On Thu, Jan 09 2014, Joonsoo Kim wrote:
> If (MAX_ORDER-1) is greater than pageblock order, there is a possibility
> to merge different migratetype pages and to be linked in unintended
> freelist.
>
> While I test CMA, CMA pages are merged and linked into MOVABLE freelist
> by above issue and then, the pages change their migratetype to UNMOVABLE by
> try_to_steal_freepages(). After that, CMA to this region always fail.
>
> To prevent this, we should not merge the page on MIGRATE_(CMA|ISOLATE)
> freelist.

This is strange.  CMA regions are always multiplies of max-pages (or
pageblocks whichever is larger), so MOVABLE free pages should never be
inside of a CMA region.

If what you're describing happens, it looks like an issue somewhere
else.

> Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 2548b42..ea99cee 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -581,6 +581,15 @@ static inline void __free_one_page(struct page *page,
>  			__mod_zone_freepage_state(zone, 1 << order,
>  						  migratetype);
>  		} else {
> +			int buddy_mt = get_buddy_migratetype(buddy);
> +
> +			/* We don't want to merge cma, isolate pages */
> +			if (unlikely(order >= pageblock_order) &&
> +				migratetype != buddy_mt &&
> +				(migratetype >= MIGRATE_PCPTYPES ||
> +				buddy_mt >= MIGRATE_PCPTYPES)) {
> +				break;
> +			}
>  			list_del(&buddy->lru);
>  			zone->free_area[order].nr_free--;
>  			rmv_page_order(buddy);
> -- 
> 1.7.9.5
>

-- 
Best regards,                                         _     _
.o. | Liege of Serenely Enlightened Majesty of      o' \,=./ `o
..o | Computer Science,  Michał “mina86” Nazarewicz    (o o)
ooo +--<mpn@google.com>--<xmpp:mina86@jabber.org>--ooO--(_)--Ooo--