blob: 585de54dbe8ccef06543b547eadfe8e4fe8e3117 [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
38
Mel Gormanb7aba692011-01-13 15:45:54 -080039#define CREATE_TRACE_POINTS
40#include <trace/events/compaction.h>
41
Mel Gorman748446b2010-05-24 14:32:27 -070042static unsigned long release_freepages(struct list_head *freelist)
43{
44 struct page *page, *next;
Vlastimil Babka6bace092014-12-10 15:43:31 -080045 unsigned long high_pfn = 0;
Mel Gorman748446b2010-05-24 14:32:27 -070046
47 list_for_each_entry_safe(page, next, freelist, lru) {
Vlastimil Babka6bace092014-12-10 15:43:31 -080048 unsigned long pfn = page_to_pfn(page);
Mel Gorman748446b2010-05-24 14:32:27 -070049 list_del(&page->lru);
50 __free_page(page);
Vlastimil Babka6bace092014-12-10 15:43:31 -080051 if (pfn > high_pfn)
52 high_pfn = pfn;
Mel Gorman748446b2010-05-24 14:32:27 -070053 }
54
Vlastimil Babka6bace092014-12-10 15:43:31 -080055 return high_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -070056}
57
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010058static void map_pages(struct list_head *list)
59{
60 struct page *page;
61
62 list_for_each_entry(page, list, lru) {
63 arch_alloc_page(page, 0);
64 kernel_map_pages(page, 1, 1);
Andrey Ryabininb8c73fc2015-02-13 14:39:28 -080065 kasan_alloc_pages(page, 0);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +010066 }
67}
68
Michal Nazarewicz47118af2011-12-29 13:09:50 +010069static inline bool migrate_async_suitable(int migratetype)
70{
71 return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
72}
73
Vlastimil Babka7d49d882014-10-09 15:27:11 -070074/*
75 * Check that the whole (or subset of) a pageblock given by the interval of
76 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
77 * with the migration of free compaction scanner. The scanners then need to
78 * use only pfn_valid_within() check for arches that allow holes within
79 * pageblocks.
80 *
81 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
82 *
83 * It's possible on some configurations to have a setup like node0 node1 node0
84 * i.e. it's possible that all pages within a zones range of pages do not
85 * belong to a single zone. We assume that a border between node0 and node1
86 * can occur within a single pageblock, but not a node0 node1 node0
87 * interleaving within a single pageblock. It is therefore sufficient to check
88 * the first and last page of a pageblock and avoid checking each individual
89 * page in a pageblock.
90 */
91static struct page *pageblock_pfn_to_page(unsigned long start_pfn,
92 unsigned long end_pfn, struct zone *zone)
93{
94 struct page *start_page;
95 struct page *end_page;
96
97 /* end_pfn is one past the range we are checking */
98 end_pfn--;
99
100 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
101 return NULL;
102
103 start_page = pfn_to_page(start_pfn);
104
105 if (page_zone(start_page) != zone)
106 return NULL;
107
108 end_page = pfn_to_page(end_pfn);
109
110 /* This gives a shorter code than deriving page_zone(end_page) */
111 if (page_zone_id(start_page) != page_zone_id(end_page))
112 return NULL;
113
114 return start_page;
115}
116
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700117#ifdef CONFIG_COMPACTION
Joonsoo Kim24e27162015-02-11 15:27:09 -0800118
119/* Do not skip compaction more than 64 times */
120#define COMPACT_MAX_DEFER_SHIFT 6
121
122/*
123 * Compaction is deferred when compaction fails to result in a page
124 * allocation success. 1 << compact_defer_limit compactions are skipped up
125 * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT
126 */
127void defer_compaction(struct zone *zone, int order)
128{
129 zone->compact_considered = 0;
130 zone->compact_defer_shift++;
131
132 if (order < zone->compact_order_failed)
133 zone->compact_order_failed = order;
134
135 if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT)
136 zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT;
137
138 trace_mm_compaction_defer_compaction(zone, order);
139}
140
141/* Returns true if compaction should be skipped this time */
142bool compaction_deferred(struct zone *zone, int order)
143{
144 unsigned long defer_limit = 1UL << zone->compact_defer_shift;
145
146 if (order < zone->compact_order_failed)
147 return false;
148
149 /* Avoid possible overflow */
150 if (++zone->compact_considered > defer_limit)
151 zone->compact_considered = defer_limit;
152
153 if (zone->compact_considered >= defer_limit)
154 return false;
155
156 trace_mm_compaction_deferred(zone, order);
157
158 return true;
159}
160
161/*
162 * Update defer tracking counters after successful compaction of given order,
163 * which means an allocation either succeeded (alloc_success == true) or is
164 * expected to succeed.
165 */
166void compaction_defer_reset(struct zone *zone, int order,
167 bool alloc_success)
168{
169 if (alloc_success) {
170 zone->compact_considered = 0;
171 zone->compact_defer_shift = 0;
172 }
173 if (order >= zone->compact_order_failed)
174 zone->compact_order_failed = order + 1;
175
176 trace_mm_compaction_defer_reset(zone, order);
177}
178
179/* Returns true if restarting compaction after many failures */
180bool compaction_restarting(struct zone *zone, int order)
181{
182 if (order < zone->compact_order_failed)
183 return false;
184
185 return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT &&
186 zone->compact_considered >= 1UL << zone->compact_defer_shift;
187}
188
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700189/* Returns true if the pageblock should be scanned for pages to isolate. */
190static inline bool isolation_suitable(struct compact_control *cc,
191 struct page *page)
192{
193 if (cc->ignore_skip_hint)
194 return true;
195
196 return !get_pageblock_skip(page);
197}
198
Vlastimil Babka02333642015-09-08 15:02:42 -0700199static void reset_cached_positions(struct zone *zone)
200{
201 zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn;
202 zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn;
203 zone->compact_cached_free_pfn = zone_end_pfn(zone);
204}
205
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700206/*
207 * This function is called to clear all cached information on pageblocks that
208 * should be skipped for page isolation when the migrate and free page scanner
209 * meet.
210 */
Mel Gorman62997022012-10-08 16:32:47 -0700211static void __reset_isolation_suitable(struct zone *zone)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700212{
213 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -0800214 unsigned long end_pfn = zone_end_pfn(zone);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700215 unsigned long pfn;
216
Mel Gorman62997022012-10-08 16:32:47 -0700217 zone->compact_blockskip_flush = false;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700218
219 /* Walk the zone and mark every pageblock as suitable for isolation */
220 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
221 struct page *page;
222
223 cond_resched();
224
225 if (!pfn_valid(pfn))
226 continue;
227
228 page = pfn_to_page(pfn);
229 if (zone != page_zone(page))
230 continue;
231
232 clear_pageblock_skip(page);
233 }
Vlastimil Babka02333642015-09-08 15:02:42 -0700234
235 reset_cached_positions(zone);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700236}
237
Mel Gorman62997022012-10-08 16:32:47 -0700238void reset_isolation_suitable(pg_data_t *pgdat)
239{
240 int zoneid;
241
242 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
243 struct zone *zone = &pgdat->node_zones[zoneid];
244 if (!populated_zone(zone))
245 continue;
246
247 /* Only flush if a full compaction finished recently */
248 if (zone->compact_blockskip_flush)
249 __reset_isolation_suitable(zone);
250 }
251}
252
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700253/*
254 * If no pages were isolated then mark this pageblock to be skipped in the
Mel Gorman62997022012-10-08 16:32:47 -0700255 * future. The information is later cleared by __reset_isolation_suitable().
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700256 */
Mel Gormanc89511a2012-10-08 16:32:45 -0700257static void update_pageblock_skip(struct compact_control *cc,
258 struct page *page, unsigned long nr_isolated,
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700259 bool migrate_scanner)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700260{
Mel Gormanc89511a2012-10-08 16:32:45 -0700261 struct zone *zone = cc->zone;
David Rientjes35979ef2014-06-04 16:08:27 -0700262 unsigned long pfn;
Joonsoo Kim6815bf32013-12-18 17:08:52 -0800263
264 if (cc->ignore_skip_hint)
265 return;
266
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700267 if (!page)
268 return;
269
David Rientjes35979ef2014-06-04 16:08:27 -0700270 if (nr_isolated)
271 return;
272
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700273 set_pageblock_skip(page);
Mel Gormanc89511a2012-10-08 16:32:45 -0700274
David Rientjes35979ef2014-06-04 16:08:27 -0700275 pfn = page_to_pfn(page);
276
277 /* Update where async and sync compaction should restart */
278 if (migrate_scanner) {
David Rientjes35979ef2014-06-04 16:08:27 -0700279 if (pfn > zone->compact_cached_migrate_pfn[0])
280 zone->compact_cached_migrate_pfn[0] = pfn;
David Rientjese0b9dae2014-06-04 16:08:28 -0700281 if (cc->mode != MIGRATE_ASYNC &&
282 pfn > zone->compact_cached_migrate_pfn[1])
David Rientjes35979ef2014-06-04 16:08:27 -0700283 zone->compact_cached_migrate_pfn[1] = pfn;
284 } else {
David Rientjes35979ef2014-06-04 16:08:27 -0700285 if (pfn < zone->compact_cached_free_pfn)
286 zone->compact_cached_free_pfn = pfn;
Mel Gormanc89511a2012-10-08 16:32:45 -0700287 }
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700288}
289#else
290static inline bool isolation_suitable(struct compact_control *cc,
291 struct page *page)
292{
293 return true;
294}
295
Mel Gormanc89511a2012-10-08 16:32:45 -0700296static void update_pageblock_skip(struct compact_control *cc,
297 struct page *page, unsigned long nr_isolated,
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700298 bool migrate_scanner)
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700299{
300}
301#endif /* CONFIG_COMPACTION */
302
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700303/*
304 * Compaction requires the taking of some coarse locks that are potentially
305 * very heavily contended. For async compaction, back out if the lock cannot
306 * be taken immediately. For sync compaction, spin on the lock if needed.
307 *
308 * Returns true if the lock is held
309 * Returns false if the lock is not held and compaction should abort
310 */
311static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags,
312 struct compact_control *cc)
Mel Gorman2a1402a2012-10-08 16:32:33 -0700313{
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700314 if (cc->mode == MIGRATE_ASYNC) {
315 if (!spin_trylock_irqsave(lock, *flags)) {
316 cc->contended = COMPACT_CONTENDED_LOCK;
317 return false;
318 }
319 } else {
320 spin_lock_irqsave(lock, *flags);
321 }
Vlastimil Babka1f9efde2014-10-09 15:27:14 -0700322
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700323 return true;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700324}
325
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100326/*
Mel Gormanc67fe372012-08-21 16:16:17 -0700327 * Compaction requires the taking of some coarse locks that are potentially
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700328 * very heavily contended. The lock should be periodically unlocked to avoid
329 * having disabled IRQs for a long time, even when there is nobody waiting on
330 * the lock. It might also be that allowing the IRQs will result in
331 * need_resched() becoming true. If scheduling is needed, async compaction
332 * aborts. Sync compaction schedules.
333 * Either compaction type will also abort if a fatal signal is pending.
334 * In either case if the lock was locked, it is dropped and not regained.
Mel Gormanc67fe372012-08-21 16:16:17 -0700335 *
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700336 * Returns true if compaction should abort due to fatal signal pending, or
337 * async compaction due to need_resched()
338 * Returns false when compaction can continue (sync compaction might have
339 * scheduled)
Mel Gormanc67fe372012-08-21 16:16:17 -0700340 */
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700341static bool compact_unlock_should_abort(spinlock_t *lock,
342 unsigned long flags, bool *locked, struct compact_control *cc)
Mel Gormanc67fe372012-08-21 16:16:17 -0700343{
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700344 if (*locked) {
345 spin_unlock_irqrestore(lock, flags);
346 *locked = false;
347 }
Vlastimil Babka1f9efde2014-10-09 15:27:14 -0700348
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700349 if (fatal_signal_pending(current)) {
350 cc->contended = COMPACT_CONTENDED_SCHED;
351 return true;
352 }
Mel Gormanc67fe372012-08-21 16:16:17 -0700353
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700354 if (need_resched()) {
David Rientjese0b9dae2014-06-04 16:08:28 -0700355 if (cc->mode == MIGRATE_ASYNC) {
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700356 cc->contended = COMPACT_CONTENDED_SCHED;
357 return true;
Mel Gormanc67fe372012-08-21 16:16:17 -0700358 }
Mel Gormanc67fe372012-08-21 16:16:17 -0700359 cond_resched();
Mel Gormanc67fe372012-08-21 16:16:17 -0700360 }
361
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700362 return false;
Mel Gormanc67fe372012-08-21 16:16:17 -0700363}
364
Vlastimil Babkabe976572014-06-04 16:10:41 -0700365/*
366 * Aside from avoiding lock contention, compaction also periodically checks
367 * need_resched() and either schedules in sync compaction or aborts async
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700368 * compaction. This is similar to what compact_unlock_should_abort() does, but
Vlastimil Babkabe976572014-06-04 16:10:41 -0700369 * is used where no lock is concerned.
370 *
371 * Returns false when no scheduling was needed, or sync compaction scheduled.
372 * Returns true when async compaction should abort.
373 */
374static inline bool compact_should_abort(struct compact_control *cc)
375{
376 /* async compaction aborts if contended */
377 if (need_resched()) {
378 if (cc->mode == MIGRATE_ASYNC) {
Vlastimil Babka1f9efde2014-10-09 15:27:14 -0700379 cc->contended = COMPACT_CONTENDED_SCHED;
Vlastimil Babkabe976572014-06-04 16:10:41 -0700380 return true;
381 }
382
383 cond_resched();
384 }
385
386 return false;
387}
388
Mel Gormanc67fe372012-08-21 16:16:17 -0700389/*
Jerome Marchand9e4be472013-11-12 15:07:12 -0800390 * Isolate free pages onto a private freelist. If @strict is true, will abort
391 * returning 0 on any invalid PFNs or non-free pages inside of the pageblock
392 * (even though it may still end up isolating some pages).
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100393 */
Mel Gormanf40d1e42012-10-08 16:32:36 -0700394static unsigned long isolate_freepages_block(struct compact_control *cc,
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700395 unsigned long *start_pfn,
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100396 unsigned long end_pfn,
397 struct list_head *freelist,
398 bool strict)
Mel Gorman748446b2010-05-24 14:32:27 -0700399{
Mel Gormanb7aba692011-01-13 15:45:54 -0800400 int nr_scanned = 0, total_isolated = 0;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700401 struct page *cursor, *valid_page = NULL;
Xiubo Lib8b2d822014-10-09 15:28:21 -0700402 unsigned long flags = 0;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700403 bool locked = false;
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700404 unsigned long blockpfn = *start_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -0700405
Mel Gorman748446b2010-05-24 14:32:27 -0700406 cursor = pfn_to_page(blockpfn);
407
Mel Gormanf40d1e42012-10-08 16:32:36 -0700408 /* Isolate free pages. */
Mel Gorman748446b2010-05-24 14:32:27 -0700409 for (; blockpfn < end_pfn; blockpfn++, cursor++) {
410 int isolated, i;
411 struct page *page = cursor;
412
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700413 /*
414 * Periodically drop the lock (if held) regardless of its
415 * contention, to give chance to IRQs. Abort if fatal signal
416 * pending or async compaction detects need_resched()
417 */
418 if (!(blockpfn % SWAP_CLUSTER_MAX)
419 && compact_unlock_should_abort(&cc->zone->lock, flags,
420 &locked, cc))
421 break;
422
Mel Gormanb7aba692011-01-13 15:45:54 -0800423 nr_scanned++;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700424 if (!pfn_valid_within(blockpfn))
Laura Abbott2af120b2014-03-10 15:49:44 -0700425 goto isolate_fail;
426
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700427 if (!valid_page)
428 valid_page = page;
Vlastimil Babka9fcd6d22015-09-08 15:02:49 -0700429
430 /*
431 * For compound pages such as THP and hugetlbfs, we can save
432 * potentially a lot of iterations if we skip them at once.
433 * The check is racy, but we can consider only valid values
434 * and the only danger is skipping too much.
435 */
436 if (PageCompound(page)) {
437 unsigned int comp_order = compound_order(page);
438
439 if (likely(comp_order < MAX_ORDER)) {
440 blockpfn += (1UL << comp_order) - 1;
441 cursor += (1UL << comp_order) - 1;
442 }
443
444 goto isolate_fail;
445 }
446
Mel Gormanf40d1e42012-10-08 16:32:36 -0700447 if (!PageBuddy(page))
Laura Abbott2af120b2014-03-10 15:49:44 -0700448 goto isolate_fail;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700449
450 /*
Vlastimil Babka69b71892014-10-09 15:27:18 -0700451 * If we already hold the lock, we can skip some rechecking.
452 * Note that if we hold the lock now, checked_pageblock was
453 * already set in some previous iteration (or strict is true),
454 * so it is correct to skip the suitable migration target
455 * recheck as well.
Mel Gormanf40d1e42012-10-08 16:32:36 -0700456 */
Vlastimil Babka69b71892014-10-09 15:27:18 -0700457 if (!locked) {
458 /*
459 * The zone lock must be held to isolate freepages.
460 * Unfortunately this is a very coarse lock and can be
461 * heavily contended if there are parallel allocations
462 * or parallel compactions. For async compaction do not
463 * spin on the lock and we acquire the lock as late as
464 * possible.
465 */
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700466 locked = compact_trylock_irqsave(&cc->zone->lock,
467 &flags, cc);
Vlastimil Babka69b71892014-10-09 15:27:18 -0700468 if (!locked)
469 break;
Mel Gormanf40d1e42012-10-08 16:32:36 -0700470
Vlastimil Babka69b71892014-10-09 15:27:18 -0700471 /* Recheck this is a buddy page under lock */
472 if (!PageBuddy(page))
473 goto isolate_fail;
474 }
Mel Gorman748446b2010-05-24 14:32:27 -0700475
476 /* Found a free page, break it into order-0 pages */
477 isolated = split_free_page(page);
478 total_isolated += isolated;
479 for (i = 0; i < isolated; i++) {
480 list_add(&page->lru, freelist);
481 page++;
482 }
483
484 /* If a page was split, advance to the end of it */
485 if (isolated) {
Joonsoo Kim932ff6b2015-02-12 14:59:53 -0800486 cc->nr_freepages += isolated;
487 if (!strict &&
488 cc->nr_migratepages <= cc->nr_freepages) {
489 blockpfn += isolated;
490 break;
491 }
492
Mel Gorman748446b2010-05-24 14:32:27 -0700493 blockpfn += isolated - 1;
494 cursor += isolated - 1;
Laura Abbott2af120b2014-03-10 15:49:44 -0700495 continue;
Mel Gorman748446b2010-05-24 14:32:27 -0700496 }
Laura Abbott2af120b2014-03-10 15:49:44 -0700497
498isolate_fail:
499 if (strict)
500 break;
501 else
502 continue;
503
Mel Gorman748446b2010-05-24 14:32:27 -0700504 }
505
Vlastimil Babka9fcd6d22015-09-08 15:02:49 -0700506 /*
507 * There is a tiny chance that we have read bogus compound_order(),
508 * so be careful to not go outside of the pageblock.
509 */
510 if (unlikely(blockpfn > end_pfn))
511 blockpfn = end_pfn;
512
Joonsoo Kime34d85f2015-02-11 15:27:04 -0800513 trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn,
514 nr_scanned, total_isolated);
515
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700516 /* Record how far we have got within the block */
517 *start_pfn = blockpfn;
518
Mel Gormanf40d1e42012-10-08 16:32:36 -0700519 /*
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.
523 */
Laura Abbott2af120b2014-03-10 15:49:44 -0700524 if (strict && blockpfn < end_pfn)
Mel Gormanf40d1e42012-10-08 16:32:36 -0700525 total_isolated = 0;
526
527 if (locked)
528 spin_unlock_irqrestore(&cc->zone->lock, flags);
529
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700530 /* Update the pageblock-skip if the whole pageblock was scanned */
531 if (blockpfn == end_pfn)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700532 update_pageblock_skip(cc, valid_page, total_isolated, false);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700533
Minchan Kim010fc292012-12-20 15:05:06 -0800534 count_compact_events(COMPACTFREE_SCANNED, nr_scanned);
Mel Gorman397487d2012-10-19 12:00:10 +0100535 if (total_isolated)
Minchan Kim010fc292012-12-20 15:05:06 -0800536 count_compact_events(COMPACTISOLATED, total_isolated);
Mel Gorman748446b2010-05-24 14:32:27 -0700537 return total_isolated;
538}
539
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100540/**
541 * isolate_freepages_range() - isolate free pages.
542 * @start_pfn: The first PFN to start isolating.
543 * @end_pfn: The one-past-last PFN.
544 *
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.
548 *
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
551 * a free page).
552 */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100553unsigned long
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700554isolate_freepages_range(struct compact_control *cc,
555 unsigned long start_pfn, unsigned long end_pfn)
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100556{
Mel Gormanf40d1e42012-10-08 16:32:36 -0700557 unsigned long isolated, pfn, block_end_pfn;
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100558 LIST_HEAD(freelist);
559
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700560 pfn = start_pfn;
561 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100562
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700563 for (; pfn < end_pfn; pfn += isolated,
564 block_end_pfn += pageblock_nr_pages) {
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700565 /* Protect pfn from changing by isolate_freepages_block */
566 unsigned long isolate_start_pfn = pfn;
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700567
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100568 block_end_pfn = min(block_end_pfn, end_pfn);
569
Joonsoo Kim58420012014-11-13 15:19:07 -0800570 /*
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.
574 */
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);
578 }
579
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700580 if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone))
581 break;
582
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700583 isolated = isolate_freepages_block(cc, &isolate_start_pfn,
584 block_end_pfn, &freelist, true);
Michal Nazarewicz85aa1252012-01-30 13:24:03 +0100585
586 /*
587 * In strict mode, isolate_freepages_block() returns 0 if
588 * there are any holes in the block (ie. invalid PFNs or
589 * non-free pages).
590 */
591 if (!isolated)
592 break;
593
594 /*
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).
598 */
599 }
600
601 /* split_free_page does not map the pages */
602 map_pages(&freelist);
603
604 if (pfn < end_pfn) {
605 /* Loop terminated early, cleanup. */
606 release_freepages(&freelist);
607 return 0;
608 }
609
610 /* We don't use freelists for anything. */
611 return pfn;
612}
613
Mel Gorman748446b2010-05-24 14:32:27 -0700614/* Update the number of anon and file isolated pages in the zone */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700615static void acct_isolated(struct zone *zone, struct compact_control *cc)
Mel Gorman748446b2010-05-24 14:32:27 -0700616{
617 struct page *page;
Minchan Kimb9e84ac2011-10-31 17:06:44 -0700618 unsigned int count[2] = { 0, };
Mel Gorman748446b2010-05-24 14:32:27 -0700619
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700620 if (list_empty(&cc->migratepages))
621 return;
622
Minchan Kimb9e84ac2011-10-31 17:06:44 -0700623 list_for_each_entry(page, &cc->migratepages, lru)
624 count[!!page_is_file_cache(page)]++;
Mel Gorman748446b2010-05-24 14:32:27 -0700625
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700626 mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
627 mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
Mel Gorman748446b2010-05-24 14:32:27 -0700628}
629
630/* Similar to reclaim, but different enough that they don't share logic */
631static bool too_many_isolated(struct zone *zone)
632{
Minchan Kimbc693042010-09-09 16:38:00 -0700633 unsigned long active, inactive, isolated;
Mel Gorman748446b2010-05-24 14:32:27 -0700634
635 inactive = zone_page_state(zone, NR_INACTIVE_FILE) +
636 zone_page_state(zone, NR_INACTIVE_ANON);
Minchan Kimbc693042010-09-09 16:38:00 -0700637 active = zone_page_state(zone, NR_ACTIVE_FILE) +
638 zone_page_state(zone, NR_ACTIVE_ANON);
Mel Gorman748446b2010-05-24 14:32:27 -0700639 isolated = zone_page_state(zone, NR_ISOLATED_FILE) +
640 zone_page_state(zone, NR_ISOLATED_ANON);
641
Minchan Kimbc693042010-09-09 16:38:00 -0700642 return isolated > (inactive + active) / 2;
Mel Gorman748446b2010-05-24 14:32:27 -0700643}
644
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100645/**
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700646 * isolate_migratepages_block() - isolate all migrate-able pages within
647 * a single pageblock
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100648 * @cc: Compaction control structure.
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700649 * @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.
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100652 *
653 * Isolate all pages that can be migrated from the range specified by
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700654 * [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
657 * than end_pfn).
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100658 *
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700659 * 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.
Mel Gorman748446b2010-05-24 14:32:27 -0700662 */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700663static unsigned long
664isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn,
665 unsigned long end_pfn, isolate_mode_t isolate_mode)
Mel Gorman748446b2010-05-24 14:32:27 -0700666{
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700667 struct zone *zone = cc->zone;
Mel Gormanb7aba692011-01-13 15:45:54 -0800668 unsigned long nr_scanned = 0, nr_isolated = 0;
Mel Gorman748446b2010-05-24 14:32:27 -0700669 struct list_head *migratelist = &cc->migratepages;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700670 struct lruvec *lruvec;
Xiubo Lib8b2d822014-10-09 15:28:21 -0700671 unsigned long flags = 0;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700672 bool locked = false;
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700673 struct page *page = NULL, *valid_page = NULL;
Joonsoo Kime34d85f2015-02-11 15:27:04 -0800674 unsigned long start_pfn = low_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -0700675
Mel Gorman748446b2010-05-24 14:32:27 -0700676 /*
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
680 */
681 while (unlikely(too_many_isolated(zone))) {
Mel Gormanf9e35b32011-06-15 15:08:52 -0700682 /* async migration should just abort */
David Rientjese0b9dae2014-06-04 16:08:28 -0700683 if (cc->mode == MIGRATE_ASYNC)
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100684 return 0;
Mel Gormanf9e35b32011-06-15 15:08:52 -0700685
Mel Gorman748446b2010-05-24 14:32:27 -0700686 congestion_wait(BLK_RW_ASYNC, HZ/10);
687
688 if (fatal_signal_pending(current))
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100689 return 0;
Mel Gorman748446b2010-05-24 14:32:27 -0700690 }
691
Vlastimil Babkabe976572014-06-04 16:10:41 -0700692 if (compact_should_abort(cc))
693 return 0;
David Rientjesaeef4b82014-06-04 16:08:31 -0700694
Mel Gorman748446b2010-05-24 14:32:27 -0700695 /* Time to isolate some pages for migration */
Mel Gorman748446b2010-05-24 14:32:27 -0700696 for (; low_pfn < end_pfn; low_pfn++) {
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700697 bool is_lru;
698
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700699 /*
700 * Periodically drop the lock (if held) regardless of its
701 * contention, to give chance to IRQs. Abort async compaction
702 * if contended.
703 */
704 if (!(low_pfn % SWAP_CLUSTER_MAX)
705 && compact_unlock_should_abort(&zone->lru_lock, flags,
706 &locked, cc))
707 break;
Mel Gormanc67fe372012-08-21 16:16:17 -0700708
Mel Gorman748446b2010-05-24 14:32:27 -0700709 if (!pfn_valid_within(low_pfn))
710 continue;
Mel Gormanb7aba692011-01-13 15:45:54 -0800711 nr_scanned++;
Mel Gorman748446b2010-05-24 14:32:27 -0700712
Mel Gorman748446b2010-05-24 14:32:27 -0700713 page = pfn_to_page(low_pfn);
Mel Gormandc908602012-02-08 17:13:38 -0800714
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700715 if (!valid_page)
716 valid_page = page;
717
Mel Gorman6c144662014-01-23 15:53:38 -0800718 /*
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700719 * Skip if free. We read page order here without zone lock
720 * which is generally unsafe, but the race window is small and
721 * the worst thing that can happen is that we skip some
722 * potential isolation targets.
Mel Gorman6c144662014-01-23 15:53:38 -0800723 */
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700724 if (PageBuddy(page)) {
725 unsigned long freepage_order = page_order_unsafe(page);
726
727 /*
728 * Without lock, we cannot be sure that what we got is
729 * a valid page order. Consider only values in the
730 * valid order range to prevent low_pfn overflow.
731 */
732 if (freepage_order > 0 && freepage_order < MAX_ORDER)
733 low_pfn += (1UL << freepage_order) - 1;
Mel Gorman748446b2010-05-24 14:32:27 -0700734 continue;
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700735 }
Mel Gorman748446b2010-05-24 14:32:27 -0700736
Mel Gorman9927af742011-01-13 15:45:59 -0800737 /*
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800738 * Check may be lockless but that's ok as we recheck later.
739 * It's possible to migrate LRU pages and balloon pages
740 * Skip any other type of page
741 */
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700742 is_lru = PageLRU(page);
743 if (!is_lru) {
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800744 if (unlikely(balloon_page_movable(page))) {
Konstantin Khlebnikovd6d86c02014-10-09 15:29:27 -0700745 if (balloon_page_isolate(page)) {
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800746 /* Successfully isolated */
Joonsoo Kimb6c75012014-04-07 15:37:07 -0700747 goto isolate_success;
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800748 }
749 }
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800750 }
Andrea Arcangelibc835012011-01-13 15:47:08 -0800751
752 /*
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700753 * Regardless of being on LRU, compound pages such as THP and
754 * hugetlbfs are not to be compacted. We can potentially save
755 * a lot of iterations if we skip them at once. The check is
756 * racy, but we can consider only valid values and the only
757 * danger is skipping too much.
Andrea Arcangelibc835012011-01-13 15:47:08 -0800758 */
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700759 if (PageCompound(page)) {
760 unsigned int comp_order = compound_order(page);
761
762 if (likely(comp_order < MAX_ORDER))
763 low_pfn += (1UL << comp_order) - 1;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700764
Mel Gorman2a1402a2012-10-08 16:32:33 -0700765 continue;
766 }
767
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700768 if (!is_lru)
769 continue;
770
David Rientjes119d6d52014-04-03 14:48:00 -0700771 /*
772 * Migration will fail if an anonymous page is pinned in memory,
773 * so avoid taking lru_lock and isolating it unnecessarily in an
774 * admittedly racy check.
775 */
776 if (!page_mapping(page) &&
777 page_count(page) > page_mapcount(page))
778 continue;
779
Vlastimil Babka69b71892014-10-09 15:27:18 -0700780 /* If we already hold the lock, we can skip some rechecking */
781 if (!locked) {
Vlastimil Babka8b44d272014-10-09 15:27:16 -0700782 locked = compact_trylock_irqsave(&zone->lru_lock,
783 &flags, cc);
Vlastimil Babka69b71892014-10-09 15:27:18 -0700784 if (!locked)
785 break;
Mel Gorman2a1402a2012-10-08 16:32:33 -0700786
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700787 /* Recheck PageLRU and PageCompound under lock */
Vlastimil Babka69b71892014-10-09 15:27:18 -0700788 if (!PageLRU(page))
789 continue;
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700790
791 /*
792 * Page become compound since the non-locked check,
793 * and it's on LRU. It can only be a THP so the order
794 * is safe to read and it's 0 for tail pages.
795 */
796 if (unlikely(PageCompound(page))) {
797 low_pfn += (1UL << compound_order(page)) - 1;
Vlastimil Babka69b71892014-10-09 15:27:18 -0700798 continue;
799 }
Andrea Arcangelibc835012011-01-13 15:47:08 -0800800 }
801
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700802 lruvec = mem_cgroup_page_lruvec(page, zone);
803
Mel Gorman748446b2010-05-24 14:32:27 -0700804 /* Try isolate the page */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700805 if (__isolate_lru_page(page, isolate_mode) != 0)
Mel Gorman748446b2010-05-24 14:32:27 -0700806 continue;
807
Vlastimil Babka29c0dde2015-09-08 15:02:46 -0700808 VM_BUG_ON_PAGE(PageCompound(page), page);
Andrea Arcangelibc835012011-01-13 15:47:08 -0800809
Mel Gorman748446b2010-05-24 14:32:27 -0700810 /* Successfully isolated */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700811 del_page_from_lru_list(page, lruvec, page_lru(page));
Joonsoo Kimb6c75012014-04-07 15:37:07 -0700812
813isolate_success:
Mel Gorman748446b2010-05-24 14:32:27 -0700814 list_add(&page->lru, migratelist);
Mel Gorman748446b2010-05-24 14:32:27 -0700815 cc->nr_migratepages++;
Mel Gormanb7aba692011-01-13 15:45:54 -0800816 nr_isolated++;
Mel Gorman748446b2010-05-24 14:32:27 -0700817
818 /* Avoid isolating too much */
Hillf Danton31b83842012-01-10 15:07:59 -0800819 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) {
820 ++low_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -0700821 break;
Hillf Danton31b83842012-01-10 15:07:59 -0800822 }
Mel Gorman748446b2010-05-24 14:32:27 -0700823 }
824
Vlastimil Babka99c0fd52014-10-09 15:27:23 -0700825 /*
826 * The PageBuddy() check could have potentially brought us outside
827 * the range to be scanned.
828 */
829 if (unlikely(low_pfn > end_pfn))
830 low_pfn = end_pfn;
831
Mel Gormanc67fe372012-08-21 16:16:17 -0700832 if (locked)
833 spin_unlock_irqrestore(&zone->lru_lock, flags);
Mel Gorman748446b2010-05-24 14:32:27 -0700834
Vlastimil Babka50b5b092014-01-21 15:51:10 -0800835 /*
836 * Update the pageblock-skip information and cached scanner pfn,
837 * if the whole pageblock was scanned without isolating any page.
Vlastimil Babka50b5b092014-01-21 15:51:10 -0800838 */
David Rientjes35979ef2014-06-04 16:08:27 -0700839 if (low_pfn == end_pfn)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700840 update_pageblock_skip(cc, valid_page, nr_isolated, true);
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700841
Joonsoo Kime34d85f2015-02-11 15:27:04 -0800842 trace_mm_compaction_isolate_migratepages(start_pfn, low_pfn,
843 nr_scanned, nr_isolated);
Mel Gormanb7aba692011-01-13 15:45:54 -0800844
Minchan Kim010fc292012-12-20 15:05:06 -0800845 count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned);
Mel Gorman397487d2012-10-19 12:00:10 +0100846 if (nr_isolated)
Minchan Kim010fc292012-12-20 15:05:06 -0800847 count_compact_events(COMPACTISOLATED, nr_isolated);
Mel Gorman397487d2012-10-19 12:00:10 +0100848
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +0100849 return low_pfn;
850}
851
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700852/**
853 * isolate_migratepages_range() - isolate migrate-able pages in a PFN range
854 * @cc: Compaction control structure.
855 * @start_pfn: The first PFN to start isolating.
856 * @end_pfn: The one-past-last PFN.
857 *
858 * Returns zero if isolation fails fatally due to e.g. pending signal.
859 * Otherwise, function returns one-past-the-last PFN of isolated page
860 * (which may be greater than end_pfn if end fell in a middle of a THP page).
861 */
862unsigned long
863isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn,
864 unsigned long end_pfn)
865{
866 unsigned long pfn, block_end_pfn;
867
868 /* Scan block by block. First and last block may be incomplete */
869 pfn = start_pfn;
870 block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages);
871
872 for (; pfn < end_pfn; pfn = block_end_pfn,
873 block_end_pfn += pageblock_nr_pages) {
874
875 block_end_pfn = min(block_end_pfn, end_pfn);
876
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700877 if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone))
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700878 continue;
879
880 pfn = isolate_migratepages_block(cc, pfn, block_end_pfn,
881 ISOLATE_UNEVICTABLE);
882
883 /*
884 * In case of fatal failure, release everything that might
885 * have been isolated in the previous iteration, and signal
886 * the failure back to caller.
887 */
888 if (!pfn) {
889 putback_movable_pages(&cc->migratepages);
890 cc->nr_migratepages = 0;
891 break;
892 }
Joonsoo Kim6ea41c02014-10-29 14:50:20 -0700893
894 if (cc->nr_migratepages == COMPACT_CLUSTER_MAX)
895 break;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700896 }
897 acct_isolated(cc->zone, cc);
898
899 return pfn;
900}
901
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100902#endif /* CONFIG_COMPACTION || CONFIG_CMA */
903#ifdef CONFIG_COMPACTION
Andrew Morton018e9a42015-04-15 16:15:20 -0700904
905/* Returns true if the page is within a block suitable for migration to */
906static bool suitable_migration_target(struct page *page)
907{
908 /* If the page is a large free page, then disallow migration */
909 if (PageBuddy(page)) {
910 /*
911 * We are checking page_order without zone->lock taken. But
912 * the only small danger is that we skip a potentially suitable
913 * pageblock, so it's not worth to check order for valid range.
914 */
915 if (page_order_unsafe(page) >= pageblock_order)
916 return false;
917 }
918
919 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
920 if (migrate_async_suitable(get_pageblock_migratetype(page)))
921 return true;
922
923 /* Otherwise skip the block */
924 return false;
925}
926
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100927/*
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -0700928 * Test whether the free scanner has reached the same or lower pageblock than
929 * the migration scanner, and compaction should thus terminate.
930 */
931static inline bool compact_scanners_met(struct compact_control *cc)
932{
933 return (cc->free_pfn >> pageblock_order)
934 <= (cc->migrate_pfn >> pageblock_order);
935}
936
937/*
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100938 * Based on information in the current compact_control, find blocks
939 * suitable for isolating free pages from and then isolate them.
940 */
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700941static void isolate_freepages(struct compact_control *cc)
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100942{
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -0700943 struct zone *zone = cc->zone;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100944 struct page *page;
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700945 unsigned long block_start_pfn; /* start of current pageblock */
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700946 unsigned long isolate_start_pfn; /* exact pfn we start at */
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700947 unsigned long block_end_pfn; /* end of current pageblock */
948 unsigned long low_pfn; /* lowest pfn scanner is able to scan */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100949 struct list_head *freelist = &cc->freepages;
950
951 /*
952 * Initialise the free scanner. The starting point is where we last
Vlastimil Babka49e068f2014-05-06 12:50:03 -0700953 * successfully isolated from, zone-cached value, or the end of the
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700954 * zone when isolating for the first time. For looping we also need
955 * this pfn aligned down to the pageblock boundary, because we do
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700956 * block_start_pfn -= pageblock_nr_pages in the for loop.
957 * For ending point, take care when isolating in last pageblock of a
958 * a zone which ends in the middle of a pageblock.
Vlastimil Babka49e068f2014-05-06 12:50:03 -0700959 * The low boundary is the end of the pageblock the migration scanner
960 * is using.
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100961 */
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700962 isolate_start_pfn = cc->free_pfn;
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700963 block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1);
964 block_end_pfn = min(block_start_pfn + pageblock_nr_pages,
965 zone_end_pfn(zone));
Vlastimil Babka7ed695e2014-01-21 15:51:09 -0800966 low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100967
968 /*
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100969 * Isolate free pages until enough are available to migrate the
970 * pages on cc->migratepages. We stop searching if the migrate
971 * and free page scanners meet or enough free pages are isolated.
972 */
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -0700973 for (; block_start_pfn >= low_pfn;
Vlastimil Babkac96b9e52014-06-04 16:07:26 -0700974 block_end_pfn = block_start_pfn,
Vlastimil Babkae14c7202014-10-09 15:27:20 -0700975 block_start_pfn -= pageblock_nr_pages,
976 isolate_start_pfn = block_start_pfn) {
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100977
David Rientjesf6ea3ad2013-09-30 13:45:03 -0700978 /*
979 * This can iterate a massively long zone without finding any
980 * suitable migration targets, so periodically check if we need
Vlastimil Babkabe976572014-06-04 16:10:41 -0700981 * to schedule, or even abort async compaction.
David Rientjesf6ea3ad2013-09-30 13:45:03 -0700982 */
Vlastimil Babkabe976572014-06-04 16:10:41 -0700983 if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages))
984 && compact_should_abort(cc))
985 break;
David Rientjesf6ea3ad2013-09-30 13:45:03 -0700986
Vlastimil Babka7d49d882014-10-09 15:27:11 -0700987 page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn,
988 zone);
989 if (!page)
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100990 continue;
991
992 /* Check the block is suitable for migration */
Linus Torvalds68e3e922012-06-03 20:05:57 -0700993 if (!suitable_migration_target(page))
Michal Nazarewiczff9543f2011-12-29 13:09:50 +0100994 continue;
Linus Torvalds68e3e922012-06-03 20:05:57 -0700995
Mel Gormanbb13ffe2012-10-08 16:32:41 -0700996 /* If isolation recently failed, do not retry */
997 if (!isolation_suitable(cc, page))
998 continue;
999
Vlastimil Babkae14c7202014-10-09 15:27:20 -07001000 /* Found a block suitable for isolating free pages from. */
Joonsoo Kim932ff6b2015-02-12 14:59:53 -08001001 isolate_freepages_block(cc, &isolate_start_pfn,
Vlastimil Babkac96b9e52014-06-04 16:07:26 -07001002 block_end_pfn, freelist, false);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001003
1004 /*
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001005 * If we isolated enough freepages, or aborted due to async
1006 * compaction being contended, terminate the loop.
Vlastimil Babkae14c7202014-10-09 15:27:20 -07001007 * Remember where the free scanner should restart next time,
1008 * which is where isolate_freepages_block() left off.
1009 * But if it scanned the whole pageblock, isolate_start_pfn
1010 * now points at block_end_pfn, which is the start of the next
1011 * pageblock.
1012 * In that case we will however want to restart at the start
1013 * of the previous pageblock.
1014 */
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001015 if ((cc->nr_freepages >= cc->nr_migratepages)
1016 || cc->contended) {
1017 if (isolate_start_pfn >= block_end_pfn)
1018 isolate_start_pfn =
1019 block_start_pfn - pageblock_nr_pages;
Vlastimil Babkabe976572014-06-04 16:10:41 -07001020 break;
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001021 } else {
1022 /*
1023 * isolate_freepages_block() should not terminate
1024 * prematurely unless contended, or isolated enough
1025 */
1026 VM_BUG_ON(isolate_start_pfn < block_end_pfn);
1027 }
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +01001028 }
1029
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001030 /* split_free_page does not map the pages */
1031 map_pages(freelist);
Michal Nazarewicz2fe86e02012-01-30 13:16:26 +01001032
Vlastimil Babka7ed695e2014-01-21 15:51:09 -08001033 /*
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001034 * Record where the free scanner will restart next time. Either we
1035 * broke from the loop and set isolate_start_pfn based on the last
1036 * call to isolate_freepages_block(), or we met the migration scanner
1037 * and the loop terminated due to isolate_start_pfn < low_pfn
Vlastimil Babka7ed695e2014-01-21 15:51:09 -08001038 */
Vlastimil Babkaf5f61a32015-09-08 15:02:39 -07001039 cc->free_pfn = isolate_start_pfn;
Mel Gorman748446b2010-05-24 14:32:27 -07001040}
1041
1042/*
1043 * This is a migrate-callback that "allocates" freepages by taking pages
1044 * from the isolated freelists in the block we are migrating to.
1045 */
1046static struct page *compaction_alloc(struct page *migratepage,
1047 unsigned long data,
1048 int **result)
1049{
1050 struct compact_control *cc = (struct compact_control *)data;
1051 struct page *freepage;
1052
Vlastimil Babkabe976572014-06-04 16:10:41 -07001053 /*
1054 * Isolate free pages if necessary, and if we are not aborting due to
1055 * contention.
1056 */
Mel Gorman748446b2010-05-24 14:32:27 -07001057 if (list_empty(&cc->freepages)) {
Vlastimil Babkabe976572014-06-04 16:10:41 -07001058 if (!cc->contended)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001059 isolate_freepages(cc);
Mel Gorman748446b2010-05-24 14:32:27 -07001060
1061 if (list_empty(&cc->freepages))
1062 return NULL;
1063 }
1064
1065 freepage = list_entry(cc->freepages.next, struct page, lru);
1066 list_del(&freepage->lru);
1067 cc->nr_freepages--;
1068
1069 return freepage;
1070}
1071
1072/*
David Rientjesd53aea32014-06-04 16:08:26 -07001073 * This is a migrate-callback that "frees" freepages back to the isolated
1074 * freelist. All pages on the freelist are from the same zone, so there is no
1075 * special handling needed for NUMA.
1076 */
1077static void compaction_free(struct page *page, unsigned long data)
1078{
1079 struct compact_control *cc = (struct compact_control *)data;
1080
1081 list_add(&page->lru, &cc->freepages);
1082 cc->nr_freepages++;
1083}
1084
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001085/* possible outcome of isolate_migratepages */
1086typedef enum {
1087 ISOLATE_ABORT, /* Abort compaction now */
1088 ISOLATE_NONE, /* No pages isolated, continue scanning */
1089 ISOLATE_SUCCESS, /* Pages isolated, migrate */
1090} isolate_migrate_t;
1091
1092/*
Eric B Munson5bbe3542015-04-15 16:13:20 -07001093 * Allow userspace to control policy on scanning the unevictable LRU for
1094 * compactable pages.
1095 */
1096int sysctl_compact_unevictable_allowed __read_mostly = 1;
1097
1098/*
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001099 * Isolate all pages that can be migrated from the first suitable block,
1100 * starting at the block pointed to by the migrate scanner pfn within
1101 * compact_control.
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001102 */
1103static isolate_migrate_t isolate_migratepages(struct zone *zone,
1104 struct compact_control *cc)
1105{
1106 unsigned long low_pfn, end_pfn;
Joonsoo Kim1a167182015-09-08 15:03:59 -07001107 unsigned long isolate_start_pfn;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001108 struct page *page;
1109 const isolate_mode_t isolate_mode =
Eric B Munson5bbe3542015-04-15 16:13:20 -07001110 (sysctl_compact_unevictable_allowed ? ISOLATE_UNEVICTABLE : 0) |
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001111 (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001112
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001113 /*
1114 * Start at where we last stopped, or beginning of the zone as
1115 * initialized by compact_zone()
1116 */
1117 low_pfn = cc->migrate_pfn;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001118
1119 /* Only scan within a pageblock boundary */
Mel Gormana9aacbc2013-02-22 16:32:25 -08001120 end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages);
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001121
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001122 /*
1123 * Iterate over whole pageblocks until we find the first suitable.
1124 * Do not cross the free scanner.
1125 */
1126 for (; end_pfn <= cc->free_pfn;
1127 low_pfn = end_pfn, end_pfn += pageblock_nr_pages) {
1128
1129 /*
1130 * This can potentially iterate a massively long zone with
1131 * many pageblocks unsuitable, so periodically check if we
1132 * need to schedule, or even abort async compaction.
1133 */
1134 if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages))
1135 && compact_should_abort(cc))
1136 break;
1137
Vlastimil Babka7d49d882014-10-09 15:27:11 -07001138 page = pageblock_pfn_to_page(low_pfn, end_pfn, zone);
1139 if (!page)
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001140 continue;
1141
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001142 /* If isolation recently failed, do not retry */
1143 if (!isolation_suitable(cc, page))
1144 continue;
1145
1146 /*
1147 * For async compaction, also only scan in MOVABLE blocks.
1148 * Async compaction is optimistic to see if the minimum amount
1149 * of work satisfies the allocation.
1150 */
1151 if (cc->mode == MIGRATE_ASYNC &&
1152 !migrate_async_suitable(get_pageblock_migratetype(page)))
1153 continue;
1154
1155 /* Perform the isolation */
Joonsoo Kim1a167182015-09-08 15:03:59 -07001156 isolate_start_pfn = low_pfn;
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001157 low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn,
1158 isolate_mode);
1159
Hugh Dickinsff599092015-02-12 15:00:28 -08001160 if (!low_pfn || cc->contended) {
1161 acct_isolated(zone, cc);
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001162 return ISOLATE_ABORT;
Hugh Dickinsff599092015-02-12 15:00:28 -08001163 }
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001164
1165 /*
Joonsoo Kim1a167182015-09-08 15:03:59 -07001166 * Record where we could have freed pages by migration and not
1167 * yet flushed them to buddy allocator.
1168 * - this is the lowest page that could have been isolated and
1169 * then freed by migration.
1170 */
1171 if (cc->nr_migratepages && !cc->last_migrated_pfn)
1172 cc->last_migrated_pfn = isolate_start_pfn;
1173
1174 /*
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001175 * Either we isolated something and proceed with migration. Or
1176 * we failed and compact_zone should decide if we should
1177 * continue or not.
1178 */
1179 break;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001180 }
1181
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001182 acct_isolated(zone, cc);
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -07001183 /* Record where migration scanner will be restarted. */
1184 cc->migrate_pfn = low_pfn;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001185
Vlastimil Babkaedc2ca62014-10-09 15:27:09 -07001186 return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE;
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001187}
1188
Yaowei Bai21c527a2015-11-05 18:47:20 -08001189/*
1190 * order == -1 is expected when compacting via
1191 * /proc/sys/vm/compact_memory
1192 */
1193static inline bool is_via_compact_memory(int order)
1194{
1195 return order == -1;
1196}
1197
Joonsoo Kim837d0262015-02-11 15:27:06 -08001198static int __compact_finished(struct zone *zone, struct compact_control *cc,
David Rientjes6d7ce552014-10-09 15:27:27 -07001199 const int migratetype)
Mel Gorman748446b2010-05-24 14:32:27 -07001200{
Mel Gorman8fb74b92013-01-11 14:32:16 -08001201 unsigned int order;
Andrea Arcangeli5a03b052011-01-13 15:47:11 -08001202 unsigned long watermark;
Mel Gorman56de7262010-05-24 14:32:30 -07001203
Vlastimil Babkabe976572014-06-04 16:10:41 -07001204 if (cc->contended || fatal_signal_pending(current))
Vlastimil Babka2d1e1042015-11-05 18:48:02 -08001205 return COMPACT_CONTENDED;
Mel Gorman748446b2010-05-24 14:32:27 -07001206
Mel Gorman753341a2012-10-08 16:32:40 -07001207 /* Compaction run completes if the migrate and free scanner meet */
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -07001208 if (compact_scanners_met(cc)) {
Vlastimil Babka55b7c4c2014-01-21 15:51:11 -08001209 /* Let the next compaction start anew. */
Vlastimil Babka02333642015-09-08 15:02:42 -07001210 reset_cached_positions(zone);
Vlastimil Babka55b7c4c2014-01-21 15:51:11 -08001211
Mel Gorman62997022012-10-08 16:32:47 -07001212 /*
1213 * Mark that the PG_migrate_skip information should be cleared
1214 * by kswapd when it goes to sleep. kswapd does not set the
1215 * flag itself as the decision to be clear should be directly
1216 * based on an allocation request.
1217 */
1218 if (!current_is_kswapd())
1219 zone->compact_blockskip_flush = true;
1220
Mel Gorman748446b2010-05-24 14:32:27 -07001221 return COMPACT_COMPLETE;
Mel Gormanbb13ffe2012-10-08 16:32:41 -07001222 }
Mel Gorman748446b2010-05-24 14:32:27 -07001223
Yaowei Bai21c527a2015-11-05 18:47:20 -08001224 if (is_via_compact_memory(cc->order))
Mel Gorman56de7262010-05-24 14:32:30 -07001225 return COMPACT_CONTINUE;
1226
Michal Hocko3957c772011-06-15 15:08:25 -07001227 /* Compaction run is not finished if the watermark is not met */
1228 watermark = low_wmark_pages(zone);
Michal Hocko3957c772011-06-15 15:08:25 -07001229
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001230 if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx,
1231 cc->alloc_flags))
Michal Hocko3957c772011-06-15 15:08:25 -07001232 return COMPACT_CONTINUE;
1233
Mel Gorman56de7262010-05-24 14:32:30 -07001234 /* Direct compactor: Is a suitable page free? */
Mel Gorman8fb74b92013-01-11 14:32:16 -08001235 for (order = cc->order; order < MAX_ORDER; order++) {
1236 struct free_area *area = &zone->free_area[order];
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001237 bool can_steal;
Mel Gorman56de7262010-05-24 14:32:30 -07001238
Mel Gorman8fb74b92013-01-11 14:32:16 -08001239 /* Job done if page is free of the right migratetype */
David Rientjes6d7ce552014-10-09 15:27:27 -07001240 if (!list_empty(&area->free_list[migratetype]))
Mel Gorman8fb74b92013-01-11 14:32:16 -08001241 return COMPACT_PARTIAL;
1242
Joonsoo Kim2149cda2015-04-14 15:45:21 -07001243#ifdef CONFIG_CMA
1244 /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */
1245 if (migratetype == MIGRATE_MOVABLE &&
1246 !list_empty(&area->free_list[MIGRATE_CMA]))
1247 return COMPACT_PARTIAL;
1248#endif
1249 /*
1250 * Job done if allocation would steal freepages from
1251 * other migratetype buddy lists.
1252 */
1253 if (find_suitable_fallback(area, order, migratetype,
1254 true, &can_steal) != -1)
Mel Gorman8fb74b92013-01-11 14:32:16 -08001255 return COMPACT_PARTIAL;
Mel Gorman56de7262010-05-24 14:32:30 -07001256 }
1257
Joonsoo Kim837d0262015-02-11 15:27:06 -08001258 return COMPACT_NO_SUITABLE_PAGE;
1259}
1260
1261static int compact_finished(struct zone *zone, struct compact_control *cc,
1262 const int migratetype)
1263{
1264 int ret;
1265
1266 ret = __compact_finished(zone, cc, migratetype);
1267 trace_mm_compaction_finished(zone, cc->order, ret);
1268 if (ret == COMPACT_NO_SUITABLE_PAGE)
1269 ret = COMPACT_CONTINUE;
1270
1271 return ret;
Mel Gorman748446b2010-05-24 14:32:27 -07001272}
1273
Mel Gorman3e7d3442011-01-13 15:45:56 -08001274/*
1275 * compaction_suitable: Is this suitable to run compaction on this zone now?
1276 * Returns
1277 * COMPACT_SKIPPED - If there are too few free pages for compaction
1278 * COMPACT_PARTIAL - If the allocation would succeed without compaction
1279 * COMPACT_CONTINUE - If compaction should run now
1280 */
Joonsoo Kim837d0262015-02-11 15:27:06 -08001281static unsigned long __compaction_suitable(struct zone *zone, int order,
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001282 int alloc_flags, int classzone_idx)
Mel Gorman3e7d3442011-01-13 15:45:56 -08001283{
1284 int fragindex;
1285 unsigned long watermark;
1286
Yaowei Bai21c527a2015-11-05 18:47:20 -08001287 if (is_via_compact_memory(order))
Michal Hocko3957c772011-06-15 15:08:25 -07001288 return COMPACT_CONTINUE;
1289
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001290 watermark = low_wmark_pages(zone);
1291 /*
1292 * If watermarks for high-order allocation are already met, there
1293 * should be no need for compaction at all.
1294 */
1295 if (zone_watermark_ok(zone, order, watermark, classzone_idx,
1296 alloc_flags))
1297 return COMPACT_PARTIAL;
1298
Michal Hocko3957c772011-06-15 15:08:25 -07001299 /*
Mel Gorman3e7d3442011-01-13 15:45:56 -08001300 * Watermarks for order-0 must be met for compaction. Note the 2UL.
1301 * This is because during migration, copies of pages need to be
1302 * allocated and for a short time, the footprint is higher
1303 */
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001304 watermark += (2UL << order);
1305 if (!zone_watermark_ok(zone, 0, watermark, classzone_idx, alloc_flags))
Mel Gorman3e7d3442011-01-13 15:45:56 -08001306 return COMPACT_SKIPPED;
1307
1308 /*
1309 * fragmentation index determines if allocation failures are due to
1310 * low memory or external fragmentation
1311 *
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001312 * index of -1000 would imply allocations might succeed depending on
1313 * watermarks, but we already failed the high-order watermark check
Mel Gorman3e7d3442011-01-13 15:45:56 -08001314 * index towards 0 implies failure is due to lack of memory
1315 * index towards 1000 implies failure is due to fragmentation
1316 *
1317 * Only compact if a failure would be due to fragmentation.
1318 */
1319 fragindex = fragmentation_index(zone, order);
1320 if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold)
Joonsoo Kim837d0262015-02-11 15:27:06 -08001321 return COMPACT_NOT_SUITABLE_ZONE;
Mel Gorman3e7d3442011-01-13 15:45:56 -08001322
Mel Gorman3e7d3442011-01-13 15:45:56 -08001323 return COMPACT_CONTINUE;
1324}
1325
Joonsoo Kim837d0262015-02-11 15:27:06 -08001326unsigned long compaction_suitable(struct zone *zone, int order,
1327 int alloc_flags, int classzone_idx)
1328{
1329 unsigned long ret;
1330
1331 ret = __compaction_suitable(zone, order, alloc_flags, classzone_idx);
1332 trace_mm_compaction_suitable(zone, order, ret);
1333 if (ret == COMPACT_NOT_SUITABLE_ZONE)
1334 ret = COMPACT_SKIPPED;
1335
1336 return ret;
1337}
1338
Mel Gorman748446b2010-05-24 14:32:27 -07001339static int compact_zone(struct zone *zone, struct compact_control *cc)
1340{
1341 int ret;
Mel Gormanc89511a2012-10-08 16:32:45 -07001342 unsigned long start_pfn = zone->zone_start_pfn;
Cody P Schafer108bcc92013-02-22 16:35:23 -08001343 unsigned long end_pfn = zone_end_pfn(zone);
David Rientjes6d7ce552014-10-09 15:27:27 -07001344 const int migratetype = gfpflags_to_migratetype(cc->gfp_mask);
David Rientjese0b9dae2014-06-04 16:08:28 -07001345 const bool sync = cc->mode != MIGRATE_ASYNC;
Mel Gorman748446b2010-05-24 14:32:27 -07001346
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001347 ret = compaction_suitable(zone, cc->order, cc->alloc_flags,
1348 cc->classzone_idx);
Mel Gorman3e7d3442011-01-13 15:45:56 -08001349 switch (ret) {
1350 case COMPACT_PARTIAL:
1351 case COMPACT_SKIPPED:
1352 /* Compaction is likely to fail */
1353 return ret;
1354 case COMPACT_CONTINUE:
1355 /* Fall through to compaction */
1356 ;
1357 }
1358
Mel Gormanc89511a2012-10-08 16:32:45 -07001359 /*
Vlastimil Babkad3132e42014-01-21 15:51:08 -08001360 * Clear pageblock skip if there were failures recently and compaction
1361 * is about to be retried after being deferred. kswapd does not do
1362 * this reset as it'll reset the cached information when going to sleep.
1363 */
1364 if (compaction_restarting(zone, cc->order) && !current_is_kswapd())
1365 __reset_isolation_suitable(zone);
1366
1367 /*
Mel Gormanc89511a2012-10-08 16:32:45 -07001368 * Setup to move all movable pages to the end of the zone. Used cached
1369 * information on where the scanners should start but check that it
1370 * is initialised by ensuring the values are within zone boundaries.
1371 */
David Rientjese0b9dae2014-06-04 16:08:28 -07001372 cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync];
Mel Gormanc89511a2012-10-08 16:32:45 -07001373 cc->free_pfn = zone->compact_cached_free_pfn;
1374 if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) {
1375 cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1);
1376 zone->compact_cached_free_pfn = cc->free_pfn;
1377 }
1378 if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) {
1379 cc->migrate_pfn = start_pfn;
David Rientjes35979ef2014-06-04 16:08:27 -07001380 zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn;
1381 zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn;
Mel Gormanc89511a2012-10-08 16:32:45 -07001382 }
Joonsoo Kim1a167182015-09-08 15:03:59 -07001383 cc->last_migrated_pfn = 0;
Mel Gorman748446b2010-05-24 14:32:27 -07001384
Joonsoo Kim16c4a092015-02-11 15:27:01 -08001385 trace_mm_compaction_begin(start_pfn, cc->migrate_pfn,
1386 cc->free_pfn, end_pfn, sync);
Mel Gorman0eb927c2014-01-21 15:51:05 -08001387
Mel Gorman748446b2010-05-24 14:32:27 -07001388 migrate_prep_local();
1389
David Rientjes6d7ce552014-10-09 15:27:27 -07001390 while ((ret = compact_finished(zone, cc, migratetype)) ==
1391 COMPACT_CONTINUE) {
Minchan Kim9d502c12011-03-22 16:30:39 -07001392 int err;
Mel Gorman748446b2010-05-24 14:32:27 -07001393
Mel Gormanf9e35b32011-06-15 15:08:52 -07001394 switch (isolate_migratepages(zone, cc)) {
1395 case ISOLATE_ABORT:
Vlastimil Babka2d1e1042015-11-05 18:48:02 -08001396 ret = COMPACT_CONTENDED;
Rafael Aquini5733c7d2012-12-11 16:02:47 -08001397 putback_movable_pages(&cc->migratepages);
Shaohua Lie64c5232012-10-08 16:32:27 -07001398 cc->nr_migratepages = 0;
Mel Gormanf9e35b32011-06-15 15:08:52 -07001399 goto out;
1400 case ISOLATE_NONE:
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001401 /*
1402 * We haven't isolated and migrated anything, but
1403 * there might still be unflushed migrations from
1404 * previous cc->order aligned block.
1405 */
1406 goto check_drain;
Mel Gormanf9e35b32011-06-15 15:08:52 -07001407 case ISOLATE_SUCCESS:
1408 ;
1409 }
Mel Gorman748446b2010-05-24 14:32:27 -07001410
David Rientjesd53aea32014-06-04 16:08:26 -07001411 err = migrate_pages(&cc->migratepages, compaction_alloc,
David Rientjese0b9dae2014-06-04 16:08:28 -07001412 compaction_free, (unsigned long)cc, cc->mode,
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001413 MR_COMPACTION);
Mel Gorman748446b2010-05-24 14:32:27 -07001414
Vlastimil Babkaf8c93012014-06-04 16:08:32 -07001415 trace_mm_compaction_migratepages(cc->nr_migratepages, err,
1416 &cc->migratepages);
Mel Gorman748446b2010-05-24 14:32:27 -07001417
Vlastimil Babkaf8c93012014-06-04 16:08:32 -07001418 /* All pages were either migrated or will be released */
1419 cc->nr_migratepages = 0;
Minchan Kim9d502c12011-03-22 16:30:39 -07001420 if (err) {
Rafael Aquini5733c7d2012-12-11 16:02:47 -08001421 putback_movable_pages(&cc->migratepages);
Vlastimil Babka7ed695e2014-01-21 15:51:09 -08001422 /*
1423 * migrate_pages() may return -ENOMEM when scanners meet
1424 * and we want compact_finished() to detect it
1425 */
Vlastimil Babkaf2849aa2015-09-08 15:02:36 -07001426 if (err == -ENOMEM && !compact_scanners_met(cc)) {
Vlastimil Babka2d1e1042015-11-05 18:48:02 -08001427 ret = COMPACT_CONTENDED;
David Rientjes4bf2bba2012-07-11 14:02:13 -07001428 goto out;
1429 }
Mel Gorman748446b2010-05-24 14:32:27 -07001430 }
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001431
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001432check_drain:
1433 /*
1434 * Has the migration scanner moved away from the previous
1435 * cc->order aligned block where we migrated from? If yes,
1436 * flush the pages that were freed, so that they can merge and
1437 * compact_finished() can detect immediately if allocation
1438 * would succeed.
1439 */
Joonsoo Kim1a167182015-09-08 15:03:59 -07001440 if (cc->order > 0 && cc->last_migrated_pfn) {
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001441 int cpu;
1442 unsigned long current_block_start =
1443 cc->migrate_pfn & ~((1UL << cc->order) - 1);
1444
Joonsoo Kim1a167182015-09-08 15:03:59 -07001445 if (cc->last_migrated_pfn < current_block_start) {
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001446 cpu = get_cpu();
1447 lru_add_drain_cpu(cpu);
1448 drain_local_pages(zone);
1449 put_cpu();
1450 /* No more flushing until we migrate again */
Joonsoo Kim1a167182015-09-08 15:03:59 -07001451 cc->last_migrated_pfn = 0;
Vlastimil Babkafdaf7f52014-12-10 15:43:34 -08001452 }
1453 }
1454
Mel Gorman748446b2010-05-24 14:32:27 -07001455 }
1456
Mel Gormanf9e35b32011-06-15 15:08:52 -07001457out:
Vlastimil Babka6bace092014-12-10 15:43:31 -08001458 /*
1459 * Release free pages and update where the free scanner should restart,
1460 * so we don't leave any returned pages behind in the next attempt.
1461 */
1462 if (cc->nr_freepages > 0) {
1463 unsigned long free_pfn = release_freepages(&cc->freepages);
1464
1465 cc->nr_freepages = 0;
1466 VM_BUG_ON(free_pfn == 0);
1467 /* The cached pfn is always the first in a pageblock */
1468 free_pfn &= ~(pageblock_nr_pages-1);
1469 /*
1470 * Only go back, not forward. The cached pfn might have been
1471 * already reset to zone end in compact_finished()
1472 */
1473 if (free_pfn > zone->compact_cached_free_pfn)
1474 zone->compact_cached_free_pfn = free_pfn;
1475 }
Mel Gorman748446b2010-05-24 14:32:27 -07001476
Joonsoo Kim16c4a092015-02-11 15:27:01 -08001477 trace_mm_compaction_end(start_pfn, cc->migrate_pfn,
1478 cc->free_pfn, end_pfn, sync, ret);
Mel Gorman0eb927c2014-01-21 15:51:05 -08001479
Vlastimil Babka2d1e1042015-11-05 18:48:02 -08001480 if (ret == COMPACT_CONTENDED)
1481 ret = COMPACT_PARTIAL;
1482
Mel Gorman748446b2010-05-24 14:32:27 -07001483 return ret;
1484}
Mel Gorman76ab0f52010-05-24 14:32:28 -07001485
David Rientjese0b9dae2014-06-04 16:08:28 -07001486static unsigned long compact_zone_order(struct zone *zone, int order,
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001487 gfp_t gfp_mask, enum migrate_mode mode, int *contended,
1488 int alloc_flags, int classzone_idx)
Mel Gorman56de7262010-05-24 14:32:30 -07001489{
Shaohua Lie64c5232012-10-08 16:32:27 -07001490 unsigned long ret;
Mel Gorman56de7262010-05-24 14:32:30 -07001491 struct compact_control cc = {
1492 .nr_freepages = 0,
1493 .nr_migratepages = 0,
1494 .order = order,
David Rientjes6d7ce552014-10-09 15:27:27 -07001495 .gfp_mask = gfp_mask,
Mel Gorman56de7262010-05-24 14:32:30 -07001496 .zone = zone,
David Rientjese0b9dae2014-06-04 16:08:28 -07001497 .mode = mode,
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001498 .alloc_flags = alloc_flags,
1499 .classzone_idx = classzone_idx,
Mel Gorman56de7262010-05-24 14:32:30 -07001500 };
1501 INIT_LIST_HEAD(&cc.freepages);
1502 INIT_LIST_HEAD(&cc.migratepages);
1503
Shaohua Lie64c5232012-10-08 16:32:27 -07001504 ret = compact_zone(zone, &cc);
1505
1506 VM_BUG_ON(!list_empty(&cc.freepages));
1507 VM_BUG_ON(!list_empty(&cc.migratepages));
1508
1509 *contended = cc.contended;
1510 return ret;
Mel Gorman56de7262010-05-24 14:32:30 -07001511}
1512
Mel Gorman5e771902010-05-24 14:32:31 -07001513int sysctl_extfrag_threshold = 500;
1514
Mel Gorman56de7262010-05-24 14:32:30 -07001515/**
1516 * try_to_compact_pages - Direct compact to satisfy a high-order allocation
Mel Gorman56de7262010-05-24 14:32:30 -07001517 * @gfp_mask: The GFP mask of the current allocation
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001518 * @order: The order of the current allocation
1519 * @alloc_flags: The allocation flags of the current allocation
1520 * @ac: The context of current allocation
David Rientjese0b9dae2014-06-04 16:08:28 -07001521 * @mode: The migration mode for async, sync light, or sync migration
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001522 * @contended: Return value that determines if compaction was aborted due to
1523 * need_resched() or lock contention
Mel Gorman56de7262010-05-24 14:32:30 -07001524 *
1525 * This is the main entry point for direct page compaction.
1526 */
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001527unsigned long try_to_compact_pages(gfp_t gfp_mask, unsigned int order,
1528 int alloc_flags, const struct alloc_context *ac,
1529 enum migrate_mode mode, int *contended)
Mel Gorman56de7262010-05-24 14:32:30 -07001530{
Mel Gorman56de7262010-05-24 14:32:30 -07001531 int may_enter_fs = gfp_mask & __GFP_FS;
1532 int may_perform_io = gfp_mask & __GFP_IO;
Mel Gorman56de7262010-05-24 14:32:30 -07001533 struct zoneref *z;
1534 struct zone *zone;
Vlastimil Babka53853e22014-10-09 15:27:02 -07001535 int rc = COMPACT_DEFERRED;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001536 int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */
1537
1538 *contended = COMPACT_CONTENDED_NONE;
Mel Gorman56de7262010-05-24 14:32:30 -07001539
Mel Gorman4ffb6332012-10-08 16:29:09 -07001540 /* Check if the GFP flags allow compaction */
Andrea Arcangelic5a73c32011-01-13 15:47:11 -08001541 if (!order || !may_enter_fs || !may_perform_io)
Vlastimil Babka53853e22014-10-09 15:27:02 -07001542 return COMPACT_SKIPPED;
Mel Gorman56de7262010-05-24 14:32:30 -07001543
Joonsoo Kim837d0262015-02-11 15:27:06 -08001544 trace_mm_compaction_try_to_compact_pages(order, gfp_mask, mode);
1545
Mel Gorman56de7262010-05-24 14:32:30 -07001546 /* Compact each zone in the list */
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001547 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
1548 ac->nodemask) {
Mel Gorman56de7262010-05-24 14:32:30 -07001549 int status;
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001550 int zone_contended;
Mel Gorman56de7262010-05-24 14:32:30 -07001551
Vlastimil Babka53853e22014-10-09 15:27:02 -07001552 if (compaction_deferred(zone, order))
1553 continue;
1554
David Rientjese0b9dae2014-06-04 16:08:28 -07001555 status = compact_zone_order(zone, order, gfp_mask, mode,
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001556 &zone_contended, alloc_flags,
1557 ac->classzone_idx);
Mel Gorman56de7262010-05-24 14:32:30 -07001558 rc = max(status, rc);
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001559 /*
1560 * It takes at least one zone that wasn't lock contended
1561 * to clear all_zones_contended.
1562 */
1563 all_zones_contended &= zone_contended;
Mel Gorman56de7262010-05-24 14:32:30 -07001564
Mel Gorman3e7d3442011-01-13 15:45:56 -08001565 /* If a normal allocation would succeed, stop compacting */
Vlastimil Babkaebff3982014-12-10 15:43:22 -08001566 if (zone_watermark_ok(zone, order, low_wmark_pages(zone),
Vlastimil Babka1a6d53a2015-02-11 15:25:44 -08001567 ac->classzone_idx, alloc_flags)) {
Vlastimil Babka53853e22014-10-09 15:27:02 -07001568 /*
1569 * We think the allocation will succeed in this zone,
1570 * but it is not certain, hence the false. The caller
1571 * will repeat this with true if allocation indeed
1572 * succeeds in this zone.
1573 */
1574 compaction_defer_reset(zone, order, false);
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001575 /*
1576 * It is possible that async compaction aborted due to
1577 * need_resched() and the watermarks were ok thanks to
1578 * somebody else freeing memory. The allocation can
1579 * however still fail so we better signal the
1580 * need_resched() contention anyway (this will not
1581 * prevent the allocation attempt).
1582 */
1583 if (zone_contended == COMPACT_CONTENDED_SCHED)
1584 *contended = COMPACT_CONTENDED_SCHED;
1585
1586 goto break_loop;
1587 }
1588
Vlastimil Babkaf8669792014-12-10 15:43:28 -08001589 if (mode != MIGRATE_ASYNC && status == COMPACT_COMPLETE) {
Vlastimil Babka53853e22014-10-09 15:27:02 -07001590 /*
1591 * We think that allocation won't succeed in this zone
1592 * so we defer compaction there. If it ends up
1593 * succeeding after all, it will be reset.
1594 */
1595 defer_compaction(zone, order);
1596 }
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001597
1598 /*
1599 * We might have stopped compacting due to need_resched() in
1600 * async compaction, or due to a fatal signal detected. In that
1601 * case do not try further zones and signal need_resched()
1602 * contention.
1603 */
1604 if ((zone_contended == COMPACT_CONTENDED_SCHED)
1605 || fatal_signal_pending(current)) {
1606 *contended = COMPACT_CONTENDED_SCHED;
1607 goto break_loop;
1608 }
1609
1610 continue;
1611break_loop:
1612 /*
1613 * We might not have tried all the zones, so be conservative
1614 * and assume they are not all lock contended.
1615 */
1616 all_zones_contended = 0;
1617 break;
Mel Gorman56de7262010-05-24 14:32:30 -07001618 }
1619
Vlastimil Babka1f9efde2014-10-09 15:27:14 -07001620 /*
1621 * If at least one zone wasn't deferred or skipped, we report if all
1622 * zones that were tried were lock contended.
1623 */
1624 if (rc > COMPACT_SKIPPED && all_zones_contended)
1625 *contended = COMPACT_CONTENDED_LOCK;
1626
Mel Gorman56de7262010-05-24 14:32:30 -07001627 return rc;
1628}
1629
1630
Mel Gorman76ab0f52010-05-24 14:32:28 -07001631/* Compact all zones within a node */
Andrew Morton7103f162013-02-22 16:32:33 -08001632static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc)
Mel Gorman76ab0f52010-05-24 14:32:28 -07001633{
1634 int zoneid;
Mel Gorman76ab0f52010-05-24 14:32:28 -07001635 struct zone *zone;
1636
Mel Gorman76ab0f52010-05-24 14:32:28 -07001637 for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
Mel Gorman76ab0f52010-05-24 14:32:28 -07001638
1639 zone = &pgdat->node_zones[zoneid];
1640 if (!populated_zone(zone))
1641 continue;
1642
Rik van Riel7be62de2012-03-21 16:33:52 -07001643 cc->nr_freepages = 0;
1644 cc->nr_migratepages = 0;
1645 cc->zone = zone;
1646 INIT_LIST_HEAD(&cc->freepages);
1647 INIT_LIST_HEAD(&cc->migratepages);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001648
Gioh Kim195b0c62015-04-15 16:13:33 -07001649 /*
1650 * When called via /proc/sys/vm/compact_memory
1651 * this makes sure we compact the whole zone regardless of
1652 * cached scanner positions.
1653 */
Yaowei Bai21c527a2015-11-05 18:47:20 -08001654 if (is_via_compact_memory(cc->order))
Gioh Kim195b0c62015-04-15 16:13:33 -07001655 __reset_isolation_suitable(zone);
1656
Yaowei Bai21c527a2015-11-05 18:47:20 -08001657 if (is_via_compact_memory(cc->order) ||
1658 !compaction_deferred(zone, cc->order))
Rik van Riel7be62de2012-03-21 16:33:52 -07001659 compact_zone(zone, cc);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001660
Rik van Riel7be62de2012-03-21 16:33:52 -07001661 VM_BUG_ON(!list_empty(&cc->freepages));
1662 VM_BUG_ON(!list_empty(&cc->migratepages));
Joonsoo Kim75469342016-01-14 15:20:48 -08001663
1664 if (is_via_compact_memory(cc->order))
1665 continue;
1666
1667 if (zone_watermark_ok(zone, cc->order,
1668 low_wmark_pages(zone), 0, 0))
1669 compaction_defer_reset(zone, cc->order, false);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001670 }
Mel Gorman76ab0f52010-05-24 14:32:28 -07001671}
1672
Andrew Morton7103f162013-02-22 16:32:33 -08001673void compact_pgdat(pg_data_t *pgdat, int order)
Rik van Riel7be62de2012-03-21 16:33:52 -07001674{
1675 struct compact_control cc = {
1676 .order = order,
David Rientjese0b9dae2014-06-04 16:08:28 -07001677 .mode = MIGRATE_ASYNC,
Rik van Riel7be62de2012-03-21 16:33:52 -07001678 };
1679
Mel Gorman3a7200a2013-09-11 14:22:19 -07001680 if (!order)
1681 return;
1682
Andrew Morton7103f162013-02-22 16:32:33 -08001683 __compact_pgdat(pgdat, &cc);
Rik van Riel7be62de2012-03-21 16:33:52 -07001684}
1685
Andrew Morton7103f162013-02-22 16:32:33 -08001686static void compact_node(int nid)
Rik van Riel7be62de2012-03-21 16:33:52 -07001687{
Rik van Riel7be62de2012-03-21 16:33:52 -07001688 struct compact_control cc = {
1689 .order = -1,
David Rientjese0b9dae2014-06-04 16:08:28 -07001690 .mode = MIGRATE_SYNC,
David Rientjes91ca9182014-04-03 14:47:23 -07001691 .ignore_skip_hint = true,
Rik van Riel7be62de2012-03-21 16:33:52 -07001692 };
1693
Andrew Morton7103f162013-02-22 16:32:33 -08001694 __compact_pgdat(NODE_DATA(nid), &cc);
Rik van Riel7be62de2012-03-21 16:33:52 -07001695}
1696
Mel Gorman76ab0f52010-05-24 14:32:28 -07001697/* Compact all nodes in the system */
Jason Liu7964c062013-01-11 14:31:47 -08001698static void compact_nodes(void)
Mel Gorman76ab0f52010-05-24 14:32:28 -07001699{
1700 int nid;
1701
Hugh Dickins8575ec22012-03-21 16:33:53 -07001702 /* Flush pending updates to the LRU lists */
1703 lru_add_drain_all();
1704
Mel Gorman76ab0f52010-05-24 14:32:28 -07001705 for_each_online_node(nid)
1706 compact_node(nid);
Mel Gorman76ab0f52010-05-24 14:32:28 -07001707}
1708
1709/* The written value is actually unused, all memory is compacted */
1710int sysctl_compact_memory;
1711
Yaowei Baifec4eb22016-01-14 15:20:09 -08001712/*
1713 * This is the entry point for compacting all nodes via
1714 * /proc/sys/vm/compact_memory
1715 */
Mel Gorman76ab0f52010-05-24 14:32:28 -07001716int sysctl_compaction_handler(struct ctl_table *table, int write,
1717 void __user *buffer, size_t *length, loff_t *ppos)
1718{
1719 if (write)
Jason Liu7964c062013-01-11 14:31:47 -08001720 compact_nodes();
Mel Gorman76ab0f52010-05-24 14:32:28 -07001721
1722 return 0;
1723}
Mel Gormaned4a6d72010-05-24 14:32:29 -07001724
Mel Gorman5e771902010-05-24 14:32:31 -07001725int sysctl_extfrag_handler(struct ctl_table *table, int write,
1726 void __user *buffer, size_t *length, loff_t *ppos)
1727{
1728 proc_dointvec_minmax(table, write, buffer, length, ppos);
1729
1730 return 0;
1731}
1732
Mel Gormaned4a6d72010-05-24 14:32:29 -07001733#if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
Rashika Kheria74e77fb2014-04-03 14:48:01 -07001734static ssize_t sysfs_compact_node(struct device *dev,
Kay Sievers10fbcf42011-12-21 14:48:43 -08001735 struct device_attribute *attr,
Mel Gormaned4a6d72010-05-24 14:32:29 -07001736 const char *buf, size_t count)
1737{
Hugh Dickins8575ec22012-03-21 16:33:53 -07001738 int nid = dev->id;
1739
1740 if (nid >= 0 && nid < nr_node_ids && node_online(nid)) {
1741 /* Flush pending updates to the LRU lists */
1742 lru_add_drain_all();
1743
1744 compact_node(nid);
1745 }
Mel Gormaned4a6d72010-05-24 14:32:29 -07001746
1747 return count;
1748}
Kay Sievers10fbcf42011-12-21 14:48:43 -08001749static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001750
1751int compaction_register_node(struct node *node)
1752{
Kay Sievers10fbcf42011-12-21 14:48:43 -08001753 return device_create_file(&node->dev, &dev_attr_compact);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001754}
1755
1756void compaction_unregister_node(struct node *node)
1757{
Kay Sievers10fbcf42011-12-21 14:48:43 -08001758 return device_remove_file(&node->dev, &dev_attr_compact);
Mel Gormaned4a6d72010-05-24 14:32:29 -07001759}
1760#endif /* CONFIG_SYSFS && CONFIG_NUMA */
Michal Nazarewiczff9543f2011-12-29 13:09:50 +01001761
1762#endif /* CONFIG_COMPACTION */