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 | ||
35979ef3 DR |
92 | zone->compact_cached_migrate_pfn[0] = start_pfn; |
93 | zone->compact_cached_migrate_pfn[1] = start_pfn; | |
c89511ab | 94 | zone->compact_cached_free_pfn = end_pfn; |
62997027 | 95 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
96 | |
97 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
98 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
99 | struct page *page; | |
100 | ||
101 | cond_resched(); | |
102 | ||
103 | if (!pfn_valid(pfn)) | |
104 | continue; | |
105 | ||
106 | page = pfn_to_page(pfn); | |
107 | if (zone != page_zone(page)) | |
108 | continue; | |
109 | ||
110 | clear_pageblock_skip(page); | |
111 | } | |
112 | } | |
113 | ||
62997027 MG |
114 | void reset_isolation_suitable(pg_data_t *pgdat) |
115 | { | |
116 | int zoneid; | |
117 | ||
118 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
119 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
120 | if (!populated_zone(zone)) | |
121 | continue; | |
122 | ||
123 | /* Only flush if a full compaction finished recently */ | |
124 | if (zone->compact_blockskip_flush) | |
125 | __reset_isolation_suitable(zone); | |
126 | } | |
127 | } | |
128 | ||
bb13ffeb MG |
129 | /* |
130 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 131 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 132 | */ |
c89511ab MG |
133 | static void update_pageblock_skip(struct compact_control *cc, |
134 | struct page *page, unsigned long nr_isolated, | |
35979ef3 | 135 | bool set_unsuitable, bool migrate_scanner) |
bb13ffeb | 136 | { |
c89511ab | 137 | struct zone *zone = cc->zone; |
35979ef3 | 138 | unsigned long pfn; |
6815bf3f JK |
139 | |
140 | if (cc->ignore_skip_hint) | |
141 | return; | |
142 | ||
bb13ffeb MG |
143 | if (!page) |
144 | return; | |
145 | ||
35979ef3 DR |
146 | if (nr_isolated) |
147 | return; | |
148 | ||
149 | /* | |
150 | * Only skip pageblocks when all forms of compaction will be known to | |
151 | * fail in the near future. | |
152 | */ | |
153 | if (set_unsuitable) | |
bb13ffeb | 154 | set_pageblock_skip(page); |
c89511ab | 155 | |
35979ef3 DR |
156 | pfn = page_to_pfn(page); |
157 | ||
158 | /* Update where async and sync compaction should restart */ | |
159 | if (migrate_scanner) { | |
160 | if (cc->finished_update_migrate) | |
161 | return; | |
162 | if (pfn > zone->compact_cached_migrate_pfn[0]) | |
163 | zone->compact_cached_migrate_pfn[0] = pfn; | |
e0b9daeb DR |
164 | if (cc->mode != MIGRATE_ASYNC && |
165 | pfn > zone->compact_cached_migrate_pfn[1]) | |
35979ef3 DR |
166 | zone->compact_cached_migrate_pfn[1] = pfn; |
167 | } else { | |
168 | if (cc->finished_update_free) | |
169 | return; | |
170 | if (pfn < zone->compact_cached_free_pfn) | |
171 | zone->compact_cached_free_pfn = pfn; | |
c89511ab | 172 | } |
bb13ffeb MG |
173 | } |
174 | #else | |
175 | static inline bool isolation_suitable(struct compact_control *cc, | |
176 | struct page *page) | |
177 | { | |
178 | return true; | |
179 | } | |
180 | ||
c89511ab MG |
181 | static void update_pageblock_skip(struct compact_control *cc, |
182 | struct page *page, unsigned long nr_isolated, | |
35979ef3 | 183 | bool set_unsuitable, bool migrate_scanner) |
bb13ffeb MG |
184 | { |
185 | } | |
186 | #endif /* CONFIG_COMPACTION */ | |
187 | ||
2a1402aa MG |
188 | static inline bool should_release_lock(spinlock_t *lock) |
189 | { | |
190 | return need_resched() || spin_is_contended(lock); | |
191 | } | |
192 | ||
c67fe375 MG |
193 | /* |
194 | * Compaction requires the taking of some coarse locks that are potentially | |
195 | * very heavily contended. Check if the process needs to be scheduled or | |
196 | * if the lock is contended. For async compaction, back out in the event | |
197 | * if contention is severe. For sync compaction, schedule. | |
198 | * | |
199 | * Returns true if the lock is held. | |
200 | * Returns false if the lock is released and compaction should abort | |
201 | */ | |
202 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, | |
203 | bool locked, struct compact_control *cc) | |
204 | { | |
2a1402aa | 205 | if (should_release_lock(lock)) { |
c67fe375 MG |
206 | if (locked) { |
207 | spin_unlock_irqrestore(lock, *flags); | |
208 | locked = false; | |
209 | } | |
210 | ||
211 | /* async aborts if taking too long or contended */ | |
e0b9daeb | 212 | if (cc->mode == MIGRATE_ASYNC) { |
e64c5237 | 213 | cc->contended = true; |
c67fe375 MG |
214 | return false; |
215 | } | |
216 | ||
217 | cond_resched(); | |
c67fe375 MG |
218 | } |
219 | ||
220 | if (!locked) | |
221 | spin_lock_irqsave(lock, *flags); | |
222 | return true; | |
223 | } | |
224 | ||
be976572 VB |
225 | /* |
226 | * Aside from avoiding lock contention, compaction also periodically checks | |
227 | * need_resched() and either schedules in sync compaction or aborts async | |
228 | * compaction. This is similar to what compact_checklock_irqsave() does, but | |
229 | * is used where no lock is concerned. | |
230 | * | |
231 | * Returns false when no scheduling was needed, or sync compaction scheduled. | |
232 | * Returns true when async compaction should abort. | |
233 | */ | |
234 | static inline bool compact_should_abort(struct compact_control *cc) | |
235 | { | |
236 | /* async compaction aborts if contended */ | |
237 | if (need_resched()) { | |
238 | if (cc->mode == MIGRATE_ASYNC) { | |
239 | cc->contended = true; | |
240 | return true; | |
241 | } | |
242 | ||
243 | cond_resched(); | |
244 | } | |
245 | ||
246 | return false; | |
247 | } | |
248 | ||
f40d1e42 MG |
249 | /* Returns true if the page is within a block suitable for migration to */ |
250 | static bool suitable_migration_target(struct page *page) | |
251 | { | |
7d348b9e | 252 | /* If the page is a large free page, then disallow migration */ |
f40d1e42 | 253 | if (PageBuddy(page) && page_order(page) >= pageblock_order) |
7d348b9e | 254 | return false; |
f40d1e42 MG |
255 | |
256 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
7d348b9e | 257 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
f40d1e42 MG |
258 | return true; |
259 | ||
260 | /* Otherwise skip the block */ | |
261 | return false; | |
262 | } | |
263 | ||
85aa125f | 264 | /* |
9e4be470 JM |
265 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
266 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
267 | * (even though it may still end up isolating some pages). | |
85aa125f | 268 | */ |
f40d1e42 MG |
269 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
270 | unsigned long blockpfn, | |
85aa125f MN |
271 | unsigned long end_pfn, |
272 | struct list_head *freelist, | |
273 | bool strict) | |
748446bb | 274 | { |
b7aba698 | 275 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 276 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
277 | unsigned long flags; |
278 | bool locked = false; | |
01ead534 | 279 | bool checked_pageblock = false; |
748446bb | 280 | |
748446bb MG |
281 | cursor = pfn_to_page(blockpfn); |
282 | ||
f40d1e42 | 283 | /* Isolate free pages. */ |
748446bb MG |
284 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
285 | int isolated, i; | |
286 | struct page *page = cursor; | |
287 | ||
b7aba698 | 288 | nr_scanned++; |
f40d1e42 | 289 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
290 | goto isolate_fail; |
291 | ||
bb13ffeb MG |
292 | if (!valid_page) |
293 | valid_page = page; | |
f40d1e42 | 294 | if (!PageBuddy(page)) |
2af120bc | 295 | goto isolate_fail; |
f40d1e42 MG |
296 | |
297 | /* | |
298 | * The zone lock must be held to isolate freepages. | |
299 | * Unfortunately this is a very coarse lock and can be | |
300 | * heavily contended if there are parallel allocations | |
301 | * or parallel compactions. For async compaction do not | |
302 | * spin on the lock and we acquire the lock as late as | |
303 | * possible. | |
304 | */ | |
305 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
306 | locked, cc); | |
307 | if (!locked) | |
308 | break; | |
309 | ||
310 | /* Recheck this is a suitable migration target under lock */ | |
01ead534 JK |
311 | if (!strict && !checked_pageblock) { |
312 | /* | |
313 | * We need to check suitability of pageblock only once | |
314 | * and this isolate_freepages_block() is called with | |
315 | * pageblock range, so just check once is sufficient. | |
316 | */ | |
317 | checked_pageblock = true; | |
318 | if (!suitable_migration_target(page)) | |
319 | break; | |
320 | } | |
748446bb | 321 | |
f40d1e42 MG |
322 | /* Recheck this is a buddy page under lock */ |
323 | if (!PageBuddy(page)) | |
2af120bc | 324 | goto isolate_fail; |
748446bb MG |
325 | |
326 | /* Found a free page, break it into order-0 pages */ | |
327 | isolated = split_free_page(page); | |
328 | total_isolated += isolated; | |
329 | for (i = 0; i < isolated; i++) { | |
330 | list_add(&page->lru, freelist); | |
331 | page++; | |
332 | } | |
333 | ||
334 | /* If a page was split, advance to the end of it */ | |
335 | if (isolated) { | |
336 | blockpfn += isolated - 1; | |
337 | cursor += isolated - 1; | |
2af120bc | 338 | continue; |
748446bb | 339 | } |
2af120bc LA |
340 | |
341 | isolate_fail: | |
342 | if (strict) | |
343 | break; | |
344 | else | |
345 | continue; | |
346 | ||
748446bb MG |
347 | } |
348 | ||
b7aba698 | 349 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
350 | |
351 | /* | |
352 | * If strict isolation is requested by CMA then check that all the | |
353 | * pages requested were isolated. If there were any failures, 0 is | |
354 | * returned and CMA will fail. | |
355 | */ | |
2af120bc | 356 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
357 | total_isolated = 0; |
358 | ||
359 | if (locked) | |
360 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
361 | ||
bb13ffeb MG |
362 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
363 | if (blockpfn == end_pfn) | |
35979ef3 DR |
364 | update_pageblock_skip(cc, valid_page, total_isolated, true, |
365 | false); | |
bb13ffeb | 366 | |
010fc29a | 367 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 368 | if (total_isolated) |
010fc29a | 369 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
370 | return total_isolated; |
371 | } | |
372 | ||
85aa125f MN |
373 | /** |
374 | * isolate_freepages_range() - isolate free pages. | |
375 | * @start_pfn: The first PFN to start isolating. | |
376 | * @end_pfn: The one-past-last PFN. | |
377 | * | |
378 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
379 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
380 | * undo its actions and return zero. | |
381 | * | |
382 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
383 | * (which may be greater then end_pfn if end fell in a middle of | |
384 | * a free page). | |
385 | */ | |
ff9543fd | 386 | unsigned long |
bb13ffeb MG |
387 | isolate_freepages_range(struct compact_control *cc, |
388 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 389 | { |
f40d1e42 | 390 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
391 | LIST_HEAD(freelist); |
392 | ||
85aa125f | 393 | for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) { |
bb13ffeb | 394 | if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn))) |
85aa125f MN |
395 | break; |
396 | ||
397 | /* | |
398 | * On subsequent iterations ALIGN() is actually not needed, | |
399 | * but we keep it that we not to complicate the code. | |
400 | */ | |
401 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
402 | block_end_pfn = min(block_end_pfn, end_pfn); | |
403 | ||
bb13ffeb | 404 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 405 | &freelist, true); |
85aa125f MN |
406 | |
407 | /* | |
408 | * In strict mode, isolate_freepages_block() returns 0 if | |
409 | * there are any holes in the block (ie. invalid PFNs or | |
410 | * non-free pages). | |
411 | */ | |
412 | if (!isolated) | |
413 | break; | |
414 | ||
415 | /* | |
416 | * If we managed to isolate pages, it is always (1 << n) * | |
417 | * pageblock_nr_pages for some non-negative n. (Max order | |
418 | * page may span two pageblocks). | |
419 | */ | |
420 | } | |
421 | ||
422 | /* split_free_page does not map the pages */ | |
423 | map_pages(&freelist); | |
424 | ||
425 | if (pfn < end_pfn) { | |
426 | /* Loop terminated early, cleanup. */ | |
427 | release_freepages(&freelist); | |
428 | return 0; | |
429 | } | |
430 | ||
431 | /* We don't use freelists for anything. */ | |
432 | return pfn; | |
433 | } | |
434 | ||
748446bb | 435 | /* Update the number of anon and file isolated pages in the zone */ |
c67fe375 | 436 | static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc) |
748446bb MG |
437 | { |
438 | struct page *page; | |
b9e84ac1 | 439 | unsigned int count[2] = { 0, }; |
748446bb | 440 | |
b9e84ac1 MK |
441 | list_for_each_entry(page, &cc->migratepages, lru) |
442 | count[!!page_is_file_cache(page)]++; | |
748446bb | 443 | |
c67fe375 MG |
444 | /* If locked we can use the interrupt unsafe versions */ |
445 | if (locked) { | |
446 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
447 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
448 | } else { | |
449 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
450 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
451 | } | |
748446bb MG |
452 | } |
453 | ||
454 | /* Similar to reclaim, but different enough that they don't share logic */ | |
455 | static bool too_many_isolated(struct zone *zone) | |
456 | { | |
bc693045 | 457 | unsigned long active, inactive, isolated; |
748446bb MG |
458 | |
459 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
460 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
461 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
462 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
463 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
464 | zone_page_state(zone, NR_ISOLATED_ANON); | |
465 | ||
bc693045 | 466 | return isolated > (inactive + active) / 2; |
748446bb MG |
467 | } |
468 | ||
2fe86e00 MN |
469 | /** |
470 | * isolate_migratepages_range() - isolate all migrate-able pages in range. | |
471 | * @zone: Zone pages are in. | |
472 | * @cc: Compaction control structure. | |
473 | * @low_pfn: The first PFN of the range. | |
474 | * @end_pfn: The one-past-the-last PFN of the range. | |
e46a2879 | 475 | * @unevictable: true if it allows to isolate unevictable pages |
2fe86e00 MN |
476 | * |
477 | * Isolate all pages that can be migrated from the range specified by | |
478 | * [low_pfn, end_pfn). Returns zero if there is a fatal signal | |
479 | * pending), otherwise PFN of the first page that was not scanned | |
480 | * (which may be both less, equal to or more then end_pfn). | |
481 | * | |
482 | * Assumes that cc->migratepages is empty and cc->nr_migratepages is | |
483 | * zero. | |
484 | * | |
485 | * Apart from cc->migratepages and cc->nr_migratetypes this function | |
486 | * does not modify any cc's fields, in particular it does not modify | |
487 | * (or read for that matter) cc->migrate_pfn. | |
748446bb | 488 | */ |
ff9543fd | 489 | unsigned long |
2fe86e00 | 490 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
e46a2879 | 491 | unsigned long low_pfn, unsigned long end_pfn, bool unevictable) |
748446bb | 492 | { |
9927af74 | 493 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
b7aba698 | 494 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 495 | struct list_head *migratelist = &cc->migratepages; |
fa9add64 | 496 | struct lruvec *lruvec; |
c67fe375 | 497 | unsigned long flags; |
2a1402aa | 498 | bool locked = false; |
bb13ffeb | 499 | struct page *page = NULL, *valid_page = NULL; |
35979ef3 | 500 | bool set_unsuitable = true; |
e0b9daeb DR |
501 | const isolate_mode_t mode = (cc->mode == MIGRATE_ASYNC ? |
502 | ISOLATE_ASYNC_MIGRATE : 0) | | |
da1c67a7 | 503 | (unevictable ? ISOLATE_UNEVICTABLE : 0); |
748446bb | 504 | |
748446bb MG |
505 | /* |
506 | * Ensure that there are not too many pages isolated from the LRU | |
507 | * list by either parallel reclaimers or compaction. If there are, | |
508 | * delay for some time until fewer pages are isolated | |
509 | */ | |
510 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 511 | /* async migration should just abort */ |
e0b9daeb | 512 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 513 | return 0; |
f9e35b3b | 514 | |
748446bb MG |
515 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
516 | ||
517 | if (fatal_signal_pending(current)) | |
2fe86e00 | 518 | return 0; |
748446bb MG |
519 | } |
520 | ||
be976572 VB |
521 | if (compact_should_abort(cc)) |
522 | return 0; | |
aeef4b83 | 523 | |
748446bb | 524 | /* Time to isolate some pages for migration */ |
748446bb | 525 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 526 | /* give a chance to irqs before checking need_resched() */ |
be1aa03b | 527 | if (locked && !(low_pfn % SWAP_CLUSTER_MAX)) { |
2a1402aa MG |
528 | if (should_release_lock(&zone->lru_lock)) { |
529 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
530 | locked = false; | |
531 | } | |
b2eef8c0 | 532 | } |
c67fe375 | 533 | |
0bf380bc MG |
534 | /* |
535 | * migrate_pfn does not necessarily start aligned to a | |
536 | * pageblock. Ensure that pfn_valid is called when moving | |
537 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
538 | * memory holes within the zone | |
539 | */ | |
540 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
541 | if (!pfn_valid(low_pfn)) { | |
542 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
543 | continue; | |
544 | } | |
545 | } | |
546 | ||
748446bb MG |
547 | if (!pfn_valid_within(low_pfn)) |
548 | continue; | |
b7aba698 | 549 | nr_scanned++; |
748446bb | 550 | |
dc908600 MG |
551 | /* |
552 | * Get the page and ensure the page is within the same zone. | |
553 | * See the comment in isolate_freepages about overlapping | |
554 | * nodes. It is deliberate that the new zone lock is not taken | |
555 | * as memory compaction should not move pages between nodes. | |
556 | */ | |
748446bb | 557 | page = pfn_to_page(low_pfn); |
dc908600 MG |
558 | if (page_zone(page) != zone) |
559 | continue; | |
560 | ||
bb13ffeb MG |
561 | if (!valid_page) |
562 | valid_page = page; | |
563 | ||
564 | /* If isolation recently failed, do not retry */ | |
565 | pageblock_nr = low_pfn >> pageblock_order; | |
c122b208 JK |
566 | if (last_pageblock_nr != pageblock_nr) { |
567 | int mt; | |
568 | ||
569 | last_pageblock_nr = pageblock_nr; | |
570 | if (!isolation_suitable(cc, page)) | |
571 | goto next_pageblock; | |
572 | ||
573 | /* | |
574 | * For async migration, also only scan in MOVABLE | |
575 | * blocks. Async migration is optimistic to see if | |
576 | * the minimum amount of work satisfies the allocation | |
577 | */ | |
578 | mt = get_pageblock_migratetype(page); | |
e0b9daeb DR |
579 | if (cc->mode == MIGRATE_ASYNC && |
580 | !migrate_async_suitable(mt)) { | |
35979ef3 | 581 | set_unsuitable = false; |
c122b208 JK |
582 | goto next_pageblock; |
583 | } | |
584 | } | |
bb13ffeb | 585 | |
6c14466c MG |
586 | /* |
587 | * Skip if free. page_order cannot be used without zone->lock | |
588 | * as nothing prevents parallel allocations or buddy merging. | |
589 | */ | |
748446bb MG |
590 | if (PageBuddy(page)) |
591 | continue; | |
592 | ||
bf6bddf1 RA |
593 | /* |
594 | * Check may be lockless but that's ok as we recheck later. | |
595 | * It's possible to migrate LRU pages and balloon pages | |
596 | * Skip any other type of page | |
597 | */ | |
598 | if (!PageLRU(page)) { | |
599 | if (unlikely(balloon_page_movable(page))) { | |
600 | if (locked && balloon_page_isolate(page)) { | |
601 | /* Successfully isolated */ | |
b6c75016 | 602 | goto isolate_success; |
bf6bddf1 RA |
603 | } |
604 | } | |
bc835011 | 605 | continue; |
bf6bddf1 | 606 | } |
bc835011 AA |
607 | |
608 | /* | |
2a1402aa MG |
609 | * PageLRU is set. lru_lock normally excludes isolation |
610 | * splitting and collapsing (collapsing has already happened | |
611 | * if PageLRU is set) but the lock is not necessarily taken | |
612 | * here and it is wasteful to take it just to check transhuge. | |
613 | * Check TransHuge without lock and skip the whole pageblock if | |
614 | * it's either a transhuge or hugetlbfs page, as calling | |
615 | * compound_order() without preventing THP from splitting the | |
616 | * page underneath us may return surprising results. | |
bc835011 | 617 | */ |
2a1402aa MG |
618 | if (PageTransHuge(page)) { |
619 | if (!locked) | |
620 | goto next_pageblock; | |
621 | low_pfn += (1 << compound_order(page)) - 1; | |
622 | continue; | |
623 | } | |
624 | ||
119d6d59 DR |
625 | /* |
626 | * Migration will fail if an anonymous page is pinned in memory, | |
627 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
628 | * admittedly racy check. | |
629 | */ | |
630 | if (!page_mapping(page) && | |
631 | page_count(page) > page_mapcount(page)) | |
632 | continue; | |
633 | ||
2a1402aa MG |
634 | /* Check if it is ok to still hold the lock */ |
635 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
636 | locked, cc); | |
637 | if (!locked || fatal_signal_pending(current)) | |
638 | break; | |
639 | ||
640 | /* Recheck PageLRU and PageTransHuge under lock */ | |
641 | if (!PageLRU(page)) | |
642 | continue; | |
bc835011 AA |
643 | if (PageTransHuge(page)) { |
644 | low_pfn += (1 << compound_order(page)) - 1; | |
645 | continue; | |
646 | } | |
647 | ||
fa9add64 HD |
648 | lruvec = mem_cgroup_page_lruvec(page, zone); |
649 | ||
748446bb | 650 | /* Try isolate the page */ |
f3fd4a61 | 651 | if (__isolate_lru_page(page, mode) != 0) |
748446bb MG |
652 | continue; |
653 | ||
309381fe | 654 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 655 | |
748446bb | 656 | /* Successfully isolated */ |
fa9add64 | 657 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
658 | |
659 | isolate_success: | |
660 | cc->finished_update_migrate = true; | |
748446bb | 661 | list_add(&page->lru, migratelist); |
748446bb | 662 | cc->nr_migratepages++; |
b7aba698 | 663 | nr_isolated++; |
748446bb MG |
664 | |
665 | /* Avoid isolating too much */ | |
31b8384a HD |
666 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
667 | ++low_pfn; | |
748446bb | 668 | break; |
31b8384a | 669 | } |
2a1402aa MG |
670 | |
671 | continue; | |
672 | ||
673 | next_pageblock: | |
a9aacbcc | 674 | low_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages) - 1; |
748446bb MG |
675 | } |
676 | ||
c67fe375 | 677 | acct_isolated(zone, locked, cc); |
748446bb | 678 | |
c67fe375 MG |
679 | if (locked) |
680 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 681 | |
50b5b094 VB |
682 | /* |
683 | * Update the pageblock-skip information and cached scanner pfn, | |
684 | * if the whole pageblock was scanned without isolating any page. | |
50b5b094 | 685 | */ |
35979ef3 DR |
686 | if (low_pfn == end_pfn) |
687 | update_pageblock_skip(cc, valid_page, nr_isolated, | |
688 | set_unsuitable, true); | |
bb13ffeb | 689 | |
b7aba698 MG |
690 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
691 | ||
010fc29a | 692 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 693 | if (nr_isolated) |
010fc29a | 694 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 695 | |
2fe86e00 MN |
696 | return low_pfn; |
697 | } | |
698 | ||
ff9543fd MN |
699 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
700 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 701 | /* |
ff9543fd MN |
702 | * Based on information in the current compact_control, find blocks |
703 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 704 | */ |
ff9543fd MN |
705 | static void isolate_freepages(struct zone *zone, |
706 | struct compact_control *cc) | |
2fe86e00 | 707 | { |
ff9543fd | 708 | struct page *page; |
c96b9e50 VB |
709 | unsigned long block_start_pfn; /* start of current pageblock */ |
710 | unsigned long block_end_pfn; /* end of current pageblock */ | |
711 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd MN |
712 | int nr_freepages = cc->nr_freepages; |
713 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 714 | |
ff9543fd MN |
715 | /* |
716 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 VB |
717 | * successfully isolated from, zone-cached value, or the end of the |
718 | * zone when isolating for the first time. We need this aligned to | |
c96b9e50 VB |
719 | * the pageblock boundary, because we do |
720 | * block_start_pfn -= pageblock_nr_pages in the for loop. | |
721 | * For ending point, take care when isolating in last pageblock of a | |
722 | * a zone which ends in the middle of a pageblock. | |
49e068f0 VB |
723 | * The low boundary is the end of the pageblock the migration scanner |
724 | * is using. | |
ff9543fd | 725 | */ |
c96b9e50 VB |
726 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
727 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, | |
728 | zone_end_pfn(zone)); | |
7ed695e0 | 729 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 730 | |
ff9543fd MN |
731 | /* |
732 | * Isolate free pages until enough are available to migrate the | |
733 | * pages on cc->migratepages. We stop searching if the migrate | |
734 | * and free page scanners meet or enough free pages are isolated. | |
735 | */ | |
c96b9e50 VB |
736 | for (; block_start_pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
737 | block_end_pfn = block_start_pfn, | |
738 | block_start_pfn -= pageblock_nr_pages) { | |
ff9543fd | 739 | unsigned long isolated; |
2fe86e00 | 740 | |
f6ea3adb DR |
741 | /* |
742 | * This can iterate a massively long zone without finding any | |
743 | * suitable migration targets, so periodically check if we need | |
be976572 | 744 | * to schedule, or even abort async compaction. |
f6ea3adb | 745 | */ |
be976572 VB |
746 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
747 | && compact_should_abort(cc)) | |
748 | break; | |
f6ea3adb | 749 | |
c96b9e50 | 750 | if (!pfn_valid(block_start_pfn)) |
ff9543fd | 751 | continue; |
2fe86e00 | 752 | |
ff9543fd MN |
753 | /* |
754 | * Check for overlapping nodes/zones. It's possible on some | |
755 | * configurations to have a setup like | |
756 | * node0 node1 node0 | |
757 | * i.e. it's possible that all pages within a zones range of | |
758 | * pages do not belong to a single zone. | |
759 | */ | |
c96b9e50 | 760 | page = pfn_to_page(block_start_pfn); |
ff9543fd MN |
761 | if (page_zone(page) != zone) |
762 | continue; | |
763 | ||
764 | /* Check the block is suitable for migration */ | |
68e3e926 | 765 | if (!suitable_migration_target(page)) |
ff9543fd | 766 | continue; |
68e3e926 | 767 | |
bb13ffeb MG |
768 | /* If isolation recently failed, do not retry */ |
769 | if (!isolation_suitable(cc, page)) | |
770 | continue; | |
771 | ||
f40d1e42 | 772 | /* Found a block suitable for isolating free pages from */ |
e9ade569 | 773 | cc->free_pfn = block_start_pfn; |
c96b9e50 VB |
774 | isolated = isolate_freepages_block(cc, block_start_pfn, |
775 | block_end_pfn, freelist, false); | |
f40d1e42 | 776 | nr_freepages += isolated; |
ff9543fd MN |
777 | |
778 | /* | |
e9ade569 VB |
779 | * Set a flag that we successfully isolated in this pageblock. |
780 | * In the next loop iteration, zone->compact_cached_free_pfn | |
781 | * will not be updated and thus it will effectively contain the | |
782 | * highest pageblock we isolated pages from. | |
ff9543fd | 783 | */ |
e9ade569 | 784 | if (isolated) |
c89511ab | 785 | cc->finished_update_free = true; |
be976572 VB |
786 | |
787 | /* | |
788 | * isolate_freepages_block() might have aborted due to async | |
789 | * compaction being contended | |
790 | */ | |
791 | if (cc->contended) | |
792 | break; | |
ff9543fd MN |
793 | } |
794 | ||
795 | /* split_free_page does not map the pages */ | |
796 | map_pages(freelist); | |
797 | ||
7ed695e0 VB |
798 | /* |
799 | * If we crossed the migrate scanner, we want to keep it that way | |
800 | * so that compact_finished() may detect this | |
801 | */ | |
c96b9e50 | 802 | if (block_start_pfn < low_pfn) |
e9ade569 | 803 | cc->free_pfn = cc->migrate_pfn; |
c96b9e50 | 804 | |
ff9543fd | 805 | cc->nr_freepages = nr_freepages; |
748446bb MG |
806 | } |
807 | ||
808 | /* | |
809 | * This is a migrate-callback that "allocates" freepages by taking pages | |
810 | * from the isolated freelists in the block we are migrating to. | |
811 | */ | |
812 | static struct page *compaction_alloc(struct page *migratepage, | |
813 | unsigned long data, | |
814 | int **result) | |
815 | { | |
816 | struct compact_control *cc = (struct compact_control *)data; | |
817 | struct page *freepage; | |
818 | ||
be976572 VB |
819 | /* |
820 | * Isolate free pages if necessary, and if we are not aborting due to | |
821 | * contention. | |
822 | */ | |
748446bb | 823 | if (list_empty(&cc->freepages)) { |
be976572 VB |
824 | if (!cc->contended) |
825 | isolate_freepages(cc->zone, cc); | |
748446bb MG |
826 | |
827 | if (list_empty(&cc->freepages)) | |
828 | return NULL; | |
829 | } | |
830 | ||
831 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
832 | list_del(&freepage->lru); | |
833 | cc->nr_freepages--; | |
834 | ||
835 | return freepage; | |
836 | } | |
837 | ||
838 | /* | |
d53aea3d DR |
839 | * This is a migrate-callback that "frees" freepages back to the isolated |
840 | * freelist. All pages on the freelist are from the same zone, so there is no | |
841 | * special handling needed for NUMA. | |
842 | */ | |
843 | static void compaction_free(struct page *page, unsigned long data) | |
844 | { | |
845 | struct compact_control *cc = (struct compact_control *)data; | |
846 | ||
847 | list_add(&page->lru, &cc->freepages); | |
848 | cc->nr_freepages++; | |
849 | } | |
850 | ||
ff9543fd MN |
851 | /* possible outcome of isolate_migratepages */ |
852 | typedef enum { | |
853 | ISOLATE_ABORT, /* Abort compaction now */ | |
854 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
855 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
856 | } isolate_migrate_t; | |
857 | ||
858 | /* | |
859 | * Isolate all pages that can be migrated from the block pointed to by | |
860 | * the migrate scanner within compact_control. | |
861 | */ | |
862 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
863 | struct compact_control *cc) | |
864 | { | |
865 | unsigned long low_pfn, end_pfn; | |
866 | ||
867 | /* Do not scan outside zone boundaries */ | |
868 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
869 | ||
870 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 871 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd MN |
872 | |
873 | /* Do not cross the free scanner or scan within a memory hole */ | |
874 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
875 | cc->migrate_pfn = end_pfn; | |
876 | return ISOLATE_NONE; | |
877 | } | |
878 | ||
879 | /* Perform the isolation */ | |
e46a2879 | 880 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
e64c5237 | 881 | if (!low_pfn || cc->contended) |
ff9543fd MN |
882 | return ISOLATE_ABORT; |
883 | ||
884 | cc->migrate_pfn = low_pfn; | |
885 | ||
886 | return ISOLATE_SUCCESS; | |
887 | } | |
888 | ||
748446bb | 889 | static int compact_finished(struct zone *zone, |
5a03b051 | 890 | struct compact_control *cc) |
748446bb | 891 | { |
8fb74b9f | 892 | unsigned int order; |
5a03b051 | 893 | unsigned long watermark; |
56de7263 | 894 | |
be976572 | 895 | if (cc->contended || fatal_signal_pending(current)) |
748446bb MG |
896 | return COMPACT_PARTIAL; |
897 | ||
753341a4 | 898 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 899 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 | 900 | /* Let the next compaction start anew. */ |
35979ef3 DR |
901 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
902 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
55b7c4c9 VB |
903 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
904 | ||
62997027 MG |
905 | /* |
906 | * Mark that the PG_migrate_skip information should be cleared | |
907 | * by kswapd when it goes to sleep. kswapd does not set the | |
908 | * flag itself as the decision to be clear should be directly | |
909 | * based on an allocation request. | |
910 | */ | |
911 | if (!current_is_kswapd()) | |
912 | zone->compact_blockskip_flush = true; | |
913 | ||
748446bb | 914 | return COMPACT_COMPLETE; |
bb13ffeb | 915 | } |
748446bb | 916 | |
82478fb7 JW |
917 | /* |
918 | * order == -1 is expected when compacting via | |
919 | * /proc/sys/vm/compact_memory | |
920 | */ | |
56de7263 MG |
921 | if (cc->order == -1) |
922 | return COMPACT_CONTINUE; | |
923 | ||
3957c776 MH |
924 | /* Compaction run is not finished if the watermark is not met */ |
925 | watermark = low_wmark_pages(zone); | |
926 | watermark += (1 << cc->order); | |
927 | ||
928 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
929 | return COMPACT_CONTINUE; | |
930 | ||
56de7263 | 931 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
932 | for (order = cc->order; order < MAX_ORDER; order++) { |
933 | struct free_area *area = &zone->free_area[order]; | |
934 | ||
935 | /* Job done if page is free of the right migratetype */ | |
936 | if (!list_empty(&area->free_list[cc->migratetype])) | |
937 | return COMPACT_PARTIAL; | |
938 | ||
939 | /* Job done if allocation would set block type */ | |
940 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
941 | return COMPACT_PARTIAL; |
942 | } | |
943 | ||
748446bb MG |
944 | return COMPACT_CONTINUE; |
945 | } | |
946 | ||
3e7d3449 MG |
947 | /* |
948 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
949 | * Returns | |
950 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
951 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
952 | * COMPACT_CONTINUE - If compaction should run now | |
953 | */ | |
954 | unsigned long compaction_suitable(struct zone *zone, int order) | |
955 | { | |
956 | int fragindex; | |
957 | unsigned long watermark; | |
958 | ||
3957c776 MH |
959 | /* |
960 | * order == -1 is expected when compacting via | |
961 | * /proc/sys/vm/compact_memory | |
962 | */ | |
963 | if (order == -1) | |
964 | return COMPACT_CONTINUE; | |
965 | ||
3e7d3449 MG |
966 | /* |
967 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
968 | * This is because during migration, copies of pages need to be | |
969 | * allocated and for a short time, the footprint is higher | |
970 | */ | |
971 | watermark = low_wmark_pages(zone) + (2UL << order); | |
972 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
973 | return COMPACT_SKIPPED; | |
974 | ||
975 | /* | |
976 | * fragmentation index determines if allocation failures are due to | |
977 | * low memory or external fragmentation | |
978 | * | |
a582a738 SL |
979 | * index of -1000 implies allocations might succeed depending on |
980 | * watermarks | |
3e7d3449 MG |
981 | * index towards 0 implies failure is due to lack of memory |
982 | * index towards 1000 implies failure is due to fragmentation | |
983 | * | |
984 | * Only compact if a failure would be due to fragmentation. | |
985 | */ | |
986 | fragindex = fragmentation_index(zone, order); | |
987 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
988 | return COMPACT_SKIPPED; | |
989 | ||
a582a738 SL |
990 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
991 | 0, 0)) | |
3e7d3449 MG |
992 | return COMPACT_PARTIAL; |
993 | ||
994 | return COMPACT_CONTINUE; | |
995 | } | |
996 | ||
748446bb MG |
997 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
998 | { | |
999 | int ret; | |
c89511ab | 1000 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 1001 | unsigned long end_pfn = zone_end_pfn(zone); |
e0b9daeb | 1002 | const bool sync = cc->mode != MIGRATE_ASYNC; |
748446bb | 1003 | |
3e7d3449 MG |
1004 | ret = compaction_suitable(zone, cc->order); |
1005 | switch (ret) { | |
1006 | case COMPACT_PARTIAL: | |
1007 | case COMPACT_SKIPPED: | |
1008 | /* Compaction is likely to fail */ | |
1009 | return ret; | |
1010 | case COMPACT_CONTINUE: | |
1011 | /* Fall through to compaction */ | |
1012 | ; | |
1013 | } | |
1014 | ||
d3132e4b VB |
1015 | /* |
1016 | * Clear pageblock skip if there were failures recently and compaction | |
1017 | * is about to be retried after being deferred. kswapd does not do | |
1018 | * this reset as it'll reset the cached information when going to sleep. | |
1019 | */ | |
1020 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1021 | __reset_isolation_suitable(zone); | |
1022 | ||
c89511ab MG |
1023 | /* |
1024 | * Setup to move all movable pages to the end of the zone. Used cached | |
1025 | * information on where the scanners should start but check that it | |
1026 | * is initialised by ensuring the values are within zone boundaries. | |
1027 | */ | |
e0b9daeb | 1028 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
c89511ab MG |
1029 | cc->free_pfn = zone->compact_cached_free_pfn; |
1030 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
1031 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
1032 | zone->compact_cached_free_pfn = cc->free_pfn; | |
1033 | } | |
1034 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1035 | cc->migrate_pfn = start_pfn; | |
35979ef3 DR |
1036 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
1037 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
c89511ab | 1038 | } |
748446bb | 1039 | |
0eb927c0 MG |
1040 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn); |
1041 | ||
748446bb MG |
1042 | migrate_prep_local(); |
1043 | ||
1044 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
9d502c1c | 1045 | int err; |
748446bb | 1046 | |
f9e35b3b MG |
1047 | switch (isolate_migratepages(zone, cc)) { |
1048 | case ISOLATE_ABORT: | |
1049 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1050 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1051 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1052 | goto out; |
1053 | case ISOLATE_NONE: | |
748446bb | 1054 | continue; |
f9e35b3b MG |
1055 | case ISOLATE_SUCCESS: |
1056 | ; | |
1057 | } | |
748446bb | 1058 | |
f8c9301f VB |
1059 | if (!cc->nr_migratepages) |
1060 | continue; | |
1061 | ||
d53aea3d | 1062 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 1063 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 1064 | MR_COMPACTION); |
748446bb | 1065 | |
f8c9301f VB |
1066 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
1067 | &cc->migratepages); | |
748446bb | 1068 | |
f8c9301f VB |
1069 | /* All pages were either migrated or will be released */ |
1070 | cc->nr_migratepages = 0; | |
9d502c1c | 1071 | if (err) { |
5733c7d1 | 1072 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
1073 | /* |
1074 | * migrate_pages() may return -ENOMEM when scanners meet | |
1075 | * and we want compact_finished() to detect it | |
1076 | */ | |
1077 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1078 | ret = COMPACT_PARTIAL; |
1079 | goto out; | |
1080 | } | |
748446bb | 1081 | } |
748446bb MG |
1082 | } |
1083 | ||
f9e35b3b | 1084 | out: |
748446bb MG |
1085 | /* Release free pages and check accounting */ |
1086 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1087 | VM_BUG_ON(cc->nr_freepages != 0); | |
1088 | ||
0eb927c0 MG |
1089 | trace_mm_compaction_end(ret); |
1090 | ||
748446bb MG |
1091 | return ret; |
1092 | } | |
76ab0f53 | 1093 | |
e0b9daeb DR |
1094 | static unsigned long compact_zone_order(struct zone *zone, int order, |
1095 | gfp_t gfp_mask, enum migrate_mode mode, bool *contended) | |
56de7263 | 1096 | { |
e64c5237 | 1097 | unsigned long ret; |
56de7263 MG |
1098 | struct compact_control cc = { |
1099 | .nr_freepages = 0, | |
1100 | .nr_migratepages = 0, | |
1101 | .order = order, | |
1102 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1103 | .zone = zone, | |
e0b9daeb | 1104 | .mode = mode, |
56de7263 MG |
1105 | }; |
1106 | INIT_LIST_HEAD(&cc.freepages); | |
1107 | INIT_LIST_HEAD(&cc.migratepages); | |
1108 | ||
e64c5237 SL |
1109 | ret = compact_zone(zone, &cc); |
1110 | ||
1111 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1112 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1113 | ||
1114 | *contended = cc.contended; | |
1115 | return ret; | |
56de7263 MG |
1116 | } |
1117 | ||
5e771905 MG |
1118 | int sysctl_extfrag_threshold = 500; |
1119 | ||
56de7263 MG |
1120 | /** |
1121 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1122 | * @zonelist: The zonelist used for the current allocation | |
1123 | * @order: The order of the current allocation | |
1124 | * @gfp_mask: The GFP mask of the current allocation | |
1125 | * @nodemask: The allowed nodes to allocate from | |
e0b9daeb | 1126 | * @mode: The migration mode for async, sync light, or sync migration |
661c4cb9 MG |
1127 | * @contended: Return value that is true if compaction was aborted due to lock contention |
1128 | * @page: Optionally capture a free page of the requested order during compaction | |
56de7263 MG |
1129 | * |
1130 | * This is the main entry point for direct page compaction. | |
1131 | */ | |
1132 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1133 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
e0b9daeb | 1134 | enum migrate_mode mode, bool *contended) |
56de7263 MG |
1135 | { |
1136 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1137 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1138 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1139 | struct zoneref *z; |
1140 | struct zone *zone; | |
1141 | int rc = COMPACT_SKIPPED; | |
d95ea5d1 | 1142 | int alloc_flags = 0; |
56de7263 | 1143 | |
4ffb6335 | 1144 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1145 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
1146 | return rc; |
1147 | ||
010fc29a | 1148 | count_compact_event(COMPACTSTALL); |
56de7263 | 1149 | |
d95ea5d1 BZ |
1150 | #ifdef CONFIG_CMA |
1151 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1152 | alloc_flags |= ALLOC_CMA; | |
1153 | #endif | |
56de7263 MG |
1154 | /* Compact each zone in the list */ |
1155 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1156 | nodemask) { | |
56de7263 MG |
1157 | int status; |
1158 | ||
e0b9daeb | 1159 | status = compact_zone_order(zone, order, gfp_mask, mode, |
8fb74b9f | 1160 | contended); |
56de7263 MG |
1161 | rc = max(status, rc); |
1162 | ||
3e7d3449 | 1163 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 BZ |
1164 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
1165 | alloc_flags)) | |
56de7263 MG |
1166 | break; |
1167 | } | |
1168 | ||
1169 | return rc; | |
1170 | } | |
1171 | ||
1172 | ||
76ab0f53 | 1173 | /* Compact all zones within a node */ |
7103f16d | 1174 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1175 | { |
1176 | int zoneid; | |
76ab0f53 MG |
1177 | struct zone *zone; |
1178 | ||
76ab0f53 | 1179 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1180 | |
1181 | zone = &pgdat->node_zones[zoneid]; | |
1182 | if (!populated_zone(zone)) | |
1183 | continue; | |
1184 | ||
7be62de9 RR |
1185 | cc->nr_freepages = 0; |
1186 | cc->nr_migratepages = 0; | |
1187 | cc->zone = zone; | |
1188 | INIT_LIST_HEAD(&cc->freepages); | |
1189 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1190 | |
aad6ec37 | 1191 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1192 | compact_zone(zone, cc); |
76ab0f53 | 1193 | |
aff62249 | 1194 | if (cc->order > 0) { |
de6c60a6 VB |
1195 | if (zone_watermark_ok(zone, cc->order, |
1196 | low_wmark_pages(zone), 0, 0)) | |
1197 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 RR |
1198 | } |
1199 | ||
7be62de9 RR |
1200 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1201 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1202 | } |
76ab0f53 MG |
1203 | } |
1204 | ||
7103f16d | 1205 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1206 | { |
1207 | struct compact_control cc = { | |
1208 | .order = order, | |
e0b9daeb | 1209 | .mode = MIGRATE_ASYNC, |
7be62de9 RR |
1210 | }; |
1211 | ||
3a7200af MG |
1212 | if (!order) |
1213 | return; | |
1214 | ||
7103f16d | 1215 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1216 | } |
1217 | ||
7103f16d | 1218 | static void compact_node(int nid) |
7be62de9 | 1219 | { |
7be62de9 RR |
1220 | struct compact_control cc = { |
1221 | .order = -1, | |
e0b9daeb | 1222 | .mode = MIGRATE_SYNC, |
91ca9186 | 1223 | .ignore_skip_hint = true, |
7be62de9 RR |
1224 | }; |
1225 | ||
7103f16d | 1226 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1227 | } |
1228 | ||
76ab0f53 | 1229 | /* Compact all nodes in the system */ |
7964c06d | 1230 | static void compact_nodes(void) |
76ab0f53 MG |
1231 | { |
1232 | int nid; | |
1233 | ||
8575ec29 HD |
1234 | /* Flush pending updates to the LRU lists */ |
1235 | lru_add_drain_all(); | |
1236 | ||
76ab0f53 MG |
1237 | for_each_online_node(nid) |
1238 | compact_node(nid); | |
76ab0f53 MG |
1239 | } |
1240 | ||
1241 | /* The written value is actually unused, all memory is compacted */ | |
1242 | int sysctl_compact_memory; | |
1243 | ||
1244 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1245 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1246 | void __user *buffer, size_t *length, loff_t *ppos) | |
1247 | { | |
1248 | if (write) | |
7964c06d | 1249 | compact_nodes(); |
76ab0f53 MG |
1250 | |
1251 | return 0; | |
1252 | } | |
ed4a6d7f | 1253 | |
5e771905 MG |
1254 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1255 | void __user *buffer, size_t *length, loff_t *ppos) | |
1256 | { | |
1257 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1258 | ||
1259 | return 0; | |
1260 | } | |
1261 | ||
ed4a6d7f | 1262 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1263 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1264 | struct device_attribute *attr, |
ed4a6d7f MG |
1265 | const char *buf, size_t count) |
1266 | { | |
8575ec29 HD |
1267 | int nid = dev->id; |
1268 | ||
1269 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1270 | /* Flush pending updates to the LRU lists */ | |
1271 | lru_add_drain_all(); | |
1272 | ||
1273 | compact_node(nid); | |
1274 | } | |
ed4a6d7f MG |
1275 | |
1276 | return count; | |
1277 | } | |
10fbcf4c | 1278 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1279 | |
1280 | int compaction_register_node(struct node *node) | |
1281 | { | |
10fbcf4c | 1282 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1283 | } |
1284 | ||
1285 | void compaction_unregister_node(struct node *node) | |
1286 | { | |
10fbcf4c | 1287 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1288 | } |
1289 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1290 | |
1291 | #endif /* CONFIG_COMPACTION */ |