blob: c5c627aae9962daf9c64d4f482c075e4cd96422a [file] [log] [blame]
Mel Gorman748446b2010-05-24 14:32:27 -07001/*
2 * linux/mm/compaction.c
3 *
4 * Memory compaction for the reduction of external fragmentation. Note that
5 * this heavily depends upon page migration to do all the real heavy
6 * lifting
7 *
8 * Copyright IBM Corp. 2007-2010 Mel Gorman <mel@csn.ul.ie>
9 */
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>
Mel Gorman76ab0f52010-05-24 14:32:28 -070015#include <linux/sysctl.h>
Mel Gormaned4a6d72010-05-24 14:32:29 -070016#include <linux/sysfs.h>
Rafael Aquinibf6bddf2012-12-11 16:02:42 -080017#include <linux/balloon_compaction.h>
Minchan Kim194159f2013-02-22 16:33:58 -080018#include <linux/page-isolation.h>
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -080019#include <linux/kasan.h>
Mel Gorman748446b2010-05-24 14:32:27 -070020#include "internal.h"
21
Minchan Kim010fc292012-12-20 15:05:06 -080022#ifdef CONFIG_COMPACTION
23static inline void count_compact_event(enum vm_event_item item)
24{
25 count_vm_event(item);
26}
27
28static inline void count_compact_events(enum vm_event_item item, long delta)
29{
30 count_vm_events(item, delta);
31}
32#else
33#define count_compact_event(item) do { } while (0)
34#define count_compact_events(item, delta) do { } while (0)
35#endif
36
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010037#if defined CONFIG_COMPACTION || defined CONFIG_CMA
Joonsoo Kim16c4a092015-02-11 15:27:01 -080038#ifdef CONFIG_TRACEPOINTS
39static const char *const compaction_status_string[] = {
40 "deferred",
41 "skipped",
42 "continue",
43 "partial",
44 "complete",
Joonsoo Kim837d0262015-02-11 15:27:06 -080045 "no_suitable_page",
46 "not_suitable_zone",
Joonsoo Kim16c4a092015-02-11 15:27:01 -080047};
48#endif
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010049
Mel Gormanb7aba692011-01-13 15:45:54 -080050#define CREATE_TRACE_POINTS
51#include <trace/events/compaction.h>
52
Mel Gorman748446b2010-05-24 14:32:27 -070053static unsigned long release_freepages(struct list_head *freelist)
54{
55 struct page *page, *next;
Vlastimil Babka6bace092014-12-10 15:43:31 -080056 unsigned long high_pfn = 0;
Mel Gorman748446b2010-05-24 14:32:27 -070057
58 list_for_each_entry_safe(page, next, freelist, lru) {
Vlastimil Babka6bace092014-12-10 15:43:31 -080059 unsigned long pfn = page_to_pfn(page);
Mel Gorman748446b2010-05-24 14:32:27 -070060 list_del(&page->lru);
61 __free_page(page);
Vlastimil Babka6bace092014-12-10 15:43:31 -080062 if (pfn > high_pfn)
63 high_pfn = pfn;
Mel Gorman748446b2010-05-24 14:32:27 -070064 }
65
Vlastimil Babka6bace092014-12-10 15:43:31 -080066 return high_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -070067}
68
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010069static void map_pages(struct list_head *list)
70{
71 struct page *page;
72
73 list_for_each_entry(page, list, lru) {
74 arch_alloc_page(page, 0);
75 kernel_map_pages(page, 1, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -080076 kasan_alloc_pages(page, 0);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010077 }
78}
79
Michal Nazarewicz47118af2011-12-29 13:09:50 +010080static inline bool migrate_async_suitable(int migratetype)
81{
82 return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
83}
84
Vlastimil Babka7d49d882014-10-09 15:27:11 -070085/*
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
90 * pageblocks.
91 *
92 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
93 *
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.
101 */
102static struct page *pageblock_pfn_to_page(unsigned long start_pfn,
103 unsigned long end_pfn, struct zone *zone)
104{
105 struct page *start_page;
106 struct page *end_page;
107
108 /* end_pfn is one past the range we are checking */
109 end_pfn--;
110
111 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
112 return NULL;
113
114 start_page = pfn_to_page(start_pfn);
115
116 if (page_zone(start_page) != zone)
117 return NULL;
118
119 end_page = pfn_to_page(end_pfn);
120
121 /* This gives a shorter code than deriving page_zone(end_page) */
122 if (page_zone_id(start_page) != page_zone_id(end_page))
123 return NULL;
124
125 return start_page;
126}
127
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700128#ifdef CONFIG_COMPACTION
Joonsoo Kim24e27162015-02-11 15:27:09 -0800129
130/* Do not skip compaction more than 64 times */
131#define COMPACT_MAX_DEFER_SHIFT 6
132
133/*
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
137 */
138void defer_compaction(struct zone *zone, int order)
139{
140 zone->compact_considered = 0;
141 zone->compact_defer_shift++;
142
143 if (order < zone->compact_order_failed)
144 zone->compact_order_failed = order;
145
146 if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT)
147 zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT;
148
149 trace_mm_compaction_defer_compaction(zone, order);
150}
151
152/* Returns true if compaction should be skipped this time */
153bool compaction_deferred(struct zone *zone, int order)
154{
155 unsigned long defer_limit = 1UL << zone->compact_defer_shift;
156
157 if (order < zone->compact_order_failed)
158 return false;
159
160 /* Avoid possible overflow */
161 if (++zone->compact_considered > defer_limit)
162 zone->compact_considered = defer_limit;
163
164 if (zone->compact_considered >= defer_limit)
165 return false;
166
167 trace_mm_compaction_deferred(zone, order);
168
169 return true;
170}
171
172/*
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.
176 */
177void compaction_defer_reset(struct zone *zone, int order,
178 bool alloc_success)
179{
180 if (alloc_success) {
181 zone->compact_considered = 0;
182 zone->compact_defer_shift = 0;
183 }
184 if (order >= zone->compact_order_failed)
185 zone->compact_order_failed = order + 1;
186
187 trace_mm_compaction_defer_reset(zone, order);
188}
189
190/* Returns true if restarting compaction after many failures */
191bool compaction_restarting(struct zone *zone, int order)
192{
193 if (order < zone->compact_order_failed)
194 return false;
195
196 return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT &&
197 zone->compact_considered >= 1UL << zone->compact_defer_shift;
198}
199
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700200/* Returns true if the pageblock should be scanned for pages to isolate. */
201static inline bool isolation_suitable(struct compact_control *cc,
202 struct page *page)
203{
204 if (cc->ignore_skip_hint)
205 return true;
206
207 return !get_pageblock_skip(page);
208}
209
Vlastimil Babka02333642015-09-08 15:02:42 -0700210static void reset_cached_positions(struct zone *zone)
211{
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);
215}
216
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700217/*
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
220 * meet.
221 */
Mel Gorman62997022012-10-08 16:32:47 -0700222static void __reset_isolation_suitable(struct zone *zone)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700223{
224 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800225 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700226 unsigned long pfn;
227
Mel Gorman62997022012-10-08 16:32:47 -0700228 zone->compact_blockskip_flush = false;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700229
230 /* Walk the zone and mark every pageblock as suitable for isolation */
231 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
232 struct page *page;
233
234 cond_resched();
235
236 if (!pfn_valid(pfn))
237 continue;
238
239 page = pfn_to_page(pfn);
240 if (zone != page_zone(page))
241 continue;
242
243 clear_pageblock_skip(page);
244 }
Vlastimil Babka02333642015-09-08 15:02:42 -0700245
246 reset_cached_positions(zone);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700247}
248
Mel Gorman62997022012-10-08 16:32:47 -0700249void reset_isolation_suitable(pg_data_t *pgdat)
250{
251 int zoneid;
252
253 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
254 struct zone *zone = &pgdat->node_zones[zoneid];
255 if (!populated_zone(zone))
256 continue;
257
258 /* Only flush if a full compaction finished recently */
259 if (zone->compact_blockskip_flush)
260 __reset_isolation_suitable(zone);
261 }
262}
263
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700264/*
265 * If no pages were isolated then mark this pageblock to be skipped in the
Mel Gorman62997022012-10-08 16:32:47 -0700266 * future. The information is later cleared by __reset_isolation_suitable().
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700267 */
Mel Gormanc89511a2012-10-08 16:32:45 -0700268static void update_pageblock_skip(struct compact_control *cc,
269 struct page *page, unsigned long nr_isolated,
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700270 bool migrate_scanner)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700271{
Mel Gormanc89511a2012-10-08 16:32:45 -0700272 struct zone *zone = cc->zone;
David Rientjes35979ef2014-06-04 16:08:27 -0700273 unsigned long pfn;
Joonsoo Kim6815bf32013-12-18 17:08:52 -0800274
275 if (cc->ignore_skip_hint)
276 return;
277
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700278 if (!page)
279 return;
280
David Rientjes35979ef2014-06-04 16:08:27 -0700281 if (nr_isolated)
282 return;
283
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700284 set_pageblock_skip(page);
Mel Gormanc89511a2012-10-08 16:32:45 -0700285
David Rientjes35979ef2014-06-04 16:08:27 -0700286 pfn = page_to_pfn(page);
287
288 /* Update where async and sync compaction should restart */
289 if (migrate_scanner) {
David Rientjes35979ef2014-06-04 16:08:27 -0700290 if (pfn > zone->compact_cached_migrate_pfn[0])
291 zone->compact_cached_migrate_pfn[0] = pfn;
David Rientjese0b9dae2014-06-04 16:08:28 -0700292 if (cc->mode != MIGRATE_ASYNC &&
293 pfn > zone->compact_cached_migrate_pfn[1])
David Rientjes35979ef2014-06-04 16:08:27 -0700294 zone->compact_cached_migrate_pfn[1] = pfn;
295 } else {
David Rientjes35979ef2014-06-04 16:08:27 -0700296 if (pfn < zone->compact_cached_free_pfn)
297 zone->compact_cached_free_pfn = pfn;
Mel Gormanc89511a2012-10-08 16:32:45 -0700298 }
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700299}
300#else
301static inline bool isolation_suitable(struct compact_control *cc,
302 struct page *page)
303{
304 return true;
305}
306
Mel Gormanc89511a2012-10-08 16:32:45 -0700307static void update_pageblock_skip(struct compact_control *cc,
308 struct page *page, unsigned long nr_isolated,
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700309 bool migrate_scanner)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700310{
311}
312#endif /* CONFIG_COMPACTION */
313
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700314/*
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.
318 *
319 * Returns true if the lock is held
320 * Returns false if the lock is not held and compaction should abort
321 */
322static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags,
323 struct compact_control *cc)
Mel Gorman2a1402a2012-10-08 16:32:33 -0700324{
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700325 if (cc->mode == MIGRATE_ASYNC) {
326 if (!spin_trylock_irqsave(lock, *flags)) {
327 cc->contended = COMPACT_CONTENDED_LOCK;
328 return false;
329 }
330 } else {
331 spin_lock_irqsave(lock, *flags);
332 }
Vlastimil Babka1f9efde2014-10-09 15:27:14 -0700333
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700334 return true;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700335}
336
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100337/*
Mel Gormanc67fe372012-08-21 16:16:17 -0700338 * Compaction requires the taking of some coarse locks that are potentially
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700339 * 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.
Mel Gormanc67fe372012-08-21 16:16:17 -0700346 *
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700347 * 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
350 * scheduled)
Mel Gormanc67fe372012-08-21 16:16:17 -0700351 */
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700352static bool compact_unlock_should_abort(spinlock_t *lock,
353 unsigned long flags, bool *locked, struct compact_control *cc)
Mel Gormanc67fe372012-08-21 16:16:17 -0700354{
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700355 if (*locked) {
356 spin_unlock_irqrestore(lock, flags);
357 *locked = false;
358 }
Vlastimil Babka1f9efde2014-10-09 15:27:14 -0700359
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700360 if (fatal_signal_pending(current)) {
361 cc->contended = COMPACT_CONTENDED_SCHED;
362 return true;
363 }
Mel Gormanc67fe372012-08-21 16:16:17 -0700364
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700365 if (need_resched()) {
David Rientjese0b9dae2014-06-04 16:08:28 -0700366 if (cc->mode == MIGRATE_ASYNC) {
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700367 cc->contended = COMPACT_CONTENDED_SCHED;
368 return true;
Mel Gormanc67fe372012-08-21 16:16:17 -0700369 }
Mel Gormanc67fe372012-08-21 16:16:17 -0700370 cond_resched();
Mel Gormanc67fe372012-08-21 16:16:17 -0700371 }
372
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700373 return false;
Mel Gormanc67fe372012-08-21 16:16:17 -0700374}
375
Vlastimil Babkabe976572014-06-04 16:10:41 -0700376/*
377 * Aside from avoiding lock contention, compaction also periodically checks
378 * need_resched() and either schedules in sync compaction or aborts async
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700379 * compaction. This is similar to what compact_unlock_should_abort() does, but
Vlastimil Babkabe976572014-06-04 16:10:41 -0700380 * is used where no lock is concerned.
381 *
382 * Returns false when no scheduling was needed, or sync compaction scheduled.
383 * Returns true when async compaction should abort.
384 */
385static inline bool compact_should_abort(struct compact_control *cc)
386{
387 /* async compaction aborts if contended */
388 if (need_resched()) {
389 if (cc->mode == MIGRATE_ASYNC) {
Vlastimil Babka1f9efde2014-10-09 15:27:14 -0700390 cc->contended = COMPACT_CONTENDED_SCHED;
Vlastimil Babkabe976572014-06-04 16:10:41 -0700391 return true;
392 }
393
394 cond_resched();
395 }
396
397 return false;
398}
399
Mel Gormanc67fe372012-08-21 16:16:17 -0700400/*
Jerome Marchand9e4be472013-11-12 15:07:12 -0800401 * 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).
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100404 */
Mel Gormanf40d1e42012-10-08 16:32:36 -0700405static unsigned long isolate_freepages_block(struct compact_control *cc,
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700406 unsigned long *start_pfn,
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100407 unsigned long end_pfn,
408 struct list_head *freelist,
409 bool strict)
Mel Gorman748446b2010-05-24 14:32:27 -0700410{
Mel Gormanb7aba692011-01-13 15:45:54 -0800411 int nr_scanned = 0, total_isolated = 0;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700412 struct page *cursor, *valid_page = NULL;
Xiubo Lib8b2d822014-10-09 15:28:21 -0700413 unsigned long flags = 0;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700414 bool locked = false;
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700415 unsigned long blockpfn = *start_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -0700416
Mel Gorman748446b2010-05-24 14:32:27 -0700417 cursor = pfn_to_page(blockpfn);
418
Mel Gormanf40d1e42012-10-08 16:32:36 -0700419 /* Isolate free pages. */
Mel Gorman748446b2010-05-24 14:32:27 -0700420 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
421 int isolated, i;
422 struct page *page = cursor;
423
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700424 /*
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()
428 */
429 if (!(blockpfn % SWAP_CLUSTER_MAX)
430 && compact_unlock_should_abort(&cc->zone->lock, flags,
431 &locked, cc))
432 break;
433
Mel Gormanb7aba692011-01-13 15:45:54 -0800434 nr_scanned++;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700435 if (!pfn_valid_within(blockpfn))
Laura Abbott2af120b2014-03-10 15:49:44 -0700436 goto isolate_fail;
437
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700438 if (!valid_page)
439 valid_page = page;
Vlastimil Babka9fcd6d22015-09-08 15:02:49 -0700440
441 /*
442 * For compound pages such as THP and hugetlbfs, we can save
443 * potentially a lot of iterations if we skip them at once.
444 * The check is racy, but we can consider only valid values
445 * and the only danger is skipping too much.
446 */
447 if (PageCompound(page)) {
448 unsigned int comp_order = compound_order(page);
449
450 if (likely(comp_order < MAX_ORDER)) {
451 blockpfn += (1UL << comp_order) - 1;
452 cursor += (1UL << comp_order) - 1;
453 }
454
455 goto isolate_fail;
456 }
457
Mel Gormanf40d1e42012-10-08 16:32:36 -0700458 if (!PageBuddy(page))
Laura Abbott2af120b2014-03-10 15:49:44 -0700459 goto isolate_fail;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700460
461 /*
Vlastimil Babka69b71892014-10-09 15:27:18 -0700462 * If we already hold the lock, we can skip some rechecking.
463 * Note that if we hold the lock now, checked_pageblock was
464 * already set in some previous iteration (or strict is true),
465 * so it is correct to skip the suitable migration target
466 * recheck as well.
Mel Gormanf40d1e42012-10-08 16:32:36 -0700467 */
Vlastimil Babka69b71892014-10-09 15:27:18 -0700468 if (!locked) {
469 /*
470 * The zone lock must be held to isolate freepages.
471 * Unfortunately this is a very coarse lock and can be
472 * heavily contended if there are parallel allocations
473 * or parallel compactions. For async compaction do not
474 * spin on the lock and we acquire the lock as late as
475 * possible.
476 */
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700477 locked = compact_trylock_irqsave(&cc->zone->lock,
478 &flags, cc);
Vlastimil Babka69b71892014-10-09 15:27:18 -0700479 if (!locked)
480 break;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700481
Vlastimil Babka69b71892014-10-09 15:27:18 -0700482 /* Recheck this is a buddy page under lock */
483 if (!PageBuddy(page))
484 goto isolate_fail;
485 }
Mel Gorman748446b2010-05-24 14:32:27 -0700486
487 /* Found a free page, break it into order-0 pages */
488 isolated = split_free_page(page);
489 total_isolated += isolated;
490 for (i = 0; i < isolated; i++) {
491 list_add(&page->lru, freelist);
492 page++;
493 }
494
495 /* If a page was split, advance to the end of it */
496 if (isolated) {
Joonsoo Kim932ff6b2015-02-12 14:59:53 -0800497 cc->nr_freepages += isolated;
498 if (!strict &&
499 cc->nr_migratepages <= cc->nr_freepages) {
500 blockpfn += isolated;
501 break;
502 }
503
Mel Gorman748446b2010-05-24 14:32:27 -0700504 blockpfn += isolated - 1;
505 cursor += isolated - 1;
Laura Abbott2af120b2014-03-10 15:49:44 -0700506 continue;
Mel Gorman748446b2010-05-24 14:32:27 -0700507 }
Laura Abbott2af120b2014-03-10 15:49:44 -0700508
509isolate_fail:
510 if (strict)
511 break;
512 else
513 continue;
514
Mel Gorman748446b2010-05-24 14:32:27 -0700515 }
516
Vlastimil Babka9fcd6d22015-09-08 15:02:49 -0700517 /*
518 * There is a tiny chance that we have read bogus compound_order(),
519 * so be careful to not go outside of the pageblock.
520 */
521 if (unlikely(blockpfn > end_pfn))
522 blockpfn = end_pfn;
523
Joonsoo Kime34d85f2015-02-11 15:27:04 -0800524 trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn,
525 nr_scanned, total_isolated);
526
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700527 /* Record how far we have got within the block */
528 *start_pfn = blockpfn;
529
Mel Gormanf40d1e42012-10-08 16:32:36 -0700530 /*
531 * If strict isolation is requested by CMA then check that all the
532 * pages requested were isolated. If there were any failures, 0 is
533 * returned and CMA will fail.
534 */
Laura Abbott2af120b2014-03-10 15:49:44 -0700535 if (strict && blockpfn < end_pfn)
Mel Gormanf40d1e42012-10-08 16:32:36 -0700536 total_isolated = 0;
537
538 if (locked)
539 spin_unlock_irqrestore(&cc->zone->lock, flags);
540
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700541 /* Update the pageblock-skip if the whole pageblock was scanned */
542 if (blockpfn == end_pfn)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700543 update_pageblock_skip(cc, valid_page, total_isolated, false);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700544
Minchan Kim010fc292012-12-20 15:05:06 -0800545 count_compact_events(COMPACTFREE_SCANNED, nr_scanned);
Mel Gorman397487d2012-10-19 12:00:10 +0100546 if (total_isolated)
Minchan Kim010fc292012-12-20 15:05:06 -0800547 count_compact_events(COMPACTISOLATED, total_isolated);
Mel Gorman748446b2010-05-24 14:32:27 -0700548 return total_isolated;
549}
550
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100551/**
552 * isolate_freepages_range() - isolate free pages.
553 * @start_pfn: The first PFN to start isolating.
554 * @end_pfn: The one-past-last PFN.
555 *
556 * Non-free pages, invalid PFNs, or zone boundaries within the
557 * [start_pfn, end_pfn) range are considered errors, cause function to
558 * undo its actions and return zero.
559 *
560 * Otherwise, function returns one-past-the-last PFN of isolated page
561 * (which may be greater then end_pfn if end fell in a middle of
562 * a free page).
563 */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100564unsigned long
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700565isolate_freepages_range(struct compact_control *cc,
566 unsigned long start_pfn, unsigned long end_pfn)
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100567{
Mel Gormanf40d1e42012-10-08 16:32:36 -0700568 unsigned long isolated, pfn, block_end_pfn;
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100569 LIST_HEAD(freelist);
570
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700571 pfn = start_pfn;
572 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100573
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700574 for (; pfn < end_pfn; pfn += isolated,
575 block_end_pfn += pageblock_nr_pages) {
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700576 /* Protect pfn from changing by isolate_freepages_block */
577 unsigned long isolate_start_pfn = pfn;
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700578
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100579 block_end_pfn = min(block_end_pfn, end_pfn);
580
Joonsoo Kim58420012014-11-13 15:19:07 -0800581 /*
582 * pfn could pass the block_end_pfn if isolated freepage
583 * is more than pageblock order. In this case, we adjust
584 * scanning range to right one.
585 */
586 if (pfn >= block_end_pfn) {
587 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
588 block_end_pfn = min(block_end_pfn, end_pfn);
589 }
590
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700591 if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone))
592 break;
593
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700594 isolated = isolate_freepages_block(cc, &isolate_start_pfn,
595 block_end_pfn, &freelist, true);
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100596
597 /*
598 * In strict mode, isolate_freepages_block() returns 0 if
599 * there are any holes in the block (ie. invalid PFNs or
600 * non-free pages).
601 */
602 if (!isolated)
603 break;
604
605 /*
606 * If we managed to isolate pages, it is always (1 << n) *
607 * pageblock_nr_pages for some non-negative n. (Max order
608 * page may span two pageblocks).
609 */
610 }
611
612 /* split_free_page does not map the pages */
613 map_pages(&freelist);
614
615 if (pfn < end_pfn) {
616 /* Loop terminated early, cleanup. */
617 release_freepages(&freelist);
618 return 0;
619 }
620
621 /* We don't use freelists for anything. */
622 return pfn;
623}
624
Mel Gorman748446b2010-05-24 14:32:27 -0700625/* Update the number of anon and file isolated pages in the zone */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700626static void acct_isolated(struct zone *zone, struct compact_control *cc)
Mel Gorman748446b2010-05-24 14:32:27 -0700627{
628 struct page *page;
Minchan Kimb9e84ac2011-10-31 17:06:44 -0700629 unsigned int count[2] = { 0, };
Mel Gorman748446b2010-05-24 14:32:27 -0700630
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700631 if (list_empty(&cc->migratepages))
632 return;
633
Minchan Kimb9e84ac2011-10-31 17:06:44 -0700634 list_for_each_entry(page, &cc->migratepages, lru)
635 count[!!page_is_file_cache(page)]++;
Mel Gorman748446b2010-05-24 14:32:27 -0700636
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700637 mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
638 mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
Mel Gorman748446b2010-05-24 14:32:27 -0700639}
640
641/* Similar to reclaim, but different enough that they don't share logic */
642static bool too_many_isolated(struct zone *zone)
643{
Minchan Kimbc693042010-09-09 16:38:00 -0700644 unsigned long active, inactive, isolated;
Mel Gorman748446b2010-05-24 14:32:27 -0700645
646 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
647 zone_page_state(zone, NR_INACTIVE_ANON);
Minchan Kimbc693042010-09-09 16:38:00 -0700648 active = zone_page_state(zone, NR_ACTIVE_FILE) +
649 zone_page_state(zone, NR_ACTIVE_ANON);
Mel Gorman748446b2010-05-24 14:32:27 -0700650 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
651 zone_page_state(zone, NR_ISOLATED_ANON);
652
Minchan Kimbc693042010-09-09 16:38:00 -0700653 return isolated > (inactive + active) / 2;
Mel Gorman748446b2010-05-24 14:32:27 -0700654}
655
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100656/**
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700657 * isolate_migratepages_block() - isolate all migrate-able pages within
658 * a single pageblock
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100659 * @cc: Compaction control structure.
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700660 * @low_pfn: The first PFN to isolate
661 * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock
662 * @isolate_mode: Isolation mode to be used.
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100663 *
664 * Isolate all pages that can be migrated from the range specified by
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700665 * [low_pfn, end_pfn). The range is expected to be within same pageblock.
666 * Returns zero if there is a fatal signal pending, otherwise PFN of the
667 * first page that was not scanned (which may be both less, equal to or more
668 * than end_pfn).
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100669 *
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700670 * The pages are isolated on cc->migratepages list (not required to be empty),
671 * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field
672 * is neither read nor updated.
Mel Gorman748446b2010-05-24 14:32:27 -0700673 */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700674static unsigned long
675isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
676 unsigned long end_pfn, isolate_mode_t isolate_mode)
Mel Gorman748446b2010-05-24 14:32:27 -0700677{
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700678 struct zone *zone = cc->zone;
Mel Gormanb7aba692011-01-13 15:45:54 -0800679 unsigned long nr_scanned = 0, nr_isolated = 0;
Mel Gorman748446b2010-05-24 14:32:27 -0700680 struct list_head *migratelist = &cc->migratepages;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700681 struct lruvec *lruvec;
Xiubo Lib8b2d822014-10-09 15:28:21 -0700682 unsigned long flags = 0;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700683 bool locked = false;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700684 struct page *page = NULL, *valid_page = NULL;
Joonsoo Kime34d85f2015-02-11 15:27:04 -0800685 unsigned long start_pfn = low_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -0700686
Mel Gorman748446b2010-05-24 14:32:27 -0700687 /*
688 * Ensure that there are not too many pages isolated from the LRU
689 * list by either parallel reclaimers or compaction. If there are,
690 * delay for some time until fewer pages are isolated
691 */
692 while (unlikely(too_many_isolated(zone))) {
Mel Gormanf9e35b32011-06-15 15:08:52 -0700693 /* async migration should just abort */
David Rientjese0b9dae2014-06-04 16:08:28 -0700694 if (cc->mode == MIGRATE_ASYNC)
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100695 return 0;
Mel Gormanf9e35b32011-06-15 15:08:52 -0700696
Mel Gorman748446b2010-05-24 14:32:27 -0700697 congestion_wait(BLK_RW_ASYNC, HZ/10);
698
699 if (fatal_signal_pending(current))
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100700 return 0;
Mel Gorman748446b2010-05-24 14:32:27 -0700701 }
702
Vlastimil Babkabe976572014-06-04 16:10:41 -0700703 if (compact_should_abort(cc))
704 return 0;
David Rientjesaeef4b82014-06-04 16:08:31 -0700705
Mel Gorman748446b2010-05-24 14:32:27 -0700706 /* Time to isolate some pages for migration */
Mel Gorman748446b2010-05-24 14:32:27 -0700707 for (; low_pfn < end_pfn; low_pfn++) {
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700708 bool is_lru;
709
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700710 /*
711 * Periodically drop the lock (if held) regardless of its
712 * contention, to give chance to IRQs. Abort async compaction
713 * if contended.
714 */
715 if (!(low_pfn % SWAP_CLUSTER_MAX)
716 && compact_unlock_should_abort(&zone->lru_lock, flags,
717 &locked, cc))
718 break;
Mel Gormanc67fe372012-08-21 16:16:17 -0700719
Mel Gorman748446b2010-05-24 14:32:27 -0700720 if (!pfn_valid_within(low_pfn))
721 continue;
Mel Gormanb7aba692011-01-13 15:45:54 -0800722 nr_scanned++;
Mel Gorman748446b2010-05-24 14:32:27 -0700723
Mel Gorman748446b2010-05-24 14:32:27 -0700724 page = pfn_to_page(low_pfn);
Mel Gormandc908602012-02-08 17:13:38 -0800725
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700726 if (!valid_page)
727 valid_page = page;
728
Mel Gorman6c144662014-01-23 15:53:38 -0800729 /*
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700730 * Skip if free. We read page order here without zone lock
731 * which is generally unsafe, but the race window is small and
732 * the worst thing that can happen is that we skip some
733 * potential isolation targets.
Mel Gorman6c144662014-01-23 15:53:38 -0800734 */
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700735 if (PageBuddy(page)) {
736 unsigned long freepage_order = page_order_unsafe(page);
737
738 /*
739 * Without lock, we cannot be sure that what we got is
740 * a valid page order. Consider only values in the
741 * valid order range to prevent low_pfn overflow.
742 */
743 if (freepage_order > 0 && freepage_order < MAX_ORDER)
744 low_pfn += (1UL << freepage_order) - 1;
Mel Gorman748446b2010-05-24 14:32:27 -0700745 continue;
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700746 }
Mel Gorman748446b2010-05-24 14:32:27 -0700747
Mel Gorman9927af742011-01-13 15:45:59 -0800748 /*
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800749 * Check may be lockless but that's ok as we recheck later.
750 * It's possible to migrate LRU pages and balloon pages
751 * Skip any other type of page
752 */
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700753 is_lru = PageLRU(page);
754 if (!is_lru) {
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800755 if (unlikely(balloon_page_movable(page))) {
Konstantin Khlebnikovd6d86c02014-10-09 15:29:27 -0700756 if (balloon_page_isolate(page)) {
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800757 /* Successfully isolated */
Joonsoo Kimb6c75012014-04-07 15:37:07 -0700758 goto isolate_success;
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800759 }
760 }
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800761 }
Andrea Arcangelibc835012011-01-13 15:47:08 -0800762
763 /*
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700764 * Regardless of being on LRU, compound pages such as THP and
765 * hugetlbfs are not to be compacted. We can potentially save
766 * a lot of iterations if we skip them at once. The check is
767 * racy, but we can consider only valid values and the only
768 * danger is skipping too much.
Andrea Arcangelibc835012011-01-13 15:47:08 -0800769 */
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700770 if (PageCompound(page)) {
771 unsigned int comp_order = compound_order(page);
772
773 if (likely(comp_order < MAX_ORDER))
774 low_pfn += (1UL << comp_order) - 1;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700775
Mel Gorman2a1402a2012-10-08 16:32:33 -0700776 continue;
777 }
778
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700779 if (!is_lru)
780 continue;
781
David Rientjes119d6d52014-04-03 14:48:00 -0700782 /*
783 * Migration will fail if an anonymous page is pinned in memory,
784 * so avoid taking lru_lock and isolating it unnecessarily in an
785 * admittedly racy check.
786 */
787 if (!page_mapping(page) &&
788 page_count(page) > page_mapcount(page))
789 continue;
790
Vlastimil Babka69b71892014-10-09 15:27:18 -0700791 /* If we already hold the lock, we can skip some rechecking */
792 if (!locked) {
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700793 locked = compact_trylock_irqsave(&zone->lru_lock,
794 &flags, cc);
Vlastimil Babka69b71892014-10-09 15:27:18 -0700795 if (!locked)
796 break;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700797
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700798 /* Recheck PageLRU and PageCompound under lock */
Vlastimil Babka69b71892014-10-09 15:27:18 -0700799 if (!PageLRU(page))
800 continue;
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700801
802 /*
803 * Page become compound since the non-locked check,
804 * and it's on LRU. It can only be a THP so the order
805 * is safe to read and it's 0 for tail pages.
806 */
807 if (unlikely(PageCompound(page))) {
808 low_pfn += (1UL << compound_order(page)) - 1;
Vlastimil Babka69b71892014-10-09 15:27:18 -0700809 continue;
810 }
Andrea Arcangelibc835012011-01-13 15:47:08 -0800811 }
812
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700813 lruvec = mem_cgroup_page_lruvec(page, zone);
814
Mel Gorman748446b2010-05-24 14:32:27 -0700815 /* Try isolate the page */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700816 if (__isolate_lru_page(page, isolate_mode) != 0)
Mel Gorman748446b2010-05-24 14:32:27 -0700817 continue;
818
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700819 VM_BUG_ON_PAGE(PageCompound(page), page);
Andrea Arcangelibc835012011-01-13 15:47:08 -0800820
Mel Gorman748446b2010-05-24 14:32:27 -0700821 /* Successfully isolated */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700822 del_page_from_lru_list(page, lruvec, page_lru(page));
Joonsoo Kimb6c75012014-04-07 15:37:07 -0700823
824isolate_success:
Mel Gorman748446b2010-05-24 14:32:27 -0700825 list_add(&page->lru, migratelist);
Mel Gorman748446b2010-05-24 14:32:27 -0700826 cc->nr_migratepages++;
Mel Gormanb7aba692011-01-13 15:45:54 -0800827 nr_isolated++;
Mel Gorman748446b2010-05-24 14:32:27 -0700828
829 /* Avoid isolating too much */
Hillf Danton31b83842012-01-10 15:07:59 -0800830 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
831 ++low_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -0700832 break;
Hillf Danton31b83842012-01-10 15:07:59 -0800833 }
Mel Gorman748446b2010-05-24 14:32:27 -0700834 }
835
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700836 /*
837 * The PageBuddy() check could have potentially brought us outside
838 * the range to be scanned.
839 */
840 if (unlikely(low_pfn > end_pfn))
841 low_pfn = end_pfn;
842
Mel Gormanc67fe372012-08-21 16:16:17 -0700843 if (locked)
844 spin_unlock_irqrestore(&zone->lru_lock, flags);
Mel Gorman748446b2010-05-24 14:32:27 -0700845
Vlastimil Babka50b5b092014-01-21 15:51:10 -0800846 /*
847 * Update the pageblock-skip information and cached scanner pfn,
848 * if the whole pageblock was scanned without isolating any page.
Vlastimil Babka50b5b092014-01-21 15:51:10 -0800849 */
David Rientjes35979ef2014-06-04 16:08:27 -0700850 if (low_pfn == end_pfn)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700851 update_pageblock_skip(cc, valid_page, nr_isolated, true);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700852
Joonsoo Kime34d85f2015-02-11 15:27:04 -0800853 trace_mm_compaction_isolate_migratepages(start_pfn, low_pfn,
854 nr_scanned, nr_isolated);
Mel Gormanb7aba692011-01-13 15:45:54 -0800855
Minchan Kim010fc292012-12-20 15:05:06 -0800856 count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned);
Mel Gorman397487d2012-10-19 12:00:10 +0100857 if (nr_isolated)
Minchan Kim010fc292012-12-20 15:05:06 -0800858 count_compact_events(COMPACTISOLATED, nr_isolated);
Mel Gorman397487d2012-10-19 12:00:10 +0100859
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100860 return low_pfn;
861}
862
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700863/**
864 * isolate_migratepages_range() - isolate migrate-able pages in a PFN range
865 * @cc: Compaction control structure.
866 * @start_pfn: The first PFN to start isolating.
867 * @end_pfn: The one-past-last PFN.
868 *
869 * Returns zero if isolation fails fatally due to e.g. pending signal.
870 * Otherwise, function returns one-past-the-last PFN of isolated page
871 * (which may be greater than end_pfn if end fell in a middle of a THP page).
872 */
873unsigned long
874isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn,
875 unsigned long end_pfn)
876{
877 unsigned long pfn, block_end_pfn;
878
879 /* Scan block by block. First and last block may be incomplete */
880 pfn = start_pfn;
881 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
882
883 for (; pfn < end_pfn; pfn = block_end_pfn,
884 block_end_pfn += pageblock_nr_pages) {
885
886 block_end_pfn = min(block_end_pfn, end_pfn);
887
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700888 if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone))
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700889 continue;
890
891 pfn = isolate_migratepages_block(cc, pfn, block_end_pfn,
892 ISOLATE_UNEVICTABLE);
893
894 /*
895 * In case of fatal failure, release everything that might
896 * have been isolated in the previous iteration, and signal
897 * the failure back to caller.
898 */
899 if (!pfn) {
900 putback_movable_pages(&cc->migratepages);
901 cc->nr_migratepages = 0;
902 break;
903 }
Joonsoo Kim6ea41c02014-10-29 14:50:20 -0700904
905 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX)
906 break;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700907 }
908 acct_isolated(cc->zone, cc);
909
910 return pfn;
911}
912
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100913#endif /* CONFIG_COMPACTION || CONFIG_CMA */
914#ifdef CONFIG_COMPACTION
Andrew Morton018e9a42015-04-15 16:15:20 -0700915
916/* Returns true if the page is within a block suitable for migration to */
917static bool suitable_migration_target(struct page *page)
918{
919 /* If the page is a large free page, then disallow migration */
920 if (PageBuddy(page)) {
921 /*
922 * We are checking page_order without zone->lock taken. But
923 * the only small danger is that we skip a potentially suitable
924 * pageblock, so it's not worth to check order for valid range.
925 */
926 if (page_order_unsafe(page) >= pageblock_order)
927 return false;
928 }
929
930 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
931 if (migrate_async_suitable(get_pageblock_migratetype(page)))
932 return true;
933
934 /* Otherwise skip the block */
935 return false;
936}
937
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100938/*
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -0700939 * Test whether the free scanner has reached the same or lower pageblock than
940 * the migration scanner, and compaction should thus terminate.
941 */
942static inline bool compact_scanners_met(struct compact_control *cc)
943{
944 return (cc->free_pfn >> pageblock_order)
945 <= (cc->migrate_pfn >> pageblock_order);
946}
947
948/*
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100949 * Based on information in the current compact_control, find blocks
950 * suitable for isolating free pages from and then isolate them.
951 */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700952static void isolate_freepages(struct compact_control *cc)
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100953{
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700954 struct zone *zone = cc->zone;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100955 struct page *page;
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700956 unsigned long block_start_pfn; /* start of current pageblock */
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700957 unsigned long isolate_start_pfn; /* exact pfn we start at */
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700958 unsigned long block_end_pfn; /* end of current pageblock */
959 unsigned long low_pfn; /* lowest pfn scanner is able to scan */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100960 struct list_head *freelist = &cc->freepages;
961
962 /*
963 * Initialise the free scanner. The starting point is where we last
Vlastimil Babka49e068f2014-05-06 12:50:03 -0700964 * successfully isolated from, zone-cached value, or the end of the
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700965 * zone when isolating for the first time. For looping we also need
966 * this pfn aligned down to the pageblock boundary, because we do
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700967 * block_start_pfn -= pageblock_nr_pages in the for loop.
968 * For ending point, take care when isolating in last pageblock of a
969 * a zone which ends in the middle of a pageblock.
Vlastimil Babka49e068f2014-05-06 12:50:03 -0700970 * The low boundary is the end of the pageblock the migration scanner
971 * is using.
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100972 */
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700973 isolate_start_pfn = cc->free_pfn;
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700974 block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1);
975 block_end_pfn = min(block_start_pfn + pageblock_nr_pages,
976 zone_end_pfn(zone));
Vlastimil Babka7ed695e2014-01-21 15:51:09 -0800977 low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100978
979 /*
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100980 * Isolate free pages until enough are available to migrate the
981 * pages on cc->migratepages. We stop searching if the migrate
982 * and free page scanners meet or enough free pages are isolated.
983 */
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -0700984 for (; block_start_pfn >= low_pfn;
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700985 block_end_pfn = block_start_pfn,
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700986 block_start_pfn -= pageblock_nr_pages,
987 isolate_start_pfn = block_start_pfn) {
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100988
David Rientjesf6ea3ad2013-09-30 13:45:03 -0700989 /*
990 * This can iterate a massively long zone without finding any
991 * suitable migration targets, so periodically check if we need
Vlastimil Babkabe976572014-06-04 16:10:41 -0700992 * to schedule, or even abort async compaction.
David Rientjesf6ea3ad2013-09-30 13:45:03 -0700993 */
Vlastimil Babkabe976572014-06-04 16:10:41 -0700994 if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages))
995 && compact_should_abort(cc))
996 break;
David Rientjesf6ea3ad2013-09-30 13:45:03 -0700997
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700998 page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn,
999 zone);
1000 if (!page)
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001001 continue;
1002
1003 /* Check the block is suitable for migration */
Linus Torvalds68e3e922012-06-03 20:05:57 -07001004 if (!suitable_migration_target(page))
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001005 continue;
Linus Torvalds68e3e922012-06-03 20:05:57 -07001006
Mel Gormanbb13ffe2012-10-08 16:32:41 -07001007 /* If isolation recently failed, do not retry */
1008 if (!isolation_suitable(cc, page))
1009 continue;
1010
Vlastimil Babkae14c7202014-10-09 15:27:20 -07001011 /* Found a block suitable for isolating free pages from. */
Joonsoo Kim932ff6b2015-02-12 14:59:53 -08001012 isolate_freepages_block(cc, &isolate_start_pfn,
Vlastimil Babkac96b9e52014-06-04 16:07:26 -07001013 block_end_pfn, freelist, false);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001014
1015 /*
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001016 * If we isolated enough freepages, or aborted due to async
1017 * compaction being contended, terminate the loop.
Vlastimil Babkae14c7202014-10-09 15:27:20 -07001018 * Remember where the free scanner should restart next time,
1019 * which is where isolate_freepages_block() left off.
1020 * But if it scanned the whole pageblock, isolate_start_pfn
1021 * now points at block_end_pfn, which is the start of the next
1022 * pageblock.
1023 * In that case we will however want to restart at the start
1024 * of the previous pageblock.
1025 */
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001026 if ((cc->nr_freepages >= cc->nr_migratepages)
1027 || cc->contended) {
1028 if (isolate_start_pfn >= block_end_pfn)
1029 isolate_start_pfn =
1030 block_start_pfn - pageblock_nr_pages;
Vlastimil Babkabe976572014-06-04 16:10:41 -07001031 break;
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001032 } else {
1033 /*
1034 * isolate_freepages_block() should not terminate
1035 * prematurely unless contended, or isolated enough
1036 */
1037 VM_BUG_ON(isolate_start_pfn < block_end_pfn);
1038 }
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +01001039 }
1040
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001041 /* split_free_page does not map the pages */
1042 map_pages(freelist);
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +01001043
Vlastimil Babka7ed695e2014-01-21 15:51:09 -08001044 /*
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001045 * Record where the free scanner will restart next time. Either we
1046 * broke from the loop and set isolate_start_pfn based on the last
1047 * call to isolate_freepages_block(), or we met the migration scanner
1048 * and the loop terminated due to isolate_start_pfn < low_pfn
Vlastimil Babka7ed695e2014-01-21 15:51:09 -08001049 */
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001050 cc->free_pfn = isolate_start_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -07001051}
1052
1053/*
1054 * This is a migrate-callback that "allocates" freepages by taking pages
1055 * from the isolated freelists in the block we are migrating to.
1056 */
1057static struct page *compaction_alloc(struct page *migratepage,
1058 unsigned long data,
1059 int **result)
1060{
1061 struct compact_control *cc = (struct compact_control *)data;
1062 struct page *freepage;
1063
Vlastimil Babkabe976572014-06-04 16:10:41 -07001064 /*
1065 * Isolate free pages if necessary, and if we are not aborting due to
1066 * contention.
1067 */
Mel Gorman748446b2010-05-24 14:32:27 -07001068 if (list_empty(&cc->freepages)) {
Vlastimil Babkabe976572014-06-04 16:10:41 -07001069 if (!cc->contended)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001070 isolate_freepages(cc);
Mel Gorman748446b2010-05-24 14:32:27 -07001071
1072 if (list_empty(&cc->freepages))
1073 return NULL;
1074 }
1075
1076 freepage = list_entry(cc->freepages.next, struct page, lru);
1077 list_del(&freepage->lru);
1078 cc->nr_freepages--;
1079
1080 return freepage;
1081}
1082
1083/*
David Rientjesd53aea32014-06-04 16:08:26 -07001084 * This is a migrate-callback that "frees" freepages back to the isolated
1085 * freelist. All pages on the freelist are from the same zone, so there is no
1086 * special handling needed for NUMA.
1087 */
1088static void compaction_free(struct page *page, unsigned long data)
1089{
1090 struct compact_control *cc = (struct compact_control *)data;
1091
1092 list_add(&page->lru, &cc->freepages);
1093 cc->nr_freepages++;
1094}
1095
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001096/* possible outcome of isolate_migratepages */
1097typedef enum {
1098 ISOLATE_ABORT, /* Abort compaction now */
1099 ISOLATE_NONE, /* No pages isolated, continue scanning */
1100 ISOLATE_SUCCESS, /* Pages isolated, migrate */
1101} isolate_migrate_t;
1102
1103/*
Eric B Munson5bbe3542015-04-15 16:13:20 -07001104 * Allow userspace to control policy on scanning the unevictable LRU for
1105 * compactable pages.
1106 */
1107int sysctl_compact_unevictable_allowed __read_mostly = 1;
1108
1109/*
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001110 * Isolate all pages that can be migrated from the first suitable block,
1111 * starting at the block pointed to by the migrate scanner pfn within
1112 * compact_control.
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001113 */
1114static isolate_migrate_t isolate_migratepages(struct zone *zone,
1115 struct compact_control *cc)
1116{
1117 unsigned long low_pfn, end_pfn;
Joonsoo Kim1a167182015-09-08 15:03:59 -07001118 unsigned long isolate_start_pfn;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001119 struct page *page;
1120 const isolate_mode_t isolate_mode =
Eric B Munson5bbe3542015-04-15 16:13:20 -07001121 (sysctl_compact_unevictable_allowed ? ISOLATE_UNEVICTABLE : 0) |
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001122 (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001123
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001124 /*
1125 * Start at where we last stopped, or beginning of the zone as
1126 * initialized by compact_zone()
1127 */
1128 low_pfn = cc->migrate_pfn;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001129
1130 /* Only scan within a pageblock boundary */
Mel Gormana9aacbc2013-02-22 16:32:25 -08001131 end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001132
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001133 /*
1134 * Iterate over whole pageblocks until we find the first suitable.
1135 * Do not cross the free scanner.
1136 */
1137 for (; end_pfn <= cc->free_pfn;
1138 low_pfn = end_pfn, end_pfn += pageblock_nr_pages) {
1139
1140 /*
1141 * This can potentially iterate a massively long zone with
1142 * many pageblocks unsuitable, so periodically check if we
1143 * need to schedule, or even abort async compaction.
1144 */
1145 if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages))
1146 && compact_should_abort(cc))
1147 break;
1148
Vlastimil Babka7d49d882014-10-09 15:27:11 -07001149 page = pageblock_pfn_to_page(low_pfn, end_pfn, zone);
1150 if (!page)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001151 continue;
1152
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001153 /* If isolation recently failed, do not retry */
1154 if (!isolation_suitable(cc, page))
1155 continue;
1156
1157 /*
1158 * For async compaction, also only scan in MOVABLE blocks.
1159 * Async compaction is optimistic to see if the minimum amount
1160 * of work satisfies the allocation.
1161 */
1162 if (cc->mode == MIGRATE_ASYNC &&
1163 !migrate_async_suitable(get_pageblock_migratetype(page)))
1164 continue;
1165
1166 /* Perform the isolation */
Joonsoo Kim1a167182015-09-08 15:03:59 -07001167 isolate_start_pfn = low_pfn;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001168 low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn,
1169 isolate_mode);
1170
Hugh Dickinsff599092015-02-12 15:00:28 -08001171 if (!low_pfn || cc->contended) {
1172 acct_isolated(zone, cc);
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001173 return ISOLATE_ABORT;
Hugh Dickinsff599092015-02-12 15:00:28 -08001174 }
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001175
1176 /*
Joonsoo Kim1a167182015-09-08 15:03:59 -07001177 * Record where we could have freed pages by migration and not
1178 * yet flushed them to buddy allocator.
1179 * - this is the lowest page that could have been isolated and
1180 * then freed by migration.
1181 */
1182 if (cc->nr_migratepages && !cc->last_migrated_pfn)
1183 cc->last_migrated_pfn = isolate_start_pfn;
1184
1185 /*
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001186 * Either we isolated something and proceed with migration. Or
1187 * we failed and compact_zone should decide if we should
1188 * continue or not.
1189 */
1190 break;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001191 }
1192
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001193 acct_isolated(zone, cc);
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -07001194 /* Record where migration scanner will be restarted. */
1195 cc->migrate_pfn = low_pfn;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001196
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001197 return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001198}
1199
Joonsoo Kim837d0262015-02-11 15:27:06 -08001200static int __compact_finished(struct zone *zone, struct compact_control *cc,
David Rientjes6d7ce552014-10-09 15:27:27 -07001201 const int migratetype)
Mel Gorman748446b2010-05-24 14:32:27 -07001202{
Mel Gorman8fb74b92013-01-11 14:32:16 -08001203 unsigned int order;
Andrea Arcangeli5a03b052011-01-13 15:47:11 -08001204 unsigned long watermark;
Mel Gorman56de7262010-05-24 14:32:30 -07001205
Vlastimil Babkabe976572014-06-04 16:10:41 -07001206 if (cc->contended || fatal_signal_pending(current))
Mel Gorman748446b2010-05-24 14:32:27 -07001207 return COMPACT_PARTIAL;
1208
Mel Gorman753341a2012-10-08 16:32:40 -07001209 /* Compaction run completes if the migrate and free scanner meet */
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -07001210 if (compact_scanners_met(cc)) {
Vlastimil Babka55b7c4c2014-01-21 15:51:11 -08001211 /* Let the next compaction start anew. */
Vlastimil Babka02333642015-09-08 15:02:42 -07001212 reset_cached_positions(zone);
Vlastimil Babka55b7c4c2014-01-21 15:51:11 -08001213
Mel Gorman62997022012-10-08 16:32:47 -07001214 /*
1215 * Mark that the PG_migrate_skip information should be cleared
1216 * by kswapd when it goes to sleep. kswapd does not set the
1217 * flag itself as the decision to be clear should be directly
1218 * based on an allocation request.
1219 */
1220 if (!current_is_kswapd())
1221 zone->compact_blockskip_flush = true;
1222
Mel Gorman748446b2010-05-24 14:32:27 -07001223 return COMPACT_COMPLETE;
Mel Gormanbb13ffe2012-10-08 16:32:41 -07001224 }
Mel Gorman748446b2010-05-24 14:32:27 -07001225
Johannes Weiner82478fb2011-01-20 14:44:21 -08001226 /*
1227 * order == -1 is expected when compacting via
1228 * /proc/sys/vm/compact_memory
1229 */
Mel Gorman56de7262010-05-24 14:32:30 -07001230 if (cc->order == -1)
1231 return COMPACT_CONTINUE;
1232
Michal Hocko3957c772011-06-15 15:08:25 -07001233 /* Compaction run is not finished if the watermark is not met */
1234 watermark = low_wmark_pages(zone);
Michal Hocko3957c772011-06-15 15:08:25 -07001235
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001236 if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx,
1237 cc->alloc_flags))
Michal Hocko3957c772011-06-15 15:08:25 -07001238 return COMPACT_CONTINUE;
1239
Mel Gorman56de7262010-05-24 14:32:30 -07001240 /* Direct compactor: Is a suitable page free? */
Mel Gorman8fb74b92013-01-11 14:32:16 -08001241 for (order = cc->order; order < MAX_ORDER; order++) {
1242 struct free_area *area = &zone->free_area[order];
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001243 bool can_steal;
Mel Gorman56de7262010-05-24 14:32:30 -07001244
Mel Gorman8fb74b92013-01-11 14:32:16 -08001245 /* Job done if page is free of the right migratetype */
David Rientjes6d7ce552014-10-09 15:27:27 -07001246 if (!list_empty(&area->free_list[migratetype]))
Mel Gorman8fb74b92013-01-11 14:32:16 -08001247 return COMPACT_PARTIAL;
1248
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001249#ifdef CONFIG_CMA
1250 /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */
1251 if (migratetype == MIGRATE_MOVABLE &&
1252 !list_empty(&area->free_list[MIGRATE_CMA]))
1253 return COMPACT_PARTIAL;
1254#endif
1255 /*
1256 * Job done if allocation would steal freepages from
1257 * other migratetype buddy lists.
1258 */
1259 if (find_suitable_fallback(area, order, migratetype,
1260 true, &can_steal) != -1)
Mel Gorman8fb74b92013-01-11 14:32:16 -08001261 return COMPACT_PARTIAL;
Mel Gorman56de7262010-05-24 14:32:30 -07001262 }
1263
Joonsoo Kim837d0262015-02-11 15:27:06 -08001264 return COMPACT_NO_SUITABLE_PAGE;
1265}
1266
1267static int compact_finished(struct zone *zone, struct compact_control *cc,
1268 const int migratetype)
1269{
1270 int ret;
1271
1272 ret = __compact_finished(zone, cc, migratetype);
1273 trace_mm_compaction_finished(zone, cc->order, ret);
1274 if (ret == COMPACT_NO_SUITABLE_PAGE)
1275 ret = COMPACT_CONTINUE;
1276
1277 return ret;
Mel Gorman748446b2010-05-24 14:32:27 -07001278}
1279
Mel Gorman3e7d3442011-01-13 15:45:56 -08001280/*
1281 * compaction_suitable: Is this suitable to run compaction on this zone now?
1282 * Returns
1283 * COMPACT_SKIPPED - If there are too few free pages for compaction
1284 * COMPACT_PARTIAL - If the allocation would succeed without compaction
1285 * COMPACT_CONTINUE - If compaction should run now
1286 */
Joonsoo Kim837d0262015-02-11 15:27:06 -08001287static unsigned long __compaction_suitable(struct zone *zone, int order,
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001288 int alloc_flags, int classzone_idx)
Mel Gorman3e7d3442011-01-13 15:45:56 -08001289{
1290 int fragindex;
1291 unsigned long watermark;
1292
1293 /*
Michal Hocko3957c772011-06-15 15:08:25 -07001294 * order == -1 is expected when compacting via
1295 * /proc/sys/vm/compact_memory
1296 */
1297 if (order == -1)
1298 return COMPACT_CONTINUE;
1299
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001300 watermark = low_wmark_pages(zone);
1301 /*
1302 * If watermarks for high-order allocation are already met, there
1303 * should be no need for compaction at all.
1304 */
1305 if (zone_watermark_ok(zone, order, watermark, classzone_idx,
1306 alloc_flags))
1307 return COMPACT_PARTIAL;
1308
Michal Hocko3957c772011-06-15 15:08:25 -07001309 /*
Mel Gorman3e7d3442011-01-13 15:45:56 -08001310 * Watermarks for order-0 must be met for compaction. Note the 2UL.
1311 * This is because during migration, copies of pages need to be
1312 * allocated and for a short time, the footprint is higher
1313 */
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001314 watermark += (2UL << order);
1315 if (!zone_watermark_ok(zone, 0, watermark, classzone_idx, alloc_flags))
Mel Gorman3e7d3442011-01-13 15:45:56 -08001316 return COMPACT_SKIPPED;
1317
1318 /*
1319 * fragmentation index determines if allocation failures are due to
1320 * low memory or external fragmentation
1321 *
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001322 * index of -1000 would imply allocations might succeed depending on
1323 * watermarks, but we already failed the high-order watermark check
Mel Gorman3e7d3442011-01-13 15:45:56 -08001324 * index towards 0 implies failure is due to lack of memory
1325 * index towards 1000 implies failure is due to fragmentation
1326 *
1327 * Only compact if a failure would be due to fragmentation.
1328 */
1329 fragindex = fragmentation_index(zone, order);
1330 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
Joonsoo Kim837d0262015-02-11 15:27:06 -08001331 return COMPACT_NOT_SUITABLE_ZONE;
Mel Gorman3e7d3442011-01-13 15:45:56 -08001332
Mel Gorman3e7d3442011-01-13 15:45:56 -08001333 return COMPACT_CONTINUE;
1334}
1335
Joonsoo Kim837d0262015-02-11 15:27:06 -08001336unsigned long compaction_suitable(struct zone *zone, int order,
1337 int alloc_flags, int classzone_idx)
1338{
1339 unsigned long ret;
1340
1341 ret = __compaction_suitable(zone, order, alloc_flags, classzone_idx);
1342 trace_mm_compaction_suitable(zone, order, ret);
1343 if (ret == COMPACT_NOT_SUITABLE_ZONE)
1344 ret = COMPACT_SKIPPED;
1345
1346 return ret;
1347}
1348
Mel Gorman748446b2010-05-24 14:32:27 -07001349static int compact_zone(struct zone *zone, struct compact_control *cc)
1350{
1351 int ret;
Mel Gormanc89511a2012-10-08 16:32:45 -07001352 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001353 unsigned long end_pfn = zone_end_pfn(zone);
David Rientjes6d7ce552014-10-09 15:27:27 -07001354 const int migratetype = gfpflags_to_migratetype(cc->gfp_mask);
David Rientjese0b9dae2014-06-04 16:08:28 -07001355 const bool sync = cc->mode != MIGRATE_ASYNC;
Mel Gorman748446b2010-05-24 14:32:27 -07001356
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001357 ret = compaction_suitable(zone, cc->order, cc->alloc_flags,
1358 cc->classzone_idx);
Mel Gorman3e7d3442011-01-13 15:45:56 -08001359 switch (ret) {
1360 case COMPACT_PARTIAL:
1361 case COMPACT_SKIPPED:
1362 /* Compaction is likely to fail */
1363 return ret;
1364 case COMPACT_CONTINUE:
1365 /* Fall through to compaction */
1366 ;
1367 }
1368
Mel Gormanc89511a2012-10-08 16:32:45 -07001369 /*
Vlastimil Babkad3132e42014-01-21 15:51:08 -08001370 * Clear pageblock skip if there were failures recently and compaction
1371 * is about to be retried after being deferred. kswapd does not do
1372 * this reset as it'll reset the cached information when going to sleep.
1373 */
1374 if (compaction_restarting(zone, cc->order) && !current_is_kswapd())
1375 __reset_isolation_suitable(zone);
1376
1377 /*
Mel Gormanc89511a2012-10-08 16:32:45 -07001378 * Setup to move all movable pages to the end of the zone. Used cached
1379 * information on where the scanners should start but check that it
1380 * is initialised by ensuring the values are within zone boundaries.
1381 */
David Rientjese0b9dae2014-06-04 16:08:28 -07001382 cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync];
Mel Gormanc89511a2012-10-08 16:32:45 -07001383 cc->free_pfn = zone->compact_cached_free_pfn;
1384 if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) {
1385 cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1);
1386 zone->compact_cached_free_pfn = cc->free_pfn;
1387 }
1388 if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) {
1389 cc->migrate_pfn = start_pfn;
David Rientjes35979ef2014-06-04 16:08:27 -07001390 zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn;
1391 zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn;
Mel Gormanc89511a2012-10-08 16:32:45 -07001392 }
Joonsoo Kim1a167182015-09-08 15:03:59 -07001393 cc->last_migrated_pfn = 0;
Mel Gorman748446b2010-05-24 14:32:27 -07001394
Joonsoo Kim16c4a092015-02-11 15:27:01 -08001395 trace_mm_compaction_begin(start_pfn, cc->migrate_pfn,
1396 cc->free_pfn, end_pfn, sync);
Mel Gorman0eb927c2014-01-21 15:51:05 -08001397
Mel Gorman748446b2010-05-24 14:32:27 -07001398 migrate_prep_local();
1399
David Rientjes6d7ce552014-10-09 15:27:27 -07001400 while ((ret = compact_finished(zone, cc, migratetype)) ==
1401 COMPACT_CONTINUE) {
Minchan Kim9d502c12011-03-22 16:30:39 -07001402 int err;
Mel Gorman748446b2010-05-24 14:32:27 -07001403
Mel Gormanf9e35b32011-06-15 15:08:52 -07001404 switch (isolate_migratepages(zone, cc)) {
1405 case ISOLATE_ABORT:
1406 ret = COMPACT_PARTIAL;
Rafael Aquini5733c7d2012-12-11 16:02:47 -08001407 putback_movable_pages(&cc->migratepages);
Shaohua Lie64c5232012-10-08 16:32:27 -07001408 cc->nr_migratepages = 0;
Mel Gormanf9e35b32011-06-15 15:08:52 -07001409 goto out;
1410 case ISOLATE_NONE:
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001411 /*
1412 * We haven't isolated and migrated anything, but
1413 * there might still be unflushed migrations from
1414 * previous cc->order aligned block.
1415 */
1416 goto check_drain;
Mel Gormanf9e35b32011-06-15 15:08:52 -07001417 case ISOLATE_SUCCESS:
1418 ;
1419 }
Mel Gorman748446b2010-05-24 14:32:27 -07001420
David Rientjesd53aea32014-06-04 16:08:26 -07001421 err = migrate_pages(&cc->migratepages, compaction_alloc,
David Rientjese0b9dae2014-06-04 16:08:28 -07001422 compaction_free, (unsigned long)cc, cc->mode,
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001423 MR_COMPACTION);
Mel Gorman748446b2010-05-24 14:32:27 -07001424
Vlastimil Babkaf8c93012014-06-04 16:08:32 -07001425 trace_mm_compaction_migratepages(cc->nr_migratepages, err,
1426 &cc->migratepages);
Mel Gorman748446b2010-05-24 14:32:27 -07001427
Vlastimil Babkaf8c93012014-06-04 16:08:32 -07001428 /* All pages were either migrated or will be released */
1429 cc->nr_migratepages = 0;
Minchan Kim9d502c12011-03-22 16:30:39 -07001430 if (err) {
Rafael Aquini5733c7d2012-12-11 16:02:47 -08001431 putback_movable_pages(&cc->migratepages);
Vlastimil Babka7ed695e2014-01-21 15:51:09 -08001432 /*
1433 * migrate_pages() may return -ENOMEM when scanners meet
1434 * and we want compact_finished() to detect it
1435 */
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -07001436 if (err == -ENOMEM && !compact_scanners_met(cc)) {
David Rientjes4bf2bba2012-07-11 14:02:13 -07001437 ret = COMPACT_PARTIAL;
1438 goto out;
1439 }
Mel Gorman748446b2010-05-24 14:32:27 -07001440 }
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001441
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001442check_drain:
1443 /*
1444 * Has the migration scanner moved away from the previous
1445 * cc->order aligned block where we migrated from? If yes,
1446 * flush the pages that were freed, so that they can merge and
1447 * compact_finished() can detect immediately if allocation
1448 * would succeed.
1449 */
Joonsoo Kim1a167182015-09-08 15:03:59 -07001450 if (cc->order > 0 && cc->last_migrated_pfn) {
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001451 int cpu;
1452 unsigned long current_block_start =
1453 cc->migrate_pfn & ~((1UL << cc->order) - 1);
1454
Joonsoo Kim1a167182015-09-08 15:03:59 -07001455 if (cc->last_migrated_pfn < current_block_start) {
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001456 cpu = get_cpu();
1457 lru_add_drain_cpu(cpu);
1458 drain_local_pages(zone);
1459 put_cpu();
1460 /* No more flushing until we migrate again */
Joonsoo Kim1a167182015-09-08 15:03:59 -07001461 cc->last_migrated_pfn = 0;
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001462 }
1463 }
1464
Mel Gorman748446b2010-05-24 14:32:27 -07001465 }
1466
Mel Gormanf9e35b32011-06-15 15:08:52 -07001467out:
Vlastimil Babka6bace092014-12-10 15:43:31 -08001468 /*
1469 * Release free pages and update where the free scanner should restart,
1470 * so we don't leave any returned pages behind in the next attempt.
1471 */
1472 if (cc->nr_freepages > 0) {
1473 unsigned long free_pfn = release_freepages(&cc->freepages);
1474
1475 cc->nr_freepages = 0;
1476 VM_BUG_ON(free_pfn == 0);
1477 /* The cached pfn is always the first in a pageblock */
1478 free_pfn &= ~(pageblock_nr_pages-1);
1479 /*
1480 * Only go back, not forward. The cached pfn might have been
1481 * already reset to zone end in compact_finished()
1482 */
1483 if (free_pfn > zone->compact_cached_free_pfn)
1484 zone->compact_cached_free_pfn = free_pfn;
1485 }
Mel Gorman748446b2010-05-24 14:32:27 -07001486
Joonsoo Kim16c4a092015-02-11 15:27:01 -08001487 trace_mm_compaction_end(start_pfn, cc->migrate_pfn,
1488 cc->free_pfn, end_pfn, sync, ret);
Mel Gorman0eb927c2014-01-21 15:51:05 -08001489
Mel Gorman748446b2010-05-24 14:32:27 -07001490 return ret;
1491}
Mel Gorman76ab0f52010-05-24 14:32:28 -07001492
David Rientjese0b9dae2014-06-04 16:08:28 -07001493static unsigned long compact_zone_order(struct zone *zone, int order,
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001494 gfp_t gfp_mask, enum migrate_mode mode, int *contended,
1495 int alloc_flags, int classzone_idx)
Mel Gorman56de7262010-05-24 14:32:30 -07001496{
Shaohua Lie64c5232012-10-08 16:32:27 -07001497 unsigned long ret;
Mel Gorman56de7262010-05-24 14:32:30 -07001498 struct compact_control cc = {
1499 .nr_freepages = 0,
1500 .nr_migratepages = 0,
1501 .order = order,
David Rientjes6d7ce552014-10-09 15:27:27 -07001502 .gfp_mask = gfp_mask,
Mel Gorman56de7262010-05-24 14:32:30 -07001503 .zone = zone,
David Rientjese0b9dae2014-06-04 16:08:28 -07001504 .mode = mode,
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001505 .alloc_flags = alloc_flags,
1506 .classzone_idx = classzone_idx,
Mel Gorman56de7262010-05-24 14:32:30 -07001507 };
1508 INIT_LIST_HEAD(&cc.freepages);
1509 INIT_LIST_HEAD(&cc.migratepages);
1510
Shaohua Lie64c5232012-10-08 16:32:27 -07001511 ret = compact_zone(zone, &cc);
1512
1513 VM_BUG_ON(!list_empty(&cc.freepages));
1514 VM_BUG_ON(!list_empty(&cc.migratepages));
1515
1516 *contended = cc.contended;
1517 return ret;
Mel Gorman56de7262010-05-24 14:32:30 -07001518}
1519
Mel Gorman5e771902010-05-24 14:32:31 -07001520int sysctl_extfrag_threshold = 500;
1521
Mel Gorman56de7262010-05-24 14:32:30 -07001522/**
1523 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
Mel Gorman56de7262010-05-24 14:32:30 -07001524 * @gfp_mask: The GFP mask of the current allocation
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001525 * @order: The order of the current allocation
1526 * @alloc_flags: The allocation flags of the current allocation
1527 * @ac: The context of current allocation
David Rientjese0b9dae2014-06-04 16:08:28 -07001528 * @mode: The migration mode for async, sync light, or sync migration
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001529 * @contended: Return value that determines if compaction was aborted due to
1530 * need_resched() or lock contention
Mel Gorman56de7262010-05-24 14:32:30 -07001531 *
1532 * This is the main entry point for direct page compaction.
1533 */
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001534unsigned long try_to_compact_pages(gfp_t gfp_mask, unsigned int order,
1535 int alloc_flags, const struct alloc_context *ac,
1536 enum migrate_mode mode, int *contended)
Mel Gorman56de7262010-05-24 14:32:30 -07001537{
Mel Gorman56de7262010-05-24 14:32:30 -07001538 int may_enter_fs = gfp_mask & __GFP_FS;
1539 int may_perform_io = gfp_mask & __GFP_IO;
Mel Gorman56de7262010-05-24 14:32:30 -07001540 struct zoneref *z;
1541 struct zone *zone;
Vlastimil Babka53853e22014-10-09 15:27:02 -07001542 int rc = COMPACT_DEFERRED;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001543 int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */
1544
1545 *contended = COMPACT_CONTENDED_NONE;
Mel Gorman56de7262010-05-24 14:32:30 -07001546
Mel Gorman4ffb6332012-10-08 16:29:09 -07001547 /* Check if the GFP flags allow compaction */
Andrea Arcangelic5a73c32011-01-13 15:47:11 -08001548 if (!order || !may_enter_fs || !may_perform_io)
Vlastimil Babka53853e22014-10-09 15:27:02 -07001549 return COMPACT_SKIPPED;
Mel Gorman56de7262010-05-24 14:32:30 -07001550
Joonsoo Kim837d0262015-02-11 15:27:06 -08001551 trace_mm_compaction_try_to_compact_pages(order, gfp_mask, mode);
1552
Mel Gorman56de7262010-05-24 14:32:30 -07001553 /* Compact each zone in the list */
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001554 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
1555 ac->nodemask) {
Mel Gorman56de7262010-05-24 14:32:30 -07001556 int status;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001557 int zone_contended;
Mel Gorman56de7262010-05-24 14:32:30 -07001558
Vlastimil Babka53853e22014-10-09 15:27:02 -07001559 if (compaction_deferred(zone, order))
1560 continue;
1561
David Rientjese0b9dae2014-06-04 16:08:28 -07001562 status = compact_zone_order(zone, order, gfp_mask, mode,
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001563 &zone_contended, alloc_flags,
1564 ac->classzone_idx);
Mel Gorman56de7262010-05-24 14:32:30 -07001565 rc = max(status, rc);
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001566 /*
1567 * It takes at least one zone that wasn't lock contended
1568 * to clear all_zones_contended.
1569 */
1570 all_zones_contended &= zone_contended;
Mel Gorman56de7262010-05-24 14:32:30 -07001571
Mel Gorman3e7d3442011-01-13 15:45:56 -08001572 /* If a normal allocation would succeed, stop compacting */
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001573 if (zone_watermark_ok(zone, order, low_wmark_pages(zone),
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001574 ac->classzone_idx, alloc_flags)) {
Vlastimil Babka53853e22014-10-09 15:27:02 -07001575 /*
1576 * We think the allocation will succeed in this zone,
1577 * but it is not certain, hence the false. The caller
1578 * will repeat this with true if allocation indeed
1579 * succeeds in this zone.
1580 */
1581 compaction_defer_reset(zone, order, false);
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001582 /*
1583 * It is possible that async compaction aborted due to
1584 * need_resched() and the watermarks were ok thanks to
1585 * somebody else freeing memory. The allocation can
1586 * however still fail so we better signal the
1587 * need_resched() contention anyway (this will not
1588 * prevent the allocation attempt).
1589 */
1590 if (zone_contended == COMPACT_CONTENDED_SCHED)
1591 *contended = COMPACT_CONTENDED_SCHED;
1592
1593 goto break_loop;
1594 }
1595
Vlastimil Babkaf8669792014-12-10 15:43:28 -08001596 if (mode != MIGRATE_ASYNC && status == COMPACT_COMPLETE) {
Vlastimil Babka53853e22014-10-09 15:27:02 -07001597 /*
1598 * We think that allocation won't succeed in this zone
1599 * so we defer compaction there. If it ends up
1600 * succeeding after all, it will be reset.
1601 */
1602 defer_compaction(zone, order);
1603 }
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001604
1605 /*
1606 * We might have stopped compacting due to need_resched() in
1607 * async compaction, or due to a fatal signal detected. In that
1608 * case do not try further zones and signal need_resched()
1609 * contention.
1610 */
1611 if ((zone_contended == COMPACT_CONTENDED_SCHED)
1612 || fatal_signal_pending(current)) {
1613 *contended = COMPACT_CONTENDED_SCHED;
1614 goto break_loop;
1615 }
1616
1617 continue;
1618break_loop:
1619 /*
1620 * We might not have tried all the zones, so be conservative
1621 * and assume they are not all lock contended.
1622 */
1623 all_zones_contended = 0;
1624 break;
Mel Gorman56de7262010-05-24 14:32:30 -07001625 }
1626
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001627 /*
1628 * If at least one zone wasn't deferred or skipped, we report if all
1629 * zones that were tried were lock contended.
1630 */
1631 if (rc > COMPACT_SKIPPED && all_zones_contended)
1632 *contended = COMPACT_CONTENDED_LOCK;
1633
Mel Gorman56de7262010-05-24 14:32:30 -07001634 return rc;
1635}
1636
1637
Mel Gorman76ab0f52010-05-24 14:32:28 -07001638/* Compact all zones within a node */
Andrew Morton7103f162013-02-22 16:32:33 -08001639static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
Mel Gorman76ab0f52010-05-24 14:32:28 -07001640{
1641 int zoneid;
Mel Gorman76ab0f52010-05-24 14:32:28 -07001642 struct zone *zone;
1643
Mel Gorman76ab0f52010-05-24 14:32:28 -07001644 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
Mel Gorman76ab0f52010-05-24 14:32:28 -07001645
1646 zone = &pgdat->node_zones[zoneid];
1647 if (!populated_zone(zone))
1648 continue;
1649
Rik van Riel7be62de2012-03-21 16:33:52 -07001650 cc->nr_freepages = 0;
1651 cc->nr_migratepages = 0;
1652 cc->zone = zone;
1653 INIT_LIST_HEAD(&cc->freepages);
1654 INIT_LIST_HEAD(&cc->migratepages);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001655
Gioh Kim195b0c62015-04-15 16:13:33 -07001656 /*
1657 * When called via /proc/sys/vm/compact_memory
1658 * this makes sure we compact the whole zone regardless of
1659 * cached scanner positions.
1660 */
1661 if (cc->order == -1)
1662 __reset_isolation_suitable(zone);
1663
Dan Carpenteraad6ec32012-03-21 16:33:54 -07001664 if (cc->order == -1 || !compaction_deferred(zone, cc->order))
Rik van Riel7be62de2012-03-21 16:33:52 -07001665 compact_zone(zone, cc);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001666
Rik van Rielaff62242012-03-21 16:33:52 -07001667 if (cc->order > 0) {
Vlastimil Babkade6c60a2014-01-21 15:51:07 -08001668 if (zone_watermark_ok(zone, cc->order,
1669 low_wmark_pages(zone), 0, 0))
1670 compaction_defer_reset(zone, cc->order, false);
Rik van Rielaff62242012-03-21 16:33:52 -07001671 }
1672
Rik van Riel7be62de2012-03-21 16:33:52 -07001673 VM_BUG_ON(!list_empty(&cc->freepages));
1674 VM_BUG_ON(!list_empty(&cc->migratepages));
Mel Gorman76ab0f52010-05-24 14:32:28 -07001675 }
Mel Gorman76ab0f52010-05-24 14:32:28 -07001676}
1677
Andrew Morton7103f162013-02-22 16:32:33 -08001678void compact_pgdat(pg_data_t *pgdat, int order)
Rik van Riel7be62de2012-03-21 16:33:52 -07001679{
1680 struct compact_control cc = {
1681 .order = order,
David Rientjese0b9dae2014-06-04 16:08:28 -07001682 .mode = MIGRATE_ASYNC,
Rik van Riel7be62de2012-03-21 16:33:52 -07001683 };
1684
Mel Gorman3a7200a2013-09-11 14:22:19 -07001685 if (!order)
1686 return;
1687
Andrew Morton7103f162013-02-22 16:32:33 -08001688 __compact_pgdat(pgdat, &cc);
Rik van Riel7be62de2012-03-21 16:33:52 -07001689}
1690
Andrew Morton7103f162013-02-22 16:32:33 -08001691static void compact_node(int nid)
Rik van Riel7be62de2012-03-21 16:33:52 -07001692{
Rik van Riel7be62de2012-03-21 16:33:52 -07001693 struct compact_control cc = {
1694 .order = -1,
David Rientjese0b9dae2014-06-04 16:08:28 -07001695 .mode = MIGRATE_SYNC,
David Rientjes91ca9182014-04-03 14:47:23 -07001696 .ignore_skip_hint = true,
Rik van Riel7be62de2012-03-21 16:33:52 -07001697 };
1698
Andrew Morton7103f162013-02-22 16:32:33 -08001699 __compact_pgdat(NODE_DATA(nid), &cc);
Rik van Riel7be62de2012-03-21 16:33:52 -07001700}
1701
Mel Gorman76ab0f52010-05-24 14:32:28 -07001702/* Compact all nodes in the system */
Jason Liu7964c062013-01-11 14:31:47 -08001703static void compact_nodes(void)
Mel Gorman76ab0f52010-05-24 14:32:28 -07001704{
1705 int nid;
1706
Hugh Dickins8575ec22012-03-21 16:33:53 -07001707 /* Flush pending updates to the LRU lists */
1708 lru_add_drain_all();
1709
Mel Gorman76ab0f52010-05-24 14:32:28 -07001710 for_each_online_node(nid)
1711 compact_node(nid);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001712}
1713
1714/* The written value is actually unused, all memory is compacted */
1715int sysctl_compact_memory;
1716
1717/* This is the entry point for compacting all nodes via /proc/sys/vm */
1718int sysctl_compaction_handler(struct ctl_table *table, int write,
1719 void __user *buffer, size_t *length, loff_t *ppos)
1720{
1721 if (write)
Jason Liu7964c062013-01-11 14:31:47 -08001722 compact_nodes();
Mel Gorman76ab0f52010-05-24 14:32:28 -07001723
1724 return 0;
1725}
Mel Gormaned4a6d72010-05-24 14:32:29 -07001726
Mel Gorman5e771902010-05-24 14:32:31 -07001727int sysctl_extfrag_handler(struct ctl_table *table, int write,
1728 void __user *buffer, size_t *length, loff_t *ppos)
1729{
1730 proc_dointvec_minmax(table, write, buffer, length, ppos);
1731
1732 return 0;
1733}
1734
Mel Gormaned4a6d72010-05-24 14:32:29 -07001735#if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
Rashika Kheria74e77fb2014-04-03 14:48:01 -07001736static ssize_t sysfs_compact_node(struct device *dev,
Kay Sievers10fbcf42011-12-21 14:48:43 -08001737 struct device_attribute *attr,
Mel Gormaned4a6d72010-05-24 14:32:29 -07001738 const char *buf, size_t count)
1739{
Hugh Dickins8575ec22012-03-21 16:33:53 -07001740 int nid = dev->id;
1741
1742 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1743 /* Flush pending updates to the LRU lists */
1744 lru_add_drain_all();
1745
1746 compact_node(nid);
1747 }
Mel Gormaned4a6d72010-05-24 14:32:29 -07001748
1749 return count;
1750}
Kay Sievers10fbcf42011-12-21 14:48:43 -08001751static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001752
1753int compaction_register_node(struct node *node)
1754{
Kay Sievers10fbcf42011-12-21 14:48:43 -08001755 return device_create_file(&node->dev, &dev_attr_compact);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001756}
1757
1758void compaction_unregister_node(struct node *node)
1759{
Kay Sievers10fbcf42011-12-21 14:48:43 -08001760 return device_remove_file(&node->dev, &dev_attr_compact);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001761}
1762#endif /* CONFIG_SYSFS && CONFIG_NUMA */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001763
1764#endif /* CONFIG_COMPACTION */