On Wednesday 05 September 2012 13:08:47 Mel Gorman wrote:

On Tue, Sep 04, 2012 at 03:26:22PM +0200, Bartlomiej Zolnierkiewicz wrote:

quoted

Isolated free pages shouldn't be accounted to NR_FREE_PAGES counter.
Fix it by properly decreasing/increasing NR_FREE_PAGES counter in
set_migratetype_isolate()/unset_migratetype_isolate() and removing
counter adjustment for isolated pages from free_one_page() and
split_free_page().

Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Michal Nazarewicz <redacted>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <redacted>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
---
 mm/page_alloc.c     |  7 +++++--
 mm/page_isolation.c | 13 ++++++++++---
 2 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index e9da55c..3acdf0f 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c

@@ -691,7 +691,8 @@ static void free_one_page(struct zone *zone, struct page *page, int order,
 	zone->pages_scanned = 0;
 
 	__free_one_page(page, zone, order, migratetype);
-	__mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
+	if (migratetype != MIGRATE_ISOLATE)
+		__mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
 	spin_unlock(&zone->lock);
 }
 

@@ -1414,7 +1415,9 @@ int split_free_page(struct page *page, bool check_wmark)
 	list_del(&page->lru);
 	zone->free_area[order].nr_free--;
 	rmv_page_order(page);
-	__mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
+
+	if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
+		__mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
 

Are you *sure* about this part? The page is already free so the
NR_FREE_PAGES counters should already be correct. It feels to me that it

The isolated page is not counted as free so the counter shouldn't be
adjusted here (IOW we shouldn't decrease the counter as it was already
decreased in set_migratetype_isolate() earlier).

should be the caller that fixes up NR_FREE_PAGES if necessary.

split_free_page() is only called from isolate_freepages_block() so
the fixup for MIGRATE_ISOLATE case can be added there if needed..

Have you tested this with THP? I have a suspicion that the free page
accounting gets broken when page migration is used there.

No I haven't tested it with THP but I don't see how it could break
because of this patch (please explain the potential failure scenario
a bit more).

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung Poland R&D Center

quoted

 	/* Split into individual pages */
 	set_page_refcounted(page);

diff --git a/mm/page_isolation.c b/mm/page_isolation.c
index 247d1f1..d210cc8 100644
--- a/mm/page_isolation.c
+++ b/mm/page_isolation.c

@@ -76,8 +76,12 @@ int set_migratetype_isolate(struct page *page)
 
 out:
 	if (!ret) {
+		unsigned long nr_pages;
+
 		set_pageblock_isolate(page);
-		move_freepages_block(zone, page, MIGRATE_ISOLATE);
+		nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE);
+
+		__mod_zone_page_state(zone, NR_FREE_PAGES, -nr_pages);
 	}
 
 	spin_unlock_irqrestore(&zone->lock, flags);

@@ -89,12 +93,15 @@ out:
 void unset_migratetype_isolate(struct page *page, unsigned migratetype)
 {
 	struct zone *zone;
-	unsigned long flags;
+	unsigned long flags, nr_pages;
+
 	zone = page_zone(page);
+

unnecessary whitespace change.

quoted

 	spin_lock_irqsave(&zone->lock, flags);
 	if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
 		goto out;
-	move_freepages_block(zone, page, migratetype);
+	nr_pages = move_freepages_block(zone, page, migratetype);
+	__mod_zone_page_state(zone, NR_FREE_PAGES, nr_pages);
 	restore_pageblock_isolate(page, migratetype);
 out:
 	spin_unlock_irqrestore(&zone->lock, flags);

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>