Commit | Line | Data |
---|---|---|
748446bb MG |
1 | /* |
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> | |
76ab0f53 | 15 | #include <linux/sysctl.h> |
ed4a6d7f | 16 | #include <linux/sysfs.h> |
bf6bddf1 | 17 | #include <linux/balloon_compaction.h> |
194159fb | 18 | #include <linux/page-isolation.h> |
748446bb MG |
19 | #include "internal.h" |
20 | ||
010fc29a MK |
21 | #ifdef CONFIG_COMPACTION |
22 | static inline void count_compact_event(enum vm_event_item item) | |
23 | { | |
24 | count_vm_event(item); | |
25 | } | |
26 | ||
27 | static inline void count_compact_events(enum vm_event_item item, long delta) | |
28 | { | |
29 | count_vm_events(item, delta); | |
30 | } | |
31 | #else | |
32 | #define count_compact_event(item) do { } while (0) | |
33 | #define count_compact_events(item, delta) do { } while (0) | |
34 | #endif | |
35 | ||
ff9543fd MN |
36 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
37 | ||
b7aba698 MG |
38 | #define CREATE_TRACE_POINTS |
39 | #include <trace/events/compaction.h> | |
40 | ||
748446bb MG |
41 | static unsigned long release_freepages(struct list_head *freelist) |
42 | { | |
43 | struct page *page, *next; | |
44 | unsigned long count = 0; | |
45 | ||
46 | list_for_each_entry_safe(page, next, freelist, lru) { | |
47 | list_del(&page->lru); | |
48 | __free_page(page); | |
49 | count++; | |
50 | } | |
51 | ||
52 | return count; | |
53 | } | |
54 | ||
ff9543fd MN |
55 | static void map_pages(struct list_head *list) |
56 | { | |
57 | struct page *page; | |
58 | ||
59 | list_for_each_entry(page, list, lru) { | |
60 | arch_alloc_page(page, 0); | |
61 | kernel_map_pages(page, 1, 1); | |
62 | } | |
63 | } | |
64 | ||
47118af0 MN |
65 | static inline bool migrate_async_suitable(int migratetype) |
66 | { | |
67 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; | |
68 | } | |
69 | ||
7d49d886 VB |
70 | /* |
71 | * Check that the whole (or subset of) a pageblock given by the interval of | |
72 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
73 | * with the migration of free compaction scanner. The scanners then need to | |
74 | * use only pfn_valid_within() check for arches that allow holes within | |
75 | * pageblocks. | |
76 | * | |
77 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
78 | * | |
79 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
80 | * i.e. it's possible that all pages within a zones range of pages do not | |
81 | * belong to a single zone. We assume that a border between node0 and node1 | |
82 | * can occur within a single pageblock, but not a node0 node1 node0 | |
83 | * interleaving within a single pageblock. It is therefore sufficient to check | |
84 | * the first and last page of a pageblock and avoid checking each individual | |
85 | * page in a pageblock. | |
86 | */ | |
87 | static struct page *pageblock_pfn_to_page(unsigned long start_pfn, | |
88 | unsigned long end_pfn, struct zone *zone) | |
89 | { | |
90 | struct page *start_page; | |
91 | struct page *end_page; | |
92 | ||
93 | /* end_pfn is one past the range we are checking */ | |
94 | end_pfn--; | |
95 | ||
96 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
97 | return NULL; | |
98 | ||
99 | start_page = pfn_to_page(start_pfn); | |
100 | ||
101 | if (page_zone(start_page) != zone) | |
102 | return NULL; | |
103 | ||
104 | end_page = pfn_to_page(end_pfn); | |
105 | ||
106 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
107 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
108 | return NULL; | |
109 | ||
110 | return start_page; | |
111 | } | |
112 | ||
bb13ffeb MG |
113 | #ifdef CONFIG_COMPACTION |
114 | /* Returns true if the pageblock should be scanned for pages to isolate. */ | |
115 | static inline bool isolation_suitable(struct compact_control *cc, | |
116 | struct page *page) | |
117 | { | |
118 | if (cc->ignore_skip_hint) | |
119 | return true; | |
120 | ||
121 | return !get_pageblock_skip(page); | |
122 | } | |
123 | ||
124 | /* | |
125 | * This function is called to clear all cached information on pageblocks that | |
126 | * should be skipped for page isolation when the migrate and free page scanner | |
127 | * meet. | |
128 | */ | |
62997027 | 129 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
130 | { |
131 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 132 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
133 | unsigned long pfn; |
134 | ||
35979ef3 DR |
135 | zone->compact_cached_migrate_pfn[0] = start_pfn; |
136 | zone->compact_cached_migrate_pfn[1] = start_pfn; | |
c89511ab | 137 | zone->compact_cached_free_pfn = end_pfn; |
62997027 | 138 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
139 | |
140 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
141 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
142 | struct page *page; | |
143 | ||
144 | cond_resched(); | |
145 | ||
146 | if (!pfn_valid(pfn)) | |
147 | continue; | |
148 | ||
149 | page = pfn_to_page(pfn); | |
150 | if (zone != page_zone(page)) | |
151 | continue; | |
152 | ||
153 | clear_pageblock_skip(page); | |
154 | } | |
155 | } | |
156 | ||
62997027 MG |
157 | void reset_isolation_suitable(pg_data_t *pgdat) |
158 | { | |
159 | int zoneid; | |
160 | ||
161 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
162 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
163 | if (!populated_zone(zone)) | |
164 | continue; | |
165 | ||
166 | /* Only flush if a full compaction finished recently */ | |
167 | if (zone->compact_blockskip_flush) | |
168 | __reset_isolation_suitable(zone); | |
169 | } | |
170 | } | |
171 | ||
bb13ffeb MG |
172 | /* |
173 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 174 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 175 | */ |
c89511ab MG |
176 | static void update_pageblock_skip(struct compact_control *cc, |
177 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 178 | bool migrate_scanner) |
bb13ffeb | 179 | { |
c89511ab | 180 | struct zone *zone = cc->zone; |
35979ef3 | 181 | unsigned long pfn; |
6815bf3f JK |
182 | |
183 | if (cc->ignore_skip_hint) | |
184 | return; | |
185 | ||
bb13ffeb MG |
186 | if (!page) |
187 | return; | |
188 | ||
35979ef3 DR |
189 | if (nr_isolated) |
190 | return; | |
191 | ||
edc2ca61 | 192 | set_pageblock_skip(page); |
c89511ab | 193 | |
35979ef3 DR |
194 | pfn = page_to_pfn(page); |
195 | ||
196 | /* Update where async and sync compaction should restart */ | |
197 | if (migrate_scanner) { | |
198 | if (cc->finished_update_migrate) | |
199 | return; | |
200 | if (pfn > zone->compact_cached_migrate_pfn[0]) | |
201 | zone->compact_cached_migrate_pfn[0] = pfn; | |
e0b9daeb DR |
202 | if (cc->mode != MIGRATE_ASYNC && |
203 | pfn > zone->compact_cached_migrate_pfn[1]) | |
35979ef3 DR |
204 | zone->compact_cached_migrate_pfn[1] = pfn; |
205 | } else { | |
206 | if (cc->finished_update_free) | |
207 | return; | |
208 | if (pfn < zone->compact_cached_free_pfn) | |
209 | zone->compact_cached_free_pfn = pfn; | |
c89511ab | 210 | } |
bb13ffeb MG |
211 | } |
212 | #else | |
213 | static inline bool isolation_suitable(struct compact_control *cc, | |
214 | struct page *page) | |
215 | { | |
216 | return true; | |
217 | } | |
218 | ||
c89511ab MG |
219 | static void update_pageblock_skip(struct compact_control *cc, |
220 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 221 | bool migrate_scanner) |
bb13ffeb MG |
222 | { |
223 | } | |
224 | #endif /* CONFIG_COMPACTION */ | |
225 | ||
1f9efdef | 226 | static int should_release_lock(spinlock_t *lock) |
2a1402aa | 227 | { |
1f9efdef VB |
228 | /* |
229 | * Sched contention has higher priority here as we may potentially | |
230 | * have to abort whole compaction ASAP. Returning with lock contention | |
231 | * means we will try another zone, and further decisions are | |
232 | * influenced only when all zones are lock contended. That means | |
233 | * potentially missing a lock contention is less critical. | |
234 | */ | |
235 | if (need_resched()) | |
236 | return COMPACT_CONTENDED_SCHED; | |
237 | else if (spin_is_contended(lock)) | |
238 | return COMPACT_CONTENDED_LOCK; | |
239 | ||
240 | return COMPACT_CONTENDED_NONE; | |
2a1402aa MG |
241 | } |
242 | ||
c67fe375 MG |
243 | /* |
244 | * Compaction requires the taking of some coarse locks that are potentially | |
245 | * very heavily contended. Check if the process needs to be scheduled or | |
246 | * if the lock is contended. For async compaction, back out in the event | |
247 | * if contention is severe. For sync compaction, schedule. | |
248 | * | |
249 | * Returns true if the lock is held. | |
250 | * Returns false if the lock is released and compaction should abort | |
251 | */ | |
252 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, | |
253 | bool locked, struct compact_control *cc) | |
254 | { | |
1f9efdef VB |
255 | int contended = should_release_lock(lock); |
256 | ||
257 | if (contended) { | |
c67fe375 MG |
258 | if (locked) { |
259 | spin_unlock_irqrestore(lock, *flags); | |
260 | locked = false; | |
261 | } | |
262 | ||
263 | /* async aborts if taking too long or contended */ | |
e0b9daeb | 264 | if (cc->mode == MIGRATE_ASYNC) { |
1f9efdef | 265 | cc->contended = contended; |
c67fe375 MG |
266 | return false; |
267 | } | |
268 | ||
269 | cond_resched(); | |
c67fe375 MG |
270 | } |
271 | ||
272 | if (!locked) | |
273 | spin_lock_irqsave(lock, *flags); | |
274 | return true; | |
275 | } | |
276 | ||
be976572 VB |
277 | /* |
278 | * Aside from avoiding lock contention, compaction also periodically checks | |
279 | * need_resched() and either schedules in sync compaction or aborts async | |
280 | * compaction. This is similar to what compact_checklock_irqsave() does, but | |
281 | * is used where no lock is concerned. | |
282 | * | |
283 | * Returns false when no scheduling was needed, or sync compaction scheduled. | |
284 | * Returns true when async compaction should abort. | |
285 | */ | |
286 | static inline bool compact_should_abort(struct compact_control *cc) | |
287 | { | |
288 | /* async compaction aborts if contended */ | |
289 | if (need_resched()) { | |
290 | if (cc->mode == MIGRATE_ASYNC) { | |
1f9efdef | 291 | cc->contended = COMPACT_CONTENDED_SCHED; |
be976572 VB |
292 | return true; |
293 | } | |
294 | ||
295 | cond_resched(); | |
296 | } | |
297 | ||
298 | return false; | |
299 | } | |
300 | ||
f40d1e42 MG |
301 | /* Returns true if the page is within a block suitable for migration to */ |
302 | static bool suitable_migration_target(struct page *page) | |
303 | { | |
7d348b9e | 304 | /* If the page is a large free page, then disallow migration */ |
f40d1e42 | 305 | if (PageBuddy(page) && page_order(page) >= pageblock_order) |
7d348b9e | 306 | return false; |
f40d1e42 MG |
307 | |
308 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
7d348b9e | 309 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
f40d1e42 MG |
310 | return true; |
311 | ||
312 | /* Otherwise skip the block */ | |
313 | return false; | |
314 | } | |
315 | ||
85aa125f | 316 | /* |
9e4be470 JM |
317 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
318 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
319 | * (even though it may still end up isolating some pages). | |
85aa125f | 320 | */ |
f40d1e42 MG |
321 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
322 | unsigned long blockpfn, | |
85aa125f MN |
323 | unsigned long end_pfn, |
324 | struct list_head *freelist, | |
325 | bool strict) | |
748446bb | 326 | { |
b7aba698 | 327 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 328 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
329 | unsigned long flags; |
330 | bool locked = false; | |
748446bb | 331 | |
748446bb MG |
332 | cursor = pfn_to_page(blockpfn); |
333 | ||
f40d1e42 | 334 | /* Isolate free pages. */ |
748446bb MG |
335 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
336 | int isolated, i; | |
337 | struct page *page = cursor; | |
338 | ||
b7aba698 | 339 | nr_scanned++; |
f40d1e42 | 340 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
341 | goto isolate_fail; |
342 | ||
bb13ffeb MG |
343 | if (!valid_page) |
344 | valid_page = page; | |
f40d1e42 | 345 | if (!PageBuddy(page)) |
2af120bc | 346 | goto isolate_fail; |
f40d1e42 MG |
347 | |
348 | /* | |
349 | * The zone lock must be held to isolate freepages. | |
350 | * Unfortunately this is a very coarse lock and can be | |
351 | * heavily contended if there are parallel allocations | |
352 | * or parallel compactions. For async compaction do not | |
353 | * spin on the lock and we acquire the lock as late as | |
354 | * possible. | |
355 | */ | |
356 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
357 | locked, cc); | |
358 | if (!locked) | |
359 | break; | |
360 | ||
f40d1e42 MG |
361 | /* Recheck this is a buddy page under lock */ |
362 | if (!PageBuddy(page)) | |
2af120bc | 363 | goto isolate_fail; |
748446bb MG |
364 | |
365 | /* Found a free page, break it into order-0 pages */ | |
366 | isolated = split_free_page(page); | |
367 | total_isolated += isolated; | |
368 | for (i = 0; i < isolated; i++) { | |
369 | list_add(&page->lru, freelist); | |
370 | page++; | |
371 | } | |
372 | ||
373 | /* If a page was split, advance to the end of it */ | |
374 | if (isolated) { | |
375 | blockpfn += isolated - 1; | |
376 | cursor += isolated - 1; | |
2af120bc | 377 | continue; |
748446bb | 378 | } |
2af120bc LA |
379 | |
380 | isolate_fail: | |
381 | if (strict) | |
382 | break; | |
383 | else | |
384 | continue; | |
385 | ||
748446bb MG |
386 | } |
387 | ||
b7aba698 | 388 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
389 | |
390 | /* | |
391 | * If strict isolation is requested by CMA then check that all the | |
392 | * pages requested were isolated. If there were any failures, 0 is | |
393 | * returned and CMA will fail. | |
394 | */ | |
2af120bc | 395 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
396 | total_isolated = 0; |
397 | ||
398 | if (locked) | |
399 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
400 | ||
bb13ffeb MG |
401 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
402 | if (blockpfn == end_pfn) | |
edc2ca61 | 403 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 404 | |
010fc29a | 405 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 406 | if (total_isolated) |
010fc29a | 407 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
408 | return total_isolated; |
409 | } | |
410 | ||
85aa125f MN |
411 | /** |
412 | * isolate_freepages_range() - isolate free pages. | |
413 | * @start_pfn: The first PFN to start isolating. | |
414 | * @end_pfn: The one-past-last PFN. | |
415 | * | |
416 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
417 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
418 | * undo its actions and return zero. | |
419 | * | |
420 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
421 | * (which may be greater then end_pfn if end fell in a middle of | |
422 | * a free page). | |
423 | */ | |
ff9543fd | 424 | unsigned long |
bb13ffeb MG |
425 | isolate_freepages_range(struct compact_control *cc, |
426 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 427 | { |
f40d1e42 | 428 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
429 | LIST_HEAD(freelist); |
430 | ||
7d49d886 VB |
431 | pfn = start_pfn; |
432 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
433 | ||
434 | for (; pfn < end_pfn; pfn += isolated, | |
435 | block_end_pfn += pageblock_nr_pages) { | |
85aa125f | 436 | |
85aa125f MN |
437 | block_end_pfn = min(block_end_pfn, end_pfn); |
438 | ||
7d49d886 VB |
439 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
440 | break; | |
441 | ||
bb13ffeb | 442 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 443 | &freelist, true); |
85aa125f MN |
444 | |
445 | /* | |
446 | * In strict mode, isolate_freepages_block() returns 0 if | |
447 | * there are any holes in the block (ie. invalid PFNs or | |
448 | * non-free pages). | |
449 | */ | |
450 | if (!isolated) | |
451 | break; | |
452 | ||
453 | /* | |
454 | * If we managed to isolate pages, it is always (1 << n) * | |
455 | * pageblock_nr_pages for some non-negative n. (Max order | |
456 | * page may span two pageblocks). | |
457 | */ | |
458 | } | |
459 | ||
460 | /* split_free_page does not map the pages */ | |
461 | map_pages(&freelist); | |
462 | ||
463 | if (pfn < end_pfn) { | |
464 | /* Loop terminated early, cleanup. */ | |
465 | release_freepages(&freelist); | |
466 | return 0; | |
467 | } | |
468 | ||
469 | /* We don't use freelists for anything. */ | |
470 | return pfn; | |
471 | } | |
472 | ||
748446bb | 473 | /* Update the number of anon and file isolated pages in the zone */ |
edc2ca61 | 474 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
748446bb MG |
475 | { |
476 | struct page *page; | |
b9e84ac1 | 477 | unsigned int count[2] = { 0, }; |
748446bb | 478 | |
edc2ca61 VB |
479 | if (list_empty(&cc->migratepages)) |
480 | return; | |
481 | ||
b9e84ac1 MK |
482 | list_for_each_entry(page, &cc->migratepages, lru) |
483 | count[!!page_is_file_cache(page)]++; | |
748446bb | 484 | |
edc2ca61 VB |
485 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
486 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
748446bb MG |
487 | } |
488 | ||
489 | /* Similar to reclaim, but different enough that they don't share logic */ | |
490 | static bool too_many_isolated(struct zone *zone) | |
491 | { | |
bc693045 | 492 | unsigned long active, inactive, isolated; |
748446bb MG |
493 | |
494 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
495 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
496 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
497 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
498 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
499 | zone_page_state(zone, NR_ISOLATED_ANON); | |
500 | ||
bc693045 | 501 | return isolated > (inactive + active) / 2; |
748446bb MG |
502 | } |
503 | ||
2fe86e00 | 504 | /** |
edc2ca61 VB |
505 | * isolate_migratepages_block() - isolate all migrate-able pages within |
506 | * a single pageblock | |
2fe86e00 | 507 | * @cc: Compaction control structure. |
edc2ca61 VB |
508 | * @low_pfn: The first PFN to isolate |
509 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock | |
510 | * @isolate_mode: Isolation mode to be used. | |
2fe86e00 MN |
511 | * |
512 | * Isolate all pages that can be migrated from the range specified by | |
edc2ca61 VB |
513 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
514 | * Returns zero if there is a fatal signal pending, otherwise PFN of the | |
515 | * first page that was not scanned (which may be both less, equal to or more | |
516 | * than end_pfn). | |
2fe86e00 | 517 | * |
edc2ca61 VB |
518 | * The pages are isolated on cc->migratepages list (not required to be empty), |
519 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field | |
520 | * is neither read nor updated. | |
748446bb | 521 | */ |
edc2ca61 VB |
522 | static unsigned long |
523 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, | |
524 | unsigned long end_pfn, isolate_mode_t isolate_mode) | |
748446bb | 525 | { |
edc2ca61 | 526 | struct zone *zone = cc->zone; |
b7aba698 | 527 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 528 | struct list_head *migratelist = &cc->migratepages; |
fa9add64 | 529 | struct lruvec *lruvec; |
c67fe375 | 530 | unsigned long flags; |
2a1402aa | 531 | bool locked = false; |
bb13ffeb | 532 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 533 | |
748446bb MG |
534 | /* |
535 | * Ensure that there are not too many pages isolated from the LRU | |
536 | * list by either parallel reclaimers or compaction. If there are, | |
537 | * delay for some time until fewer pages are isolated | |
538 | */ | |
539 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 540 | /* async migration should just abort */ |
e0b9daeb | 541 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 542 | return 0; |
f9e35b3b | 543 | |
748446bb MG |
544 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
545 | ||
546 | if (fatal_signal_pending(current)) | |
2fe86e00 | 547 | return 0; |
748446bb MG |
548 | } |
549 | ||
be976572 VB |
550 | if (compact_should_abort(cc)) |
551 | return 0; | |
aeef4b83 | 552 | |
748446bb | 553 | /* Time to isolate some pages for migration */ |
748446bb | 554 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 555 | /* give a chance to irqs before checking need_resched() */ |
be1aa03b | 556 | if (locked && !(low_pfn % SWAP_CLUSTER_MAX)) { |
2a1402aa MG |
557 | if (should_release_lock(&zone->lru_lock)) { |
558 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
559 | locked = false; | |
560 | } | |
b2eef8c0 | 561 | } |
c67fe375 | 562 | |
748446bb MG |
563 | if (!pfn_valid_within(low_pfn)) |
564 | continue; | |
b7aba698 | 565 | nr_scanned++; |
748446bb | 566 | |
748446bb | 567 | page = pfn_to_page(low_pfn); |
dc908600 | 568 | |
bb13ffeb MG |
569 | if (!valid_page) |
570 | valid_page = page; | |
571 | ||
6c14466c MG |
572 | /* |
573 | * Skip if free. page_order cannot be used without zone->lock | |
574 | * as nothing prevents parallel allocations or buddy merging. | |
575 | */ | |
748446bb MG |
576 | if (PageBuddy(page)) |
577 | continue; | |
578 | ||
bf6bddf1 RA |
579 | /* |
580 | * Check may be lockless but that's ok as we recheck later. | |
581 | * It's possible to migrate LRU pages and balloon pages | |
582 | * Skip any other type of page | |
583 | */ | |
584 | if (!PageLRU(page)) { | |
585 | if (unlikely(balloon_page_movable(page))) { | |
586 | if (locked && balloon_page_isolate(page)) { | |
587 | /* Successfully isolated */ | |
b6c75016 | 588 | goto isolate_success; |
bf6bddf1 RA |
589 | } |
590 | } | |
bc835011 | 591 | continue; |
bf6bddf1 | 592 | } |
bc835011 AA |
593 | |
594 | /* | |
2a1402aa MG |
595 | * PageLRU is set. lru_lock normally excludes isolation |
596 | * splitting and collapsing (collapsing has already happened | |
597 | * if PageLRU is set) but the lock is not necessarily taken | |
598 | * here and it is wasteful to take it just to check transhuge. | |
599 | * Check TransHuge without lock and skip the whole pageblock if | |
600 | * it's either a transhuge or hugetlbfs page, as calling | |
601 | * compound_order() without preventing THP from splitting the | |
602 | * page underneath us may return surprising results. | |
bc835011 | 603 | */ |
2a1402aa MG |
604 | if (PageTransHuge(page)) { |
605 | if (!locked) | |
edc2ca61 VB |
606 | low_pfn = ALIGN(low_pfn + 1, |
607 | pageblock_nr_pages) - 1; | |
608 | else | |
609 | low_pfn += (1 << compound_order(page)) - 1; | |
610 | ||
2a1402aa MG |
611 | continue; |
612 | } | |
613 | ||
119d6d59 DR |
614 | /* |
615 | * Migration will fail if an anonymous page is pinned in memory, | |
616 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
617 | * admittedly racy check. | |
618 | */ | |
619 | if (!page_mapping(page) && | |
620 | page_count(page) > page_mapcount(page)) | |
621 | continue; | |
622 | ||
2a1402aa MG |
623 | /* Check if it is ok to still hold the lock */ |
624 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
625 | locked, cc); | |
626 | if (!locked || fatal_signal_pending(current)) | |
627 | break; | |
628 | ||
629 | /* Recheck PageLRU and PageTransHuge under lock */ | |
630 | if (!PageLRU(page)) | |
631 | continue; | |
bc835011 AA |
632 | if (PageTransHuge(page)) { |
633 | low_pfn += (1 << compound_order(page)) - 1; | |
634 | continue; | |
635 | } | |
636 | ||
fa9add64 HD |
637 | lruvec = mem_cgroup_page_lruvec(page, zone); |
638 | ||
748446bb | 639 | /* Try isolate the page */ |
edc2ca61 | 640 | if (__isolate_lru_page(page, isolate_mode) != 0) |
748446bb MG |
641 | continue; |
642 | ||
309381fe | 643 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 644 | |
748446bb | 645 | /* Successfully isolated */ |
fa9add64 | 646 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
647 | |
648 | isolate_success: | |
649 | cc->finished_update_migrate = true; | |
748446bb | 650 | list_add(&page->lru, migratelist); |
748446bb | 651 | cc->nr_migratepages++; |
b7aba698 | 652 | nr_isolated++; |
748446bb MG |
653 | |
654 | /* Avoid isolating too much */ | |
31b8384a HD |
655 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
656 | ++low_pfn; | |
748446bb | 657 | break; |
31b8384a | 658 | } |
748446bb MG |
659 | } |
660 | ||
c67fe375 MG |
661 | if (locked) |
662 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 663 | |
50b5b094 VB |
664 | /* |
665 | * Update the pageblock-skip information and cached scanner pfn, | |
666 | * if the whole pageblock was scanned without isolating any page. | |
50b5b094 | 667 | */ |
35979ef3 | 668 | if (low_pfn == end_pfn) |
edc2ca61 | 669 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 670 | |
b7aba698 MG |
671 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
672 | ||
010fc29a | 673 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 674 | if (nr_isolated) |
010fc29a | 675 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 676 | |
2fe86e00 MN |
677 | return low_pfn; |
678 | } | |
679 | ||
edc2ca61 VB |
680 | /** |
681 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range | |
682 | * @cc: Compaction control structure. | |
683 | * @start_pfn: The first PFN to start isolating. | |
684 | * @end_pfn: The one-past-last PFN. | |
685 | * | |
686 | * Returns zero if isolation fails fatally due to e.g. pending signal. | |
687 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
688 | * (which may be greater than end_pfn if end fell in a middle of a THP page). | |
689 | */ | |
690 | unsigned long | |
691 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, | |
692 | unsigned long end_pfn) | |
693 | { | |
694 | unsigned long pfn, block_end_pfn; | |
695 | ||
696 | /* Scan block by block. First and last block may be incomplete */ | |
697 | pfn = start_pfn; | |
698 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
699 | ||
700 | for (; pfn < end_pfn; pfn = block_end_pfn, | |
701 | block_end_pfn += pageblock_nr_pages) { | |
702 | ||
703 | block_end_pfn = min(block_end_pfn, end_pfn); | |
704 | ||
7d49d886 | 705 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
edc2ca61 VB |
706 | continue; |
707 | ||
708 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, | |
709 | ISOLATE_UNEVICTABLE); | |
710 | ||
711 | /* | |
712 | * In case of fatal failure, release everything that might | |
713 | * have been isolated in the previous iteration, and signal | |
714 | * the failure back to caller. | |
715 | */ | |
716 | if (!pfn) { | |
717 | putback_movable_pages(&cc->migratepages); | |
718 | cc->nr_migratepages = 0; | |
719 | break; | |
720 | } | |
721 | } | |
722 | acct_isolated(cc->zone, cc); | |
723 | ||
724 | return pfn; | |
725 | } | |
726 | ||
ff9543fd MN |
727 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
728 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 729 | /* |
ff9543fd MN |
730 | * Based on information in the current compact_control, find blocks |
731 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 732 | */ |
edc2ca61 | 733 | static void isolate_freepages(struct compact_control *cc) |
2fe86e00 | 734 | { |
edc2ca61 | 735 | struct zone *zone = cc->zone; |
ff9543fd | 736 | struct page *page; |
c96b9e50 VB |
737 | unsigned long block_start_pfn; /* start of current pageblock */ |
738 | unsigned long block_end_pfn; /* end of current pageblock */ | |
739 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd MN |
740 | int nr_freepages = cc->nr_freepages; |
741 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 742 | |
ff9543fd MN |
743 | /* |
744 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 VB |
745 | * successfully isolated from, zone-cached value, or the end of the |
746 | * zone when isolating for the first time. We need this aligned to | |
c96b9e50 VB |
747 | * the pageblock boundary, because we do |
748 | * block_start_pfn -= pageblock_nr_pages in the for loop. | |
749 | * For ending point, take care when isolating in last pageblock of a | |
750 | * a zone which ends in the middle of a pageblock. | |
49e068f0 VB |
751 | * The low boundary is the end of the pageblock the migration scanner |
752 | * is using. | |
ff9543fd | 753 | */ |
c96b9e50 VB |
754 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
755 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, | |
756 | zone_end_pfn(zone)); | |
7ed695e0 | 757 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 758 | |
ff9543fd MN |
759 | /* |
760 | * Isolate free pages until enough are available to migrate the | |
761 | * pages on cc->migratepages. We stop searching if the migrate | |
762 | * and free page scanners meet or enough free pages are isolated. | |
763 | */ | |
c96b9e50 VB |
764 | for (; block_start_pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
765 | block_end_pfn = block_start_pfn, | |
766 | block_start_pfn -= pageblock_nr_pages) { | |
ff9543fd | 767 | unsigned long isolated; |
2fe86e00 | 768 | |
f6ea3adb DR |
769 | /* |
770 | * This can iterate a massively long zone without finding any | |
771 | * suitable migration targets, so periodically check if we need | |
be976572 | 772 | * to schedule, or even abort async compaction. |
f6ea3adb | 773 | */ |
be976572 VB |
774 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
775 | && compact_should_abort(cc)) | |
776 | break; | |
f6ea3adb | 777 | |
7d49d886 VB |
778 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
779 | zone); | |
780 | if (!page) | |
ff9543fd MN |
781 | continue; |
782 | ||
783 | /* Check the block is suitable for migration */ | |
68e3e926 | 784 | if (!suitable_migration_target(page)) |
ff9543fd | 785 | continue; |
68e3e926 | 786 | |
bb13ffeb MG |
787 | /* If isolation recently failed, do not retry */ |
788 | if (!isolation_suitable(cc, page)) | |
789 | continue; | |
790 | ||
f40d1e42 | 791 | /* Found a block suitable for isolating free pages from */ |
e9ade569 | 792 | cc->free_pfn = block_start_pfn; |
c96b9e50 VB |
793 | isolated = isolate_freepages_block(cc, block_start_pfn, |
794 | block_end_pfn, freelist, false); | |
f40d1e42 | 795 | nr_freepages += isolated; |
ff9543fd MN |
796 | |
797 | /* | |
e9ade569 VB |
798 | * Set a flag that we successfully isolated in this pageblock. |
799 | * In the next loop iteration, zone->compact_cached_free_pfn | |
800 | * will not be updated and thus it will effectively contain the | |
801 | * highest pageblock we isolated pages from. | |
ff9543fd | 802 | */ |
e9ade569 | 803 | if (isolated) |
c89511ab | 804 | cc->finished_update_free = true; |
be976572 VB |
805 | |
806 | /* | |
807 | * isolate_freepages_block() might have aborted due to async | |
808 | * compaction being contended | |
809 | */ | |
810 | if (cc->contended) | |
811 | break; | |
ff9543fd MN |
812 | } |
813 | ||
814 | /* split_free_page does not map the pages */ | |
815 | map_pages(freelist); | |
816 | ||
7ed695e0 VB |
817 | /* |
818 | * If we crossed the migrate scanner, we want to keep it that way | |
819 | * so that compact_finished() may detect this | |
820 | */ | |
c96b9e50 | 821 | if (block_start_pfn < low_pfn) |
e9ade569 | 822 | cc->free_pfn = cc->migrate_pfn; |
c96b9e50 | 823 | |
ff9543fd | 824 | cc->nr_freepages = nr_freepages; |
748446bb MG |
825 | } |
826 | ||
827 | /* | |
828 | * This is a migrate-callback that "allocates" freepages by taking pages | |
829 | * from the isolated freelists in the block we are migrating to. | |
830 | */ | |
831 | static struct page *compaction_alloc(struct page *migratepage, | |
832 | unsigned long data, | |
833 | int **result) | |
834 | { | |
835 | struct compact_control *cc = (struct compact_control *)data; | |
836 | struct page *freepage; | |
837 | ||
be976572 VB |
838 | /* |
839 | * Isolate free pages if necessary, and if we are not aborting due to | |
840 | * contention. | |
841 | */ | |
748446bb | 842 | if (list_empty(&cc->freepages)) { |
be976572 | 843 | if (!cc->contended) |
edc2ca61 | 844 | isolate_freepages(cc); |
748446bb MG |
845 | |
846 | if (list_empty(&cc->freepages)) | |
847 | return NULL; | |
848 | } | |
849 | ||
850 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
851 | list_del(&freepage->lru); | |
852 | cc->nr_freepages--; | |
853 | ||
854 | return freepage; | |
855 | } | |
856 | ||
857 | /* | |
d53aea3d DR |
858 | * This is a migrate-callback that "frees" freepages back to the isolated |
859 | * freelist. All pages on the freelist are from the same zone, so there is no | |
860 | * special handling needed for NUMA. | |
861 | */ | |
862 | static void compaction_free(struct page *page, unsigned long data) | |
863 | { | |
864 | struct compact_control *cc = (struct compact_control *)data; | |
865 | ||
866 | list_add(&page->lru, &cc->freepages); | |
867 | cc->nr_freepages++; | |
868 | } | |
869 | ||
ff9543fd MN |
870 | /* possible outcome of isolate_migratepages */ |
871 | typedef enum { | |
872 | ISOLATE_ABORT, /* Abort compaction now */ | |
873 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
874 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
875 | } isolate_migrate_t; | |
876 | ||
877 | /* | |
edc2ca61 VB |
878 | * Isolate all pages that can be migrated from the first suitable block, |
879 | * starting at the block pointed to by the migrate scanner pfn within | |
880 | * compact_control. | |
ff9543fd MN |
881 | */ |
882 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
883 | struct compact_control *cc) | |
884 | { | |
885 | unsigned long low_pfn, end_pfn; | |
edc2ca61 VB |
886 | struct page *page; |
887 | const isolate_mode_t isolate_mode = | |
888 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); | |
ff9543fd | 889 | |
edc2ca61 VB |
890 | /* |
891 | * Start at where we last stopped, or beginning of the zone as | |
892 | * initialized by compact_zone() | |
893 | */ | |
894 | low_pfn = cc->migrate_pfn; | |
ff9543fd MN |
895 | |
896 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 897 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd | 898 | |
edc2ca61 VB |
899 | /* |
900 | * Iterate over whole pageblocks until we find the first suitable. | |
901 | * Do not cross the free scanner. | |
902 | */ | |
903 | for (; end_pfn <= cc->free_pfn; | |
904 | low_pfn = end_pfn, end_pfn += pageblock_nr_pages) { | |
ff9543fd | 905 | |
edc2ca61 VB |
906 | /* |
907 | * This can potentially iterate a massively long zone with | |
908 | * many pageblocks unsuitable, so periodically check if we | |
909 | * need to schedule, or even abort async compaction. | |
910 | */ | |
911 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) | |
912 | && compact_should_abort(cc)) | |
913 | break; | |
ff9543fd | 914 | |
7d49d886 VB |
915 | page = pageblock_pfn_to_page(low_pfn, end_pfn, zone); |
916 | if (!page) | |
edc2ca61 VB |
917 | continue; |
918 | ||
edc2ca61 VB |
919 | /* If isolation recently failed, do not retry */ |
920 | if (!isolation_suitable(cc, page)) | |
921 | continue; | |
922 | ||
923 | /* | |
924 | * For async compaction, also only scan in MOVABLE blocks. | |
925 | * Async compaction is optimistic to see if the minimum amount | |
926 | * of work satisfies the allocation. | |
927 | */ | |
928 | if (cc->mode == MIGRATE_ASYNC && | |
929 | !migrate_async_suitable(get_pageblock_migratetype(page))) | |
930 | continue; | |
931 | ||
932 | /* Perform the isolation */ | |
933 | low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn, | |
934 | isolate_mode); | |
935 | ||
936 | if (!low_pfn || cc->contended) | |
937 | return ISOLATE_ABORT; | |
938 | ||
939 | /* | |
940 | * Either we isolated something and proceed with migration. Or | |
941 | * we failed and compact_zone should decide if we should | |
942 | * continue or not. | |
943 | */ | |
944 | break; | |
945 | } | |
946 | ||
947 | acct_isolated(zone, cc); | |
948 | /* Record where migration scanner will be restarted */ | |
ff9543fd MN |
949 | cc->migrate_pfn = low_pfn; |
950 | ||
edc2ca61 | 951 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
ff9543fd MN |
952 | } |
953 | ||
748446bb | 954 | static int compact_finished(struct zone *zone, |
5a03b051 | 955 | struct compact_control *cc) |
748446bb | 956 | { |
8fb74b9f | 957 | unsigned int order; |
5a03b051 | 958 | unsigned long watermark; |
56de7263 | 959 | |
be976572 | 960 | if (cc->contended || fatal_signal_pending(current)) |
748446bb MG |
961 | return COMPACT_PARTIAL; |
962 | ||
753341a4 | 963 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 964 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 | 965 | /* Let the next compaction start anew. */ |
35979ef3 DR |
966 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
967 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
55b7c4c9 VB |
968 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
969 | ||
62997027 MG |
970 | /* |
971 | * Mark that the PG_migrate_skip information should be cleared | |
972 | * by kswapd when it goes to sleep. kswapd does not set the | |
973 | * flag itself as the decision to be clear should be directly | |
974 | * based on an allocation request. | |
975 | */ | |
976 | if (!current_is_kswapd()) | |
977 | zone->compact_blockskip_flush = true; | |
978 | ||
748446bb | 979 | return COMPACT_COMPLETE; |
bb13ffeb | 980 | } |
748446bb | 981 | |
82478fb7 JW |
982 | /* |
983 | * order == -1 is expected when compacting via | |
984 | * /proc/sys/vm/compact_memory | |
985 | */ | |
56de7263 MG |
986 | if (cc->order == -1) |
987 | return COMPACT_CONTINUE; | |
988 | ||
3957c776 MH |
989 | /* Compaction run is not finished if the watermark is not met */ |
990 | watermark = low_wmark_pages(zone); | |
991 | watermark += (1 << cc->order); | |
992 | ||
993 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
994 | return COMPACT_CONTINUE; | |
995 | ||
56de7263 | 996 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
997 | for (order = cc->order; order < MAX_ORDER; order++) { |
998 | struct free_area *area = &zone->free_area[order]; | |
999 | ||
1000 | /* Job done if page is free of the right migratetype */ | |
1001 | if (!list_empty(&area->free_list[cc->migratetype])) | |
1002 | return COMPACT_PARTIAL; | |
1003 | ||
1004 | /* Job done if allocation would set block type */ | |
1005 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
1006 | return COMPACT_PARTIAL; |
1007 | } | |
1008 | ||
748446bb MG |
1009 | return COMPACT_CONTINUE; |
1010 | } | |
1011 | ||
3e7d3449 MG |
1012 | /* |
1013 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
1014 | * Returns | |
1015 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
1016 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
1017 | * COMPACT_CONTINUE - If compaction should run now | |
1018 | */ | |
1019 | unsigned long compaction_suitable(struct zone *zone, int order) | |
1020 | { | |
1021 | int fragindex; | |
1022 | unsigned long watermark; | |
1023 | ||
3957c776 MH |
1024 | /* |
1025 | * order == -1 is expected when compacting via | |
1026 | * /proc/sys/vm/compact_memory | |
1027 | */ | |
1028 | if (order == -1) | |
1029 | return COMPACT_CONTINUE; | |
1030 | ||
3e7d3449 MG |
1031 | /* |
1032 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
1033 | * This is because during migration, copies of pages need to be | |
1034 | * allocated and for a short time, the footprint is higher | |
1035 | */ | |
1036 | watermark = low_wmark_pages(zone) + (2UL << order); | |
1037 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1038 | return COMPACT_SKIPPED; | |
1039 | ||
1040 | /* | |
1041 | * fragmentation index determines if allocation failures are due to | |
1042 | * low memory or external fragmentation | |
1043 | * | |
a582a738 SL |
1044 | * index of -1000 implies allocations might succeed depending on |
1045 | * watermarks | |
3e7d3449 MG |
1046 | * index towards 0 implies failure is due to lack of memory |
1047 | * index towards 1000 implies failure is due to fragmentation | |
1048 | * | |
1049 | * Only compact if a failure would be due to fragmentation. | |
1050 | */ | |
1051 | fragindex = fragmentation_index(zone, order); | |
1052 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
1053 | return COMPACT_SKIPPED; | |
1054 | ||
a582a738 SL |
1055 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
1056 | 0, 0)) | |
3e7d3449 MG |
1057 | return COMPACT_PARTIAL; |
1058 | ||
1059 | return COMPACT_CONTINUE; | |
1060 | } | |
1061 | ||
748446bb MG |
1062 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
1063 | { | |
1064 | int ret; | |
c89511ab | 1065 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 1066 | unsigned long end_pfn = zone_end_pfn(zone); |
e0b9daeb | 1067 | const bool sync = cc->mode != MIGRATE_ASYNC; |
748446bb | 1068 | |
3e7d3449 MG |
1069 | ret = compaction_suitable(zone, cc->order); |
1070 | switch (ret) { | |
1071 | case COMPACT_PARTIAL: | |
1072 | case COMPACT_SKIPPED: | |
1073 | /* Compaction is likely to fail */ | |
1074 | return ret; | |
1075 | case COMPACT_CONTINUE: | |
1076 | /* Fall through to compaction */ | |
1077 | ; | |
1078 | } | |
1079 | ||
d3132e4b VB |
1080 | /* |
1081 | * Clear pageblock skip if there were failures recently and compaction | |
1082 | * is about to be retried after being deferred. kswapd does not do | |
1083 | * this reset as it'll reset the cached information when going to sleep. | |
1084 | */ | |
1085 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1086 | __reset_isolation_suitable(zone); | |
1087 | ||
c89511ab MG |
1088 | /* |
1089 | * Setup to move all movable pages to the end of the zone. Used cached | |
1090 | * information on where the scanners should start but check that it | |
1091 | * is initialised by ensuring the values are within zone boundaries. | |
1092 | */ | |
e0b9daeb | 1093 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
c89511ab MG |
1094 | cc->free_pfn = zone->compact_cached_free_pfn; |
1095 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
1096 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
1097 | zone->compact_cached_free_pfn = cc->free_pfn; | |
1098 | } | |
1099 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1100 | cc->migrate_pfn = start_pfn; | |
35979ef3 DR |
1101 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
1102 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
c89511ab | 1103 | } |
748446bb | 1104 | |
0eb927c0 MG |
1105 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn); |
1106 | ||
748446bb MG |
1107 | migrate_prep_local(); |
1108 | ||
1109 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
9d502c1c | 1110 | int err; |
748446bb | 1111 | |
f9e35b3b MG |
1112 | switch (isolate_migratepages(zone, cc)) { |
1113 | case ISOLATE_ABORT: | |
1114 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1115 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1116 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1117 | goto out; |
1118 | case ISOLATE_NONE: | |
748446bb | 1119 | continue; |
f9e35b3b MG |
1120 | case ISOLATE_SUCCESS: |
1121 | ; | |
1122 | } | |
748446bb | 1123 | |
d53aea3d | 1124 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 1125 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 1126 | MR_COMPACTION); |
748446bb | 1127 | |
f8c9301f VB |
1128 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
1129 | &cc->migratepages); | |
748446bb | 1130 | |
f8c9301f VB |
1131 | /* All pages were either migrated or will be released */ |
1132 | cc->nr_migratepages = 0; | |
9d502c1c | 1133 | if (err) { |
5733c7d1 | 1134 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
1135 | /* |
1136 | * migrate_pages() may return -ENOMEM when scanners meet | |
1137 | * and we want compact_finished() to detect it | |
1138 | */ | |
1139 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1140 | ret = COMPACT_PARTIAL; |
1141 | goto out; | |
1142 | } | |
748446bb | 1143 | } |
748446bb MG |
1144 | } |
1145 | ||
f9e35b3b | 1146 | out: |
748446bb MG |
1147 | /* Release free pages and check accounting */ |
1148 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1149 | VM_BUG_ON(cc->nr_freepages != 0); | |
1150 | ||
0eb927c0 MG |
1151 | trace_mm_compaction_end(ret); |
1152 | ||
748446bb MG |
1153 | return ret; |
1154 | } | |
76ab0f53 | 1155 | |
e0b9daeb | 1156 | static unsigned long compact_zone_order(struct zone *zone, int order, |
1f9efdef | 1157 | gfp_t gfp_mask, enum migrate_mode mode, int *contended) |
56de7263 | 1158 | { |
e64c5237 | 1159 | unsigned long ret; |
56de7263 MG |
1160 | struct compact_control cc = { |
1161 | .nr_freepages = 0, | |
1162 | .nr_migratepages = 0, | |
1163 | .order = order, | |
1164 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1165 | .zone = zone, | |
e0b9daeb | 1166 | .mode = mode, |
56de7263 MG |
1167 | }; |
1168 | INIT_LIST_HEAD(&cc.freepages); | |
1169 | INIT_LIST_HEAD(&cc.migratepages); | |
1170 | ||
e64c5237 SL |
1171 | ret = compact_zone(zone, &cc); |
1172 | ||
1173 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1174 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1175 | ||
1176 | *contended = cc.contended; | |
1177 | return ret; | |
56de7263 MG |
1178 | } |
1179 | ||
5e771905 MG |
1180 | int sysctl_extfrag_threshold = 500; |
1181 | ||
56de7263 MG |
1182 | /** |
1183 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1184 | * @zonelist: The zonelist used for the current allocation | |
1185 | * @order: The order of the current allocation | |
1186 | * @gfp_mask: The GFP mask of the current allocation | |
1187 | * @nodemask: The allowed nodes to allocate from | |
e0b9daeb | 1188 | * @mode: The migration mode for async, sync light, or sync migration |
1f9efdef VB |
1189 | * @contended: Return value that determines if compaction was aborted due to |
1190 | * need_resched() or lock contention | |
53853e2d | 1191 | * @candidate_zone: Return the zone where we think allocation should succeed |
56de7263 MG |
1192 | * |
1193 | * This is the main entry point for direct page compaction. | |
1194 | */ | |
1195 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1196 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
1f9efdef | 1197 | enum migrate_mode mode, int *contended, |
53853e2d | 1198 | struct zone **candidate_zone) |
56de7263 MG |
1199 | { |
1200 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1201 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1202 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1203 | struct zoneref *z; |
1204 | struct zone *zone; | |
53853e2d | 1205 | int rc = COMPACT_DEFERRED; |
d95ea5d1 | 1206 | int alloc_flags = 0; |
1f9efdef VB |
1207 | int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ |
1208 | ||
1209 | *contended = COMPACT_CONTENDED_NONE; | |
56de7263 | 1210 | |
4ffb6335 | 1211 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1212 | if (!order || !may_enter_fs || !may_perform_io) |
53853e2d | 1213 | return COMPACT_SKIPPED; |
56de7263 | 1214 | |
d95ea5d1 BZ |
1215 | #ifdef CONFIG_CMA |
1216 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1217 | alloc_flags |= ALLOC_CMA; | |
1218 | #endif | |
56de7263 MG |
1219 | /* Compact each zone in the list */ |
1220 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1221 | nodemask) { | |
56de7263 | 1222 | int status; |
1f9efdef | 1223 | int zone_contended; |
56de7263 | 1224 | |
53853e2d VB |
1225 | if (compaction_deferred(zone, order)) |
1226 | continue; | |
1227 | ||
e0b9daeb | 1228 | status = compact_zone_order(zone, order, gfp_mask, mode, |
1f9efdef | 1229 | &zone_contended); |
56de7263 | 1230 | rc = max(status, rc); |
1f9efdef VB |
1231 | /* |
1232 | * It takes at least one zone that wasn't lock contended | |
1233 | * to clear all_zones_contended. | |
1234 | */ | |
1235 | all_zones_contended &= zone_contended; | |
56de7263 | 1236 | |
3e7d3449 | 1237 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 | 1238 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
53853e2d VB |
1239 | alloc_flags)) { |
1240 | *candidate_zone = zone; | |
1241 | /* | |
1242 | * We think the allocation will succeed in this zone, | |
1243 | * but it is not certain, hence the false. The caller | |
1244 | * will repeat this with true if allocation indeed | |
1245 | * succeeds in this zone. | |
1246 | */ | |
1247 | compaction_defer_reset(zone, order, false); | |
1f9efdef VB |
1248 | /* |
1249 | * It is possible that async compaction aborted due to | |
1250 | * need_resched() and the watermarks were ok thanks to | |
1251 | * somebody else freeing memory. The allocation can | |
1252 | * however still fail so we better signal the | |
1253 | * need_resched() contention anyway (this will not | |
1254 | * prevent the allocation attempt). | |
1255 | */ | |
1256 | if (zone_contended == COMPACT_CONTENDED_SCHED) | |
1257 | *contended = COMPACT_CONTENDED_SCHED; | |
1258 | ||
1259 | goto break_loop; | |
1260 | } | |
1261 | ||
1262 | if (mode != MIGRATE_ASYNC) { | |
53853e2d VB |
1263 | /* |
1264 | * We think that allocation won't succeed in this zone | |
1265 | * so we defer compaction there. If it ends up | |
1266 | * succeeding after all, it will be reset. | |
1267 | */ | |
1268 | defer_compaction(zone, order); | |
1269 | } | |
1f9efdef VB |
1270 | |
1271 | /* | |
1272 | * We might have stopped compacting due to need_resched() in | |
1273 | * async compaction, or due to a fatal signal detected. In that | |
1274 | * case do not try further zones and signal need_resched() | |
1275 | * contention. | |
1276 | */ | |
1277 | if ((zone_contended == COMPACT_CONTENDED_SCHED) | |
1278 | || fatal_signal_pending(current)) { | |
1279 | *contended = COMPACT_CONTENDED_SCHED; | |
1280 | goto break_loop; | |
1281 | } | |
1282 | ||
1283 | continue; | |
1284 | break_loop: | |
1285 | /* | |
1286 | * We might not have tried all the zones, so be conservative | |
1287 | * and assume they are not all lock contended. | |
1288 | */ | |
1289 | all_zones_contended = 0; | |
1290 | break; | |
56de7263 MG |
1291 | } |
1292 | ||
1f9efdef VB |
1293 | /* |
1294 | * If at least one zone wasn't deferred or skipped, we report if all | |
1295 | * zones that were tried were lock contended. | |
1296 | */ | |
1297 | if (rc > COMPACT_SKIPPED && all_zones_contended) | |
1298 | *contended = COMPACT_CONTENDED_LOCK; | |
1299 | ||
56de7263 MG |
1300 | return rc; |
1301 | } | |
1302 | ||
1303 | ||
76ab0f53 | 1304 | /* Compact all zones within a node */ |
7103f16d | 1305 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1306 | { |
1307 | int zoneid; | |
76ab0f53 MG |
1308 | struct zone *zone; |
1309 | ||
76ab0f53 | 1310 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1311 | |
1312 | zone = &pgdat->node_zones[zoneid]; | |
1313 | if (!populated_zone(zone)) | |
1314 | continue; | |
1315 | ||
7be62de9 RR |
1316 | cc->nr_freepages = 0; |
1317 | cc->nr_migratepages = 0; | |
1318 | cc->zone = zone; | |
1319 | INIT_LIST_HEAD(&cc->freepages); | |
1320 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1321 | |
aad6ec37 | 1322 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1323 | compact_zone(zone, cc); |
76ab0f53 | 1324 | |
aff62249 | 1325 | if (cc->order > 0) { |
de6c60a6 VB |
1326 | if (zone_watermark_ok(zone, cc->order, |
1327 | low_wmark_pages(zone), 0, 0)) | |
1328 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 RR |
1329 | } |
1330 | ||
7be62de9 RR |
1331 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1332 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1333 | } |
76ab0f53 MG |
1334 | } |
1335 | ||
7103f16d | 1336 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1337 | { |
1338 | struct compact_control cc = { | |
1339 | .order = order, | |
e0b9daeb | 1340 | .mode = MIGRATE_ASYNC, |
7be62de9 RR |
1341 | }; |
1342 | ||
3a7200af MG |
1343 | if (!order) |
1344 | return; | |
1345 | ||
7103f16d | 1346 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1347 | } |
1348 | ||
7103f16d | 1349 | static void compact_node(int nid) |
7be62de9 | 1350 | { |
7be62de9 RR |
1351 | struct compact_control cc = { |
1352 | .order = -1, | |
e0b9daeb | 1353 | .mode = MIGRATE_SYNC, |
91ca9186 | 1354 | .ignore_skip_hint = true, |
7be62de9 RR |
1355 | }; |
1356 | ||
7103f16d | 1357 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1358 | } |
1359 | ||
76ab0f53 | 1360 | /* Compact all nodes in the system */ |
7964c06d | 1361 | static void compact_nodes(void) |
76ab0f53 MG |
1362 | { |
1363 | int nid; | |
1364 | ||
8575ec29 HD |
1365 | /* Flush pending updates to the LRU lists */ |
1366 | lru_add_drain_all(); | |
1367 | ||
76ab0f53 MG |
1368 | for_each_online_node(nid) |
1369 | compact_node(nid); | |
76ab0f53 MG |
1370 | } |
1371 | ||
1372 | /* The written value is actually unused, all memory is compacted */ | |
1373 | int sysctl_compact_memory; | |
1374 | ||
1375 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1376 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1377 | void __user *buffer, size_t *length, loff_t *ppos) | |
1378 | { | |
1379 | if (write) | |
7964c06d | 1380 | compact_nodes(); |
76ab0f53 MG |
1381 | |
1382 | return 0; | |
1383 | } | |
ed4a6d7f | 1384 | |
5e771905 MG |
1385 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1386 | void __user *buffer, size_t *length, loff_t *ppos) | |
1387 | { | |
1388 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1389 | ||
1390 | return 0; | |
1391 | } | |
1392 | ||
ed4a6d7f | 1393 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1394 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1395 | struct device_attribute *attr, |
ed4a6d7f MG |
1396 | const char *buf, size_t count) |
1397 | { | |
8575ec29 HD |
1398 | int nid = dev->id; |
1399 | ||
1400 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1401 | /* Flush pending updates to the LRU lists */ | |
1402 | lru_add_drain_all(); | |
1403 | ||
1404 | compact_node(nid); | |
1405 | } | |
ed4a6d7f MG |
1406 | |
1407 | return count; | |
1408 | } | |
10fbcf4c | 1409 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1410 | |
1411 | int compaction_register_node(struct node *node) | |
1412 | { | |
10fbcf4c | 1413 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1414 | } |
1415 | ||
1416 | void compaction_unregister_node(struct node *node) | |
1417 | { | |
10fbcf4c | 1418 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1419 | } |
1420 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1421 | |
1422 | #endif /* CONFIG_COMPACTION */ |