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
);
211 * This function is called to clear all cached information on pageblocks that
212 * should be skipped for page isolation when the migrate and free page scanner
215 static void __reset_isolation_suitable(struct zone
*zone
)
217 unsigned long start_pfn
= zone
->zone_start_pfn
;
218 unsigned long end_pfn
= zone_end_pfn(zone
);
221 zone
->compact_cached_migrate_pfn
[0] = start_pfn
;
222 zone
->compact_cached_migrate_pfn
[1] = start_pfn
;
223 zone
->compact_cached_free_pfn
= end_pfn
;
224 zone
->compact_blockskip_flush
= false;
226 /* Walk the zone and mark every pageblock as suitable for isolation */
227 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
235 page
= pfn_to_page(pfn
);
236 if (zone
!= page_zone(page
))
239 clear_pageblock_skip(page
);
243 void reset_isolation_suitable(pg_data_t
*pgdat
)
247 for (zoneid
= 0; zoneid
< MAX_NR_ZONES
; zoneid
++) {
248 struct zone
*zone
= &pgdat
->node_zones
[zoneid
];
249 if (!populated_zone(zone
))
252 /* Only flush if a full compaction finished recently */
253 if (zone
->compact_blockskip_flush
)
254 __reset_isolation_suitable(zone
);
259 * If no pages were isolated then mark this pageblock to be skipped in the
260 * future. The information is later cleared by __reset_isolation_suitable().
262 static void update_pageblock_skip(struct compact_control
*cc
,
263 struct page
*page
, unsigned long nr_isolated
,
264 bool migrate_scanner
)
266 struct zone
*zone
= cc
->zone
;
269 if (cc
->ignore_skip_hint
)
278 set_pageblock_skip(page
);
280 pfn
= page_to_pfn(page
);
282 /* Update where async and sync compaction should restart */
283 if (migrate_scanner
) {
284 if (pfn
> zone
->compact_cached_migrate_pfn
[0])
285 zone
->compact_cached_migrate_pfn
[0] = pfn
;
286 if (cc
->mode
!= MIGRATE_ASYNC
&&
287 pfn
> zone
->compact_cached_migrate_pfn
[1])
288 zone
->compact_cached_migrate_pfn
[1] = pfn
;
290 if (pfn
< zone
->compact_cached_free_pfn
)
291 zone
->compact_cached_free_pfn
= pfn
;
295 static inline bool isolation_suitable(struct compact_control
*cc
,
301 static void update_pageblock_skip(struct compact_control
*cc
,
302 struct page
*page
, unsigned long nr_isolated
,
303 bool migrate_scanner
)
306 #endif /* CONFIG_COMPACTION */
309 * Compaction requires the taking of some coarse locks that are potentially
310 * very heavily contended. For async compaction, back out if the lock cannot
311 * be taken immediately. For sync compaction, spin on the lock if needed.
313 * Returns true if the lock is held
314 * Returns false if the lock is not held and compaction should abort
316 static bool compact_trylock_irqsave(spinlock_t
*lock
, unsigned long *flags
,
317 struct compact_control
*cc
)
319 if (cc
->mode
== MIGRATE_ASYNC
) {
320 if (!spin_trylock_irqsave(lock
, *flags
)) {
321 cc
->contended
= COMPACT_CONTENDED_LOCK
;
325 spin_lock_irqsave(lock
, *flags
);
332 * Compaction requires the taking of some coarse locks that are potentially
333 * very heavily contended. The lock should be periodically unlocked to avoid
334 * having disabled IRQs for a long time, even when there is nobody waiting on
335 * the lock. It might also be that allowing the IRQs will result in
336 * need_resched() becoming true. If scheduling is needed, async compaction
337 * aborts. Sync compaction schedules.
338 * Either compaction type will also abort if a fatal signal is pending.
339 * In either case if the lock was locked, it is dropped and not regained.
341 * Returns true if compaction should abort due to fatal signal pending, or
342 * async compaction due to need_resched()
343 * Returns false when compaction can continue (sync compaction might have
346 static bool compact_unlock_should_abort(spinlock_t
*lock
,
347 unsigned long flags
, bool *locked
, struct compact_control
*cc
)
350 spin_unlock_irqrestore(lock
, flags
);
354 if (fatal_signal_pending(current
)) {
355 cc
->contended
= COMPACT_CONTENDED_SCHED
;
359 if (need_resched()) {
360 if (cc
->mode
== MIGRATE_ASYNC
) {
361 cc
->contended
= COMPACT_CONTENDED_SCHED
;
371 * Aside from avoiding lock contention, compaction also periodically checks
372 * need_resched() and either schedules in sync compaction or aborts async
373 * compaction. This is similar to what compact_unlock_should_abort() does, but
374 * is used where no lock is concerned.
376 * Returns false when no scheduling was needed, or sync compaction scheduled.
377 * Returns true when async compaction should abort.
379 static inline bool compact_should_abort(struct compact_control
*cc
)
381 /* async compaction aborts if contended */
382 if (need_resched()) {
383 if (cc
->mode
== MIGRATE_ASYNC
) {
384 cc
->contended
= COMPACT_CONTENDED_SCHED
;
395 * Isolate free pages onto a private freelist. If @strict is true, will abort
396 * returning 0 on any invalid PFNs or non-free pages inside of the pageblock
397 * (even though it may still end up isolating some pages).
399 static unsigned long isolate_freepages_block(struct compact_control
*cc
,
400 unsigned long *start_pfn
,
401 unsigned long end_pfn
,
402 struct list_head
*freelist
,
405 int nr_scanned
= 0, total_isolated
= 0;
406 struct page
*cursor
, *valid_page
= NULL
;
407 unsigned long flags
= 0;
409 unsigned long blockpfn
= *start_pfn
;
411 cursor
= pfn_to_page(blockpfn
);
413 /* Isolate free pages. */
414 for (; blockpfn
< end_pfn
; blockpfn
++, cursor
++) {
416 struct page
*page
= cursor
;
419 * Periodically drop the lock (if held) regardless of its
420 * contention, to give chance to IRQs. Abort if fatal signal
421 * pending or async compaction detects need_resched()
423 if (!(blockpfn
% SWAP_CLUSTER_MAX
)
424 && compact_unlock_should_abort(&cc
->zone
->lock
, flags
,
429 if (!pfn_valid_within(blockpfn
))
434 if (!PageBuddy(page
))
438 * If we already hold the lock, we can skip some rechecking.
439 * Note that if we hold the lock now, checked_pageblock was
440 * already set in some previous iteration (or strict is true),
441 * so it is correct to skip the suitable migration target
446 * The zone lock must be held to isolate freepages.
447 * Unfortunately this is a very coarse lock and can be
448 * heavily contended if there are parallel allocations
449 * or parallel compactions. For async compaction do not
450 * spin on the lock and we acquire the lock as late as
453 locked
= compact_trylock_irqsave(&cc
->zone
->lock
,
458 /* Recheck this is a buddy page under lock */
459 if (!PageBuddy(page
))
463 /* Found a free page, break it into order-0 pages */
464 isolated
= split_free_page(page
);
465 total_isolated
+= isolated
;
466 for (i
= 0; i
< isolated
; i
++) {
467 list_add(&page
->lru
, freelist
);
471 /* If a page was split, advance to the end of it */
473 cc
->nr_freepages
+= isolated
;
475 cc
->nr_migratepages
<= cc
->nr_freepages
) {
476 blockpfn
+= isolated
;
480 blockpfn
+= isolated
- 1;
481 cursor
+= isolated
- 1;
493 trace_mm_compaction_isolate_freepages(*start_pfn
, blockpfn
,
494 nr_scanned
, total_isolated
);
496 /* Record how far we have got within the block */
497 *start_pfn
= blockpfn
;
500 * If strict isolation is requested by CMA then check that all the
501 * pages requested were isolated. If there were any failures, 0 is
502 * returned and CMA will fail.
504 if (strict
&& blockpfn
< end_pfn
)
508 spin_unlock_irqrestore(&cc
->zone
->lock
, flags
);
510 /* Update the pageblock-skip if the whole pageblock was scanned */
511 if (blockpfn
== end_pfn
)
512 update_pageblock_skip(cc
, valid_page
, total_isolated
, false);
514 count_compact_events(COMPACTFREE_SCANNED
, nr_scanned
);
516 count_compact_events(COMPACTISOLATED
, total_isolated
);
517 return total_isolated
;
521 * isolate_freepages_range() - isolate free pages.
522 * @start_pfn: The first PFN to start isolating.
523 * @end_pfn: The one-past-last PFN.
525 * Non-free pages, invalid PFNs, or zone boundaries within the
526 * [start_pfn, end_pfn) range are considered errors, cause function to
527 * undo its actions and return zero.
529 * Otherwise, function returns one-past-the-last PFN of isolated page
530 * (which may be greater then end_pfn if end fell in a middle of
534 isolate_freepages_range(struct compact_control
*cc
,
535 unsigned long start_pfn
, unsigned long end_pfn
)
537 unsigned long isolated
, pfn
, block_end_pfn
;
541 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
543 for (; pfn
< end_pfn
; pfn
+= isolated
,
544 block_end_pfn
+= pageblock_nr_pages
) {
545 /* Protect pfn from changing by isolate_freepages_block */
546 unsigned long isolate_start_pfn
= pfn
;
548 block_end_pfn
= min(block_end_pfn
, end_pfn
);
551 * pfn could pass the block_end_pfn if isolated freepage
552 * is more than pageblock order. In this case, we adjust
553 * scanning range to right one.
555 if (pfn
>= block_end_pfn
) {
556 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
557 block_end_pfn
= min(block_end_pfn
, end_pfn
);
560 if (!pageblock_pfn_to_page(pfn
, block_end_pfn
, cc
->zone
))
563 isolated
= isolate_freepages_block(cc
, &isolate_start_pfn
,
564 block_end_pfn
, &freelist
, true);
567 * In strict mode, isolate_freepages_block() returns 0 if
568 * there are any holes in the block (ie. invalid PFNs or
575 * If we managed to isolate pages, it is always (1 << n) *
576 * pageblock_nr_pages for some non-negative n. (Max order
577 * page may span two pageblocks).
581 /* split_free_page does not map the pages */
582 map_pages(&freelist
);
585 /* Loop terminated early, cleanup. */
586 release_freepages(&freelist
);
590 /* We don't use freelists for anything. */
594 /* Update the number of anon and file isolated pages in the zone */
595 static void acct_isolated(struct zone
*zone
, struct compact_control
*cc
)
598 unsigned int count
[2] = { 0, };
600 if (list_empty(&cc
->migratepages
))
603 list_for_each_entry(page
, &cc
->migratepages
, lru
)
604 count
[!!page_is_file_cache(page
)]++;
606 mod_zone_page_state(zone
, NR_ISOLATED_ANON
, count
[0]);
607 mod_zone_page_state(zone
, NR_ISOLATED_FILE
, count
[1]);
610 /* Similar to reclaim, but different enough that they don't share logic */
611 static bool too_many_isolated(struct zone
*zone
)
613 unsigned long active
, inactive
, isolated
;
615 inactive
= zone_page_state(zone
, NR_INACTIVE_FILE
) +
616 zone_page_state(zone
, NR_INACTIVE_ANON
);
617 active
= zone_page_state(zone
, NR_ACTIVE_FILE
) +
618 zone_page_state(zone
, NR_ACTIVE_ANON
);
619 isolated
= zone_page_state(zone
, NR_ISOLATED_FILE
) +
620 zone_page_state(zone
, NR_ISOLATED_ANON
);
622 return isolated
> (inactive
+ active
) / 2;
626 * isolate_migratepages_block() - isolate all migrate-able pages within
628 * @cc: Compaction control structure.
629 * @low_pfn: The first PFN to isolate
630 * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock
631 * @isolate_mode: Isolation mode to be used.
633 * Isolate all pages that can be migrated from the range specified by
634 * [low_pfn, end_pfn). The range is expected to be within same pageblock.
635 * Returns zero if there is a fatal signal pending, otherwise PFN of the
636 * first page that was not scanned (which may be both less, equal to or more
639 * The pages are isolated on cc->migratepages list (not required to be empty),
640 * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field
641 * is neither read nor updated.
644 isolate_migratepages_block(struct compact_control
*cc
, unsigned long low_pfn
,
645 unsigned long end_pfn
, isolate_mode_t isolate_mode
)
647 struct zone
*zone
= cc
->zone
;
648 unsigned long nr_scanned
= 0, nr_isolated
= 0;
649 struct list_head
*migratelist
= &cc
->migratepages
;
650 struct lruvec
*lruvec
;
651 unsigned long flags
= 0;
653 struct page
*page
= NULL
, *valid_page
= NULL
;
654 unsigned long start_pfn
= low_pfn
;
657 * Ensure that there are not too many pages isolated from the LRU
658 * list by either parallel reclaimers or compaction. If there are,
659 * delay for some time until fewer pages are isolated
661 while (unlikely(too_many_isolated(zone
))) {
662 /* async migration should just abort */
663 if (cc
->mode
== MIGRATE_ASYNC
)
666 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
668 if (fatal_signal_pending(current
))
672 if (compact_should_abort(cc
))
675 /* Time to isolate some pages for migration */
676 for (; low_pfn
< end_pfn
; low_pfn
++) {
678 * Periodically drop the lock (if held) regardless of its
679 * contention, to give chance to IRQs. Abort async compaction
682 if (!(low_pfn
% SWAP_CLUSTER_MAX
)
683 && compact_unlock_should_abort(&zone
->lru_lock
, flags
,
687 if (!pfn_valid_within(low_pfn
))
691 page
= pfn_to_page(low_pfn
);
697 * Skip if free. We read page order here without zone lock
698 * which is generally unsafe, but the race window is small and
699 * the worst thing that can happen is that we skip some
700 * potential isolation targets.
702 if (PageBuddy(page
)) {
703 unsigned long freepage_order
= page_order_unsafe(page
);
706 * Without lock, we cannot be sure that what we got is
707 * a valid page order. Consider only values in the
708 * valid order range to prevent low_pfn overflow.
710 if (freepage_order
> 0 && freepage_order
< MAX_ORDER
)
711 low_pfn
+= (1UL << freepage_order
) - 1;
716 * Check may be lockless but that's ok as we recheck later.
717 * It's possible to migrate LRU pages and balloon pages
718 * Skip any other type of page
720 if (!PageLRU(page
)) {
721 if (unlikely(balloon_page_movable(page
))) {
722 if (balloon_page_isolate(page
)) {
723 /* Successfully isolated */
724 goto isolate_success
;
731 * PageLRU is set. lru_lock normally excludes isolation
732 * splitting and collapsing (collapsing has already happened
733 * if PageLRU is set) but the lock is not necessarily taken
734 * here and it is wasteful to take it just to check transhuge.
735 * Check TransHuge without lock and skip the whole pageblock if
736 * it's either a transhuge or hugetlbfs page, as calling
737 * compound_order() without preventing THP from splitting the
738 * page underneath us may return surprising results.
740 if (PageTransHuge(page
)) {
742 low_pfn
= ALIGN(low_pfn
+ 1,
743 pageblock_nr_pages
) - 1;
745 low_pfn
+= (1 << compound_order(page
)) - 1;
751 * Migration will fail if an anonymous page is pinned in memory,
752 * so avoid taking lru_lock and isolating it unnecessarily in an
753 * admittedly racy check.
755 if (!page_mapping(page
) &&
756 page_count(page
) > page_mapcount(page
))
759 /* If we already hold the lock, we can skip some rechecking */
761 locked
= compact_trylock_irqsave(&zone
->lru_lock
,
766 /* Recheck PageLRU and PageTransHuge under lock */
769 if (PageTransHuge(page
)) {
770 low_pfn
+= (1 << compound_order(page
)) - 1;
775 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
777 /* Try isolate the page */
778 if (__isolate_lru_page(page
, isolate_mode
) != 0)
781 VM_BUG_ON_PAGE(PageTransCompound(page
), page
);
783 /* Successfully isolated */
784 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
787 list_add(&page
->lru
, migratelist
);
788 cc
->nr_migratepages
++;
791 /* Avoid isolating too much */
792 if (cc
->nr_migratepages
== COMPACT_CLUSTER_MAX
) {
799 * The PageBuddy() check could have potentially brought us outside
800 * the range to be scanned.
802 if (unlikely(low_pfn
> end_pfn
))
806 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
809 * Update the pageblock-skip information and cached scanner pfn,
810 * if the whole pageblock was scanned without isolating any page.
812 if (low_pfn
== end_pfn
)
813 update_pageblock_skip(cc
, valid_page
, nr_isolated
, true);
815 trace_mm_compaction_isolate_migratepages(start_pfn
, low_pfn
,
816 nr_scanned
, nr_isolated
);
818 count_compact_events(COMPACTMIGRATE_SCANNED
, nr_scanned
);
820 count_compact_events(COMPACTISOLATED
, nr_isolated
);
826 * isolate_migratepages_range() - isolate migrate-able pages in a PFN range
827 * @cc: Compaction control structure.
828 * @start_pfn: The first PFN to start isolating.
829 * @end_pfn: The one-past-last PFN.
831 * Returns zero if isolation fails fatally due to e.g. pending signal.
832 * Otherwise, function returns one-past-the-last PFN of isolated page
833 * (which may be greater than end_pfn if end fell in a middle of a THP page).
836 isolate_migratepages_range(struct compact_control
*cc
, unsigned long start_pfn
,
837 unsigned long end_pfn
)
839 unsigned long pfn
, block_end_pfn
;
841 /* Scan block by block. First and last block may be incomplete */
843 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
845 for (; pfn
< end_pfn
; pfn
= block_end_pfn
,
846 block_end_pfn
+= pageblock_nr_pages
) {
848 block_end_pfn
= min(block_end_pfn
, end_pfn
);
850 if (!pageblock_pfn_to_page(pfn
, block_end_pfn
, cc
->zone
))
853 pfn
= isolate_migratepages_block(cc
, pfn
, block_end_pfn
,
854 ISOLATE_UNEVICTABLE
);
857 * In case of fatal failure, release everything that might
858 * have been isolated in the previous iteration, and signal
859 * the failure back to caller.
862 putback_movable_pages(&cc
->migratepages
);
863 cc
->nr_migratepages
= 0;
867 if (cc
->nr_migratepages
== COMPACT_CLUSTER_MAX
)
870 acct_isolated(cc
->zone
, cc
);
875 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
876 #ifdef CONFIG_COMPACTION
878 /* Returns true if the page is within a block suitable for migration to */
879 static bool suitable_migration_target(struct page
*page
)
881 /* If the page is a large free page, then disallow migration */
882 if (PageBuddy(page
)) {
884 * We are checking page_order without zone->lock taken. But
885 * the only small danger is that we skip a potentially suitable
886 * pageblock, so it's not worth to check order for valid range.
888 if (page_order_unsafe(page
) >= pageblock_order
)
892 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
893 if (migrate_async_suitable(get_pageblock_migratetype(page
)))
896 /* Otherwise skip the block */
901 * Test whether the free scanner has reached the same or lower pageblock than
902 * the migration scanner, and compaction should thus terminate.
904 static inline bool compact_scanners_met(struct compact_control
*cc
)
906 return (cc
->free_pfn
>> pageblock_order
)
907 <= (cc
->migrate_pfn
>> pageblock_order
);
911 * Based on information in the current compact_control, find blocks
912 * suitable for isolating free pages from and then isolate them.
914 static void isolate_freepages(struct compact_control
*cc
)
916 struct zone
*zone
= cc
->zone
;
918 unsigned long block_start_pfn
; /* start of current pageblock */
919 unsigned long isolate_start_pfn
; /* exact pfn we start at */
920 unsigned long block_end_pfn
; /* end of current pageblock */
921 unsigned long low_pfn
; /* lowest pfn scanner is able to scan */
922 struct list_head
*freelist
= &cc
->freepages
;
925 * Initialise the free scanner. The starting point is where we last
926 * successfully isolated from, zone-cached value, or the end of the
927 * zone when isolating for the first time. For looping we also need
928 * this pfn aligned down to the pageblock boundary, because we do
929 * block_start_pfn -= pageblock_nr_pages in the for loop.
930 * For ending point, take care when isolating in last pageblock of a
931 * a zone which ends in the middle of a pageblock.
932 * The low boundary is the end of the pageblock the migration scanner
935 isolate_start_pfn
= cc
->free_pfn
;
936 block_start_pfn
= cc
->free_pfn
& ~(pageblock_nr_pages
-1);
937 block_end_pfn
= min(block_start_pfn
+ pageblock_nr_pages
,
939 low_pfn
= ALIGN(cc
->migrate_pfn
+ 1, pageblock_nr_pages
);
942 * Isolate free pages until enough are available to migrate the
943 * pages on cc->migratepages. We stop searching if the migrate
944 * and free page scanners meet or enough free pages are isolated.
946 for (; block_start_pfn
>= low_pfn
&&
947 cc
->nr_migratepages
> cc
->nr_freepages
;
948 block_end_pfn
= block_start_pfn
,
949 block_start_pfn
-= pageblock_nr_pages
,
950 isolate_start_pfn
= block_start_pfn
) {
953 * This can iterate a massively long zone without finding any
954 * suitable migration targets, so periodically check if we need
955 * to schedule, or even abort async compaction.
957 if (!(block_start_pfn
% (SWAP_CLUSTER_MAX
* pageblock_nr_pages
))
958 && compact_should_abort(cc
))
961 page
= pageblock_pfn_to_page(block_start_pfn
, block_end_pfn
,
966 /* Check the block is suitable for migration */
967 if (!suitable_migration_target(page
))
970 /* If isolation recently failed, do not retry */
971 if (!isolation_suitable(cc
, page
))
974 /* Found a block suitable for isolating free pages from. */
975 isolate_freepages_block(cc
, &isolate_start_pfn
,
976 block_end_pfn
, freelist
, false);
979 * Remember where the free scanner should restart next time,
980 * which is where isolate_freepages_block() left off.
981 * But if it scanned the whole pageblock, isolate_start_pfn
982 * now points at block_end_pfn, which is the start of the next
984 * In that case we will however want to restart at the start
985 * of the previous pageblock.
987 cc
->free_pfn
= (isolate_start_pfn
< block_end_pfn
) ?
989 block_start_pfn
- pageblock_nr_pages
;
992 * isolate_freepages_block() might have aborted due to async
993 * compaction being contended
999 /* split_free_page does not map the pages */
1000 map_pages(freelist
);
1003 * If we crossed the migrate scanner, we want to keep it that way
1004 * so that compact_finished() may detect this
1006 if (block_start_pfn
< low_pfn
)
1007 cc
->free_pfn
= cc
->migrate_pfn
;
1011 * This is a migrate-callback that "allocates" freepages by taking pages
1012 * from the isolated freelists in the block we are migrating to.
1014 static struct page
*compaction_alloc(struct page
*migratepage
,
1018 struct compact_control
*cc
= (struct compact_control
*)data
;
1019 struct page
*freepage
;
1022 * Isolate free pages if necessary, and if we are not aborting due to
1025 if (list_empty(&cc
->freepages
)) {
1027 isolate_freepages(cc
);
1029 if (list_empty(&cc
->freepages
))
1033 freepage
= list_entry(cc
->freepages
.next
, struct page
, lru
);
1034 list_del(&freepage
->lru
);
1041 * This is a migrate-callback that "frees" freepages back to the isolated
1042 * freelist. All pages on the freelist are from the same zone, so there is no
1043 * special handling needed for NUMA.
1045 static void compaction_free(struct page
*page
, unsigned long data
)
1047 struct compact_control
*cc
= (struct compact_control
*)data
;
1049 list_add(&page
->lru
, &cc
->freepages
);
1053 /* possible outcome of isolate_migratepages */
1055 ISOLATE_ABORT
, /* Abort compaction now */
1056 ISOLATE_NONE
, /* No pages isolated, continue scanning */
1057 ISOLATE_SUCCESS
, /* Pages isolated, migrate */
1058 } isolate_migrate_t
;
1061 * Allow userspace to control policy on scanning the unevictable LRU for
1062 * compactable pages.
1064 int sysctl_compact_unevictable_allowed __read_mostly
= 1;
1067 * Isolate all pages that can be migrated from the first suitable block,
1068 * starting at the block pointed to by the migrate scanner pfn within
1071 static isolate_migrate_t
isolate_migratepages(struct zone
*zone
,
1072 struct compact_control
*cc
)
1074 unsigned long low_pfn
, end_pfn
;
1076 const isolate_mode_t isolate_mode
=
1077 (sysctl_compact_unevictable_allowed
? ISOLATE_UNEVICTABLE
: 0) |
1078 (cc
->mode
== MIGRATE_ASYNC
? ISOLATE_ASYNC_MIGRATE
: 0);
1081 * Start at where we last stopped, or beginning of the zone as
1082 * initialized by compact_zone()
1084 low_pfn
= cc
->migrate_pfn
;
1086 /* Only scan within a pageblock boundary */
1087 end_pfn
= ALIGN(low_pfn
+ 1, pageblock_nr_pages
);
1090 * Iterate over whole pageblocks until we find the first suitable.
1091 * Do not cross the free scanner.
1093 for (; end_pfn
<= cc
->free_pfn
;
1094 low_pfn
= end_pfn
, end_pfn
+= pageblock_nr_pages
) {
1097 * This can potentially iterate a massively long zone with
1098 * many pageblocks unsuitable, so periodically check if we
1099 * need to schedule, or even abort async compaction.
1101 if (!(low_pfn
% (SWAP_CLUSTER_MAX
* pageblock_nr_pages
))
1102 && compact_should_abort(cc
))
1105 page
= pageblock_pfn_to_page(low_pfn
, end_pfn
, zone
);
1109 /* If isolation recently failed, do not retry */
1110 if (!isolation_suitable(cc
, page
))
1114 * For async compaction, also only scan in MOVABLE blocks.
1115 * Async compaction is optimistic to see if the minimum amount
1116 * of work satisfies the allocation.
1118 if (cc
->mode
== MIGRATE_ASYNC
&&
1119 !migrate_async_suitable(get_pageblock_migratetype(page
)))
1122 /* Perform the isolation */
1123 low_pfn
= isolate_migratepages_block(cc
, low_pfn
, end_pfn
,
1126 if (!low_pfn
|| cc
->contended
) {
1127 acct_isolated(zone
, cc
);
1128 return ISOLATE_ABORT
;
1132 * Either we isolated something and proceed with migration. Or
1133 * we failed and compact_zone should decide if we should
1139 acct_isolated(zone
, cc
);
1140 /* Record where migration scanner will be restarted. */
1141 cc
->migrate_pfn
= low_pfn
;
1143 return cc
->nr_migratepages
? ISOLATE_SUCCESS
: ISOLATE_NONE
;
1146 static int __compact_finished(struct zone
*zone
, struct compact_control
*cc
,
1147 const int migratetype
)
1150 unsigned long watermark
;
1152 if (cc
->contended
|| fatal_signal_pending(current
))
1153 return COMPACT_PARTIAL
;
1155 /* Compaction run completes if the migrate and free scanner meet */
1156 if (compact_scanners_met(cc
)) {
1157 /* Let the next compaction start anew. */
1158 zone
->compact_cached_migrate_pfn
[0] = zone
->zone_start_pfn
;
1159 zone
->compact_cached_migrate_pfn
[1] = zone
->zone_start_pfn
;
1160 zone
->compact_cached_free_pfn
= zone_end_pfn(zone
);
1163 * Mark that the PG_migrate_skip information should be cleared
1164 * by kswapd when it goes to sleep. kswapd does not set the
1165 * flag itself as the decision to be clear should be directly
1166 * based on an allocation request.
1168 if (!current_is_kswapd())
1169 zone
->compact_blockskip_flush
= true;
1171 return COMPACT_COMPLETE
;
1175 * order == -1 is expected when compacting via
1176 * /proc/sys/vm/compact_memory
1178 if (cc
->order
== -1)
1179 return COMPACT_CONTINUE
;
1181 /* Compaction run is not finished if the watermark is not met */
1182 watermark
= low_wmark_pages(zone
);
1184 if (!zone_watermark_ok(zone
, cc
->order
, watermark
, cc
->classzone_idx
,
1186 return COMPACT_CONTINUE
;
1188 /* Direct compactor: Is a suitable page free? */
1189 for (order
= cc
->order
; order
< MAX_ORDER
; order
++) {
1190 struct free_area
*area
= &zone
->free_area
[order
];
1193 /* Job done if page is free of the right migratetype */
1194 if (!list_empty(&area
->free_list
[migratetype
]))
1195 return COMPACT_PARTIAL
;
1198 /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */
1199 if (migratetype
== MIGRATE_MOVABLE
&&
1200 !list_empty(&area
->free_list
[MIGRATE_CMA
]))
1201 return COMPACT_PARTIAL
;
1204 * Job done if allocation would steal freepages from
1205 * other migratetype buddy lists.
1207 if (find_suitable_fallback(area
, order
, migratetype
,
1208 true, &can_steal
) != -1)
1209 return COMPACT_PARTIAL
;
1212 return COMPACT_NO_SUITABLE_PAGE
;
1215 static int compact_finished(struct zone
*zone
, struct compact_control
*cc
,
1216 const int migratetype
)
1220 ret
= __compact_finished(zone
, cc
, migratetype
);
1221 trace_mm_compaction_finished(zone
, cc
->order
, ret
);
1222 if (ret
== COMPACT_NO_SUITABLE_PAGE
)
1223 ret
= COMPACT_CONTINUE
;
1229 * compaction_suitable: Is this suitable to run compaction on this zone now?
1231 * COMPACT_SKIPPED - If there are too few free pages for compaction
1232 * COMPACT_PARTIAL - If the allocation would succeed without compaction
1233 * COMPACT_CONTINUE - If compaction should run now
1235 static unsigned long __compaction_suitable(struct zone
*zone
, int order
,
1236 int alloc_flags
, int classzone_idx
)
1239 unsigned long watermark
;
1242 * order == -1 is expected when compacting via
1243 * /proc/sys/vm/compact_memory
1246 return COMPACT_CONTINUE
;
1248 watermark
= low_wmark_pages(zone
);
1250 * If watermarks for high-order allocation are already met, there
1251 * should be no need for compaction at all.
1253 if (zone_watermark_ok(zone
, order
, watermark
, classzone_idx
,
1255 return COMPACT_PARTIAL
;
1258 * Watermarks for order-0 must be met for compaction. Note the 2UL.
1259 * This is because during migration, copies of pages need to be
1260 * allocated and for a short time, the footprint is higher
1262 watermark
+= (2UL << order
);
1263 if (!zone_watermark_ok(zone
, 0, watermark
, classzone_idx
, alloc_flags
))
1264 return COMPACT_SKIPPED
;
1267 * fragmentation index determines if allocation failures are due to
1268 * low memory or external fragmentation
1270 * index of -1000 would imply allocations might succeed depending on
1271 * watermarks, but we already failed the high-order watermark check
1272 * index towards 0 implies failure is due to lack of memory
1273 * index towards 1000 implies failure is due to fragmentation
1275 * Only compact if a failure would be due to fragmentation.
1277 fragindex
= fragmentation_index(zone
, order
);
1278 if (fragindex
>= 0 && fragindex
<= sysctl_extfrag_threshold
)
1279 return COMPACT_NOT_SUITABLE_ZONE
;
1281 return COMPACT_CONTINUE
;
1284 unsigned long compaction_suitable(struct zone
*zone
, int order
,
1285 int alloc_flags
, int classzone_idx
)
1289 ret
= __compaction_suitable(zone
, order
, alloc_flags
, classzone_idx
);
1290 trace_mm_compaction_suitable(zone
, order
, ret
);
1291 if (ret
== COMPACT_NOT_SUITABLE_ZONE
)
1292 ret
= COMPACT_SKIPPED
;
1297 static int compact_zone(struct zone
*zone
, struct compact_control
*cc
)
1300 unsigned long start_pfn
= zone
->zone_start_pfn
;
1301 unsigned long end_pfn
= zone_end_pfn(zone
);
1302 const int migratetype
= gfpflags_to_migratetype(cc
->gfp_mask
);
1303 const bool sync
= cc
->mode
!= MIGRATE_ASYNC
;
1304 unsigned long last_migrated_pfn
= 0;
1306 ret
= compaction_suitable(zone
, cc
->order
, cc
->alloc_flags
,
1309 case COMPACT_PARTIAL
:
1310 case COMPACT_SKIPPED
:
1311 /* Compaction is likely to fail */
1313 case COMPACT_CONTINUE
:
1314 /* Fall through to compaction */
1319 * Clear pageblock skip if there were failures recently and compaction
1320 * is about to be retried after being deferred. kswapd does not do
1321 * this reset as it'll reset the cached information when going to sleep.
1323 if (compaction_restarting(zone
, cc
->order
) && !current_is_kswapd())
1324 __reset_isolation_suitable(zone
);
1327 * Setup to move all movable pages to the end of the zone. Used cached
1328 * information on where the scanners should start but check that it
1329 * is initialised by ensuring the values are within zone boundaries.
1331 cc
->migrate_pfn
= zone
->compact_cached_migrate_pfn
[sync
];
1332 cc
->free_pfn
= zone
->compact_cached_free_pfn
;
1333 if (cc
->free_pfn
< start_pfn
|| cc
->free_pfn
> end_pfn
) {
1334 cc
->free_pfn
= end_pfn
& ~(pageblock_nr_pages
-1);
1335 zone
->compact_cached_free_pfn
= cc
->free_pfn
;
1337 if (cc
->migrate_pfn
< start_pfn
|| cc
->migrate_pfn
> end_pfn
) {
1338 cc
->migrate_pfn
= start_pfn
;
1339 zone
->compact_cached_migrate_pfn
[0] = cc
->migrate_pfn
;
1340 zone
->compact_cached_migrate_pfn
[1] = cc
->migrate_pfn
;
1343 trace_mm_compaction_begin(start_pfn
, cc
->migrate_pfn
,
1344 cc
->free_pfn
, end_pfn
, sync
);
1346 migrate_prep_local();
1348 while ((ret
= compact_finished(zone
, cc
, migratetype
)) ==
1351 unsigned long isolate_start_pfn
= cc
->migrate_pfn
;
1353 switch (isolate_migratepages(zone
, cc
)) {
1355 ret
= COMPACT_PARTIAL
;
1356 putback_movable_pages(&cc
->migratepages
);
1357 cc
->nr_migratepages
= 0;
1361 * We haven't isolated and migrated anything, but
1362 * there might still be unflushed migrations from
1363 * previous cc->order aligned block.
1366 case ISOLATE_SUCCESS
:
1370 err
= migrate_pages(&cc
->migratepages
, compaction_alloc
,
1371 compaction_free
, (unsigned long)cc
, cc
->mode
,
1374 trace_mm_compaction_migratepages(cc
->nr_migratepages
, err
,
1377 /* All pages were either migrated or will be released */
1378 cc
->nr_migratepages
= 0;
1380 putback_movable_pages(&cc
->migratepages
);
1382 * migrate_pages() may return -ENOMEM when scanners meet
1383 * and we want compact_finished() to detect it
1385 if (err
== -ENOMEM
&& !compact_scanners_met(cc
)) {
1386 ret
= COMPACT_PARTIAL
;
1392 * Record where we could have freed pages by migration and not
1393 * yet flushed them to buddy allocator. We use the pfn that
1394 * isolate_migratepages() started from in this loop iteration
1395 * - this is the lowest page that could have been isolated and
1396 * then freed by migration.
1398 if (!last_migrated_pfn
)
1399 last_migrated_pfn
= isolate_start_pfn
;
1403 * Has the migration scanner moved away from the previous
1404 * cc->order aligned block where we migrated from? If yes,
1405 * flush the pages that were freed, so that they can merge and
1406 * compact_finished() can detect immediately if allocation
1409 if (cc
->order
> 0 && last_migrated_pfn
) {
1411 unsigned long current_block_start
=
1412 cc
->migrate_pfn
& ~((1UL << cc
->order
) - 1);
1414 if (last_migrated_pfn
< current_block_start
) {
1416 lru_add_drain_cpu(cpu
);
1417 drain_local_pages(zone
);
1419 /* No more flushing until we migrate again */
1420 last_migrated_pfn
= 0;
1428 * Release free pages and update where the free scanner should restart,
1429 * so we don't leave any returned pages behind in the next attempt.
1431 if (cc
->nr_freepages
> 0) {
1432 unsigned long free_pfn
= release_freepages(&cc
->freepages
);
1434 cc
->nr_freepages
= 0;
1435 VM_BUG_ON(free_pfn
== 0);
1436 /* The cached pfn is always the first in a pageblock */
1437 free_pfn
&= ~(pageblock_nr_pages
-1);
1439 * Only go back, not forward. The cached pfn might have been
1440 * already reset to zone end in compact_finished()
1442 if (free_pfn
> zone
->compact_cached_free_pfn
)
1443 zone
->compact_cached_free_pfn
= free_pfn
;
1446 trace_mm_compaction_end(start_pfn
, cc
->migrate_pfn
,
1447 cc
->free_pfn
, end_pfn
, sync
, ret
);
1452 static unsigned long compact_zone_order(struct zone
*zone
, int order
,
1453 gfp_t gfp_mask
, enum migrate_mode mode
, int *contended
,
1454 int alloc_flags
, int classzone_idx
)
1457 struct compact_control cc
= {
1459 .nr_migratepages
= 0,
1461 .gfp_mask
= gfp_mask
,
1464 .alloc_flags
= alloc_flags
,
1465 .classzone_idx
= classzone_idx
,
1467 INIT_LIST_HEAD(&cc
.freepages
);
1468 INIT_LIST_HEAD(&cc
.migratepages
);
1470 ret
= compact_zone(zone
, &cc
);
1472 VM_BUG_ON(!list_empty(&cc
.freepages
));
1473 VM_BUG_ON(!list_empty(&cc
.migratepages
));
1475 *contended
= cc
.contended
;
1479 int sysctl_extfrag_threshold
= 500;
1482 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
1483 * @gfp_mask: The GFP mask of the current allocation
1484 * @order: The order of the current allocation
1485 * @alloc_flags: The allocation flags of the current allocation
1486 * @ac: The context of current allocation
1487 * @mode: The migration mode for async, sync light, or sync migration
1488 * @contended: Return value that determines if compaction was aborted due to
1489 * need_resched() or lock contention
1491 * This is the main entry point for direct page compaction.
1493 unsigned long try_to_compact_pages(gfp_t gfp_mask
, unsigned int order
,
1494 int alloc_flags
, const struct alloc_context
*ac
,
1495 enum migrate_mode mode
, int *contended
)
1497 int may_enter_fs
= gfp_mask
& __GFP_FS
;
1498 int may_perform_io
= gfp_mask
& __GFP_IO
;
1501 int rc
= COMPACT_DEFERRED
;
1502 int all_zones_contended
= COMPACT_CONTENDED_LOCK
; /* init for &= op */
1504 *contended
= COMPACT_CONTENDED_NONE
;
1506 /* Check if the GFP flags allow compaction */
1507 if (!order
|| !may_enter_fs
|| !may_perform_io
)
1508 return COMPACT_SKIPPED
;
1510 trace_mm_compaction_try_to_compact_pages(order
, gfp_mask
, mode
);
1512 /* Compact each zone in the list */
1513 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
1518 if (compaction_deferred(zone
, order
))
1521 status
= compact_zone_order(zone
, order
, gfp_mask
, mode
,
1522 &zone_contended
, alloc_flags
,
1524 rc
= max(status
, rc
);
1526 * It takes at least one zone that wasn't lock contended
1527 * to clear all_zones_contended.
1529 all_zones_contended
&= zone_contended
;
1531 /* If a normal allocation would succeed, stop compacting */
1532 if (zone_watermark_ok(zone
, order
, low_wmark_pages(zone
),
1533 ac
->classzone_idx
, alloc_flags
)) {
1535 * We think the allocation will succeed in this zone,
1536 * but it is not certain, hence the false. The caller
1537 * will repeat this with true if allocation indeed
1538 * succeeds in this zone.
1540 compaction_defer_reset(zone
, order
, false);
1542 * It is possible that async compaction aborted due to
1543 * need_resched() and the watermarks were ok thanks to
1544 * somebody else freeing memory. The allocation can
1545 * however still fail so we better signal the
1546 * need_resched() contention anyway (this will not
1547 * prevent the allocation attempt).
1549 if (zone_contended
== COMPACT_CONTENDED_SCHED
)
1550 *contended
= COMPACT_CONTENDED_SCHED
;
1555 if (mode
!= MIGRATE_ASYNC
&& status
== COMPACT_COMPLETE
) {
1557 * We think that allocation won't succeed in this zone
1558 * so we defer compaction there. If it ends up
1559 * succeeding after all, it will be reset.
1561 defer_compaction(zone
, order
);
1565 * We might have stopped compacting due to need_resched() in
1566 * async compaction, or due to a fatal signal detected. In that
1567 * case do not try further zones and signal need_resched()
1570 if ((zone_contended
== COMPACT_CONTENDED_SCHED
)
1571 || fatal_signal_pending(current
)) {
1572 *contended
= COMPACT_CONTENDED_SCHED
;
1579 * We might not have tried all the zones, so be conservative
1580 * and assume they are not all lock contended.
1582 all_zones_contended
= 0;
1587 * If at least one zone wasn't deferred or skipped, we report if all
1588 * zones that were tried were lock contended.
1590 if (rc
> COMPACT_SKIPPED
&& all_zones_contended
)
1591 *contended
= COMPACT_CONTENDED_LOCK
;
1597 /* Compact all zones within a node */
1598 static void __compact_pgdat(pg_data_t
*pgdat
, struct compact_control
*cc
)
1603 for (zoneid
= 0; zoneid
< MAX_NR_ZONES
; zoneid
++) {
1605 zone
= &pgdat
->node_zones
[zoneid
];
1606 if (!populated_zone(zone
))
1609 cc
->nr_freepages
= 0;
1610 cc
->nr_migratepages
= 0;
1612 INIT_LIST_HEAD(&cc
->freepages
);
1613 INIT_LIST_HEAD(&cc
->migratepages
);
1616 * When called via /proc/sys/vm/compact_memory
1617 * this makes sure we compact the whole zone regardless of
1618 * cached scanner positions.
1620 if (cc
->order
== -1)
1621 __reset_isolation_suitable(zone
);
1623 if (cc
->order
== -1 || !compaction_deferred(zone
, cc
->order
))
1624 compact_zone(zone
, cc
);
1626 if (cc
->order
> 0) {
1627 if (zone_watermark_ok(zone
, cc
->order
,
1628 low_wmark_pages(zone
), 0, 0))
1629 compaction_defer_reset(zone
, cc
->order
, false);
1632 VM_BUG_ON(!list_empty(&cc
->freepages
));
1633 VM_BUG_ON(!list_empty(&cc
->migratepages
));
1637 void compact_pgdat(pg_data_t
*pgdat
, int order
)
1639 struct compact_control cc
= {
1641 .mode
= MIGRATE_ASYNC
,
1647 __compact_pgdat(pgdat
, &cc
);
1650 static void compact_node(int nid
)
1652 struct compact_control cc
= {
1654 .mode
= MIGRATE_SYNC
,
1655 .ignore_skip_hint
= true,
1658 __compact_pgdat(NODE_DATA(nid
), &cc
);
1661 /* Compact all nodes in the system */
1662 static void compact_nodes(void)
1666 /* Flush pending updates to the LRU lists */
1667 lru_add_drain_all();
1669 for_each_online_node(nid
)
1673 /* The written value is actually unused, all memory is compacted */
1674 int sysctl_compact_memory
;
1676 /* This is the entry point for compacting all nodes via /proc/sys/vm */
1677 int sysctl_compaction_handler(struct ctl_table
*table
, int write
,
1678 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
1686 int sysctl_extfrag_handler(struct ctl_table
*table
, int write
,
1687 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
1689 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
1694 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
1695 static ssize_t
sysfs_compact_node(struct device
*dev
,
1696 struct device_attribute
*attr
,
1697 const char *buf
, size_t count
)
1701 if (nid
>= 0 && nid
< nr_node_ids
&& node_online(nid
)) {
1702 /* Flush pending updates to the LRU lists */
1703 lru_add_drain_all();
1710 static DEVICE_ATTR(compact
, S_IWUSR
, NULL
, sysfs_compact_node
);
1712 int compaction_register_node(struct node
*node
)
1714 return device_create_file(&node
->dev
, &dev_attr_compact
);
1717 void compaction_unregister_node(struct node
*node
)
1719 return device_remove_file(&node
->dev
, &dev_attr_compact
);
1721 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
1723 #endif /* CONFIG_COMPACTION */