2 * linux/mm/compaction.c
4 * Memory compaction for the reduction of external fragmentation. Note that
5 * this heavily depends upon page migration to do all the real heavy
8 * Copyright IBM Corp. 2007-2010 Mel Gorman <mel@csn.ul.ie>
10 #include <linux/swap.h>
11 #include <linux/migrate.h>
12 #include <linux/compaction.h>
13 #include <linux/mm_inline.h>
14 #include <linux/backing-dev.h>
15 #include <linux/sysctl.h>
16 #include <linux/sysfs.h>
17 #include <linux/balloon_compaction.h>
18 #include <linux/page-isolation.h>
21 #ifdef CONFIG_COMPACTION
22 static inline void count_compact_event(enum vm_event_item item
)
27 static inline void count_compact_events(enum vm_event_item item
, long delta
)
29 count_vm_events(item
, delta
);
32 #define count_compact_event(item) do { } while (0)
33 #define count_compact_events(item, delta) do { } while (0)
36 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
37 #ifdef CONFIG_TRACEPOINTS
38 static const char *const compaction_status_string
[] = {
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/compaction.h>
52 static unsigned long release_freepages(struct list_head
*freelist
)
54 struct page
*page
, *next
;
55 unsigned long high_pfn
= 0;
57 list_for_each_entry_safe(page
, next
, freelist
, lru
) {
58 unsigned long pfn
= page_to_pfn(page
);
68 static void map_pages(struct list_head
*list
)
72 list_for_each_entry(page
, list
, lru
) {
73 arch_alloc_page(page
, 0);
74 kernel_map_pages(page
, 1, 1);
78 static inline bool migrate_async_suitable(int migratetype
)
80 return is_migrate_cma(migratetype
) || migratetype
== MIGRATE_MOVABLE
;
84 * Check that the whole (or subset of) a pageblock given by the interval of
85 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
86 * with the migration of free compaction scanner. The scanners then need to
87 * use only pfn_valid_within() check for arches that allow holes within
90 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
92 * It's possible on some configurations to have a setup like node0 node1 node0
93 * i.e. it's possible that all pages within a zones range of pages do not
94 * belong to a single zone. We assume that a border between node0 and node1
95 * can occur within a single pageblock, but not a node0 node1 node0
96 * interleaving within a single pageblock. It is therefore sufficient to check
97 * the first and last page of a pageblock and avoid checking each individual
98 * page in a pageblock.
100 static struct page
*pageblock_pfn_to_page(unsigned long start_pfn
,
101 unsigned long end_pfn
, struct zone
*zone
)
103 struct page
*start_page
;
104 struct page
*end_page
;
106 /* end_pfn is one past the range we are checking */
109 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
112 start_page
= pfn_to_page(start_pfn
);
114 if (page_zone(start_page
) != zone
)
117 end_page
= pfn_to_page(end_pfn
);
119 /* This gives a shorter code than deriving page_zone(end_page) */
120 if (page_zone_id(start_page
) != page_zone_id(end_page
))
126 #ifdef CONFIG_COMPACTION
128 /* Do not skip compaction more than 64 times */
129 #define COMPACT_MAX_DEFER_SHIFT 6
132 * Compaction is deferred when compaction fails to result in a page
133 * allocation success. 1 << compact_defer_limit compactions are skipped up
134 * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT
136 void defer_compaction(struct zone
*zone
, int order
)
138 zone
->compact_considered
= 0;
139 zone
->compact_defer_shift
++;
141 if (order
< zone
->compact_order_failed
)
142 zone
->compact_order_failed
= order
;
144 if (zone
->compact_defer_shift
> COMPACT_MAX_DEFER_SHIFT
)
145 zone
->compact_defer_shift
= COMPACT_MAX_DEFER_SHIFT
;
147 trace_mm_compaction_defer_compaction(zone
, order
);
150 /* Returns true if compaction should be skipped this time */
151 bool compaction_deferred(struct zone
*zone
, int order
)
153 unsigned long defer_limit
= 1UL << zone
->compact_defer_shift
;
155 if (order
< zone
->compact_order_failed
)
158 /* Avoid possible overflow */
159 if (++zone
->compact_considered
> defer_limit
)
160 zone
->compact_considered
= defer_limit
;
162 if (zone
->compact_considered
>= defer_limit
)
165 trace_mm_compaction_deferred(zone
, order
);
171 * Update defer tracking counters after successful compaction of given order,
172 * which means an allocation either succeeded (alloc_success == true) or is
173 * expected to succeed.
175 void compaction_defer_reset(struct zone
*zone
, int order
,
179 zone
->compact_considered
= 0;
180 zone
->compact_defer_shift
= 0;
182 if (order
>= zone
->compact_order_failed
)
183 zone
->compact_order_failed
= order
+ 1;
185 trace_mm_compaction_defer_reset(zone
, order
);
188 /* Returns true if restarting compaction after many failures */
189 bool compaction_restarting(struct zone
*zone
, int order
)
191 if (order
< zone
->compact_order_failed
)
194 return zone
->compact_defer_shift
== COMPACT_MAX_DEFER_SHIFT
&&
195 zone
->compact_considered
>= 1UL << zone
->compact_defer_shift
;
198 /* Returns true if the pageblock should be scanned for pages to isolate. */
199 static inline bool isolation_suitable(struct compact_control
*cc
,
202 if (cc
->ignore_skip_hint
)
205 return !get_pageblock_skip(page
);
209 * This function is called to clear all cached information on pageblocks that
210 * should be skipped for page isolation when the migrate and free page scanner
213 static void __reset_isolation_suitable(struct zone
*zone
)
215 unsigned long start_pfn
= zone
->zone_start_pfn
;
216 unsigned long end_pfn
= zone_end_pfn(zone
);
219 zone
->compact_cached_migrate_pfn
[0] = start_pfn
;
220 zone
->compact_cached_migrate_pfn
[1] = start_pfn
;
221 zone
->compact_cached_free_pfn
= end_pfn
;
222 zone
->compact_blockskip_flush
= false;
224 /* Walk the zone and mark every pageblock as suitable for isolation */
225 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
233 page
= pfn_to_page(pfn
);
234 if (zone
!= page_zone(page
))
237 clear_pageblock_skip(page
);
241 void reset_isolation_suitable(pg_data_t
*pgdat
)
245 for (zoneid
= 0; zoneid
< MAX_NR_ZONES
; zoneid
++) {
246 struct zone
*zone
= &pgdat
->node_zones
[zoneid
];
247 if (!populated_zone(zone
))
250 /* Only flush if a full compaction finished recently */
251 if (zone
->compact_blockskip_flush
)
252 __reset_isolation_suitable(zone
);
257 * If no pages were isolated then mark this pageblock to be skipped in the
258 * future. The information is later cleared by __reset_isolation_suitable().
260 static void update_pageblock_skip(struct compact_control
*cc
,
261 struct page
*page
, unsigned long nr_isolated
,
262 bool migrate_scanner
)
264 struct zone
*zone
= cc
->zone
;
267 if (cc
->ignore_skip_hint
)
276 set_pageblock_skip(page
);
278 pfn
= page_to_pfn(page
);
280 /* Update where async and sync compaction should restart */
281 if (migrate_scanner
) {
282 if (pfn
> zone
->compact_cached_migrate_pfn
[0])
283 zone
->compact_cached_migrate_pfn
[0] = pfn
;
284 if (cc
->mode
!= MIGRATE_ASYNC
&&
285 pfn
> zone
->compact_cached_migrate_pfn
[1])
286 zone
->compact_cached_migrate_pfn
[1] = pfn
;
288 if (pfn
< zone
->compact_cached_free_pfn
)
289 zone
->compact_cached_free_pfn
= pfn
;
293 static inline bool isolation_suitable(struct compact_control
*cc
,
299 static void update_pageblock_skip(struct compact_control
*cc
,
300 struct page
*page
, unsigned long nr_isolated
,
301 bool migrate_scanner
)
304 #endif /* CONFIG_COMPACTION */
307 * Compaction requires the taking of some coarse locks that are potentially
308 * very heavily contended. For async compaction, back out if the lock cannot
309 * be taken immediately. For sync compaction, spin on the lock if needed.
311 * Returns true if the lock is held
312 * Returns false if the lock is not held and compaction should abort
314 static bool compact_trylock_irqsave(spinlock_t
*lock
, unsigned long *flags
,
315 struct compact_control
*cc
)
317 if (cc
->mode
== MIGRATE_ASYNC
) {
318 if (!spin_trylock_irqsave(lock
, *flags
)) {
319 cc
->contended
= COMPACT_CONTENDED_LOCK
;
323 spin_lock_irqsave(lock
, *flags
);
330 * Compaction requires the taking of some coarse locks that are potentially
331 * very heavily contended. The lock should be periodically unlocked to avoid
332 * having disabled IRQs for a long time, even when there is nobody waiting on
333 * the lock. It might also be that allowing the IRQs will result in
334 * need_resched() becoming true. If scheduling is needed, async compaction
335 * aborts. Sync compaction schedules.
336 * Either compaction type will also abort if a fatal signal is pending.
337 * In either case if the lock was locked, it is dropped and not regained.
339 * Returns true if compaction should abort due to fatal signal pending, or
340 * async compaction due to need_resched()
341 * Returns false when compaction can continue (sync compaction might have
344 static bool compact_unlock_should_abort(spinlock_t
*lock
,
345 unsigned long flags
, bool *locked
, struct compact_control
*cc
)
348 spin_unlock_irqrestore(lock
, flags
);
352 if (fatal_signal_pending(current
)) {
353 cc
->contended
= COMPACT_CONTENDED_SCHED
;
357 if (need_resched()) {
358 if (cc
->mode
== MIGRATE_ASYNC
) {
359 cc
->contended
= COMPACT_CONTENDED_SCHED
;
369 * Aside from avoiding lock contention, compaction also periodically checks
370 * need_resched() and either schedules in sync compaction or aborts async
371 * compaction. This is similar to what compact_unlock_should_abort() does, but
372 * is used where no lock is concerned.
374 * Returns false when no scheduling was needed, or sync compaction scheduled.
375 * Returns true when async compaction should abort.
377 static inline bool compact_should_abort(struct compact_control
*cc
)
379 /* async compaction aborts if contended */
380 if (need_resched()) {
381 if (cc
->mode
== MIGRATE_ASYNC
) {
382 cc
->contended
= COMPACT_CONTENDED_SCHED
;
392 /* Returns true if the page is within a block suitable for migration to */
393 static bool suitable_migration_target(struct page
*page
)
395 /* If the page is a large free page, then disallow migration */
396 if (PageBuddy(page
)) {
398 * We are checking page_order without zone->lock taken. But
399 * the only small danger is that we skip a potentially suitable
400 * pageblock, so it's not worth to check order for valid range.
402 if (page_order_unsafe(page
) >= pageblock_order
)
406 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
407 if (migrate_async_suitable(get_pageblock_migratetype(page
)))
410 /* Otherwise skip the block */
415 * Isolate free pages onto a private freelist. If @strict is true, will abort
416 * returning 0 on any invalid PFNs or non-free pages inside of the pageblock
417 * (even though it may still end up isolating some pages).
419 static unsigned long isolate_freepages_block(struct compact_control
*cc
,
420 unsigned long *start_pfn
,
421 unsigned long end_pfn
,
422 struct list_head
*freelist
,
425 int nr_scanned
= 0, total_isolated
= 0;
426 struct page
*cursor
, *valid_page
= NULL
;
427 unsigned long flags
= 0;
429 unsigned long blockpfn
= *start_pfn
;
431 cursor
= pfn_to_page(blockpfn
);
433 /* Isolate free pages. */
434 for (; blockpfn
< end_pfn
; blockpfn
++, cursor
++) {
436 struct page
*page
= cursor
;
439 * Periodically drop the lock (if held) regardless of its
440 * contention, to give chance to IRQs. Abort if fatal signal
441 * pending or async compaction detects need_resched()
443 if (!(blockpfn
% SWAP_CLUSTER_MAX
)
444 && compact_unlock_should_abort(&cc
->zone
->lock
, flags
,
449 if (!pfn_valid_within(blockpfn
))
454 if (!PageBuddy(page
))
458 * If we already hold the lock, we can skip some rechecking.
459 * Note that if we hold the lock now, checked_pageblock was
460 * already set in some previous iteration (or strict is true),
461 * so it is correct to skip the suitable migration target
466 * The zone lock must be held to isolate freepages.
467 * Unfortunately this is a very coarse lock and can be
468 * heavily contended if there are parallel allocations
469 * or parallel compactions. For async compaction do not
470 * spin on the lock and we acquire the lock as late as
473 locked
= compact_trylock_irqsave(&cc
->zone
->lock
,
478 /* Recheck this is a buddy page under lock */
479 if (!PageBuddy(page
))
483 /* Found a free page, break it into order-0 pages */
484 isolated
= split_free_page(page
);
485 total_isolated
+= isolated
;
486 for (i
= 0; i
< isolated
; i
++) {
487 list_add(&page
->lru
, freelist
);
491 /* If a page was split, advance to the end of it */
493 cc
->nr_freepages
+= isolated
;
495 cc
->nr_migratepages
<= cc
->nr_freepages
) {
496 blockpfn
+= isolated
;
500 blockpfn
+= isolated
- 1;
501 cursor
+= isolated
- 1;
513 trace_mm_compaction_isolate_freepages(*start_pfn
, blockpfn
,
514 nr_scanned
, total_isolated
);
516 /* Record how far we have got within the block */
517 *start_pfn
= blockpfn
;
520 * If strict isolation is requested by CMA then check that all the
521 * pages requested were isolated. If there were any failures, 0 is
522 * returned and CMA will fail.
524 if (strict
&& blockpfn
< end_pfn
)
528 spin_unlock_irqrestore(&cc
->zone
->lock
, flags
);
530 /* Update the pageblock-skip if the whole pageblock was scanned */
531 if (blockpfn
== end_pfn
)
532 update_pageblock_skip(cc
, valid_page
, total_isolated
, false);
534 count_compact_events(COMPACTFREE_SCANNED
, nr_scanned
);
536 count_compact_events(COMPACTISOLATED
, total_isolated
);
537 return total_isolated
;
541 * isolate_freepages_range() - isolate free pages.
542 * @start_pfn: The first PFN to start isolating.
543 * @end_pfn: The one-past-last PFN.
545 * Non-free pages, invalid PFNs, or zone boundaries within the
546 * [start_pfn, end_pfn) range are considered errors, cause function to
547 * undo its actions and return zero.
549 * Otherwise, function returns one-past-the-last PFN of isolated page
550 * (which may be greater then end_pfn if end fell in a middle of
554 isolate_freepages_range(struct compact_control
*cc
,
555 unsigned long start_pfn
, unsigned long end_pfn
)
557 unsigned long isolated
, pfn
, block_end_pfn
;
561 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
563 for (; pfn
< end_pfn
; pfn
+= isolated
,
564 block_end_pfn
+= pageblock_nr_pages
) {
565 /* Protect pfn from changing by isolate_freepages_block */
566 unsigned long isolate_start_pfn
= pfn
;
568 block_end_pfn
= min(block_end_pfn
, end_pfn
);
571 * pfn could pass the block_end_pfn if isolated freepage
572 * is more than pageblock order. In this case, we adjust
573 * scanning range to right one.
575 if (pfn
>= block_end_pfn
) {
576 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
577 block_end_pfn
= min(block_end_pfn
, end_pfn
);
580 if (!pageblock_pfn_to_page(pfn
, block_end_pfn
, cc
->zone
))
583 isolated
= isolate_freepages_block(cc
, &isolate_start_pfn
,
584 block_end_pfn
, &freelist
, true);
587 * In strict mode, isolate_freepages_block() returns 0 if
588 * there are any holes in the block (ie. invalid PFNs or
595 * If we managed to isolate pages, it is always (1 << n) *
596 * pageblock_nr_pages for some non-negative n. (Max order
597 * page may span two pageblocks).
601 /* split_free_page does not map the pages */
602 map_pages(&freelist
);
605 /* Loop terminated early, cleanup. */
606 release_freepages(&freelist
);
610 /* We don't use freelists for anything. */
614 /* Update the number of anon and file isolated pages in the zone */
615 static void acct_isolated(struct zone
*zone
, struct compact_control
*cc
)
618 unsigned int count
[2] = { 0, };
620 if (list_empty(&cc
->migratepages
))
623 list_for_each_entry(page
, &cc
->migratepages
, lru
)
624 count
[!!page_is_file_cache(page
)]++;
626 mod_zone_page_state(zone
, NR_ISOLATED_ANON
, count
[0]);
627 mod_zone_page_state(zone
, NR_ISOLATED_FILE
, count
[1]);
630 /* Similar to reclaim, but different enough that they don't share logic */
631 static bool too_many_isolated(struct zone
*zone
)
633 unsigned long active
, inactive
, isolated
;
635 inactive
= zone_page_state(zone
, NR_INACTIVE_FILE
) +
636 zone_page_state(zone
, NR_INACTIVE_ANON
);
637 active
= zone_page_state(zone
, NR_ACTIVE_FILE
) +
638 zone_page_state(zone
, NR_ACTIVE_ANON
);
639 isolated
= zone_page_state(zone
, NR_ISOLATED_FILE
) +
640 zone_page_state(zone
, NR_ISOLATED_ANON
);
642 return isolated
> (inactive
+ active
) / 2;
646 * isolate_migratepages_block() - isolate all migrate-able pages within
648 * @cc: Compaction control structure.
649 * @low_pfn: The first PFN to isolate
650 * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock
651 * @isolate_mode: Isolation mode to be used.
653 * Isolate all pages that can be migrated from the range specified by
654 * [low_pfn, end_pfn). The range is expected to be within same pageblock.
655 * Returns zero if there is a fatal signal pending, otherwise PFN of the
656 * first page that was not scanned (which may be both less, equal to or more
659 * The pages are isolated on cc->migratepages list (not required to be empty),
660 * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field
661 * is neither read nor updated.
664 isolate_migratepages_block(struct compact_control
*cc
, unsigned long low_pfn
,
665 unsigned long end_pfn
, isolate_mode_t isolate_mode
)
667 struct zone
*zone
= cc
->zone
;
668 unsigned long nr_scanned
= 0, nr_isolated
= 0;
669 struct list_head
*migratelist
= &cc
->migratepages
;
670 struct lruvec
*lruvec
;
671 unsigned long flags
= 0;
673 struct page
*page
= NULL
, *valid_page
= NULL
;
674 unsigned long start_pfn
= low_pfn
;
677 * Ensure that there are not too many pages isolated from the LRU
678 * list by either parallel reclaimers or compaction. If there are,
679 * delay for some time until fewer pages are isolated
681 while (unlikely(too_many_isolated(zone
))) {
682 /* async migration should just abort */
683 if (cc
->mode
== MIGRATE_ASYNC
)
686 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
688 if (fatal_signal_pending(current
))
692 if (compact_should_abort(cc
))
695 /* Time to isolate some pages for migration */
696 for (; low_pfn
< end_pfn
; low_pfn
++) {
698 * Periodically drop the lock (if held) regardless of its
699 * contention, to give chance to IRQs. Abort async compaction
702 if (!(low_pfn
% SWAP_CLUSTER_MAX
)
703 && compact_unlock_should_abort(&zone
->lru_lock
, flags
,
707 if (!pfn_valid_within(low_pfn
))
711 page
= pfn_to_page(low_pfn
);
717 * Skip if free. We read page order here without zone lock
718 * which is generally unsafe, but the race window is small and
719 * the worst thing that can happen is that we skip some
720 * potential isolation targets.
722 if (PageBuddy(page
)) {
723 unsigned long freepage_order
= page_order_unsafe(page
);
726 * Without lock, we cannot be sure that what we got is
727 * a valid page order. Consider only values in the
728 * valid order range to prevent low_pfn overflow.
730 if (freepage_order
> 0 && freepage_order
< MAX_ORDER
)
731 low_pfn
+= (1UL << freepage_order
) - 1;
736 * Check may be lockless but that's ok as we recheck later.
737 * It's possible to migrate LRU pages and balloon pages
738 * Skip any other type of page
740 if (!PageLRU(page
)) {
741 if (unlikely(balloon_page_movable(page
))) {
742 if (balloon_page_isolate(page
)) {
743 /* Successfully isolated */
744 goto isolate_success
;
751 * PageLRU is set. lru_lock normally excludes isolation
752 * splitting and collapsing (collapsing has already happened
753 * if PageLRU is set) but the lock is not necessarily taken
754 * here and it is wasteful to take it just to check transhuge.
755 * Check TransHuge without lock and skip the whole pageblock if
756 * it's either a transhuge or hugetlbfs page, as calling
757 * compound_order() without preventing THP from splitting the
758 * page underneath us may return surprising results.
760 if (PageTransHuge(page
)) {
762 low_pfn
= ALIGN(low_pfn
+ 1,
763 pageblock_nr_pages
) - 1;
765 low_pfn
+= (1 << compound_order(page
)) - 1;
771 * Migration will fail if an anonymous page is pinned in memory,
772 * so avoid taking lru_lock and isolating it unnecessarily in an
773 * admittedly racy check.
775 if (!page_mapping(page
) &&
776 page_count(page
) > page_mapcount(page
))
779 /* If we already hold the lock, we can skip some rechecking */
781 locked
= compact_trylock_irqsave(&zone
->lru_lock
,
786 /* Recheck PageLRU and PageTransHuge under lock */
789 if (PageTransHuge(page
)) {
790 low_pfn
+= (1 << compound_order(page
)) - 1;
795 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
797 /* Try isolate the page */
798 if (__isolate_lru_page(page
, isolate_mode
) != 0)
801 VM_BUG_ON_PAGE(PageTransCompound(page
), page
);
803 /* Successfully isolated */
804 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
807 list_add(&page
->lru
, migratelist
);
808 cc
->nr_migratepages
++;
811 /* Avoid isolating too much */
812 if (cc
->nr_migratepages
== COMPACT_CLUSTER_MAX
) {
819 * The PageBuddy() check could have potentially brought us outside
820 * the range to be scanned.
822 if (unlikely(low_pfn
> end_pfn
))
826 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
829 * Update the pageblock-skip information and cached scanner pfn,
830 * if the whole pageblock was scanned without isolating any page.
832 if (low_pfn
== end_pfn
)
833 update_pageblock_skip(cc
, valid_page
, nr_isolated
, true);
835 trace_mm_compaction_isolate_migratepages(start_pfn
, low_pfn
,
836 nr_scanned
, nr_isolated
);
838 count_compact_events(COMPACTMIGRATE_SCANNED
, nr_scanned
);
840 count_compact_events(COMPACTISOLATED
, nr_isolated
);
846 * isolate_migratepages_range() - isolate migrate-able pages in a PFN range
847 * @cc: Compaction control structure.
848 * @start_pfn: The first PFN to start isolating.
849 * @end_pfn: The one-past-last PFN.
851 * Returns zero if isolation fails fatally due to e.g. pending signal.
852 * Otherwise, function returns one-past-the-last PFN of isolated page
853 * (which may be greater than end_pfn if end fell in a middle of a THP page).
856 isolate_migratepages_range(struct compact_control
*cc
, unsigned long start_pfn
,
857 unsigned long end_pfn
)
859 unsigned long pfn
, block_end_pfn
;
861 /* Scan block by block. First and last block may be incomplete */
863 block_end_pfn
= ALIGN(pfn
+ 1, pageblock_nr_pages
);
865 for (; pfn
< end_pfn
; pfn
= block_end_pfn
,
866 block_end_pfn
+= pageblock_nr_pages
) {
868 block_end_pfn
= min(block_end_pfn
, end_pfn
);
870 if (!pageblock_pfn_to_page(pfn
, block_end_pfn
, cc
->zone
))
873 pfn
= isolate_migratepages_block(cc
, pfn
, block_end_pfn
,
874 ISOLATE_UNEVICTABLE
);
877 * In case of fatal failure, release everything that might
878 * have been isolated in the previous iteration, and signal
879 * the failure back to caller.
882 putback_movable_pages(&cc
->migratepages
);
883 cc
->nr_migratepages
= 0;
887 if (cc
->nr_migratepages
== COMPACT_CLUSTER_MAX
)
890 acct_isolated(cc
->zone
, cc
);
895 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
896 #ifdef CONFIG_COMPACTION
898 * Based on information in the current compact_control, find blocks
899 * suitable for isolating free pages from and then isolate them.
901 static void isolate_freepages(struct compact_control
*cc
)
903 struct zone
*zone
= cc
->zone
;
905 unsigned long block_start_pfn
; /* start of current pageblock */
906 unsigned long isolate_start_pfn
; /* exact pfn we start at */
907 unsigned long block_end_pfn
; /* end of current pageblock */
908 unsigned long low_pfn
; /* lowest pfn scanner is able to scan */
909 struct list_head
*freelist
= &cc
->freepages
;
912 * Initialise the free scanner. The starting point is where we last
913 * successfully isolated from, zone-cached value, or the end of the
914 * zone when isolating for the first time. For looping we also need
915 * this pfn aligned down to the pageblock boundary, because we do
916 * block_start_pfn -= pageblock_nr_pages in the for loop.
917 * For ending point, take care when isolating in last pageblock of a
918 * a zone which ends in the middle of a pageblock.
919 * The low boundary is the end of the pageblock the migration scanner
922 isolate_start_pfn
= cc
->free_pfn
;
923 block_start_pfn
= cc
->free_pfn
& ~(pageblock_nr_pages
-1);
924 block_end_pfn
= min(block_start_pfn
+ pageblock_nr_pages
,
926 low_pfn
= ALIGN(cc
->migrate_pfn
+ 1, pageblock_nr_pages
);
929 * Isolate free pages until enough are available to migrate the
930 * pages on cc->migratepages. We stop searching if the migrate
931 * and free page scanners meet or enough free pages are isolated.
933 for (; block_start_pfn
>= low_pfn
&&
934 cc
->nr_migratepages
> cc
->nr_freepages
;
935 block_end_pfn
= block_start_pfn
,
936 block_start_pfn
-= pageblock_nr_pages
,
937 isolate_start_pfn
= block_start_pfn
) {
940 * This can iterate a massively long zone without finding any
941 * suitable migration targets, so periodically check if we need
942 * to schedule, or even abort async compaction.
944 if (!(block_start_pfn
% (SWAP_CLUSTER_MAX
* pageblock_nr_pages
))
945 && compact_should_abort(cc
))
948 page
= pageblock_pfn_to_page(block_start_pfn
, block_end_pfn
,
953 /* Check the block is suitable for migration */
954 if (!suitable_migration_target(page
))
957 /* If isolation recently failed, do not retry */
958 if (!isolation_suitable(cc
, page
))
961 /* Found a block suitable for isolating free pages from. */
962 isolate_freepages_block(cc
, &isolate_start_pfn
,
963 block_end_pfn
, freelist
, false);
966 * Remember where the free scanner should restart next time,
967 * which is where isolate_freepages_block() left off.
968 * But if it scanned the whole pageblock, isolate_start_pfn
969 * now points at block_end_pfn, which is the start of the next
971 * In that case we will however want to restart at the start
972 * of the previous pageblock.
974 cc
->free_pfn
= (isolate_start_pfn
< block_end_pfn
) ?
976 block_start_pfn
- pageblock_nr_pages
;
979 * isolate_freepages_block() might have aborted due to async
980 * compaction being contended
986 /* split_free_page does not map the pages */
990 * If we crossed the migrate scanner, we want to keep it that way
991 * so that compact_finished() may detect this
993 if (block_start_pfn
< low_pfn
)
994 cc
->free_pfn
= cc
->migrate_pfn
;
998 * This is a migrate-callback that "allocates" freepages by taking pages
999 * from the isolated freelists in the block we are migrating to.
1001 static struct page
*compaction_alloc(struct page
*migratepage
,
1005 struct compact_control
*cc
= (struct compact_control
*)data
;
1006 struct page
*freepage
;
1009 * Isolate free pages if necessary, and if we are not aborting due to
1012 if (list_empty(&cc
->freepages
)) {
1014 isolate_freepages(cc
);
1016 if (list_empty(&cc
->freepages
))
1020 freepage
= list_entry(cc
->freepages
.next
, struct page
, lru
);
1021 list_del(&freepage
->lru
);
1028 * This is a migrate-callback that "frees" freepages back to the isolated
1029 * freelist. All pages on the freelist are from the same zone, so there is no
1030 * special handling needed for NUMA.
1032 static void compaction_free(struct page
*page
, unsigned long data
)
1034 struct compact_control
*cc
= (struct compact_control
*)data
;
1036 list_add(&page
->lru
, &cc
->freepages
);
1040 /* possible outcome of isolate_migratepages */
1042 ISOLATE_ABORT
, /* Abort compaction now */
1043 ISOLATE_NONE
, /* No pages isolated, continue scanning */
1044 ISOLATE_SUCCESS
, /* Pages isolated, migrate */
1045 } isolate_migrate_t
;
1048 * Isolate all pages that can be migrated from the first suitable block,
1049 * starting at the block pointed to by the migrate scanner pfn within
1052 static isolate_migrate_t
isolate_migratepages(struct zone
*zone
,
1053 struct compact_control
*cc
)
1055 unsigned long low_pfn
, end_pfn
;
1057 const isolate_mode_t isolate_mode
=
1058 (cc
->mode
== MIGRATE_ASYNC
? ISOLATE_ASYNC_MIGRATE
: 0);
1061 * Start at where we last stopped, or beginning of the zone as
1062 * initialized by compact_zone()
1064 low_pfn
= cc
->migrate_pfn
;
1066 /* Only scan within a pageblock boundary */
1067 end_pfn
= ALIGN(low_pfn
+ 1, pageblock_nr_pages
);
1070 * Iterate over whole pageblocks until we find the first suitable.
1071 * Do not cross the free scanner.
1073 for (; end_pfn
<= cc
->free_pfn
;
1074 low_pfn
= end_pfn
, end_pfn
+= pageblock_nr_pages
) {
1077 * This can potentially iterate a massively long zone with
1078 * many pageblocks unsuitable, so periodically check if we
1079 * need to schedule, or even abort async compaction.
1081 if (!(low_pfn
% (SWAP_CLUSTER_MAX
* pageblock_nr_pages
))
1082 && compact_should_abort(cc
))
1085 page
= pageblock_pfn_to_page(low_pfn
, end_pfn
, zone
);
1089 /* If isolation recently failed, do not retry */
1090 if (!isolation_suitable(cc
, page
))
1094 * For async compaction, also only scan in MOVABLE blocks.
1095 * Async compaction is optimistic to see if the minimum amount
1096 * of work satisfies the allocation.
1098 if (cc
->mode
== MIGRATE_ASYNC
&&
1099 !migrate_async_suitable(get_pageblock_migratetype(page
)))
1102 /* Perform the isolation */
1103 low_pfn
= isolate_migratepages_block(cc
, low_pfn
, end_pfn
,
1106 if (!low_pfn
|| cc
->contended
) {
1107 acct_isolated(zone
, cc
);
1108 return ISOLATE_ABORT
;
1112 * Either we isolated something and proceed with migration. Or
1113 * we failed and compact_zone should decide if we should
1119 acct_isolated(zone
, cc
);
1121 * Record where migration scanner will be restarted. If we end up in
1122 * the same pageblock as the free scanner, make the scanners fully
1123 * meet so that compact_finished() terminates compaction.
1125 cc
->migrate_pfn
= (end_pfn
<= cc
->free_pfn
) ? low_pfn
: cc
->free_pfn
;
1127 return cc
->nr_migratepages
? ISOLATE_SUCCESS
: ISOLATE_NONE
;
1130 static int __compact_finished(struct zone
*zone
, struct compact_control
*cc
,
1131 const int migratetype
)
1134 unsigned long watermark
;
1136 if (cc
->contended
|| fatal_signal_pending(current
))
1137 return COMPACT_PARTIAL
;
1139 /* Compaction run completes if the migrate and free scanner meet */
1140 if (cc
->free_pfn
<= cc
->migrate_pfn
) {
1141 /* Let the next compaction start anew. */
1142 zone
->compact_cached_migrate_pfn
[0] = zone
->zone_start_pfn
;
1143 zone
->compact_cached_migrate_pfn
[1] = zone
->zone_start_pfn
;
1144 zone
->compact_cached_free_pfn
= zone_end_pfn(zone
);
1147 * Mark that the PG_migrate_skip information should be cleared
1148 * by kswapd when it goes to sleep. kswapd does not set the
1149 * flag itself as the decision to be clear should be directly
1150 * based on an allocation request.
1152 if (!current_is_kswapd())
1153 zone
->compact_blockskip_flush
= true;
1155 return COMPACT_COMPLETE
;
1159 * order == -1 is expected when compacting via
1160 * /proc/sys/vm/compact_memory
1162 if (cc
->order
== -1)
1163 return COMPACT_CONTINUE
;
1165 /* Compaction run is not finished if the watermark is not met */
1166 watermark
= low_wmark_pages(zone
);
1168 if (!zone_watermark_ok(zone
, cc
->order
, watermark
, cc
->classzone_idx
,
1170 return COMPACT_CONTINUE
;
1172 /* Direct compactor: Is a suitable page free? */
1173 for (order
= cc
->order
; order
< MAX_ORDER
; order
++) {
1174 struct free_area
*area
= &zone
->free_area
[order
];
1176 /* Job done if page is free of the right migratetype */
1177 if (!list_empty(&area
->free_list
[migratetype
]))
1178 return COMPACT_PARTIAL
;
1180 /* Job done if allocation would set block type */
1181 if (order
>= pageblock_order
&& area
->nr_free
)
1182 return COMPACT_PARTIAL
;
1185 return COMPACT_NO_SUITABLE_PAGE
;
1188 static int compact_finished(struct zone
*zone
, struct compact_control
*cc
,
1189 const int migratetype
)
1193 ret
= __compact_finished(zone
, cc
, migratetype
);
1194 trace_mm_compaction_finished(zone
, cc
->order
, ret
);
1195 if (ret
== COMPACT_NO_SUITABLE_PAGE
)
1196 ret
= COMPACT_CONTINUE
;
1202 * compaction_suitable: Is this suitable to run compaction on this zone now?
1204 * COMPACT_SKIPPED - If there are too few free pages for compaction
1205 * COMPACT_PARTIAL - If the allocation would succeed without compaction
1206 * COMPACT_CONTINUE - If compaction should run now
1208 static unsigned long __compaction_suitable(struct zone
*zone
, int order
,
1209 int alloc_flags
, int classzone_idx
)
1212 unsigned long watermark
;
1215 * order == -1 is expected when compacting via
1216 * /proc/sys/vm/compact_memory
1219 return COMPACT_CONTINUE
;
1221 watermark
= low_wmark_pages(zone
);
1223 * If watermarks for high-order allocation are already met, there
1224 * should be no need for compaction at all.
1226 if (zone_watermark_ok(zone
, order
, watermark
, classzone_idx
,
1228 return COMPACT_PARTIAL
;
1231 * Watermarks for order-0 must be met for compaction. Note the 2UL.
1232 * This is because during migration, copies of pages need to be
1233 * allocated and for a short time, the footprint is higher
1235 watermark
+= (2UL << order
);
1236 if (!zone_watermark_ok(zone
, 0, watermark
, classzone_idx
, alloc_flags
))
1237 return COMPACT_SKIPPED
;
1240 * fragmentation index determines if allocation failures are due to
1241 * low memory or external fragmentation
1243 * index of -1000 would imply allocations might succeed depending on
1244 * watermarks, but we already failed the high-order watermark check
1245 * index towards 0 implies failure is due to lack of memory
1246 * index towards 1000 implies failure is due to fragmentation
1248 * Only compact if a failure would be due to fragmentation.
1250 fragindex
= fragmentation_index(zone
, order
);
1251 if (fragindex
>= 0 && fragindex
<= sysctl_extfrag_threshold
)
1252 return COMPACT_NOT_SUITABLE_ZONE
;
1254 return COMPACT_CONTINUE
;
1257 unsigned long compaction_suitable(struct zone
*zone
, int order
,
1258 int alloc_flags
, int classzone_idx
)
1262 ret
= __compaction_suitable(zone
, order
, alloc_flags
, classzone_idx
);
1263 trace_mm_compaction_suitable(zone
, order
, ret
);
1264 if (ret
== COMPACT_NOT_SUITABLE_ZONE
)
1265 ret
= COMPACT_SKIPPED
;
1270 static int compact_zone(struct zone
*zone
, struct compact_control
*cc
)
1273 unsigned long start_pfn
= zone
->zone_start_pfn
;
1274 unsigned long end_pfn
= zone_end_pfn(zone
);
1275 const int migratetype
= gfpflags_to_migratetype(cc
->gfp_mask
);
1276 const bool sync
= cc
->mode
!= MIGRATE_ASYNC
;
1277 unsigned long last_migrated_pfn
= 0;
1279 ret
= compaction_suitable(zone
, cc
->order
, cc
->alloc_flags
,
1282 case COMPACT_PARTIAL
:
1283 case COMPACT_SKIPPED
:
1284 /* Compaction is likely to fail */
1286 case COMPACT_CONTINUE
:
1287 /* Fall through to compaction */
1292 * Clear pageblock skip if there were failures recently and compaction
1293 * is about to be retried after being deferred. kswapd does not do
1294 * this reset as it'll reset the cached information when going to sleep.
1296 if (compaction_restarting(zone
, cc
->order
) && !current_is_kswapd())
1297 __reset_isolation_suitable(zone
);
1300 * Setup to move all movable pages to the end of the zone. Used cached
1301 * information on where the scanners should start but check that it
1302 * is initialised by ensuring the values are within zone boundaries.
1304 cc
->migrate_pfn
= zone
->compact_cached_migrate_pfn
[sync
];
1305 cc
->free_pfn
= zone
->compact_cached_free_pfn
;
1306 if (cc
->free_pfn
< start_pfn
|| cc
->free_pfn
> end_pfn
) {
1307 cc
->free_pfn
= end_pfn
& ~(pageblock_nr_pages
-1);
1308 zone
->compact_cached_free_pfn
= cc
->free_pfn
;
1310 if (cc
->migrate_pfn
< start_pfn
|| cc
->migrate_pfn
> end_pfn
) {
1311 cc
->migrate_pfn
= start_pfn
;
1312 zone
->compact_cached_migrate_pfn
[0] = cc
->migrate_pfn
;
1313 zone
->compact_cached_migrate_pfn
[1] = cc
->migrate_pfn
;
1316 trace_mm_compaction_begin(start_pfn
, cc
->migrate_pfn
,
1317 cc
->free_pfn
, end_pfn
, sync
);
1319 migrate_prep_local();
1321 while ((ret
= compact_finished(zone
, cc
, migratetype
)) ==
1324 unsigned long isolate_start_pfn
= cc
->migrate_pfn
;
1326 switch (isolate_migratepages(zone
, cc
)) {
1328 ret
= COMPACT_PARTIAL
;
1329 putback_movable_pages(&cc
->migratepages
);
1330 cc
->nr_migratepages
= 0;
1334 * We haven't isolated and migrated anything, but
1335 * there might still be unflushed migrations from
1336 * previous cc->order aligned block.
1339 case ISOLATE_SUCCESS
:
1343 err
= migrate_pages(&cc
->migratepages
, compaction_alloc
,
1344 compaction_free
, (unsigned long)cc
, cc
->mode
,
1347 trace_mm_compaction_migratepages(cc
->nr_migratepages
, err
,
1350 /* All pages were either migrated or will be released */
1351 cc
->nr_migratepages
= 0;
1353 putback_movable_pages(&cc
->migratepages
);
1355 * migrate_pages() may return -ENOMEM when scanners meet
1356 * and we want compact_finished() to detect it
1358 if (err
== -ENOMEM
&& cc
->free_pfn
> cc
->migrate_pfn
) {
1359 ret
= COMPACT_PARTIAL
;
1365 * Record where we could have freed pages by migration and not
1366 * yet flushed them to buddy allocator. We use the pfn that
1367 * isolate_migratepages() started from in this loop iteration
1368 * - this is the lowest page that could have been isolated and
1369 * then freed by migration.
1371 if (!last_migrated_pfn
)
1372 last_migrated_pfn
= isolate_start_pfn
;
1376 * Has the migration scanner moved away from the previous
1377 * cc->order aligned block where we migrated from? If yes,
1378 * flush the pages that were freed, so that they can merge and
1379 * compact_finished() can detect immediately if allocation
1382 if (cc
->order
> 0 && last_migrated_pfn
) {
1384 unsigned long current_block_start
=
1385 cc
->migrate_pfn
& ~((1UL << cc
->order
) - 1);
1387 if (last_migrated_pfn
< current_block_start
) {
1389 lru_add_drain_cpu(cpu
);
1390 drain_local_pages(zone
);
1392 /* No more flushing until we migrate again */
1393 last_migrated_pfn
= 0;
1401 * Release free pages and update where the free scanner should restart,
1402 * so we don't leave any returned pages behind in the next attempt.
1404 if (cc
->nr_freepages
> 0) {
1405 unsigned long free_pfn
= release_freepages(&cc
->freepages
);
1407 cc
->nr_freepages
= 0;
1408 VM_BUG_ON(free_pfn
== 0);
1409 /* The cached pfn is always the first in a pageblock */
1410 free_pfn
&= ~(pageblock_nr_pages
-1);
1412 * Only go back, not forward. The cached pfn might have been
1413 * already reset to zone end in compact_finished()
1415 if (free_pfn
> zone
->compact_cached_free_pfn
)
1416 zone
->compact_cached_free_pfn
= free_pfn
;
1419 trace_mm_compaction_end(start_pfn
, cc
->migrate_pfn
,
1420 cc
->free_pfn
, end_pfn
, sync
, ret
);
1425 static unsigned long compact_zone_order(struct zone
*zone
, int order
,
1426 gfp_t gfp_mask
, enum migrate_mode mode
, int *contended
,
1427 int alloc_flags
, int classzone_idx
)
1430 struct compact_control cc
= {
1432 .nr_migratepages
= 0,
1434 .gfp_mask
= gfp_mask
,
1437 .alloc_flags
= alloc_flags
,
1438 .classzone_idx
= classzone_idx
,
1440 INIT_LIST_HEAD(&cc
.freepages
);
1441 INIT_LIST_HEAD(&cc
.migratepages
);
1443 ret
= compact_zone(zone
, &cc
);
1445 VM_BUG_ON(!list_empty(&cc
.freepages
));
1446 VM_BUG_ON(!list_empty(&cc
.migratepages
));
1448 *contended
= cc
.contended
;
1452 int sysctl_extfrag_threshold
= 500;
1455 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
1456 * @gfp_mask: The GFP mask of the current allocation
1457 * @order: The order of the current allocation
1458 * @alloc_flags: The allocation flags of the current allocation
1459 * @ac: The context of current allocation
1460 * @mode: The migration mode for async, sync light, or sync migration
1461 * @contended: Return value that determines if compaction was aborted due to
1462 * need_resched() or lock contention
1464 * This is the main entry point for direct page compaction.
1466 unsigned long try_to_compact_pages(gfp_t gfp_mask
, unsigned int order
,
1467 int alloc_flags
, const struct alloc_context
*ac
,
1468 enum migrate_mode mode
, int *contended
)
1470 int may_enter_fs
= gfp_mask
& __GFP_FS
;
1471 int may_perform_io
= gfp_mask
& __GFP_IO
;
1474 int rc
= COMPACT_DEFERRED
;
1475 int all_zones_contended
= COMPACT_CONTENDED_LOCK
; /* init for &= op */
1477 *contended
= COMPACT_CONTENDED_NONE
;
1479 /* Check if the GFP flags allow compaction */
1480 if (!order
|| !may_enter_fs
|| !may_perform_io
)
1481 return COMPACT_SKIPPED
;
1483 trace_mm_compaction_try_to_compact_pages(order
, gfp_mask
, mode
);
1485 /* Compact each zone in the list */
1486 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
1491 if (compaction_deferred(zone
, order
))
1494 status
= compact_zone_order(zone
, order
, gfp_mask
, mode
,
1495 &zone_contended
, alloc_flags
,
1497 rc
= max(status
, rc
);
1499 * It takes at least one zone that wasn't lock contended
1500 * to clear all_zones_contended.
1502 all_zones_contended
&= zone_contended
;
1504 /* If a normal allocation would succeed, stop compacting */
1505 if (zone_watermark_ok(zone
, order
, low_wmark_pages(zone
),
1506 ac
->classzone_idx
, alloc_flags
)) {
1508 * We think the allocation will succeed in this zone,
1509 * but it is not certain, hence the false. The caller
1510 * will repeat this with true if allocation indeed
1511 * succeeds in this zone.
1513 compaction_defer_reset(zone
, order
, false);
1515 * It is possible that async compaction aborted due to
1516 * need_resched() and the watermarks were ok thanks to
1517 * somebody else freeing memory. The allocation can
1518 * however still fail so we better signal the
1519 * need_resched() contention anyway (this will not
1520 * prevent the allocation attempt).
1522 if (zone_contended
== COMPACT_CONTENDED_SCHED
)
1523 *contended
= COMPACT_CONTENDED_SCHED
;
1528 if (mode
!= MIGRATE_ASYNC
&& status
== COMPACT_COMPLETE
) {
1530 * We think that allocation won't succeed in this zone
1531 * so we defer compaction there. If it ends up
1532 * succeeding after all, it will be reset.
1534 defer_compaction(zone
, order
);
1538 * We might have stopped compacting due to need_resched() in
1539 * async compaction, or due to a fatal signal detected. In that
1540 * case do not try further zones and signal need_resched()
1543 if ((zone_contended
== COMPACT_CONTENDED_SCHED
)
1544 || fatal_signal_pending(current
)) {
1545 *contended
= COMPACT_CONTENDED_SCHED
;
1552 * We might not have tried all the zones, so be conservative
1553 * and assume they are not all lock contended.
1555 all_zones_contended
= 0;
1560 * If at least one zone wasn't deferred or skipped, we report if all
1561 * zones that were tried were lock contended.
1563 if (rc
> COMPACT_SKIPPED
&& all_zones_contended
)
1564 *contended
= COMPACT_CONTENDED_LOCK
;
1570 /* Compact all zones within a node */
1571 static void __compact_pgdat(pg_data_t
*pgdat
, struct compact_control
*cc
)
1576 for (zoneid
= 0; zoneid
< MAX_NR_ZONES
; zoneid
++) {
1578 zone
= &pgdat
->node_zones
[zoneid
];
1579 if (!populated_zone(zone
))
1582 cc
->nr_freepages
= 0;
1583 cc
->nr_migratepages
= 0;
1585 INIT_LIST_HEAD(&cc
->freepages
);
1586 INIT_LIST_HEAD(&cc
->migratepages
);
1588 if (cc
->order
== -1 || !compaction_deferred(zone
, cc
->order
))
1589 compact_zone(zone
, cc
);
1591 if (cc
->order
> 0) {
1592 if (zone_watermark_ok(zone
, cc
->order
,
1593 low_wmark_pages(zone
), 0, 0))
1594 compaction_defer_reset(zone
, cc
->order
, false);
1597 VM_BUG_ON(!list_empty(&cc
->freepages
));
1598 VM_BUG_ON(!list_empty(&cc
->migratepages
));
1602 void compact_pgdat(pg_data_t
*pgdat
, int order
)
1604 struct compact_control cc
= {
1606 .mode
= MIGRATE_ASYNC
,
1612 __compact_pgdat(pgdat
, &cc
);
1615 static void compact_node(int nid
)
1617 struct compact_control cc
= {
1619 .mode
= MIGRATE_SYNC
,
1620 .ignore_skip_hint
= true,
1623 __compact_pgdat(NODE_DATA(nid
), &cc
);
1626 /* Compact all nodes in the system */
1627 static void compact_nodes(void)
1631 /* Flush pending updates to the LRU lists */
1632 lru_add_drain_all();
1634 for_each_online_node(nid
)
1638 /* The written value is actually unused, all memory is compacted */
1639 int sysctl_compact_memory
;
1641 /* This is the entry point for compacting all nodes via /proc/sys/vm */
1642 int sysctl_compaction_handler(struct ctl_table
*table
, int write
,
1643 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
1651 int sysctl_extfrag_handler(struct ctl_table
*table
, int write
,
1652 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
1654 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
1659 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
1660 static ssize_t
sysfs_compact_node(struct device
*dev
,
1661 struct device_attribute
*attr
,
1662 const char *buf
, size_t count
)
1666 if (nid
>= 0 && nid
< nr_node_ids
&& node_online(nid
)) {
1667 /* Flush pending updates to the LRU lists */
1668 lru_add_drain_all();
1675 static DEVICE_ATTR(compact
, S_IWUSR
, NULL
, sysfs_compact_node
);
1677 int compaction_register_node(struct node
*node
)
1679 return device_create_file(&node
->dev
, &dev_attr_compact
);
1682 void compaction_unregister_node(struct node
*node
)
1684 return device_remove_file(&node
->dev
, &dev_attr_compact
);
1686 #endif /* CONFIG_SYSFS && CONFIG_NUMA */
1688 #endif /* CONFIG_COMPACTION */