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 | ||
6fa3eb70 S |
21 | #ifdef CONFIG_HAS_EARLYSUSPEND |
22 | #include <linux/earlysuspend.h> | |
23 | #endif | |
24 | ||
010fc29a MK |
25 | #ifdef CONFIG_COMPACTION |
26 | static inline void count_compact_event(enum vm_event_item item) | |
27 | { | |
28 | count_vm_event(item); | |
29 | } | |
30 | ||
31 | static inline void count_compact_events(enum vm_event_item item, long delta) | |
32 | { | |
33 | count_vm_events(item, delta); | |
34 | } | |
35 | #else | |
36 | #define count_compact_event(item) do { } while (0) | |
37 | #define count_compact_events(item, delta) do { } while (0) | |
38 | #endif | |
39 | ||
ff9543fd MN |
40 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
41 | ||
b7aba698 MG |
42 | #define CREATE_TRACE_POINTS |
43 | #include <trace/events/compaction.h> | |
44 | ||
748446bb MG |
45 | static unsigned long release_freepages(struct list_head *freelist) |
46 | { | |
47 | struct page *page, *next; | |
48 | unsigned long count = 0; | |
49 | ||
50 | list_for_each_entry_safe(page, next, freelist, lru) { | |
51 | list_del(&page->lru); | |
52 | __free_page(page); | |
53 | count++; | |
54 | } | |
55 | ||
56 | return count; | |
57 | } | |
58 | ||
ff9543fd MN |
59 | static void map_pages(struct list_head *list) |
60 | { | |
61 | struct page *page; | |
62 | ||
63 | list_for_each_entry(page, list, lru) { | |
64 | arch_alloc_page(page, 0); | |
65 | kernel_map_pages(page, 1, 1); | |
66 | } | |
67 | } | |
68 | ||
47118af0 MN |
69 | static inline bool migrate_async_suitable(int migratetype) |
70 | { | |
71 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; | |
72 | } | |
73 | ||
bb13ffeb MG |
74 | #ifdef CONFIG_COMPACTION |
75 | /* Returns true if the pageblock should be scanned for pages to isolate. */ | |
76 | static inline bool isolation_suitable(struct compact_control *cc, | |
77 | struct page *page) | |
78 | { | |
79 | if (cc->ignore_skip_hint) | |
80 | return true; | |
81 | ||
82 | return !get_pageblock_skip(page); | |
83 | } | |
84 | ||
85 | /* | |
86 | * This function is called to clear all cached information on pageblocks that | |
87 | * should be skipped for page isolation when the migrate and free page scanner | |
88 | * meet. | |
89 | */ | |
62997027 | 90 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
91 | { |
92 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 93 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
94 | unsigned long pfn; |
95 | ||
c89511ab MG |
96 | zone->compact_cached_migrate_pfn = start_pfn; |
97 | zone->compact_cached_free_pfn = end_pfn; | |
62997027 | 98 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
99 | |
100 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
101 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
102 | struct page *page; | |
103 | ||
104 | cond_resched(); | |
105 | ||
106 | if (!pfn_valid(pfn)) | |
107 | continue; | |
108 | ||
109 | page = pfn_to_page(pfn); | |
110 | if (zone != page_zone(page)) | |
111 | continue; | |
112 | ||
113 | clear_pageblock_skip(page); | |
114 | } | |
115 | } | |
116 | ||
62997027 MG |
117 | void reset_isolation_suitable(pg_data_t *pgdat) |
118 | { | |
119 | int zoneid; | |
120 | ||
121 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
122 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
123 | if (!populated_zone(zone)) | |
124 | continue; | |
125 | ||
126 | /* Only flush if a full compaction finished recently */ | |
127 | if (zone->compact_blockskip_flush) | |
128 | __reset_isolation_suitable(zone); | |
129 | } | |
130 | } | |
131 | ||
bb13ffeb MG |
132 | /* |
133 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 134 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 135 | */ |
c89511ab MG |
136 | static void update_pageblock_skip(struct compact_control *cc, |
137 | struct page *page, unsigned long nr_isolated, | |
138 | bool migrate_scanner) | |
bb13ffeb | 139 | { |
c89511ab | 140 | struct zone *zone = cc->zone; |
31eb5f24 JK |
141 | |
142 | if (cc->ignore_skip_hint) | |
143 | return; | |
144 | ||
bb13ffeb MG |
145 | if (!page) |
146 | return; | |
147 | ||
c89511ab MG |
148 | if (!nr_isolated) { |
149 | unsigned long pfn = page_to_pfn(page); | |
bb13ffeb | 150 | set_pageblock_skip(page); |
c89511ab MG |
151 | |
152 | /* Update where compaction should restart */ | |
153 | if (migrate_scanner) { | |
154 | if (!cc->finished_update_migrate && | |
155 | pfn > zone->compact_cached_migrate_pfn) | |
156 | zone->compact_cached_migrate_pfn = pfn; | |
157 | } else { | |
158 | if (!cc->finished_update_free && | |
159 | pfn < zone->compact_cached_free_pfn) | |
160 | zone->compact_cached_free_pfn = pfn; | |
161 | } | |
162 | } | |
bb13ffeb MG |
163 | } |
164 | #else | |
165 | static inline bool isolation_suitable(struct compact_control *cc, | |
166 | struct page *page) | |
167 | { | |
168 | return true; | |
169 | } | |
170 | ||
c89511ab MG |
171 | static void update_pageblock_skip(struct compact_control *cc, |
172 | struct page *page, unsigned long nr_isolated, | |
173 | bool migrate_scanner) | |
bb13ffeb MG |
174 | { |
175 | } | |
176 | #endif /* CONFIG_COMPACTION */ | |
177 | ||
2a1402aa MG |
178 | static inline bool should_release_lock(spinlock_t *lock) |
179 | { | |
180 | return need_resched() || spin_is_contended(lock); | |
181 | } | |
182 | ||
c67fe375 MG |
183 | /* |
184 | * Compaction requires the taking of some coarse locks that are potentially | |
185 | * very heavily contended. Check if the process needs to be scheduled or | |
186 | * if the lock is contended. For async compaction, back out in the event | |
187 | * if contention is severe. For sync compaction, schedule. | |
188 | * | |
189 | * Returns true if the lock is held. | |
190 | * Returns false if the lock is released and compaction should abort | |
191 | */ | |
192 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, | |
193 | bool locked, struct compact_control *cc) | |
194 | { | |
2a1402aa | 195 | if (should_release_lock(lock)) { |
c67fe375 MG |
196 | if (locked) { |
197 | spin_unlock_irqrestore(lock, *flags); | |
198 | locked = false; | |
199 | } | |
200 | ||
201 | /* async aborts if taking too long or contended */ | |
202 | if (!cc->sync) { | |
e64c5237 | 203 | cc->contended = true; |
c67fe375 MG |
204 | return false; |
205 | } | |
206 | ||
207 | cond_resched(); | |
c67fe375 MG |
208 | } |
209 | ||
210 | if (!locked) | |
211 | spin_lock_irqsave(lock, *flags); | |
212 | return true; | |
213 | } | |
214 | ||
215 | static inline bool compact_trylock_irqsave(spinlock_t *lock, | |
216 | unsigned long *flags, struct compact_control *cc) | |
217 | { | |
218 | return compact_checklock_irqsave(lock, flags, false, cc); | |
219 | } | |
220 | ||
f40d1e42 MG |
221 | /* Returns true if the page is within a block suitable for migration to */ |
222 | static bool suitable_migration_target(struct page *page) | |
223 | { | |
224 | int migratetype = get_pageblock_migratetype(page); | |
225 | ||
226 | /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ | |
194159fb MK |
227 | if (migratetype == MIGRATE_RESERVE) |
228 | return false; | |
229 | ||
230 | if (is_migrate_isolate(migratetype)) | |
f40d1e42 MG |
231 | return false; |
232 | ||
6fa3eb70 S |
233 | if (is_migrate_mtkpasr(migratetype)) |
234 | return false; | |
235 | ||
f40d1e42 MG |
236 | /* If the page is a large free page, then allow migration */ |
237 | if (PageBuddy(page) && page_order(page) >= pageblock_order) | |
238 | return true; | |
239 | ||
240 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
241 | if (migrate_async_suitable(migratetype)) | |
242 | return true; | |
243 | ||
244 | /* Otherwise skip the block */ | |
245 | return false; | |
246 | } | |
247 | ||
85aa125f MN |
248 | /* |
249 | * Isolate free pages onto a private freelist. Caller must hold zone->lock. | |
250 | * If @strict is true, will abort returning 0 on any invalid PFNs or non-free | |
251 | * pages inside of the pageblock (even though it may still end up isolating | |
252 | * some pages). | |
253 | */ | |
f40d1e42 MG |
254 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
255 | unsigned long blockpfn, | |
85aa125f MN |
256 | unsigned long end_pfn, |
257 | struct list_head *freelist, | |
258 | bool strict) | |
748446bb | 259 | { |
b7aba698 | 260 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 261 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
262 | unsigned long flags; |
263 | bool locked = false; | |
748446bb | 264 | |
748446bb MG |
265 | cursor = pfn_to_page(blockpfn); |
266 | ||
f40d1e42 | 267 | /* Isolate free pages. */ |
748446bb MG |
268 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
269 | int isolated, i; | |
270 | struct page *page = cursor; | |
271 | ||
b7aba698 | 272 | nr_scanned++; |
f40d1e42 | 273 | if (!pfn_valid_within(blockpfn)) |
8c54de8f LA |
274 | goto isolate_fail; |
275 | ||
bb13ffeb MG |
276 | if (!valid_page) |
277 | valid_page = page; | |
f40d1e42 | 278 | if (!PageBuddy(page)) |
8c54de8f | 279 | goto isolate_fail; |
f40d1e42 MG |
280 | |
281 | /* | |
282 | * The zone lock must be held to isolate freepages. | |
283 | * Unfortunately this is a very coarse lock and can be | |
284 | * heavily contended if there are parallel allocations | |
285 | * or parallel compactions. For async compaction do not | |
286 | * spin on the lock and we acquire the lock as late as | |
287 | * possible. | |
288 | */ | |
289 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
290 | locked, cc); | |
291 | if (!locked) | |
292 | break; | |
293 | ||
294 | /* Recheck this is a suitable migration target under lock */ | |
295 | if (!strict && !suitable_migration_target(page)) | |
296 | break; | |
748446bb | 297 | |
f40d1e42 MG |
298 | /* Recheck this is a buddy page under lock */ |
299 | if (!PageBuddy(page)) | |
8c54de8f | 300 | goto isolate_fail; |
748446bb MG |
301 | |
302 | /* Found a free page, break it into order-0 pages */ | |
303 | isolated = split_free_page(page); | |
304 | total_isolated += isolated; | |
305 | for (i = 0; i < isolated; i++) { | |
306 | list_add(&page->lru, freelist); | |
307 | page++; | |
308 | } | |
309 | ||
310 | /* If a page was split, advance to the end of it */ | |
311 | if (isolated) { | |
312 | blockpfn += isolated - 1; | |
313 | cursor += isolated - 1; | |
8c54de8f | 314 | continue; |
748446bb | 315 | } |
8c54de8f LA |
316 | |
317 | isolate_fail: | |
318 | if (strict) | |
319 | break; | |
320 | else | |
321 | continue; | |
322 | ||
748446bb MG |
323 | } |
324 | ||
b7aba698 | 325 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
326 | |
327 | /* | |
328 | * If strict isolation is requested by CMA then check that all the | |
329 | * pages requested were isolated. If there were any failures, 0 is | |
330 | * returned and CMA will fail. | |
331 | */ | |
8c54de8f | 332 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
333 | total_isolated = 0; |
334 | ||
335 | if (locked) | |
336 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
337 | ||
bb13ffeb MG |
338 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
339 | if (blockpfn == end_pfn) | |
c89511ab | 340 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 341 | |
010fc29a | 342 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 343 | if (total_isolated) |
010fc29a | 344 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
345 | return total_isolated; |
346 | } | |
347 | ||
85aa125f MN |
348 | /** |
349 | * isolate_freepages_range() - isolate free pages. | |
350 | * @start_pfn: The first PFN to start isolating. | |
351 | * @end_pfn: The one-past-last PFN. | |
352 | * | |
353 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
354 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
355 | * undo its actions and return zero. | |
356 | * | |
357 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
358 | * (which may be greater then end_pfn if end fell in a middle of | |
359 | * a free page). | |
360 | */ | |
ff9543fd | 361 | unsigned long |
bb13ffeb MG |
362 | isolate_freepages_range(struct compact_control *cc, |
363 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 364 | { |
f40d1e42 | 365 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
366 | LIST_HEAD(freelist); |
367 | ||
85aa125f | 368 | for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) { |
bb13ffeb | 369 | if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn))) |
85aa125f MN |
370 | break; |
371 | ||
372 | /* | |
373 | * On subsequent iterations ALIGN() is actually not needed, | |
374 | * but we keep it that we not to complicate the code. | |
375 | */ | |
376 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
377 | block_end_pfn = min(block_end_pfn, end_pfn); | |
378 | ||
bb13ffeb | 379 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 380 | &freelist, true); |
85aa125f MN |
381 | |
382 | /* | |
383 | * In strict mode, isolate_freepages_block() returns 0 if | |
384 | * there are any holes in the block (ie. invalid PFNs or | |
385 | * non-free pages). | |
386 | */ | |
387 | if (!isolated) | |
388 | break; | |
389 | ||
390 | /* | |
391 | * If we managed to isolate pages, it is always (1 << n) * | |
392 | * pageblock_nr_pages for some non-negative n. (Max order | |
393 | * page may span two pageblocks). | |
394 | */ | |
395 | } | |
396 | ||
397 | /* split_free_page does not map the pages */ | |
398 | map_pages(&freelist); | |
399 | ||
400 | if (pfn < end_pfn) { | |
401 | /* Loop terminated early, cleanup. */ | |
402 | release_freepages(&freelist); | |
403 | return 0; | |
404 | } | |
405 | ||
406 | /* We don't use freelists for anything. */ | |
407 | return pfn; | |
408 | } | |
409 | ||
748446bb | 410 | /* Update the number of anon and file isolated pages in the zone */ |
c67fe375 | 411 | static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc) |
748446bb MG |
412 | { |
413 | struct page *page; | |
b9e84ac1 | 414 | unsigned int count[2] = { 0, }; |
748446bb | 415 | |
b9e84ac1 MK |
416 | list_for_each_entry(page, &cc->migratepages, lru) |
417 | count[!!page_is_file_cache(page)]++; | |
748446bb | 418 | |
c67fe375 MG |
419 | /* If locked we can use the interrupt unsafe versions */ |
420 | if (locked) { | |
421 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
422 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
423 | } else { | |
424 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
425 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
426 | } | |
748446bb MG |
427 | } |
428 | ||
429 | /* Similar to reclaim, but different enough that they don't share logic */ | |
430 | static bool too_many_isolated(struct zone *zone) | |
431 | { | |
bc693045 | 432 | unsigned long active, inactive, isolated; |
748446bb MG |
433 | |
434 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
435 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
436 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
437 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
438 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
439 | zone_page_state(zone, NR_ISOLATED_ANON); | |
440 | ||
bc693045 | 441 | return isolated > (inactive + active) / 2; |
748446bb MG |
442 | } |
443 | ||
2fe86e00 MN |
444 | /** |
445 | * isolate_migratepages_range() - isolate all migrate-able pages in range. | |
446 | * @zone: Zone pages are in. | |
447 | * @cc: Compaction control structure. | |
448 | * @low_pfn: The first PFN of the range. | |
449 | * @end_pfn: The one-past-the-last PFN of the range. | |
e46a2879 | 450 | * @unevictable: true if it allows to isolate unevictable pages |
2fe86e00 MN |
451 | * |
452 | * Isolate all pages that can be migrated from the range specified by | |
453 | * [low_pfn, end_pfn). Returns zero if there is a fatal signal | |
454 | * pending), otherwise PFN of the first page that was not scanned | |
455 | * (which may be both less, equal to or more then end_pfn). | |
456 | * | |
457 | * Assumes that cc->migratepages is empty and cc->nr_migratepages is | |
458 | * zero. | |
459 | * | |
460 | * Apart from cc->migratepages and cc->nr_migratetypes this function | |
461 | * does not modify any cc's fields, in particular it does not modify | |
462 | * (or read for that matter) cc->migrate_pfn. | |
748446bb | 463 | */ |
ff9543fd | 464 | unsigned long |
2fe86e00 | 465 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
e46a2879 | 466 | unsigned long low_pfn, unsigned long end_pfn, bool unevictable) |
748446bb | 467 | { |
9927af74 | 468 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
b7aba698 | 469 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 470 | struct list_head *migratelist = &cc->migratepages; |
f3fd4a61 | 471 | isolate_mode_t mode = 0; |
fa9add64 | 472 | struct lruvec *lruvec; |
c67fe375 | 473 | unsigned long flags; |
2a1402aa | 474 | bool locked = false; |
bb13ffeb | 475 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 476 | |
748446bb MG |
477 | /* |
478 | * Ensure that there are not too many pages isolated from the LRU | |
479 | * list by either parallel reclaimers or compaction. If there are, | |
480 | * delay for some time until fewer pages are isolated | |
481 | */ | |
482 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 483 | /* async migration should just abort */ |
68e3e926 | 484 | if (!cc->sync) |
2fe86e00 | 485 | return 0; |
f9e35b3b | 486 | |
748446bb MG |
487 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
488 | ||
489 | if (fatal_signal_pending(current)) | |
2fe86e00 | 490 | return 0; |
748446bb MG |
491 | } |
492 | ||
493 | /* Time to isolate some pages for migration */ | |
b2eef8c0 | 494 | cond_resched(); |
748446bb | 495 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 496 | /* give a chance to irqs before checking need_resched() */ |
2a1402aa MG |
497 | if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) { |
498 | if (should_release_lock(&zone->lru_lock)) { | |
499 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
500 | locked = false; | |
501 | } | |
b2eef8c0 | 502 | } |
c67fe375 | 503 | |
0bf380bc MG |
504 | /* |
505 | * migrate_pfn does not necessarily start aligned to a | |
506 | * pageblock. Ensure that pfn_valid is called when moving | |
507 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
508 | * memory holes within the zone | |
509 | */ | |
510 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
511 | if (!pfn_valid(low_pfn)) { | |
512 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
513 | continue; | |
514 | } | |
515 | } | |
516 | ||
748446bb MG |
517 | if (!pfn_valid_within(low_pfn)) |
518 | continue; | |
b7aba698 | 519 | nr_scanned++; |
748446bb | 520 | |
dc908600 MG |
521 | /* |
522 | * Get the page and ensure the page is within the same zone. | |
523 | * See the comment in isolate_freepages about overlapping | |
524 | * nodes. It is deliberate that the new zone lock is not taken | |
525 | * as memory compaction should not move pages between nodes. | |
526 | */ | |
748446bb | 527 | page = pfn_to_page(low_pfn); |
dc908600 MG |
528 | if (page_zone(page) != zone) |
529 | continue; | |
530 | ||
bb13ffeb MG |
531 | if (!valid_page) |
532 | valid_page = page; | |
533 | ||
534 | /* If isolation recently failed, do not retry */ | |
535 | pageblock_nr = low_pfn >> pageblock_order; | |
536 | if (!isolation_suitable(cc, page)) | |
537 | goto next_pageblock; | |
538 | ||
dc908600 | 539 | /* Skip if free */ |
748446bb MG |
540 | if (PageBuddy(page)) |
541 | continue; | |
542 | ||
9927af74 MG |
543 | /* |
544 | * For async migration, also only scan in MOVABLE blocks. Async | |
545 | * migration is optimistic to see if the minimum amount of work | |
546 | * satisfies the allocation | |
547 | */ | |
68e3e926 | 548 | if (!cc->sync && last_pageblock_nr != pageblock_nr && |
47118af0 | 549 | !migrate_async_suitable(get_pageblock_migratetype(page))) { |
c89511ab | 550 | cc->finished_update_migrate = true; |
2a1402aa | 551 | goto next_pageblock; |
9927af74 MG |
552 | } |
553 | ||
bf6bddf1 RA |
554 | /* |
555 | * Check may be lockless but that's ok as we recheck later. | |
556 | * It's possible to migrate LRU pages and balloon pages | |
557 | * Skip any other type of page | |
558 | */ | |
559 | if (!PageLRU(page)) { | |
560 | if (unlikely(balloon_page_movable(page))) { | |
561 | if (locked && balloon_page_isolate(page)) { | |
562 | /* Successfully isolated */ | |
563 | cc->finished_update_migrate = true; | |
564 | list_add(&page->lru, migratelist); | |
565 | cc->nr_migratepages++; | |
566 | nr_isolated++; | |
567 | goto check_compact_cluster; | |
568 | } | |
569 | } | |
bc835011 | 570 | continue; |
bf6bddf1 | 571 | } |
bc835011 AA |
572 | |
573 | /* | |
2a1402aa MG |
574 | * PageLRU is set. lru_lock normally excludes isolation |
575 | * splitting and collapsing (collapsing has already happened | |
576 | * if PageLRU is set) but the lock is not necessarily taken | |
577 | * here and it is wasteful to take it just to check transhuge. | |
578 | * Check TransHuge without lock and skip the whole pageblock if | |
579 | * it's either a transhuge or hugetlbfs page, as calling | |
580 | * compound_order() without preventing THP from splitting the | |
581 | * page underneath us may return surprising results. | |
bc835011 | 582 | */ |
2a1402aa MG |
583 | if (PageTransHuge(page)) { |
584 | if (!locked) | |
585 | goto next_pageblock; | |
586 | low_pfn += (1 << compound_order(page)) - 1; | |
587 | continue; | |
588 | } | |
589 | ||
590 | /* Check if it is ok to still hold the lock */ | |
591 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
592 | locked, cc); | |
593 | if (!locked || fatal_signal_pending(current)) | |
594 | break; | |
595 | ||
596 | /* Recheck PageLRU and PageTransHuge under lock */ | |
597 | if (!PageLRU(page)) | |
598 | continue; | |
bc835011 AA |
599 | if (PageTransHuge(page)) { |
600 | low_pfn += (1 << compound_order(page)) - 1; | |
601 | continue; | |
602 | } | |
603 | ||
68e3e926 | 604 | if (!cc->sync) |
c8244935 MG |
605 | mode |= ISOLATE_ASYNC_MIGRATE; |
606 | ||
e46a2879 MK |
607 | if (unevictable) |
608 | mode |= ISOLATE_UNEVICTABLE; | |
609 | ||
fa9add64 HD |
610 | lruvec = mem_cgroup_page_lruvec(page, zone); |
611 | ||
748446bb | 612 | /* Try isolate the page */ |
f3fd4a61 | 613 | if (__isolate_lru_page(page, mode) != 0) |
748446bb MG |
614 | continue; |
615 | ||
bc835011 AA |
616 | VM_BUG_ON(PageTransCompound(page)); |
617 | ||
748446bb | 618 | /* Successfully isolated */ |
c89511ab | 619 | cc->finished_update_migrate = true; |
fa9add64 | 620 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
748446bb | 621 | list_add(&page->lru, migratelist); |
748446bb | 622 | cc->nr_migratepages++; |
b7aba698 | 623 | nr_isolated++; |
748446bb | 624 | |
bf6bddf1 | 625 | check_compact_cluster: |
748446bb | 626 | /* Avoid isolating too much */ |
31b8384a HD |
627 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
628 | ++low_pfn; | |
748446bb | 629 | break; |
31b8384a | 630 | } |
2a1402aa MG |
631 | |
632 | continue; | |
633 | ||
634 | next_pageblock: | |
a9aacbcc | 635 | low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1; |
2a1402aa | 636 | last_pageblock_nr = pageblock_nr; |
748446bb MG |
637 | } |
638 | ||
c67fe375 | 639 | acct_isolated(zone, locked, cc); |
748446bb | 640 | |
c67fe375 MG |
641 | if (locked) |
642 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 643 | |
bb13ffeb MG |
644 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
645 | if (low_pfn == end_pfn) | |
c89511ab | 646 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 647 | |
b7aba698 MG |
648 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
649 | ||
010fc29a | 650 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 651 | if (nr_isolated) |
010fc29a | 652 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 653 | |
2fe86e00 MN |
654 | return low_pfn; |
655 | } | |
656 | ||
ff9543fd MN |
657 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
658 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 659 | /* |
ff9543fd MN |
660 | * Based on information in the current compact_control, find blocks |
661 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 662 | */ |
ff9543fd MN |
663 | static void isolate_freepages(struct zone *zone, |
664 | struct compact_control *cc) | |
2fe86e00 | 665 | { |
ff9543fd | 666 | struct page *page; |
ee7f7615 | 667 | unsigned long high_pfn, low_pfn, pfn, z_end_pfn; |
ff9543fd MN |
668 | int nr_freepages = cc->nr_freepages; |
669 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 670 | |
ff9543fd MN |
671 | /* |
672 | * Initialise the free scanner. The starting point is where we last | |
ee7f7615 VB |
673 | * successfully isolated from, zone-cached value, or the end of the |
674 | * zone when isolating for the first time. We need this aligned to | |
675 | * the pageblock boundary, because we do pfn -= pageblock_nr_pages | |
676 | * in the for loop. | |
677 | * The low boundary is the end of the pageblock the migration scanner | |
678 | * is using. | |
ff9543fd | 679 | */ |
ee7f7615 | 680 | pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
87c4475c | 681 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 682 | |
ff9543fd MN |
683 | /* |
684 | * Take care that if the migration scanner is at the end of the zone | |
685 | * that the free scanner does not accidentally move to the next zone | |
686 | * in the next isolation cycle. | |
687 | */ | |
688 | high_pfn = min(low_pfn, pfn); | |
2fe86e00 | 689 | |
108bcc96 | 690 | z_end_pfn = zone_end_pfn(zone); |
2fe86e00 | 691 | |
ff9543fd MN |
692 | /* |
693 | * Isolate free pages until enough are available to migrate the | |
694 | * pages on cc->migratepages. We stop searching if the migrate | |
695 | * and free page scanners meet or enough free pages are isolated. | |
696 | */ | |
87c4475c | 697 | for (; pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
ff9543fd MN |
698 | pfn -= pageblock_nr_pages) { |
699 | unsigned long isolated; | |
ee7f7615 | 700 | unsigned long end_pfn; |
2fe86e00 | 701 | |
ff9543fd MN |
702 | if (!pfn_valid(pfn)) |
703 | continue; | |
2fe86e00 | 704 | |
ff9543fd MN |
705 | /* |
706 | * Check for overlapping nodes/zones. It's possible on some | |
707 | * configurations to have a setup like | |
708 | * node0 node1 node0 | |
709 | * i.e. it's possible that all pages within a zones range of | |
710 | * pages do not belong to a single zone. | |
711 | */ | |
712 | page = pfn_to_page(pfn); | |
713 | if (page_zone(page) != zone) | |
714 | continue; | |
715 | ||
716 | /* Check the block is suitable for migration */ | |
68e3e926 | 717 | if (!suitable_migration_target(page)) |
ff9543fd | 718 | continue; |
68e3e926 | 719 | |
bb13ffeb MG |
720 | /* If isolation recently failed, do not retry */ |
721 | if (!isolation_suitable(cc, page)) | |
722 | continue; | |
723 | ||
f40d1e42 | 724 | /* Found a block suitable for isolating free pages from */ |
ff9543fd | 725 | isolated = 0; |
60177d31 MG |
726 | |
727 | /* | |
ee7f7615 VB |
728 | * Take care when isolating in last pageblock of a zone which |
729 | * ends in the middle of a pageblock. | |
60177d31 | 730 | */ |
ee7f7615 | 731 | end_pfn = min(pfn + pageblock_nr_pages, z_end_pfn); |
f40d1e42 MG |
732 | isolated = isolate_freepages_block(cc, pfn, end_pfn, |
733 | freelist, false); | |
734 | nr_freepages += isolated; | |
ff9543fd MN |
735 | |
736 | /* | |
737 | * Record the highest PFN we isolated pages from. When next | |
738 | * looking for free pages, the search will restart here as | |
739 | * page migration may have returned some pages to the allocator | |
740 | */ | |
c89511ab MG |
741 | if (isolated) { |
742 | cc->finished_update_free = true; | |
ff9543fd | 743 | high_pfn = max(high_pfn, pfn); |
c89511ab | 744 | } |
ff9543fd MN |
745 | } |
746 | ||
747 | /* split_free_page does not map the pages */ | |
748 | map_pages(freelist); | |
749 | ||
87c4475c VB |
750 | /* |
751 | * If we crossed the migrate scanner, we want to keep it that way | |
752 | * so that compact_finished() may detect this | |
753 | */ | |
754 | if (pfn < low_pfn) | |
755 | cc->free_pfn = max(pfn, zone->zone_start_pfn); | |
756 | else | |
757 | cc->free_pfn = high_pfn; | |
ff9543fd | 758 | cc->nr_freepages = nr_freepages; |
748446bb MG |
759 | } |
760 | ||
761 | /* | |
762 | * This is a migrate-callback that "allocates" freepages by taking pages | |
763 | * from the isolated freelists in the block we are migrating to. | |
764 | */ | |
765 | static struct page *compaction_alloc(struct page *migratepage, | |
766 | unsigned long data, | |
767 | int **result) | |
768 | { | |
769 | struct compact_control *cc = (struct compact_control *)data; | |
770 | struct page *freepage; | |
771 | ||
772 | /* Isolate free pages if necessary */ | |
773 | if (list_empty(&cc->freepages)) { | |
774 | isolate_freepages(cc->zone, cc); | |
775 | ||
776 | if (list_empty(&cc->freepages)) | |
777 | return NULL; | |
778 | } | |
779 | ||
780 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
781 | list_del(&freepage->lru); | |
782 | cc->nr_freepages--; | |
783 | ||
784 | return freepage; | |
785 | } | |
786 | ||
787 | /* | |
788 | * We cannot control nr_migratepages and nr_freepages fully when migration is | |
789 | * running as migrate_pages() has no knowledge of compact_control. When | |
790 | * migration is complete, we count the number of pages on the lists by hand. | |
791 | */ | |
792 | static void update_nr_listpages(struct compact_control *cc) | |
793 | { | |
794 | int nr_migratepages = 0; | |
795 | int nr_freepages = 0; | |
796 | struct page *page; | |
797 | ||
798 | list_for_each_entry(page, &cc->migratepages, lru) | |
799 | nr_migratepages++; | |
800 | list_for_each_entry(page, &cc->freepages, lru) | |
801 | nr_freepages++; | |
802 | ||
803 | cc->nr_migratepages = nr_migratepages; | |
804 | cc->nr_freepages = nr_freepages; | |
805 | } | |
806 | ||
ff9543fd MN |
807 | /* possible outcome of isolate_migratepages */ |
808 | typedef enum { | |
809 | ISOLATE_ABORT, /* Abort compaction now */ | |
810 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
811 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
812 | } isolate_migrate_t; | |
813 | ||
814 | /* | |
815 | * Isolate all pages that can be migrated from the block pointed to by | |
816 | * the migrate scanner within compact_control. | |
817 | */ | |
818 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
819 | struct compact_control *cc) | |
820 | { | |
821 | unsigned long low_pfn, end_pfn; | |
822 | ||
823 | /* Do not scan outside zone boundaries */ | |
824 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
825 | ||
826 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 827 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd MN |
828 | |
829 | /* Do not cross the free scanner or scan within a memory hole */ | |
830 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
831 | cc->migrate_pfn = end_pfn; | |
832 | return ISOLATE_NONE; | |
833 | } | |
834 | ||
835 | /* Perform the isolation */ | |
e46a2879 | 836 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
e64c5237 | 837 | if (!low_pfn || cc->contended) |
ff9543fd MN |
838 | return ISOLATE_ABORT; |
839 | ||
840 | cc->migrate_pfn = low_pfn; | |
841 | ||
842 | return ISOLATE_SUCCESS; | |
843 | } | |
844 | ||
748446bb | 845 | static int compact_finished(struct zone *zone, |
5a03b051 | 846 | struct compact_control *cc) |
748446bb | 847 | { |
8fb74b9f | 848 | unsigned int order; |
5a03b051 | 849 | unsigned long watermark; |
56de7263 | 850 | |
748446bb MG |
851 | if (fatal_signal_pending(current)) |
852 | return COMPACT_PARTIAL; | |
853 | ||
753341a4 | 854 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 855 | if (cc->free_pfn <= cc->migrate_pfn) { |
62997027 MG |
856 | /* |
857 | * Mark that the PG_migrate_skip information should be cleared | |
858 | * by kswapd when it goes to sleep. kswapd does not set the | |
859 | * flag itself as the decision to be clear should be directly | |
860 | * based on an allocation request. | |
861 | */ | |
862 | if (!current_is_kswapd()) | |
863 | zone->compact_blockskip_flush = true; | |
864 | ||
748446bb | 865 | return COMPACT_COMPLETE; |
bb13ffeb | 866 | } |
748446bb | 867 | |
82478fb7 JW |
868 | /* |
869 | * order == -1 is expected when compacting via | |
870 | * /proc/sys/vm/compact_memory | |
871 | */ | |
56de7263 MG |
872 | if (cc->order == -1) |
873 | return COMPACT_CONTINUE; | |
874 | ||
3957c776 MH |
875 | /* Compaction run is not finished if the watermark is not met */ |
876 | watermark = low_wmark_pages(zone); | |
877 | watermark += (1 << cc->order); | |
878 | ||
879 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
880 | return COMPACT_CONTINUE; | |
881 | ||
56de7263 | 882 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
883 | for (order = cc->order; order < MAX_ORDER; order++) { |
884 | struct free_area *area = &zone->free_area[order]; | |
885 | ||
886 | /* Job done if page is free of the right migratetype */ | |
887 | if (!list_empty(&area->free_list[cc->migratetype])) | |
888 | return COMPACT_PARTIAL; | |
889 | ||
890 | /* Job done if allocation would set block type */ | |
2295074e | 891 | if (order >= pageblock_order && area->nr_free) |
56de7263 MG |
892 | return COMPACT_PARTIAL; |
893 | } | |
894 | ||
748446bb MG |
895 | return COMPACT_CONTINUE; |
896 | } | |
897 | ||
3e7d3449 MG |
898 | /* |
899 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
900 | * Returns | |
901 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
902 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
903 | * COMPACT_CONTINUE - If compaction should run now | |
904 | */ | |
905 | unsigned long compaction_suitable(struct zone *zone, int order) | |
906 | { | |
907 | int fragindex; | |
908 | unsigned long watermark; | |
909 | ||
3957c776 MH |
910 | /* |
911 | * order == -1 is expected when compacting via | |
912 | * /proc/sys/vm/compact_memory | |
913 | */ | |
914 | if (order == -1) | |
915 | return COMPACT_CONTINUE; | |
916 | ||
3e7d3449 MG |
917 | /* |
918 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
919 | * This is because during migration, copies of pages need to be | |
920 | * allocated and for a short time, the footprint is higher | |
921 | */ | |
922 | watermark = low_wmark_pages(zone) + (2UL << order); | |
923 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
924 | return COMPACT_SKIPPED; | |
925 | ||
926 | /* | |
927 | * fragmentation index determines if allocation failures are due to | |
928 | * low memory or external fragmentation | |
929 | * | |
a582a738 SL |
930 | * index of -1000 implies allocations might succeed depending on |
931 | * watermarks | |
3e7d3449 MG |
932 | * index towards 0 implies failure is due to lack of memory |
933 | * index towards 1000 implies failure is due to fragmentation | |
934 | * | |
935 | * Only compact if a failure would be due to fragmentation. | |
936 | */ | |
937 | fragindex = fragmentation_index(zone, order); | |
938 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
939 | return COMPACT_SKIPPED; | |
940 | ||
a582a738 SL |
941 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
942 | 0, 0)) | |
3e7d3449 MG |
943 | return COMPACT_PARTIAL; |
944 | ||
945 | return COMPACT_CONTINUE; | |
946 | } | |
947 | ||
748446bb MG |
948 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
949 | { | |
950 | int ret; | |
c89511ab | 951 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 952 | unsigned long end_pfn = zone_end_pfn(zone); |
748446bb | 953 | |
3e7d3449 MG |
954 | ret = compaction_suitable(zone, cc->order); |
955 | switch (ret) { | |
956 | case COMPACT_PARTIAL: | |
957 | case COMPACT_SKIPPED: | |
958 | /* Compaction is likely to fail */ | |
959 | return ret; | |
960 | case COMPACT_CONTINUE: | |
961 | /* Fall through to compaction */ | |
962 | ; | |
963 | } | |
964 | ||
e6798424 VB |
965 | /* |
966 | * Clear pageblock skip if there were failures recently and compaction | |
967 | * is about to be retried after being deferred. kswapd does not do | |
968 | * this reset as it'll reset the cached information when going to sleep. | |
969 | */ | |
970 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
971 | __reset_isolation_suitable(zone); | |
972 | ||
c89511ab MG |
973 | /* |
974 | * Setup to move all movable pages to the end of the zone. Used cached | |
975 | * information on where the scanners should start but check that it | |
976 | * is initialised by ensuring the values are within zone boundaries. | |
977 | */ | |
978 | cc->migrate_pfn = zone->compact_cached_migrate_pfn; | |
979 | cc->free_pfn = zone->compact_cached_free_pfn; | |
980 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
981 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
982 | zone->compact_cached_free_pfn = cc->free_pfn; | |
983 | } | |
984 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
985 | cc->migrate_pfn = start_pfn; | |
986 | zone->compact_cached_migrate_pfn = cc->migrate_pfn; | |
987 | } | |
748446bb MG |
988 | |
989 | migrate_prep_local(); | |
990 | ||
991 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
992 | unsigned long nr_migrate, nr_remaining; | |
9d502c1c | 993 | int err; |
748446bb | 994 | |
f9e35b3b MG |
995 | switch (isolate_migratepages(zone, cc)) { |
996 | case ISOLATE_ABORT: | |
997 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 998 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 999 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1000 | goto out; |
1001 | case ISOLATE_NONE: | |
748446bb | 1002 | continue; |
f9e35b3b MG |
1003 | case ISOLATE_SUCCESS: |
1004 | ; | |
1005 | } | |
748446bb MG |
1006 | |
1007 | nr_migrate = cc->nr_migratepages; | |
9d502c1c | 1008 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
9c620e2b | 1009 | (unsigned long)cc, |
7b2a2d4a MG |
1010 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC, |
1011 | MR_COMPACTION); | |
748446bb MG |
1012 | update_nr_listpages(cc); |
1013 | nr_remaining = cc->nr_migratepages; | |
1014 | ||
b7aba698 MG |
1015 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
1016 | nr_remaining); | |
748446bb | 1017 | |
5733c7d1 | 1018 | /* Release isolated pages not migrated */ |
9d502c1c | 1019 | if (err) { |
5733c7d1 | 1020 | putback_movable_pages(&cc->migratepages); |
748446bb | 1021 | cc->nr_migratepages = 0; |
87c4475c VB |
1022 | /* |
1023 | * migrate_pages() may return -ENOMEM when scanners meet | |
1024 | * and we want compact_finished() to detect it | |
1025 | */ | |
1026 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1027 | ret = COMPACT_PARTIAL; |
1028 | goto out; | |
1029 | } | |
748446bb | 1030 | } |
748446bb MG |
1031 | } |
1032 | ||
f9e35b3b | 1033 | out: |
748446bb MG |
1034 | /* Release free pages and check accounting */ |
1035 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1036 | VM_BUG_ON(cc->nr_freepages != 0); | |
1037 | ||
1038 | return ret; | |
1039 | } | |
76ab0f53 | 1040 | |
d43a87e6 | 1041 | static unsigned long compact_zone_order(struct zone *zone, |
5a03b051 | 1042 | int order, gfp_t gfp_mask, |
8fb74b9f | 1043 | bool sync, bool *contended) |
56de7263 | 1044 | { |
e64c5237 | 1045 | unsigned long ret; |
56de7263 MG |
1046 | struct compact_control cc = { |
1047 | .nr_freepages = 0, | |
1048 | .nr_migratepages = 0, | |
1049 | .order = order, | |
1050 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1051 | .zone = zone, | |
68e3e926 | 1052 | .sync = sync, |
56de7263 MG |
1053 | }; |
1054 | INIT_LIST_HEAD(&cc.freepages); | |
1055 | INIT_LIST_HEAD(&cc.migratepages); | |
1056 | ||
e64c5237 SL |
1057 | ret = compact_zone(zone, &cc); |
1058 | ||
1059 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1060 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1061 | ||
1062 | *contended = cc.contended; | |
1063 | return ret; | |
56de7263 MG |
1064 | } |
1065 | ||
5e771905 MG |
1066 | int sysctl_extfrag_threshold = 500; |
1067 | ||
56de7263 MG |
1068 | /** |
1069 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1070 | * @zonelist: The zonelist used for the current allocation | |
1071 | * @order: The order of the current allocation | |
1072 | * @gfp_mask: The GFP mask of the current allocation | |
1073 | * @nodemask: The allowed nodes to allocate from | |
77f1fe6b | 1074 | * @sync: Whether migration is synchronous or not |
661c4cb9 MG |
1075 | * @contended: Return value that is true if compaction was aborted due to lock contention |
1076 | * @page: Optionally capture a free page of the requested order during compaction | |
56de7263 MG |
1077 | * |
1078 | * This is the main entry point for direct page compaction. | |
1079 | */ | |
1080 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1081 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
8fb74b9f | 1082 | bool sync, bool *contended) |
56de7263 MG |
1083 | { |
1084 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1085 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1086 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1087 | struct zoneref *z; |
1088 | struct zone *zone; | |
1089 | int rc = COMPACT_SKIPPED; | |
d95ea5d1 | 1090 | int alloc_flags = 0; |
56de7263 | 1091 | |
4ffb6335 | 1092 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1093 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
1094 | return rc; |
1095 | ||
010fc29a | 1096 | count_compact_event(COMPACTSTALL); |
56de7263 | 1097 | |
d95ea5d1 BZ |
1098 | #ifdef CONFIG_CMA |
1099 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1100 | alloc_flags |= ALLOC_CMA; | |
1101 | #endif | |
56de7263 MG |
1102 | /* Compact each zone in the list */ |
1103 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1104 | nodemask) { | |
56de7263 MG |
1105 | int status; |
1106 | ||
c67fe375 | 1107 | status = compact_zone_order(zone, order, gfp_mask, sync, |
8fb74b9f | 1108 | contended); |
56de7263 MG |
1109 | rc = max(status, rc); |
1110 | ||
3e7d3449 | 1111 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 BZ |
1112 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
1113 | alloc_flags)) | |
56de7263 MG |
1114 | break; |
1115 | } | |
1116 | ||
1117 | return rc; | |
1118 | } | |
1119 | ||
1120 | ||
76ab0f53 | 1121 | /* Compact all zones within a node */ |
7103f16d | 1122 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1123 | { |
1124 | int zoneid; | |
76ab0f53 MG |
1125 | struct zone *zone; |
1126 | ||
76ab0f53 | 1127 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1128 | |
1129 | zone = &pgdat->node_zones[zoneid]; | |
1130 | if (!populated_zone(zone)) | |
1131 | continue; | |
1132 | ||
7be62de9 RR |
1133 | cc->nr_freepages = 0; |
1134 | cc->nr_migratepages = 0; | |
1135 | cc->zone = zone; | |
1136 | INIT_LIST_HEAD(&cc->freepages); | |
1137 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1138 | |
aad6ec37 | 1139 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1140 | compact_zone(zone, cc); |
76ab0f53 | 1141 | |
aff62249 RR |
1142 | if (cc->order > 0) { |
1143 | int ok = zone_watermark_ok(zone, cc->order, | |
1144 | low_wmark_pages(zone), 0, 0); | |
c81758fb | 1145 | if (ok && cc->order >= zone->compact_order_failed) |
aff62249 RR |
1146 | zone->compact_order_failed = cc->order + 1; |
1147 | /* Currently async compaction is never deferred. */ | |
68e3e926 | 1148 | else if (!ok && cc->sync) |
aff62249 RR |
1149 | defer_compaction(zone, cc->order); |
1150 | } | |
1151 | ||
7be62de9 RR |
1152 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1153 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1154 | } |
76ab0f53 MG |
1155 | } |
1156 | ||
7103f16d | 1157 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1158 | { |
1159 | struct compact_control cc = { | |
1160 | .order = order, | |
68e3e926 | 1161 | .sync = false, |
7be62de9 RR |
1162 | }; |
1163 | ||
7103f16d | 1164 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1165 | } |
1166 | ||
7103f16d | 1167 | static void compact_node(int nid) |
7be62de9 | 1168 | { |
7be62de9 RR |
1169 | struct compact_control cc = { |
1170 | .order = -1, | |
68e3e926 | 1171 | .sync = true, |
7be62de9 RR |
1172 | }; |
1173 | ||
7103f16d | 1174 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1175 | } |
1176 | ||
76ab0f53 | 1177 | /* Compact all nodes in the system */ |
7964c06d | 1178 | static void compact_nodes(void) |
76ab0f53 MG |
1179 | { |
1180 | int nid; | |
1181 | ||
8575ec29 HD |
1182 | /* Flush pending updates to the LRU lists */ |
1183 | lru_add_drain_all(); | |
1184 | ||
76ab0f53 MG |
1185 | for_each_online_node(nid) |
1186 | compact_node(nid); | |
76ab0f53 MG |
1187 | } |
1188 | ||
1189 | /* The written value is actually unused, all memory is compacted */ | |
1190 | int sysctl_compact_memory; | |
1191 | ||
1192 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1193 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1194 | void __user *buffer, size_t *length, loff_t *ppos) | |
1195 | { | |
1196 | if (write) | |
7964c06d | 1197 | compact_nodes(); |
76ab0f53 MG |
1198 | |
1199 | return 0; | |
1200 | } | |
ed4a6d7f | 1201 | |
5e771905 MG |
1202 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1203 | void __user *buffer, size_t *length, loff_t *ppos) | |
1204 | { | |
1205 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1206 | ||
1207 | return 0; | |
1208 | } | |
1209 | ||
ed4a6d7f | 1210 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
10fbcf4c KS |
1211 | ssize_t sysfs_compact_node(struct device *dev, |
1212 | struct device_attribute *attr, | |
ed4a6d7f MG |
1213 | const char *buf, size_t count) |
1214 | { | |
8575ec29 HD |
1215 | int nid = dev->id; |
1216 | ||
1217 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1218 | /* Flush pending updates to the LRU lists */ | |
1219 | lru_add_drain_all(); | |
1220 | ||
1221 | compact_node(nid); | |
1222 | } | |
ed4a6d7f MG |
1223 | |
1224 | return count; | |
1225 | } | |
10fbcf4c | 1226 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1227 | |
1228 | int compaction_register_node(struct node *node) | |
1229 | { | |
10fbcf4c | 1230 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1231 | } |
1232 | ||
1233 | void compaction_unregister_node(struct node *node) | |
1234 | { | |
10fbcf4c | 1235 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1236 | } |
1237 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1238 | |
1239 | #endif /* CONFIG_COMPACTION */ | |
6fa3eb70 S |
1240 | |
1241 | #ifdef CONFIG_HAS_EARLYSUSPEND | |
1242 | extern void drop_pagecache(void); | |
1243 | //extern void kick_lmk_from_compaction(gfp_t); | |
1244 | static void kick_compaction_early_suspend(struct early_suspend *h) | |
1245 | { | |
1246 | struct zone *z = &NODE_DATA(0)->node_zones[ZONE_NORMAL]; | |
1247 | int status; | |
1248 | int retry = 3; | |
1249 | int safe_order = THREAD_SIZE_ORDER + 1; | |
1250 | bool contended; | |
1251 | gfp_t gfp_mask = GFP_KERNEL; | |
1252 | ||
1253 | /* Check whether gfp is restricted. */ | |
1254 | if (gfp_mask != (gfp_mask & gfp_allowed_mask)) { | |
1255 | printk("XXXXXX GFP is restricted! XXXXXX\n"); | |
1256 | return; | |
1257 | } | |
1258 | ||
1259 | /* We try retry times at most. */ | |
1260 | while (retry > 0) { | |
1261 | /* If it is safe under low watermark, then break. */ | |
1262 | if (zone_watermark_ok(z, safe_order, low_wmark_pages(z), 0, 0)) | |
1263 | break; | |
1264 | status = compact_zone_order(z, safe_order, gfp_mask, true, &contended); | |
1265 | --retry; | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | static void kick_compaction_late_resume(struct early_suspend *h) | |
1270 | { | |
1271 | /* Do nothing */ | |
1272 | } | |
1273 | ||
1274 | static struct early_suspend kick_compaction_early_suspend_desc = { | |
1275 | .level = EARLY_SUSPEND_LEVEL_DISABLE_FB + 1, | |
1276 | .suspend = kick_compaction_early_suspend, | |
1277 | .resume = kick_compaction_late_resume, | |
1278 | }; | |
1279 | ||
1280 | static int __init compaction_init(void) | |
1281 | { | |
1282 | printk("@@@@@@ [%s] Register early suspend callback @@@@@@\n",__FUNCTION__); | |
1283 | register_early_suspend(&kick_compaction_early_suspend_desc); | |
1284 | return 0; | |
1285 | } | |
1286 | static void __exit compaction_exit(void) | |
1287 | { | |
1288 | printk("@@@@@@ [%s] Unregister early suspend callback @@@@@@\n",__FUNCTION__); | |
1289 | unregister_early_suspend(&kick_compaction_early_suspend_desc); | |
1290 | } | |
1291 | ||
1292 | module_init(compaction_init); | |
1293 | module_exit(compaction_exit); | |
1294 | #endif |