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; |
bb13ffeb MG |
137 | if (!page) |
138 | return; | |
139 | ||
c89511ab MG |
140 | if (!nr_isolated) { |
141 | unsigned long pfn = page_to_pfn(page); | |
bb13ffeb | 142 | set_pageblock_skip(page); |
c89511ab MG |
143 | |
144 | /* Update where compaction should restart */ | |
145 | if (migrate_scanner) { | |
146 | if (!cc->finished_update_migrate && | |
147 | pfn > zone->compact_cached_migrate_pfn) | |
148 | zone->compact_cached_migrate_pfn = pfn; | |
149 | } else { | |
150 | if (!cc->finished_update_free && | |
151 | pfn < zone->compact_cached_free_pfn) | |
152 | zone->compact_cached_free_pfn = pfn; | |
153 | } | |
154 | } | |
bb13ffeb MG |
155 | } |
156 | #else | |
157 | static inline bool isolation_suitable(struct compact_control *cc, | |
158 | struct page *page) | |
159 | { | |
160 | return true; | |
161 | } | |
162 | ||
c89511ab MG |
163 | static void update_pageblock_skip(struct compact_control *cc, |
164 | struct page *page, unsigned long nr_isolated, | |
165 | bool migrate_scanner) | |
bb13ffeb MG |
166 | { |
167 | } | |
168 | #endif /* CONFIG_COMPACTION */ | |
169 | ||
2a1402aa MG |
170 | static inline bool should_release_lock(spinlock_t *lock) |
171 | { | |
172 | return need_resched() || spin_is_contended(lock); | |
173 | } | |
174 | ||
c67fe375 MG |
175 | /* |
176 | * Compaction requires the taking of some coarse locks that are potentially | |
177 | * very heavily contended. Check if the process needs to be scheduled or | |
178 | * if the lock is contended. For async compaction, back out in the event | |
179 | * if contention is severe. For sync compaction, schedule. | |
180 | * | |
181 | * Returns true if the lock is held. | |
182 | * Returns false if the lock is released and compaction should abort | |
183 | */ | |
184 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, | |
185 | bool locked, struct compact_control *cc) | |
186 | { | |
2a1402aa | 187 | if (should_release_lock(lock)) { |
c67fe375 MG |
188 | if (locked) { |
189 | spin_unlock_irqrestore(lock, *flags); | |
190 | locked = false; | |
191 | } | |
192 | ||
193 | /* async aborts if taking too long or contended */ | |
194 | if (!cc->sync) { | |
e64c5237 | 195 | cc->contended = true; |
c67fe375 MG |
196 | return false; |
197 | } | |
198 | ||
199 | cond_resched(); | |
c67fe375 MG |
200 | } |
201 | ||
202 | if (!locked) | |
203 | spin_lock_irqsave(lock, *flags); | |
204 | return true; | |
205 | } | |
206 | ||
207 | static inline bool compact_trylock_irqsave(spinlock_t *lock, | |
208 | unsigned long *flags, struct compact_control *cc) | |
209 | { | |
210 | return compact_checklock_irqsave(lock, flags, false, cc); | |
211 | } | |
212 | ||
f40d1e42 MG |
213 | /* Returns true if the page is within a block suitable for migration to */ |
214 | static bool suitable_migration_target(struct page *page) | |
215 | { | |
216 | int migratetype = get_pageblock_migratetype(page); | |
217 | ||
218 | /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ | |
194159fb MK |
219 | if (migratetype == MIGRATE_RESERVE) |
220 | return false; | |
221 | ||
222 | if (is_migrate_isolate(migratetype)) | |
f40d1e42 MG |
223 | return false; |
224 | ||
225 | /* If the page is a large free page, then allow migration */ | |
226 | if (PageBuddy(page) && page_order(page) >= pageblock_order) | |
227 | return true; | |
228 | ||
229 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
230 | if (migrate_async_suitable(migratetype)) | |
231 | return true; | |
232 | ||
233 | /* Otherwise skip the block */ | |
234 | return false; | |
235 | } | |
236 | ||
85aa125f MN |
237 | /* |
238 | * Isolate free pages onto a private freelist. Caller must hold zone->lock. | |
239 | * If @strict is true, will abort returning 0 on any invalid PFNs or non-free | |
240 | * pages inside of the pageblock (even though it may still end up isolating | |
241 | * some pages). | |
242 | */ | |
f40d1e42 MG |
243 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
244 | unsigned long blockpfn, | |
85aa125f MN |
245 | unsigned long end_pfn, |
246 | struct list_head *freelist, | |
247 | bool strict) | |
748446bb | 248 | { |
b7aba698 | 249 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 250 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
251 | unsigned long nr_strict_required = end_pfn - blockpfn; |
252 | unsigned long flags; | |
253 | bool locked = false; | |
748446bb | 254 | |
748446bb MG |
255 | cursor = pfn_to_page(blockpfn); |
256 | ||
f40d1e42 | 257 | /* Isolate free pages. */ |
748446bb MG |
258 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
259 | int isolated, i; | |
260 | struct page *page = cursor; | |
261 | ||
b7aba698 | 262 | nr_scanned++; |
f40d1e42 MG |
263 | if (!pfn_valid_within(blockpfn)) |
264 | continue; | |
bb13ffeb MG |
265 | if (!valid_page) |
266 | valid_page = page; | |
f40d1e42 MG |
267 | if (!PageBuddy(page)) |
268 | continue; | |
269 | ||
270 | /* | |
271 | * The zone lock must be held to isolate freepages. | |
272 | * Unfortunately this is a very coarse lock and can be | |
273 | * heavily contended if there are parallel allocations | |
274 | * or parallel compactions. For async compaction do not | |
275 | * spin on the lock and we acquire the lock as late as | |
276 | * possible. | |
277 | */ | |
278 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
279 | locked, cc); | |
280 | if (!locked) | |
281 | break; | |
282 | ||
283 | /* Recheck this is a suitable migration target under lock */ | |
284 | if (!strict && !suitable_migration_target(page)) | |
285 | break; | |
748446bb | 286 | |
f40d1e42 MG |
287 | /* Recheck this is a buddy page under lock */ |
288 | if (!PageBuddy(page)) | |
748446bb MG |
289 | continue; |
290 | ||
291 | /* Found a free page, break it into order-0 pages */ | |
292 | isolated = split_free_page(page); | |
85aa125f | 293 | if (!isolated && strict) |
f40d1e42 | 294 | break; |
748446bb MG |
295 | total_isolated += isolated; |
296 | for (i = 0; i < isolated; i++) { | |
297 | list_add(&page->lru, freelist); | |
298 | page++; | |
299 | } | |
300 | ||
301 | /* If a page was split, advance to the end of it */ | |
302 | if (isolated) { | |
303 | blockpfn += isolated - 1; | |
304 | cursor += isolated - 1; | |
305 | } | |
306 | } | |
307 | ||
b7aba698 | 308 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
309 | |
310 | /* | |
311 | * If strict isolation is requested by CMA then check that all the | |
312 | * pages requested were isolated. If there were any failures, 0 is | |
313 | * returned and CMA will fail. | |
314 | */ | |
0db63d7e | 315 | if (strict && nr_strict_required > total_isolated) |
f40d1e42 MG |
316 | total_isolated = 0; |
317 | ||
318 | if (locked) | |
319 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
320 | ||
bb13ffeb MG |
321 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
322 | if (blockpfn == end_pfn) | |
c89511ab | 323 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 324 | |
010fc29a | 325 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 326 | if (total_isolated) |
010fc29a | 327 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
328 | return total_isolated; |
329 | } | |
330 | ||
85aa125f MN |
331 | /** |
332 | * isolate_freepages_range() - isolate free pages. | |
333 | * @start_pfn: The first PFN to start isolating. | |
334 | * @end_pfn: The one-past-last PFN. | |
335 | * | |
336 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
337 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
338 | * undo its actions and return zero. | |
339 | * | |
340 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
341 | * (which may be greater then end_pfn if end fell in a middle of | |
342 | * a free page). | |
343 | */ | |
ff9543fd | 344 | unsigned long |
bb13ffeb MG |
345 | isolate_freepages_range(struct compact_control *cc, |
346 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 347 | { |
f40d1e42 | 348 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
349 | LIST_HEAD(freelist); |
350 | ||
85aa125f | 351 | for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) { |
bb13ffeb | 352 | if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn))) |
85aa125f MN |
353 | break; |
354 | ||
355 | /* | |
356 | * On subsequent iterations ALIGN() is actually not needed, | |
357 | * but we keep it that we not to complicate the code. | |
358 | */ | |
359 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
360 | block_end_pfn = min(block_end_pfn, end_pfn); | |
361 | ||
bb13ffeb | 362 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 363 | &freelist, true); |
85aa125f MN |
364 | |
365 | /* | |
366 | * In strict mode, isolate_freepages_block() returns 0 if | |
367 | * there are any holes in the block (ie. invalid PFNs or | |
368 | * non-free pages). | |
369 | */ | |
370 | if (!isolated) | |
371 | break; | |
372 | ||
373 | /* | |
374 | * If we managed to isolate pages, it is always (1 << n) * | |
375 | * pageblock_nr_pages for some non-negative n. (Max order | |
376 | * page may span two pageblocks). | |
377 | */ | |
378 | } | |
379 | ||
380 | /* split_free_page does not map the pages */ | |
381 | map_pages(&freelist); | |
382 | ||
383 | if (pfn < end_pfn) { | |
384 | /* Loop terminated early, cleanup. */ | |
385 | release_freepages(&freelist); | |
386 | return 0; | |
387 | } | |
388 | ||
389 | /* We don't use freelists for anything. */ | |
390 | return pfn; | |
391 | } | |
392 | ||
748446bb | 393 | /* Update the number of anon and file isolated pages in the zone */ |
c67fe375 | 394 | static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc) |
748446bb MG |
395 | { |
396 | struct page *page; | |
b9e84ac1 | 397 | unsigned int count[2] = { 0, }; |
748446bb | 398 | |
b9e84ac1 MK |
399 | list_for_each_entry(page, &cc->migratepages, lru) |
400 | count[!!page_is_file_cache(page)]++; | |
748446bb | 401 | |
c67fe375 MG |
402 | /* If locked we can use the interrupt unsafe versions */ |
403 | if (locked) { | |
404 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
405 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
406 | } else { | |
407 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
408 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
409 | } | |
748446bb MG |
410 | } |
411 | ||
412 | /* Similar to reclaim, but different enough that they don't share logic */ | |
413 | static bool too_many_isolated(struct zone *zone) | |
414 | { | |
bc693045 | 415 | unsigned long active, inactive, isolated; |
748446bb MG |
416 | |
417 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
418 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
419 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
420 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
421 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
422 | zone_page_state(zone, NR_ISOLATED_ANON); | |
423 | ||
bc693045 | 424 | return isolated > (inactive + active) / 2; |
748446bb MG |
425 | } |
426 | ||
2fe86e00 MN |
427 | /** |
428 | * isolate_migratepages_range() - isolate all migrate-able pages in range. | |
429 | * @zone: Zone pages are in. | |
430 | * @cc: Compaction control structure. | |
431 | * @low_pfn: The first PFN of the range. | |
432 | * @end_pfn: The one-past-the-last PFN of the range. | |
e46a2879 | 433 | * @unevictable: true if it allows to isolate unevictable pages |
2fe86e00 MN |
434 | * |
435 | * Isolate all pages that can be migrated from the range specified by | |
436 | * [low_pfn, end_pfn). Returns zero if there is a fatal signal | |
437 | * pending), otherwise PFN of the first page that was not scanned | |
438 | * (which may be both less, equal to or more then end_pfn). | |
439 | * | |
440 | * Assumes that cc->migratepages is empty and cc->nr_migratepages is | |
441 | * zero. | |
442 | * | |
443 | * Apart from cc->migratepages and cc->nr_migratetypes this function | |
444 | * does not modify any cc's fields, in particular it does not modify | |
445 | * (or read for that matter) cc->migrate_pfn. | |
748446bb | 446 | */ |
ff9543fd | 447 | unsigned long |
2fe86e00 | 448 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
e46a2879 | 449 | unsigned long low_pfn, unsigned long end_pfn, bool unevictable) |
748446bb | 450 | { |
9927af74 | 451 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
b7aba698 | 452 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 453 | struct list_head *migratelist = &cc->migratepages; |
f3fd4a61 | 454 | isolate_mode_t mode = 0; |
fa9add64 | 455 | struct lruvec *lruvec; |
c67fe375 | 456 | unsigned long flags; |
2a1402aa | 457 | bool locked = false; |
bb13ffeb | 458 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 459 | |
748446bb MG |
460 | /* |
461 | * Ensure that there are not too many pages isolated from the LRU | |
462 | * list by either parallel reclaimers or compaction. If there are, | |
463 | * delay for some time until fewer pages are isolated | |
464 | */ | |
465 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 466 | /* async migration should just abort */ |
68e3e926 | 467 | if (!cc->sync) |
2fe86e00 | 468 | return 0; |
f9e35b3b | 469 | |
748446bb MG |
470 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
471 | ||
472 | if (fatal_signal_pending(current)) | |
2fe86e00 | 473 | return 0; |
748446bb MG |
474 | } |
475 | ||
476 | /* Time to isolate some pages for migration */ | |
b2eef8c0 | 477 | cond_resched(); |
748446bb | 478 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 479 | /* give a chance to irqs before checking need_resched() */ |
2a1402aa MG |
480 | if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) { |
481 | if (should_release_lock(&zone->lru_lock)) { | |
482 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
483 | locked = false; | |
484 | } | |
b2eef8c0 | 485 | } |
c67fe375 | 486 | |
0bf380bc MG |
487 | /* |
488 | * migrate_pfn does not necessarily start aligned to a | |
489 | * pageblock. Ensure that pfn_valid is called when moving | |
490 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
491 | * memory holes within the zone | |
492 | */ | |
493 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
494 | if (!pfn_valid(low_pfn)) { | |
495 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
496 | continue; | |
497 | } | |
498 | } | |
499 | ||
748446bb MG |
500 | if (!pfn_valid_within(low_pfn)) |
501 | continue; | |
b7aba698 | 502 | nr_scanned++; |
748446bb | 503 | |
dc908600 MG |
504 | /* |
505 | * Get the page and ensure the page is within the same zone. | |
506 | * See the comment in isolate_freepages about overlapping | |
507 | * nodes. It is deliberate that the new zone lock is not taken | |
508 | * as memory compaction should not move pages between nodes. | |
509 | */ | |
748446bb | 510 | page = pfn_to_page(low_pfn); |
dc908600 MG |
511 | if (page_zone(page) != zone) |
512 | continue; | |
513 | ||
bb13ffeb MG |
514 | if (!valid_page) |
515 | valid_page = page; | |
516 | ||
517 | /* If isolation recently failed, do not retry */ | |
518 | pageblock_nr = low_pfn >> pageblock_order; | |
519 | if (!isolation_suitable(cc, page)) | |
520 | goto next_pageblock; | |
521 | ||
dc908600 | 522 | /* Skip if free */ |
748446bb MG |
523 | if (PageBuddy(page)) |
524 | continue; | |
525 | ||
9927af74 MG |
526 | /* |
527 | * For async migration, also only scan in MOVABLE blocks. Async | |
528 | * migration is optimistic to see if the minimum amount of work | |
529 | * satisfies the allocation | |
530 | */ | |
68e3e926 | 531 | if (!cc->sync && last_pageblock_nr != pageblock_nr && |
47118af0 | 532 | !migrate_async_suitable(get_pageblock_migratetype(page))) { |
c89511ab | 533 | cc->finished_update_migrate = true; |
2a1402aa | 534 | goto next_pageblock; |
9927af74 MG |
535 | } |
536 | ||
bf6bddf1 RA |
537 | /* |
538 | * Check may be lockless but that's ok as we recheck later. | |
539 | * It's possible to migrate LRU pages and balloon pages | |
540 | * Skip any other type of page | |
541 | */ | |
542 | if (!PageLRU(page)) { | |
543 | if (unlikely(balloon_page_movable(page))) { | |
544 | if (locked && balloon_page_isolate(page)) { | |
545 | /* Successfully isolated */ | |
546 | cc->finished_update_migrate = true; | |
547 | list_add(&page->lru, migratelist); | |
548 | cc->nr_migratepages++; | |
549 | nr_isolated++; | |
550 | goto check_compact_cluster; | |
551 | } | |
552 | } | |
bc835011 | 553 | continue; |
bf6bddf1 | 554 | } |
bc835011 AA |
555 | |
556 | /* | |
2a1402aa MG |
557 | * PageLRU is set. lru_lock normally excludes isolation |
558 | * splitting and collapsing (collapsing has already happened | |
559 | * if PageLRU is set) but the lock is not necessarily taken | |
560 | * here and it is wasteful to take it just to check transhuge. | |
561 | * Check TransHuge without lock and skip the whole pageblock if | |
562 | * it's either a transhuge or hugetlbfs page, as calling | |
563 | * compound_order() without preventing THP from splitting the | |
564 | * page underneath us may return surprising results. | |
bc835011 | 565 | */ |
2a1402aa MG |
566 | if (PageTransHuge(page)) { |
567 | if (!locked) | |
568 | goto next_pageblock; | |
569 | low_pfn += (1 << compound_order(page)) - 1; | |
570 | continue; | |
571 | } | |
572 | ||
573 | /* Check if it is ok to still hold the lock */ | |
574 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
575 | locked, cc); | |
576 | if (!locked || fatal_signal_pending(current)) | |
577 | break; | |
578 | ||
579 | /* Recheck PageLRU and PageTransHuge under lock */ | |
580 | if (!PageLRU(page)) | |
581 | continue; | |
bc835011 AA |
582 | if (PageTransHuge(page)) { |
583 | low_pfn += (1 << compound_order(page)) - 1; | |
584 | continue; | |
585 | } | |
586 | ||
68e3e926 | 587 | if (!cc->sync) |
c8244935 MG |
588 | mode |= ISOLATE_ASYNC_MIGRATE; |
589 | ||
e46a2879 MK |
590 | if (unevictable) |
591 | mode |= ISOLATE_UNEVICTABLE; | |
592 | ||
fa9add64 HD |
593 | lruvec = mem_cgroup_page_lruvec(page, zone); |
594 | ||
748446bb | 595 | /* Try isolate the page */ |
f3fd4a61 | 596 | if (__isolate_lru_page(page, mode) != 0) |
748446bb MG |
597 | continue; |
598 | ||
bc835011 AA |
599 | VM_BUG_ON(PageTransCompound(page)); |
600 | ||
748446bb | 601 | /* Successfully isolated */ |
c89511ab | 602 | cc->finished_update_migrate = true; |
fa9add64 | 603 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
748446bb | 604 | list_add(&page->lru, migratelist); |
748446bb | 605 | cc->nr_migratepages++; |
b7aba698 | 606 | nr_isolated++; |
748446bb | 607 | |
bf6bddf1 | 608 | check_compact_cluster: |
748446bb | 609 | /* Avoid isolating too much */ |
31b8384a HD |
610 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
611 | ++low_pfn; | |
748446bb | 612 | break; |
31b8384a | 613 | } |
2a1402aa MG |
614 | |
615 | continue; | |
616 | ||
617 | next_pageblock: | |
a9aacbcc | 618 | low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1; |
2a1402aa | 619 | last_pageblock_nr = pageblock_nr; |
748446bb MG |
620 | } |
621 | ||
c67fe375 | 622 | acct_isolated(zone, locked, cc); |
748446bb | 623 | |
c67fe375 MG |
624 | if (locked) |
625 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 626 | |
bb13ffeb MG |
627 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
628 | if (low_pfn == end_pfn) | |
c89511ab | 629 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 630 | |
b7aba698 MG |
631 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
632 | ||
010fc29a | 633 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 634 | if (nr_isolated) |
010fc29a | 635 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 636 | |
2fe86e00 MN |
637 | return low_pfn; |
638 | } | |
639 | ||
ff9543fd MN |
640 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
641 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 642 | /* |
ff9543fd MN |
643 | * Based on information in the current compact_control, find blocks |
644 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 645 | */ |
ff9543fd MN |
646 | static void isolate_freepages(struct zone *zone, |
647 | struct compact_control *cc) | |
2fe86e00 | 648 | { |
ff9543fd | 649 | struct page *page; |
108bcc96 | 650 | unsigned long high_pfn, low_pfn, pfn, z_end_pfn, end_pfn; |
ff9543fd MN |
651 | int nr_freepages = cc->nr_freepages; |
652 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 653 | |
ff9543fd MN |
654 | /* |
655 | * Initialise the free scanner. The starting point is where we last | |
656 | * scanned from (or the end of the zone if starting). The low point | |
657 | * is the end of the pageblock the migration scanner is using. | |
658 | */ | |
659 | pfn = cc->free_pfn; | |
660 | low_pfn = cc->migrate_pfn + pageblock_nr_pages; | |
2fe86e00 | 661 | |
ff9543fd MN |
662 | /* |
663 | * Take care that if the migration scanner is at the end of the zone | |
664 | * that the free scanner does not accidentally move to the next zone | |
665 | * in the next isolation cycle. | |
666 | */ | |
667 | high_pfn = min(low_pfn, pfn); | |
2fe86e00 | 668 | |
108bcc96 | 669 | z_end_pfn = zone_end_pfn(zone); |
2fe86e00 | 670 | |
ff9543fd MN |
671 | /* |
672 | * Isolate free pages until enough are available to migrate the | |
673 | * pages on cc->migratepages. We stop searching if the migrate | |
674 | * and free page scanners meet or enough free pages are isolated. | |
675 | */ | |
676 | for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages; | |
677 | pfn -= pageblock_nr_pages) { | |
678 | unsigned long isolated; | |
2fe86e00 | 679 | |
ff9543fd MN |
680 | if (!pfn_valid(pfn)) |
681 | continue; | |
2fe86e00 | 682 | |
ff9543fd MN |
683 | /* |
684 | * Check for overlapping nodes/zones. It's possible on some | |
685 | * configurations to have a setup like | |
686 | * node0 node1 node0 | |
687 | * i.e. it's possible that all pages within a zones range of | |
688 | * pages do not belong to a single zone. | |
689 | */ | |
690 | page = pfn_to_page(pfn); | |
691 | if (page_zone(page) != zone) | |
692 | continue; | |
693 | ||
694 | /* Check the block is suitable for migration */ | |
68e3e926 | 695 | if (!suitable_migration_target(page)) |
ff9543fd | 696 | continue; |
68e3e926 | 697 | |
bb13ffeb MG |
698 | /* If isolation recently failed, do not retry */ |
699 | if (!isolation_suitable(cc, page)) | |
700 | continue; | |
701 | ||
f40d1e42 | 702 | /* Found a block suitable for isolating free pages from */ |
ff9543fd | 703 | isolated = 0; |
60177d31 MG |
704 | |
705 | /* | |
706 | * As pfn may not start aligned, pfn+pageblock_nr_page | |
707 | * may cross a MAX_ORDER_NR_PAGES boundary and miss | |
708 | * a pfn_valid check. Ensure isolate_freepages_block() | |
709 | * only scans within a pageblock | |
710 | */ | |
711 | end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
108bcc96 | 712 | end_pfn = min(end_pfn, z_end_pfn); |
f40d1e42 MG |
713 | isolated = isolate_freepages_block(cc, pfn, end_pfn, |
714 | freelist, false); | |
715 | nr_freepages += isolated; | |
ff9543fd MN |
716 | |
717 | /* | |
718 | * Record the highest PFN we isolated pages from. When next | |
719 | * looking for free pages, the search will restart here as | |
720 | * page migration may have returned some pages to the allocator | |
721 | */ | |
c89511ab MG |
722 | if (isolated) { |
723 | cc->finished_update_free = true; | |
ff9543fd | 724 | high_pfn = max(high_pfn, pfn); |
c89511ab | 725 | } |
ff9543fd MN |
726 | } |
727 | ||
728 | /* split_free_page does not map the pages */ | |
729 | map_pages(freelist); | |
730 | ||
731 | cc->free_pfn = high_pfn; | |
732 | cc->nr_freepages = nr_freepages; | |
748446bb MG |
733 | } |
734 | ||
735 | /* | |
736 | * This is a migrate-callback that "allocates" freepages by taking pages | |
737 | * from the isolated freelists in the block we are migrating to. | |
738 | */ | |
739 | static struct page *compaction_alloc(struct page *migratepage, | |
740 | unsigned long data, | |
741 | int **result) | |
742 | { | |
743 | struct compact_control *cc = (struct compact_control *)data; | |
744 | struct page *freepage; | |
745 | ||
746 | /* Isolate free pages if necessary */ | |
747 | if (list_empty(&cc->freepages)) { | |
748 | isolate_freepages(cc->zone, cc); | |
749 | ||
750 | if (list_empty(&cc->freepages)) | |
751 | return NULL; | |
752 | } | |
753 | ||
754 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
755 | list_del(&freepage->lru); | |
756 | cc->nr_freepages--; | |
757 | ||
758 | return freepage; | |
759 | } | |
760 | ||
761 | /* | |
762 | * We cannot control nr_migratepages and nr_freepages fully when migration is | |
763 | * running as migrate_pages() has no knowledge of compact_control. When | |
764 | * migration is complete, we count the number of pages on the lists by hand. | |
765 | */ | |
766 | static void update_nr_listpages(struct compact_control *cc) | |
767 | { | |
768 | int nr_migratepages = 0; | |
769 | int nr_freepages = 0; | |
770 | struct page *page; | |
771 | ||
772 | list_for_each_entry(page, &cc->migratepages, lru) | |
773 | nr_migratepages++; | |
774 | list_for_each_entry(page, &cc->freepages, lru) | |
775 | nr_freepages++; | |
776 | ||
777 | cc->nr_migratepages = nr_migratepages; | |
778 | cc->nr_freepages = nr_freepages; | |
779 | } | |
780 | ||
ff9543fd MN |
781 | /* possible outcome of isolate_migratepages */ |
782 | typedef enum { | |
783 | ISOLATE_ABORT, /* Abort compaction now */ | |
784 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
785 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
786 | } isolate_migrate_t; | |
787 | ||
788 | /* | |
789 | * Isolate all pages that can be migrated from the block pointed to by | |
790 | * the migrate scanner within compact_control. | |
791 | */ | |
792 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
793 | struct compact_control *cc) | |
794 | { | |
795 | unsigned long low_pfn, end_pfn; | |
796 | ||
797 | /* Do not scan outside zone boundaries */ | |
798 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
799 | ||
800 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 801 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd MN |
802 | |
803 | /* Do not cross the free scanner or scan within a memory hole */ | |
804 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
805 | cc->migrate_pfn = end_pfn; | |
806 | return ISOLATE_NONE; | |
807 | } | |
808 | ||
809 | /* Perform the isolation */ | |
e46a2879 | 810 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
e64c5237 | 811 | if (!low_pfn || cc->contended) |
ff9543fd MN |
812 | return ISOLATE_ABORT; |
813 | ||
814 | cc->migrate_pfn = low_pfn; | |
815 | ||
816 | return ISOLATE_SUCCESS; | |
817 | } | |
818 | ||
748446bb | 819 | static int compact_finished(struct zone *zone, |
5a03b051 | 820 | struct compact_control *cc) |
748446bb | 821 | { |
8fb74b9f | 822 | unsigned int order; |
5a03b051 | 823 | unsigned long watermark; |
56de7263 | 824 | |
748446bb MG |
825 | if (fatal_signal_pending(current)) |
826 | return COMPACT_PARTIAL; | |
827 | ||
753341a4 | 828 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 829 | if (cc->free_pfn <= cc->migrate_pfn) { |
62997027 MG |
830 | /* |
831 | * Mark that the PG_migrate_skip information should be cleared | |
832 | * by kswapd when it goes to sleep. kswapd does not set the | |
833 | * flag itself as the decision to be clear should be directly | |
834 | * based on an allocation request. | |
835 | */ | |
836 | if (!current_is_kswapd()) | |
837 | zone->compact_blockskip_flush = true; | |
838 | ||
748446bb | 839 | return COMPACT_COMPLETE; |
bb13ffeb | 840 | } |
748446bb | 841 | |
82478fb7 JW |
842 | /* |
843 | * order == -1 is expected when compacting via | |
844 | * /proc/sys/vm/compact_memory | |
845 | */ | |
56de7263 MG |
846 | if (cc->order == -1) |
847 | return COMPACT_CONTINUE; | |
848 | ||
3957c776 MH |
849 | /* Compaction run is not finished if the watermark is not met */ |
850 | watermark = low_wmark_pages(zone); | |
851 | watermark += (1 << cc->order); | |
852 | ||
853 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
854 | return COMPACT_CONTINUE; | |
855 | ||
56de7263 | 856 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
857 | for (order = cc->order; order < MAX_ORDER; order++) { |
858 | struct free_area *area = &zone->free_area[order]; | |
859 | ||
860 | /* Job done if page is free of the right migratetype */ | |
861 | if (!list_empty(&area->free_list[cc->migratetype])) | |
862 | return COMPACT_PARTIAL; | |
863 | ||
864 | /* Job done if allocation would set block type */ | |
865 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
866 | return COMPACT_PARTIAL; |
867 | } | |
868 | ||
748446bb MG |
869 | return COMPACT_CONTINUE; |
870 | } | |
871 | ||
3e7d3449 MG |
872 | /* |
873 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
874 | * Returns | |
875 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
876 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
877 | * COMPACT_CONTINUE - If compaction should run now | |
878 | */ | |
879 | unsigned long compaction_suitable(struct zone *zone, int order) | |
880 | { | |
881 | int fragindex; | |
882 | unsigned long watermark; | |
883 | ||
3957c776 MH |
884 | /* |
885 | * order == -1 is expected when compacting via | |
886 | * /proc/sys/vm/compact_memory | |
887 | */ | |
888 | if (order == -1) | |
889 | return COMPACT_CONTINUE; | |
890 | ||
3e7d3449 MG |
891 | /* |
892 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
893 | * This is because during migration, copies of pages need to be | |
894 | * allocated and for a short time, the footprint is higher | |
895 | */ | |
896 | watermark = low_wmark_pages(zone) + (2UL << order); | |
897 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
898 | return COMPACT_SKIPPED; | |
899 | ||
900 | /* | |
901 | * fragmentation index determines if allocation failures are due to | |
902 | * low memory or external fragmentation | |
903 | * | |
a582a738 SL |
904 | * index of -1000 implies allocations might succeed depending on |
905 | * watermarks | |
3e7d3449 MG |
906 | * index towards 0 implies failure is due to lack of memory |
907 | * index towards 1000 implies failure is due to fragmentation | |
908 | * | |
909 | * Only compact if a failure would be due to fragmentation. | |
910 | */ | |
911 | fragindex = fragmentation_index(zone, order); | |
912 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
913 | return COMPACT_SKIPPED; | |
914 | ||
a582a738 SL |
915 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
916 | 0, 0)) | |
3e7d3449 MG |
917 | return COMPACT_PARTIAL; |
918 | ||
919 | return COMPACT_CONTINUE; | |
920 | } | |
921 | ||
748446bb MG |
922 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
923 | { | |
924 | int ret; | |
c89511ab | 925 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 926 | unsigned long end_pfn = zone_end_pfn(zone); |
748446bb | 927 | |
3e7d3449 MG |
928 | ret = compaction_suitable(zone, cc->order); |
929 | switch (ret) { | |
930 | case COMPACT_PARTIAL: | |
931 | case COMPACT_SKIPPED: | |
932 | /* Compaction is likely to fail */ | |
933 | return ret; | |
934 | case COMPACT_CONTINUE: | |
935 | /* Fall through to compaction */ | |
936 | ; | |
937 | } | |
938 | ||
c89511ab MG |
939 | /* |
940 | * Setup to move all movable pages to the end of the zone. Used cached | |
941 | * information on where the scanners should start but check that it | |
942 | * is initialised by ensuring the values are within zone boundaries. | |
943 | */ | |
944 | cc->migrate_pfn = zone->compact_cached_migrate_pfn; | |
945 | cc->free_pfn = zone->compact_cached_free_pfn; | |
946 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
947 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
948 | zone->compact_cached_free_pfn = cc->free_pfn; | |
949 | } | |
950 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
951 | cc->migrate_pfn = start_pfn; | |
952 | zone->compact_cached_migrate_pfn = cc->migrate_pfn; | |
953 | } | |
748446bb | 954 | |
62997027 MG |
955 | /* |
956 | * Clear pageblock skip if there were failures recently and compaction | |
957 | * is about to be retried after being deferred. kswapd does not do | |
958 | * this reset as it'll reset the cached information when going to sleep. | |
959 | */ | |
960 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
961 | __reset_isolation_suitable(zone); | |
bb13ffeb | 962 | |
748446bb MG |
963 | migrate_prep_local(); |
964 | ||
965 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
966 | unsigned long nr_migrate, nr_remaining; | |
9d502c1c | 967 | int err; |
748446bb | 968 | |
f9e35b3b MG |
969 | switch (isolate_migratepages(zone, cc)) { |
970 | case ISOLATE_ABORT: | |
971 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 972 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 973 | cc->nr_migratepages = 0; |
f9e35b3b MG |
974 | goto out; |
975 | case ISOLATE_NONE: | |
748446bb | 976 | continue; |
f9e35b3b MG |
977 | case ISOLATE_SUCCESS: |
978 | ; | |
979 | } | |
748446bb MG |
980 | |
981 | nr_migrate = cc->nr_migratepages; | |
9d502c1c | 982 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
9c620e2b | 983 | (unsigned long)cc, |
7b2a2d4a MG |
984 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC, |
985 | MR_COMPACTION); | |
748446bb MG |
986 | update_nr_listpages(cc); |
987 | nr_remaining = cc->nr_migratepages; | |
988 | ||
b7aba698 MG |
989 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
990 | nr_remaining); | |
748446bb | 991 | |
5733c7d1 | 992 | /* Release isolated pages not migrated */ |
9d502c1c | 993 | if (err) { |
5733c7d1 | 994 | putback_movable_pages(&cc->migratepages); |
748446bb | 995 | cc->nr_migratepages = 0; |
4bf2bba3 DR |
996 | if (err == -ENOMEM) { |
997 | ret = COMPACT_PARTIAL; | |
998 | goto out; | |
999 | } | |
748446bb | 1000 | } |
748446bb MG |
1001 | } |
1002 | ||
f9e35b3b | 1003 | out: |
748446bb MG |
1004 | /* Release free pages and check accounting */ |
1005 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1006 | VM_BUG_ON(cc->nr_freepages != 0); | |
1007 | ||
1008 | return ret; | |
1009 | } | |
76ab0f53 | 1010 | |
d43a87e6 | 1011 | static unsigned long compact_zone_order(struct zone *zone, |
5a03b051 | 1012 | int order, gfp_t gfp_mask, |
8fb74b9f | 1013 | bool sync, bool *contended) |
56de7263 | 1014 | { |
e64c5237 | 1015 | unsigned long ret; |
56de7263 MG |
1016 | struct compact_control cc = { |
1017 | .nr_freepages = 0, | |
1018 | .nr_migratepages = 0, | |
1019 | .order = order, | |
1020 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1021 | .zone = zone, | |
68e3e926 | 1022 | .sync = sync, |
56de7263 MG |
1023 | }; |
1024 | INIT_LIST_HEAD(&cc.freepages); | |
1025 | INIT_LIST_HEAD(&cc.migratepages); | |
1026 | ||
e64c5237 SL |
1027 | ret = compact_zone(zone, &cc); |
1028 | ||
1029 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1030 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1031 | ||
1032 | *contended = cc.contended; | |
1033 | return ret; | |
56de7263 MG |
1034 | } |
1035 | ||
5e771905 MG |
1036 | int sysctl_extfrag_threshold = 500; |
1037 | ||
56de7263 MG |
1038 | /** |
1039 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1040 | * @zonelist: The zonelist used for the current allocation | |
1041 | * @order: The order of the current allocation | |
1042 | * @gfp_mask: The GFP mask of the current allocation | |
1043 | * @nodemask: The allowed nodes to allocate from | |
77f1fe6b | 1044 | * @sync: Whether migration is synchronous or not |
661c4cb9 MG |
1045 | * @contended: Return value that is true if compaction was aborted due to lock contention |
1046 | * @page: Optionally capture a free page of the requested order during compaction | |
56de7263 MG |
1047 | * |
1048 | * This is the main entry point for direct page compaction. | |
1049 | */ | |
1050 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1051 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
8fb74b9f | 1052 | bool sync, bool *contended) |
56de7263 MG |
1053 | { |
1054 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1055 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1056 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1057 | struct zoneref *z; |
1058 | struct zone *zone; | |
1059 | int rc = COMPACT_SKIPPED; | |
d95ea5d1 | 1060 | int alloc_flags = 0; |
56de7263 | 1061 | |
4ffb6335 | 1062 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1063 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
1064 | return rc; |
1065 | ||
010fc29a | 1066 | count_compact_event(COMPACTSTALL); |
56de7263 | 1067 | |
d95ea5d1 BZ |
1068 | #ifdef CONFIG_CMA |
1069 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1070 | alloc_flags |= ALLOC_CMA; | |
1071 | #endif | |
56de7263 MG |
1072 | /* Compact each zone in the list */ |
1073 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1074 | nodemask) { | |
56de7263 MG |
1075 | int status; |
1076 | ||
c67fe375 | 1077 | status = compact_zone_order(zone, order, gfp_mask, sync, |
8fb74b9f | 1078 | contended); |
56de7263 MG |
1079 | rc = max(status, rc); |
1080 | ||
3e7d3449 | 1081 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 BZ |
1082 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
1083 | alloc_flags)) | |
56de7263 MG |
1084 | break; |
1085 | } | |
1086 | ||
1087 | return rc; | |
1088 | } | |
1089 | ||
1090 | ||
76ab0f53 | 1091 | /* Compact all zones within a node */ |
7103f16d | 1092 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1093 | { |
1094 | int zoneid; | |
76ab0f53 MG |
1095 | struct zone *zone; |
1096 | ||
76ab0f53 | 1097 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1098 | |
1099 | zone = &pgdat->node_zones[zoneid]; | |
1100 | if (!populated_zone(zone)) | |
1101 | continue; | |
1102 | ||
7be62de9 RR |
1103 | cc->nr_freepages = 0; |
1104 | cc->nr_migratepages = 0; | |
1105 | cc->zone = zone; | |
1106 | INIT_LIST_HEAD(&cc->freepages); | |
1107 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1108 | |
aad6ec37 | 1109 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1110 | compact_zone(zone, cc); |
76ab0f53 | 1111 | |
aff62249 RR |
1112 | if (cc->order > 0) { |
1113 | int ok = zone_watermark_ok(zone, cc->order, | |
1114 | low_wmark_pages(zone), 0, 0); | |
c81758fb | 1115 | if (ok && cc->order >= zone->compact_order_failed) |
aff62249 RR |
1116 | zone->compact_order_failed = cc->order + 1; |
1117 | /* Currently async compaction is never deferred. */ | |
68e3e926 | 1118 | else if (!ok && cc->sync) |
aff62249 RR |
1119 | defer_compaction(zone, cc->order); |
1120 | } | |
1121 | ||
7be62de9 RR |
1122 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1123 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1124 | } |
76ab0f53 MG |
1125 | } |
1126 | ||
7103f16d | 1127 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1128 | { |
1129 | struct compact_control cc = { | |
1130 | .order = order, | |
68e3e926 | 1131 | .sync = false, |
7be62de9 RR |
1132 | }; |
1133 | ||
7103f16d | 1134 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1135 | } |
1136 | ||
7103f16d | 1137 | static void compact_node(int nid) |
7be62de9 | 1138 | { |
7be62de9 RR |
1139 | struct compact_control cc = { |
1140 | .order = -1, | |
68e3e926 | 1141 | .sync = true, |
7be62de9 RR |
1142 | }; |
1143 | ||
7103f16d | 1144 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1145 | } |
1146 | ||
76ab0f53 | 1147 | /* Compact all nodes in the system */ |
7964c06d | 1148 | static void compact_nodes(void) |
76ab0f53 MG |
1149 | { |
1150 | int nid; | |
1151 | ||
8575ec29 HD |
1152 | /* Flush pending updates to the LRU lists */ |
1153 | lru_add_drain_all(); | |
1154 | ||
76ab0f53 MG |
1155 | for_each_online_node(nid) |
1156 | compact_node(nid); | |
76ab0f53 MG |
1157 | } |
1158 | ||
1159 | /* The written value is actually unused, all memory is compacted */ | |
1160 | int sysctl_compact_memory; | |
1161 | ||
1162 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1163 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1164 | void __user *buffer, size_t *length, loff_t *ppos) | |
1165 | { | |
1166 | if (write) | |
7964c06d | 1167 | compact_nodes(); |
76ab0f53 MG |
1168 | |
1169 | return 0; | |
1170 | } | |
ed4a6d7f | 1171 | |
5e771905 MG |
1172 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1173 | void __user *buffer, size_t *length, loff_t *ppos) | |
1174 | { | |
1175 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1176 | ||
1177 | return 0; | |
1178 | } | |
1179 | ||
ed4a6d7f | 1180 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
10fbcf4c KS |
1181 | ssize_t sysfs_compact_node(struct device *dev, |
1182 | struct device_attribute *attr, | |
ed4a6d7f MG |
1183 | const char *buf, size_t count) |
1184 | { | |
8575ec29 HD |
1185 | int nid = dev->id; |
1186 | ||
1187 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1188 | /* Flush pending updates to the LRU lists */ | |
1189 | lru_add_drain_all(); | |
1190 | ||
1191 | compact_node(nid); | |
1192 | } | |
ed4a6d7f MG |
1193 | |
1194 | return count; | |
1195 | } | |
10fbcf4c | 1196 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1197 | |
1198 | int compaction_register_node(struct node *node) | |
1199 | { | |
10fbcf4c | 1200 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1201 | } |
1202 | ||
1203 | void compaction_unregister_node(struct node *node) | |
1204 | { | |
10fbcf4c | 1205 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1206 | } |
1207 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1208 | |
1209 | #endif /* CONFIG_COMPACTION */ |