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 | 237 | /* |
9e4be470 JM |
238 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
239 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
240 | * (even though it may still end up isolating some pages). | |
85aa125f | 241 | */ |
f40d1e42 MG |
242 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
243 | unsigned long blockpfn, | |
85aa125f MN |
244 | unsigned long end_pfn, |
245 | struct list_head *freelist, | |
246 | bool strict) | |
748446bb | 247 | { |
b7aba698 | 248 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 249 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
250 | unsigned long nr_strict_required = end_pfn - blockpfn; |
251 | unsigned long flags; | |
252 | bool locked = false; | |
748446bb | 253 | |
748446bb MG |
254 | cursor = pfn_to_page(blockpfn); |
255 | ||
f40d1e42 | 256 | /* Isolate free pages. */ |
748446bb MG |
257 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
258 | int isolated, i; | |
259 | struct page *page = cursor; | |
260 | ||
b7aba698 | 261 | nr_scanned++; |
f40d1e42 MG |
262 | if (!pfn_valid_within(blockpfn)) |
263 | continue; | |
bb13ffeb MG |
264 | if (!valid_page) |
265 | valid_page = page; | |
f40d1e42 MG |
266 | if (!PageBuddy(page)) |
267 | continue; | |
268 | ||
269 | /* | |
270 | * The zone lock must be held to isolate freepages. | |
271 | * Unfortunately this is a very coarse lock and can be | |
272 | * heavily contended if there are parallel allocations | |
273 | * or parallel compactions. For async compaction do not | |
274 | * spin on the lock and we acquire the lock as late as | |
275 | * possible. | |
276 | */ | |
277 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
278 | locked, cc); | |
279 | if (!locked) | |
280 | break; | |
281 | ||
282 | /* Recheck this is a suitable migration target under lock */ | |
283 | if (!strict && !suitable_migration_target(page)) | |
284 | break; | |
748446bb | 285 | |
f40d1e42 MG |
286 | /* Recheck this is a buddy page under lock */ |
287 | if (!PageBuddy(page)) | |
748446bb MG |
288 | continue; |
289 | ||
290 | /* Found a free page, break it into order-0 pages */ | |
291 | isolated = split_free_page(page); | |
85aa125f | 292 | if (!isolated && strict) |
f40d1e42 | 293 | break; |
748446bb MG |
294 | total_isolated += isolated; |
295 | for (i = 0; i < isolated; i++) { | |
296 | list_add(&page->lru, freelist); | |
297 | page++; | |
298 | } | |
299 | ||
300 | /* If a page was split, advance to the end of it */ | |
301 | if (isolated) { | |
302 | blockpfn += isolated - 1; | |
303 | cursor += isolated - 1; | |
304 | } | |
305 | } | |
306 | ||
b7aba698 | 307 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
308 | |
309 | /* | |
310 | * If strict isolation is requested by CMA then check that all the | |
311 | * pages requested were isolated. If there were any failures, 0 is | |
312 | * returned and CMA will fail. | |
313 | */ | |
0db63d7e | 314 | if (strict && nr_strict_required > total_isolated) |
f40d1e42 MG |
315 | total_isolated = 0; |
316 | ||
317 | if (locked) | |
318 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
319 | ||
bb13ffeb MG |
320 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
321 | if (blockpfn == end_pfn) | |
c89511ab | 322 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 323 | |
010fc29a | 324 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 325 | if (total_isolated) |
010fc29a | 326 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
327 | return total_isolated; |
328 | } | |
329 | ||
85aa125f MN |
330 | /** |
331 | * isolate_freepages_range() - isolate free pages. | |
332 | * @start_pfn: The first PFN to start isolating. | |
333 | * @end_pfn: The one-past-last PFN. | |
334 | * | |
335 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
336 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
337 | * undo its actions and return zero. | |
338 | * | |
339 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
340 | * (which may be greater then end_pfn if end fell in a middle of | |
341 | * a free page). | |
342 | */ | |
ff9543fd | 343 | unsigned long |
bb13ffeb MG |
344 | isolate_freepages_range(struct compact_control *cc, |
345 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 346 | { |
f40d1e42 | 347 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
348 | LIST_HEAD(freelist); |
349 | ||
85aa125f | 350 | for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) { |
bb13ffeb | 351 | if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn))) |
85aa125f MN |
352 | break; |
353 | ||
354 | /* | |
355 | * On subsequent iterations ALIGN() is actually not needed, | |
356 | * but we keep it that we not to complicate the code. | |
357 | */ | |
358 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
359 | block_end_pfn = min(block_end_pfn, end_pfn); | |
360 | ||
bb13ffeb | 361 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 362 | &freelist, true); |
85aa125f MN |
363 | |
364 | /* | |
365 | * In strict mode, isolate_freepages_block() returns 0 if | |
366 | * there are any holes in the block (ie. invalid PFNs or | |
367 | * non-free pages). | |
368 | */ | |
369 | if (!isolated) | |
370 | break; | |
371 | ||
372 | /* | |
373 | * If we managed to isolate pages, it is always (1 << n) * | |
374 | * pageblock_nr_pages for some non-negative n. (Max order | |
375 | * page may span two pageblocks). | |
376 | */ | |
377 | } | |
378 | ||
379 | /* split_free_page does not map the pages */ | |
380 | map_pages(&freelist); | |
381 | ||
382 | if (pfn < end_pfn) { | |
383 | /* Loop terminated early, cleanup. */ | |
384 | release_freepages(&freelist); | |
385 | return 0; | |
386 | } | |
387 | ||
388 | /* We don't use freelists for anything. */ | |
389 | return pfn; | |
390 | } | |
391 | ||
748446bb | 392 | /* Update the number of anon and file isolated pages in the zone */ |
c67fe375 | 393 | static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc) |
748446bb MG |
394 | { |
395 | struct page *page; | |
b9e84ac1 | 396 | unsigned int count[2] = { 0, }; |
748446bb | 397 | |
b9e84ac1 MK |
398 | list_for_each_entry(page, &cc->migratepages, lru) |
399 | count[!!page_is_file_cache(page)]++; | |
748446bb | 400 | |
c67fe375 MG |
401 | /* If locked we can use the interrupt unsafe versions */ |
402 | if (locked) { | |
403 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
404 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
405 | } else { | |
406 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
407 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
408 | } | |
748446bb MG |
409 | } |
410 | ||
411 | /* Similar to reclaim, but different enough that they don't share logic */ | |
412 | static bool too_many_isolated(struct zone *zone) | |
413 | { | |
bc693045 | 414 | unsigned long active, inactive, isolated; |
748446bb MG |
415 | |
416 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
417 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
418 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
419 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
420 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
421 | zone_page_state(zone, NR_ISOLATED_ANON); | |
422 | ||
bc693045 | 423 | return isolated > (inactive + active) / 2; |
748446bb MG |
424 | } |
425 | ||
2fe86e00 MN |
426 | /** |
427 | * isolate_migratepages_range() - isolate all migrate-able pages in range. | |
428 | * @zone: Zone pages are in. | |
429 | * @cc: Compaction control structure. | |
430 | * @low_pfn: The first PFN of the range. | |
431 | * @end_pfn: The one-past-the-last PFN of the range. | |
e46a2879 | 432 | * @unevictable: true if it allows to isolate unevictable pages |
2fe86e00 MN |
433 | * |
434 | * Isolate all pages that can be migrated from the range specified by | |
435 | * [low_pfn, end_pfn). Returns zero if there is a fatal signal | |
436 | * pending), otherwise PFN of the first page that was not scanned | |
437 | * (which may be both less, equal to or more then end_pfn). | |
438 | * | |
439 | * Assumes that cc->migratepages is empty and cc->nr_migratepages is | |
440 | * zero. | |
441 | * | |
442 | * Apart from cc->migratepages and cc->nr_migratetypes this function | |
443 | * does not modify any cc's fields, in particular it does not modify | |
444 | * (or read for that matter) cc->migrate_pfn. | |
748446bb | 445 | */ |
ff9543fd | 446 | unsigned long |
2fe86e00 | 447 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
e46a2879 | 448 | unsigned long low_pfn, unsigned long end_pfn, bool unevictable) |
748446bb | 449 | { |
9927af74 | 450 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
b7aba698 | 451 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 452 | struct list_head *migratelist = &cc->migratepages; |
f3fd4a61 | 453 | isolate_mode_t mode = 0; |
fa9add64 | 454 | struct lruvec *lruvec; |
c67fe375 | 455 | unsigned long flags; |
2a1402aa | 456 | bool locked = false; |
bb13ffeb | 457 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 458 | |
748446bb MG |
459 | /* |
460 | * Ensure that there are not too many pages isolated from the LRU | |
461 | * list by either parallel reclaimers or compaction. If there are, | |
462 | * delay for some time until fewer pages are isolated | |
463 | */ | |
464 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 465 | /* async migration should just abort */ |
68e3e926 | 466 | if (!cc->sync) |
2fe86e00 | 467 | return 0; |
f9e35b3b | 468 | |
748446bb MG |
469 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
470 | ||
471 | if (fatal_signal_pending(current)) | |
2fe86e00 | 472 | return 0; |
748446bb MG |
473 | } |
474 | ||
475 | /* Time to isolate some pages for migration */ | |
b2eef8c0 | 476 | cond_resched(); |
748446bb | 477 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 478 | /* give a chance to irqs before checking need_resched() */ |
2a1402aa MG |
479 | if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) { |
480 | if (should_release_lock(&zone->lru_lock)) { | |
481 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
482 | locked = false; | |
483 | } | |
b2eef8c0 | 484 | } |
c67fe375 | 485 | |
0bf380bc MG |
486 | /* |
487 | * migrate_pfn does not necessarily start aligned to a | |
488 | * pageblock. Ensure that pfn_valid is called when moving | |
489 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
490 | * memory holes within the zone | |
491 | */ | |
492 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
493 | if (!pfn_valid(low_pfn)) { | |
494 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
495 | continue; | |
496 | } | |
497 | } | |
498 | ||
748446bb MG |
499 | if (!pfn_valid_within(low_pfn)) |
500 | continue; | |
b7aba698 | 501 | nr_scanned++; |
748446bb | 502 | |
dc908600 MG |
503 | /* |
504 | * Get the page and ensure the page is within the same zone. | |
505 | * See the comment in isolate_freepages about overlapping | |
506 | * nodes. It is deliberate that the new zone lock is not taken | |
507 | * as memory compaction should not move pages between nodes. | |
508 | */ | |
748446bb | 509 | page = pfn_to_page(low_pfn); |
dc908600 MG |
510 | if (page_zone(page) != zone) |
511 | continue; | |
512 | ||
bb13ffeb MG |
513 | if (!valid_page) |
514 | valid_page = page; | |
515 | ||
516 | /* If isolation recently failed, do not retry */ | |
517 | pageblock_nr = low_pfn >> pageblock_order; | |
518 | if (!isolation_suitable(cc, page)) | |
519 | goto next_pageblock; | |
520 | ||
dc908600 | 521 | /* Skip if free */ |
748446bb MG |
522 | if (PageBuddy(page)) |
523 | continue; | |
524 | ||
9927af74 MG |
525 | /* |
526 | * For async migration, also only scan in MOVABLE blocks. Async | |
527 | * migration is optimistic to see if the minimum amount of work | |
528 | * satisfies the allocation | |
529 | */ | |
68e3e926 | 530 | if (!cc->sync && last_pageblock_nr != pageblock_nr && |
47118af0 | 531 | !migrate_async_suitable(get_pageblock_migratetype(page))) { |
c89511ab | 532 | cc->finished_update_migrate = true; |
2a1402aa | 533 | goto next_pageblock; |
9927af74 MG |
534 | } |
535 | ||
bf6bddf1 RA |
536 | /* |
537 | * Check may be lockless but that's ok as we recheck later. | |
538 | * It's possible to migrate LRU pages and balloon pages | |
539 | * Skip any other type of page | |
540 | */ | |
541 | if (!PageLRU(page)) { | |
542 | if (unlikely(balloon_page_movable(page))) { | |
543 | if (locked && balloon_page_isolate(page)) { | |
544 | /* Successfully isolated */ | |
545 | cc->finished_update_migrate = true; | |
546 | list_add(&page->lru, migratelist); | |
547 | cc->nr_migratepages++; | |
548 | nr_isolated++; | |
549 | goto check_compact_cluster; | |
550 | } | |
551 | } | |
bc835011 | 552 | continue; |
bf6bddf1 | 553 | } |
bc835011 AA |
554 | |
555 | /* | |
2a1402aa MG |
556 | * PageLRU is set. lru_lock normally excludes isolation |
557 | * splitting and collapsing (collapsing has already happened | |
558 | * if PageLRU is set) but the lock is not necessarily taken | |
559 | * here and it is wasteful to take it just to check transhuge. | |
560 | * Check TransHuge without lock and skip the whole pageblock if | |
561 | * it's either a transhuge or hugetlbfs page, as calling | |
562 | * compound_order() without preventing THP from splitting the | |
563 | * page underneath us may return surprising results. | |
bc835011 | 564 | */ |
2a1402aa MG |
565 | if (PageTransHuge(page)) { |
566 | if (!locked) | |
567 | goto next_pageblock; | |
568 | low_pfn += (1 << compound_order(page)) - 1; | |
569 | continue; | |
570 | } | |
571 | ||
572 | /* Check if it is ok to still hold the lock */ | |
573 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
574 | locked, cc); | |
575 | if (!locked || fatal_signal_pending(current)) | |
576 | break; | |
577 | ||
578 | /* Recheck PageLRU and PageTransHuge under lock */ | |
579 | if (!PageLRU(page)) | |
580 | continue; | |
bc835011 AA |
581 | if (PageTransHuge(page)) { |
582 | low_pfn += (1 << compound_order(page)) - 1; | |
583 | continue; | |
584 | } | |
585 | ||
68e3e926 | 586 | if (!cc->sync) |
c8244935 MG |
587 | mode |= ISOLATE_ASYNC_MIGRATE; |
588 | ||
e46a2879 MK |
589 | if (unevictable) |
590 | mode |= ISOLATE_UNEVICTABLE; | |
591 | ||
fa9add64 HD |
592 | lruvec = mem_cgroup_page_lruvec(page, zone); |
593 | ||
748446bb | 594 | /* Try isolate the page */ |
f3fd4a61 | 595 | if (__isolate_lru_page(page, mode) != 0) |
748446bb MG |
596 | continue; |
597 | ||
bc835011 AA |
598 | VM_BUG_ON(PageTransCompound(page)); |
599 | ||
748446bb | 600 | /* Successfully isolated */ |
c89511ab | 601 | cc->finished_update_migrate = true; |
fa9add64 | 602 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
748446bb | 603 | list_add(&page->lru, migratelist); |
748446bb | 604 | cc->nr_migratepages++; |
b7aba698 | 605 | nr_isolated++; |
748446bb | 606 | |
bf6bddf1 | 607 | check_compact_cluster: |
748446bb | 608 | /* Avoid isolating too much */ |
31b8384a HD |
609 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
610 | ++low_pfn; | |
748446bb | 611 | break; |
31b8384a | 612 | } |
2a1402aa MG |
613 | |
614 | continue; | |
615 | ||
616 | next_pageblock: | |
a9aacbcc | 617 | low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1; |
2a1402aa | 618 | last_pageblock_nr = pageblock_nr; |
748446bb MG |
619 | } |
620 | ||
c67fe375 | 621 | acct_isolated(zone, locked, cc); |
748446bb | 622 | |
c67fe375 MG |
623 | if (locked) |
624 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 625 | |
bb13ffeb MG |
626 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
627 | if (low_pfn == end_pfn) | |
c89511ab | 628 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 629 | |
b7aba698 MG |
630 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
631 | ||
010fc29a | 632 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 633 | if (nr_isolated) |
010fc29a | 634 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 635 | |
2fe86e00 MN |
636 | return low_pfn; |
637 | } | |
638 | ||
ff9543fd MN |
639 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
640 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 641 | /* |
ff9543fd MN |
642 | * Based on information in the current compact_control, find blocks |
643 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 644 | */ |
ff9543fd MN |
645 | static void isolate_freepages(struct zone *zone, |
646 | struct compact_control *cc) | |
2fe86e00 | 647 | { |
ff9543fd | 648 | struct page *page; |
108bcc96 | 649 | unsigned long high_pfn, low_pfn, pfn, z_end_pfn, end_pfn; |
ff9543fd MN |
650 | int nr_freepages = cc->nr_freepages; |
651 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 652 | |
ff9543fd MN |
653 | /* |
654 | * Initialise the free scanner. The starting point is where we last | |
655 | * scanned from (or the end of the zone if starting). The low point | |
656 | * is the end of the pageblock the migration scanner is using. | |
657 | */ | |
658 | pfn = cc->free_pfn; | |
659 | low_pfn = cc->migrate_pfn + pageblock_nr_pages; | |
2fe86e00 | 660 | |
ff9543fd MN |
661 | /* |
662 | * Take care that if the migration scanner is at the end of the zone | |
663 | * that the free scanner does not accidentally move to the next zone | |
664 | * in the next isolation cycle. | |
665 | */ | |
666 | high_pfn = min(low_pfn, pfn); | |
2fe86e00 | 667 | |
108bcc96 | 668 | z_end_pfn = zone_end_pfn(zone); |
2fe86e00 | 669 | |
ff9543fd MN |
670 | /* |
671 | * Isolate free pages until enough are available to migrate the | |
672 | * pages on cc->migratepages. We stop searching if the migrate | |
673 | * and free page scanners meet or enough free pages are isolated. | |
674 | */ | |
675 | for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages; | |
676 | pfn -= pageblock_nr_pages) { | |
677 | unsigned long isolated; | |
2fe86e00 | 678 | |
f6ea3adb DR |
679 | /* |
680 | * This can iterate a massively long zone without finding any | |
681 | * suitable migration targets, so periodically check if we need | |
682 | * to schedule. | |
683 | */ | |
684 | cond_resched(); | |
685 | ||
ff9543fd MN |
686 | if (!pfn_valid(pfn)) |
687 | continue; | |
2fe86e00 | 688 | |
ff9543fd MN |
689 | /* |
690 | * Check for overlapping nodes/zones. It's possible on some | |
691 | * configurations to have a setup like | |
692 | * node0 node1 node0 | |
693 | * i.e. it's possible that all pages within a zones range of | |
694 | * pages do not belong to a single zone. | |
695 | */ | |
696 | page = pfn_to_page(pfn); | |
697 | if (page_zone(page) != zone) | |
698 | continue; | |
699 | ||
700 | /* Check the block is suitable for migration */ | |
68e3e926 | 701 | if (!suitable_migration_target(page)) |
ff9543fd | 702 | continue; |
68e3e926 | 703 | |
bb13ffeb MG |
704 | /* If isolation recently failed, do not retry */ |
705 | if (!isolation_suitable(cc, page)) | |
706 | continue; | |
707 | ||
f40d1e42 | 708 | /* Found a block suitable for isolating free pages from */ |
ff9543fd | 709 | isolated = 0; |
60177d31 MG |
710 | |
711 | /* | |
712 | * As pfn may not start aligned, pfn+pageblock_nr_page | |
713 | * may cross a MAX_ORDER_NR_PAGES boundary and miss | |
714 | * a pfn_valid check. Ensure isolate_freepages_block() | |
715 | * only scans within a pageblock | |
716 | */ | |
717 | end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
108bcc96 | 718 | end_pfn = min(end_pfn, z_end_pfn); |
f40d1e42 MG |
719 | isolated = isolate_freepages_block(cc, pfn, end_pfn, |
720 | freelist, false); | |
721 | nr_freepages += isolated; | |
ff9543fd MN |
722 | |
723 | /* | |
724 | * Record the highest PFN we isolated pages from. When next | |
725 | * looking for free pages, the search will restart here as | |
726 | * page migration may have returned some pages to the allocator | |
727 | */ | |
c89511ab MG |
728 | if (isolated) { |
729 | cc->finished_update_free = true; | |
ff9543fd | 730 | high_pfn = max(high_pfn, pfn); |
c89511ab | 731 | } |
ff9543fd MN |
732 | } |
733 | ||
734 | /* split_free_page does not map the pages */ | |
735 | map_pages(freelist); | |
736 | ||
737 | cc->free_pfn = high_pfn; | |
738 | cc->nr_freepages = nr_freepages; | |
748446bb MG |
739 | } |
740 | ||
741 | /* | |
742 | * This is a migrate-callback that "allocates" freepages by taking pages | |
743 | * from the isolated freelists in the block we are migrating to. | |
744 | */ | |
745 | static struct page *compaction_alloc(struct page *migratepage, | |
746 | unsigned long data, | |
747 | int **result) | |
748 | { | |
749 | struct compact_control *cc = (struct compact_control *)data; | |
750 | struct page *freepage; | |
751 | ||
752 | /* Isolate free pages if necessary */ | |
753 | if (list_empty(&cc->freepages)) { | |
754 | isolate_freepages(cc->zone, cc); | |
755 | ||
756 | if (list_empty(&cc->freepages)) | |
757 | return NULL; | |
758 | } | |
759 | ||
760 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
761 | list_del(&freepage->lru); | |
762 | cc->nr_freepages--; | |
763 | ||
764 | return freepage; | |
765 | } | |
766 | ||
767 | /* | |
768 | * We cannot control nr_migratepages and nr_freepages fully when migration is | |
769 | * running as migrate_pages() has no knowledge of compact_control. When | |
770 | * migration is complete, we count the number of pages on the lists by hand. | |
771 | */ | |
772 | static void update_nr_listpages(struct compact_control *cc) | |
773 | { | |
774 | int nr_migratepages = 0; | |
775 | int nr_freepages = 0; | |
776 | struct page *page; | |
777 | ||
778 | list_for_each_entry(page, &cc->migratepages, lru) | |
779 | nr_migratepages++; | |
780 | list_for_each_entry(page, &cc->freepages, lru) | |
781 | nr_freepages++; | |
782 | ||
783 | cc->nr_migratepages = nr_migratepages; | |
784 | cc->nr_freepages = nr_freepages; | |
785 | } | |
786 | ||
ff9543fd MN |
787 | /* possible outcome of isolate_migratepages */ |
788 | typedef enum { | |
789 | ISOLATE_ABORT, /* Abort compaction now */ | |
790 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
791 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
792 | } isolate_migrate_t; | |
793 | ||
794 | /* | |
795 | * Isolate all pages that can be migrated from the block pointed to by | |
796 | * the migrate scanner within compact_control. | |
797 | */ | |
798 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
799 | struct compact_control *cc) | |
800 | { | |
801 | unsigned long low_pfn, end_pfn; | |
802 | ||
803 | /* Do not scan outside zone boundaries */ | |
804 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
805 | ||
806 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 807 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd MN |
808 | |
809 | /* Do not cross the free scanner or scan within a memory hole */ | |
810 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
811 | cc->migrate_pfn = end_pfn; | |
812 | return ISOLATE_NONE; | |
813 | } | |
814 | ||
815 | /* Perform the isolation */ | |
e46a2879 | 816 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
e64c5237 | 817 | if (!low_pfn || cc->contended) |
ff9543fd MN |
818 | return ISOLATE_ABORT; |
819 | ||
820 | cc->migrate_pfn = low_pfn; | |
821 | ||
822 | return ISOLATE_SUCCESS; | |
823 | } | |
824 | ||
748446bb | 825 | static int compact_finished(struct zone *zone, |
5a03b051 | 826 | struct compact_control *cc) |
748446bb | 827 | { |
8fb74b9f | 828 | unsigned int order; |
5a03b051 | 829 | unsigned long watermark; |
56de7263 | 830 | |
748446bb MG |
831 | if (fatal_signal_pending(current)) |
832 | return COMPACT_PARTIAL; | |
833 | ||
753341a4 | 834 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 835 | if (cc->free_pfn <= cc->migrate_pfn) { |
62997027 MG |
836 | /* |
837 | * Mark that the PG_migrate_skip information should be cleared | |
838 | * by kswapd when it goes to sleep. kswapd does not set the | |
839 | * flag itself as the decision to be clear should be directly | |
840 | * based on an allocation request. | |
841 | */ | |
842 | if (!current_is_kswapd()) | |
843 | zone->compact_blockskip_flush = true; | |
844 | ||
748446bb | 845 | return COMPACT_COMPLETE; |
bb13ffeb | 846 | } |
748446bb | 847 | |
82478fb7 JW |
848 | /* |
849 | * order == -1 is expected when compacting via | |
850 | * /proc/sys/vm/compact_memory | |
851 | */ | |
56de7263 MG |
852 | if (cc->order == -1) |
853 | return COMPACT_CONTINUE; | |
854 | ||
3957c776 MH |
855 | /* Compaction run is not finished if the watermark is not met */ |
856 | watermark = low_wmark_pages(zone); | |
857 | watermark += (1 << cc->order); | |
858 | ||
859 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
860 | return COMPACT_CONTINUE; | |
861 | ||
56de7263 | 862 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
863 | for (order = cc->order; order < MAX_ORDER; order++) { |
864 | struct free_area *area = &zone->free_area[order]; | |
865 | ||
866 | /* Job done if page is free of the right migratetype */ | |
867 | if (!list_empty(&area->free_list[cc->migratetype])) | |
868 | return COMPACT_PARTIAL; | |
869 | ||
870 | /* Job done if allocation would set block type */ | |
871 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
872 | return COMPACT_PARTIAL; |
873 | } | |
874 | ||
748446bb MG |
875 | return COMPACT_CONTINUE; |
876 | } | |
877 | ||
3e7d3449 MG |
878 | /* |
879 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
880 | * Returns | |
881 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
882 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
883 | * COMPACT_CONTINUE - If compaction should run now | |
884 | */ | |
885 | unsigned long compaction_suitable(struct zone *zone, int order) | |
886 | { | |
887 | int fragindex; | |
888 | unsigned long watermark; | |
889 | ||
3957c776 MH |
890 | /* |
891 | * order == -1 is expected when compacting via | |
892 | * /proc/sys/vm/compact_memory | |
893 | */ | |
894 | if (order == -1) | |
895 | return COMPACT_CONTINUE; | |
896 | ||
3e7d3449 MG |
897 | /* |
898 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
899 | * This is because during migration, copies of pages need to be | |
900 | * allocated and for a short time, the footprint is higher | |
901 | */ | |
902 | watermark = low_wmark_pages(zone) + (2UL << order); | |
903 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
904 | return COMPACT_SKIPPED; | |
905 | ||
906 | /* | |
907 | * fragmentation index determines if allocation failures are due to | |
908 | * low memory or external fragmentation | |
909 | * | |
a582a738 SL |
910 | * index of -1000 implies allocations might succeed depending on |
911 | * watermarks | |
3e7d3449 MG |
912 | * index towards 0 implies failure is due to lack of memory |
913 | * index towards 1000 implies failure is due to fragmentation | |
914 | * | |
915 | * Only compact if a failure would be due to fragmentation. | |
916 | */ | |
917 | fragindex = fragmentation_index(zone, order); | |
918 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
919 | return COMPACT_SKIPPED; | |
920 | ||
a582a738 SL |
921 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
922 | 0, 0)) | |
3e7d3449 MG |
923 | return COMPACT_PARTIAL; |
924 | ||
925 | return COMPACT_CONTINUE; | |
926 | } | |
927 | ||
748446bb MG |
928 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
929 | { | |
930 | int ret; | |
c89511ab | 931 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 932 | unsigned long end_pfn = zone_end_pfn(zone); |
748446bb | 933 | |
3e7d3449 MG |
934 | ret = compaction_suitable(zone, cc->order); |
935 | switch (ret) { | |
936 | case COMPACT_PARTIAL: | |
937 | case COMPACT_SKIPPED: | |
938 | /* Compaction is likely to fail */ | |
939 | return ret; | |
940 | case COMPACT_CONTINUE: | |
941 | /* Fall through to compaction */ | |
942 | ; | |
943 | } | |
944 | ||
c89511ab MG |
945 | /* |
946 | * Setup to move all movable pages to the end of the zone. Used cached | |
947 | * information on where the scanners should start but check that it | |
948 | * is initialised by ensuring the values are within zone boundaries. | |
949 | */ | |
950 | cc->migrate_pfn = zone->compact_cached_migrate_pfn; | |
951 | cc->free_pfn = zone->compact_cached_free_pfn; | |
952 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
953 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
954 | zone->compact_cached_free_pfn = cc->free_pfn; | |
955 | } | |
956 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
957 | cc->migrate_pfn = start_pfn; | |
958 | zone->compact_cached_migrate_pfn = cc->migrate_pfn; | |
959 | } | |
748446bb | 960 | |
62997027 MG |
961 | /* |
962 | * Clear pageblock skip if there were failures recently and compaction | |
963 | * is about to be retried after being deferred. kswapd does not do | |
964 | * this reset as it'll reset the cached information when going to sleep. | |
965 | */ | |
966 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
967 | __reset_isolation_suitable(zone); | |
bb13ffeb | 968 | |
748446bb MG |
969 | migrate_prep_local(); |
970 | ||
971 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
972 | unsigned long nr_migrate, nr_remaining; | |
9d502c1c | 973 | int err; |
748446bb | 974 | |
f9e35b3b MG |
975 | switch (isolate_migratepages(zone, cc)) { |
976 | case ISOLATE_ABORT: | |
977 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 978 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 979 | cc->nr_migratepages = 0; |
f9e35b3b MG |
980 | goto out; |
981 | case ISOLATE_NONE: | |
748446bb | 982 | continue; |
f9e35b3b MG |
983 | case ISOLATE_SUCCESS: |
984 | ; | |
985 | } | |
748446bb MG |
986 | |
987 | nr_migrate = cc->nr_migratepages; | |
9d502c1c | 988 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
9c620e2b | 989 | (unsigned long)cc, |
7b2a2d4a MG |
990 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC, |
991 | MR_COMPACTION); | |
748446bb MG |
992 | update_nr_listpages(cc); |
993 | nr_remaining = cc->nr_migratepages; | |
994 | ||
b7aba698 MG |
995 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
996 | nr_remaining); | |
748446bb | 997 | |
5733c7d1 | 998 | /* Release isolated pages not migrated */ |
9d502c1c | 999 | if (err) { |
5733c7d1 | 1000 | putback_movable_pages(&cc->migratepages); |
748446bb | 1001 | cc->nr_migratepages = 0; |
4bf2bba3 DR |
1002 | if (err == -ENOMEM) { |
1003 | ret = COMPACT_PARTIAL; | |
1004 | goto out; | |
1005 | } | |
748446bb | 1006 | } |
748446bb MG |
1007 | } |
1008 | ||
f9e35b3b | 1009 | out: |
748446bb MG |
1010 | /* Release free pages and check accounting */ |
1011 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1012 | VM_BUG_ON(cc->nr_freepages != 0); | |
1013 | ||
1014 | return ret; | |
1015 | } | |
76ab0f53 | 1016 | |
d43a87e6 | 1017 | static unsigned long compact_zone_order(struct zone *zone, |
5a03b051 | 1018 | int order, gfp_t gfp_mask, |
8fb74b9f | 1019 | bool sync, bool *contended) |
56de7263 | 1020 | { |
e64c5237 | 1021 | unsigned long ret; |
56de7263 MG |
1022 | struct compact_control cc = { |
1023 | .nr_freepages = 0, | |
1024 | .nr_migratepages = 0, | |
1025 | .order = order, | |
1026 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1027 | .zone = zone, | |
68e3e926 | 1028 | .sync = sync, |
56de7263 MG |
1029 | }; |
1030 | INIT_LIST_HEAD(&cc.freepages); | |
1031 | INIT_LIST_HEAD(&cc.migratepages); | |
1032 | ||
e64c5237 SL |
1033 | ret = compact_zone(zone, &cc); |
1034 | ||
1035 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1036 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1037 | ||
1038 | *contended = cc.contended; | |
1039 | return ret; | |
56de7263 MG |
1040 | } |
1041 | ||
5e771905 MG |
1042 | int sysctl_extfrag_threshold = 500; |
1043 | ||
56de7263 MG |
1044 | /** |
1045 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1046 | * @zonelist: The zonelist used for the current allocation | |
1047 | * @order: The order of the current allocation | |
1048 | * @gfp_mask: The GFP mask of the current allocation | |
1049 | * @nodemask: The allowed nodes to allocate from | |
77f1fe6b | 1050 | * @sync: Whether migration is synchronous or not |
661c4cb9 MG |
1051 | * @contended: Return value that is true if compaction was aborted due to lock contention |
1052 | * @page: Optionally capture a free page of the requested order during compaction | |
56de7263 MG |
1053 | * |
1054 | * This is the main entry point for direct page compaction. | |
1055 | */ | |
1056 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1057 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
8fb74b9f | 1058 | bool sync, bool *contended) |
56de7263 MG |
1059 | { |
1060 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1061 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1062 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1063 | struct zoneref *z; |
1064 | struct zone *zone; | |
1065 | int rc = COMPACT_SKIPPED; | |
d95ea5d1 | 1066 | int alloc_flags = 0; |
56de7263 | 1067 | |
4ffb6335 | 1068 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1069 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
1070 | return rc; |
1071 | ||
010fc29a | 1072 | count_compact_event(COMPACTSTALL); |
56de7263 | 1073 | |
d95ea5d1 BZ |
1074 | #ifdef CONFIG_CMA |
1075 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1076 | alloc_flags |= ALLOC_CMA; | |
1077 | #endif | |
56de7263 MG |
1078 | /* Compact each zone in the list */ |
1079 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1080 | nodemask) { | |
56de7263 MG |
1081 | int status; |
1082 | ||
c67fe375 | 1083 | status = compact_zone_order(zone, order, gfp_mask, sync, |
8fb74b9f | 1084 | contended); |
56de7263 MG |
1085 | rc = max(status, rc); |
1086 | ||
3e7d3449 | 1087 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 BZ |
1088 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
1089 | alloc_flags)) | |
56de7263 MG |
1090 | break; |
1091 | } | |
1092 | ||
1093 | return rc; | |
1094 | } | |
1095 | ||
1096 | ||
76ab0f53 | 1097 | /* Compact all zones within a node */ |
7103f16d | 1098 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1099 | { |
1100 | int zoneid; | |
76ab0f53 MG |
1101 | struct zone *zone; |
1102 | ||
76ab0f53 | 1103 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1104 | |
1105 | zone = &pgdat->node_zones[zoneid]; | |
1106 | if (!populated_zone(zone)) | |
1107 | continue; | |
1108 | ||
7be62de9 RR |
1109 | cc->nr_freepages = 0; |
1110 | cc->nr_migratepages = 0; | |
1111 | cc->zone = zone; | |
1112 | INIT_LIST_HEAD(&cc->freepages); | |
1113 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1114 | |
aad6ec37 | 1115 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1116 | compact_zone(zone, cc); |
76ab0f53 | 1117 | |
aff62249 RR |
1118 | if (cc->order > 0) { |
1119 | int ok = zone_watermark_ok(zone, cc->order, | |
1120 | low_wmark_pages(zone), 0, 0); | |
c81758fb | 1121 | if (ok && cc->order >= zone->compact_order_failed) |
aff62249 RR |
1122 | zone->compact_order_failed = cc->order + 1; |
1123 | /* Currently async compaction is never deferred. */ | |
68e3e926 | 1124 | else if (!ok && cc->sync) |
aff62249 RR |
1125 | defer_compaction(zone, cc->order); |
1126 | } | |
1127 | ||
7be62de9 RR |
1128 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1129 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1130 | } |
76ab0f53 MG |
1131 | } |
1132 | ||
7103f16d | 1133 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1134 | { |
1135 | struct compact_control cc = { | |
1136 | .order = order, | |
68e3e926 | 1137 | .sync = false, |
7be62de9 RR |
1138 | }; |
1139 | ||
3a7200af MG |
1140 | if (!order) |
1141 | return; | |
1142 | ||
7103f16d | 1143 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1144 | } |
1145 | ||
7103f16d | 1146 | static void compact_node(int nid) |
7be62de9 | 1147 | { |
7be62de9 RR |
1148 | struct compact_control cc = { |
1149 | .order = -1, | |
68e3e926 | 1150 | .sync = true, |
7be62de9 RR |
1151 | }; |
1152 | ||
7103f16d | 1153 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1154 | } |
1155 | ||
76ab0f53 | 1156 | /* Compact all nodes in the system */ |
7964c06d | 1157 | static void compact_nodes(void) |
76ab0f53 MG |
1158 | { |
1159 | int nid; | |
1160 | ||
8575ec29 HD |
1161 | /* Flush pending updates to the LRU lists */ |
1162 | lru_add_drain_all(); | |
1163 | ||
76ab0f53 MG |
1164 | for_each_online_node(nid) |
1165 | compact_node(nid); | |
76ab0f53 MG |
1166 | } |
1167 | ||
1168 | /* The written value is actually unused, all memory is compacted */ | |
1169 | int sysctl_compact_memory; | |
1170 | ||
1171 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1172 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1173 | void __user *buffer, size_t *length, loff_t *ppos) | |
1174 | { | |
1175 | if (write) | |
7964c06d | 1176 | compact_nodes(); |
76ab0f53 MG |
1177 | |
1178 | return 0; | |
1179 | } | |
ed4a6d7f | 1180 | |
5e771905 MG |
1181 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1182 | void __user *buffer, size_t *length, loff_t *ppos) | |
1183 | { | |
1184 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1185 | ||
1186 | return 0; | |
1187 | } | |
1188 | ||
ed4a6d7f | 1189 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
10fbcf4c KS |
1190 | ssize_t sysfs_compact_node(struct device *dev, |
1191 | struct device_attribute *attr, | |
ed4a6d7f MG |
1192 | const char *buf, size_t count) |
1193 | { | |
8575ec29 HD |
1194 | int nid = dev->id; |
1195 | ||
1196 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1197 | /* Flush pending updates to the LRU lists */ | |
1198 | lru_add_drain_all(); | |
1199 | ||
1200 | compact_node(nid); | |
1201 | } | |
ed4a6d7f MG |
1202 | |
1203 | return count; | |
1204 | } | |
10fbcf4c | 1205 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1206 | |
1207 | int compaction_register_node(struct node *node) | |
1208 | { | |
10fbcf4c | 1209 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1210 | } |
1211 | ||
1212 | void compaction_unregister_node(struct node *node) | |
1213 | { | |
10fbcf4c | 1214 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1215 | } |
1216 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1217 | |
1218 | #endif /* CONFIG_COMPACTION */ |