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>
17 #include <linux/balloon_compaction.h>
18 #include <linux/page-isolation.h>
21 #ifdef CONFIG_COMPACTION
22 static inline void count_compact_event(enum vm_event_item item)
27 static inline void count_compact_events(enum vm_event_item item, long delta)
29 count_vm_events(item, delta);
32 #define count_compact_event(item) do { } while (0)
33 #define count_compact_events(item, delta) do { } while (0)
36 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/compaction.h>
41 static unsigned long release_freepages(struct list_head *freelist)
43 struct page *page, *next;
44 unsigned long count = 0;
46 list_for_each_entry_safe(page, next, freelist, lru) {
55 static void map_pages(struct list_head *list)
59 list_for_each_entry(page, list, lru) {
60 arch_alloc_page(page, 0);
61 kernel_map_pages(page, 1, 1);
65 static inline bool migrate_async_suitable(int migratetype)
67 return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
70 #ifdef CONFIG_COMPACTION
71 /* Returns true if the pageblock should be scanned for pages to isolate. */
72 static inline bool isolation_suitable(struct compact_control *cc,
75 if (cc->ignore_skip_hint)
78 return !get_pageblock_skip(page);
82 * This function is called to clear all cached information on pageblocks that
83 * should be skipped for page isolation when the migrate and free page scanner
86 static void __reset_isolation_suitable(struct zone *zone)
88 unsigned long start_pfn = zone->zone_start_pfn;
89 unsigned long end_pfn = zone_end_pfn(zone);
92 zone->compact_cached_migrate_pfn = start_pfn;
93 zone->compact_cached_free_pfn = end_pfn;
94 zone->compact_blockskip_flush = false;
96 /* Walk the zone and mark every pageblock as suitable for isolation */
97 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
105 page = pfn_to_page(pfn);
106 if (zone != page_zone(page))
109 clear_pageblock_skip(page);
113 void reset_isolation_suitable(pg_data_t *pgdat)
117 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
118 struct zone *zone = &pgdat->node_zones[zoneid];
119 if (!populated_zone(zone))
122 /* Only flush if a full compaction finished recently */
123 if (zone->compact_blockskip_flush)
124 __reset_isolation_suitable(zone);
129 * If no pages were isolated then mark this pageblock to be skipped in the
130 * future. The information is later cleared by __reset_isolation_suitable().
132 static void update_pageblock_skip(struct compact_control *cc,
133 struct page *page, unsigned long nr_isolated,
134 bool migrate_scanner)
136 struct zone *zone = cc->zone;
141 unsigned long pfn = page_to_pfn(page);
142 set_pageblock_skip(page);
144 /* Update where compaction should restart */
145 if (migrate_scanner) {
146 if (!cc->finished_update_migrate &&
147 pfn > zone->compact_cached_migrate_pfn)
148 zone->compact_cached_migrate_pfn = pfn;
150 if (!cc->finished_update_free &&
151 pfn < zone->compact_cached_free_pfn)
152 zone->compact_cached_free_pfn = pfn;
157 static inline bool isolation_suitable(struct compact_control *cc,
163 static void update_pageblock_skip(struct compact_control *cc,
164 struct page *page, unsigned long nr_isolated,
165 bool migrate_scanner)
168 #endif /* CONFIG_COMPACTION */
170 static inline bool should_release_lock(spinlock_t *lock)
172 return need_resched() || spin_is_contended(lock);
176 * Compaction requires the taking of some coarse locks that are potentially
177 * very heavily contended. Check if the process needs to be scheduled or
178 * if the lock is contended. For async compaction, back out in the event
179 * if contention is severe. For sync compaction, schedule.
181 * Returns true if the lock is held.
182 * Returns false if the lock is released and compaction should abort
184 static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags,
185 bool locked, struct compact_control *cc)
187 if (should_release_lock(lock)) {
189 spin_unlock_irqrestore(lock, *flags);
193 /* async aborts if taking too long or contended */
195 cc->contended = true;
203 spin_lock_irqsave(lock, *flags);
207 static inline bool compact_trylock_irqsave(spinlock_t *lock,
208 unsigned long *flags, struct compact_control *cc)
210 return compact_checklock_irqsave(lock, flags, false, cc);
213 /* Returns true if the page is within a block suitable for migration to */
214 static bool suitable_migration_target(struct page *page)
216 int migratetype = get_pageblock_migratetype(page);
218 /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
219 if (migratetype == MIGRATE_RESERVE)
222 if (is_migrate_isolate(migratetype))
225 /* If the page is a large free page, then allow migration */
226 if (PageBuddy(page) && page_order(page) >= pageblock_order)
229 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
230 if (migrate_async_suitable(migratetype))
233 /* Otherwise skip the block */
238 * Isolate free pages onto a private freelist. Caller must hold zone->lock.
239 * If @strict is true, will abort returning 0 on any invalid PFNs or non-free
240 * pages inside of the pageblock (even though it may still end up isolating
243 static unsigned long isolate_freepages_block(struct compact_control *cc,
244 unsigned long blockpfn,
245 unsigned long end_pfn,
246 struct list_head *freelist,
249 int nr_scanned = 0, total_isolated = 0;
250 struct page *cursor, *valid_page = NULL;
251 unsigned long nr_strict_required = end_pfn - blockpfn;
255 cursor = pfn_to_page(blockpfn);
257 /* Isolate free pages. */
258 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
260 struct page *page = cursor;
263 if (!pfn_valid_within(blockpfn))
267 if (!PageBuddy(page))
271 * The zone lock must be held to isolate freepages.
272 * Unfortunately this is a very coarse lock and can be
273 * heavily contended if there are parallel allocations
274 * or parallel compactions. For async compaction do not
275 * spin on the lock and we acquire the lock as late as
278 locked = compact_checklock_irqsave(&cc->zone->lock, &flags,
283 /* Recheck this is a suitable migration target under lock */
284 if (!strict && !suitable_migration_target(page))
287 /* Recheck this is a buddy page under lock */
288 if (!PageBuddy(page))
291 /* Found a free page, break it into order-0 pages */
292 isolated = split_free_page(page);
293 if (!isolated && strict)
295 total_isolated += isolated;
296 for (i = 0; i < isolated; i++) {
297 list_add(&page->lru, freelist);
301 /* If a page was split, advance to the end of it */
303 blockpfn += isolated - 1;
304 cursor += isolated - 1;
308 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
311 * If strict isolation is requested by CMA then check that all the
312 * pages requested were isolated. If there were any failures, 0 is
313 * returned and CMA will fail.
315 if (strict && nr_strict_required > total_isolated)
319 spin_unlock_irqrestore(&cc->zone->lock, flags);
321 /* Update the pageblock-skip if the whole pageblock was scanned */
322 if (blockpfn == end_pfn)
323 update_pageblock_skip(cc, valid_page, total_isolated, false);
325 count_compact_events(COMPACTFREE_SCANNED, nr_scanned);
327 count_compact_events(COMPACTISOLATED, total_isolated);
328 return total_isolated;
332 * isolate_freepages_range() - isolate free pages.
333 * @start_pfn: The first PFN to start isolating.
334 * @end_pfn: The one-past-last PFN.
336 * Non-free pages, invalid PFNs, or zone boundaries within the
337 * [start_pfn, end_pfn) range are considered errors, cause function to
338 * undo its actions and return zero.
340 * Otherwise, function returns one-past-the-last PFN of isolated page
341 * (which may be greater then end_pfn if end fell in a middle of
345 isolate_freepages_range(struct compact_control *cc,
346 unsigned long start_pfn, unsigned long end_pfn)
348 unsigned long isolated, pfn, block_end_pfn;
351 for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
352 if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn)))
356 * On subsequent iterations ALIGN() is actually not needed,
357 * but we keep it that we not to complicate the code.
359 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
360 block_end_pfn = min(block_end_pfn, end_pfn);
362 isolated = isolate_freepages_block(cc, pfn, block_end_pfn,
366 * In strict mode, isolate_freepages_block() returns 0 if
367 * there are any holes in the block (ie. invalid PFNs or
374 * If we managed to isolate pages, it is always (1 << n) *
375 * pageblock_nr_pages for some non-negative n. (Max order
376 * page may span two pageblocks).
380 /* split_free_page does not map the pages */
381 map_pages(&freelist);
384 /* Loop terminated early, cleanup. */
385 release_freepages(&freelist);
389 /* We don't use freelists for anything. */
393 /* Update the number of anon and file isolated pages in the zone */
394 static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc)
397 unsigned int count[2] = { 0, };
399 list_for_each_entry(page, &cc->migratepages, lru)
400 count[!!page_is_file_cache(page)]++;
402 /* If locked we can use the interrupt unsafe versions */
404 __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
405 __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
407 mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
408 mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
412 /* Similar to reclaim, but different enough that they don't share logic */
413 static bool too_many_isolated(struct zone *zone)
415 unsigned long active, inactive, isolated;
417 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
418 zone_page_state(zone, NR_INACTIVE_ANON);
419 active = zone_page_state(zone, NR_ACTIVE_FILE) +
420 zone_page_state(zone, NR_ACTIVE_ANON);
421 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
422 zone_page_state(zone, NR_ISOLATED_ANON);
424 return isolated > (inactive + active) / 2;
428 * isolate_migratepages_range() - isolate all migrate-able pages in range.
429 * @zone: Zone pages are in.
430 * @cc: Compaction control structure.
431 * @low_pfn: The first PFN of the range.
432 * @end_pfn: The one-past-the-last PFN of the range.
433 * @unevictable: true if it allows to isolate unevictable pages
435 * Isolate all pages that can be migrated from the range specified by
436 * [low_pfn, end_pfn). Returns zero if there is a fatal signal
437 * pending), otherwise PFN of the first page that was not scanned
438 * (which may be both less, equal to or more then end_pfn).
440 * Assumes that cc->migratepages is empty and cc->nr_migratepages is
443 * Apart from cc->migratepages and cc->nr_migratetypes this function
444 * does not modify any cc's fields, in particular it does not modify
445 * (or read for that matter) cc->migrate_pfn.
448 isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
449 unsigned long low_pfn, unsigned long end_pfn, bool unevictable)
451 unsigned long last_pageblock_nr = 0, pageblock_nr;
452 unsigned long nr_scanned = 0, nr_isolated = 0;
453 struct list_head *migratelist = &cc->migratepages;
454 isolate_mode_t mode = 0;
455 struct lruvec *lruvec;
458 struct page *page = NULL, *valid_page = NULL;
461 * Ensure that there are not too many pages isolated from the LRU
462 * list by either parallel reclaimers or compaction. If there are,
463 * delay for some time until fewer pages are isolated
465 while (unlikely(too_many_isolated(zone))) {
466 /* async migration should just abort */
470 congestion_wait(BLK_RW_ASYNC, HZ/10);
472 if (fatal_signal_pending(current))
476 /* Time to isolate some pages for migration */
478 for (; low_pfn < end_pfn; low_pfn++) {
479 /* give a chance to irqs before checking need_resched() */
480 if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) {
481 if (should_release_lock(&zone->lru_lock)) {
482 spin_unlock_irqrestore(&zone->lru_lock, flags);
488 * migrate_pfn does not necessarily start aligned to a
489 * pageblock. Ensure that pfn_valid is called when moving
490 * into a new MAX_ORDER_NR_PAGES range in case of large
491 * memory holes within the zone
493 if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
494 if (!pfn_valid(low_pfn)) {
495 low_pfn += MAX_ORDER_NR_PAGES - 1;
500 if (!pfn_valid_within(low_pfn))
505 * Get the page and ensure the page is within the same zone.
506 * See the comment in isolate_freepages about overlapping
507 * nodes. It is deliberate that the new zone lock is not taken
508 * as memory compaction should not move pages between nodes.
510 page = pfn_to_page(low_pfn);
511 if (page_zone(page) != zone)
517 /* If isolation recently failed, do not retry */
518 pageblock_nr = low_pfn >> pageblock_order;
519 if (!isolation_suitable(cc, page))
527 * For async migration, also only scan in MOVABLE blocks. Async
528 * migration is optimistic to see if the minimum amount of work
529 * satisfies the allocation
531 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
532 !migrate_async_suitable(get_pageblock_migratetype(page))) {
533 cc->finished_update_migrate = true;
538 * Check may be lockless but that's ok as we recheck later.
539 * It's possible to migrate LRU pages and balloon pages
540 * Skip any other type of page
542 if (!PageLRU(page)) {
543 if (unlikely(balloon_page_movable(page))) {
544 if (locked && balloon_page_isolate(page)) {
545 /* Successfully isolated */
546 cc->finished_update_migrate = true;
547 list_add(&page->lru, migratelist);
548 cc->nr_migratepages++;
550 goto check_compact_cluster;
557 * PageLRU is set. lru_lock normally excludes isolation
558 * splitting and collapsing (collapsing has already happened
559 * if PageLRU is set) but the lock is not necessarily taken
560 * here and it is wasteful to take it just to check transhuge.
561 * Check TransHuge without lock and skip the whole pageblock if
562 * it's either a transhuge or hugetlbfs page, as calling
563 * compound_order() without preventing THP from splitting the
564 * page underneath us may return surprising results.
566 if (PageTransHuge(page)) {
569 low_pfn += (1 << compound_order(page)) - 1;
573 /* Check if it is ok to still hold the lock */
574 locked = compact_checklock_irqsave(&zone->lru_lock, &flags,
576 if (!locked || fatal_signal_pending(current))
579 /* Recheck PageLRU and PageTransHuge under lock */
582 if (PageTransHuge(page)) {
583 low_pfn += (1 << compound_order(page)) - 1;
588 mode |= ISOLATE_ASYNC_MIGRATE;
591 mode |= ISOLATE_UNEVICTABLE;
593 lruvec = mem_cgroup_page_lruvec(page, zone);
595 /* Try isolate the page */
596 if (__isolate_lru_page(page, mode) != 0)
599 VM_BUG_ON(PageTransCompound(page));
601 /* Successfully isolated */
602 cc->finished_update_migrate = true;
603 del_page_from_lru_list(page, lruvec, page_lru(page));
604 list_add(&page->lru, migratelist);
605 cc->nr_migratepages++;
608 check_compact_cluster:
609 /* Avoid isolating too much */
610 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
618 low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1;
619 last_pageblock_nr = pageblock_nr;
622 acct_isolated(zone, locked, cc);
625 spin_unlock_irqrestore(&zone->lru_lock, flags);
627 /* Update the pageblock-skip if the whole pageblock was scanned */
628 if (low_pfn == end_pfn)
629 update_pageblock_skip(cc, valid_page, nr_isolated, true);
631 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
633 count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned);
635 count_compact_events(COMPACTISOLATED, nr_isolated);
640 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
641 #ifdef CONFIG_COMPACTION
643 * Based on information in the current compact_control, find blocks
644 * suitable for isolating free pages from and then isolate them.
646 static void isolate_freepages(struct zone *zone,
647 struct compact_control *cc)
650 unsigned long high_pfn, low_pfn, pfn, z_end_pfn, end_pfn;
651 int nr_freepages = cc->nr_freepages;
652 struct list_head *freelist = &cc->freepages;
655 * Initialise the free scanner. The starting point is where we last
656 * scanned from (or the end of the zone if starting). The low point
657 * is the end of the pageblock the migration scanner is using.
660 low_pfn = cc->migrate_pfn + pageblock_nr_pages;
663 * Take care that if the migration scanner is at the end of the zone
664 * that the free scanner does not accidentally move to the next zone
665 * in the next isolation cycle.
667 high_pfn = min(low_pfn, pfn);
669 z_end_pfn = zone_end_pfn(zone);
672 * Isolate free pages until enough are available to migrate the
673 * pages on cc->migratepages. We stop searching if the migrate
674 * and free page scanners meet or enough free pages are isolated.
676 for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
677 pfn -= pageblock_nr_pages) {
678 unsigned long isolated;
684 * Check for overlapping nodes/zones. It's possible on some
685 * configurations to have a setup like
687 * i.e. it's possible that all pages within a zones range of
688 * pages do not belong to a single zone.
690 page = pfn_to_page(pfn);
691 if (page_zone(page) != zone)
694 /* Check the block is suitable for migration */
695 if (!suitable_migration_target(page))
698 /* If isolation recently failed, do not retry */
699 if (!isolation_suitable(cc, page))
702 /* Found a block suitable for isolating free pages from */
706 * As pfn may not start aligned, pfn+pageblock_nr_page
707 * may cross a MAX_ORDER_NR_PAGES boundary and miss
708 * a pfn_valid check. Ensure isolate_freepages_block()
709 * only scans within a pageblock
711 end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
712 end_pfn = min(end_pfn, z_end_pfn);
713 isolated = isolate_freepages_block(cc, pfn, end_pfn,
715 nr_freepages += isolated;
718 * Record the highest PFN we isolated pages from. When next
719 * looking for free pages, the search will restart here as
720 * page migration may have returned some pages to the allocator
723 cc->finished_update_free = true;
724 high_pfn = max(high_pfn, pfn);
728 /* split_free_page does not map the pages */
731 cc->free_pfn = high_pfn;
732 cc->nr_freepages = nr_freepages;
736 * This is a migrate-callback that "allocates" freepages by taking pages
737 * from the isolated freelists in the block we are migrating to.
739 static struct page *compaction_alloc(struct page *migratepage,
743 struct compact_control *cc = (struct compact_control *)data;
744 struct page *freepage;
746 /* Isolate free pages if necessary */
747 if (list_empty(&cc->freepages)) {
748 isolate_freepages(cc->zone, cc);
750 if (list_empty(&cc->freepages))
754 freepage = list_entry(cc->freepages.next, struct page, lru);
755 list_del(&freepage->lru);
762 * We cannot control nr_migratepages and nr_freepages fully when migration is
763 * running as migrate_pages() has no knowledge of compact_control. When
764 * migration is complete, we count the number of pages on the lists by hand.
766 static void update_nr_listpages(struct compact_control *cc)
768 int nr_migratepages = 0;
769 int nr_freepages = 0;
772 list_for_each_entry(page, &cc->migratepages, lru)
774 list_for_each_entry(page, &cc->freepages, lru)
777 cc->nr_migratepages = nr_migratepages;
778 cc->nr_freepages = nr_freepages;
781 /* possible outcome of isolate_migratepages */
783 ISOLATE_ABORT, /* Abort compaction now */
784 ISOLATE_NONE, /* No pages isolated, continue scanning */
785 ISOLATE_SUCCESS, /* Pages isolated, migrate */
789 * Isolate all pages that can be migrated from the block pointed to by
790 * the migrate scanner within compact_control.
792 static isolate_migrate_t isolate_migratepages(struct zone *zone,
793 struct compact_control *cc)
795 unsigned long low_pfn, end_pfn;
797 /* Do not scan outside zone boundaries */
798 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
800 /* Only scan within a pageblock boundary */
801 end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages);
803 /* Do not cross the free scanner or scan within a memory hole */
804 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
805 cc->migrate_pfn = end_pfn;
809 /* Perform the isolation */
810 low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false);
811 if (!low_pfn || cc->contended)
812 return ISOLATE_ABORT;
814 cc->migrate_pfn = low_pfn;
816 return ISOLATE_SUCCESS;
819 static int compact_finished(struct zone *zone,
820 struct compact_control *cc)
823 unsigned long watermark;
825 if (fatal_signal_pending(current))
826 return COMPACT_PARTIAL;
828 /* Compaction run completes if the migrate and free scanner meet */
829 if (cc->free_pfn <= cc->migrate_pfn) {
831 * Mark that the PG_migrate_skip information should be cleared
832 * by kswapd when it goes to sleep. kswapd does not set the
833 * flag itself as the decision to be clear should be directly
834 * based on an allocation request.
836 if (!current_is_kswapd())
837 zone->compact_blockskip_flush = true;
839 return COMPACT_COMPLETE;
843 * order == -1 is expected when compacting via
844 * /proc/sys/vm/compact_memory
847 return COMPACT_CONTINUE;
849 /* Compaction run is not finished if the watermark is not met */
850 watermark = low_wmark_pages(zone);
851 watermark += (1 << cc->order);
853 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
854 return COMPACT_CONTINUE;
856 /* Direct compactor: Is a suitable page free? */
857 for (order = cc->order; order < MAX_ORDER; order++) {
858 struct free_area *area = &zone->free_area[order];
860 /* Job done if page is free of the right migratetype */
861 if (!list_empty(&area->free_list[cc->migratetype]))
862 return COMPACT_PARTIAL;
864 /* Job done if allocation would set block type */
865 if (cc->order >= pageblock_order && area->nr_free)
866 return COMPACT_PARTIAL;
869 return COMPACT_CONTINUE;
873 * compaction_suitable: Is this suitable to run compaction on this zone now?
875 * COMPACT_SKIPPED - If there are too few free pages for compaction
876 * COMPACT_PARTIAL - If the allocation would succeed without compaction
877 * COMPACT_CONTINUE - If compaction should run now
879 unsigned long compaction_suitable(struct zone *zone, int order)
882 unsigned long watermark;
885 * order == -1 is expected when compacting via
886 * /proc/sys/vm/compact_memory
889 return COMPACT_CONTINUE;
892 * Watermarks for order-0 must be met for compaction. Note the 2UL.
893 * This is because during migration, copies of pages need to be
894 * allocated and for a short time, the footprint is higher
896 watermark = low_wmark_pages(zone) + (2UL << order);
897 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
898 return COMPACT_SKIPPED;
901 * fragmentation index determines if allocation failures are due to
902 * low memory or external fragmentation
904 * index of -1000 implies allocations might succeed depending on
906 * index towards 0 implies failure is due to lack of memory
907 * index towards 1000 implies failure is due to fragmentation
909 * Only compact if a failure would be due to fragmentation.
911 fragindex = fragmentation_index(zone, order);
912 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
913 return COMPACT_SKIPPED;
915 if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
917 return COMPACT_PARTIAL;
919 return COMPACT_CONTINUE;
922 static int compact_zone(struct zone *zone, struct compact_control *cc)
925 unsigned long start_pfn = zone->zone_start_pfn;
926 unsigned long end_pfn = zone_end_pfn(zone);
928 ret = compaction_suitable(zone, cc->order);
930 case COMPACT_PARTIAL:
931 case COMPACT_SKIPPED:
932 /* Compaction is likely to fail */
934 case COMPACT_CONTINUE:
935 /* Fall through to compaction */
940 * Setup to move all movable pages to the end of the zone. Used cached
941 * information on where the scanners should start but check that it
942 * is initialised by ensuring the values are within zone boundaries.
944 cc->migrate_pfn = zone->compact_cached_migrate_pfn;
945 cc->free_pfn = zone->compact_cached_free_pfn;
946 if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) {
947 cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1);
948 zone->compact_cached_free_pfn = cc->free_pfn;
950 if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) {
951 cc->migrate_pfn = start_pfn;
952 zone->compact_cached_migrate_pfn = cc->migrate_pfn;
956 * Clear pageblock skip if there were failures recently and compaction
957 * is about to be retried after being deferred. kswapd does not do
958 * this reset as it'll reset the cached information when going to sleep.
960 if (compaction_restarting(zone, cc->order) && !current_is_kswapd())
961 __reset_isolation_suitable(zone);
963 migrate_prep_local();
965 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
966 unsigned long nr_migrate, nr_remaining;
969 switch (isolate_migratepages(zone, cc)) {
971 ret = COMPACT_PARTIAL;
972 putback_movable_pages(&cc->migratepages);
973 cc->nr_migratepages = 0;
977 case ISOLATE_SUCCESS:
981 nr_migrate = cc->nr_migratepages;
982 err = migrate_pages(&cc->migratepages, compaction_alloc,
984 cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC,
986 update_nr_listpages(cc);
987 nr_remaining = cc->nr_migratepages;
989 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
992 /* Release isolated pages not migrated */
994 putback_movable_pages(&cc->migratepages);
995 cc->nr_migratepages = 0;
996 if (err == -ENOMEM) {
997 ret = COMPACT_PARTIAL;
1004 /* Release free pages and check accounting */
1005 cc->nr_freepages -= release_freepages(&cc->freepages);
1006 VM_BUG_ON(cc->nr_freepages != 0);
1011 static unsigned long compact_zone_order(struct zone *zone,
1012 int order, gfp_t gfp_mask,
1013 bool sync, bool *contended)
1016 struct compact_control cc = {
1018 .nr_migratepages = 0,
1020 .migratetype = allocflags_to_migratetype(gfp_mask),
1024 INIT_LIST_HEAD(&cc.freepages);
1025 INIT_LIST_HEAD(&cc.migratepages);
1027 ret = compact_zone(zone, &cc);
1029 VM_BUG_ON(!list_empty(&cc.freepages));
1030 VM_BUG_ON(!list_empty(&cc.migratepages));
1032 *contended = cc.contended;
1036 int sysctl_extfrag_threshold = 500;
1039 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
1040 * @zonelist: The zonelist used for the current allocation
1041 * @order: The order of the current allocation
1042 * @gfp_mask: The GFP mask of the current allocation
1043 * @nodemask: The allowed nodes to allocate from
1044 * @sync: Whether migration is synchronous or not
1045 * @contended: Return value that is true if compaction was aborted due to lock contention
1046 * @page: Optionally capture a free page of the requested order during compaction
1048 * This is the main entry point for direct page compaction.
1050 unsigned long try_to_compact_pages(struct zonelist *zonelist,
1051 int order, gfp_t gfp_mask, nodemask_t *nodemask,
1052 bool sync, bool *contended)
1054 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1055 int may_enter_fs = gfp_mask & __GFP_FS;
1056 int may_perform_io = gfp_mask & __GFP_IO;
1059 int rc = COMPACT_SKIPPED;
1060 int alloc_flags = 0;
1062 /* Check if the GFP flags allow compaction */
1063 if (!order || !may_enter_fs || !may_perform_io)
1066 count_compact_event(COMPACTSTALL);
1069 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
1070 alloc_flags |= ALLOC_CMA;
1072 /* Compact each zone in the list */
1073 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
1077 status = compact_zone_order(zone, order, gfp_mask, sync,
1079 rc = max(status, rc);
1081 /* If a normal allocation would succeed, stop compacting */
1082 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0,
1091 /* Compact all zones within a node */
1092 static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
1097 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
1099 zone = &pgdat->node_zones[zoneid];
1100 if (!populated_zone(zone))
1103 cc->nr_freepages = 0;
1104 cc->nr_migratepages = 0;
1106 INIT_LIST_HEAD(&cc->freepages);
1107 INIT_LIST_HEAD(&cc->migratepages);
1109 if (cc->order == -1 || !compaction_deferred(zone, cc->order))
1110 compact_zone(zone, cc);
1112 if (cc->order > 0) {
1113 int ok = zone_watermark_ok(zone, cc->order,
1114 low_wmark_pages(zone), 0, 0);
1115 if (ok && cc->order >= zone->compact_order_failed)
1116 zone->compact_order_failed = cc->order + 1;
1117 /* Currently async compaction is never deferred. */
1118 else if (!ok && cc->sync)
1119 defer_compaction(zone, cc->order);
1122 VM_BUG_ON(!list_empty(&cc->freepages));
1123 VM_BUG_ON(!list_empty(&cc->migratepages));
1127 void compact_pgdat(pg_data_t *pgdat, int order)
1129 struct compact_control cc = {
1137 __compact_pgdat(pgdat, &cc);
1140 static void compact_node(int nid)
1142 struct compact_control cc = {
1147 __compact_pgdat(NODE_DATA(nid), &cc);
1150 /* Compact all nodes in the system */
1151 static void compact_nodes(void)
1155 /* Flush pending updates to the LRU lists */
1156 lru_add_drain_all();
1158 for_each_online_node(nid)
1162 /* The written value is actually unused, all memory is compacted */
1163 int sysctl_compact_memory;
1165 /* This is the entry point for compacting all nodes via /proc/sys/vm */
1166 int sysctl_compaction_handler(struct ctl_table *table, int write,
1167 void __user *buffer, size_t *length, loff_t *ppos)
1175 int sysctl_extfrag_handler(struct ctl_table *table, int write,
1176 void __user *buffer, size_t *length, loff_t *ppos)
1178 proc_dointvec_minmax(table, write, buffer, length, ppos);
1183 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
1184 ssize_t sysfs_compact_node(struct device *dev,
1185 struct device_attribute *attr,
1186 const char *buf, size_t count)
1190 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1191 /* Flush pending updates to the LRU lists */
1192 lru_add_drain_all();
1199 static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
1201 int compaction_register_node(struct node *node)
1203 return device_create_file(&node->dev, &dev_attr_compact);
1206 void compaction_unregister_node(struct node *node)
1208 return device_remove_file(&node->dev, &dev_attr_compact);
1210 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
1212 #endif /* CONFIG_COMPACTION */