]>
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 | { | |
7d348b9e | 220 | /* If the page is a large free page, then disallow migration */ |
f40d1e42 | 221 | if (PageBuddy(page) && page_order(page) >= pageblock_order) |
7d348b9e | 222 | return false; |
f40d1e42 MG |
223 | |
224 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
7d348b9e | 225 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
f40d1e42 MG |
226 | return true; |
227 | ||
228 | /* Otherwise skip the block */ | |
229 | return false; | |
230 | } | |
231 | ||
85aa125f | 232 | /* |
9e4be470 JM |
233 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
234 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
235 | * (even though it may still end up isolating some pages). | |
85aa125f | 236 | */ |
f40d1e42 MG |
237 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
238 | unsigned long blockpfn, | |
85aa125f MN |
239 | unsigned long end_pfn, |
240 | struct list_head *freelist, | |
241 | bool strict) | |
748446bb | 242 | { |
b7aba698 | 243 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 244 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
245 | unsigned long flags; |
246 | bool locked = false; | |
01ead534 | 247 | bool checked_pageblock = false; |
748446bb | 248 | |
748446bb MG |
249 | cursor = pfn_to_page(blockpfn); |
250 | ||
f40d1e42 | 251 | /* Isolate free pages. */ |
748446bb MG |
252 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
253 | int isolated, i; | |
254 | struct page *page = cursor; | |
255 | ||
b7aba698 | 256 | nr_scanned++; |
f40d1e42 | 257 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
258 | goto isolate_fail; |
259 | ||
bb13ffeb MG |
260 | if (!valid_page) |
261 | valid_page = page; | |
f40d1e42 | 262 | if (!PageBuddy(page)) |
2af120bc | 263 | goto isolate_fail; |
f40d1e42 MG |
264 | |
265 | /* | |
266 | * The zone lock must be held to isolate freepages. | |
267 | * Unfortunately this is a very coarse lock and can be | |
268 | * heavily contended if there are parallel allocations | |
269 | * or parallel compactions. For async compaction do not | |
270 | * spin on the lock and we acquire the lock as late as | |
271 | * possible. | |
272 | */ | |
273 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
274 | locked, cc); | |
275 | if (!locked) | |
276 | break; | |
277 | ||
278 | /* Recheck this is a suitable migration target under lock */ | |
01ead534 JK |
279 | if (!strict && !checked_pageblock) { |
280 | /* | |
281 | * We need to check suitability of pageblock only once | |
282 | * and this isolate_freepages_block() is called with | |
283 | * pageblock range, so just check once is sufficient. | |
284 | */ | |
285 | checked_pageblock = true; | |
286 | if (!suitable_migration_target(page)) | |
287 | break; | |
288 | } | |
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; |
fa9add64 | 463 | struct lruvec *lruvec; |
c67fe375 | 464 | unsigned long flags; |
2a1402aa | 465 | bool locked = false; |
bb13ffeb | 466 | struct page *page = NULL, *valid_page = NULL; |
50b5b094 | 467 | bool skipped_async_unsuitable = false; |
da1c67a7 DR |
468 | const isolate_mode_t mode = (!cc->sync ? ISOLATE_ASYNC_MIGRATE : 0) | |
469 | (unevictable ? ISOLATE_UNEVICTABLE : 0); | |
748446bb | 470 | |
748446bb MG |
471 | /* |
472 | * Ensure that there are not too many pages isolated from the LRU | |
473 | * list by either parallel reclaimers or compaction. If there are, | |
474 | * delay for some time until fewer pages are isolated | |
475 | */ | |
476 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 477 | /* async migration should just abort */ |
68e3e926 | 478 | if (!cc->sync) |
2fe86e00 | 479 | return 0; |
f9e35b3b | 480 | |
748446bb MG |
481 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
482 | ||
483 | if (fatal_signal_pending(current)) | |
2fe86e00 | 484 | return 0; |
748446bb MG |
485 | } |
486 | ||
487 | /* Time to isolate some pages for migration */ | |
b2eef8c0 | 488 | cond_resched(); |
748446bb | 489 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 490 | /* give a chance to irqs before checking need_resched() */ |
be1aa03b | 491 | if (locked && !(low_pfn % SWAP_CLUSTER_MAX)) { |
2a1402aa MG |
492 | if (should_release_lock(&zone->lru_lock)) { |
493 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
494 | locked = false; | |
495 | } | |
b2eef8c0 | 496 | } |
c67fe375 | 497 | |
0bf380bc MG |
498 | /* |
499 | * migrate_pfn does not necessarily start aligned to a | |
500 | * pageblock. Ensure that pfn_valid is called when moving | |
501 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
502 | * memory holes within the zone | |
503 | */ | |
504 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
505 | if (!pfn_valid(low_pfn)) { | |
506 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
507 | continue; | |
508 | } | |
509 | } | |
510 | ||
748446bb MG |
511 | if (!pfn_valid_within(low_pfn)) |
512 | continue; | |
b7aba698 | 513 | nr_scanned++; |
748446bb | 514 | |
dc908600 MG |
515 | /* |
516 | * Get the page and ensure the page is within the same zone. | |
517 | * See the comment in isolate_freepages about overlapping | |
518 | * nodes. It is deliberate that the new zone lock is not taken | |
519 | * as memory compaction should not move pages between nodes. | |
520 | */ | |
748446bb | 521 | page = pfn_to_page(low_pfn); |
dc908600 MG |
522 | if (page_zone(page) != zone) |
523 | continue; | |
524 | ||
bb13ffeb MG |
525 | if (!valid_page) |
526 | valid_page = page; | |
527 | ||
528 | /* If isolation recently failed, do not retry */ | |
529 | pageblock_nr = low_pfn >> pageblock_order; | |
c122b208 JK |
530 | if (last_pageblock_nr != pageblock_nr) { |
531 | int mt; | |
532 | ||
533 | last_pageblock_nr = pageblock_nr; | |
534 | if (!isolation_suitable(cc, page)) | |
535 | goto next_pageblock; | |
536 | ||
537 | /* | |
538 | * For async migration, also only scan in MOVABLE | |
539 | * blocks. Async migration is optimistic to see if | |
540 | * the minimum amount of work satisfies the allocation | |
541 | */ | |
542 | mt = get_pageblock_migratetype(page); | |
543 | if (!cc->sync && !migrate_async_suitable(mt)) { | |
544 | cc->finished_update_migrate = true; | |
545 | skipped_async_unsuitable = true; | |
546 | goto next_pageblock; | |
547 | } | |
548 | } | |
bb13ffeb | 549 | |
6c14466c MG |
550 | /* |
551 | * Skip if free. page_order cannot be used without zone->lock | |
552 | * as nothing prevents parallel allocations or buddy merging. | |
553 | */ | |
748446bb MG |
554 | if (PageBuddy(page)) |
555 | continue; | |
556 | ||
bf6bddf1 RA |
557 | /* |
558 | * Check may be lockless but that's ok as we recheck later. | |
559 | * It's possible to migrate LRU pages and balloon pages | |
560 | * Skip any other type of page | |
561 | */ | |
562 | if (!PageLRU(page)) { | |
563 | if (unlikely(balloon_page_movable(page))) { | |
564 | if (locked && balloon_page_isolate(page)) { | |
565 | /* Successfully isolated */ | |
b6c75016 | 566 | goto isolate_success; |
bf6bddf1 RA |
567 | } |
568 | } | |
bc835011 | 569 | continue; |
bf6bddf1 | 570 | } |
bc835011 AA |
571 | |
572 | /* | |
2a1402aa MG |
573 | * PageLRU is set. lru_lock normally excludes isolation |
574 | * splitting and collapsing (collapsing has already happened | |
575 | * if PageLRU is set) but the lock is not necessarily taken | |
576 | * here and it is wasteful to take it just to check transhuge. | |
577 | * Check TransHuge without lock and skip the whole pageblock if | |
578 | * it's either a transhuge or hugetlbfs page, as calling | |
579 | * compound_order() without preventing THP from splitting the | |
580 | * page underneath us may return surprising results. | |
bc835011 | 581 | */ |
2a1402aa MG |
582 | if (PageTransHuge(page)) { |
583 | if (!locked) | |
584 | goto next_pageblock; | |
585 | low_pfn += (1 << compound_order(page)) - 1; | |
586 | continue; | |
587 | } | |
588 | ||
119d6d59 DR |
589 | /* |
590 | * Migration will fail if an anonymous page is pinned in memory, | |
591 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
592 | * admittedly racy check. | |
593 | */ | |
594 | if (!page_mapping(page) && | |
595 | page_count(page) > page_mapcount(page)) | |
596 | continue; | |
597 | ||
2a1402aa MG |
598 | /* Check if it is ok to still hold the lock */ |
599 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
600 | locked, cc); | |
601 | if (!locked || fatal_signal_pending(current)) | |
602 | break; | |
603 | ||
604 | /* Recheck PageLRU and PageTransHuge under lock */ | |
605 | if (!PageLRU(page)) | |
606 | continue; | |
bc835011 AA |
607 | if (PageTransHuge(page)) { |
608 | low_pfn += (1 << compound_order(page)) - 1; | |
609 | continue; | |
610 | } | |
611 | ||
fa9add64 HD |
612 | lruvec = mem_cgroup_page_lruvec(page, zone); |
613 | ||
748446bb | 614 | /* Try isolate the page */ |
f3fd4a61 | 615 | if (__isolate_lru_page(page, mode) != 0) |
748446bb MG |
616 | continue; |
617 | ||
309381fe | 618 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 619 | |
748446bb | 620 | /* Successfully isolated */ |
fa9add64 | 621 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
622 | |
623 | isolate_success: | |
624 | cc->finished_update_migrate = true; | |
748446bb | 625 | list_add(&page->lru, migratelist); |
748446bb | 626 | cc->nr_migratepages++; |
b7aba698 | 627 | nr_isolated++; |
748446bb MG |
628 | |
629 | /* Avoid isolating too much */ | |
31b8384a HD |
630 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
631 | ++low_pfn; | |
748446bb | 632 | break; |
31b8384a | 633 | } |
2a1402aa MG |
634 | |
635 | continue; | |
636 | ||
637 | next_pageblock: | |
a9aacbcc | 638 | low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1; |
748446bb MG |
639 | } |
640 | ||
c67fe375 | 641 | acct_isolated(zone, locked, cc); |
748446bb | 642 | |
c67fe375 MG |
643 | if (locked) |
644 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 645 | |
50b5b094 VB |
646 | /* |
647 | * Update the pageblock-skip information and cached scanner pfn, | |
648 | * if the whole pageblock was scanned without isolating any page. | |
649 | * This is not done when pageblock was skipped due to being unsuitable | |
650 | * for async compaction, so that eventual sync compaction can try. | |
651 | */ | |
652 | if (low_pfn == end_pfn && !skipped_async_unsuitable) | |
c89511ab | 653 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 654 | |
b7aba698 MG |
655 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
656 | ||
010fc29a | 657 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 658 | if (nr_isolated) |
010fc29a | 659 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 660 | |
2fe86e00 MN |
661 | return low_pfn; |
662 | } | |
663 | ||
ff9543fd MN |
664 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
665 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 666 | /* |
ff9543fd MN |
667 | * Based on information in the current compact_control, find blocks |
668 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 669 | */ |
ff9543fd MN |
670 | static void isolate_freepages(struct zone *zone, |
671 | struct compact_control *cc) | |
2fe86e00 | 672 | { |
ff9543fd | 673 | struct page *page; |
108bcc96 | 674 | unsigned long high_pfn, low_pfn, pfn, z_end_pfn, end_pfn; |
ff9543fd MN |
675 | int nr_freepages = cc->nr_freepages; |
676 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 677 | |
ff9543fd MN |
678 | /* |
679 | * Initialise the free scanner. The starting point is where we last | |
680 | * scanned from (or the end of the zone if starting). The low point | |
681 | * is the end of the pageblock the migration scanner is using. | |
682 | */ | |
683 | pfn = cc->free_pfn; | |
7ed695e0 | 684 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 685 | |
ff9543fd MN |
686 | /* |
687 | * Take care that if the migration scanner is at the end of the zone | |
688 | * that the free scanner does not accidentally move to the next zone | |
689 | * in the next isolation cycle. | |
690 | */ | |
691 | high_pfn = min(low_pfn, pfn); | |
2fe86e00 | 692 | |
108bcc96 | 693 | z_end_pfn = zone_end_pfn(zone); |
2fe86e00 | 694 | |
ff9543fd MN |
695 | /* |
696 | * Isolate free pages until enough are available to migrate the | |
697 | * pages on cc->migratepages. We stop searching if the migrate | |
698 | * and free page scanners meet or enough free pages are isolated. | |
699 | */ | |
7ed695e0 | 700 | for (; pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
ff9543fd MN |
701 | pfn -= pageblock_nr_pages) { |
702 | unsigned long isolated; | |
2fe86e00 | 703 | |
f6ea3adb DR |
704 | /* |
705 | * This can iterate a massively long zone without finding any | |
706 | * suitable migration targets, so periodically check if we need | |
707 | * to schedule. | |
708 | */ | |
709 | cond_resched(); | |
710 | ||
ff9543fd MN |
711 | if (!pfn_valid(pfn)) |
712 | continue; | |
2fe86e00 | 713 | |
ff9543fd MN |
714 | /* |
715 | * Check for overlapping nodes/zones. It's possible on some | |
716 | * configurations to have a setup like | |
717 | * node0 node1 node0 | |
718 | * i.e. it's possible that all pages within a zones range of | |
719 | * pages do not belong to a single zone. | |
720 | */ | |
721 | page = pfn_to_page(pfn); | |
722 | if (page_zone(page) != zone) | |
723 | continue; | |
724 | ||
725 | /* Check the block is suitable for migration */ | |
68e3e926 | 726 | if (!suitable_migration_target(page)) |
ff9543fd | 727 | continue; |
68e3e926 | 728 | |
bb13ffeb MG |
729 | /* If isolation recently failed, do not retry */ |
730 | if (!isolation_suitable(cc, page)) | |
731 | continue; | |
732 | ||
f40d1e42 | 733 | /* Found a block suitable for isolating free pages from */ |
ff9543fd | 734 | isolated = 0; |
60177d31 MG |
735 | |
736 | /* | |
737 | * As pfn may not start aligned, pfn+pageblock_nr_page | |
738 | * may cross a MAX_ORDER_NR_PAGES boundary and miss | |
739 | * a pfn_valid check. Ensure isolate_freepages_block() | |
740 | * only scans within a pageblock | |
741 | */ | |
742 | end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
108bcc96 | 743 | end_pfn = min(end_pfn, z_end_pfn); |
f40d1e42 MG |
744 | isolated = isolate_freepages_block(cc, pfn, end_pfn, |
745 | freelist, false); | |
746 | nr_freepages += isolated; | |
ff9543fd MN |
747 | |
748 | /* | |
749 | * Record the highest PFN we isolated pages from. When next | |
750 | * looking for free pages, the search will restart here as | |
751 | * page migration may have returned some pages to the allocator | |
752 | */ | |
c89511ab MG |
753 | if (isolated) { |
754 | cc->finished_update_free = true; | |
ff9543fd | 755 | high_pfn = max(high_pfn, pfn); |
c89511ab | 756 | } |
ff9543fd MN |
757 | } |
758 | ||
759 | /* split_free_page does not map the pages */ | |
760 | map_pages(freelist); | |
761 | ||
7ed695e0 VB |
762 | /* |
763 | * If we crossed the migrate scanner, we want to keep it that way | |
764 | * so that compact_finished() may detect this | |
765 | */ | |
766 | if (pfn < low_pfn) | |
767 | cc->free_pfn = max(pfn, zone->zone_start_pfn); | |
768 | else | |
769 | cc->free_pfn = high_pfn; | |
ff9543fd | 770 | cc->nr_freepages = nr_freepages; |
748446bb MG |
771 | } |
772 | ||
773 | /* | |
774 | * This is a migrate-callback that "allocates" freepages by taking pages | |
775 | * from the isolated freelists in the block we are migrating to. | |
776 | */ | |
777 | static struct page *compaction_alloc(struct page *migratepage, | |
778 | unsigned long data, | |
779 | int **result) | |
780 | { | |
781 | struct compact_control *cc = (struct compact_control *)data; | |
782 | struct page *freepage; | |
783 | ||
784 | /* Isolate free pages if necessary */ | |
785 | if (list_empty(&cc->freepages)) { | |
786 | isolate_freepages(cc->zone, cc); | |
787 | ||
788 | if (list_empty(&cc->freepages)) | |
789 | return NULL; | |
790 | } | |
791 | ||
792 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
793 | list_del(&freepage->lru); | |
794 | cc->nr_freepages--; | |
795 | ||
796 | return freepage; | |
797 | } | |
798 | ||
799 | /* | |
800 | * We cannot control nr_migratepages and nr_freepages fully when migration is | |
801 | * running as migrate_pages() has no knowledge of compact_control. When | |
802 | * migration is complete, we count the number of pages on the lists by hand. | |
803 | */ | |
804 | static void update_nr_listpages(struct compact_control *cc) | |
805 | { | |
806 | int nr_migratepages = 0; | |
807 | int nr_freepages = 0; | |
808 | struct page *page; | |
809 | ||
810 | list_for_each_entry(page, &cc->migratepages, lru) | |
811 | nr_migratepages++; | |
812 | list_for_each_entry(page, &cc->freepages, lru) | |
813 | nr_freepages++; | |
814 | ||
815 | cc->nr_migratepages = nr_migratepages; | |
816 | cc->nr_freepages = nr_freepages; | |
817 | } | |
818 | ||
ff9543fd MN |
819 | /* possible outcome of isolate_migratepages */ |
820 | typedef enum { | |
821 | ISOLATE_ABORT, /* Abort compaction now */ | |
822 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
823 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
824 | } isolate_migrate_t; | |
825 | ||
826 | /* | |
827 | * Isolate all pages that can be migrated from the block pointed to by | |
828 | * the migrate scanner within compact_control. | |
829 | */ | |
830 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
831 | struct compact_control *cc) | |
832 | { | |
833 | unsigned long low_pfn, end_pfn; | |
834 | ||
835 | /* Do not scan outside zone boundaries */ | |
836 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
837 | ||
838 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 839 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd MN |
840 | |
841 | /* Do not cross the free scanner or scan within a memory hole */ | |
842 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
843 | cc->migrate_pfn = end_pfn; | |
844 | return ISOLATE_NONE; | |
845 | } | |
846 | ||
847 | /* Perform the isolation */ | |
e46a2879 | 848 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
e64c5237 | 849 | if (!low_pfn || cc->contended) |
ff9543fd MN |
850 | return ISOLATE_ABORT; |
851 | ||
852 | cc->migrate_pfn = low_pfn; | |
853 | ||
854 | return ISOLATE_SUCCESS; | |
855 | } | |
856 | ||
748446bb | 857 | static int compact_finished(struct zone *zone, |
5a03b051 | 858 | struct compact_control *cc) |
748446bb | 859 | { |
8fb74b9f | 860 | unsigned int order; |
5a03b051 | 861 | unsigned long watermark; |
56de7263 | 862 | |
748446bb MG |
863 | if (fatal_signal_pending(current)) |
864 | return COMPACT_PARTIAL; | |
865 | ||
753341a4 | 866 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 867 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 VB |
868 | /* Let the next compaction start anew. */ |
869 | zone->compact_cached_migrate_pfn = zone->zone_start_pfn; | |
870 | zone->compact_cached_free_pfn = zone_end_pfn(zone); | |
871 | ||
62997027 MG |
872 | /* |
873 | * Mark that the PG_migrate_skip information should be cleared | |
874 | * by kswapd when it goes to sleep. kswapd does not set the | |
875 | * flag itself as the decision to be clear should be directly | |
876 | * based on an allocation request. | |
877 | */ | |
878 | if (!current_is_kswapd()) | |
879 | zone->compact_blockskip_flush = true; | |
880 | ||
748446bb | 881 | return COMPACT_COMPLETE; |
bb13ffeb | 882 | } |
748446bb | 883 | |
82478fb7 JW |
884 | /* |
885 | * order == -1 is expected when compacting via | |
886 | * /proc/sys/vm/compact_memory | |
887 | */ | |
56de7263 MG |
888 | if (cc->order == -1) |
889 | return COMPACT_CONTINUE; | |
890 | ||
3957c776 MH |
891 | /* Compaction run is not finished if the watermark is not met */ |
892 | watermark = low_wmark_pages(zone); | |
893 | watermark += (1 << cc->order); | |
894 | ||
895 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
896 | return COMPACT_CONTINUE; | |
897 | ||
56de7263 | 898 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
899 | for (order = cc->order; order < MAX_ORDER; order++) { |
900 | struct free_area *area = &zone->free_area[order]; | |
901 | ||
902 | /* Job done if page is free of the right migratetype */ | |
903 | if (!list_empty(&area->free_list[cc->migratetype])) | |
904 | return COMPACT_PARTIAL; | |
905 | ||
906 | /* Job done if allocation would set block type */ | |
907 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
908 | return COMPACT_PARTIAL; |
909 | } | |
910 | ||
748446bb MG |
911 | return COMPACT_CONTINUE; |
912 | } | |
913 | ||
3e7d3449 MG |
914 | /* |
915 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
916 | * Returns | |
917 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
918 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
919 | * COMPACT_CONTINUE - If compaction should run now | |
920 | */ | |
921 | unsigned long compaction_suitable(struct zone *zone, int order) | |
922 | { | |
923 | int fragindex; | |
924 | unsigned long watermark; | |
925 | ||
3957c776 MH |
926 | /* |
927 | * order == -1 is expected when compacting via | |
928 | * /proc/sys/vm/compact_memory | |
929 | */ | |
930 | if (order == -1) | |
931 | return COMPACT_CONTINUE; | |
932 | ||
3e7d3449 MG |
933 | /* |
934 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
935 | * This is because during migration, copies of pages need to be | |
936 | * allocated and for a short time, the footprint is higher | |
937 | */ | |
938 | watermark = low_wmark_pages(zone) + (2UL << order); | |
939 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
940 | return COMPACT_SKIPPED; | |
941 | ||
942 | /* | |
943 | * fragmentation index determines if allocation failures are due to | |
944 | * low memory or external fragmentation | |
945 | * | |
a582a738 SL |
946 | * index of -1000 implies allocations might succeed depending on |
947 | * watermarks | |
3e7d3449 MG |
948 | * index towards 0 implies failure is due to lack of memory |
949 | * index towards 1000 implies failure is due to fragmentation | |
950 | * | |
951 | * Only compact if a failure would be due to fragmentation. | |
952 | */ | |
953 | fragindex = fragmentation_index(zone, order); | |
954 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
955 | return COMPACT_SKIPPED; | |
956 | ||
a582a738 SL |
957 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
958 | 0, 0)) | |
3e7d3449 MG |
959 | return COMPACT_PARTIAL; |
960 | ||
961 | return COMPACT_CONTINUE; | |
962 | } | |
963 | ||
748446bb MG |
964 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
965 | { | |
966 | int ret; | |
c89511ab | 967 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 968 | unsigned long end_pfn = zone_end_pfn(zone); |
748446bb | 969 | |
3e7d3449 MG |
970 | ret = compaction_suitable(zone, cc->order); |
971 | switch (ret) { | |
972 | case COMPACT_PARTIAL: | |
973 | case COMPACT_SKIPPED: | |
974 | /* Compaction is likely to fail */ | |
975 | return ret; | |
976 | case COMPACT_CONTINUE: | |
977 | /* Fall through to compaction */ | |
978 | ; | |
979 | } | |
980 | ||
d3132e4b VB |
981 | /* |
982 | * Clear pageblock skip if there were failures recently and compaction | |
983 | * is about to be retried after being deferred. kswapd does not do | |
984 | * this reset as it'll reset the cached information when going to sleep. | |
985 | */ | |
986 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
987 | __reset_isolation_suitable(zone); | |
988 | ||
c89511ab MG |
989 | /* |
990 | * Setup to move all movable pages to the end of the zone. Used cached | |
991 | * information on where the scanners should start but check that it | |
992 | * is initialised by ensuring the values are within zone boundaries. | |
993 | */ | |
994 | cc->migrate_pfn = zone->compact_cached_migrate_pfn; | |
995 | cc->free_pfn = zone->compact_cached_free_pfn; | |
996 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
997 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
998 | zone->compact_cached_free_pfn = cc->free_pfn; | |
999 | } | |
1000 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1001 | cc->migrate_pfn = start_pfn; | |
1002 | zone->compact_cached_migrate_pfn = cc->migrate_pfn; | |
1003 | } | |
748446bb | 1004 | |
0eb927c0 MG |
1005 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn); |
1006 | ||
748446bb MG |
1007 | migrate_prep_local(); |
1008 | ||
1009 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
1010 | unsigned long nr_migrate, nr_remaining; | |
9d502c1c | 1011 | int err; |
748446bb | 1012 | |
f9e35b3b MG |
1013 | switch (isolate_migratepages(zone, cc)) { |
1014 | case ISOLATE_ABORT: | |
1015 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1016 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1017 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1018 | goto out; |
1019 | case ISOLATE_NONE: | |
748446bb | 1020 | continue; |
f9e35b3b MG |
1021 | case ISOLATE_SUCCESS: |
1022 | ; | |
1023 | } | |
748446bb MG |
1024 | |
1025 | nr_migrate = cc->nr_migratepages; | |
9d502c1c | 1026 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
9c620e2b | 1027 | (unsigned long)cc, |
7b2a2d4a MG |
1028 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC, |
1029 | MR_COMPACTION); | |
748446bb MG |
1030 | update_nr_listpages(cc); |
1031 | nr_remaining = cc->nr_migratepages; | |
1032 | ||
b7aba698 MG |
1033 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
1034 | nr_remaining); | |
748446bb | 1035 | |
5733c7d1 | 1036 | /* Release isolated pages not migrated */ |
9d502c1c | 1037 | if (err) { |
5733c7d1 | 1038 | putback_movable_pages(&cc->migratepages); |
748446bb | 1039 | cc->nr_migratepages = 0; |
7ed695e0 VB |
1040 | /* |
1041 | * migrate_pages() may return -ENOMEM when scanners meet | |
1042 | * and we want compact_finished() to detect it | |
1043 | */ | |
1044 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1045 | ret = COMPACT_PARTIAL; |
1046 | goto out; | |
1047 | } | |
748446bb | 1048 | } |
748446bb MG |
1049 | } |
1050 | ||
f9e35b3b | 1051 | out: |
748446bb MG |
1052 | /* Release free pages and check accounting */ |
1053 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1054 | VM_BUG_ON(cc->nr_freepages != 0); | |
1055 | ||
0eb927c0 MG |
1056 | trace_mm_compaction_end(ret); |
1057 | ||
748446bb MG |
1058 | return ret; |
1059 | } | |
76ab0f53 | 1060 | |
d43a87e6 | 1061 | static unsigned long compact_zone_order(struct zone *zone, |
5a03b051 | 1062 | int order, gfp_t gfp_mask, |
8fb74b9f | 1063 | bool sync, bool *contended) |
56de7263 | 1064 | { |
e64c5237 | 1065 | unsigned long ret; |
56de7263 MG |
1066 | struct compact_control cc = { |
1067 | .nr_freepages = 0, | |
1068 | .nr_migratepages = 0, | |
1069 | .order = order, | |
1070 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1071 | .zone = zone, | |
68e3e926 | 1072 | .sync = sync, |
56de7263 MG |
1073 | }; |
1074 | INIT_LIST_HEAD(&cc.freepages); | |
1075 | INIT_LIST_HEAD(&cc.migratepages); | |
1076 | ||
e64c5237 SL |
1077 | ret = compact_zone(zone, &cc); |
1078 | ||
1079 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1080 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1081 | ||
1082 | *contended = cc.contended; | |
1083 | return ret; | |
56de7263 MG |
1084 | } |
1085 | ||
5e771905 MG |
1086 | int sysctl_extfrag_threshold = 500; |
1087 | ||
56de7263 MG |
1088 | /** |
1089 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1090 | * @zonelist: The zonelist used for the current allocation | |
1091 | * @order: The order of the current allocation | |
1092 | * @gfp_mask: The GFP mask of the current allocation | |
1093 | * @nodemask: The allowed nodes to allocate from | |
77f1fe6b | 1094 | * @sync: Whether migration is synchronous or not |
661c4cb9 MG |
1095 | * @contended: Return value that is true if compaction was aborted due to lock contention |
1096 | * @page: Optionally capture a free page of the requested order during compaction | |
56de7263 MG |
1097 | * |
1098 | * This is the main entry point for direct page compaction. | |
1099 | */ | |
1100 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1101 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
8fb74b9f | 1102 | bool sync, bool *contended) |
56de7263 MG |
1103 | { |
1104 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1105 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1106 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1107 | struct zoneref *z; |
1108 | struct zone *zone; | |
1109 | int rc = COMPACT_SKIPPED; | |
d95ea5d1 | 1110 | int alloc_flags = 0; |
56de7263 | 1111 | |
4ffb6335 | 1112 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1113 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
1114 | return rc; |
1115 | ||
010fc29a | 1116 | count_compact_event(COMPACTSTALL); |
56de7263 | 1117 | |
d95ea5d1 BZ |
1118 | #ifdef CONFIG_CMA |
1119 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1120 | alloc_flags |= ALLOC_CMA; | |
1121 | #endif | |
56de7263 MG |
1122 | /* Compact each zone in the list */ |
1123 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1124 | nodemask) { | |
56de7263 MG |
1125 | int status; |
1126 | ||
c67fe375 | 1127 | status = compact_zone_order(zone, order, gfp_mask, sync, |
8fb74b9f | 1128 | contended); |
56de7263 MG |
1129 | rc = max(status, rc); |
1130 | ||
3e7d3449 | 1131 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 BZ |
1132 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
1133 | alloc_flags)) | |
56de7263 MG |
1134 | break; |
1135 | } | |
1136 | ||
1137 | return rc; | |
1138 | } | |
1139 | ||
1140 | ||
76ab0f53 | 1141 | /* Compact all zones within a node */ |
7103f16d | 1142 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1143 | { |
1144 | int zoneid; | |
76ab0f53 MG |
1145 | struct zone *zone; |
1146 | ||
76ab0f53 | 1147 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1148 | |
1149 | zone = &pgdat->node_zones[zoneid]; | |
1150 | if (!populated_zone(zone)) | |
1151 | continue; | |
1152 | ||
7be62de9 RR |
1153 | cc->nr_freepages = 0; |
1154 | cc->nr_migratepages = 0; | |
1155 | cc->zone = zone; | |
1156 | INIT_LIST_HEAD(&cc->freepages); | |
1157 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1158 | |
aad6ec37 | 1159 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1160 | compact_zone(zone, cc); |
76ab0f53 | 1161 | |
aff62249 | 1162 | if (cc->order > 0) { |
de6c60a6 VB |
1163 | if (zone_watermark_ok(zone, cc->order, |
1164 | low_wmark_pages(zone), 0, 0)) | |
1165 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 | 1166 | /* Currently async compaction is never deferred. */ |
de6c60a6 | 1167 | else if (cc->sync) |
aff62249 RR |
1168 | defer_compaction(zone, cc->order); |
1169 | } | |
1170 | ||
7be62de9 RR |
1171 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1172 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1173 | } |
76ab0f53 MG |
1174 | } |
1175 | ||
7103f16d | 1176 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1177 | { |
1178 | struct compact_control cc = { | |
1179 | .order = order, | |
68e3e926 | 1180 | .sync = false, |
7be62de9 RR |
1181 | }; |
1182 | ||
3a7200af MG |
1183 | if (!order) |
1184 | return; | |
1185 | ||
7103f16d | 1186 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1187 | } |
1188 | ||
7103f16d | 1189 | static void compact_node(int nid) |
7be62de9 | 1190 | { |
7be62de9 RR |
1191 | struct compact_control cc = { |
1192 | .order = -1, | |
68e3e926 | 1193 | .sync = true, |
91ca9186 | 1194 | .ignore_skip_hint = true, |
7be62de9 RR |
1195 | }; |
1196 | ||
7103f16d | 1197 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1198 | } |
1199 | ||
76ab0f53 | 1200 | /* Compact all nodes in the system */ |
7964c06d | 1201 | static void compact_nodes(void) |
76ab0f53 MG |
1202 | { |
1203 | int nid; | |
1204 | ||
8575ec29 HD |
1205 | /* Flush pending updates to the LRU lists */ |
1206 | lru_add_drain_all(); | |
1207 | ||
76ab0f53 MG |
1208 | for_each_online_node(nid) |
1209 | compact_node(nid); | |
76ab0f53 MG |
1210 | } |
1211 | ||
1212 | /* The written value is actually unused, all memory is compacted */ | |
1213 | int sysctl_compact_memory; | |
1214 | ||
1215 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1216 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1217 | void __user *buffer, size_t *length, loff_t *ppos) | |
1218 | { | |
1219 | if (write) | |
7964c06d | 1220 | compact_nodes(); |
76ab0f53 MG |
1221 | |
1222 | return 0; | |
1223 | } | |
ed4a6d7f | 1224 | |
5e771905 MG |
1225 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1226 | void __user *buffer, size_t *length, loff_t *ppos) | |
1227 | { | |
1228 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1229 | ||
1230 | return 0; | |
1231 | } | |
1232 | ||
ed4a6d7f | 1233 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1234 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1235 | struct device_attribute *attr, |
ed4a6d7f MG |
1236 | const char *buf, size_t count) |
1237 | { | |
8575ec29 HD |
1238 | int nid = dev->id; |
1239 | ||
1240 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1241 | /* Flush pending updates to the LRU lists */ | |
1242 | lru_add_drain_all(); | |
1243 | ||
1244 | compact_node(nid); | |
1245 | } | |
ed4a6d7f MG |
1246 | |
1247 | return count; | |
1248 | } | |
10fbcf4c | 1249 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1250 | |
1251 | int compaction_register_node(struct node *node) | |
1252 | { | |
10fbcf4c | 1253 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1254 | } |
1255 | ||
1256 | void compaction_unregister_node(struct node *node) | |
1257 | { | |
10fbcf4c | 1258 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1259 | } |
1260 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1261 | |
1262 | #endif /* CONFIG_COMPACTION */ |