(10/31/13 12:24 AM), kosaki.motohiro@xxxxxxxxx wrote: > From: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx> > > When __rmqueue_fallback() don't find out a free block with the same size > of required, it splits a larger page and puts back rest peiece of the page > to free list. > > But it has one serious mistake. When putting back, __rmqueue_fallback() > always use start_migratetype if type is not CMA. However, __rmqueue_fallback() > is only called when all of start_migratetype queue are empty. That said, > __rmqueue_fallback always put back memory to wrong queue except > try_to_steal_freepages() changed pageblock type (i.e. requested size is > smaller than half of page block). Finally, antifragmentation framework > increase fragmenation instead of decrease. > > Mel's original anti fragmentation do the right thing. But commit 47118af076 > (mm: mmzone: MIGRATE_CMA migration type added) broke it. > > This patch restores sane and old behavior. > > Cc: Mel Gorman <mgorman@xxxxxxx> > Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx> > --- > mm/page_alloc.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index dd886fa..ea7bb9a 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -1101,7 +1101,7 @@ __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) > */ > expand(zone, page, order, current_order, area, > is_migrate_cma(migratetype) > - ? migratetype : start_migratetype); > + ? migratetype : new_type); Oops, this can be simplified to following because try_to_steal_freepages() has cma check. - expand(zone, page, order, current_order, area, - is_migrate_cma(migratetype) - ? migratetype : start_migratetype); + expand(zone, page, order, current_order, area, new_type); -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>