Commit dd5d241e authored by Mel Gorman's avatar Mel Gorman Committed by Linus Torvalds
Browse files

page-allocator: always change pageblock ownership when anti-fragmentation is disabled

On low-memory systems, anti-fragmentation gets disabled as fragmentation
cannot be avoided on a sufficiently large boundary to be worthwhile.  Once
disabled, there is a period of time when all the pageblocks are marked
MOVABLE and the expectation is that they get marked UNMOVABLE at each call
to __rmqueue_fallback().

However, when MAX_ORDER is large the pageblocks do not change ownership
because the normal criteria are not met.  This has the effect of
prematurely breaking up too many large contiguous blocks.  This is most
serious on NOMMU systems which depend on high-order allocations to boot.
This patch causes pageblocks to change ownership on every fallback when
anti-fragmentation is disabled.  This prevents the large blocks being
prematurely broken up.

This is a fix to commit 49255c61

allocator: move check for disabled anti-fragmentation out of fastpath] and
the problem affects 2.6.31-rc8.
Signed-off-by: default avatarMel Gorman <>
Tested-by: default avatarPaul Mundt <>
Cc: David Howells <>
Cc: Pekka Enberg <>
Acked-by: default avatarGreg Ungerer <>
Signed-off-by: default avatarAndrew Morton <>
Signed-off-by: default avatarLinus Torvalds <>
parent a190887b
......@@ -817,13 +817,15 @@ __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
* agressive about taking ownership of free pages
if (unlikely(current_order >= (pageblock_order >> 1)) ||
start_migratetype == MIGRATE_RECLAIMABLE) {
start_migratetype == MIGRATE_RECLAIMABLE ||
page_group_by_mobility_disabled) {
unsigned long pages;
pages = move_freepages_block(zone, page,
/* Claim the whole block if over half of it is free */
if (pages >= (1 << (pageblock_order-1)))
if (pages >= (1 << (pageblock_order-1)) ||
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment