]>
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 | ||
bb13ffeb MG |
70 | #ifdef CONFIG_COMPACTION |
71 | /* Returns true if the pageblock should be scanned for pages to isolate. */ | |
72 | static inline bool isolation_suitable(struct compact_control *cc, | |
73 | struct page *page) | |
74 | { | |
75 | if (cc->ignore_skip_hint) | |
76 | return true; | |
77 | ||
78 | return !get_pageblock_skip(page); | |
79 | } | |
80 | ||
81 | /* | |
82 | * This function is called to clear all cached information on pageblocks that | |
83 | * should be skipped for page isolation when the migrate and free page scanner | |
84 | * meet. | |
85 | */ | |
62997027 | 86 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
87 | { |
88 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 89 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
90 | unsigned long pfn; |
91 | ||
c89511ab MG |
92 | zone->compact_cached_migrate_pfn = start_pfn; |
93 | zone->compact_cached_free_pfn = end_pfn; | |
62997027 | 94 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
95 | |
96 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
97 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
98 | struct page *page; | |
99 | ||
100 | cond_resched(); | |
101 | ||
102 | if (!pfn_valid(pfn)) | |
103 | continue; | |
104 | ||
105 | page = pfn_to_page(pfn); | |
106 | if (zone != page_zone(page)) | |
107 | continue; | |
108 | ||
109 | clear_pageblock_skip(page); | |
110 | } | |
111 | } | |
112 | ||
62997027 MG |
113 | void reset_isolation_suitable(pg_data_t *pgdat) |
114 | { | |
115 | int zoneid; | |
116 | ||
117 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
118 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
119 | if (!populated_zone(zone)) | |
120 | continue; | |
121 | ||
122 | /* Only flush if a full compaction finished recently */ | |
123 | if (zone->compact_blockskip_flush) | |
124 | __reset_isolation_suitable(zone); | |
125 | } | |
126 | } | |
127 | ||
bb13ffeb MG |
128 | /* |
129 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 130 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 131 | */ |
c89511ab MG |
132 | static void update_pageblock_skip(struct compact_control *cc, |
133 | struct page *page, unsigned long nr_isolated, | |
134 | bool migrate_scanner) | |
bb13ffeb | 135 | { |
c89511ab | 136 | struct zone *zone = cc->zone; |
6815bf3f JK |
137 | |
138 | if (cc->ignore_skip_hint) | |
139 | return; | |
140 | ||
bb13ffeb MG |
141 | if (!page) |
142 | return; | |
143 | ||
c89511ab MG |
144 | if (!nr_isolated) { |
145 | unsigned long pfn = page_to_pfn(page); | |
bb13ffeb | 146 | set_pageblock_skip(page); |
c89511ab MG |
147 | |
148 | /* Update where compaction should restart */ | |
149 | if (migrate_scanner) { | |
150 | if (!cc->finished_update_migrate && | |
151 | pfn > zone->compact_cached_migrate_pfn) | |
152 | zone->compact_cached_migrate_pfn = pfn; | |
153 | } else { | |
154 | if (!cc->finished_update_free && | |
155 | pfn < zone->compact_cached_free_pfn) | |
156 | zone->compact_cached_free_pfn = pfn; | |
157 | } | |
158 | } | |
bb13ffeb MG |
159 | } |
160 | #else | |
161 | static inline bool isolation_suitable(struct compact_control *cc, | |
162 | struct page *page) | |
163 | { | |
164 | return true; | |
165 | } | |
166 | ||
c89511ab MG |
167 | static void update_pageblock_skip(struct compact_control *cc, |
168 | struct page *page, unsigned long nr_isolated, | |
169 | bool migrate_scanner) | |
bb13ffeb MG |
170 | { |
171 | } | |
172 | #endif /* CONFIG_COMPACTION */ | |
173 | ||
2a1402aa MG |
174 | static inline bool should_release_lock(spinlock_t *lock) |
175 | { | |
176 | return need_resched() || spin_is_contended(lock); | |
177 | } | |
178 | ||
c67fe375 MG |
179 | /* |
180 | * Compaction requires the taking of some coarse locks that are potentially | |
181 | * very heavily contended. Check if the process needs to be scheduled or | |
182 | * if the lock is contended. For async compaction, back out in the event | |
183 | * if contention is severe. For sync compaction, schedule. | |
184 | * | |
185 | * Returns true if the lock is held. | |
186 | * Returns false if the lock is released and compaction should abort | |
187 | */ | |
188 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, | |
189 | bool locked, struct compact_control *cc) | |
190 | { | |
2a1402aa | 191 | if (should_release_lock(lock)) { |
c67fe375 MG |
192 | if (locked) { |
193 | spin_unlock_irqrestore(lock, *flags); | |
194 | locked = false; | |
195 | } | |
196 | ||
197 | /* async aborts if taking too long or contended */ | |
198 | if (!cc->sync) { | |
e64c5237 | 199 | cc->contended = true; |
c67fe375 MG |
200 | return false; |
201 | } | |
202 | ||
203 | cond_resched(); | |
c67fe375 MG |
204 | } |
205 | ||
206 | if (!locked) | |
207 | spin_lock_irqsave(lock, *flags); | |
208 | return true; | |
209 | } | |
210 | ||
211 | static inline bool compact_trylock_irqsave(spinlock_t *lock, | |
212 | unsigned long *flags, struct compact_control *cc) | |
213 | { | |
214 | return compact_checklock_irqsave(lock, flags, false, cc); | |
215 | } | |
216 | ||
f40d1e42 MG |
217 | /* Returns true if the page is within a block suitable for migration to */ |
218 | static bool suitable_migration_target(struct page *page) | |
219 | { | |
220 | int migratetype = get_pageblock_migratetype(page); | |
221 | ||
222 | /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ | |
194159fb MK |
223 | if (migratetype == MIGRATE_RESERVE) |
224 | return false; | |
225 | ||
226 | if (is_migrate_isolate(migratetype)) | |
f40d1e42 MG |
227 | return false; |
228 | ||
229 | /* If the page is a large free page, then allow migration */ | |
230 | if (PageBuddy(page) && page_order(page) >= pageblock_order) | |
231 | return true; | |
232 | ||
233 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
234 | if (migrate_async_suitable(migratetype)) | |
235 | return true; | |
236 | ||
237 | /* Otherwise skip the block */ | |
238 | return false; | |
239 | } | |
240 | ||
85aa125f | 241 | /* |
9e4be470 JM |
242 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
243 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
244 | * (even though it may still end up isolating some pages). | |
85aa125f | 245 | */ |
f40d1e42 MG |
246 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
247 | unsigned long blockpfn, | |
85aa125f MN |
248 | unsigned long end_pfn, |
249 | struct list_head *freelist, | |
250 | bool strict) | |
748446bb | 251 | { |
b7aba698 | 252 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 253 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
254 | unsigned long flags; |
255 | bool locked = false; | |
748446bb | 256 | |
748446bb MG |
257 | cursor = pfn_to_page(blockpfn); |
258 | ||
f40d1e42 | 259 | /* Isolate free pages. */ |
748446bb MG |
260 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
261 | int isolated, i; | |
262 | struct page *page = cursor; | |
263 | ||
b7aba698 | 264 | nr_scanned++; |
f40d1e42 | 265 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
266 | goto isolate_fail; |
267 | ||
bb13ffeb MG |
268 | if (!valid_page) |
269 | valid_page = page; | |
f40d1e42 | 270 | if (!PageBuddy(page)) |
2af120bc | 271 | goto isolate_fail; |
f40d1e42 MG |
272 | |
273 | /* | |
274 | * The zone lock must be held to isolate freepages. | |
275 | * Unfortunately this is a very coarse lock and can be | |
276 | * heavily contended if there are parallel allocations | |
277 | * or parallel compactions. For async compaction do not | |
278 | * spin on the lock and we acquire the lock as late as | |
279 | * possible. | |
280 | */ | |
281 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
282 | locked, cc); | |
283 | if (!locked) | |
284 | break; | |
285 | ||
286 | /* Recheck this is a suitable migration target under lock */ | |
287 | if (!strict && !suitable_migration_target(page)) | |
288 | break; | |
748446bb | 289 | |
f40d1e42 MG |
290 | /* Recheck this is a buddy page under lock */ |
291 | if (!PageBuddy(page)) | |
2af120bc | 292 | goto isolate_fail; |
748446bb MG |
293 | |
294 | /* Found a free page, break it into order-0 pages */ | |
295 | isolated = split_free_page(page); | |
296 | total_isolated += isolated; | |
297 | for (i = 0; i < isolated; i++) { | |
298 | list_add(&page->lru, freelist); | |
299 | page++; | |
300 | } | |
301 | ||
302 | /* If a page was split, advance to the end of it */ | |
303 | if (isolated) { | |
304 | blockpfn += isolated - 1; | |
305 | cursor += isolated - 1; | |
2af120bc | 306 | continue; |
748446bb | 307 | } |
2af120bc LA |
308 | |
309 | isolate_fail: | |
310 | if (strict) | |
311 | break; | |
312 | else | |
313 | continue; | |
314 | ||
748446bb MG |
315 | } |
316 | ||
b7aba698 | 317 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
318 | |
319 | /* | |
320 | * If strict isolation is requested by CMA then check that all the | |
321 | * pages requested were isolated. If there were any failures, 0 is | |
322 | * returned and CMA will fail. | |
323 | */ | |
2af120bc | 324 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
325 | total_isolated = 0; |
326 | ||
327 | if (locked) | |
328 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
329 | ||
bb13ffeb MG |
330 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
331 | if (blockpfn == end_pfn) | |
c89511ab | 332 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 333 | |
010fc29a | 334 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 335 | if (total_isolated) |
010fc29a | 336 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
337 | return total_isolated; |
338 | } | |
339 | ||
85aa125f MN |
340 | /** |
341 | * isolate_freepages_range() - isolate free pages. | |
342 | * @start_pfn: The first PFN to start isolating. | |
343 | * @end_pfn: The one-past-last PFN. | |
344 | * | |
345 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
346 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
347 | * undo its actions and return zero. | |
348 | * | |
349 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
350 | * (which may be greater then end_pfn if end fell in a middle of | |
351 | * a free page). | |
352 | */ | |
ff9543fd | 353 | unsigned long |
bb13ffeb MG |
354 | isolate_freepages_range(struct compact_control *cc, |
355 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 356 | { |
f40d1e42 | 357 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
358 | LIST_HEAD(freelist); |
359 | ||
85aa125f | 360 | for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) { |
bb13ffeb | 361 | if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn))) |
85aa125f MN |
362 | break; |
363 | ||
364 | /* | |
365 | * On subsequent iterations ALIGN() is actually not needed, | |
366 | * but we keep it that we not to complicate the code. | |
367 | */ | |
368 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
369 | block_end_pfn = min(block_end_pfn, end_pfn); | |
370 | ||
bb13ffeb | 371 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 372 | &freelist, true); |
85aa125f MN |
373 | |
374 | /* | |
375 | * In strict mode, isolate_freepages_block() returns 0 if | |
376 | * there are any holes in the block (ie. invalid PFNs or | |
377 | * non-free pages). | |
378 | */ | |
379 | if (!isolated) | |
380 | break; | |
381 | ||
382 | /* | |
383 | * If we managed to isolate pages, it is always (1 << n) * | |
384 | * pageblock_nr_pages for some non-negative n. (Max order | |
385 | * page may span two pageblocks). | |
386 | */ | |
387 | } | |
388 | ||
389 | /* split_free_page does not map the pages */ | |
390 | map_pages(&freelist); | |
391 | ||
392 | if (pfn < end_pfn) { | |
393 | /* Loop terminated early, cleanup. */ | |
394 | release_freepages(&freelist); | |
395 | return 0; | |
396 | } | |
397 | ||
398 | /* We don't use freelists for anything. */ | |
399 | return pfn; | |
400 | } | |
401 | ||
748446bb | 402 | /* Update the number of anon and file isolated pages in the zone */ |
c67fe375 | 403 | static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc) |
748446bb MG |
404 | { |
405 | struct page *page; | |
b9e84ac1 | 406 | unsigned int count[2] = { 0, }; |
748446bb | 407 | |
b9e84ac1 MK |
408 | list_for_each_entry(page, &cc->migratepages, lru) |
409 | count[!!page_is_file_cache(page)]++; | |
748446bb | 410 | |
c67fe375 MG |
411 | /* If locked we can use the interrupt unsafe versions */ |
412 | if (locked) { | |
413 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
414 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
415 | } else { | |
416 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
417 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
418 | } | |
748446bb MG |
419 | } |
420 | ||
421 | /* Similar to reclaim, but different enough that they don't share logic */ | |
422 | static bool too_many_isolated(struct zone *zone) | |
423 | { | |
bc693045 | 424 | unsigned long active, inactive, isolated; |
748446bb MG |
425 | |
426 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
427 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
428 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
429 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
430 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
431 | zone_page_state(zone, NR_ISOLATED_ANON); | |
432 | ||
bc693045 | 433 | return isolated > (inactive + active) / 2; |
748446bb MG |
434 | } |
435 | ||
2fe86e00 MN |
436 | /** |
437 | * isolate_migratepages_range() - isolate all migrate-able pages in range. | |
438 | * @zone: Zone pages are in. | |
439 | * @cc: Compaction control structure. | |
440 | * @low_pfn: The first PFN of the range. | |
441 | * @end_pfn: The one-past-the-last PFN of the range. | |
e46a2879 | 442 | * @unevictable: true if it allows to isolate unevictable pages |
2fe86e00 MN |
443 | * |
444 | * Isolate all pages that can be migrated from the range specified by | |
445 | * [low_pfn, end_pfn). Returns zero if there is a fatal signal | |
446 | * pending), otherwise PFN of the first page that was not scanned | |
447 | * (which may be both less, equal to or more then end_pfn). | |
448 | * | |
449 | * Assumes that cc->migratepages is empty and cc->nr_migratepages is | |
450 | * zero. | |
451 | * | |
452 | * Apart from cc->migratepages and cc->nr_migratetypes this function | |
453 | * does not modify any cc's fields, in particular it does not modify | |
454 | * (or read for that matter) cc->migrate_pfn. | |
748446bb | 455 | */ |
ff9543fd | 456 | unsigned long |
2fe86e00 | 457 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
e46a2879 | 458 | unsigned long low_pfn, unsigned long end_pfn, bool unevictable) |
748446bb | 459 | { |
9927af74 | 460 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
b7aba698 | 461 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 462 | struct list_head *migratelist = &cc->migratepages; |
f3fd4a61 | 463 | isolate_mode_t mode = 0; |
fa9add64 | 464 | struct lruvec *lruvec; |
c67fe375 | 465 | unsigned long flags; |
2a1402aa | 466 | bool locked = false; |
bb13ffeb | 467 | struct page *page = NULL, *valid_page = NULL; |
50b5b094 | 468 | bool skipped_async_unsuitable = false; |
748446bb | 469 | |
748446bb MG |
470 | /* |
471 | * Ensure that there are not too many pages isolated from the LRU | |
472 | * list by either parallel reclaimers or compaction. If there are, | |
473 | * delay for some time until fewer pages are isolated | |
474 | */ | |
475 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 476 | /* async migration should just abort */ |
68e3e926 | 477 | if (!cc->sync) |
2fe86e00 | 478 | return 0; |
f9e35b3b | 479 | |
748446bb MG |
480 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
481 | ||
482 | if (fatal_signal_pending(current)) | |
2fe86e00 | 483 | return 0; |
748446bb MG |
484 | } |
485 | ||
486 | /* Time to isolate some pages for migration */ | |
b2eef8c0 | 487 | cond_resched(); |
748446bb | 488 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 489 | /* give a chance to irqs before checking need_resched() */ |
2a1402aa MG |
490 | if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) { |
491 | if (should_release_lock(&zone->lru_lock)) { | |
492 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
493 | locked = false; | |
494 | } | |
b2eef8c0 | 495 | } |
c67fe375 | 496 | |
0bf380bc MG |
497 | /* |
498 | * migrate_pfn does not necessarily start aligned to a | |
499 | * pageblock. Ensure that pfn_valid is called when moving | |
500 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
501 | * memory holes within the zone | |
502 | */ | |
503 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
504 | if (!pfn_valid(low_pfn)) { | |
505 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
506 | continue; | |
507 | } | |
508 | } | |
509 | ||
748446bb MG |
510 | if (!pfn_valid_within(low_pfn)) |
511 | continue; | |
b7aba698 | 512 | nr_scanned++; |
748446bb | 513 | |
dc908600 MG |
514 | /* |
515 | * Get the page and ensure the page is within the same zone. | |
516 | * See the comment in isolate_freepages about overlapping | |
517 | * nodes. It is deliberate that the new zone lock is not taken | |
518 | * as memory compaction should not move pages between nodes. | |
519 | */ | |
748446bb | 520 | page = pfn_to_page(low_pfn); |
dc908600 MG |
521 | if (page_zone(page) != zone) |
522 | continue; | |
523 | ||
bb13ffeb MG |
524 | if (!valid_page) |
525 | valid_page = page; | |
526 | ||
527 | /* If isolation recently failed, do not retry */ | |
528 | pageblock_nr = low_pfn >> pageblock_order; | |
529 | if (!isolation_suitable(cc, page)) | |
530 | goto next_pageblock; | |
531 | ||
6c14466c MG |
532 | /* |
533 | * Skip if free. page_order cannot be used without zone->lock | |
534 | * as nothing prevents parallel allocations or buddy merging. | |
535 | */ | |
748446bb MG |
536 | if (PageBuddy(page)) |
537 | continue; | |
538 | ||
9927af74 MG |
539 | /* |
540 | * For async migration, also only scan in MOVABLE blocks. Async | |
541 | * migration is optimistic to see if the minimum amount of work | |
542 | * satisfies the allocation | |
543 | */ | |
68e3e926 | 544 | if (!cc->sync && last_pageblock_nr != pageblock_nr && |
47118af0 | 545 | !migrate_async_suitable(get_pageblock_migratetype(page))) { |
c89511ab | 546 | cc->finished_update_migrate = true; |
50b5b094 | 547 | skipped_async_unsuitable = true; |
2a1402aa | 548 | goto next_pageblock; |
9927af74 MG |
549 | } |
550 | ||
bf6bddf1 RA |
551 | /* |
552 | * Check may be lockless but that's ok as we recheck later. | |
553 | * It's possible to migrate LRU pages and balloon pages | |
554 | * Skip any other type of page | |
555 | */ | |
556 | if (!PageLRU(page)) { | |
557 | if (unlikely(balloon_page_movable(page))) { | |
558 | if (locked && balloon_page_isolate(page)) { | |
559 | /* Successfully isolated */ | |
560 | cc->finished_update_migrate = true; | |
561 | list_add(&page->lru, migratelist); | |
562 | cc->nr_migratepages++; | |
563 | nr_isolated++; | |
564 | goto check_compact_cluster; | |
565 | } | |
566 | } | |
bc835011 | 567 | continue; |
bf6bddf1 | 568 | } |
bc835011 AA |
569 | |
570 | /* | |
2a1402aa MG |
571 | * PageLRU is set. lru_lock normally excludes isolation |
572 | * splitting and collapsing (collapsing has already happened | |
573 | * if PageLRU is set) but the lock is not necessarily taken | |
574 | * here and it is wasteful to take it just to check transhuge. | |
575 | * Check TransHuge without lock and skip the whole pageblock if | |
576 | * it's either a transhuge or hugetlbfs page, as calling | |
577 | * compound_order() without preventing THP from splitting the | |
578 | * page underneath us may return surprising results. | |
bc835011 | 579 | */ |
2a1402aa MG |
580 | if (PageTransHuge(page)) { |
581 | if (!locked) | |
582 | goto next_pageblock; | |
583 | low_pfn += (1 << compound_order(page)) - 1; | |
584 | continue; | |
585 | } | |
586 | ||
587 | /* Check if it is ok to still hold the lock */ | |
588 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
589 | locked, cc); | |
590 | if (!locked || fatal_signal_pending(current)) | |
591 | break; | |
592 | ||
593 | /* Recheck PageLRU and PageTransHuge under lock */ | |
594 | if (!PageLRU(page)) | |
595 | continue; | |
bc835011 AA |
596 | if (PageTransHuge(page)) { |
597 | low_pfn += (1 << compound_order(page)) - 1; | |
598 | continue; | |
599 | } | |
600 | ||
68e3e926 | 601 | if (!cc->sync) |
c8244935 MG |
602 | mode |= ISOLATE_ASYNC_MIGRATE; |
603 | ||
e46a2879 MK |
604 | if (unevictable) |
605 | mode |= ISOLATE_UNEVICTABLE; | |
606 | ||
fa9add64 HD |
607 | lruvec = mem_cgroup_page_lruvec(page, zone); |
608 | ||
748446bb | 609 | /* Try isolate the page */ |
f3fd4a61 | 610 | if (__isolate_lru_page(page, mode) != 0) |
748446bb MG |
611 | continue; |
612 | ||
309381fe | 613 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 614 | |
748446bb | 615 | /* Successfully isolated */ |
c89511ab | 616 | cc->finished_update_migrate = true; |
fa9add64 | 617 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
748446bb | 618 | list_add(&page->lru, migratelist); |
748446bb | 619 | cc->nr_migratepages++; |
b7aba698 | 620 | nr_isolated++; |
748446bb | 621 | |
bf6bddf1 | 622 | check_compact_cluster: |
748446bb | 623 | /* Avoid isolating too much */ |
31b8384a HD |
624 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
625 | ++low_pfn; | |
748446bb | 626 | break; |
31b8384a | 627 | } |
2a1402aa MG |
628 | |
629 | continue; | |
630 | ||
631 | next_pageblock: | |
a9aacbcc | 632 | low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1; |
2a1402aa | 633 | last_pageblock_nr = pageblock_nr; |
748446bb MG |
634 | } |
635 | ||
c67fe375 | 636 | acct_isolated(zone, locked, cc); |
748446bb | 637 | |
c67fe375 MG |
638 | if (locked) |
639 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 640 | |
50b5b094 VB |
641 | /* |
642 | * Update the pageblock-skip information and cached scanner pfn, | |
643 | * if the whole pageblock was scanned without isolating any page. | |
644 | * This is not done when pageblock was skipped due to being unsuitable | |
645 | * for async compaction, so that eventual sync compaction can try. | |
646 | */ | |
647 | if (low_pfn == end_pfn && !skipped_async_unsuitable) | |
c89511ab | 648 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 649 | |
b7aba698 MG |
650 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
651 | ||
010fc29a | 652 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 653 | if (nr_isolated) |
010fc29a | 654 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 655 | |
2fe86e00 MN |
656 | return low_pfn; |
657 | } | |
658 | ||
ff9543fd MN |
659 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
660 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 661 | /* |
ff9543fd MN |
662 | * Based on information in the current compact_control, find blocks |
663 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 664 | */ |
ff9543fd MN |
665 | static void isolate_freepages(struct zone *zone, |
666 | struct compact_control *cc) | |
2fe86e00 | 667 | { |
ff9543fd | 668 | struct page *page; |
108bcc96 | 669 | unsigned long high_pfn, low_pfn, pfn, z_end_pfn, end_pfn; |
ff9543fd MN |
670 | int nr_freepages = cc->nr_freepages; |
671 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 672 | |
ff9543fd MN |
673 | /* |
674 | * Initialise the free scanner. The starting point is where we last | |
675 | * scanned from (or the end of the zone if starting). The low point | |
676 | * is the end of the pageblock the migration scanner is using. | |
677 | */ | |
678 | pfn = cc->free_pfn; | |
7ed695e0 | 679 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 680 | |
ff9543fd MN |
681 | /* |
682 | * Take care that if the migration scanner is at the end of the zone | |
683 | * that the free scanner does not accidentally move to the next zone | |
684 | * in the next isolation cycle. | |
685 | */ | |
686 | high_pfn = min(low_pfn, pfn); | |
2fe86e00 | 687 | |
108bcc96 | 688 | z_end_pfn = zone_end_pfn(zone); |
2fe86e00 | 689 | |
ff9543fd MN |
690 | /* |
691 | * Isolate free pages until enough are available to migrate the | |
692 | * pages on cc->migratepages. We stop searching if the migrate | |
693 | * and free page scanners meet or enough free pages are isolated. | |
694 | */ | |
7ed695e0 | 695 | for (; pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
ff9543fd MN |
696 | pfn -= pageblock_nr_pages) { |
697 | unsigned long isolated; | |
2fe86e00 | 698 | |
f6ea3adb DR |
699 | /* |
700 | * This can iterate a massively long zone without finding any | |
701 | * suitable migration targets, so periodically check if we need | |
702 | * to schedule. | |
703 | */ | |
704 | cond_resched(); | |
705 | ||
ff9543fd MN |
706 | if (!pfn_valid(pfn)) |
707 | continue; | |
2fe86e00 | 708 | |
ff9543fd MN |
709 | /* |
710 | * Check for overlapping nodes/zones. It's possible on some | |
711 | * configurations to have a setup like | |
712 | * node0 node1 node0 | |
713 | * i.e. it's possible that all pages within a zones range of | |
714 | * pages do not belong to a single zone. | |
715 | */ | |
716 | page = pfn_to_page(pfn); | |
717 | if (page_zone(page) != zone) | |
718 | continue; | |
719 | ||
720 | /* Check the block is suitable for migration */ | |
68e3e926 | 721 | if (!suitable_migration_target(page)) |
ff9543fd | 722 | continue; |
68e3e926 | 723 | |
bb13ffeb MG |
724 | /* If isolation recently failed, do not retry */ |
725 | if (!isolation_suitable(cc, page)) | |
726 | continue; | |
727 | ||
f40d1e42 | 728 | /* Found a block suitable for isolating free pages from */ |
ff9543fd | 729 | isolated = 0; |
60177d31 MG |
730 | |
731 | /* | |
732 | * As pfn may not start aligned, pfn+pageblock_nr_page | |
733 | * may cross a MAX_ORDER_NR_PAGES boundary and miss | |
734 | * a pfn_valid check. Ensure isolate_freepages_block() | |
735 | * only scans within a pageblock | |
736 | */ | |
737 | end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
108bcc96 | 738 | end_pfn = min(end_pfn, z_end_pfn); |
f40d1e42 MG |
739 | isolated = isolate_freepages_block(cc, pfn, end_pfn, |
740 | freelist, false); | |
741 | nr_freepages += isolated; | |
ff9543fd MN |
742 | |
743 | /* | |
744 | * Record the highest PFN we isolated pages from. When next | |
745 | * looking for free pages, the search will restart here as | |
746 | * page migration may have returned some pages to the allocator | |
747 | */ | |
c89511ab MG |
748 | if (isolated) { |
749 | cc->finished_update_free = true; | |
ff9543fd | 750 | high_pfn = max(high_pfn, pfn); |
c89511ab | 751 | } |
ff9543fd MN |
752 | } |
753 | ||
754 | /* split_free_page does not map the pages */ | |
755 | map_pages(freelist); | |
756 | ||
7ed695e0 VB |
757 | /* |
758 | * If we crossed the migrate scanner, we want to keep it that way | |
759 | * so that compact_finished() may detect this | |
760 | */ | |
761 | if (pfn < low_pfn) | |
762 | cc->free_pfn = max(pfn, zone->zone_start_pfn); | |
763 | else | |
764 | cc->free_pfn = high_pfn; | |
ff9543fd | 765 | cc->nr_freepages = nr_freepages; |
748446bb MG |
766 | } |
767 | ||
768 | /* | |
769 | * This is a migrate-callback that "allocates" freepages by taking pages | |
770 | * from the isolated freelists in the block we are migrating to. | |
771 | */ | |
772 | static struct page *compaction_alloc(struct page *migratepage, | |
773 | unsigned long data, | |
774 | int **result) | |
775 | { | |
776 | struct compact_control *cc = (struct compact_control *)data; | |
777 | struct page *freepage; | |
778 | ||
779 | /* Isolate free pages if necessary */ | |
780 | if (list_empty(&cc->freepages)) { | |
781 | isolate_freepages(cc->zone, cc); | |
782 | ||
783 | if (list_empty(&cc->freepages)) | |
784 | return NULL; | |
785 | } | |
786 | ||
787 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
788 | list_del(&freepage->lru); | |
789 | cc->nr_freepages--; | |
790 | ||
791 | return freepage; | |
792 | } | |
793 | ||
794 | /* | |
795 | * We cannot control nr_migratepages and nr_freepages fully when migration is | |
796 | * running as migrate_pages() has no knowledge of compact_control. When | |
797 | * migration is complete, we count the number of pages on the lists by hand. | |
798 | */ | |
799 | static void update_nr_listpages(struct compact_control *cc) | |
800 | { | |
801 | int nr_migratepages = 0; | |
802 | int nr_freepages = 0; | |
803 | struct page *page; | |
804 | ||
805 | list_for_each_entry(page, &cc->migratepages, lru) | |
806 | nr_migratepages++; | |
807 | list_for_each_entry(page, &cc->freepages, lru) | |
808 | nr_freepages++; | |
809 | ||
810 | cc->nr_migratepages = nr_migratepages; | |
811 | cc->nr_freepages = nr_freepages; | |
812 | } | |
813 | ||
ff9543fd MN |
814 | /* possible outcome of isolate_migratepages */ |
815 | typedef enum { | |
816 | ISOLATE_ABORT, /* Abort compaction now */ | |
817 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
818 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
819 | } isolate_migrate_t; | |
820 | ||
821 | /* | |
822 | * Isolate all pages that can be migrated from the block pointed to by | |
823 | * the migrate scanner within compact_control. | |
824 | */ | |
825 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
826 | struct compact_control *cc) | |
827 | { | |
828 | unsigned long low_pfn, end_pfn; | |
829 | ||
830 | /* Do not scan outside zone boundaries */ | |
831 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
832 | ||
833 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 834 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd MN |
835 | |
836 | /* Do not cross the free scanner or scan within a memory hole */ | |
837 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
838 | cc->migrate_pfn = end_pfn; | |
839 | return ISOLATE_NONE; | |
840 | } | |
841 | ||
842 | /* Perform the isolation */ | |
e46a2879 | 843 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
e64c5237 | 844 | if (!low_pfn || cc->contended) |
ff9543fd MN |
845 | return ISOLATE_ABORT; |
846 | ||
847 | cc->migrate_pfn = low_pfn; | |
848 | ||
849 | return ISOLATE_SUCCESS; | |
850 | } | |
851 | ||
748446bb | 852 | static int compact_finished(struct zone *zone, |
5a03b051 | 853 | struct compact_control *cc) |
748446bb | 854 | { |
8fb74b9f | 855 | unsigned int order; |
5a03b051 | 856 | unsigned long watermark; |
56de7263 | 857 | |
748446bb MG |
858 | if (fatal_signal_pending(current)) |
859 | return COMPACT_PARTIAL; | |
860 | ||
753341a4 | 861 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 862 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 VB |
863 | /* Let the next compaction start anew. */ |
864 | zone->compact_cached_migrate_pfn = zone->zone_start_pfn; | |
865 | zone->compact_cached_free_pfn = zone_end_pfn(zone); | |
866 | ||
62997027 MG |
867 | /* |
868 | * Mark that the PG_migrate_skip information should be cleared | |
869 | * by kswapd when it goes to sleep. kswapd does not set the | |
870 | * flag itself as the decision to be clear should be directly | |
871 | * based on an allocation request. | |
872 | */ | |
873 | if (!current_is_kswapd()) | |
874 | zone->compact_blockskip_flush = true; | |
875 | ||
748446bb | 876 | return COMPACT_COMPLETE; |
bb13ffeb | 877 | } |
748446bb | 878 | |
82478fb7 JW |
879 | /* |
880 | * order == -1 is expected when compacting via | |
881 | * /proc/sys/vm/compact_memory | |
882 | */ | |
56de7263 MG |
883 | if (cc->order == -1) |
884 | return COMPACT_CONTINUE; | |
885 | ||
3957c776 MH |
886 | /* Compaction run is not finished if the watermark is not met */ |
887 | watermark = low_wmark_pages(zone); | |
888 | watermark += (1 << cc->order); | |
889 | ||
890 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
891 | return COMPACT_CONTINUE; | |
892 | ||
56de7263 | 893 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
894 | for (order = cc->order; order < MAX_ORDER; order++) { |
895 | struct free_area *area = &zone->free_area[order]; | |
896 | ||
897 | /* Job done if page is free of the right migratetype */ | |
898 | if (!list_empty(&area->free_list[cc->migratetype])) | |
899 | return COMPACT_PARTIAL; | |
900 | ||
901 | /* Job done if allocation would set block type */ | |
902 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
903 | return COMPACT_PARTIAL; |
904 | } | |
905 | ||
748446bb MG |
906 | return COMPACT_CONTINUE; |
907 | } | |
908 | ||
3e7d3449 MG |
909 | /* |
910 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
911 | * Returns | |
912 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
913 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
914 | * COMPACT_CONTINUE - If compaction should run now | |
915 | */ | |
916 | unsigned long compaction_suitable(struct zone *zone, int order) | |
917 | { | |
918 | int fragindex; | |
919 | unsigned long watermark; | |
920 | ||
3957c776 MH |
921 | /* |
922 | * order == -1 is expected when compacting via | |
923 | * /proc/sys/vm/compact_memory | |
924 | */ | |
925 | if (order == -1) | |
926 | return COMPACT_CONTINUE; | |
927 | ||
3e7d3449 MG |
928 | /* |
929 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
930 | * This is because during migration, copies of pages need to be | |
931 | * allocated and for a short time, the footprint is higher | |
932 | */ | |
933 | watermark = low_wmark_pages(zone) + (2UL << order); | |
934 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
935 | return COMPACT_SKIPPED; | |
936 | ||
937 | /* | |
938 | * fragmentation index determines if allocation failures are due to | |
939 | * low memory or external fragmentation | |
940 | * | |
a582a738 SL |
941 | * index of -1000 implies allocations might succeed depending on |
942 | * watermarks | |
3e7d3449 MG |
943 | * index towards 0 implies failure is due to lack of memory |
944 | * index towards 1000 implies failure is due to fragmentation | |
945 | * | |
946 | * Only compact if a failure would be due to fragmentation. | |
947 | */ | |
948 | fragindex = fragmentation_index(zone, order); | |
949 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
950 | return COMPACT_SKIPPED; | |
951 | ||
a582a738 SL |
952 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
953 | 0, 0)) | |
3e7d3449 MG |
954 | return COMPACT_PARTIAL; |
955 | ||
956 | return COMPACT_CONTINUE; | |
957 | } | |
958 | ||
748446bb MG |
959 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
960 | { | |
961 | int ret; | |
c89511ab | 962 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 963 | unsigned long end_pfn = zone_end_pfn(zone); |
748446bb | 964 | |
3e7d3449 MG |
965 | ret = compaction_suitable(zone, cc->order); |
966 | switch (ret) { | |
967 | case COMPACT_PARTIAL: | |
968 | case COMPACT_SKIPPED: | |
969 | /* Compaction is likely to fail */ | |
970 | return ret; | |
971 | case COMPACT_CONTINUE: | |
972 | /* Fall through to compaction */ | |
973 | ; | |
974 | } | |
975 | ||
d3132e4b VB |
976 | /* |
977 | * Clear pageblock skip if there were failures recently and compaction | |
978 | * is about to be retried after being deferred. kswapd does not do | |
979 | * this reset as it'll reset the cached information when going to sleep. | |
980 | */ | |
981 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
982 | __reset_isolation_suitable(zone); | |
983 | ||
c89511ab MG |
984 | /* |
985 | * Setup to move all movable pages to the end of the zone. Used cached | |
986 | * information on where the scanners should start but check that it | |
987 | * is initialised by ensuring the values are within zone boundaries. | |
988 | */ | |
989 | cc->migrate_pfn = zone->compact_cached_migrate_pfn; | |
990 | cc->free_pfn = zone->compact_cached_free_pfn; | |
991 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
992 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
993 | zone->compact_cached_free_pfn = cc->free_pfn; | |
994 | } | |
995 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
996 | cc->migrate_pfn = start_pfn; | |
997 | zone->compact_cached_migrate_pfn = cc->migrate_pfn; | |
998 | } | |
748446bb | 999 | |
0eb927c0 MG |
1000 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn); |
1001 | ||
748446bb MG |
1002 | migrate_prep_local(); |
1003 | ||
1004 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
1005 | unsigned long nr_migrate, nr_remaining; | |
9d502c1c | 1006 | int err; |
748446bb | 1007 | |
f9e35b3b MG |
1008 | switch (isolate_migratepages(zone, cc)) { |
1009 | case ISOLATE_ABORT: | |
1010 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1011 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1012 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1013 | goto out; |
1014 | case ISOLATE_NONE: | |
748446bb | 1015 | continue; |
f9e35b3b MG |
1016 | case ISOLATE_SUCCESS: |
1017 | ; | |
1018 | } | |
748446bb MG |
1019 | |
1020 | nr_migrate = cc->nr_migratepages; | |
9d502c1c | 1021 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
9c620e2b | 1022 | (unsigned long)cc, |
7b2a2d4a MG |
1023 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC, |
1024 | MR_COMPACTION); | |
748446bb MG |
1025 | update_nr_listpages(cc); |
1026 | nr_remaining = cc->nr_migratepages; | |
1027 | ||
b7aba698 MG |
1028 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
1029 | nr_remaining); | |
748446bb | 1030 | |
5733c7d1 | 1031 | /* Release isolated pages not migrated */ |
9d502c1c | 1032 | if (err) { |
5733c7d1 | 1033 | putback_movable_pages(&cc->migratepages); |
748446bb | 1034 | cc->nr_migratepages = 0; |
7ed695e0 VB |
1035 | /* |
1036 | * migrate_pages() may return -ENOMEM when scanners meet | |
1037 | * and we want compact_finished() to detect it | |
1038 | */ | |
1039 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1040 | ret = COMPACT_PARTIAL; |
1041 | goto out; | |
1042 | } | |
748446bb | 1043 | } |
748446bb MG |
1044 | } |
1045 | ||
f9e35b3b | 1046 | out: |
748446bb MG |
1047 | /* Release free pages and check accounting */ |
1048 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1049 | VM_BUG_ON(cc->nr_freepages != 0); | |
1050 | ||
0eb927c0 MG |
1051 | trace_mm_compaction_end(ret); |
1052 | ||
748446bb MG |
1053 | return ret; |
1054 | } | |
76ab0f53 | 1055 | |
d43a87e6 | 1056 | static unsigned long compact_zone_order(struct zone *zone, |
5a03b051 | 1057 | int order, gfp_t gfp_mask, |
8fb74b9f | 1058 | bool sync, bool *contended) |
56de7263 | 1059 | { |
e64c5237 | 1060 | unsigned long ret; |
56de7263 MG |
1061 | struct compact_control cc = { |
1062 | .nr_freepages = 0, | |
1063 | .nr_migratepages = 0, | |
1064 | .order = order, | |
1065 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1066 | .zone = zone, | |
68e3e926 | 1067 | .sync = sync, |
56de7263 MG |
1068 | }; |
1069 | INIT_LIST_HEAD(&cc.freepages); | |
1070 | INIT_LIST_HEAD(&cc.migratepages); | |
1071 | ||
e64c5237 SL |
1072 | ret = compact_zone(zone, &cc); |
1073 | ||
1074 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1075 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1076 | ||
1077 | *contended = cc.contended; | |
1078 | return ret; | |
56de7263 MG |
1079 | } |
1080 | ||
5e771905 MG |
1081 | int sysctl_extfrag_threshold = 500; |
1082 | ||
56de7263 MG |
1083 | /** |
1084 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1085 | * @zonelist: The zonelist used for the current allocation | |
1086 | * @order: The order of the current allocation | |
1087 | * @gfp_mask: The GFP mask of the current allocation | |
1088 | * @nodemask: The allowed nodes to allocate from | |
77f1fe6b | 1089 | * @sync: Whether migration is synchronous or not |
661c4cb9 MG |
1090 | * @contended: Return value that is true if compaction was aborted due to lock contention |
1091 | * @page: Optionally capture a free page of the requested order during compaction | |
56de7263 MG |
1092 | * |
1093 | * This is the main entry point for direct page compaction. | |
1094 | */ | |
1095 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1096 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
8fb74b9f | 1097 | bool sync, bool *contended) |
56de7263 MG |
1098 | { |
1099 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1100 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1101 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1102 | struct zoneref *z; |
1103 | struct zone *zone; | |
1104 | int rc = COMPACT_SKIPPED; | |
d95ea5d1 | 1105 | int alloc_flags = 0; |
56de7263 | 1106 | |
4ffb6335 | 1107 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1108 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
1109 | return rc; |
1110 | ||
010fc29a | 1111 | count_compact_event(COMPACTSTALL); |
56de7263 | 1112 | |
d95ea5d1 BZ |
1113 | #ifdef CONFIG_CMA |
1114 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1115 | alloc_flags |= ALLOC_CMA; | |
1116 | #endif | |
56de7263 MG |
1117 | /* Compact each zone in the list */ |
1118 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1119 | nodemask) { | |
56de7263 MG |
1120 | int status; |
1121 | ||
c67fe375 | 1122 | status = compact_zone_order(zone, order, gfp_mask, sync, |
8fb74b9f | 1123 | contended); |
56de7263 MG |
1124 | rc = max(status, rc); |
1125 | ||
3e7d3449 | 1126 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 BZ |
1127 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
1128 | alloc_flags)) | |
56de7263 MG |
1129 | break; |
1130 | } | |
1131 | ||
1132 | return rc; | |
1133 | } | |
1134 | ||
1135 | ||
76ab0f53 | 1136 | /* Compact all zones within a node */ |
7103f16d | 1137 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1138 | { |
1139 | int zoneid; | |
76ab0f53 MG |
1140 | struct zone *zone; |
1141 | ||
76ab0f53 | 1142 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1143 | |
1144 | zone = &pgdat->node_zones[zoneid]; | |
1145 | if (!populated_zone(zone)) | |
1146 | continue; | |
1147 | ||
7be62de9 RR |
1148 | cc->nr_freepages = 0; |
1149 | cc->nr_migratepages = 0; | |
1150 | cc->zone = zone; | |
1151 | INIT_LIST_HEAD(&cc->freepages); | |
1152 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1153 | |
aad6ec37 | 1154 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1155 | compact_zone(zone, cc); |
76ab0f53 | 1156 | |
aff62249 | 1157 | if (cc->order > 0) { |
de6c60a6 VB |
1158 | if (zone_watermark_ok(zone, cc->order, |
1159 | low_wmark_pages(zone), 0, 0)) | |
1160 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 | 1161 | /* Currently async compaction is never deferred. */ |
de6c60a6 | 1162 | else if (cc->sync) |
aff62249 RR |
1163 | defer_compaction(zone, cc->order); |
1164 | } | |
1165 | ||
7be62de9 RR |
1166 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1167 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1168 | } |
76ab0f53 MG |
1169 | } |
1170 | ||
7103f16d | 1171 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1172 | { |
1173 | struct compact_control cc = { | |
1174 | .order = order, | |
68e3e926 | 1175 | .sync = false, |
7be62de9 RR |
1176 | }; |
1177 | ||
3a7200af MG |
1178 | if (!order) |
1179 | return; | |
1180 | ||
7103f16d | 1181 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1182 | } |
1183 | ||
7103f16d | 1184 | static void compact_node(int nid) |
7be62de9 | 1185 | { |
7be62de9 RR |
1186 | struct compact_control cc = { |
1187 | .order = -1, | |
68e3e926 | 1188 | .sync = true, |
7be62de9 RR |
1189 | }; |
1190 | ||
7103f16d | 1191 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1192 | } |
1193 | ||
76ab0f53 | 1194 | /* Compact all nodes in the system */ |
7964c06d | 1195 | static void compact_nodes(void) |
76ab0f53 MG |
1196 | { |
1197 | int nid; | |
1198 | ||
8575ec29 HD |
1199 | /* Flush pending updates to the LRU lists */ |
1200 | lru_add_drain_all(); | |
1201 | ||
76ab0f53 MG |
1202 | for_each_online_node(nid) |
1203 | compact_node(nid); | |
76ab0f53 MG |
1204 | } |
1205 | ||
1206 | /* The written value is actually unused, all memory is compacted */ | |
1207 | int sysctl_compact_memory; | |
1208 | ||
1209 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1210 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1211 | void __user *buffer, size_t *length, loff_t *ppos) | |
1212 | { | |
1213 | if (write) | |
7964c06d | 1214 | compact_nodes(); |
76ab0f53 MG |
1215 | |
1216 | return 0; | |
1217 | } | |
ed4a6d7f | 1218 | |
5e771905 MG |
1219 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1220 | void __user *buffer, size_t *length, loff_t *ppos) | |
1221 | { | |
1222 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1223 | ||
1224 | return 0; | |
1225 | } | |
1226 | ||
ed4a6d7f | 1227 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
10fbcf4c KS |
1228 | ssize_t sysfs_compact_node(struct device *dev, |
1229 | struct device_attribute *attr, | |
ed4a6d7f MG |
1230 | const char *buf, size_t count) |
1231 | { | |
8575ec29 HD |
1232 | int nid = dev->id; |
1233 | ||
1234 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1235 | /* Flush pending updates to the LRU lists */ | |
1236 | lru_add_drain_all(); | |
1237 | ||
1238 | compact_node(nid); | |
1239 | } | |
ed4a6d7f MG |
1240 | |
1241 | return count; | |
1242 | } | |
10fbcf4c | 1243 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1244 | |
1245 | int compaction_register_node(struct node *node) | |
1246 | { | |
10fbcf4c | 1247 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1248 | } |
1249 | ||
1250 | void compaction_unregister_node(struct node *node) | |
1251 | { | |
10fbcf4c | 1252 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1253 | } |
1254 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1255 | |
1256 | #endif /* CONFIG_COMPACTION */ |