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 */
78 if (fatal_signal_pending(current))
83 spin_lock_irqsave(lock, *flags);
87 static inline bool compact_trylock_irqsave(spinlock_t *lock,
88 unsigned long *flags, struct compact_control *cc)
90 return compact_checklock_irqsave(lock, flags, false, cc);
93 static void compact_capture_page(struct compact_control *cc)
96 int mtype, mtype_low, mtype_high;
98 if (!cc->page || *cc->page)
102 * For MIGRATE_MOVABLE allocations we capture a suitable page ASAP
103 * regardless of the migratetype of the freelist is is captured from.
104 * This is fine because the order for a high-order MIGRATE_MOVABLE
105 * allocation is typically at least a pageblock size and overall
106 * fragmentation is not impaired. Other allocation types must
107 * capture pages from their own migratelist because otherwise they
108 * could pollute other pageblocks like MIGRATE_MOVABLE with
109 * difficult to move pages and making fragmentation worse overall.
111 if (cc->migratetype == MIGRATE_MOVABLE) {
113 mtype_high = MIGRATE_PCPTYPES;
115 mtype_low = cc->migratetype;
116 mtype_high = cc->migratetype + 1;
119 /* Speculatively examine the free lists without zone lock */
120 for (mtype = mtype_low; mtype < mtype_high; mtype++) {
122 for (order = cc->order; order < MAX_ORDER; order++) {
124 struct free_area *area;
125 area = &(cc->zone->free_area[order]);
126 if (list_empty(&area->free_list[mtype]))
129 /* Take the lock and attempt capture of the page */
130 if (!compact_trylock_irqsave(&cc->zone->lock, &flags, cc))
132 if (!list_empty(&area->free_list[mtype])) {
133 page = list_entry(area->free_list[mtype].next,
135 if (capture_free_page(page, cc->order, mtype)) {
136 spin_unlock_irqrestore(&cc->zone->lock,
142 spin_unlock_irqrestore(&cc->zone->lock, flags);
148 * Isolate free pages onto a private freelist. Caller must hold zone->lock.
149 * If @strict is true, will abort returning 0 on any invalid PFNs or non-free
150 * pages inside of the pageblock (even though it may still end up isolating
153 static unsigned long isolate_freepages_block(unsigned long blockpfn,
154 unsigned long end_pfn,
155 struct list_head *freelist,
158 int nr_scanned = 0, total_isolated = 0;
161 cursor = pfn_to_page(blockpfn);
163 /* Isolate free pages. This assumes the block is valid */
164 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
166 struct page *page = cursor;
168 if (!pfn_valid_within(blockpfn)) {
175 if (!PageBuddy(page)) {
181 /* Found a free page, break it into order-0 pages */
182 isolated = split_free_page(page);
183 if (!isolated && strict)
185 total_isolated += isolated;
186 for (i = 0; i < isolated; i++) {
187 list_add(&page->lru, freelist);
191 /* If a page was split, advance to the end of it */
193 blockpfn += isolated - 1;
194 cursor += isolated - 1;
198 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
199 return total_isolated;
203 * isolate_freepages_range() - isolate free pages.
204 * @start_pfn: The first PFN to start isolating.
205 * @end_pfn: The one-past-last PFN.
207 * Non-free pages, invalid PFNs, or zone boundaries within the
208 * [start_pfn, end_pfn) range are considered errors, cause function to
209 * undo its actions and return zero.
211 * Otherwise, function returns one-past-the-last PFN of isolated page
212 * (which may be greater then end_pfn if end fell in a middle of
216 isolate_freepages_range(unsigned long start_pfn, unsigned long end_pfn)
218 unsigned long isolated, pfn, block_end_pfn, flags;
219 struct zone *zone = NULL;
222 if (pfn_valid(start_pfn))
223 zone = page_zone(pfn_to_page(start_pfn));
225 for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
226 if (!pfn_valid(pfn) || zone != page_zone(pfn_to_page(pfn)))
230 * On subsequent iterations ALIGN() is actually not needed,
231 * but we keep it that we not to complicate the code.
233 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
234 block_end_pfn = min(block_end_pfn, end_pfn);
236 spin_lock_irqsave(&zone->lock, flags);
237 isolated = isolate_freepages_block(pfn, block_end_pfn,
239 spin_unlock_irqrestore(&zone->lock, flags);
242 * In strict mode, isolate_freepages_block() returns 0 if
243 * there are any holes in the block (ie. invalid PFNs or
250 * If we managed to isolate pages, it is always (1 << n) *
251 * pageblock_nr_pages for some non-negative n. (Max order
252 * page may span two pageblocks).
256 /* split_free_page does not map the pages */
257 map_pages(&freelist);
260 /* Loop terminated early, cleanup. */
261 release_freepages(&freelist);
265 /* We don't use freelists for anything. */
269 /* Update the number of anon and file isolated pages in the zone */
270 static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc)
273 unsigned int count[2] = { 0, };
275 list_for_each_entry(page, &cc->migratepages, lru)
276 count[!!page_is_file_cache(page)]++;
278 /* If locked we can use the interrupt unsafe versions */
280 __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
281 __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
283 mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
284 mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
288 /* Similar to reclaim, but different enough that they don't share logic */
289 static bool too_many_isolated(struct zone *zone)
291 unsigned long active, inactive, isolated;
293 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
294 zone_page_state(zone, NR_INACTIVE_ANON);
295 active = zone_page_state(zone, NR_ACTIVE_FILE) +
296 zone_page_state(zone, NR_ACTIVE_ANON);
297 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
298 zone_page_state(zone, NR_ISOLATED_ANON);
300 return isolated > (inactive + active) / 2;
304 * isolate_migratepages_range() - isolate all migrate-able pages in range.
305 * @zone: Zone pages are in.
306 * @cc: Compaction control structure.
307 * @low_pfn: The first PFN of the range.
308 * @end_pfn: The one-past-the-last PFN of the range.
310 * Isolate all pages that can be migrated from the range specified by
311 * [low_pfn, end_pfn). Returns zero if there is a fatal signal
312 * pending), otherwise PFN of the first page that was not scanned
313 * (which may be both less, equal to or more then end_pfn).
315 * Assumes that cc->migratepages is empty and cc->nr_migratepages is
318 * Apart from cc->migratepages and cc->nr_migratetypes this function
319 * does not modify any cc's fields, in particular it does not modify
320 * (or read for that matter) cc->migrate_pfn.
323 isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
324 unsigned long low_pfn, unsigned long end_pfn)
326 unsigned long last_pageblock_nr = 0, pageblock_nr;
327 unsigned long nr_scanned = 0, nr_isolated = 0;
328 struct list_head *migratelist = &cc->migratepages;
329 isolate_mode_t mode = 0;
330 struct lruvec *lruvec;
335 * Ensure that there are not too many pages isolated from the LRU
336 * list by either parallel reclaimers or compaction. If there are,
337 * delay for some time until fewer pages are isolated
339 while (unlikely(too_many_isolated(zone))) {
340 /* async migration should just abort */
344 congestion_wait(BLK_RW_ASYNC, HZ/10);
346 if (fatal_signal_pending(current))
350 /* Time to isolate some pages for migration */
352 spin_lock_irqsave(&zone->lru_lock, flags);
354 for (; low_pfn < end_pfn; low_pfn++) {
357 /* give a chance to irqs before checking need_resched() */
358 if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
359 spin_unlock_irqrestore(&zone->lru_lock, flags);
363 /* Check if it is ok to still hold the lock */
364 locked = compact_checklock_irqsave(&zone->lru_lock, &flags,
370 * migrate_pfn does not necessarily start aligned to a
371 * pageblock. Ensure that pfn_valid is called when moving
372 * into a new MAX_ORDER_NR_PAGES range in case of large
373 * memory holes within the zone
375 if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
376 if (!pfn_valid(low_pfn)) {
377 low_pfn += MAX_ORDER_NR_PAGES - 1;
382 if (!pfn_valid_within(low_pfn))
387 * Get the page and ensure the page is within the same zone.
388 * See the comment in isolate_freepages about overlapping
389 * nodes. It is deliberate that the new zone lock is not taken
390 * as memory compaction should not move pages between nodes.
392 page = pfn_to_page(low_pfn);
393 if (page_zone(page) != zone)
401 * For async migration, also only scan in MOVABLE blocks. Async
402 * migration is optimistic to see if the minimum amount of work
403 * satisfies the allocation
405 pageblock_nr = low_pfn >> pageblock_order;
406 if (!cc->sync && last_pageblock_nr != pageblock_nr &&
407 !migrate_async_suitable(get_pageblock_migratetype(page))) {
408 low_pfn += pageblock_nr_pages;
409 low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1;
410 last_pageblock_nr = pageblock_nr;
418 * PageLRU is set, and lru_lock excludes isolation,
419 * splitting and collapsing (collapsing has already
420 * happened if PageLRU is set).
422 if (PageTransHuge(page)) {
423 low_pfn += (1 << compound_order(page)) - 1;
428 mode |= ISOLATE_ASYNC_MIGRATE;
430 lruvec = mem_cgroup_page_lruvec(page, zone);
432 /* Try isolate the page */
433 if (__isolate_lru_page(page, mode) != 0)
436 VM_BUG_ON(PageTransCompound(page));
438 /* Successfully isolated */
439 del_page_from_lru_list(page, lruvec, page_lru(page));
440 list_add(&page->lru, migratelist);
441 cc->nr_migratepages++;
444 /* Avoid isolating too much */
445 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
451 acct_isolated(zone, locked, cc);
454 spin_unlock_irqrestore(&zone->lru_lock, flags);
456 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
461 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
462 #ifdef CONFIG_COMPACTION
464 /* Returns true if the page is within a block suitable for migration to */
465 static bool suitable_migration_target(struct page *page)
468 int migratetype = get_pageblock_migratetype(page);
470 /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
471 if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE)
474 /* If the page is a large free page, then allow migration */
475 if (PageBuddy(page) && page_order(page) >= pageblock_order)
478 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
479 if (migrate_async_suitable(migratetype))
482 /* Otherwise skip the block */
487 * Returns the start pfn of the last page block in a zone. This is the starting
488 * point for full compaction of a zone. Compaction searches for free pages from
489 * the end of each zone, while isolate_freepages_block scans forward inside each
492 static unsigned long start_free_pfn(struct zone *zone)
494 unsigned long free_pfn;
495 free_pfn = zone->zone_start_pfn + zone->spanned_pages;
496 free_pfn &= ~(pageblock_nr_pages-1);
501 * Based on information in the current compact_control, find blocks
502 * suitable for isolating free pages from and then isolate them.
504 static void isolate_freepages(struct zone *zone,
505 struct compact_control *cc)
508 unsigned long high_pfn, low_pfn, pfn, zone_end_pfn, end_pfn;
510 int nr_freepages = cc->nr_freepages;
511 struct list_head *freelist = &cc->freepages;
514 * Initialise the free scanner. The starting point is where we last
515 * scanned from (or the end of the zone if starting). The low point
516 * is the end of the pageblock the migration scanner is using.
519 low_pfn = cc->migrate_pfn + pageblock_nr_pages;
522 * Take care that if the migration scanner is at the end of the zone
523 * that the free scanner does not accidentally move to the next zone
524 * in the next isolation cycle.
526 high_pfn = min(low_pfn, pfn);
528 zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
531 * Isolate free pages until enough are available to migrate the
532 * pages on cc->migratepages. We stop searching if the migrate
533 * and free page scanners meet or enough free pages are isolated.
535 for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
536 pfn -= pageblock_nr_pages) {
537 unsigned long isolated;
543 * Check for overlapping nodes/zones. It's possible on some
544 * configurations to have a setup like
546 * i.e. it's possible that all pages within a zones range of
547 * pages do not belong to a single zone.
549 page = pfn_to_page(pfn);
550 if (page_zone(page) != zone)
553 /* Check the block is suitable for migration */
554 if (!suitable_migration_target(page))
558 * Found a block suitable for isolating free pages from. Now
559 * we disabled interrupts, double check things are ok and
560 * isolate the pages. This is to minimise the time IRQs
566 * The zone lock must be held to isolate freepages. This
567 * unfortunately this is a very coarse lock and can be
568 * heavily contended if there are parallel allocations
569 * or parallel compactions. For async compaction do not
572 if (!compact_trylock_irqsave(&zone->lock, &flags, cc))
574 if (suitable_migration_target(page)) {
575 end_pfn = min(pfn + pageblock_nr_pages, zone_end_pfn);
576 isolated = isolate_freepages_block(pfn, end_pfn,
578 nr_freepages += isolated;
580 spin_unlock_irqrestore(&zone->lock, flags);
583 * Record the highest PFN we isolated pages from. When next
584 * looking for free pages, the search will restart here as
585 * page migration may have returned some pages to the allocator
588 high_pfn = max(high_pfn, pfn);
591 * If the free scanner has wrapped, update
592 * compact_cached_free_pfn to point to the highest
593 * pageblock with free pages. This reduces excessive
594 * scanning of full pageblocks near the end of the
597 if (cc->order > 0 && cc->wrapped)
598 zone->compact_cached_free_pfn = high_pfn;
602 /* split_free_page does not map the pages */
605 cc->free_pfn = high_pfn;
606 cc->nr_freepages = nr_freepages;
608 /* If compact_cached_free_pfn is reset then set it now */
609 if (cc->order > 0 && !cc->wrapped &&
610 zone->compact_cached_free_pfn == start_free_pfn(zone))
611 zone->compact_cached_free_pfn = high_pfn;
615 * This is a migrate-callback that "allocates" freepages by taking pages
616 * from the isolated freelists in the block we are migrating to.
618 static struct page *compaction_alloc(struct page *migratepage,
622 struct compact_control *cc = (struct compact_control *)data;
623 struct page *freepage;
625 /* Isolate free pages if necessary */
626 if (list_empty(&cc->freepages)) {
627 isolate_freepages(cc->zone, cc);
629 if (list_empty(&cc->freepages))
633 freepage = list_entry(cc->freepages.next, struct page, lru);
634 list_del(&freepage->lru);
641 * We cannot control nr_migratepages and nr_freepages fully when migration is
642 * running as migrate_pages() has no knowledge of compact_control. When
643 * migration is complete, we count the number of pages on the lists by hand.
645 static void update_nr_listpages(struct compact_control *cc)
647 int nr_migratepages = 0;
648 int nr_freepages = 0;
651 list_for_each_entry(page, &cc->migratepages, lru)
653 list_for_each_entry(page, &cc->freepages, lru)
656 cc->nr_migratepages = nr_migratepages;
657 cc->nr_freepages = nr_freepages;
660 /* possible outcome of isolate_migratepages */
662 ISOLATE_ABORT, /* Abort compaction now */
663 ISOLATE_NONE, /* No pages isolated, continue scanning */
664 ISOLATE_SUCCESS, /* Pages isolated, migrate */
668 * Isolate all pages that can be migrated from the block pointed to by
669 * the migrate scanner within compact_control.
671 static isolate_migrate_t isolate_migratepages(struct zone *zone,
672 struct compact_control *cc)
674 unsigned long low_pfn, end_pfn;
676 /* Do not scan outside zone boundaries */
677 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
679 /* Only scan within a pageblock boundary */
680 end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages);
682 /* Do not cross the free scanner or scan within a memory hole */
683 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
684 cc->migrate_pfn = end_pfn;
688 /* Perform the isolation */
689 low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn);
690 if (!low_pfn || cc->contended)
691 return ISOLATE_ABORT;
693 cc->migrate_pfn = low_pfn;
695 return ISOLATE_SUCCESS;
698 static int compact_finished(struct zone *zone,
699 struct compact_control *cc)
701 unsigned long watermark;
703 if (fatal_signal_pending(current))
704 return COMPACT_PARTIAL;
707 * A full (order == -1) compaction run starts at the beginning and
708 * end of a zone; it completes when the migrate and free scanner meet.
709 * A partial (order > 0) compaction can start with the free scanner
710 * at a random point in the zone, and may have to restart.
712 if (cc->free_pfn <= cc->migrate_pfn) {
713 if (cc->order > 0 && !cc->wrapped) {
714 /* We started partway through; restart at the end. */
715 unsigned long free_pfn = start_free_pfn(zone);
716 zone->compact_cached_free_pfn = free_pfn;
717 cc->free_pfn = free_pfn;
719 return COMPACT_CONTINUE;
721 return COMPACT_COMPLETE;
724 /* We wrapped around and ended up where we started. */
725 if (cc->wrapped && cc->free_pfn <= cc->start_free_pfn)
726 return COMPACT_COMPLETE;
729 * order == -1 is expected when compacting via
730 * /proc/sys/vm/compact_memory
733 return COMPACT_CONTINUE;
735 /* Compaction run is not finished if the watermark is not met */
736 watermark = low_wmark_pages(zone);
737 watermark += (1 << cc->order);
739 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
740 return COMPACT_CONTINUE;
742 /* Direct compactor: Is a suitable page free? */
744 /* Was a suitable page captured? */
746 return COMPACT_PARTIAL;
749 for (order = cc->order; order < MAX_ORDER; order++) {
750 struct free_area *area = &zone->free_area[cc->order];
751 /* Job done if page is free of the right migratetype */
752 if (!list_empty(&area->free_list[cc->migratetype]))
753 return COMPACT_PARTIAL;
755 /* Job done if allocation would set block type */
756 if (cc->order >= pageblock_order && area->nr_free)
757 return COMPACT_PARTIAL;
761 return COMPACT_CONTINUE;
765 * compaction_suitable: Is this suitable to run compaction on this zone now?
767 * COMPACT_SKIPPED - If there are too few free pages for compaction
768 * COMPACT_PARTIAL - If the allocation would succeed without compaction
769 * COMPACT_CONTINUE - If compaction should run now
771 unsigned long compaction_suitable(struct zone *zone, int order)
774 unsigned long watermark;
777 * order == -1 is expected when compacting via
778 * /proc/sys/vm/compact_memory
781 return COMPACT_CONTINUE;
784 * Watermarks for order-0 must be met for compaction. Note the 2UL.
785 * This is because during migration, copies of pages need to be
786 * allocated and for a short time, the footprint is higher
788 watermark = low_wmark_pages(zone) + (2UL << order);
789 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
790 return COMPACT_SKIPPED;
793 * fragmentation index determines if allocation failures are due to
794 * low memory or external fragmentation
796 * index of -1000 implies allocations might succeed depending on
798 * index towards 0 implies failure is due to lack of memory
799 * index towards 1000 implies failure is due to fragmentation
801 * Only compact if a failure would be due to fragmentation.
803 fragindex = fragmentation_index(zone, order);
804 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
805 return COMPACT_SKIPPED;
807 if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
809 return COMPACT_PARTIAL;
811 return COMPACT_CONTINUE;
814 static int compact_zone(struct zone *zone, struct compact_control *cc)
818 ret = compaction_suitable(zone, cc->order);
820 case COMPACT_PARTIAL:
821 case COMPACT_SKIPPED:
822 /* Compaction is likely to fail */
824 case COMPACT_CONTINUE:
825 /* Fall through to compaction */
829 /* Setup to move all movable pages to the end of the zone */
830 cc->migrate_pfn = zone->zone_start_pfn;
833 /* Incremental compaction. Start where the last one stopped. */
834 cc->free_pfn = zone->compact_cached_free_pfn;
835 cc->start_free_pfn = cc->free_pfn;
837 /* Order == -1 starts at the end of the zone. */
838 cc->free_pfn = start_free_pfn(zone);
841 migrate_prep_local();
843 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
844 unsigned long nr_migrate, nr_remaining;
847 switch (isolate_migratepages(zone, cc)) {
849 ret = COMPACT_PARTIAL;
853 case ISOLATE_SUCCESS:
857 nr_migrate = cc->nr_migratepages;
858 err = migrate_pages(&cc->migratepages, compaction_alloc,
859 (unsigned long)cc, false,
860 cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC);
861 update_nr_listpages(cc);
862 nr_remaining = cc->nr_migratepages;
864 count_vm_event(COMPACTBLOCKS);
865 count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining);
867 count_vm_events(COMPACTPAGEFAILED, nr_remaining);
868 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
871 /* Release LRU pages not migrated */
873 putback_lru_pages(&cc->migratepages);
874 cc->nr_migratepages = 0;
875 if (err == -ENOMEM) {
876 ret = COMPACT_PARTIAL;
881 /* Capture a page now if it is a suitable size */
882 compact_capture_page(cc);
886 /* Release free pages and check accounting */
887 cc->nr_freepages -= release_freepages(&cc->freepages);
888 VM_BUG_ON(cc->nr_freepages != 0);
893 static unsigned long compact_zone_order(struct zone *zone,
894 int order, gfp_t gfp_mask,
895 bool sync, bool *contended,
899 struct compact_control cc = {
901 .nr_migratepages = 0,
903 .migratetype = allocflags_to_migratetype(gfp_mask),
908 INIT_LIST_HEAD(&cc.freepages);
909 INIT_LIST_HEAD(&cc.migratepages);
911 ret = compact_zone(zone, &cc);
913 *contended = cc.contended;
917 int sysctl_extfrag_threshold = 500;
920 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
921 * @zonelist: The zonelist used for the current allocation
922 * @order: The order of the current allocation
923 * @gfp_mask: The GFP mask of the current allocation
924 * @nodemask: The allowed nodes to allocate from
925 * @sync: Whether migration is synchronous or not
927 * This is the main entry point for direct page compaction.
929 unsigned long try_to_compact_pages(struct zonelist *zonelist,
930 int order, gfp_t gfp_mask, nodemask_t *nodemask,
931 bool sync, bool *contended, struct page **page)
933 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
934 int may_enter_fs = gfp_mask & __GFP_FS;
935 int may_perform_io = gfp_mask & __GFP_IO;
938 int rc = COMPACT_SKIPPED;
941 /* Check if the GFP flags allow compaction */
942 if (!order || !may_enter_fs || !may_perform_io)
945 count_vm_event(COMPACTSTALL);
948 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
949 alloc_flags |= ALLOC_CMA;
951 /* Compact each zone in the list */
952 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
956 status = compact_zone_order(zone, order, gfp_mask, sync,
958 rc = max(status, rc);
960 /* If a normal allocation would succeed, stop compacting */
961 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0,
970 /* Compact all zones within a node */
971 static int __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
976 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
978 zone = &pgdat->node_zones[zoneid];
979 if (!populated_zone(zone))
982 cc->nr_freepages = 0;
983 cc->nr_migratepages = 0;
985 INIT_LIST_HEAD(&cc->freepages);
986 INIT_LIST_HEAD(&cc->migratepages);
988 if (cc->order == -1 || !compaction_deferred(zone, cc->order))
989 compact_zone(zone, cc);
992 int ok = zone_watermark_ok(zone, cc->order,
993 low_wmark_pages(zone), 0, 0);
994 if (ok && cc->order >= zone->compact_order_failed)
995 zone->compact_order_failed = cc->order + 1;
996 /* Currently async compaction is never deferred. */
997 else if (!ok && cc->sync)
998 defer_compaction(zone, cc->order);
1001 VM_BUG_ON(!list_empty(&cc->freepages));
1002 VM_BUG_ON(!list_empty(&cc->migratepages));
1008 int compact_pgdat(pg_data_t *pgdat, int order)
1010 struct compact_control cc = {
1016 return __compact_pgdat(pgdat, &cc);
1019 static int compact_node(int nid)
1021 struct compact_control cc = {
1027 return __compact_pgdat(NODE_DATA(nid), &cc);
1030 /* Compact all nodes in the system */
1031 static int compact_nodes(void)
1035 /* Flush pending updates to the LRU lists */
1036 lru_add_drain_all();
1038 for_each_online_node(nid)
1041 return COMPACT_COMPLETE;
1044 /* The written value is actually unused, all memory is compacted */
1045 int sysctl_compact_memory;
1047 /* This is the entry point for compacting all nodes via /proc/sys/vm */
1048 int sysctl_compaction_handler(struct ctl_table *table, int write,
1049 void __user *buffer, size_t *length, loff_t *ppos)
1052 return compact_nodes();
1057 int sysctl_extfrag_handler(struct ctl_table *table, int write,
1058 void __user *buffer, size_t *length, loff_t *ppos)
1060 proc_dointvec_minmax(table, write, buffer, length, ppos);
1065 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
1066 ssize_t sysfs_compact_node(struct device *dev,
1067 struct device_attribute *attr,
1068 const char *buf, size_t count)
1072 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1073 /* Flush pending updates to the LRU lists */
1074 lru_add_drain_all();
1081 static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
1083 int compaction_register_node(struct node *node)
1085 return device_create_file(&node->dev, &dev_attr_compact);
1088 void compaction_unregister_node(struct node *node)
1090 return device_remove_file(&node->dev, &dev_attr_compact);
1092 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
1094 #endif /* CONFIG_COMPACTION */