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