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>
19 #include <linux/kasan.h>
22 #ifdef CONFIG_COMPACTION
23 static inline void count_compact_event(enum vm_event_item item
)
28 static inline void count_compact_events(enum vm_event_item item
, long delta
)
30 count_vm_events(item
, delta
);
33 #define count_compact_event(item) do { } while (0)
34 #define count_compact_events(item, delta) do { } while (0)
37 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
38 #ifdef CONFIG_TRACEPOINTS
39 static const char *const compaction_status_string
[] = {
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/compaction.h>
53 static unsigned long release_freepages(struct list_head
*freelist
)
55 struct page
*page
, *next
;
56 unsigned long high_pfn
= 0;
58 list_for_each_entry_safe(page
, next
, freelist
, lru
) {
59 unsigned long pfn
= page_to_pfn(page
);
69 static void map_pages(struct list_head
*list
)
73 list_for_each_entry(page
, list
, lru
) {
74 arch_alloc_page(page
, 0);
75 kernel_map_pages(page
, 1, 1);
76 kasan_alloc_pages(page
, 0);
80 static inline bool migrate_async_suitable(int migratetype
)
82 return is_migrate_cma(migratetype
) || migratetype
== MIGRATE_MOVABLE
;
86 * Check that the whole (or subset of) a pageblock given by the interval of
87 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
88 * with the migration of free compaction scanner. The scanners then need to
89 * use only pfn_valid_within() check for arches that allow holes within
92 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
94 * It's possible on some configurations to have a setup like node0 node1 node0
95 * i.e. it's possible that all pages within a zones range of pages do not
96 * belong to a single zone. We assume that a border between node0 and node1
97 * can occur within a single pageblock, but not a node0 node1 node0
98 * interleaving within a single pageblock. It is therefore sufficient to check
99 * the first and last page of a pageblock and avoid checking each individual
100 * page in a pageblock.
102 static struct page
*pageblock_pfn_to_page(unsigned long start_pfn
,
103 unsigned long end_pfn
, struct zone
*zone
)
105 struct page
*start_page
;
106 struct page
*end_page
;
108 /* end_pfn is one past the range we are checking */
111 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
114 start_page
= pfn_to_page(start_pfn
);
116 if (page_zone(start_page
) != zone
)
119 end_page
= pfn_to_page(end_pfn
);
121 /* This gives a shorter code than deriving page_zone(end_page) */
122 if (page_zone_id(start_page
) != page_zone_id(end_page
))
128 #ifdef CONFIG_COMPACTION
130 /* Do not skip compaction more than 64 times */
131 #define COMPACT_MAX_DEFER_SHIFT 6
134 * Compaction is deferred when compaction fails to result in a page
135 * allocation success. 1 << compact_defer_limit compactions are skipped up
136 * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT
138 void defer_compaction(struct zone
*zone
, int order
)
140 zone
->compact_considered
= 0;
141 zone
->compact_defer_shift
++;
143 if (order
< zone
->compact_order_failed
)
144 zone
->compact_order_failed
= order
;
146 if (zone
->compact_defer_shift
> COMPACT_MAX_DEFER_SHIFT
)
147 zone
->compact_defer_shift
= COMPACT_MAX_DEFER_SHIFT
;
149 trace_mm_compaction_defer_compaction(zone
, order
);
152 /* Returns true if compaction should be skipped this time */
153 bool compaction_deferred(struct zone
*zone
, int order
)
155 unsigned long defer_limit
= 1UL << zone
->compact_defer_shift
;
157 if (order
< zone
->compact_order_failed
)
160 /* Avoid possible overflow */
161 if (++zone
->compact_considered
> defer_limit
)
162 zone
->compact_considered
= defer_limit
;
164 if (zone
->compact_considered
>= defer_limit
)
167 trace_mm_compaction_deferred(zone
, order
);
173 * Update defer tracking counters after successful compaction of given order,
174 * which means an allocation either succeeded (alloc_success == true) or is
175 * expected to succeed.
177 void compaction_defer_reset(struct zone
*zone
, int order
,
181 zone
->compact_considered
= 0;
182 zone
->compact_defer_shift
= 0;
184 if (order
>= zone
->compact_order_failed
)
185 zone
->compact_order_failed
= order
+ 1;
187 trace_mm_compaction_defer_reset(zone
, order
);
190 /* Returns true if restarting compaction after many failures */
191 bool compaction_restarting(struct zone
*zone
, int order
)
193 if (order
< zone
->compact_order_failed
)
196 return zone
->compact_defer_shift
== COMPACT_MAX_DEFER_SHIFT
&&
197 zone
->compact_considered
>= 1UL << zone
->compact_defer_shift
;
200 /* Returns true if the pageblock should be scanned for pages to isolate. */
201 static inline bool isolation_suitable(struct compact_control
*cc
,
204 if (cc
->ignore_skip_hint
)
207 return !get_pageblock_skip(page
);
210 static void reset_cached_positions(struct zone
*zone
)
212 zone
->compact_cached_migrate_pfn
[0] = zone
->zone_start_pfn
;
213 zone
->compact_cached_migrate_pfn
[1] = zone
->zone_start_pfn
;
214 zone
->compact_cached_free_pfn
= zone_end_pfn(zone
);
218 * This function is called to clear all cached information on pageblocks that
219 * should be skipped for page isolation when the migrate and free page scanner
222 static void __reset_isolation_suitable(struct zone
*zone
)
224 unsigned long start_pfn
= zone
->zone_start_pfn
;
225 unsigned long end_pfn
= zone_end_pfn(zone
);
228 zone
->compact_blockskip_flush
= false;
230 /* Walk the zone and mark every pageblock as suitable for isolation */
231 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
239 page
= pfn_to_page(pfn
);
240 if (zone
!= page_zone(page
))
243 clear_pageblock_skip(page
);
246 reset_cached_positions(zone
);
249 void reset_isolation_suitable(pg_data_t
*pgdat
)
253 for (zoneid
= 0; zoneid
< MAX_NR_ZONES
; zoneid
++) {
254 struct zone
*zone
= &pgdat
->node_zones
[zoneid
];
255 if (!populated_zone(zone
))
258 /* Only flush if a full compaction finished recently */
259 if (zone
->compact_blockskip_flush
)
260 __reset_isolation_suitable(zone
);
265 * If no pages were isolated then mark this pageblock to be skipped in the
266 * future. The information is later cleared by __reset_isolation_suitable().
268 static void update_pageblock_skip(struct compact_control
*cc
,
269 struct page
*page
, unsigned long nr_isolated
,
270 bool migrate_scanner
)
272 struct zone
*zone
= cc
->zone
;
275 if (cc
->ignore_skip_hint
)
284 set_pageblock_skip(page
);
286 pfn
= page_to_pfn(page
);
288 /* Update where async and sync compaction should restart */
289 if (migrate_scanner
) {
290 if (pfn
> zone
->compact_cached_migrate_pfn
[0])
291 zone
->compact_cached_migrate_pfn
[0] = pfn
;
292 if (cc
->mode
!= MIGRATE_ASYNC
&&
293 pfn
> zone
->compact_cached_migrate_pfn
[1])
294 zone
->compact_cached_migrate_pfn
[1] = pfn
;
296 if (pfn
< zone
->compact_cached_free_pfn
)
297 zone
->compact_cached_free_pfn
= pfn
;
301 static inline bool isolation_suitable(struct compact_control
*cc
,
307 static void update_pageblock_skip(struct compact_control
*cc
,
308 struct page
*page
, unsigned long nr_isolated
,
309 bool migrate_scanner
)
312 #endif /* CONFIG_COMPACTION */
315 * Compaction requires the taking of some coarse locks that are potentially
316 * very heavily contended. For async compaction, back out if the lock cannot
317 * be taken immediately. For sync compaction, spin on the lock if needed.
319 * Returns true if the lock is held
320 * Returns false if the lock is not held and compaction should abort
322 static bool compact_trylock_irqsave(spinlock_t
*lock
, unsigned long *flags
,
323 struct compact_control
*cc
)
325 if (cc
->mode
== MIGRATE_ASYNC
) {
326 if (!spin_trylock_irqsave(lock
, *flags
)) {
327 cc
->contended
= COMPACT_CONTENDED_LOCK
;
331 spin_lock_irqsave(lock
, *flags
);
338 * Compaction requires the taking of some coarse locks that are potentially
339 * very heavily contended. The lock should be periodically unlocked to avoid
340 * having disabled IRQs for a long time, even when there is nobody waiting on
341 * the lock. It might also be that allowing the IRQs will result in
342 * need_resched() becoming true. If scheduling is needed, async compaction
343 * aborts. Sync compaction schedules.
344 * Either compaction type will also abort if a fatal signal is pending.
345 * In either case if the lock was locked, it is dropped and not regained.
347 * Returns true if compaction should abort due to fatal signal pending, or
348 * async compaction due to need_resched()
349 * Returns false when compaction can continue (sync compaction might have
352 static bool compact_unlock_should_abort(spinlock_t
*lock
,
353 unsigned long flags
, bool *locked
, struct compact_control
*cc
)
356 spin_unlock_irqrestore(lock
, flags
);
360 if (fatal_signal_pending(current
)) {
361 cc
->contended
= COMPACT_CONTENDED_SCHED
;
365 if (need_resched()) {
366 if (cc
->mode
== MIGRATE_ASYNC
) {
367 cc
->contended
= COMPACT_CONTENDED_SCHED
;
377 * Aside from avoiding lock contention, compaction also periodically checks
378 * need_resched() and either schedules in sync compaction or aborts async
379 * compaction. This is similar to what compact_unlock_should_abort() does, but
380 * is used where no lock is concerned.
382 * Returns false when no scheduling was needed, or sync compaction scheduled.
383 * Returns true when async compaction should abort.
385 static inline bool compact_should_abort(struct compact_control
*cc
)
387 /* async compaction aborts if contended */
388 if (need_resched()) {
389 if (cc
->mode
== MIGRATE_ASYNC
) {
390 cc
->contended
= COMPACT_CONTENDED_SCHED
;
401 * Isolate free pages onto a private freelist. If @strict is true, will abort
402 * returning 0 on any invalid PFNs or non-free pages inside of the pageblock
403 * (even though it may still end up isolating some pages).
405 static unsigned long isolate_freepages_block(struct compact_control
*cc
,
406 unsigned long *start_pfn
,
407 unsigned long end_pfn
,
408 struct list_head
*freelist
,
411 int nr_scanned
= 0, total_isolated
= 0;
412 struct page
*cursor
, *valid_page
= NULL
;
413 unsigned long flags
= 0;
415 unsigned long blockpfn
= *start_pfn
;
417 cursor
= pfn_to_page(blockpfn
);
419 /* Isolate free pages. */
420 for (; blockpfn
< end_pfn
; blockpfn
++, cursor
++) {
422 struct page
*page
= cursor
;
425 * Periodically drop the lock (if held) regardless of its
426 * contention, to give chance to IRQs. Abort if fatal signal
427 * pending or async compaction detects need_resched()
429 if (!(blockpfn
% SWAP_CLUSTER_MAX
)
430 && compact_unlock_should_abort(&cc
->zone
->lock
, flags
,
435 if (!pfn_valid_within(blockpfn
))
440 if (!PageBuddy(page
))
444 * If we already hold the lock, we can skip some rechecking.
445 * Note that if we hold the lock now, checked_pageblock was
446 * already set in some previous iteration (or strict is true),
447 * so it is correct to skip the suitable migration target
452 * The zone lock must be held to isolate freepages.
453 * Unfortunately this is a very coarse lock and can be
454 * heavily contended if there are parallel allocations
455 * or parallel compactions. For async compaction do not
456 * spin on the lock and we acquire the lock as late as
459 locked
= compact_trylock_irqsave(&cc
->zone
->lock
,
464 /* Recheck this is a buddy page under lock */
465 if (!PageBuddy(page
))
469 /* Found a free page, break it into order-0 pages */
470 isolated
= split_free_page(page
);
471 total_isolated
+= isolated
;
472 for (i
= 0; i
< isolated
; i
++) {
473 list_add(&page
->lru
, freelist
);
477 /* If a page was split, advance to the end of it */
479 cc
->nr_freepages
+= isolated
;
481 cc
->nr_migratepages
<= cc
->nr_freepages
) {
482 blockpfn
+= isolated
;
486 blockpfn
+= isolated
- 1;
487 cursor
+= isolated
- 1;
499 trace_mm_compaction_isolate_freepages(*start_pfn
, blockpfn
,
500 nr_scanned
, total_isolated
);
502 /* Record how far we have got within the block */
503 *start_pfn
= blockpfn
;
506 * If strict isolation is requested by CMA then check that all the
507 * pages requested were isolated. If there were any failures, 0 is
508 * returned and CMA will fail.
510 if (strict
&& blockpfn
< end_pfn
)
514 spin_unlock_irqrestore(&cc
->zone
->lock
, flags
);
516 /* Update the pageblock-skip if the whole pageblock was scanned */
517 if (blockpfn
== end_pfn
)
518 update_pageblock_skip(cc
, valid_page
, total_isolated
, false);
520 count_compact_events(COMPACTFREE_SCANNED
, nr_scanned
);
522 count_compact_events(COMPACTISOLATED
, total_isolated
);
523 return total_isolated
;
527 * isolate_freepages_range() - isolate free pages.
528 * @start_pfn: The first PFN to start isolating.
529 * @end_pfn: The one-past-last PFN.
531 * Non-free pages, invalid PFNs, or zone boundaries within the
532 * [start_pfn, end_pfn) range are considered errors, cause function to
533 * undo its actions and return zero.
535 * Otherwise, function returns one-past-the-last PFN of isolated page
536 * (which may be greater then end_pfn if end fell in a middle of
540 isolate_freepages_range(struct compact_control
*cc
,
541 unsigned long start_pfn
, unsigned long end_pfn
)
543 unsigned long isolated
, pfn
, block_end_pfn
;
547 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
549 for (; pfn
< end_pfn
; pfn
+= isolated
,
550 block_end_pfn
+= pageblock_nr_pages
) {
551 /* Protect pfn from changing by isolate_freepages_block */
552 unsigned long isolate_start_pfn
= pfn
;
554 block_end_pfn
= min(block_end_pfn
, end_pfn
);
557 * pfn could pass the block_end_pfn if isolated freepage
558 * is more than pageblock order. In this case, we adjust
559 * scanning range to right one.
561 if (pfn
>= block_end_pfn
) {
562 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
563 block_end_pfn
= min(block_end_pfn
, end_pfn
);
566 if (!pageblock_pfn_to_page(pfn
, block_end_pfn
, cc
->zone
))
569 isolated
= isolate_freepages_block(cc
, &isolate_start_pfn
,
570 block_end_pfn
, &freelist
, true);
573 * In strict mode, isolate_freepages_block() returns 0 if
574 * there are any holes in the block (ie. invalid PFNs or
581 * If we managed to isolate pages, it is always (1 << n) *
582 * pageblock_nr_pages for some non-negative n. (Max order
583 * page may span two pageblocks).
587 /* split_free_page does not map the pages */
588 map_pages(&freelist
);
591 /* Loop terminated early, cleanup. */
592 release_freepages(&freelist
);
596 /* We don't use freelists for anything. */
600 /* Update the number of anon and file isolated pages in the zone */
601 static void acct_isolated(struct zone
*zone
, struct compact_control
*cc
)
604 unsigned int count
[2] = { 0, };
606 if (list_empty(&cc
->migratepages
))
609 list_for_each_entry(page
, &cc
->migratepages
, lru
)
610 count
[!!page_is_file_cache(page
)]++;
612 mod_zone_page_state(zone
, NR_ISOLATED_ANON
, count
[0]);
613 mod_zone_page_state(zone
, NR_ISOLATED_FILE
, count
[1]);
616 /* Similar to reclaim, but different enough that they don't share logic */
617 static bool too_many_isolated(struct zone
*zone
)
619 unsigned long active
, inactive
, isolated
;
621 inactive
= zone_page_state(zone
, NR_INACTIVE_FILE
) +
622 zone_page_state(zone
, NR_INACTIVE_ANON
);
623 active
= zone_page_state(zone
, NR_ACTIVE_FILE
) +
624 zone_page_state(zone
, NR_ACTIVE_ANON
);
625 isolated
= zone_page_state(zone
, NR_ISOLATED_FILE
) +
626 zone_page_state(zone
, NR_ISOLATED_ANON
);
628 return isolated
> (inactive
+ active
) / 2;
632 * isolate_migratepages_block() - isolate all migrate-able pages within
634 * @cc: Compaction control structure.
635 * @low_pfn: The first PFN to isolate
636 * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock
637 * @isolate_mode: Isolation mode to be used.
639 * Isolate all pages that can be migrated from the range specified by
640 * [low_pfn, end_pfn). The range is expected to be within same pageblock.
641 * Returns zero if there is a fatal signal pending, otherwise PFN of the
642 * first page that was not scanned (which may be both less, equal to or more
645 * The pages are isolated on cc->migratepages list (not required to be empty),
646 * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field
647 * is neither read nor updated.
650 isolate_migratepages_block(struct compact_control
*cc
, unsigned long low_pfn
,
651 unsigned long end_pfn
, isolate_mode_t isolate_mode
)
653 struct zone
*zone
= cc
->zone
;
654 unsigned long nr_scanned
= 0, nr_isolated
= 0;
655 struct list_head
*migratelist
= &cc
->migratepages
;
656 struct lruvec
*lruvec
;
657 unsigned long flags
= 0;
659 struct page
*page
= NULL
, *valid_page
= NULL
;
660 unsigned long start_pfn
= low_pfn
;
663 * Ensure that there are not too many pages isolated from the LRU
664 * list by either parallel reclaimers or compaction. If there are,
665 * delay for some time until fewer pages are isolated
667 while (unlikely(too_many_isolated(zone
))) {
668 /* async migration should just abort */
669 if (cc
->mode
== MIGRATE_ASYNC
)
672 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
674 if (fatal_signal_pending(current
))
678 if (compact_should_abort(cc
))
681 /* Time to isolate some pages for migration */
682 for (; low_pfn
< end_pfn
; low_pfn
++) {
684 * Periodically drop the lock (if held) regardless of its
685 * contention, to give chance to IRQs. Abort async compaction
688 if (!(low_pfn
% SWAP_CLUSTER_MAX
)
689 && compact_unlock_should_abort(&zone
->lru_lock
, flags
,
693 if (!pfn_valid_within(low_pfn
))
697 page
= pfn_to_page(low_pfn
);
703 * Skip if free. We read page order here without zone lock
704 * which is generally unsafe, but the race window is small and
705 * the worst thing that can happen is that we skip some
706 * potential isolation targets.
708 if (PageBuddy(page
)) {
709 unsigned long freepage_order
= page_order_unsafe(page
);
712 * Without lock, we cannot be sure that what we got is
713 * a valid page order. Consider only values in the
714 * valid order range to prevent low_pfn overflow.
716 if (freepage_order
> 0 && freepage_order
< MAX_ORDER
)
717 low_pfn
+= (1UL << freepage_order
) - 1;
722 * Check may be lockless but that's ok as we recheck later.
723 * It's possible to migrate LRU pages and balloon pages
724 * Skip any other type of page
726 if (!PageLRU(page
)) {
727 if (unlikely(balloon_page_movable(page
))) {
728 if (balloon_page_isolate(page
)) {
729 /* Successfully isolated */
730 goto isolate_success
;
737 * PageLRU is set. lru_lock normally excludes isolation
738 * splitting and collapsing (collapsing has already happened
739 * if PageLRU is set) but the lock is not necessarily taken
740 * here and it is wasteful to take it just to check transhuge.
741 * Check TransHuge without lock and skip the whole pageblock if
742 * it's either a transhuge or hugetlbfs page, as calling
743 * compound_order() without preventing THP from splitting the
744 * page underneath us may return surprising results.
746 if (PageTransHuge(page
)) {
748 low_pfn
= ALIGN(low_pfn
+ 1,
749 pageblock_nr_pages
) - 1;
751 low_pfn
+= (1 << compound_order(page
)) - 1;
757 * Migration will fail if an anonymous page is pinned in memory,
758 * so avoid taking lru_lock and isolating it unnecessarily in an
759 * admittedly racy check.
761 if (!page_mapping(page
) &&
762 page_count(page
) > page_mapcount(page
))
765 /* If we already hold the lock, we can skip some rechecking */
767 locked
= compact_trylock_irqsave(&zone
->lru_lock
,
772 /* Recheck PageLRU and PageTransHuge under lock */
775 if (PageTransHuge(page
)) {
776 low_pfn
+= (1 << compound_order(page
)) - 1;
781 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
783 /* Try isolate the page */
784 if (__isolate_lru_page(page
, isolate_mode
) != 0)
787 VM_BUG_ON_PAGE(PageTransCompound(page
), page
);
789 /* Successfully isolated */
790 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
793 list_add(&page
->lru
, migratelist
);
794 cc
->nr_migratepages
++;
797 /* Avoid isolating too much */
798 if (cc
->nr_migratepages
== COMPACT_CLUSTER_MAX
) {
805 * The PageBuddy() check could have potentially brought us outside
806 * the range to be scanned.
808 if (unlikely(low_pfn
> end_pfn
))
812 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
815 * Update the pageblock-skip information and cached scanner pfn,
816 * if the whole pageblock was scanned without isolating any page.
818 if (low_pfn
== end_pfn
)
819 update_pageblock_skip(cc
, valid_page
, nr_isolated
, true);
821 trace_mm_compaction_isolate_migratepages(start_pfn
, low_pfn
,
822 nr_scanned
, nr_isolated
);
824 count_compact_events(COMPACTMIGRATE_SCANNED
, nr_scanned
);
826 count_compact_events(COMPACTISOLATED
, nr_isolated
);
832 * isolate_migratepages_range() - isolate migrate-able pages in a PFN range
833 * @cc: Compaction control structure.
834 * @start_pfn: The first PFN to start isolating.
835 * @end_pfn: The one-past-last PFN.
837 * Returns zero if isolation fails fatally due to e.g. pending signal.
838 * Otherwise, function returns one-past-the-last PFN of isolated page
839 * (which may be greater than end_pfn if end fell in a middle of a THP page).
842 isolate_migratepages_range(struct compact_control
*cc
, unsigned long start_pfn
,
843 unsigned long end_pfn
)
845 unsigned long pfn
, block_end_pfn
;
847 /* Scan block by block. First and last block may be incomplete */
849 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
851 for (; pfn
< end_pfn
; pfn
= block_end_pfn
,
852 block_end_pfn
+= pageblock_nr_pages
) {
854 block_end_pfn
= min(block_end_pfn
, end_pfn
);
856 if (!pageblock_pfn_to_page(pfn
, block_end_pfn
, cc
->zone
))
859 pfn
= isolate_migratepages_block(cc
, pfn
, block_end_pfn
,
860 ISOLATE_UNEVICTABLE
);
863 * In case of fatal failure, release everything that might
864 * have been isolated in the previous iteration, and signal
865 * the failure back to caller.
868 putback_movable_pages(&cc
->migratepages
);
869 cc
->nr_migratepages
= 0;
873 if (cc
->nr_migratepages
== COMPACT_CLUSTER_MAX
)
876 acct_isolated(cc
->zone
, cc
);
881 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
882 #ifdef CONFIG_COMPACTION
884 /* Returns true if the page is within a block suitable for migration to */
885 static bool suitable_migration_target(struct page
*page
)
887 /* If the page is a large free page, then disallow migration */
888 if (PageBuddy(page
)) {
890 * We are checking page_order without zone->lock taken. But
891 * the only small danger is that we skip a potentially suitable
892 * pageblock, so it's not worth to check order for valid range.
894 if (page_order_unsafe(page
) >= pageblock_order
)
898 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
899 if (migrate_async_suitable(get_pageblock_migratetype(page
)))
902 /* Otherwise skip the block */
907 * Test whether the free scanner has reached the same or lower pageblock than
908 * the migration scanner, and compaction should thus terminate.
910 static inline bool compact_scanners_met(struct compact_control
*cc
)
912 return (cc
->free_pfn
>> pageblock_order
)
913 <= (cc
->migrate_pfn
>> pageblock_order
);
917 * Based on information in the current compact_control, find blocks
918 * suitable for isolating free pages from and then isolate them.
920 static void isolate_freepages(struct compact_control
*cc
)
922 struct zone
*zone
= cc
->zone
;
924 unsigned long block_start_pfn
; /* start of current pageblock */
925 unsigned long isolate_start_pfn
; /* exact pfn we start at */
926 unsigned long block_end_pfn
; /* end of current pageblock */
927 unsigned long low_pfn
; /* lowest pfn scanner is able to scan */
928 struct list_head
*freelist
= &cc
->freepages
;
931 * Initialise the free scanner. The starting point is where we last
932 * successfully isolated from, zone-cached value, or the end of the
933 * zone when isolating for the first time. For looping we also need
934 * this pfn aligned down to the pageblock boundary, because we do
935 * block_start_pfn -= pageblock_nr_pages in the for loop.
936 * For ending point, take care when isolating in last pageblock of a
937 * a zone which ends in the middle of a pageblock.
938 * The low boundary is the end of the pageblock the migration scanner
941 isolate_start_pfn
= cc
->free_pfn
;
942 block_start_pfn
= cc
->free_pfn
& ~(pageblock_nr_pages
-1);
943 block_end_pfn
= min(block_start_pfn
+ pageblock_nr_pages
,
945 low_pfn
= ALIGN(cc
->migrate_pfn
+ 1, pageblock_nr_pages
);
948 * Isolate free pages until enough are available to migrate the
949 * pages on cc->migratepages. We stop searching if the migrate
950 * and free page scanners meet or enough free pages are isolated.
952 for (; block_start_pfn
>= low_pfn
;
953 block_end_pfn
= block_start_pfn
,
954 block_start_pfn
-= pageblock_nr_pages
,
955 isolate_start_pfn
= block_start_pfn
) {
958 * This can iterate a massively long zone without finding any
959 * suitable migration targets, so periodically check if we need
960 * to schedule, or even abort async compaction.
962 if (!(block_start_pfn
% (SWAP_CLUSTER_MAX
* pageblock_nr_pages
))
963 && compact_should_abort(cc
))
966 page
= pageblock_pfn_to_page(block_start_pfn
, block_end_pfn
,
971 /* Check the block is suitable for migration */
972 if (!suitable_migration_target(page
))
975 /* If isolation recently failed, do not retry */
976 if (!isolation_suitable(cc
, page
))
979 /* Found a block suitable for isolating free pages from. */
980 isolate_freepages_block(cc
, &isolate_start_pfn
,
981 block_end_pfn
, freelist
, false);
984 * If we isolated enough freepages, or aborted due to async
985 * compaction being contended, terminate the loop.
986 * Remember where the free scanner should restart next time,
987 * which is where isolate_freepages_block() left off.
988 * But if it scanned the whole pageblock, isolate_start_pfn
989 * now points at block_end_pfn, which is the start of the next
991 * In that case we will however want to restart at the start
992 * of the previous pageblock.
994 if ((cc
->nr_freepages
>= cc
->nr_migratepages
)
996 if (isolate_start_pfn
>= block_end_pfn
)
998 block_start_pfn
- pageblock_nr_pages
;
1002 * isolate_freepages_block() should not terminate
1003 * prematurely unless contended, or isolated enough
1005 VM_BUG_ON(isolate_start_pfn
< block_end_pfn
);
1009 /* split_free_page does not map the pages */
1010 map_pages(freelist
);
1013 * Record where the free scanner will restart next time. Either we
1014 * broke from the loop and set isolate_start_pfn based on the last
1015 * call to isolate_freepages_block(), or we met the migration scanner
1016 * and the loop terminated due to isolate_start_pfn < low_pfn
1018 cc
->free_pfn
= isolate_start_pfn
;
1022 * This is a migrate-callback that "allocates" freepages by taking pages
1023 * from the isolated freelists in the block we are migrating to.
1025 static struct page
*compaction_alloc(struct page
*migratepage
,
1029 struct compact_control
*cc
= (struct compact_control
*)data
;
1030 struct page
*freepage
;
1033 * Isolate free pages if necessary, and if we are not aborting due to
1036 if (list_empty(&cc
->freepages
)) {
1038 isolate_freepages(cc
);
1040 if (list_empty(&cc
->freepages
))
1044 freepage
= list_entry(cc
->freepages
.next
, struct page
, lru
);
1045 list_del(&freepage
->lru
);
1052 * This is a migrate-callback that "frees" freepages back to the isolated
1053 * freelist. All pages on the freelist are from the same zone, so there is no
1054 * special handling needed for NUMA.
1056 static void compaction_free(struct page
*page
, unsigned long data
)
1058 struct compact_control
*cc
= (struct compact_control
*)data
;
1060 list_add(&page
->lru
, &cc
->freepages
);
1064 /* possible outcome of isolate_migratepages */
1066 ISOLATE_ABORT
, /* Abort compaction now */
1067 ISOLATE_NONE
, /* No pages isolated, continue scanning */
1068 ISOLATE_SUCCESS
, /* Pages isolated, migrate */
1069 } isolate_migrate_t
;
1072 * Allow userspace to control policy on scanning the unevictable LRU for
1073 * compactable pages.
1075 int sysctl_compact_unevictable_allowed __read_mostly
= 1;
1078 * Isolate all pages that can be migrated from the first suitable block,
1079 * starting at the block pointed to by the migrate scanner pfn within
1082 static isolate_migrate_t
isolate_migratepages(struct zone
*zone
,
1083 struct compact_control
*cc
)
1085 unsigned long low_pfn
, end_pfn
;
1087 const isolate_mode_t isolate_mode
=
1088 (sysctl_compact_unevictable_allowed
? ISOLATE_UNEVICTABLE
: 0) |
1089 (cc
->mode
== MIGRATE_ASYNC
? ISOLATE_ASYNC_MIGRATE
: 0);
1092 * Start at where we last stopped, or beginning of the zone as
1093 * initialized by compact_zone()
1095 low_pfn
= cc
->migrate_pfn
;
1097 /* Only scan within a pageblock boundary */
1098 end_pfn
= ALIGN(low_pfn
+ 1, pageblock_nr_pages
);
1101 * Iterate over whole pageblocks until we find the first suitable.
1102 * Do not cross the free scanner.
1104 for (; end_pfn
<= cc
->free_pfn
;
1105 low_pfn
= end_pfn
, end_pfn
+= pageblock_nr_pages
) {
1108 * This can potentially iterate a massively long zone with
1109 * many pageblocks unsuitable, so periodically check if we
1110 * need to schedule, or even abort async compaction.
1112 if (!(low_pfn
% (SWAP_CLUSTER_MAX
* pageblock_nr_pages
))
1113 && compact_should_abort(cc
))
1116 page
= pageblock_pfn_to_page(low_pfn
, end_pfn
, zone
);
1120 /* If isolation recently failed, do not retry */
1121 if (!isolation_suitable(cc
, page
))
1125 * For async compaction, also only scan in MOVABLE blocks.
1126 * Async compaction is optimistic to see if the minimum amount
1127 * of work satisfies the allocation.
1129 if (cc
->mode
== MIGRATE_ASYNC
&&
1130 !migrate_async_suitable(get_pageblock_migratetype(page
)))
1133 /* Perform the isolation */
1134 low_pfn
= isolate_migratepages_block(cc
, low_pfn
, end_pfn
,
1137 if (!low_pfn
|| cc
->contended
) {
1138 acct_isolated(zone
, cc
);
1139 return ISOLATE_ABORT
;
1143 * Either we isolated something and proceed with migration. Or
1144 * we failed and compact_zone should decide if we should
1150 acct_isolated(zone
, cc
);
1151 /* Record where migration scanner will be restarted. */
1152 cc
->migrate_pfn
= low_pfn
;
1154 return cc
->nr_migratepages
? ISOLATE_SUCCESS
: ISOLATE_NONE
;
1157 static int __compact_finished(struct zone
*zone
, struct compact_control
*cc
,
1158 const int migratetype
)
1161 unsigned long watermark
;
1163 if (cc
->contended
|| fatal_signal_pending(current
))
1164 return COMPACT_PARTIAL
;
1166 /* Compaction run completes if the migrate and free scanner meet */
1167 if (compact_scanners_met(cc
)) {
1168 /* Let the next compaction start anew. */
1169 reset_cached_positions(zone
);
1172 * Mark that the PG_migrate_skip information should be cleared
1173 * by kswapd when it goes to sleep. kswapd does not set the
1174 * flag itself as the decision to be clear should be directly
1175 * based on an allocation request.
1177 if (!current_is_kswapd())
1178 zone
->compact_blockskip_flush
= true;
1180 return COMPACT_COMPLETE
;
1184 * order == -1 is expected when compacting via
1185 * /proc/sys/vm/compact_memory
1187 if (cc
->order
== -1)
1188 return COMPACT_CONTINUE
;
1190 /* Compaction run is not finished if the watermark is not met */
1191 watermark
= low_wmark_pages(zone
);
1193 if (!zone_watermark_ok(zone
, cc
->order
, watermark
, cc
->classzone_idx
,
1195 return COMPACT_CONTINUE
;
1197 /* Direct compactor: Is a suitable page free? */
1198 for (order
= cc
->order
; order
< MAX_ORDER
; order
++) {
1199 struct free_area
*area
= &zone
->free_area
[order
];
1202 /* Job done if page is free of the right migratetype */
1203 if (!list_empty(&area
->free_list
[migratetype
]))
1204 return COMPACT_PARTIAL
;
1207 /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */
1208 if (migratetype
== MIGRATE_MOVABLE
&&
1209 !list_empty(&area
->free_list
[MIGRATE_CMA
]))
1210 return COMPACT_PARTIAL
;
1213 * Job done if allocation would steal freepages from
1214 * other migratetype buddy lists.
1216 if (find_suitable_fallback(area
, order
, migratetype
,
1217 true, &can_steal
) != -1)
1218 return COMPACT_PARTIAL
;
1221 return COMPACT_NO_SUITABLE_PAGE
;
1224 static int compact_finished(struct zone
*zone
, struct compact_control
*cc
,
1225 const int migratetype
)
1229 ret
= __compact_finished(zone
, cc
, migratetype
);
1230 trace_mm_compaction_finished(zone
, cc
->order
, ret
);
1231 if (ret
== COMPACT_NO_SUITABLE_PAGE
)
1232 ret
= COMPACT_CONTINUE
;
1238 * compaction_suitable: Is this suitable to run compaction on this zone now?
1240 * COMPACT_SKIPPED - If there are too few free pages for compaction
1241 * COMPACT_PARTIAL - If the allocation would succeed without compaction
1242 * COMPACT_CONTINUE - If compaction should run now
1244 static unsigned long __compaction_suitable(struct zone
*zone
, int order
,
1245 int alloc_flags
, int classzone_idx
)
1248 unsigned long watermark
;
1251 * order == -1 is expected when compacting via
1252 * /proc/sys/vm/compact_memory
1255 return COMPACT_CONTINUE
;
1257 watermark
= low_wmark_pages(zone
);
1259 * If watermarks for high-order allocation are already met, there
1260 * should be no need for compaction at all.
1262 if (zone_watermark_ok(zone
, order
, watermark
, classzone_idx
,
1264 return COMPACT_PARTIAL
;
1267 * Watermarks for order-0 must be met for compaction. Note the 2UL.
1268 * This is because during migration, copies of pages need to be
1269 * allocated and for a short time, the footprint is higher
1271 watermark
+= (2UL << order
);
1272 if (!zone_watermark_ok(zone
, 0, watermark
, classzone_idx
, alloc_flags
))
1273 return COMPACT_SKIPPED
;
1276 * fragmentation index determines if allocation failures are due to
1277 * low memory or external fragmentation
1279 * index of -1000 would imply allocations might succeed depending on
1280 * watermarks, but we already failed the high-order watermark check
1281 * index towards 0 implies failure is due to lack of memory
1282 * index towards 1000 implies failure is due to fragmentation
1284 * Only compact if a failure would be due to fragmentation.
1286 fragindex
= fragmentation_index(zone
, order
);
1287 if (fragindex
>= 0 && fragindex
<= sysctl_extfrag_threshold
)
1288 return COMPACT_NOT_SUITABLE_ZONE
;
1290 return COMPACT_CONTINUE
;
1293 unsigned long compaction_suitable(struct zone
*zone
, int order
,
1294 int alloc_flags
, int classzone_idx
)
1298 ret
= __compaction_suitable(zone
, order
, alloc_flags
, classzone_idx
);
1299 trace_mm_compaction_suitable(zone
, order
, ret
);
1300 if (ret
== COMPACT_NOT_SUITABLE_ZONE
)
1301 ret
= COMPACT_SKIPPED
;
1306 static int compact_zone(struct zone
*zone
, struct compact_control
*cc
)
1309 unsigned long start_pfn
= zone
->zone_start_pfn
;
1310 unsigned long end_pfn
= zone_end_pfn(zone
);
1311 const int migratetype
= gfpflags_to_migratetype(cc
->gfp_mask
);
1312 const bool sync
= cc
->mode
!= MIGRATE_ASYNC
;
1313 unsigned long last_migrated_pfn
= 0;
1315 ret
= compaction_suitable(zone
, cc
->order
, cc
->alloc_flags
,
1318 case COMPACT_PARTIAL
:
1319 case COMPACT_SKIPPED
:
1320 /* Compaction is likely to fail */
1322 case COMPACT_CONTINUE
:
1323 /* Fall through to compaction */
1328 * Clear pageblock skip if there were failures recently and compaction
1329 * is about to be retried after being deferred. kswapd does not do
1330 * this reset as it'll reset the cached information when going to sleep.
1332 if (compaction_restarting(zone
, cc
->order
) && !current_is_kswapd())
1333 __reset_isolation_suitable(zone
);
1336 * Setup to move all movable pages to the end of the zone. Used cached
1337 * information on where the scanners should start but check that it
1338 * is initialised by ensuring the values are within zone boundaries.
1340 cc
->migrate_pfn
= zone
->compact_cached_migrate_pfn
[sync
];
1341 cc
->free_pfn
= zone
->compact_cached_free_pfn
;
1342 if (cc
->free_pfn
< start_pfn
|| cc
->free_pfn
> end_pfn
) {
1343 cc
->free_pfn
= end_pfn
& ~(pageblock_nr_pages
-1);
1344 zone
->compact_cached_free_pfn
= cc
->free_pfn
;
1346 if (cc
->migrate_pfn
< start_pfn
|| cc
->migrate_pfn
> end_pfn
) {
1347 cc
->migrate_pfn
= start_pfn
;
1348 zone
->compact_cached_migrate_pfn
[0] = cc
->migrate_pfn
;
1349 zone
->compact_cached_migrate_pfn
[1] = cc
->migrate_pfn
;
1352 trace_mm_compaction_begin(start_pfn
, cc
->migrate_pfn
,
1353 cc
->free_pfn
, end_pfn
, sync
);
1355 migrate_prep_local();
1357 while ((ret
= compact_finished(zone
, cc
, migratetype
)) ==
1360 unsigned long isolate_start_pfn
= cc
->migrate_pfn
;
1362 switch (isolate_migratepages(zone
, cc
)) {
1364 ret
= COMPACT_PARTIAL
;
1365 putback_movable_pages(&cc
->migratepages
);
1366 cc
->nr_migratepages
= 0;
1370 * We haven't isolated and migrated anything, but
1371 * there might still be unflushed migrations from
1372 * previous cc->order aligned block.
1375 case ISOLATE_SUCCESS
:
1379 err
= migrate_pages(&cc
->migratepages
, compaction_alloc
,
1380 compaction_free
, (unsigned long)cc
, cc
->mode
,
1383 trace_mm_compaction_migratepages(cc
->nr_migratepages
, err
,
1386 /* All pages were either migrated or will be released */
1387 cc
->nr_migratepages
= 0;
1389 putback_movable_pages(&cc
->migratepages
);
1391 * migrate_pages() may return -ENOMEM when scanners meet
1392 * and we want compact_finished() to detect it
1394 if (err
== -ENOMEM
&& !compact_scanners_met(cc
)) {
1395 ret
= COMPACT_PARTIAL
;
1401 * Record where we could have freed pages by migration and not
1402 * yet flushed them to buddy allocator. We use the pfn that
1403 * isolate_migratepages() started from in this loop iteration
1404 * - this is the lowest page that could have been isolated and
1405 * then freed by migration.
1407 if (!last_migrated_pfn
)
1408 last_migrated_pfn
= isolate_start_pfn
;
1412 * Has the migration scanner moved away from the previous
1413 * cc->order aligned block where we migrated from? If yes,
1414 * flush the pages that were freed, so that they can merge and
1415 * compact_finished() can detect immediately if allocation
1418 if (cc
->order
> 0 && last_migrated_pfn
) {
1420 unsigned long current_block_start
=
1421 cc
->migrate_pfn
& ~((1UL << cc
->order
) - 1);
1423 if (last_migrated_pfn
< current_block_start
) {
1425 lru_add_drain_cpu(cpu
);
1426 drain_local_pages(zone
);
1428 /* No more flushing until we migrate again */
1429 last_migrated_pfn
= 0;
1437 * Release free pages and update where the free scanner should restart,
1438 * so we don't leave any returned pages behind in the next attempt.
1440 if (cc
->nr_freepages
> 0) {
1441 unsigned long free_pfn
= release_freepages(&cc
->freepages
);
1443 cc
->nr_freepages
= 0;
1444 VM_BUG_ON(free_pfn
== 0);
1445 /* The cached pfn is always the first in a pageblock */
1446 free_pfn
&= ~(pageblock_nr_pages
-1);
1448 * Only go back, not forward. The cached pfn might have been
1449 * already reset to zone end in compact_finished()
1451 if (free_pfn
> zone
->compact_cached_free_pfn
)
1452 zone
->compact_cached_free_pfn
= free_pfn
;
1455 trace_mm_compaction_end(start_pfn
, cc
->migrate_pfn
,
1456 cc
->free_pfn
, end_pfn
, sync
, ret
);
1461 static unsigned long compact_zone_order(struct zone
*zone
, int order
,
1462 gfp_t gfp_mask
, enum migrate_mode mode
, int *contended
,
1463 int alloc_flags
, int classzone_idx
)
1466 struct compact_control cc
= {
1468 .nr_migratepages
= 0,
1470 .gfp_mask
= gfp_mask
,
1473 .alloc_flags
= alloc_flags
,
1474 .classzone_idx
= classzone_idx
,
1476 INIT_LIST_HEAD(&cc
.freepages
);
1477 INIT_LIST_HEAD(&cc
.migratepages
);
1479 ret
= compact_zone(zone
, &cc
);
1481 VM_BUG_ON(!list_empty(&cc
.freepages
));
1482 VM_BUG_ON(!list_empty(&cc
.migratepages
));
1484 *contended
= cc
.contended
;
1488 int sysctl_extfrag_threshold
= 500;
1491 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
1492 * @gfp_mask: The GFP mask of the current allocation
1493 * @order: The order of the current allocation
1494 * @alloc_flags: The allocation flags of the current allocation
1495 * @ac: The context of current allocation
1496 * @mode: The migration mode for async, sync light, or sync migration
1497 * @contended: Return value that determines if compaction was aborted due to
1498 * need_resched() or lock contention
1500 * This is the main entry point for direct page compaction.
1502 unsigned long try_to_compact_pages(gfp_t gfp_mask
, unsigned int order
,
1503 int alloc_flags
, const struct alloc_context
*ac
,
1504 enum migrate_mode mode
, int *contended
)
1506 int may_enter_fs
= gfp_mask
& __GFP_FS
;
1507 int may_perform_io
= gfp_mask
& __GFP_IO
;
1510 int rc
= COMPACT_DEFERRED
;
1511 int all_zones_contended
= COMPACT_CONTENDED_LOCK
; /* init for &= op */
1513 *contended
= COMPACT_CONTENDED_NONE
;
1515 /* Check if the GFP flags allow compaction */
1516 if (!order
|| !may_enter_fs
|| !may_perform_io
)
1517 return COMPACT_SKIPPED
;
1519 trace_mm_compaction_try_to_compact_pages(order
, gfp_mask
, mode
);
1521 /* Compact each zone in the list */
1522 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
1527 if (compaction_deferred(zone
, order
))
1530 status
= compact_zone_order(zone
, order
, gfp_mask
, mode
,
1531 &zone_contended
, alloc_flags
,
1533 rc
= max(status
, rc
);
1535 * It takes at least one zone that wasn't lock contended
1536 * to clear all_zones_contended.
1538 all_zones_contended
&= zone_contended
;
1540 /* If a normal allocation would succeed, stop compacting */
1541 if (zone_watermark_ok(zone
, order
, low_wmark_pages(zone
),
1542 ac
->classzone_idx
, alloc_flags
)) {
1544 * We think the allocation will succeed in this zone,
1545 * but it is not certain, hence the false. The caller
1546 * will repeat this with true if allocation indeed
1547 * succeeds in this zone.
1549 compaction_defer_reset(zone
, order
, false);
1551 * It is possible that async compaction aborted due to
1552 * need_resched() and the watermarks were ok thanks to
1553 * somebody else freeing memory. The allocation can
1554 * however still fail so we better signal the
1555 * need_resched() contention anyway (this will not
1556 * prevent the allocation attempt).
1558 if (zone_contended
== COMPACT_CONTENDED_SCHED
)
1559 *contended
= COMPACT_CONTENDED_SCHED
;
1564 if (mode
!= MIGRATE_ASYNC
&& status
== COMPACT_COMPLETE
) {
1566 * We think that allocation won't succeed in this zone
1567 * so we defer compaction there. If it ends up
1568 * succeeding after all, it will be reset.
1570 defer_compaction(zone
, order
);
1574 * We might have stopped compacting due to need_resched() in
1575 * async compaction, or due to a fatal signal detected. In that
1576 * case do not try further zones and signal need_resched()
1579 if ((zone_contended
== COMPACT_CONTENDED_SCHED
)
1580 || fatal_signal_pending(current
)) {
1581 *contended
= COMPACT_CONTENDED_SCHED
;
1588 * We might not have tried all the zones, so be conservative
1589 * and assume they are not all lock contended.
1591 all_zones_contended
= 0;
1596 * If at least one zone wasn't deferred or skipped, we report if all
1597 * zones that were tried were lock contended.
1599 if (rc
> COMPACT_SKIPPED
&& all_zones_contended
)
1600 *contended
= COMPACT_CONTENDED_LOCK
;
1606 /* Compact all zones within a node */
1607 static void __compact_pgdat(pg_data_t
*pgdat
, struct compact_control
*cc
)
1612 for (zoneid
= 0; zoneid
< MAX_NR_ZONES
; zoneid
++) {
1614 zone
= &pgdat
->node_zones
[zoneid
];
1615 if (!populated_zone(zone
))
1618 cc
->nr_freepages
= 0;
1619 cc
->nr_migratepages
= 0;
1621 INIT_LIST_HEAD(&cc
->freepages
);
1622 INIT_LIST_HEAD(&cc
->migratepages
);
1625 * When called via /proc/sys/vm/compact_memory
1626 * this makes sure we compact the whole zone regardless of
1627 * cached scanner positions.
1629 if (cc
->order
== -1)
1630 __reset_isolation_suitable(zone
);
1632 if (cc
->order
== -1 || !compaction_deferred(zone
, cc
->order
))
1633 compact_zone(zone
, cc
);
1635 if (cc
->order
> 0) {
1636 if (zone_watermark_ok(zone
, cc
->order
,
1637 low_wmark_pages(zone
), 0, 0))
1638 compaction_defer_reset(zone
, cc
->order
, false);
1641 VM_BUG_ON(!list_empty(&cc
->freepages
));
1642 VM_BUG_ON(!list_empty(&cc
->migratepages
));
1646 void compact_pgdat(pg_data_t
*pgdat
, int order
)
1648 struct compact_control cc
= {
1650 .mode
= MIGRATE_ASYNC
,
1656 __compact_pgdat(pgdat
, &cc
);
1659 static void compact_node(int nid
)
1661 struct compact_control cc
= {
1663 .mode
= MIGRATE_SYNC
,
1664 .ignore_skip_hint
= true,
1667 __compact_pgdat(NODE_DATA(nid
), &cc
);
1670 /* Compact all nodes in the system */
1671 static void compact_nodes(void)
1675 /* Flush pending updates to the LRU lists */
1676 lru_add_drain_all();
1678 for_each_online_node(nid
)
1682 /* The written value is actually unused, all memory is compacted */
1683 int sysctl_compact_memory
;
1685 /* This is the entry point for compacting all nodes via /proc/sys/vm */
1686 int sysctl_compaction_handler(struct ctl_table
*table
, int write
,
1687 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
1695 int sysctl_extfrag_handler(struct ctl_table
*table
, int write
,
1696 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
1698 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
1703 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
1704 static ssize_t
sysfs_compact_node(struct device
*dev
,
1705 struct device_attribute
*attr
,
1706 const char *buf
, size_t count
)
1710 if (nid
>= 0 && nid
< nr_node_ids
&& node_online(nid
)) {
1711 /* Flush pending updates to the LRU lists */
1712 lru_add_drain_all();
1719 static DEVICE_ATTR(compact
, S_IWUSR
, NULL
, sysfs_compact_node
);
1721 int compaction_register_node(struct node
*node
)
1723 return device_create_file(&node
->dev
, &dev_attr_compact
);
1726 void compaction_unregister_node(struct node
*node
)
1728 return device_remove_file(&node
->dev
, &dev_attr_compact
);
1730 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
1732 #endif /* CONFIG_COMPACTION */