]>
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 | ||
7d49d886 VB |
70 | /* |
71 | * Check that the whole (or subset of) a pageblock given by the interval of | |
72 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
73 | * with the migration of free compaction scanner. The scanners then need to | |
74 | * use only pfn_valid_within() check for arches that allow holes within | |
75 | * pageblocks. | |
76 | * | |
77 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
78 | * | |
79 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
80 | * i.e. it's possible that all pages within a zones range of pages do not | |
81 | * belong to a single zone. We assume that a border between node0 and node1 | |
82 | * can occur within a single pageblock, but not a node0 node1 node0 | |
83 | * interleaving within a single pageblock. It is therefore sufficient to check | |
84 | * the first and last page of a pageblock and avoid checking each individual | |
85 | * page in a pageblock. | |
86 | */ | |
87 | static struct page *pageblock_pfn_to_page(unsigned long start_pfn, | |
88 | unsigned long end_pfn, struct zone *zone) | |
89 | { | |
90 | struct page *start_page; | |
91 | struct page *end_page; | |
92 | ||
93 | /* end_pfn is one past the range we are checking */ | |
94 | end_pfn--; | |
95 | ||
96 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
97 | return NULL; | |
98 | ||
99 | start_page = pfn_to_page(start_pfn); | |
100 | ||
101 | if (page_zone(start_page) != zone) | |
102 | return NULL; | |
103 | ||
104 | end_page = pfn_to_page(end_pfn); | |
105 | ||
106 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
107 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
108 | return NULL; | |
109 | ||
110 | return start_page; | |
111 | } | |
112 | ||
bb13ffeb MG |
113 | #ifdef CONFIG_COMPACTION |
114 | /* Returns true if the pageblock should be scanned for pages to isolate. */ | |
115 | static inline bool isolation_suitable(struct compact_control *cc, | |
116 | struct page *page) | |
117 | { | |
118 | if (cc->ignore_skip_hint) | |
119 | return true; | |
120 | ||
121 | return !get_pageblock_skip(page); | |
122 | } | |
123 | ||
124 | /* | |
125 | * This function is called to clear all cached information on pageblocks that | |
126 | * should be skipped for page isolation when the migrate and free page scanner | |
127 | * meet. | |
128 | */ | |
62997027 | 129 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
130 | { |
131 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 132 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
133 | unsigned long pfn; |
134 | ||
35979ef3 DR |
135 | zone->compact_cached_migrate_pfn[0] = start_pfn; |
136 | zone->compact_cached_migrate_pfn[1] = start_pfn; | |
c89511ab | 137 | zone->compact_cached_free_pfn = end_pfn; |
62997027 | 138 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
139 | |
140 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
141 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
142 | struct page *page; | |
143 | ||
144 | cond_resched(); | |
145 | ||
146 | if (!pfn_valid(pfn)) | |
147 | continue; | |
148 | ||
149 | page = pfn_to_page(pfn); | |
150 | if (zone != page_zone(page)) | |
151 | continue; | |
152 | ||
153 | clear_pageblock_skip(page); | |
154 | } | |
155 | } | |
156 | ||
62997027 MG |
157 | void reset_isolation_suitable(pg_data_t *pgdat) |
158 | { | |
159 | int zoneid; | |
160 | ||
161 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
162 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
163 | if (!populated_zone(zone)) | |
164 | continue; | |
165 | ||
166 | /* Only flush if a full compaction finished recently */ | |
167 | if (zone->compact_blockskip_flush) | |
168 | __reset_isolation_suitable(zone); | |
169 | } | |
170 | } | |
171 | ||
bb13ffeb MG |
172 | /* |
173 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 174 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 175 | */ |
c89511ab MG |
176 | static void update_pageblock_skip(struct compact_control *cc, |
177 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 178 | bool migrate_scanner) |
bb13ffeb | 179 | { |
c89511ab | 180 | struct zone *zone = cc->zone; |
35979ef3 | 181 | unsigned long pfn; |
6815bf3f JK |
182 | |
183 | if (cc->ignore_skip_hint) | |
184 | return; | |
185 | ||
bb13ffeb MG |
186 | if (!page) |
187 | return; | |
188 | ||
35979ef3 DR |
189 | if (nr_isolated) |
190 | return; | |
191 | ||
edc2ca61 | 192 | set_pageblock_skip(page); |
c89511ab | 193 | |
35979ef3 DR |
194 | pfn = page_to_pfn(page); |
195 | ||
196 | /* Update where async and sync compaction should restart */ | |
197 | if (migrate_scanner) { | |
198 | if (cc->finished_update_migrate) | |
199 | return; | |
200 | if (pfn > zone->compact_cached_migrate_pfn[0]) | |
201 | zone->compact_cached_migrate_pfn[0] = pfn; | |
e0b9daeb DR |
202 | if (cc->mode != MIGRATE_ASYNC && |
203 | pfn > zone->compact_cached_migrate_pfn[1]) | |
35979ef3 DR |
204 | zone->compact_cached_migrate_pfn[1] = pfn; |
205 | } else { | |
206 | if (cc->finished_update_free) | |
207 | return; | |
208 | if (pfn < zone->compact_cached_free_pfn) | |
209 | zone->compact_cached_free_pfn = pfn; | |
c89511ab | 210 | } |
bb13ffeb MG |
211 | } |
212 | #else | |
213 | static inline bool isolation_suitable(struct compact_control *cc, | |
214 | struct page *page) | |
215 | { | |
216 | return true; | |
217 | } | |
218 | ||
c89511ab MG |
219 | static void update_pageblock_skip(struct compact_control *cc, |
220 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 221 | bool migrate_scanner) |
bb13ffeb MG |
222 | { |
223 | } | |
224 | #endif /* CONFIG_COMPACTION */ | |
225 | ||
2a1402aa MG |
226 | static inline bool should_release_lock(spinlock_t *lock) |
227 | { | |
228 | return need_resched() || spin_is_contended(lock); | |
229 | } | |
230 | ||
c67fe375 MG |
231 | /* |
232 | * Compaction requires the taking of some coarse locks that are potentially | |
233 | * very heavily contended. Check if the process needs to be scheduled or | |
234 | * if the lock is contended. For async compaction, back out in the event | |
235 | * if contention is severe. For sync compaction, schedule. | |
236 | * | |
237 | * Returns true if the lock is held. | |
238 | * Returns false if the lock is released and compaction should abort | |
239 | */ | |
240 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, | |
241 | bool locked, struct compact_control *cc) | |
242 | { | |
2a1402aa | 243 | if (should_release_lock(lock)) { |
c67fe375 MG |
244 | if (locked) { |
245 | spin_unlock_irqrestore(lock, *flags); | |
246 | locked = false; | |
247 | } | |
248 | ||
249 | /* async aborts if taking too long or contended */ | |
e0b9daeb | 250 | if (cc->mode == MIGRATE_ASYNC) { |
e64c5237 | 251 | cc->contended = true; |
c67fe375 MG |
252 | return false; |
253 | } | |
254 | ||
255 | cond_resched(); | |
c67fe375 MG |
256 | } |
257 | ||
258 | if (!locked) | |
259 | spin_lock_irqsave(lock, *flags); | |
260 | return true; | |
261 | } | |
262 | ||
be976572 VB |
263 | /* |
264 | * Aside from avoiding lock contention, compaction also periodically checks | |
265 | * need_resched() and either schedules in sync compaction or aborts async | |
266 | * compaction. This is similar to what compact_checklock_irqsave() does, but | |
267 | * is used where no lock is concerned. | |
268 | * | |
269 | * Returns false when no scheduling was needed, or sync compaction scheduled. | |
270 | * Returns true when async compaction should abort. | |
271 | */ | |
272 | static inline bool compact_should_abort(struct compact_control *cc) | |
273 | { | |
274 | /* async compaction aborts if contended */ | |
275 | if (need_resched()) { | |
276 | if (cc->mode == MIGRATE_ASYNC) { | |
277 | cc->contended = true; | |
278 | return true; | |
279 | } | |
280 | ||
281 | cond_resched(); | |
282 | } | |
283 | ||
284 | return false; | |
285 | } | |
286 | ||
f40d1e42 MG |
287 | /* Returns true if the page is within a block suitable for migration to */ |
288 | static bool suitable_migration_target(struct page *page) | |
289 | { | |
7d348b9e | 290 | /* If the page is a large free page, then disallow migration */ |
f40d1e42 | 291 | if (PageBuddy(page) && page_order(page) >= pageblock_order) |
7d348b9e | 292 | return false; |
f40d1e42 MG |
293 | |
294 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
7d348b9e | 295 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
f40d1e42 MG |
296 | return true; |
297 | ||
298 | /* Otherwise skip the block */ | |
299 | return false; | |
300 | } | |
301 | ||
85aa125f | 302 | /* |
9e4be470 JM |
303 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
304 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
305 | * (even though it may still end up isolating some pages). | |
85aa125f | 306 | */ |
f40d1e42 MG |
307 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
308 | unsigned long blockpfn, | |
85aa125f MN |
309 | unsigned long end_pfn, |
310 | struct list_head *freelist, | |
311 | bool strict) | |
748446bb | 312 | { |
b7aba698 | 313 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 314 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
315 | unsigned long flags; |
316 | bool locked = false; | |
748446bb | 317 | |
748446bb MG |
318 | cursor = pfn_to_page(blockpfn); |
319 | ||
f40d1e42 | 320 | /* Isolate free pages. */ |
748446bb MG |
321 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
322 | int isolated, i; | |
323 | struct page *page = cursor; | |
324 | ||
b7aba698 | 325 | nr_scanned++; |
f40d1e42 | 326 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
327 | goto isolate_fail; |
328 | ||
bb13ffeb MG |
329 | if (!valid_page) |
330 | valid_page = page; | |
f40d1e42 | 331 | if (!PageBuddy(page)) |
2af120bc | 332 | goto isolate_fail; |
f40d1e42 MG |
333 | |
334 | /* | |
335 | * The zone lock must be held to isolate freepages. | |
336 | * Unfortunately this is a very coarse lock and can be | |
337 | * heavily contended if there are parallel allocations | |
338 | * or parallel compactions. For async compaction do not | |
339 | * spin on the lock and we acquire the lock as late as | |
340 | * possible. | |
341 | */ | |
342 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
343 | locked, cc); | |
344 | if (!locked) | |
345 | break; | |
346 | ||
f40d1e42 MG |
347 | /* Recheck this is a buddy page under lock */ |
348 | if (!PageBuddy(page)) | |
2af120bc | 349 | goto isolate_fail; |
748446bb MG |
350 | |
351 | /* Found a free page, break it into order-0 pages */ | |
352 | isolated = split_free_page(page); | |
353 | total_isolated += isolated; | |
354 | for (i = 0; i < isolated; i++) { | |
355 | list_add(&page->lru, freelist); | |
356 | page++; | |
357 | } | |
358 | ||
359 | /* If a page was split, advance to the end of it */ | |
360 | if (isolated) { | |
361 | blockpfn += isolated - 1; | |
362 | cursor += isolated - 1; | |
2af120bc | 363 | continue; |
748446bb | 364 | } |
2af120bc LA |
365 | |
366 | isolate_fail: | |
367 | if (strict) | |
368 | break; | |
369 | else | |
370 | continue; | |
371 | ||
748446bb MG |
372 | } |
373 | ||
b7aba698 | 374 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
375 | |
376 | /* | |
377 | * If strict isolation is requested by CMA then check that all the | |
378 | * pages requested were isolated. If there were any failures, 0 is | |
379 | * returned and CMA will fail. | |
380 | */ | |
2af120bc | 381 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
382 | total_isolated = 0; |
383 | ||
384 | if (locked) | |
385 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
386 | ||
bb13ffeb MG |
387 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
388 | if (blockpfn == end_pfn) | |
edc2ca61 | 389 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 390 | |
010fc29a | 391 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 392 | if (total_isolated) |
010fc29a | 393 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
394 | return total_isolated; |
395 | } | |
396 | ||
85aa125f MN |
397 | /** |
398 | * isolate_freepages_range() - isolate free pages. | |
399 | * @start_pfn: The first PFN to start isolating. | |
400 | * @end_pfn: The one-past-last PFN. | |
401 | * | |
402 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
403 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
404 | * undo its actions and return zero. | |
405 | * | |
406 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
407 | * (which may be greater then end_pfn if end fell in a middle of | |
408 | * a free page). | |
409 | */ | |
ff9543fd | 410 | unsigned long |
bb13ffeb MG |
411 | isolate_freepages_range(struct compact_control *cc, |
412 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 413 | { |
f40d1e42 | 414 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
415 | LIST_HEAD(freelist); |
416 | ||
7d49d886 VB |
417 | pfn = start_pfn; |
418 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
419 | ||
420 | for (; pfn < end_pfn; pfn += isolated, | |
421 | block_end_pfn += pageblock_nr_pages) { | |
85aa125f | 422 | |
85aa125f MN |
423 | block_end_pfn = min(block_end_pfn, end_pfn); |
424 | ||
7d49d886 VB |
425 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
426 | break; | |
427 | ||
bb13ffeb | 428 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 429 | &freelist, true); |
85aa125f MN |
430 | |
431 | /* | |
432 | * In strict mode, isolate_freepages_block() returns 0 if | |
433 | * there are any holes in the block (ie. invalid PFNs or | |
434 | * non-free pages). | |
435 | */ | |
436 | if (!isolated) | |
437 | break; | |
438 | ||
439 | /* | |
440 | * If we managed to isolate pages, it is always (1 << n) * | |
441 | * pageblock_nr_pages for some non-negative n. (Max order | |
442 | * page may span two pageblocks). | |
443 | */ | |
444 | } | |
445 | ||
446 | /* split_free_page does not map the pages */ | |
447 | map_pages(&freelist); | |
448 | ||
449 | if (pfn < end_pfn) { | |
450 | /* Loop terminated early, cleanup. */ | |
451 | release_freepages(&freelist); | |
452 | return 0; | |
453 | } | |
454 | ||
455 | /* We don't use freelists for anything. */ | |
456 | return pfn; | |
457 | } | |
458 | ||
748446bb | 459 | /* Update the number of anon and file isolated pages in the zone */ |
edc2ca61 | 460 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
748446bb MG |
461 | { |
462 | struct page *page; | |
b9e84ac1 | 463 | unsigned int count[2] = { 0, }; |
748446bb | 464 | |
edc2ca61 VB |
465 | if (list_empty(&cc->migratepages)) |
466 | return; | |
467 | ||
b9e84ac1 MK |
468 | list_for_each_entry(page, &cc->migratepages, lru) |
469 | count[!!page_is_file_cache(page)]++; | |
748446bb | 470 | |
edc2ca61 VB |
471 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
472 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
748446bb MG |
473 | } |
474 | ||
475 | /* Similar to reclaim, but different enough that they don't share logic */ | |
476 | static bool too_many_isolated(struct zone *zone) | |
477 | { | |
bc693045 | 478 | unsigned long active, inactive, isolated; |
748446bb MG |
479 | |
480 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
481 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
482 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
483 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
484 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
485 | zone_page_state(zone, NR_ISOLATED_ANON); | |
486 | ||
bc693045 | 487 | return isolated > (inactive + active) / 2; |
748446bb MG |
488 | } |
489 | ||
2fe86e00 | 490 | /** |
edc2ca61 VB |
491 | * isolate_migratepages_block() - isolate all migrate-able pages within |
492 | * a single pageblock | |
2fe86e00 | 493 | * @cc: Compaction control structure. |
edc2ca61 VB |
494 | * @low_pfn: The first PFN to isolate |
495 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock | |
496 | * @isolate_mode: Isolation mode to be used. | |
2fe86e00 MN |
497 | * |
498 | * Isolate all pages that can be migrated from the range specified by | |
edc2ca61 VB |
499 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
500 | * Returns zero if there is a fatal signal pending, otherwise PFN of the | |
501 | * first page that was not scanned (which may be both less, equal to or more | |
502 | * than end_pfn). | |
2fe86e00 | 503 | * |
edc2ca61 VB |
504 | * The pages are isolated on cc->migratepages list (not required to be empty), |
505 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field | |
506 | * is neither read nor updated. | |
748446bb | 507 | */ |
edc2ca61 VB |
508 | static unsigned long |
509 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, | |
510 | unsigned long end_pfn, isolate_mode_t isolate_mode) | |
748446bb | 511 | { |
edc2ca61 | 512 | struct zone *zone = cc->zone; |
b7aba698 | 513 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 514 | struct list_head *migratelist = &cc->migratepages; |
fa9add64 | 515 | struct lruvec *lruvec; |
c67fe375 | 516 | unsigned long flags; |
2a1402aa | 517 | bool locked = false; |
bb13ffeb | 518 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 519 | |
748446bb MG |
520 | /* |
521 | * Ensure that there are not too many pages isolated from the LRU | |
522 | * list by either parallel reclaimers or compaction. If there are, | |
523 | * delay for some time until fewer pages are isolated | |
524 | */ | |
525 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 526 | /* async migration should just abort */ |
e0b9daeb | 527 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 528 | return 0; |
f9e35b3b | 529 | |
748446bb MG |
530 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
531 | ||
532 | if (fatal_signal_pending(current)) | |
2fe86e00 | 533 | return 0; |
748446bb MG |
534 | } |
535 | ||
be976572 VB |
536 | if (compact_should_abort(cc)) |
537 | return 0; | |
aeef4b83 | 538 | |
748446bb | 539 | /* Time to isolate some pages for migration */ |
748446bb | 540 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 541 | /* give a chance to irqs before checking need_resched() */ |
be1aa03b | 542 | if (locked && !(low_pfn % SWAP_CLUSTER_MAX)) { |
2a1402aa MG |
543 | if (should_release_lock(&zone->lru_lock)) { |
544 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
545 | locked = false; | |
546 | } | |
b2eef8c0 | 547 | } |
c67fe375 | 548 | |
748446bb MG |
549 | if (!pfn_valid_within(low_pfn)) |
550 | continue; | |
b7aba698 | 551 | nr_scanned++; |
748446bb | 552 | |
748446bb | 553 | page = pfn_to_page(low_pfn); |
dc908600 | 554 | |
bb13ffeb MG |
555 | if (!valid_page) |
556 | valid_page = page; | |
557 | ||
6c14466c MG |
558 | /* |
559 | * Skip if free. page_order cannot be used without zone->lock | |
560 | * as nothing prevents parallel allocations or buddy merging. | |
561 | */ | |
748446bb MG |
562 | if (PageBuddy(page)) |
563 | continue; | |
564 | ||
bf6bddf1 RA |
565 | /* |
566 | * Check may be lockless but that's ok as we recheck later. | |
567 | * It's possible to migrate LRU pages and balloon pages | |
568 | * Skip any other type of page | |
569 | */ | |
570 | if (!PageLRU(page)) { | |
571 | if (unlikely(balloon_page_movable(page))) { | |
572 | if (locked && balloon_page_isolate(page)) { | |
573 | /* Successfully isolated */ | |
b6c75016 | 574 | goto isolate_success; |
bf6bddf1 RA |
575 | } |
576 | } | |
bc835011 | 577 | continue; |
bf6bddf1 | 578 | } |
bc835011 AA |
579 | |
580 | /* | |
2a1402aa MG |
581 | * PageLRU is set. lru_lock normally excludes isolation |
582 | * splitting and collapsing (collapsing has already happened | |
583 | * if PageLRU is set) but the lock is not necessarily taken | |
584 | * here and it is wasteful to take it just to check transhuge. | |
585 | * Check TransHuge without lock and skip the whole pageblock if | |
586 | * it's either a transhuge or hugetlbfs page, as calling | |
587 | * compound_order() without preventing THP from splitting the | |
588 | * page underneath us may return surprising results. | |
bc835011 | 589 | */ |
2a1402aa MG |
590 | if (PageTransHuge(page)) { |
591 | if (!locked) | |
edc2ca61 VB |
592 | low_pfn = ALIGN(low_pfn + 1, |
593 | pageblock_nr_pages) - 1; | |
594 | else | |
595 | low_pfn += (1 << compound_order(page)) - 1; | |
596 | ||
2a1402aa MG |
597 | continue; |
598 | } | |
599 | ||
119d6d59 DR |
600 | /* |
601 | * Migration will fail if an anonymous page is pinned in memory, | |
602 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
603 | * admittedly racy check. | |
604 | */ | |
605 | if (!page_mapping(page) && | |
606 | page_count(page) > page_mapcount(page)) | |
607 | continue; | |
608 | ||
2a1402aa MG |
609 | /* Check if it is ok to still hold the lock */ |
610 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
611 | locked, cc); | |
612 | if (!locked || fatal_signal_pending(current)) | |
613 | break; | |
614 | ||
615 | /* Recheck PageLRU and PageTransHuge under lock */ | |
616 | if (!PageLRU(page)) | |
617 | continue; | |
bc835011 AA |
618 | if (PageTransHuge(page)) { |
619 | low_pfn += (1 << compound_order(page)) - 1; | |
620 | continue; | |
621 | } | |
622 | ||
fa9add64 HD |
623 | lruvec = mem_cgroup_page_lruvec(page, zone); |
624 | ||
748446bb | 625 | /* Try isolate the page */ |
edc2ca61 | 626 | if (__isolate_lru_page(page, isolate_mode) != 0) |
748446bb MG |
627 | continue; |
628 | ||
309381fe | 629 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 630 | |
748446bb | 631 | /* Successfully isolated */ |
fa9add64 | 632 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
633 | |
634 | isolate_success: | |
635 | cc->finished_update_migrate = true; | |
748446bb | 636 | list_add(&page->lru, migratelist); |
748446bb | 637 | cc->nr_migratepages++; |
b7aba698 | 638 | nr_isolated++; |
748446bb MG |
639 | |
640 | /* Avoid isolating too much */ | |
31b8384a HD |
641 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
642 | ++low_pfn; | |
748446bb | 643 | break; |
31b8384a | 644 | } |
748446bb MG |
645 | } |
646 | ||
c67fe375 MG |
647 | if (locked) |
648 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 649 | |
50b5b094 VB |
650 | /* |
651 | * Update the pageblock-skip information and cached scanner pfn, | |
652 | * if the whole pageblock was scanned without isolating any page. | |
50b5b094 | 653 | */ |
35979ef3 | 654 | if (low_pfn == end_pfn) |
edc2ca61 | 655 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 656 | |
b7aba698 MG |
657 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
658 | ||
010fc29a | 659 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 660 | if (nr_isolated) |
010fc29a | 661 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 662 | |
2fe86e00 MN |
663 | return low_pfn; |
664 | } | |
665 | ||
edc2ca61 VB |
666 | /** |
667 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range | |
668 | * @cc: Compaction control structure. | |
669 | * @start_pfn: The first PFN to start isolating. | |
670 | * @end_pfn: The one-past-last PFN. | |
671 | * | |
672 | * Returns zero if isolation fails fatally due to e.g. pending signal. | |
673 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
674 | * (which may be greater than end_pfn if end fell in a middle of a THP page). | |
675 | */ | |
676 | unsigned long | |
677 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, | |
678 | unsigned long end_pfn) | |
679 | { | |
680 | unsigned long pfn, block_end_pfn; | |
681 | ||
682 | /* Scan block by block. First and last block may be incomplete */ | |
683 | pfn = start_pfn; | |
684 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
685 | ||
686 | for (; pfn < end_pfn; pfn = block_end_pfn, | |
687 | block_end_pfn += pageblock_nr_pages) { | |
688 | ||
689 | block_end_pfn = min(block_end_pfn, end_pfn); | |
690 | ||
7d49d886 | 691 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
edc2ca61 VB |
692 | continue; |
693 | ||
694 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, | |
695 | ISOLATE_UNEVICTABLE); | |
696 | ||
697 | /* | |
698 | * In case of fatal failure, release everything that might | |
699 | * have been isolated in the previous iteration, and signal | |
700 | * the failure back to caller. | |
701 | */ | |
702 | if (!pfn) { | |
703 | putback_movable_pages(&cc->migratepages); | |
704 | cc->nr_migratepages = 0; | |
705 | break; | |
706 | } | |
707 | } | |
708 | acct_isolated(cc->zone, cc); | |
709 | ||
710 | return pfn; | |
711 | } | |
712 | ||
ff9543fd MN |
713 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
714 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 715 | /* |
ff9543fd MN |
716 | * Based on information in the current compact_control, find blocks |
717 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 718 | */ |
edc2ca61 | 719 | static void isolate_freepages(struct compact_control *cc) |
2fe86e00 | 720 | { |
edc2ca61 | 721 | struct zone *zone = cc->zone; |
ff9543fd | 722 | struct page *page; |
c96b9e50 VB |
723 | unsigned long block_start_pfn; /* start of current pageblock */ |
724 | unsigned long block_end_pfn; /* end of current pageblock */ | |
725 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd MN |
726 | int nr_freepages = cc->nr_freepages; |
727 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 728 | |
ff9543fd MN |
729 | /* |
730 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 VB |
731 | * successfully isolated from, zone-cached value, or the end of the |
732 | * zone when isolating for the first time. We need this aligned to | |
c96b9e50 VB |
733 | * the pageblock boundary, because we do |
734 | * block_start_pfn -= pageblock_nr_pages in the for loop. | |
735 | * For ending point, take care when isolating in last pageblock of a | |
736 | * a zone which ends in the middle of a pageblock. | |
49e068f0 VB |
737 | * The low boundary is the end of the pageblock the migration scanner |
738 | * is using. | |
ff9543fd | 739 | */ |
c96b9e50 VB |
740 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
741 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, | |
742 | zone_end_pfn(zone)); | |
7ed695e0 | 743 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 744 | |
ff9543fd MN |
745 | /* |
746 | * Isolate free pages until enough are available to migrate the | |
747 | * pages on cc->migratepages. We stop searching if the migrate | |
748 | * and free page scanners meet or enough free pages are isolated. | |
749 | */ | |
c96b9e50 VB |
750 | for (; block_start_pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
751 | block_end_pfn = block_start_pfn, | |
752 | block_start_pfn -= pageblock_nr_pages) { | |
ff9543fd | 753 | unsigned long isolated; |
2fe86e00 | 754 | |
f6ea3adb DR |
755 | /* |
756 | * This can iterate a massively long zone without finding any | |
757 | * suitable migration targets, so periodically check if we need | |
be976572 | 758 | * to schedule, or even abort async compaction. |
f6ea3adb | 759 | */ |
be976572 VB |
760 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
761 | && compact_should_abort(cc)) | |
762 | break; | |
f6ea3adb | 763 | |
7d49d886 VB |
764 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
765 | zone); | |
766 | if (!page) | |
ff9543fd MN |
767 | continue; |
768 | ||
769 | /* Check the block is suitable for migration */ | |
68e3e926 | 770 | if (!suitable_migration_target(page)) |
ff9543fd | 771 | continue; |
68e3e926 | 772 | |
bb13ffeb MG |
773 | /* If isolation recently failed, do not retry */ |
774 | if (!isolation_suitable(cc, page)) | |
775 | continue; | |
776 | ||
f40d1e42 | 777 | /* Found a block suitable for isolating free pages from */ |
e9ade569 | 778 | cc->free_pfn = block_start_pfn; |
c96b9e50 VB |
779 | isolated = isolate_freepages_block(cc, block_start_pfn, |
780 | block_end_pfn, freelist, false); | |
f40d1e42 | 781 | nr_freepages += isolated; |
ff9543fd MN |
782 | |
783 | /* | |
e9ade569 VB |
784 | * Set a flag that we successfully isolated in this pageblock. |
785 | * In the next loop iteration, zone->compact_cached_free_pfn | |
786 | * will not be updated and thus it will effectively contain the | |
787 | * highest pageblock we isolated pages from. | |
ff9543fd | 788 | */ |
e9ade569 | 789 | if (isolated) |
c89511ab | 790 | cc->finished_update_free = true; |
be976572 VB |
791 | |
792 | /* | |
793 | * isolate_freepages_block() might have aborted due to async | |
794 | * compaction being contended | |
795 | */ | |
796 | if (cc->contended) | |
797 | break; | |
ff9543fd MN |
798 | } |
799 | ||
800 | /* split_free_page does not map the pages */ | |
801 | map_pages(freelist); | |
802 | ||
7ed695e0 VB |
803 | /* |
804 | * If we crossed the migrate scanner, we want to keep it that way | |
805 | * so that compact_finished() may detect this | |
806 | */ | |
c96b9e50 | 807 | if (block_start_pfn < low_pfn) |
e9ade569 | 808 | cc->free_pfn = cc->migrate_pfn; |
c96b9e50 | 809 | |
ff9543fd | 810 | cc->nr_freepages = nr_freepages; |
748446bb MG |
811 | } |
812 | ||
813 | /* | |
814 | * This is a migrate-callback that "allocates" freepages by taking pages | |
815 | * from the isolated freelists in the block we are migrating to. | |
816 | */ | |
817 | static struct page *compaction_alloc(struct page *migratepage, | |
818 | unsigned long data, | |
819 | int **result) | |
820 | { | |
821 | struct compact_control *cc = (struct compact_control *)data; | |
822 | struct page *freepage; | |
823 | ||
be976572 VB |
824 | /* |
825 | * Isolate free pages if necessary, and if we are not aborting due to | |
826 | * contention. | |
827 | */ | |
748446bb | 828 | if (list_empty(&cc->freepages)) { |
be976572 | 829 | if (!cc->contended) |
edc2ca61 | 830 | isolate_freepages(cc); |
748446bb MG |
831 | |
832 | if (list_empty(&cc->freepages)) | |
833 | return NULL; | |
834 | } | |
835 | ||
836 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
837 | list_del(&freepage->lru); | |
838 | cc->nr_freepages--; | |
839 | ||
840 | return freepage; | |
841 | } | |
842 | ||
843 | /* | |
d53aea3d DR |
844 | * This is a migrate-callback that "frees" freepages back to the isolated |
845 | * freelist. All pages on the freelist are from the same zone, so there is no | |
846 | * special handling needed for NUMA. | |
847 | */ | |
848 | static void compaction_free(struct page *page, unsigned long data) | |
849 | { | |
850 | struct compact_control *cc = (struct compact_control *)data; | |
851 | ||
852 | list_add(&page->lru, &cc->freepages); | |
853 | cc->nr_freepages++; | |
854 | } | |
855 | ||
ff9543fd MN |
856 | /* possible outcome of isolate_migratepages */ |
857 | typedef enum { | |
858 | ISOLATE_ABORT, /* Abort compaction now */ | |
859 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
860 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
861 | } isolate_migrate_t; | |
862 | ||
863 | /* | |
edc2ca61 VB |
864 | * Isolate all pages that can be migrated from the first suitable block, |
865 | * starting at the block pointed to by the migrate scanner pfn within | |
866 | * compact_control. | |
ff9543fd MN |
867 | */ |
868 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
869 | struct compact_control *cc) | |
870 | { | |
871 | unsigned long low_pfn, end_pfn; | |
edc2ca61 VB |
872 | struct page *page; |
873 | const isolate_mode_t isolate_mode = | |
874 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); | |
ff9543fd | 875 | |
edc2ca61 VB |
876 | /* |
877 | * Start at where we last stopped, or beginning of the zone as | |
878 | * initialized by compact_zone() | |
879 | */ | |
880 | low_pfn = cc->migrate_pfn; | |
ff9543fd MN |
881 | |
882 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 883 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd | 884 | |
edc2ca61 VB |
885 | /* |
886 | * Iterate over whole pageblocks until we find the first suitable. | |
887 | * Do not cross the free scanner. | |
888 | */ | |
889 | for (; end_pfn <= cc->free_pfn; | |
890 | low_pfn = end_pfn, end_pfn += pageblock_nr_pages) { | |
ff9543fd | 891 | |
edc2ca61 VB |
892 | /* |
893 | * This can potentially iterate a massively long zone with | |
894 | * many pageblocks unsuitable, so periodically check if we | |
895 | * need to schedule, or even abort async compaction. | |
896 | */ | |
897 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) | |
898 | && compact_should_abort(cc)) | |
899 | break; | |
ff9543fd | 900 | |
7d49d886 VB |
901 | page = pageblock_pfn_to_page(low_pfn, end_pfn, zone); |
902 | if (!page) | |
edc2ca61 VB |
903 | continue; |
904 | ||
edc2ca61 VB |
905 | /* If isolation recently failed, do not retry */ |
906 | if (!isolation_suitable(cc, page)) | |
907 | continue; | |
908 | ||
909 | /* | |
910 | * For async compaction, also only scan in MOVABLE blocks. | |
911 | * Async compaction is optimistic to see if the minimum amount | |
912 | * of work satisfies the allocation. | |
913 | */ | |
914 | if (cc->mode == MIGRATE_ASYNC && | |
915 | !migrate_async_suitable(get_pageblock_migratetype(page))) | |
916 | continue; | |
917 | ||
918 | /* Perform the isolation */ | |
919 | low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn, | |
920 | isolate_mode); | |
921 | ||
922 | if (!low_pfn || cc->contended) | |
923 | return ISOLATE_ABORT; | |
924 | ||
925 | /* | |
926 | * Either we isolated something and proceed with migration. Or | |
927 | * we failed and compact_zone should decide if we should | |
928 | * continue or not. | |
929 | */ | |
930 | break; | |
931 | } | |
932 | ||
933 | acct_isolated(zone, cc); | |
934 | /* Record where migration scanner will be restarted */ | |
ff9543fd MN |
935 | cc->migrate_pfn = low_pfn; |
936 | ||
edc2ca61 | 937 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
ff9543fd MN |
938 | } |
939 | ||
748446bb | 940 | static int compact_finished(struct zone *zone, |
5a03b051 | 941 | struct compact_control *cc) |
748446bb | 942 | { |
8fb74b9f | 943 | unsigned int order; |
5a03b051 | 944 | unsigned long watermark; |
56de7263 | 945 | |
be976572 | 946 | if (cc->contended || fatal_signal_pending(current)) |
748446bb MG |
947 | return COMPACT_PARTIAL; |
948 | ||
753341a4 | 949 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 950 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 | 951 | /* Let the next compaction start anew. */ |
35979ef3 DR |
952 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
953 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
55b7c4c9 VB |
954 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
955 | ||
62997027 MG |
956 | /* |
957 | * Mark that the PG_migrate_skip information should be cleared | |
958 | * by kswapd when it goes to sleep. kswapd does not set the | |
959 | * flag itself as the decision to be clear should be directly | |
960 | * based on an allocation request. | |
961 | */ | |
962 | if (!current_is_kswapd()) | |
963 | zone->compact_blockskip_flush = true; | |
964 | ||
748446bb | 965 | return COMPACT_COMPLETE; |
bb13ffeb | 966 | } |
748446bb | 967 | |
82478fb7 JW |
968 | /* |
969 | * order == -1 is expected when compacting via | |
970 | * /proc/sys/vm/compact_memory | |
971 | */ | |
56de7263 MG |
972 | if (cc->order == -1) |
973 | return COMPACT_CONTINUE; | |
974 | ||
3957c776 MH |
975 | /* Compaction run is not finished if the watermark is not met */ |
976 | watermark = low_wmark_pages(zone); | |
977 | watermark += (1 << cc->order); | |
978 | ||
979 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
980 | return COMPACT_CONTINUE; | |
981 | ||
56de7263 | 982 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
983 | for (order = cc->order; order < MAX_ORDER; order++) { |
984 | struct free_area *area = &zone->free_area[order]; | |
985 | ||
986 | /* Job done if page is free of the right migratetype */ | |
987 | if (!list_empty(&area->free_list[cc->migratetype])) | |
988 | return COMPACT_PARTIAL; | |
989 | ||
990 | /* Job done if allocation would set block type */ | |
991 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
992 | return COMPACT_PARTIAL; |
993 | } | |
994 | ||
748446bb MG |
995 | return COMPACT_CONTINUE; |
996 | } | |
997 | ||
3e7d3449 MG |
998 | /* |
999 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
1000 | * Returns | |
1001 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
1002 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
1003 | * COMPACT_CONTINUE - If compaction should run now | |
1004 | */ | |
1005 | unsigned long compaction_suitable(struct zone *zone, int order) | |
1006 | { | |
1007 | int fragindex; | |
1008 | unsigned long watermark; | |
1009 | ||
3957c776 MH |
1010 | /* |
1011 | * order == -1 is expected when compacting via | |
1012 | * /proc/sys/vm/compact_memory | |
1013 | */ | |
1014 | if (order == -1) | |
1015 | return COMPACT_CONTINUE; | |
1016 | ||
3e7d3449 MG |
1017 | /* |
1018 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
1019 | * This is because during migration, copies of pages need to be | |
1020 | * allocated and for a short time, the footprint is higher | |
1021 | */ | |
1022 | watermark = low_wmark_pages(zone) + (2UL << order); | |
1023 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1024 | return COMPACT_SKIPPED; | |
1025 | ||
1026 | /* | |
1027 | * fragmentation index determines if allocation failures are due to | |
1028 | * low memory or external fragmentation | |
1029 | * | |
a582a738 SL |
1030 | * index of -1000 implies allocations might succeed depending on |
1031 | * watermarks | |
3e7d3449 MG |
1032 | * index towards 0 implies failure is due to lack of memory |
1033 | * index towards 1000 implies failure is due to fragmentation | |
1034 | * | |
1035 | * Only compact if a failure would be due to fragmentation. | |
1036 | */ | |
1037 | fragindex = fragmentation_index(zone, order); | |
1038 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
1039 | return COMPACT_SKIPPED; | |
1040 | ||
a582a738 SL |
1041 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
1042 | 0, 0)) | |
3e7d3449 MG |
1043 | return COMPACT_PARTIAL; |
1044 | ||
1045 | return COMPACT_CONTINUE; | |
1046 | } | |
1047 | ||
748446bb MG |
1048 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
1049 | { | |
1050 | int ret; | |
c89511ab | 1051 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 1052 | unsigned long end_pfn = zone_end_pfn(zone); |
e0b9daeb | 1053 | const bool sync = cc->mode != MIGRATE_ASYNC; |
748446bb | 1054 | |
3e7d3449 MG |
1055 | ret = compaction_suitable(zone, cc->order); |
1056 | switch (ret) { | |
1057 | case COMPACT_PARTIAL: | |
1058 | case COMPACT_SKIPPED: | |
1059 | /* Compaction is likely to fail */ | |
1060 | return ret; | |
1061 | case COMPACT_CONTINUE: | |
1062 | /* Fall through to compaction */ | |
1063 | ; | |
1064 | } | |
1065 | ||
d3132e4b VB |
1066 | /* |
1067 | * Clear pageblock skip if there were failures recently and compaction | |
1068 | * is about to be retried after being deferred. kswapd does not do | |
1069 | * this reset as it'll reset the cached information when going to sleep. | |
1070 | */ | |
1071 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1072 | __reset_isolation_suitable(zone); | |
1073 | ||
c89511ab MG |
1074 | /* |
1075 | * Setup to move all movable pages to the end of the zone. Used cached | |
1076 | * information on where the scanners should start but check that it | |
1077 | * is initialised by ensuring the values are within zone boundaries. | |
1078 | */ | |
e0b9daeb | 1079 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
c89511ab MG |
1080 | cc->free_pfn = zone->compact_cached_free_pfn; |
1081 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
1082 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
1083 | zone->compact_cached_free_pfn = cc->free_pfn; | |
1084 | } | |
1085 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1086 | cc->migrate_pfn = start_pfn; | |
35979ef3 DR |
1087 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
1088 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
c89511ab | 1089 | } |
748446bb | 1090 | |
0eb927c0 MG |
1091 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn); |
1092 | ||
748446bb MG |
1093 | migrate_prep_local(); |
1094 | ||
1095 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
9d502c1c | 1096 | int err; |
748446bb | 1097 | |
f9e35b3b MG |
1098 | switch (isolate_migratepages(zone, cc)) { |
1099 | case ISOLATE_ABORT: | |
1100 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1101 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1102 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1103 | goto out; |
1104 | case ISOLATE_NONE: | |
748446bb | 1105 | continue; |
f9e35b3b MG |
1106 | case ISOLATE_SUCCESS: |
1107 | ; | |
1108 | } | |
748446bb | 1109 | |
d53aea3d | 1110 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 1111 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 1112 | MR_COMPACTION); |
748446bb | 1113 | |
f8c9301f VB |
1114 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
1115 | &cc->migratepages); | |
748446bb | 1116 | |
f8c9301f VB |
1117 | /* All pages were either migrated or will be released */ |
1118 | cc->nr_migratepages = 0; | |
9d502c1c | 1119 | if (err) { |
5733c7d1 | 1120 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
1121 | /* |
1122 | * migrate_pages() may return -ENOMEM when scanners meet | |
1123 | * and we want compact_finished() to detect it | |
1124 | */ | |
1125 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1126 | ret = COMPACT_PARTIAL; |
1127 | goto out; | |
1128 | } | |
748446bb | 1129 | } |
748446bb MG |
1130 | } |
1131 | ||
f9e35b3b | 1132 | out: |
748446bb MG |
1133 | /* Release free pages and check accounting */ |
1134 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1135 | VM_BUG_ON(cc->nr_freepages != 0); | |
1136 | ||
0eb927c0 MG |
1137 | trace_mm_compaction_end(ret); |
1138 | ||
748446bb MG |
1139 | return ret; |
1140 | } | |
76ab0f53 | 1141 | |
e0b9daeb DR |
1142 | static unsigned long compact_zone_order(struct zone *zone, int order, |
1143 | gfp_t gfp_mask, enum migrate_mode mode, bool *contended) | |
56de7263 | 1144 | { |
e64c5237 | 1145 | unsigned long ret; |
56de7263 MG |
1146 | struct compact_control cc = { |
1147 | .nr_freepages = 0, | |
1148 | .nr_migratepages = 0, | |
1149 | .order = order, | |
1150 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1151 | .zone = zone, | |
e0b9daeb | 1152 | .mode = mode, |
56de7263 MG |
1153 | }; |
1154 | INIT_LIST_HEAD(&cc.freepages); | |
1155 | INIT_LIST_HEAD(&cc.migratepages); | |
1156 | ||
e64c5237 SL |
1157 | ret = compact_zone(zone, &cc); |
1158 | ||
1159 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1160 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1161 | ||
1162 | *contended = cc.contended; | |
1163 | return ret; | |
56de7263 MG |
1164 | } |
1165 | ||
5e771905 MG |
1166 | int sysctl_extfrag_threshold = 500; |
1167 | ||
56de7263 MG |
1168 | /** |
1169 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1170 | * @zonelist: The zonelist used for the current allocation | |
1171 | * @order: The order of the current allocation | |
1172 | * @gfp_mask: The GFP mask of the current allocation | |
1173 | * @nodemask: The allowed nodes to allocate from | |
e0b9daeb | 1174 | * @mode: The migration mode for async, sync light, or sync migration |
661c4cb9 | 1175 | * @contended: Return value that is true if compaction was aborted due to lock contention |
53853e2d | 1176 | * @candidate_zone: Return the zone where we think allocation should succeed |
56de7263 MG |
1177 | * |
1178 | * This is the main entry point for direct page compaction. | |
1179 | */ | |
1180 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1181 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
53853e2d VB |
1182 | enum migrate_mode mode, bool *contended, |
1183 | struct zone **candidate_zone) | |
56de7263 MG |
1184 | { |
1185 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1186 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1187 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1188 | struct zoneref *z; |
1189 | struct zone *zone; | |
53853e2d | 1190 | int rc = COMPACT_DEFERRED; |
d95ea5d1 | 1191 | int alloc_flags = 0; |
56de7263 | 1192 | |
4ffb6335 | 1193 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1194 | if (!order || !may_enter_fs || !may_perform_io) |
53853e2d | 1195 | return COMPACT_SKIPPED; |
56de7263 | 1196 | |
d95ea5d1 BZ |
1197 | #ifdef CONFIG_CMA |
1198 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1199 | alloc_flags |= ALLOC_CMA; | |
1200 | #endif | |
56de7263 MG |
1201 | /* Compact each zone in the list */ |
1202 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1203 | nodemask) { | |
56de7263 MG |
1204 | int status; |
1205 | ||
53853e2d VB |
1206 | if (compaction_deferred(zone, order)) |
1207 | continue; | |
1208 | ||
e0b9daeb | 1209 | status = compact_zone_order(zone, order, gfp_mask, mode, |
8fb74b9f | 1210 | contended); |
56de7263 MG |
1211 | rc = max(status, rc); |
1212 | ||
3e7d3449 | 1213 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 | 1214 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
53853e2d VB |
1215 | alloc_flags)) { |
1216 | *candidate_zone = zone; | |
1217 | /* | |
1218 | * We think the allocation will succeed in this zone, | |
1219 | * but it is not certain, hence the false. The caller | |
1220 | * will repeat this with true if allocation indeed | |
1221 | * succeeds in this zone. | |
1222 | */ | |
1223 | compaction_defer_reset(zone, order, false); | |
56de7263 | 1224 | break; |
53853e2d VB |
1225 | } else if (mode != MIGRATE_ASYNC) { |
1226 | /* | |
1227 | * We think that allocation won't succeed in this zone | |
1228 | * so we defer compaction there. If it ends up | |
1229 | * succeeding after all, it will be reset. | |
1230 | */ | |
1231 | defer_compaction(zone, order); | |
1232 | } | |
56de7263 MG |
1233 | } |
1234 | ||
1235 | return rc; | |
1236 | } | |
1237 | ||
1238 | ||
76ab0f53 | 1239 | /* Compact all zones within a node */ |
7103f16d | 1240 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1241 | { |
1242 | int zoneid; | |
76ab0f53 MG |
1243 | struct zone *zone; |
1244 | ||
76ab0f53 | 1245 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1246 | |
1247 | zone = &pgdat->node_zones[zoneid]; | |
1248 | if (!populated_zone(zone)) | |
1249 | continue; | |
1250 | ||
7be62de9 RR |
1251 | cc->nr_freepages = 0; |
1252 | cc->nr_migratepages = 0; | |
1253 | cc->zone = zone; | |
1254 | INIT_LIST_HEAD(&cc->freepages); | |
1255 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1256 | |
aad6ec37 | 1257 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1258 | compact_zone(zone, cc); |
76ab0f53 | 1259 | |
aff62249 | 1260 | if (cc->order > 0) { |
de6c60a6 VB |
1261 | if (zone_watermark_ok(zone, cc->order, |
1262 | low_wmark_pages(zone), 0, 0)) | |
1263 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 RR |
1264 | } |
1265 | ||
7be62de9 RR |
1266 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1267 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1268 | } |
76ab0f53 MG |
1269 | } |
1270 | ||
7103f16d | 1271 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1272 | { |
1273 | struct compact_control cc = { | |
1274 | .order = order, | |
e0b9daeb | 1275 | .mode = MIGRATE_ASYNC, |
7be62de9 RR |
1276 | }; |
1277 | ||
3a7200af MG |
1278 | if (!order) |
1279 | return; | |
1280 | ||
7103f16d | 1281 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1282 | } |
1283 | ||
7103f16d | 1284 | static void compact_node(int nid) |
7be62de9 | 1285 | { |
7be62de9 RR |
1286 | struct compact_control cc = { |
1287 | .order = -1, | |
e0b9daeb | 1288 | .mode = MIGRATE_SYNC, |
91ca9186 | 1289 | .ignore_skip_hint = true, |
7be62de9 RR |
1290 | }; |
1291 | ||
7103f16d | 1292 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1293 | } |
1294 | ||
76ab0f53 | 1295 | /* Compact all nodes in the system */ |
7964c06d | 1296 | static void compact_nodes(void) |
76ab0f53 MG |
1297 | { |
1298 | int nid; | |
1299 | ||
8575ec29 HD |
1300 | /* Flush pending updates to the LRU lists */ |
1301 | lru_add_drain_all(); | |
1302 | ||
76ab0f53 MG |
1303 | for_each_online_node(nid) |
1304 | compact_node(nid); | |
76ab0f53 MG |
1305 | } |
1306 | ||
1307 | /* The written value is actually unused, all memory is compacted */ | |
1308 | int sysctl_compact_memory; | |
1309 | ||
1310 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1311 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1312 | void __user *buffer, size_t *length, loff_t *ppos) | |
1313 | { | |
1314 | if (write) | |
7964c06d | 1315 | compact_nodes(); |
76ab0f53 MG |
1316 | |
1317 | return 0; | |
1318 | } | |
ed4a6d7f | 1319 | |
5e771905 MG |
1320 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1321 | void __user *buffer, size_t *length, loff_t *ppos) | |
1322 | { | |
1323 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1324 | ||
1325 | return 0; | |
1326 | } | |
1327 | ||
ed4a6d7f | 1328 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1329 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1330 | struct device_attribute *attr, |
ed4a6d7f MG |
1331 | const char *buf, size_t count) |
1332 | { | |
8575ec29 HD |
1333 | int nid = dev->id; |
1334 | ||
1335 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1336 | /* Flush pending updates to the LRU lists */ | |
1337 | lru_add_drain_all(); | |
1338 | ||
1339 | compact_node(nid); | |
1340 | } | |
ed4a6d7f MG |
1341 | |
1342 | return count; | |
1343 | } | |
10fbcf4c | 1344 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1345 | |
1346 | int compaction_register_node(struct node *node) | |
1347 | { | |
10fbcf4c | 1348 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1349 | } |
1350 | ||
1351 | void compaction_unregister_node(struct node *node) | |
1352 | { | |
10fbcf4c | 1353 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1354 | } |
1355 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1356 | |
1357 | #endif /* CONFIG_COMPACTION */ |