2 * linux/mm/compaction.c
4 * Memory compaction for the reduction of external fragmentation. Note that
5 * this heavily depends upon page migration to do all the real heavy
8 * Copyright IBM Corp. 2007-2010 Mel Gorman <mel@csn.ul.ie>
10 #include <linux/swap.h>
11 #include <linux/migrate.h>
12 #include <linux/compaction.h>
13 #include <linux/mm_inline.h>
14 #include <linux/backing-dev.h>
15 #include <linux/sysctl.h>
16 #include <linux/sysfs.h>
19 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
21 #define CREATE_TRACE_POINTS
22 #include <trace/events/compaction.h>
24 static unsigned long release_freepages(struct list_head *freelist)
26 struct page *page, *next;
27 unsigned long count = 0;
29 list_for_each_entry_safe(page, next, freelist, lru) {
38 static void map_pages(struct list_head *list)
42 list_for_each_entry(page, list, lru) {
43 arch_alloc_page(page, 0);
44 kernel_map_pages(page, 1, 1);
48 static inline bool migrate_async_suitable(int migratetype)
50 return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
54 * Compaction requires the taking of some coarse locks that are potentially
55 * very heavily contended. Check if the process needs to be scheduled or
56 * if the lock is contended. For async compaction, back out in the event
57 * if contention is severe. For sync compaction, schedule.
59 * Returns true if the lock is held.
60 * Returns false if the lock is released and compaction should abort
62 static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags,
63 bool locked, struct compact_control *cc)
65 if (need_resched() || spin_is_contended(lock)) {
67 spin_unlock_irqrestore(lock, *flags);
71 /* async aborts if taking too long or contended */
74 *cc->contended = true;
79 if (fatal_signal_pending(current))
84 spin_lock_irqsave(lock, *flags);
88 static inline bool compact_trylock_irqsave(spinlock_t *lock,
89 unsigned long *flags, struct compact_control *cc)
91 return compact_checklock_irqsave(lock, flags, false, cc);
95 * Isolate free pages onto a private freelist. Caller must hold zone->lock.
96 * If @strict is true, will abort returning 0 on any invalid PFNs or non-free
97 * pages inside of the pageblock (even though it may still end up isolating
100 static unsigned long isolate_freepages_block(unsigned long blockpfn,
101 unsigned long end_pfn,
102 struct list_head *freelist,
105 int nr_scanned = 0, total_isolated = 0;
108 cursor = pfn_to_page(blockpfn);
110 /* Isolate free pages. This assumes the block is valid */
111 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
113 struct page *page = cursor;
115 if (!pfn_valid_within(blockpfn)) {
122 if (!PageBuddy(page)) {
128 /* Found a free page, break it into order-0 pages */
129 isolated = split_free_page(page);
130 if (!isolated && strict)
132 total_isolated += isolated;
133 for (i = 0; i < isolated; i++) {
134 list_add(&page->lru, freelist);
138 /* If a page was split, advance to the end of it */
140 blockpfn += isolated - 1;
141 cursor += isolated - 1;
145 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
146 return total_isolated;
150 * isolate_freepages_range() - isolate free pages.
151 * @start_pfn: The first PFN to start isolating.
152 * @end_pfn: The one-past-last PFN.
154 * Non-free pages, invalid PFNs, or zone boundaries within the
155 * [start_pfn, end_pfn) range are considered errors, cause function to
156 * undo its actions and return zero.
158 * Otherwise, function returns one-past-the-last PFN of isolated page
159 * (which may be greater then end_pfn if end fell in a middle of
163 isolate_freepages_range(unsigned long start_pfn, unsigned long end_pfn)
165 unsigned long isolated, pfn, block_end_pfn, flags;
166 struct zone *zone = NULL;
169 if (pfn_valid(start_pfn))
170 zone = page_zone(pfn_to_page(start_pfn));
172 for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
173 if (!pfn_valid(pfn) || zone != page_zone(pfn_to_page(pfn)))
177 * On subsequent iterations ALIGN() is actually not needed,
178 * but we keep it that we not to complicate the code.
180 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
181 block_end_pfn = min(block_end_pfn, end_pfn);
183 spin_lock_irqsave(&zone->lock, flags);
184 isolated = isolate_freepages_block(pfn, block_end_pfn,
186 spin_unlock_irqrestore(&zone->lock, flags);
189 * In strict mode, isolate_freepages_block() returns 0 if
190 * there are any holes in the block (ie. invalid PFNs or
197 * If we managed to isolate pages, it is always (1 << n) *
198 * pageblock_nr_pages for some non-negative n. (Max order
199 * page may span two pageblocks).
203 /* split_free_page does not map the pages */
204 map_pages(&freelist);
207 /* Loop terminated early, cleanup. */
208 release_freepages(&freelist);
212 /* We don't use freelists for anything. */
216 /* Update the number of anon and file isolated pages in the zone */
217 static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc)
220 unsigned int count[2] = { 0, };
222 list_for_each_entry(page, &cc->migratepages, lru)
223 count[!!page_is_file_cache(page)]++;
225 /* If locked we can use the interrupt unsafe versions */
227 __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
228 __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
230 mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
231 mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
235 /* Similar to reclaim, but different enough that they don't share logic */
236 static bool too_many_isolated(struct zone *zone)
238 unsigned long active, inactive, isolated;
240 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
241 zone_page_state(zone, NR_INACTIVE_ANON);
242 active = zone_page_state(zone, NR_ACTIVE_FILE) +
243 zone_page_state(zone, NR_ACTIVE_ANON);
244 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
245 zone_page_state(zone, NR_ISOLATED_ANON);
247 return isolated > (inactive + active) / 2;
251 * isolate_migratepages_range() - isolate all migrate-able pages in range.
252 * @zone: Zone pages are in.
253 * @cc: Compaction control structure.
254 * @low_pfn: The first PFN of the range.
255 * @end_pfn: The one-past-the-last PFN of the range.
257 * Isolate all pages that can be migrated from the range specified by
258 * [low_pfn, end_pfn). Returns zero if there is a fatal signal
259 * pending), otherwise PFN of the first page that was not scanned
260 * (which may be both less, equal to or more then end_pfn).
262 * Assumes that cc->migratepages is empty and cc->nr_migratepages is
265 * Apart from cc->migratepages and cc->nr_migratetypes this function
266 * does not modify any cc's fields, in particular it does not modify
267 * (or read for that matter) cc->migrate_pfn.
270 isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
271 unsigned long low_pfn, unsigned long end_pfn)
273 unsigned long last_pageblock_nr = 0, pageblock_nr;
274 unsigned long nr_scanned = 0, nr_isolated = 0;
275 struct list_head *migratelist = &cc->migratepages;
276 isolate_mode_t mode = 0;
277 struct lruvec *lruvec;
282 * Ensure that there are not too many pages isolated from the LRU
283 * list by either parallel reclaimers or compaction. If there are,
284 * delay for some time until fewer pages are isolated
286 while (unlikely(too_many_isolated(zone))) {
287 /* async migration should just abort */
291 congestion_wait(BLK_RW_ASYNC, HZ/10);
293 if (fatal_signal_pending(current))
297 /* Time to isolate some pages for migration */
299 spin_lock_irqsave(&zone->lru_lock, flags);
301 for (; low_pfn < end_pfn; low_pfn++) {
304 /* give a chance to irqs before checking need_resched() */
305 if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
306 spin_unlock_irqrestore(&zone->lru_lock, flags);
310 /* Check if it is ok to still hold the lock */
311 locked = compact_checklock_irqsave(&zone->lru_lock, &flags,
317 * migrate_pfn does not necessarily start aligned to a
318 * pageblock. Ensure that pfn_valid is called when moving
319 * into a new MAX_ORDER_NR_PAGES range in case of large
320 * memory holes within the zone
322 if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
323 if (!pfn_valid(low_pfn)) {
324 low_pfn += MAX_ORDER_NR_PAGES - 1;
329 if (!pfn_valid_within(low_pfn))
334 * Get the page and ensure the page is within the same zone.
335 * See the comment in isolate_freepages about overlapping
336 * nodes. It is deliberate that the new zone lock is not taken
337 * as memory compaction should not move pages between nodes.
339 page = pfn_to_page(low_pfn);
340 if (page_zone(page) != zone)
348 * For async migration, also only scan in MOVABLE blocks. Async
349 * migration is optimistic to see if the minimum amount of work
350 * satisfies the allocation
352 pageblock_nr = low_pfn >> pageblock_order;
353 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
354 !migrate_async_suitable(get_pageblock_migratetype(page))) {
355 low_pfn += pageblock_nr_pages;
356 low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1;
357 last_pageblock_nr = pageblock_nr;
365 * PageLRU is set, and lru_lock excludes isolation,
366 * splitting and collapsing (collapsing has already
367 * happened if PageLRU is set).
369 if (PageTransHuge(page)) {
370 low_pfn += (1 << compound_order(page)) - 1;
375 mode |= ISOLATE_ASYNC_MIGRATE;
377 lruvec = mem_cgroup_page_lruvec(page, zone);
379 /* Try isolate the page */
380 if (__isolate_lru_page(page, mode) != 0)
383 VM_BUG_ON(PageTransCompound(page));
385 /* Successfully isolated */
386 del_page_from_lru_list(page, lruvec, page_lru(page));
387 list_add(&page->lru, migratelist);
388 cc->nr_migratepages++;
391 /* Avoid isolating too much */
392 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
398 acct_isolated(zone, locked, cc);
401 spin_unlock_irqrestore(&zone->lru_lock, flags);
403 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
408 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
409 #ifdef CONFIG_COMPACTION
411 /* Returns true if the page is within a block suitable for migration to */
412 static bool suitable_migration_target(struct page *page)
415 int migratetype = get_pageblock_migratetype(page);
417 /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
418 if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE)
421 /* If the page is a large free page, then allow migration */
422 if (PageBuddy(page) && page_order(page) >= pageblock_order)
425 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
426 if (migrate_async_suitable(migratetype))
429 /* Otherwise skip the block */
434 * Returns the start pfn of the last page block in a zone. This is the starting
435 * point for full compaction of a zone. Compaction searches for free pages from
436 * the end of each zone, while isolate_freepages_block scans forward inside each
439 static unsigned long start_free_pfn(struct zone *zone)
441 unsigned long free_pfn;
442 free_pfn = zone->zone_start_pfn + zone->spanned_pages;
443 free_pfn &= ~(pageblock_nr_pages-1);
448 * Based on information in the current compact_control, find blocks
449 * suitable for isolating free pages from and then isolate them.
451 static void isolate_freepages(struct zone *zone,
452 struct compact_control *cc)
455 unsigned long high_pfn, low_pfn, pfn, zone_end_pfn, end_pfn;
457 int nr_freepages = cc->nr_freepages;
458 struct list_head *freelist = &cc->freepages;
461 * Initialise the free scanner. The starting point is where we last
462 * scanned from (or the end of the zone if starting). The low point
463 * is the end of the pageblock the migration scanner is using.
466 low_pfn = cc->migrate_pfn + pageblock_nr_pages;
469 * Take care that if the migration scanner is at the end of the zone
470 * that the free scanner does not accidentally move to the next zone
471 * in the next isolation cycle.
473 high_pfn = min(low_pfn, pfn);
475 zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
478 * Isolate free pages until enough are available to migrate the
479 * pages on cc->migratepages. We stop searching if the migrate
480 * and free page scanners meet or enough free pages are isolated.
482 for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
483 pfn -= pageblock_nr_pages) {
484 unsigned long isolated;
490 * Check for overlapping nodes/zones. It's possible on some
491 * configurations to have a setup like
493 * i.e. it's possible that all pages within a zones range of
494 * pages do not belong to a single zone.
496 page = pfn_to_page(pfn);
497 if (page_zone(page) != zone)
500 /* Check the block is suitable for migration */
501 if (!suitable_migration_target(page))
505 * Found a block suitable for isolating free pages from. Now
506 * we disabled interrupts, double check things are ok and
507 * isolate the pages. This is to minimise the time IRQs
513 * The zone lock must be held to isolate freepages. This
514 * unfortunately this is a very coarse lock and can be
515 * heavily contended if there are parallel allocations
516 * or parallel compactions. For async compaction do not
519 if (!compact_trylock_irqsave(&zone->lock, &flags, cc))
521 if (suitable_migration_target(page)) {
522 end_pfn = min(pfn + pageblock_nr_pages, zone_end_pfn);
523 isolated = isolate_freepages_block(pfn, end_pfn,
525 nr_freepages += isolated;
527 spin_unlock_irqrestore(&zone->lock, flags);
530 * Record the highest PFN we isolated pages from. When next
531 * looking for free pages, the search will restart here as
532 * page migration may have returned some pages to the allocator
535 high_pfn = max(high_pfn, pfn);
538 * If the free scanner has wrapped, update
539 * compact_cached_free_pfn to point to the highest
540 * pageblock with free pages. This reduces excessive
541 * scanning of full pageblocks near the end of the
544 if (cc->order > 0 && cc->wrapped)
545 zone->compact_cached_free_pfn = high_pfn;
549 /* split_free_page does not map the pages */
552 cc->free_pfn = high_pfn;
553 cc->nr_freepages = nr_freepages;
555 /* If compact_cached_free_pfn is reset then set it now */
556 if (cc->order > 0 && !cc->wrapped &&
557 zone->compact_cached_free_pfn == start_free_pfn(zone))
558 zone->compact_cached_free_pfn = high_pfn;
562 * This is a migrate-callback that "allocates" freepages by taking pages
563 * from the isolated freelists in the block we are migrating to.
565 static struct page *compaction_alloc(struct page *migratepage,
569 struct compact_control *cc = (struct compact_control *)data;
570 struct page *freepage;
572 /* Isolate free pages if necessary */
573 if (list_empty(&cc->freepages)) {
574 isolate_freepages(cc->zone, cc);
576 if (list_empty(&cc->freepages))
580 freepage = list_entry(cc->freepages.next, struct page, lru);
581 list_del(&freepage->lru);
588 * We cannot control nr_migratepages and nr_freepages fully when migration is
589 * running as migrate_pages() has no knowledge of compact_control. When
590 * migration is complete, we count the number of pages on the lists by hand.
592 static void update_nr_listpages(struct compact_control *cc)
594 int nr_migratepages = 0;
595 int nr_freepages = 0;
598 list_for_each_entry(page, &cc->migratepages, lru)
600 list_for_each_entry(page, &cc->freepages, lru)
603 cc->nr_migratepages = nr_migratepages;
604 cc->nr_freepages = nr_freepages;
607 /* possible outcome of isolate_migratepages */
609 ISOLATE_ABORT, /* Abort compaction now */
610 ISOLATE_NONE, /* No pages isolated, continue scanning */
611 ISOLATE_SUCCESS, /* Pages isolated, migrate */
615 * Isolate all pages that can be migrated from the block pointed to by
616 * the migrate scanner within compact_control.
618 static isolate_migrate_t isolate_migratepages(struct zone *zone,
619 struct compact_control *cc)
621 unsigned long low_pfn, end_pfn;
623 /* Do not scan outside zone boundaries */
624 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
626 /* Only scan within a pageblock boundary */
627 end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages);
629 /* Do not cross the free scanner or scan within a memory hole */
630 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
631 cc->migrate_pfn = end_pfn;
635 /* Perform the isolation */
636 low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn);
638 return ISOLATE_ABORT;
640 cc->migrate_pfn = low_pfn;
642 return ISOLATE_SUCCESS;
645 static int compact_finished(struct zone *zone,
646 struct compact_control *cc)
649 unsigned long watermark;
651 if (fatal_signal_pending(current))
652 return COMPACT_PARTIAL;
655 * A full (order == -1) compaction run starts at the beginning and
656 * end of a zone; it completes when the migrate and free scanner meet.
657 * A partial (order > 0) compaction can start with the free scanner
658 * at a random point in the zone, and may have to restart.
660 if (cc->free_pfn <= cc->migrate_pfn) {
661 if (cc->order > 0 && !cc->wrapped) {
662 /* We started partway through; restart at the end. */
663 unsigned long free_pfn = start_free_pfn(zone);
664 zone->compact_cached_free_pfn = free_pfn;
665 cc->free_pfn = free_pfn;
667 return COMPACT_CONTINUE;
669 return COMPACT_COMPLETE;
672 /* We wrapped around and ended up where we started. */
673 if (cc->wrapped && cc->free_pfn <= cc->start_free_pfn)
674 return COMPACT_COMPLETE;
677 * order == -1 is expected when compacting via
678 * /proc/sys/vm/compact_memory
681 return COMPACT_CONTINUE;
683 /* Compaction run is not finished if the watermark is not met */
684 watermark = low_wmark_pages(zone);
685 watermark += (1 << cc->order);
687 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
688 return COMPACT_CONTINUE;
690 /* Direct compactor: Is a suitable page free? */
691 for (order = cc->order; order < MAX_ORDER; order++) {
692 /* Job done if page is free of the right migratetype */
693 if (!list_empty(&zone->free_area[order].free_list[cc->migratetype]))
694 return COMPACT_PARTIAL;
696 /* Job done if allocation would set block type */
697 if (order >= pageblock_order && zone->free_area[order].nr_free)
698 return COMPACT_PARTIAL;
701 return COMPACT_CONTINUE;
705 * compaction_suitable: Is this suitable to run compaction on this zone now?
707 * COMPACT_SKIPPED - If there are too few free pages for compaction
708 * COMPACT_PARTIAL - If the allocation would succeed without compaction
709 * COMPACT_CONTINUE - If compaction should run now
711 unsigned long compaction_suitable(struct zone *zone, int order)
714 unsigned long watermark;
717 * order == -1 is expected when compacting via
718 * /proc/sys/vm/compact_memory
721 return COMPACT_CONTINUE;
724 * Watermarks for order-0 must be met for compaction. Note the 2UL.
725 * This is because during migration, copies of pages need to be
726 * allocated and for a short time, the footprint is higher
728 watermark = low_wmark_pages(zone) + (2UL << order);
729 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
730 return COMPACT_SKIPPED;
733 * fragmentation index determines if allocation failures are due to
734 * low memory or external fragmentation
736 * index of -1000 implies allocations might succeed depending on
738 * index towards 0 implies failure is due to lack of memory
739 * index towards 1000 implies failure is due to fragmentation
741 * Only compact if a failure would be due to fragmentation.
743 fragindex = fragmentation_index(zone, order);
744 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
745 return COMPACT_SKIPPED;
747 if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
749 return COMPACT_PARTIAL;
751 return COMPACT_CONTINUE;
754 static int compact_zone(struct zone *zone, struct compact_control *cc)
758 ret = compaction_suitable(zone, cc->order);
760 case COMPACT_PARTIAL:
761 case COMPACT_SKIPPED:
762 /* Compaction is likely to fail */
764 case COMPACT_CONTINUE:
765 /* Fall through to compaction */
769 /* Setup to move all movable pages to the end of the zone */
770 cc->migrate_pfn = zone->zone_start_pfn;
773 /* Incremental compaction. Start where the last one stopped. */
774 cc->free_pfn = zone->compact_cached_free_pfn;
775 cc->start_free_pfn = cc->free_pfn;
777 /* Order == -1 starts at the end of the zone. */
778 cc->free_pfn = start_free_pfn(zone);
781 migrate_prep_local();
783 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
784 unsigned long nr_migrate, nr_remaining;
787 switch (isolate_migratepages(zone, cc)) {
789 ret = COMPACT_PARTIAL;
793 case ISOLATE_SUCCESS:
797 nr_migrate = cc->nr_migratepages;
798 err = migrate_pages(&cc->migratepages, compaction_alloc,
799 (unsigned long)cc, false,
800 cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC);
801 update_nr_listpages(cc);
802 nr_remaining = cc->nr_migratepages;
804 count_vm_event(COMPACTBLOCKS);
805 count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining);
807 count_vm_events(COMPACTPAGEFAILED, nr_remaining);
808 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
811 /* Release LRU pages not migrated */
813 putback_lru_pages(&cc->migratepages);
814 cc->nr_migratepages = 0;
815 if (err == -ENOMEM) {
816 ret = COMPACT_PARTIAL;
823 /* Release free pages and check accounting */
824 cc->nr_freepages -= release_freepages(&cc->freepages);
825 VM_BUG_ON(cc->nr_freepages != 0);
830 static unsigned long compact_zone_order(struct zone *zone,
831 int order, gfp_t gfp_mask,
832 bool sync, bool *contended)
834 struct compact_control cc = {
836 .nr_migratepages = 0,
838 .migratetype = allocflags_to_migratetype(gfp_mask),
841 .contended = contended,
843 INIT_LIST_HEAD(&cc.freepages);
844 INIT_LIST_HEAD(&cc.migratepages);
846 return compact_zone(zone, &cc);
849 int sysctl_extfrag_threshold = 500;
852 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
853 * @zonelist: The zonelist used for the current allocation
854 * @order: The order of the current allocation
855 * @gfp_mask: The GFP mask of the current allocation
856 * @nodemask: The allowed nodes to allocate from
857 * @sync: Whether migration is synchronous or not
859 * This is the main entry point for direct page compaction.
861 unsigned long try_to_compact_pages(struct zonelist *zonelist,
862 int order, gfp_t gfp_mask, nodemask_t *nodemask,
863 bool sync, bool *contended)
865 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
866 int may_enter_fs = gfp_mask & __GFP_FS;
867 int may_perform_io = gfp_mask & __GFP_IO;
870 int rc = COMPACT_SKIPPED;
873 * Check whether it is worth even starting compaction. The order check is
874 * made because an assumption is made that the page allocator can satisfy
875 * the "cheaper" orders without taking special steps
877 if (!order || !may_enter_fs || !may_perform_io)
880 count_vm_event(COMPACTSTALL);
882 /* Compact each zone in the list */
883 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
887 status = compact_zone_order(zone, order, gfp_mask, sync,
889 rc = max(status, rc);
891 /* If a normal allocation would succeed, stop compacting */
892 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0))
900 /* Compact all zones within a node */
901 static int __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
906 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
908 zone = &pgdat->node_zones[zoneid];
909 if (!populated_zone(zone))
912 cc->nr_freepages = 0;
913 cc->nr_migratepages = 0;
915 INIT_LIST_HEAD(&cc->freepages);
916 INIT_LIST_HEAD(&cc->migratepages);
918 if (cc->order == -1 || !compaction_deferred(zone, cc->order))
919 compact_zone(zone, cc);
922 int ok = zone_watermark_ok(zone, cc->order,
923 low_wmark_pages(zone), 0, 0);
924 if (ok && cc->order >= zone->compact_order_failed)
925 zone->compact_order_failed = cc->order + 1;
926 /* Currently async compaction is never deferred. */
927 else if (!ok && cc->sync)
928 defer_compaction(zone, cc->order);
931 VM_BUG_ON(!list_empty(&cc->freepages));
932 VM_BUG_ON(!list_empty(&cc->migratepages));
938 int compact_pgdat(pg_data_t *pgdat, int order)
940 struct compact_control cc = {
945 return __compact_pgdat(pgdat, &cc);
948 static int compact_node(int nid)
950 struct compact_control cc = {
955 return __compact_pgdat(NODE_DATA(nid), &cc);
958 /* Compact all nodes in the system */
959 static int compact_nodes(void)
963 /* Flush pending updates to the LRU lists */
966 for_each_online_node(nid)
969 return COMPACT_COMPLETE;
972 /* The written value is actually unused, all memory is compacted */
973 int sysctl_compact_memory;
975 /* This is the entry point for compacting all nodes via /proc/sys/vm */
976 int sysctl_compaction_handler(struct ctl_table *table, int write,
977 void __user *buffer, size_t *length, loff_t *ppos)
980 return compact_nodes();
985 int sysctl_extfrag_handler(struct ctl_table *table, int write,
986 void __user *buffer, size_t *length, loff_t *ppos)
988 proc_dointvec_minmax(table, write, buffer, length, ppos);
993 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
994 ssize_t sysfs_compact_node(struct device *dev,
995 struct device_attribute *attr,
996 const char *buf, size_t count)
1000 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1001 /* Flush pending updates to the LRU lists */
1002 lru_add_drain_all();
1009 static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
1011 int compaction_register_node(struct node *node)
1013 return device_create_file(&node->dev, &dev_attr_compact);
1016 void compaction_unregister_node(struct node *node)
1018 return device_remove_file(&node->dev, &dev_attr_compact);
1020 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
1022 #endif /* CONFIG_COMPACTION */