 15, 2011 at 04:50:20PM -0700, Hugh Dickins wrote:
> On Wed, 15 Jun 2011, [email protected] wrote:
> > From: Andrea Arcangeli <[email protected]>
> > 
> > swapcache will reach the below code path in migrate_page_move_mapping,
> > and swapcache is accounted as NR_FILE_PAGES but it's not accounted as
> > NR_SHMEM.
> > 
> > This can make the NR_SHMEM counter underflow.
> > 
> > Signed-off-by: Andrea Arcangeli <[email protected]>
> > Reviewed-by: Minchan Kim <[email protected]>
> > Acked-by: Mel Gorman <[email protected]>
> > Cc: KOSAKI Motohiro <[email protected]>
> > Cc: Hugh Dickins <[email protected]>
> > Cc: <[email protected]>
> > Signed-off-by: Andrew Morton <[email protected]>
> > ---
> > 
> >  mm/migrate.c |    2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff -puN mm/migrate.c~mm-migratec-dont-account-swapcache-as-shmem 
> > mm/migrate.c
> > --- a/mm/migrate.c~mm-migratec-dont-account-swapcache-as-shmem
> > +++ a/mm/migrate.c
> > @@ -288,7 +288,7 @@ static int migrate_page_move_mapping(str
> >      */
> >     __dec_zone_page_state(page, NR_FILE_PAGES);
> >     __inc_zone_page_state(newpage, NR_FILE_PAGES);
> > -   if (PageSwapBacked(page)) {
> > +   if (mapping != &swapper_space && PageSwapBacked(page)) {
> >             __dec_zone_page_state(page, NR_SHMEM);
> >             __inc_zone_page_state(newpage, NR_SHMEM);
> >     }
> > _
> 
> This one breaks the CONFIG_MIGRATION=y CONFIG_SWAP=n build,
> because of the reference to swapper_space.
> 
> Noticing that now, I was about to follow up with a patch changing test to
>       if (PageSwapBacked(page) && !PageSwapCache(page)) {
> as Mel had the good instinct to suggest originally.
> 
> But then I read the comment above it which says "Note that anonymous
> pages are accounted for via NR_FILE_PAGES and NR_ANON_PAGES if they
> are mapped to swap space".  So shouldn't we actually be adjusting
> NR_ANON_PAGES in that case??

That I already thought of, NR_ANON_PAGES is handled all by rmap, and
when we are at that point of migration we know there's not a single
pte mapped so it doesn't need refiling to the other zone because it's
not accounted as anon.

I agree about the build failure. It wasn't obvious to me that
swapper_space wouldn't build with CONFIG_SWAP=n, sorry.

> If I'm not too late, let's hold back this patch for the moment.

Here a version using PageSwapCache to make CONFIG_SWAP=n happy ;)

===
Subject: migrate: don't account swapcache as shmem

From: Andrea Arcangeli <[email protected]>

swapcache will reach the below code path in migrate_page_move_mapping,
and swapcache is accounted as NR_FILE_PAGES but it's not accounted as
NR_SHMEM.

Hugh pointed out we must use PageSwapCache instead of comparing
mapping to &swapper_space, to avoid build failure with CONFIG_SWAP=n.

Signed-off-by: Andrea Arcangeli <[email protected]>
---

diff --git a/mm/migrate.c b/mm/migrate.c
index e4a5c91..2597a27 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -288,7 +288,7 @@ static int migrate_page_move_mapping(struct address_space 
*mapping,
         */
        __dec_zone_page_state(page, NR_FILE_PAGES);
        __inc_zone_page_state(newpage, NR_FILE_PAGES);
-       if (PageSwapBacked(page)) {
+       if (!PageSwapCache(page) && PageSwapBacked(page)) {
                __dec_zone_page_state(page, NR_SHMEM);
                __inc_zone_page_state(newpage, NR_SHMEM);
        }


_______________________________________________
stable mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/stable

Reply via email to