On Fri, Nov 18, 2011 at 05:43:08PM +0100, Marek Szyprowski wrote:

quoted

From: Michal Nazarewicz <redacted>

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 9dd443d..58d1a2e 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c

@@ -628,6 +628,18 @@ static void free_pcppages_bulk(struct zone *zone, int count,
 			page = list_entry(list->prev, struct page, lru);
 			/* must delete as __free_one_page list manipulates */
 			list_del(&page->lru);
+
+			/*
+			 * When page is isolated in set_migratetype_isolate()
+			 * function it's page_private is not changed since the
+			 * function has no way of knowing if it can touch it.
+			 * This means that when a page is on PCP list, it's
+			 * page_private no longer matches the desired migrate
+			 * type.
+			 */
+			if (get_pageblock_migratetype(page) == MIGRATE_ISOLATE)
+				set_page_private(page, MIGRATE_ISOLATE);
+

On Mon, 12 Dec 2011 14:42:35 +0100, Mel Gorman [off-list ref] wrote:

How much of a problem is this in practice?

IIRC, this lead to allocation being made from area marked as isolated
or some such.

[...] I'd go as far to say that it would be preferable to drain the
per-CPU lists after you set pageblocks MIGRATE_ISOLATE. The IPIs also have
overhead but it will be incurred for the rare rather than the common case.

I'll look into that.

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