Be more agressive about stealing when MIGRATE_RECLAIMABLE allocations fallback
authorMel Gorman <mel@csn.ul.ie>
Tue, 16 Oct 2007 08:25:55 +0000 (01:25 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Tue, 16 Oct 2007 16:43:00 +0000 (09:43 -0700)
MIGRATE_RECLAIMABLE allocations tend to be very bursty in nature like when
updatedb starts.  It is likely this will occur in situations where MAX_ORDER
blocks of pages are not free.  This means that updatedb can scatter
MIGRATE_RECLAIMABLE pages throughout the address space.  This patch is more
agressive about stealing blocks of pages for MIGRATE_RECLAIMABLE.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/page_alloc.c

index e1d87ee1d9c6fac46de70ba00fe55980f3685349..b864584c92b467d384b21ab93685052328542cab 100644 (file)
@@ -821,11 +821,23 @@ retry:
 
                        /*
                         * If breaking a large block of pages, move all free
 
                        /*
                         * If breaking a large block of pages, move all free
-                        * pages to the preferred allocation list
+                        * pages to the preferred allocation list. If falling
+                        * back for a reclaimable kernel allocation, be more
+                        * agressive about taking ownership of free pages
                         */
                         */
-                       if (unlikely(current_order >= MAX_ORDER / 2)) {
+                       if (unlikely(current_order >= MAX_ORDER / 2) ||
+                                       start_migratetype == MIGRATE_RECLAIMABLE) {
+                               unsigned long pages;
+                               pages = move_freepages_block(zone, page,
+                                                               start_migratetype);
+
+                               /* Claim the whole block if over half of it is free */
+                               if ((pages << current_order) >= (1 << (MAX_ORDER-2)) &&
+                                               migratetype != MIGRATE_HIGHATOMIC)
+                                       set_pageblock_migratetype(page,
+                                                               start_migratetype);
+
                                migratetype = start_migratetype;
                                migratetype = start_migratetype;
-                               move_freepages_block(zone, page, migratetype);
                        }
 
                        /* Remove the page from the freelists */
                        }
 
                        /* Remove the page from the freelists */