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
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 * Isolate free pages onto a private freelist. Caller must hold zone->lock.
55 * If @strict is true, will abort returning 0 on any invalid PFNs or non-free
56 * pages inside of the pageblock (even though it may still end up isolating
59 static unsigned long isolate_freepages_block(unsigned long blockpfn,
60 unsigned long end_pfn,
61 struct list_head *freelist,
64 int nr_scanned = 0, total_isolated = 0;
67 cursor = pfn_to_page(blockpfn);
69 /* Isolate free pages. This assumes the block is valid */
70 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
72 struct page *page = cursor;
74 if (!pfn_valid_within(blockpfn)) {
81 if (!PageBuddy(page)) {
87 /* Found a free page, break it into order-0 pages */
88 isolated = split_free_page(page);
89 if (!isolated && strict)
91 total_isolated += isolated;
92 for (i = 0; i < isolated; i++) {
93 list_add(&page->lru, freelist);
97 /* If a page was split, advance to the end of it */
99 blockpfn += isolated - 1;
100 cursor += isolated - 1;
104 trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated);
105 return total_isolated;
109 * isolate_freepages_range() - isolate free pages.
110 * @start_pfn: The first PFN to start isolating.
111 * @end_pfn: The one-past-last PFN.
113 * Non-free pages, invalid PFNs, or zone boundaries within the
114 * [start_pfn, end_pfn) range are considered errors, cause function to
115 * undo its actions and return zero.
117 * Otherwise, function returns one-past-the-last PFN of isolated page
118 * (which may be greater then end_pfn if end fell in a middle of
122 isolate_freepages_range(unsigned long start_pfn, unsigned long end_pfn)
124 unsigned long isolated, pfn, block_end_pfn, flags;
125 struct zone *zone = NULL;
128 if (pfn_valid(start_pfn))
129 zone = page_zone(pfn_to_page(start_pfn));
131 for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) {
132 if (!pfn_valid(pfn) || zone != page_zone(pfn_to_page(pfn)))
136 * On subsequent iterations ALIGN() is actually not needed,
137 * but we keep it that we not to complicate the code.
139 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
140 block_end_pfn = min(block_end_pfn, end_pfn);
142 spin_lock_irqsave(&zone->lock, flags);
143 isolated = isolate_freepages_block(pfn, block_end_pfn,
145 spin_unlock_irqrestore(&zone->lock, flags);
148 * In strict mode, isolate_freepages_block() returns 0 if
149 * there are any holes in the block (ie. invalid PFNs or
156 * If we managed to isolate pages, it is always (1 << n) *
157 * pageblock_nr_pages for some non-negative n. (Max order
158 * page may span two pageblocks).
162 /* split_free_page does not map the pages */
163 map_pages(&freelist);
166 /* Loop terminated early, cleanup. */
167 release_freepages(&freelist);
171 /* We don't use freelists for anything. */
175 /* Update the number of anon and file isolated pages in the zone */
176 static void acct_isolated(struct zone *zone, struct compact_control *cc)
179 unsigned int count[2] = { 0, };
181 list_for_each_entry(page, &cc->migratepages, lru)
182 count[!!page_is_file_cache(page)]++;
184 __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
185 __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
188 /* Similar to reclaim, but different enough that they don't share logic */
189 static bool too_many_isolated(struct zone *zone)
191 unsigned long active, inactive, isolated;
193 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
194 zone_page_state(zone, NR_INACTIVE_ANON);
195 active = zone_page_state(zone, NR_ACTIVE_FILE) +
196 zone_page_state(zone, NR_ACTIVE_ANON);
197 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
198 zone_page_state(zone, NR_ISOLATED_ANON);
200 return isolated > (inactive + active) / 2;
204 * isolate_migratepages_range() - isolate all migrate-able pages in range.
205 * @zone: Zone pages are in.
206 * @cc: Compaction control structure.
207 * @low_pfn: The first PFN of the range.
208 * @end_pfn: The one-past-the-last PFN of the range.
210 * Isolate all pages that can be migrated from the range specified by
211 * [low_pfn, end_pfn). Returns zero if there is a fatal signal
212 * pending), otherwise PFN of the first page that was not scanned
213 * (which may be both less, equal to or more then end_pfn).
215 * Assumes that cc->migratepages is empty and cc->nr_migratepages is
218 * Apart from cc->migratepages and cc->nr_migratetypes this function
219 * does not modify any cc's fields, in particular it does not modify
220 * (or read for that matter) cc->migrate_pfn.
223 isolate_migratepages_range(struct zone *zone, struct compact_control *cc,
224 unsigned long low_pfn, unsigned long end_pfn)
226 unsigned long last_pageblock_nr = 0, pageblock_nr;
227 unsigned long nr_scanned = 0, nr_isolated = 0;
228 struct list_head *migratelist = &cc->migratepages;
229 isolate_mode_t mode = ISOLATE_ACTIVE|ISOLATE_INACTIVE;
232 * Ensure that there are not too many pages isolated from the LRU
233 * list by either parallel reclaimers or compaction. If there are,
234 * delay for some time until fewer pages are isolated
236 while (unlikely(too_many_isolated(zone))) {
237 /* async migration should just abort */
238 if (cc->mode != COMPACT_SYNC)
241 congestion_wait(BLK_RW_ASYNC, HZ/10);
243 if (fatal_signal_pending(current))
247 /* Time to isolate some pages for migration */
249 spin_lock_irq(&zone->lru_lock);
250 for (; low_pfn < end_pfn; low_pfn++) {
254 /* give a chance to irqs before checking need_resched() */
255 if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
256 spin_unlock_irq(&zone->lru_lock);
259 if (need_resched() || spin_is_contended(&zone->lru_lock)) {
261 spin_unlock_irq(&zone->lru_lock);
263 spin_lock_irq(&zone->lru_lock);
264 if (fatal_signal_pending(current))
267 spin_lock_irq(&zone->lru_lock);
270 * migrate_pfn does not necessarily start aligned to a
271 * pageblock. Ensure that pfn_valid is called when moving
272 * into a new MAX_ORDER_NR_PAGES range in case of large
273 * memory holes within the zone
275 if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
276 if (!pfn_valid(low_pfn)) {
277 low_pfn += MAX_ORDER_NR_PAGES - 1;
282 if (!pfn_valid_within(low_pfn))
287 * Get the page and ensure the page is within the same zone.
288 * See the comment in isolate_freepages about overlapping
289 * nodes. It is deliberate that the new zone lock is not taken
290 * as memory compaction should not move pages between nodes.
292 page = pfn_to_page(low_pfn);
293 if (page_zone(page) != zone)
301 * For async migration, also only scan in MOVABLE blocks. Async
302 * migration is optimistic to see if the minimum amount of work
303 * satisfies the allocation
305 pageblock_nr = low_pfn >> pageblock_order;
306 if (cc->mode != COMPACT_SYNC &&
307 last_pageblock_nr != pageblock_nr &&
308 !migrate_async_suitable(get_pageblock_migratetype(page))) {
309 low_pfn += pageblock_nr_pages;
310 low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1;
311 last_pageblock_nr = pageblock_nr;
319 * PageLRU is set, and lru_lock excludes isolation,
320 * splitting and collapsing (collapsing has already
321 * happened if PageLRU is set).
323 if (PageTransHuge(page)) {
324 low_pfn += (1 << compound_order(page)) - 1;
328 if (cc->mode != COMPACT_SYNC)
329 mode |= ISOLATE_ASYNC_MIGRATE;
331 /* Try isolate the page */
332 if (__isolate_lru_page(page, mode, 0) != 0)
335 VM_BUG_ON(PageTransCompound(page));
337 /* Successfully isolated */
338 del_page_from_lru_list(zone, page, page_lru(page));
339 list_add(&page->lru, migratelist);
340 cc->nr_migratepages++;
343 /* Avoid isolating too much */
344 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
350 acct_isolated(zone, cc);
352 spin_unlock_irq(&zone->lru_lock);
354 trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated);
359 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
360 #ifdef CONFIG_COMPACTION
362 * Returns true if MIGRATE_UNMOVABLE pageblock was successfully
363 * converted to MIGRATE_MOVABLE type, false otherwise.
365 static bool rescue_unmovable_pageblock(struct page *page)
367 unsigned long pfn, start_pfn, end_pfn;
368 struct page *start_page, *end_page;
370 pfn = page_to_pfn(page);
371 start_pfn = pfn & ~(pageblock_nr_pages - 1);
372 end_pfn = start_pfn + pageblock_nr_pages;
374 start_page = pfn_to_page(start_pfn);
375 end_page = pfn_to_page(end_pfn);
377 /* Do not deal with pageblocks that overlap zones */
378 if (page_zone(start_page) != page_zone(end_page))
381 for (page = start_page, pfn = start_pfn; page < end_page; pfn++,
383 if (!pfn_valid_within(pfn))
386 if (PageBuddy(page)) {
387 int order = page_order(page);
389 pfn += (1 << order) - 1;
390 page += (1 << order) - 1;
393 } else if (page_count(page) == 0 || PageLRU(page))
399 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
400 move_freepages_block(page_zone(page), page, MIGRATE_MOVABLE);
405 GOOD_AS_MIGRATION_TARGET,
406 FAIL_UNMOVABLE_TARGET,
411 * Returns GOOD_AS_MIGRATION_TARGET if the page is within a block
412 * suitable for migration to, FAIL_UNMOVABLE_TARGET if the page
413 * is within a MIGRATE_UNMOVABLE block, FAIL_BAD_TARGET otherwise.
415 static enum smt_result suitable_migration_target(struct page *page,
416 struct compact_control *cc)
419 int migratetype = get_pageblock_migratetype(page);
421 /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */
422 if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE)
423 return FAIL_BAD_TARGET;
425 /* If the page is a large free page, then allow migration */
426 if (PageBuddy(page) && page_order(page) >= pageblock_order)
427 return GOOD_AS_MIGRATION_TARGET;
429 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
430 if (cc->mode != COMPACT_ASYNC_UNMOVABLE &&
431 migrate_async_suitable(migratetype))
432 return GOOD_AS_MIGRATION_TARGET;
434 if (cc->mode == COMPACT_ASYNC_MOVABLE &&
435 migratetype == MIGRATE_UNMOVABLE)
436 return FAIL_UNMOVABLE_TARGET;
438 if (cc->mode != COMPACT_ASYNC_MOVABLE &&
439 migratetype == MIGRATE_UNMOVABLE &&
440 rescue_unmovable_pageblock(page))
441 return GOOD_AS_MIGRATION_TARGET;
443 /* Otherwise skip the block */
444 return FAIL_BAD_TARGET;
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_freepages() may be called more than once during
479 * compact_zone_order() run and we want only the most recent
482 cc->nr_pageblocks_skipped = 0;
485 * Isolate free pages until enough are available to migrate the
486 * pages on cc->migratepages. We stop searching if the migrate
487 * and free page scanners meet or enough free pages are isolated.
489 for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages;
490 pfn -= pageblock_nr_pages) {
491 unsigned long isolated;
498 * Check for overlapping nodes/zones. It's possible on some
499 * configurations to have a setup like
501 * i.e. it's possible that all pages within a zones range of
502 * pages do not belong to a single zone.
504 page = pfn_to_page(pfn);
505 if (page_zone(page) != zone)
508 /* Check the block is suitable for migration */
509 ret = suitable_migration_target(page, cc);
510 if (ret != GOOD_AS_MIGRATION_TARGET) {
511 if (ret == FAIL_UNMOVABLE_TARGET)
512 cc->nr_pageblocks_skipped++;
516 * Found a block suitable for isolating free pages from. Now
517 * we disabled interrupts, double check things are ok and
518 * isolate the pages. This is to minimise the time IRQs
522 spin_lock_irqsave(&zone->lock, flags);
523 ret = suitable_migration_target(page, cc);
524 if (ret == GOOD_AS_MIGRATION_TARGET) {
525 end_pfn = min(pfn + pageblock_nr_pages, zone_end_pfn);
526 isolated = isolate_freepages_block(pfn, end_pfn,
528 nr_freepages += isolated;
529 } else if (ret == FAIL_UNMOVABLE_TARGET)
530 cc->nr_pageblocks_skipped++;
531 spin_unlock_irqrestore(&zone->lock, flags);
534 * Record the highest PFN we isolated pages from. When next
535 * looking for free pages, the search will restart here as
536 * page migration may have returned some pages to the allocator
539 high_pfn = max(high_pfn, pfn);
542 /* split_free_page does not map the pages */
545 cc->free_pfn = high_pfn;
546 cc->nr_freepages = nr_freepages;
550 * This is a migrate-callback that "allocates" freepages by taking pages
551 * from the isolated freelists in the block we are migrating to.
553 static struct page *compaction_alloc(struct page *migratepage,
557 struct compact_control *cc = (struct compact_control *)data;
558 struct page *freepage;
560 /* Isolate free pages if necessary */
561 if (list_empty(&cc->freepages)) {
562 isolate_freepages(cc->zone, cc);
564 if (list_empty(&cc->freepages))
568 freepage = list_entry(cc->freepages.next, struct page, lru);
569 list_del(&freepage->lru);
576 * We cannot control nr_migratepages and nr_freepages fully when migration is
577 * running as migrate_pages() has no knowledge of compact_control. When
578 * migration is complete, we count the number of pages on the lists by hand.
580 static void update_nr_listpages(struct compact_control *cc)
582 int nr_migratepages = 0;
583 int nr_freepages = 0;
586 list_for_each_entry(page, &cc->migratepages, lru)
588 list_for_each_entry(page, &cc->freepages, lru)
591 cc->nr_migratepages = nr_migratepages;
592 cc->nr_freepages = nr_freepages;
595 /* possible outcome of isolate_migratepages */
597 ISOLATE_ABORT, /* Abort compaction now */
598 ISOLATE_NONE, /* No pages isolated, continue scanning */
599 ISOLATE_SUCCESS, /* Pages isolated, migrate */
603 * Isolate all pages that can be migrated from the block pointed to by
604 * the migrate scanner within compact_control.
606 static isolate_migrate_t isolate_migratepages(struct zone *zone,
607 struct compact_control *cc)
609 unsigned long low_pfn, end_pfn;
611 /* Do not scan outside zone boundaries */
612 low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
614 /* Only scan within a pageblock boundary */
615 end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages);
617 /* Do not cross the free scanner or scan within a memory hole */
618 if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) {
619 cc->migrate_pfn = end_pfn;
623 /* Perform the isolation */
624 low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn);
626 return ISOLATE_ABORT;
628 cc->migrate_pfn = low_pfn;
630 return ISOLATE_SUCCESS;
633 static int compact_finished(struct zone *zone,
634 struct compact_control *cc)
637 unsigned long watermark;
639 if (fatal_signal_pending(current))
640 return COMPACT_PARTIAL;
642 /* Compaction run completes if the migrate and free scanner meet */
643 if (cc->free_pfn <= cc->migrate_pfn)
644 return COMPACT_COMPLETE;
647 * order == -1 is expected when compacting via
648 * /proc/sys/vm/compact_memory
651 return COMPACT_CONTINUE;
653 /* Compaction run is not finished if the watermark is not met */
654 watermark = low_wmark_pages(zone);
655 watermark += (1 << cc->order);
657 if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
658 return COMPACT_CONTINUE;
660 /* Direct compactor: Is a suitable page free? */
661 for (order = cc->order; order < MAX_ORDER; order++) {
662 /* Job done if page is free of the right migratetype */
663 if (!list_empty(&zone->free_area[order].free_list[cc->migratetype]))
664 return COMPACT_PARTIAL;
666 /* Job done if allocation would set block type */
667 if (order >= pageblock_order && zone->free_area[order].nr_free)
668 return COMPACT_PARTIAL;
671 return COMPACT_CONTINUE;
675 * compaction_suitable: Is this suitable to run compaction on this zone now?
677 * COMPACT_SKIPPED - If there are too few free pages for compaction
678 * COMPACT_PARTIAL - If the allocation would succeed without compaction
679 * COMPACT_CONTINUE - If compaction should run now
681 unsigned long compaction_suitable(struct zone *zone, int order)
684 unsigned long watermark;
687 * order == -1 is expected when compacting via
688 * /proc/sys/vm/compact_memory
691 return COMPACT_CONTINUE;
694 * Watermarks for order-0 must be met for compaction. Note the 2UL.
695 * This is because during migration, copies of pages need to be
696 * allocated and for a short time, the footprint is higher
698 watermark = low_wmark_pages(zone) + (2UL << order);
699 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
700 return COMPACT_SKIPPED;
703 * fragmentation index determines if allocation failures are due to
704 * low memory or external fragmentation
706 * index of -1000 implies allocations might succeed depending on
708 * index towards 0 implies failure is due to lack of memory
709 * index towards 1000 implies failure is due to fragmentation
711 * Only compact if a failure would be due to fragmentation.
713 fragindex = fragmentation_index(zone, order);
714 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
715 return COMPACT_SKIPPED;
717 if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark,
719 return COMPACT_PARTIAL;
721 return COMPACT_CONTINUE;
724 static int compact_zone(struct zone *zone, struct compact_control *cc)
728 ret = compaction_suitable(zone, cc->order);
730 case COMPACT_PARTIAL:
731 case COMPACT_SKIPPED:
732 /* Compaction is likely to fail */
734 case COMPACT_CONTINUE:
735 /* Fall through to compaction */
739 /* Setup to move all movable pages to the end of the zone */
740 cc->migrate_pfn = zone->zone_start_pfn;
741 cc->free_pfn = cc->migrate_pfn + zone->spanned_pages;
742 cc->free_pfn &= ~(pageblock_nr_pages-1);
744 migrate_prep_local();
746 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
747 unsigned long nr_migrate, nr_remaining;
750 switch (isolate_migratepages(zone, cc)) {
752 ret = COMPACT_PARTIAL;
756 case ISOLATE_SUCCESS:
760 nr_migrate = cc->nr_migratepages;
761 err = migrate_pages(&cc->migratepages, compaction_alloc,
762 (unsigned long)&cc->freepages, false,
763 (cc->mode == COMPACT_SYNC) ? MIGRATE_SYNC_LIGHT
765 update_nr_listpages(cc);
766 nr_remaining = cc->nr_migratepages;
768 count_vm_event(COMPACTBLOCKS);
769 count_vm_events(COMPACTPAGES, nr_migrate - nr_remaining);
771 count_vm_events(COMPACTPAGEFAILED, nr_remaining);
772 trace_mm_compaction_migratepages(nr_migrate - nr_remaining,
775 /* Release LRU pages not migrated */
777 putback_lru_pages(&cc->migratepages);
778 cc->nr_migratepages = 0;
784 /* Release free pages and check accounting */
785 cc->nr_freepages -= release_freepages(&cc->freepages);
786 VM_BUG_ON(cc->nr_freepages != 0);
791 static unsigned long compact_zone_order(struct zone *zone,
792 int order, gfp_t gfp_mask,
793 enum compact_mode mode,
794 unsigned long *nr_pageblocks_skipped)
796 struct compact_control cc = {
798 .nr_migratepages = 0,
800 .migratetype = allocflags_to_migratetype(gfp_mask),
806 INIT_LIST_HEAD(&cc.freepages);
807 INIT_LIST_HEAD(&cc.migratepages);
809 rc = compact_zone(zone, &cc);
810 *nr_pageblocks_skipped = cc.nr_pageblocks_skipped;
815 int sysctl_extfrag_threshold = 500;
818 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
819 * @zonelist: The zonelist used for the current allocation
820 * @order: The order of the current allocation
821 * @gfp_mask: The GFP mask of the current allocation
822 * @nodemask: The allowed nodes to allocate from
823 * @sync: Whether migration is synchronous or not
825 * This is the main entry point for direct page compaction.
827 unsigned long try_to_compact_pages(struct zonelist *zonelist,
828 int order, gfp_t gfp_mask, nodemask_t *nodemask,
831 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
832 int may_enter_fs = gfp_mask & __GFP_FS;
833 int may_perform_io = gfp_mask & __GFP_IO;
836 int rc = COMPACT_SKIPPED;
837 unsigned long nr_pageblocks_skipped;
838 enum compact_mode mode;
841 * Check whether it is worth even starting compaction. The order check is
842 * made because an assumption is made that the page allocator can satisfy
843 * the "cheaper" orders without taking special steps
845 if (!order || !may_enter_fs || !may_perform_io)
848 count_vm_event(COMPACTSTALL);
850 /* Compact each zone in the list */
851 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
855 mode = sync ? COMPACT_SYNC : COMPACT_ASYNC_MOVABLE;
857 status = compact_zone_order(zone, order, gfp_mask, mode,
858 &nr_pageblocks_skipped);
859 rc = max(status, rc);
861 /* If a normal allocation would succeed, stop compacting */
862 if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, 0))
865 if (rc == COMPACT_COMPLETE && mode == COMPACT_ASYNC_MOVABLE) {
866 if (nr_pageblocks_skipped) {
867 mode = COMPACT_ASYNC_UNMOVABLE;
877 /* Compact all zones within a node */
878 static int __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
883 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
885 zone = &pgdat->node_zones[zoneid];
886 if (!populated_zone(zone))
889 cc->nr_freepages = 0;
890 cc->nr_migratepages = 0;
892 INIT_LIST_HEAD(&cc->freepages);
893 INIT_LIST_HEAD(&cc->migratepages);
895 if (cc->order == -1 || !compaction_deferred(zone, cc->order))
896 compact_zone(zone, cc);
899 int ok = zone_watermark_ok(zone, cc->order,
900 low_wmark_pages(zone), 0, 0);
901 if (ok && cc->order > zone->compact_order_failed)
902 zone->compact_order_failed = cc->order + 1;
903 /* Currently async compaction is never deferred. */
904 else if (!ok && cc->mode == COMPACT_SYNC)
905 defer_compaction(zone, cc->order);
908 VM_BUG_ON(!list_empty(&cc->freepages));
909 VM_BUG_ON(!list_empty(&cc->migratepages));
915 int compact_pgdat(pg_data_t *pgdat, int order)
917 struct compact_control cc = {
919 .mode = COMPACT_ASYNC_MOVABLE,
922 return __compact_pgdat(pgdat, &cc);
925 static int compact_node(int nid)
927 struct compact_control cc = {
929 .mode = COMPACT_SYNC,
932 return __compact_pgdat(NODE_DATA(nid), &cc);
935 /* Compact all nodes in the system */
936 static int compact_nodes(void)
940 /* Flush pending updates to the LRU lists */
943 for_each_online_node(nid)
946 return COMPACT_COMPLETE;
949 /* The written value is actually unused, all memory is compacted */
950 int sysctl_compact_memory;
952 /* This is the entry point for compacting all nodes via /proc/sys/vm */
953 int sysctl_compaction_handler(struct ctl_table *table, int write,
954 void __user *buffer, size_t *length, loff_t *ppos)
957 return compact_nodes();
962 int sysctl_extfrag_handler(struct ctl_table *table, int write,
963 void __user *buffer, size_t *length, loff_t *ppos)
965 proc_dointvec_minmax(table, write, buffer, length, ppos);
970 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
971 ssize_t sysfs_compact_node(struct device *dev,
972 struct device_attribute *attr,
973 const char *buf, size_t count)
977 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
978 /* Flush pending updates to the LRU lists */
986 static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
988 int compaction_register_node(struct node *node)
990 return device_create_file(&node->dev, &dev_attr_compact);
993 void compaction_unregister_node(struct node *node)
995 return device_remove_file(&node->dev, &dev_attr_compact);
997 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
999 #endif /* CONFIG_COMPACTION */