]>
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> |
748446bb MG |
18 | #include "internal.h" |
19 | ||
ff9543fd MN |
20 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
21 | ||
b7aba698 MG |
22 | #define CREATE_TRACE_POINTS |
23 | #include <trace/events/compaction.h> | |
24 | ||
748446bb MG |
25 | static unsigned long release_freepages(struct list_head *freelist) |
26 | { | |
27 | struct page *page, *next; | |
28 | unsigned long count = 0; | |
29 | ||
30 | list_for_each_entry_safe(page, next, freelist, lru) { | |
31 | list_del(&page->lru); | |
32 | __free_page(page); | |
33 | count++; | |
34 | } | |
35 | ||
36 | return count; | |
37 | } | |
38 | ||
ff9543fd MN |
39 | static void map_pages(struct list_head *list) |
40 | { | |
41 | struct page *page; | |
42 | ||
43 | list_for_each_entry(page, list, lru) { | |
44 | arch_alloc_page(page, 0); | |
45 | kernel_map_pages(page, 1, 1); | |
46 | } | |
47 | } | |
48 | ||
47118af0 MN |
49 | static inline bool migrate_async_suitable(int migratetype) |
50 | { | |
51 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; | |
52 | } | |
53 | ||
bb13ffeb MG |
54 | #ifdef CONFIG_COMPACTION |
55 | /* Returns true if the pageblock should be scanned for pages to isolate. */ | |
56 | static inline bool isolation_suitable(struct compact_control *cc, | |
57 | struct page *page) | |
58 | { | |
59 | if (cc->ignore_skip_hint) | |
60 | return true; | |
61 | ||
62 | return !get_pageblock_skip(page); | |
63 | } | |
64 | ||
65 | /* | |
66 | * This function is called to clear all cached information on pageblocks that | |
67 | * should be skipped for page isolation when the migrate and free page scanner | |
68 | * meet. | |
69 | */ | |
62997027 | 70 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
71 | { |
72 | unsigned long start_pfn = zone->zone_start_pfn; | |
73 | unsigned long end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
74 | unsigned long pfn; | |
75 | ||
c89511ab MG |
76 | zone->compact_cached_migrate_pfn = start_pfn; |
77 | zone->compact_cached_free_pfn = end_pfn; | |
62997027 | 78 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
79 | |
80 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
81 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
82 | struct page *page; | |
83 | ||
84 | cond_resched(); | |
85 | ||
86 | if (!pfn_valid(pfn)) | |
87 | continue; | |
88 | ||
89 | page = pfn_to_page(pfn); | |
90 | if (zone != page_zone(page)) | |
91 | continue; | |
92 | ||
93 | clear_pageblock_skip(page); | |
94 | } | |
95 | } | |
96 | ||
62997027 MG |
97 | void reset_isolation_suitable(pg_data_t *pgdat) |
98 | { | |
99 | int zoneid; | |
100 | ||
101 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
102 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
103 | if (!populated_zone(zone)) | |
104 | continue; | |
105 | ||
106 | /* Only flush if a full compaction finished recently */ | |
107 | if (zone->compact_blockskip_flush) | |
108 | __reset_isolation_suitable(zone); | |
109 | } | |
110 | } | |
111 | ||
bb13ffeb MG |
112 | /* |
113 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 114 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 115 | */ |
c89511ab MG |
116 | static void update_pageblock_skip(struct compact_control *cc, |
117 | struct page *page, unsigned long nr_isolated, | |
118 | bool migrate_scanner) | |
bb13ffeb | 119 | { |
c89511ab | 120 | struct zone *zone = cc->zone; |
bb13ffeb MG |
121 | if (!page) |
122 | return; | |
123 | ||
c89511ab MG |
124 | if (!nr_isolated) { |
125 | unsigned long pfn = page_to_pfn(page); | |
bb13ffeb | 126 | set_pageblock_skip(page); |
c89511ab MG |
127 | |
128 | /* Update where compaction should restart */ | |
129 | if (migrate_scanner) { | |
130 | if (!cc->finished_update_migrate && | |
131 | pfn > zone->compact_cached_migrate_pfn) | |
132 | zone->compact_cached_migrate_pfn = pfn; | |
133 | } else { | |
134 | if (!cc->finished_update_free && | |
135 | pfn < zone->compact_cached_free_pfn) | |
136 | zone->compact_cached_free_pfn = pfn; | |
137 | } | |
138 | } | |
bb13ffeb MG |
139 | } |
140 | #else | |
141 | static inline bool isolation_suitable(struct compact_control *cc, | |
142 | struct page *page) | |
143 | { | |
144 | return true; | |
145 | } | |
146 | ||
c89511ab MG |
147 | static void update_pageblock_skip(struct compact_control *cc, |
148 | struct page *page, unsigned long nr_isolated, | |
149 | bool migrate_scanner) | |
bb13ffeb MG |
150 | { |
151 | } | |
152 | #endif /* CONFIG_COMPACTION */ | |
153 | ||
2a1402aa MG |
154 | static inline bool should_release_lock(spinlock_t *lock) |
155 | { | |
156 | return need_resched() || spin_is_contended(lock); | |
157 | } | |
158 | ||
c67fe375 MG |
159 | /* |
160 | * Compaction requires the taking of some coarse locks that are potentially | |
161 | * very heavily contended. Check if the process needs to be scheduled or | |
162 | * if the lock is contended. For async compaction, back out in the event | |
163 | * if contention is severe. For sync compaction, schedule. | |
164 | * | |
165 | * Returns true if the lock is held. | |
166 | * Returns false if the lock is released and compaction should abort | |
167 | */ | |
168 | static bool compact_checklock_irqsave(spinlock_t *lock, unsigned long *flags, | |
169 | bool locked, struct compact_control *cc) | |
170 | { | |
2a1402aa | 171 | if (should_release_lock(lock)) { |
c67fe375 MG |
172 | if (locked) { |
173 | spin_unlock_irqrestore(lock, *flags); | |
174 | locked = false; | |
175 | } | |
176 | ||
177 | /* async aborts if taking too long or contended */ | |
178 | if (!cc->sync) { | |
e64c5237 | 179 | cc->contended = true; |
c67fe375 MG |
180 | return false; |
181 | } | |
182 | ||
183 | cond_resched(); | |
c67fe375 MG |
184 | } |
185 | ||
186 | if (!locked) | |
187 | spin_lock_irqsave(lock, *flags); | |
188 | return true; | |
189 | } | |
190 | ||
191 | static inline bool compact_trylock_irqsave(spinlock_t *lock, | |
192 | unsigned long *flags, struct compact_control *cc) | |
193 | { | |
194 | return compact_checklock_irqsave(lock, flags, false, cc); | |
195 | } | |
196 | ||
f40d1e42 MG |
197 | /* Returns true if the page is within a block suitable for migration to */ |
198 | static bool suitable_migration_target(struct page *page) | |
199 | { | |
200 | int migratetype = get_pageblock_migratetype(page); | |
201 | ||
202 | /* Don't interfere with memory hot-remove or the min_free_kbytes blocks */ | |
203 | if (migratetype == MIGRATE_ISOLATE || migratetype == MIGRATE_RESERVE) | |
204 | return false; | |
205 | ||
206 | /* If the page is a large free page, then allow migration */ | |
207 | if (PageBuddy(page) && page_order(page) >= pageblock_order) | |
208 | return true; | |
209 | ||
210 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
211 | if (migrate_async_suitable(migratetype)) | |
212 | return true; | |
213 | ||
214 | /* Otherwise skip the block */ | |
215 | return false; | |
216 | } | |
217 | ||
85aa125f MN |
218 | /* |
219 | * Isolate free pages onto a private freelist. Caller must hold zone->lock. | |
220 | * If @strict is true, will abort returning 0 on any invalid PFNs or non-free | |
221 | * pages inside of the pageblock (even though it may still end up isolating | |
222 | * some pages). | |
223 | */ | |
f40d1e42 MG |
224 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
225 | unsigned long blockpfn, | |
85aa125f MN |
226 | unsigned long end_pfn, |
227 | struct list_head *freelist, | |
228 | bool strict) | |
748446bb | 229 | { |
b7aba698 | 230 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 231 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
232 | unsigned long nr_strict_required = end_pfn - blockpfn; |
233 | unsigned long flags; | |
234 | bool locked = false; | |
748446bb | 235 | |
748446bb MG |
236 | cursor = pfn_to_page(blockpfn); |
237 | ||
f40d1e42 | 238 | /* Isolate free pages. */ |
748446bb MG |
239 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
240 | int isolated, i; | |
241 | struct page *page = cursor; | |
242 | ||
b7aba698 | 243 | nr_scanned++; |
f40d1e42 MG |
244 | if (!pfn_valid_within(blockpfn)) |
245 | continue; | |
bb13ffeb MG |
246 | if (!valid_page) |
247 | valid_page = page; | |
f40d1e42 MG |
248 | if (!PageBuddy(page)) |
249 | continue; | |
250 | ||
251 | /* | |
252 | * The zone lock must be held to isolate freepages. | |
253 | * Unfortunately this is a very coarse lock and can be | |
254 | * heavily contended if there are parallel allocations | |
255 | * or parallel compactions. For async compaction do not | |
256 | * spin on the lock and we acquire the lock as late as | |
257 | * possible. | |
258 | */ | |
259 | locked = compact_checklock_irqsave(&cc->zone->lock, &flags, | |
260 | locked, cc); | |
261 | if (!locked) | |
262 | break; | |
263 | ||
264 | /* Recheck this is a suitable migration target under lock */ | |
265 | if (!strict && !suitable_migration_target(page)) | |
266 | break; | |
748446bb | 267 | |
f40d1e42 MG |
268 | /* Recheck this is a buddy page under lock */ |
269 | if (!PageBuddy(page)) | |
748446bb MG |
270 | continue; |
271 | ||
272 | /* Found a free page, break it into order-0 pages */ | |
273 | isolated = split_free_page(page); | |
85aa125f | 274 | if (!isolated && strict) |
f40d1e42 | 275 | break; |
748446bb MG |
276 | total_isolated += isolated; |
277 | for (i = 0; i < isolated; i++) { | |
278 | list_add(&page->lru, freelist); | |
279 | page++; | |
280 | } | |
281 | ||
282 | /* If a page was split, advance to the end of it */ | |
283 | if (isolated) { | |
284 | blockpfn += isolated - 1; | |
285 | cursor += isolated - 1; | |
286 | } | |
287 | } | |
288 | ||
b7aba698 | 289 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
290 | |
291 | /* | |
292 | * If strict isolation is requested by CMA then check that all the | |
293 | * pages requested were isolated. If there were any failures, 0 is | |
294 | * returned and CMA will fail. | |
295 | */ | |
0db63d7e | 296 | if (strict && nr_strict_required > total_isolated) |
f40d1e42 MG |
297 | total_isolated = 0; |
298 | ||
299 | if (locked) | |
300 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
301 | ||
bb13ffeb MG |
302 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
303 | if (blockpfn == end_pfn) | |
c89511ab | 304 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 305 | |
397487db MG |
306 | count_vm_events(COMPACTFREE_SCANNED, nr_scanned); |
307 | if (total_isolated) | |
308 | count_vm_events(COMPACTISOLATED, total_isolated); | |
309 | ||
748446bb MG |
310 | return total_isolated; |
311 | } | |
312 | ||
85aa125f MN |
313 | /** |
314 | * isolate_freepages_range() - isolate free pages. | |
315 | * @start_pfn: The first PFN to start isolating. | |
316 | * @end_pfn: The one-past-last PFN. | |
317 | * | |
318 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
319 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
320 | * undo its actions and return zero. | |
321 | * | |
322 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
323 | * (which may be greater then end_pfn if end fell in a middle of | |
324 | * a free page). | |
325 | */ | |
ff9543fd | 326 | unsigned long |
bb13ffeb MG |
327 | isolate_freepages_range(struct compact_control *cc, |
328 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 329 | { |
f40d1e42 | 330 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
331 | LIST_HEAD(freelist); |
332 | ||
85aa125f | 333 | for (pfn = start_pfn; pfn < end_pfn; pfn += isolated) { |
bb13ffeb | 334 | if (!pfn_valid(pfn) || cc->zone != page_zone(pfn_to_page(pfn))) |
85aa125f MN |
335 | break; |
336 | ||
337 | /* | |
338 | * On subsequent iterations ALIGN() is actually not needed, | |
339 | * but we keep it that we not to complicate the code. | |
340 | */ | |
341 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
342 | block_end_pfn = min(block_end_pfn, end_pfn); | |
343 | ||
bb13ffeb | 344 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 345 | &freelist, true); |
85aa125f MN |
346 | |
347 | /* | |
348 | * In strict mode, isolate_freepages_block() returns 0 if | |
349 | * there are any holes in the block (ie. invalid PFNs or | |
350 | * non-free pages). | |
351 | */ | |
352 | if (!isolated) | |
353 | break; | |
354 | ||
355 | /* | |
356 | * If we managed to isolate pages, it is always (1 << n) * | |
357 | * pageblock_nr_pages for some non-negative n. (Max order | |
358 | * page may span two pageblocks). | |
359 | */ | |
360 | } | |
361 | ||
362 | /* split_free_page does not map the pages */ | |
363 | map_pages(&freelist); | |
364 | ||
365 | if (pfn < end_pfn) { | |
366 | /* Loop terminated early, cleanup. */ | |
367 | release_freepages(&freelist); | |
368 | return 0; | |
369 | } | |
370 | ||
371 | /* We don't use freelists for anything. */ | |
372 | return pfn; | |
373 | } | |
374 | ||
748446bb | 375 | /* Update the number of anon and file isolated pages in the zone */ |
c67fe375 | 376 | static void acct_isolated(struct zone *zone, bool locked, struct compact_control *cc) |
748446bb MG |
377 | { |
378 | struct page *page; | |
b9e84ac1 | 379 | unsigned int count[2] = { 0, }; |
748446bb | 380 | |
b9e84ac1 MK |
381 | list_for_each_entry(page, &cc->migratepages, lru) |
382 | count[!!page_is_file_cache(page)]++; | |
748446bb | 383 | |
c67fe375 MG |
384 | /* If locked we can use the interrupt unsafe versions */ |
385 | if (locked) { | |
386 | __mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
387 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
388 | } else { | |
389 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); | |
390 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
391 | } | |
748446bb MG |
392 | } |
393 | ||
394 | /* Similar to reclaim, but different enough that they don't share logic */ | |
395 | static bool too_many_isolated(struct zone *zone) | |
396 | { | |
bc693045 | 397 | unsigned long active, inactive, isolated; |
748446bb MG |
398 | |
399 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
400 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
401 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
402 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
403 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
404 | zone_page_state(zone, NR_ISOLATED_ANON); | |
405 | ||
bc693045 | 406 | return isolated > (inactive + active) / 2; |
748446bb MG |
407 | } |
408 | ||
2fe86e00 MN |
409 | /** |
410 | * isolate_migratepages_range() - isolate all migrate-able pages in range. | |
411 | * @zone: Zone pages are in. | |
412 | * @cc: Compaction control structure. | |
413 | * @low_pfn: The first PFN of the range. | |
414 | * @end_pfn: The one-past-the-last PFN of the range. | |
e46a2879 | 415 | * @unevictable: true if it allows to isolate unevictable pages |
2fe86e00 MN |
416 | * |
417 | * Isolate all pages that can be migrated from the range specified by | |
418 | * [low_pfn, end_pfn). Returns zero if there is a fatal signal | |
419 | * pending), otherwise PFN of the first page that was not scanned | |
420 | * (which may be both less, equal to or more then end_pfn). | |
421 | * | |
422 | * Assumes that cc->migratepages is empty and cc->nr_migratepages is | |
423 | * zero. | |
424 | * | |
425 | * Apart from cc->migratepages and cc->nr_migratetypes this function | |
426 | * does not modify any cc's fields, in particular it does not modify | |
427 | * (or read for that matter) cc->migrate_pfn. | |
748446bb | 428 | */ |
ff9543fd | 429 | unsigned long |
2fe86e00 | 430 | isolate_migratepages_range(struct zone *zone, struct compact_control *cc, |
e46a2879 | 431 | unsigned long low_pfn, unsigned long end_pfn, bool unevictable) |
748446bb | 432 | { |
9927af74 | 433 | unsigned long last_pageblock_nr = 0, pageblock_nr; |
b7aba698 | 434 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 435 | struct list_head *migratelist = &cc->migratepages; |
f3fd4a61 | 436 | isolate_mode_t mode = 0; |
fa9add64 | 437 | struct lruvec *lruvec; |
c67fe375 | 438 | unsigned long flags; |
2a1402aa | 439 | bool locked = false; |
bb13ffeb | 440 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 441 | |
748446bb MG |
442 | /* |
443 | * Ensure that there are not too many pages isolated from the LRU | |
444 | * list by either parallel reclaimers or compaction. If there are, | |
445 | * delay for some time until fewer pages are isolated | |
446 | */ | |
447 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 448 | /* async migration should just abort */ |
68e3e926 | 449 | if (!cc->sync) |
2fe86e00 | 450 | return 0; |
f9e35b3b | 451 | |
748446bb MG |
452 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
453 | ||
454 | if (fatal_signal_pending(current)) | |
2fe86e00 | 455 | return 0; |
748446bb MG |
456 | } |
457 | ||
458 | /* Time to isolate some pages for migration */ | |
b2eef8c0 | 459 | cond_resched(); |
748446bb | 460 | for (; low_pfn < end_pfn; low_pfn++) { |
b2eef8c0 | 461 | /* give a chance to irqs before checking need_resched() */ |
2a1402aa MG |
462 | if (locked && !((low_pfn+1) % SWAP_CLUSTER_MAX)) { |
463 | if (should_release_lock(&zone->lru_lock)) { | |
464 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
465 | locked = false; | |
466 | } | |
b2eef8c0 | 467 | } |
c67fe375 | 468 | |
0bf380bc MG |
469 | /* |
470 | * migrate_pfn does not necessarily start aligned to a | |
471 | * pageblock. Ensure that pfn_valid is called when moving | |
472 | * into a new MAX_ORDER_NR_PAGES range in case of large | |
473 | * memory holes within the zone | |
474 | */ | |
475 | if ((low_pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) { | |
476 | if (!pfn_valid(low_pfn)) { | |
477 | low_pfn += MAX_ORDER_NR_PAGES - 1; | |
478 | continue; | |
479 | } | |
480 | } | |
481 | ||
748446bb MG |
482 | if (!pfn_valid_within(low_pfn)) |
483 | continue; | |
b7aba698 | 484 | nr_scanned++; |
748446bb | 485 | |
dc908600 MG |
486 | /* |
487 | * Get the page and ensure the page is within the same zone. | |
488 | * See the comment in isolate_freepages about overlapping | |
489 | * nodes. It is deliberate that the new zone lock is not taken | |
490 | * as memory compaction should not move pages between nodes. | |
491 | */ | |
748446bb | 492 | page = pfn_to_page(low_pfn); |
dc908600 MG |
493 | if (page_zone(page) != zone) |
494 | continue; | |
495 | ||
bb13ffeb MG |
496 | if (!valid_page) |
497 | valid_page = page; | |
498 | ||
499 | /* If isolation recently failed, do not retry */ | |
500 | pageblock_nr = low_pfn >> pageblock_order; | |
501 | if (!isolation_suitable(cc, page)) | |
502 | goto next_pageblock; | |
503 | ||
dc908600 | 504 | /* Skip if free */ |
748446bb MG |
505 | if (PageBuddy(page)) |
506 | continue; | |
507 | ||
9927af74 MG |
508 | /* |
509 | * For async migration, also only scan in MOVABLE blocks. Async | |
510 | * migration is optimistic to see if the minimum amount of work | |
511 | * satisfies the allocation | |
512 | */ | |
68e3e926 | 513 | if (!cc->sync && last_pageblock_nr != pageblock_nr && |
47118af0 | 514 | !migrate_async_suitable(get_pageblock_migratetype(page))) { |
c89511ab | 515 | cc->finished_update_migrate = true; |
2a1402aa | 516 | goto next_pageblock; |
9927af74 MG |
517 | } |
518 | ||
bf6bddf1 RA |
519 | /* |
520 | * Check may be lockless but that's ok as we recheck later. | |
521 | * It's possible to migrate LRU pages and balloon pages | |
522 | * Skip any other type of page | |
523 | */ | |
524 | if (!PageLRU(page)) { | |
525 | if (unlikely(balloon_page_movable(page))) { | |
526 | if (locked && balloon_page_isolate(page)) { | |
527 | /* Successfully isolated */ | |
528 | cc->finished_update_migrate = true; | |
529 | list_add(&page->lru, migratelist); | |
530 | cc->nr_migratepages++; | |
531 | nr_isolated++; | |
532 | goto check_compact_cluster; | |
533 | } | |
534 | } | |
bc835011 | 535 | continue; |
bf6bddf1 | 536 | } |
bc835011 AA |
537 | |
538 | /* | |
2a1402aa MG |
539 | * PageLRU is set. lru_lock normally excludes isolation |
540 | * splitting and collapsing (collapsing has already happened | |
541 | * if PageLRU is set) but the lock is not necessarily taken | |
542 | * here and it is wasteful to take it just to check transhuge. | |
543 | * Check TransHuge without lock and skip the whole pageblock if | |
544 | * it's either a transhuge or hugetlbfs page, as calling | |
545 | * compound_order() without preventing THP from splitting the | |
546 | * page underneath us may return surprising results. | |
bc835011 | 547 | */ |
2a1402aa MG |
548 | if (PageTransHuge(page)) { |
549 | if (!locked) | |
550 | goto next_pageblock; | |
551 | low_pfn += (1 << compound_order(page)) - 1; | |
552 | continue; | |
553 | } | |
554 | ||
555 | /* Check if it is ok to still hold the lock */ | |
556 | locked = compact_checklock_irqsave(&zone->lru_lock, &flags, | |
557 | locked, cc); | |
558 | if (!locked || fatal_signal_pending(current)) | |
559 | break; | |
560 | ||
561 | /* Recheck PageLRU and PageTransHuge under lock */ | |
562 | if (!PageLRU(page)) | |
563 | continue; | |
bc835011 AA |
564 | if (PageTransHuge(page)) { |
565 | low_pfn += (1 << compound_order(page)) - 1; | |
566 | continue; | |
567 | } | |
568 | ||
68e3e926 | 569 | if (!cc->sync) |
c8244935 MG |
570 | mode |= ISOLATE_ASYNC_MIGRATE; |
571 | ||
e46a2879 MK |
572 | if (unevictable) |
573 | mode |= ISOLATE_UNEVICTABLE; | |
574 | ||
fa9add64 HD |
575 | lruvec = mem_cgroup_page_lruvec(page, zone); |
576 | ||
748446bb | 577 | /* Try isolate the page */ |
f3fd4a61 | 578 | if (__isolate_lru_page(page, mode) != 0) |
748446bb MG |
579 | continue; |
580 | ||
bc835011 AA |
581 | VM_BUG_ON(PageTransCompound(page)); |
582 | ||
748446bb | 583 | /* Successfully isolated */ |
c89511ab | 584 | cc->finished_update_migrate = true; |
fa9add64 | 585 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
748446bb | 586 | list_add(&page->lru, migratelist); |
748446bb | 587 | cc->nr_migratepages++; |
b7aba698 | 588 | nr_isolated++; |
748446bb | 589 | |
bf6bddf1 | 590 | check_compact_cluster: |
748446bb | 591 | /* Avoid isolating too much */ |
31b8384a HD |
592 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
593 | ++low_pfn; | |
748446bb | 594 | break; |
31b8384a | 595 | } |
2a1402aa MG |
596 | |
597 | continue; | |
598 | ||
599 | next_pageblock: | |
600 | low_pfn += pageblock_nr_pages; | |
601 | low_pfn = ALIGN(low_pfn, pageblock_nr_pages) - 1; | |
602 | last_pageblock_nr = pageblock_nr; | |
748446bb MG |
603 | } |
604 | ||
c67fe375 | 605 | acct_isolated(zone, locked, cc); |
748446bb | 606 | |
c67fe375 MG |
607 | if (locked) |
608 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 609 | |
bb13ffeb MG |
610 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
611 | if (low_pfn == end_pfn) | |
c89511ab | 612 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 613 | |
b7aba698 MG |
614 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
615 | ||
397487db MG |
616 | count_vm_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
617 | if (nr_isolated) | |
618 | count_vm_events(COMPACTISOLATED, nr_isolated); | |
619 | ||
2fe86e00 MN |
620 | return low_pfn; |
621 | } | |
622 | ||
ff9543fd MN |
623 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
624 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 625 | /* |
ff9543fd MN |
626 | * Based on information in the current compact_control, find blocks |
627 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 628 | */ |
ff9543fd MN |
629 | static void isolate_freepages(struct zone *zone, |
630 | struct compact_control *cc) | |
2fe86e00 | 631 | { |
ff9543fd MN |
632 | struct page *page; |
633 | unsigned long high_pfn, low_pfn, pfn, zone_end_pfn, end_pfn; | |
ff9543fd MN |
634 | int nr_freepages = cc->nr_freepages; |
635 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 636 | |
ff9543fd MN |
637 | /* |
638 | * Initialise the free scanner. The starting point is where we last | |
639 | * scanned from (or the end of the zone if starting). The low point | |
640 | * is the end of the pageblock the migration scanner is using. | |
641 | */ | |
642 | pfn = cc->free_pfn; | |
643 | low_pfn = cc->migrate_pfn + pageblock_nr_pages; | |
2fe86e00 | 644 | |
ff9543fd MN |
645 | /* |
646 | * Take care that if the migration scanner is at the end of the zone | |
647 | * that the free scanner does not accidentally move to the next zone | |
648 | * in the next isolation cycle. | |
649 | */ | |
650 | high_pfn = min(low_pfn, pfn); | |
2fe86e00 | 651 | |
ff9543fd | 652 | zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages; |
2fe86e00 | 653 | |
ff9543fd MN |
654 | /* |
655 | * Isolate free pages until enough are available to migrate the | |
656 | * pages on cc->migratepages. We stop searching if the migrate | |
657 | * and free page scanners meet or enough free pages are isolated. | |
658 | */ | |
659 | for (; pfn > low_pfn && cc->nr_migratepages > nr_freepages; | |
660 | pfn -= pageblock_nr_pages) { | |
661 | unsigned long isolated; | |
2fe86e00 | 662 | |
ff9543fd MN |
663 | if (!pfn_valid(pfn)) |
664 | continue; | |
2fe86e00 | 665 | |
ff9543fd MN |
666 | /* |
667 | * Check for overlapping nodes/zones. It's possible on some | |
668 | * configurations to have a setup like | |
669 | * node0 node1 node0 | |
670 | * i.e. it's possible that all pages within a zones range of | |
671 | * pages do not belong to a single zone. | |
672 | */ | |
673 | page = pfn_to_page(pfn); | |
674 | if (page_zone(page) != zone) | |
675 | continue; | |
676 | ||
677 | /* Check the block is suitable for migration */ | |
68e3e926 | 678 | if (!suitable_migration_target(page)) |
ff9543fd | 679 | continue; |
68e3e926 | 680 | |
bb13ffeb MG |
681 | /* If isolation recently failed, do not retry */ |
682 | if (!isolation_suitable(cc, page)) | |
683 | continue; | |
684 | ||
f40d1e42 | 685 | /* Found a block suitable for isolating free pages from */ |
ff9543fd | 686 | isolated = 0; |
60177d31 MG |
687 | |
688 | /* | |
689 | * As pfn may not start aligned, pfn+pageblock_nr_page | |
690 | * may cross a MAX_ORDER_NR_PAGES boundary and miss | |
691 | * a pfn_valid check. Ensure isolate_freepages_block() | |
692 | * only scans within a pageblock | |
693 | */ | |
694 | end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
695 | end_pfn = min(end_pfn, zone_end_pfn); | |
f40d1e42 MG |
696 | isolated = isolate_freepages_block(cc, pfn, end_pfn, |
697 | freelist, false); | |
698 | nr_freepages += isolated; | |
ff9543fd MN |
699 | |
700 | /* | |
701 | * Record the highest PFN we isolated pages from. When next | |
702 | * looking for free pages, the search will restart here as | |
703 | * page migration may have returned some pages to the allocator | |
704 | */ | |
c89511ab MG |
705 | if (isolated) { |
706 | cc->finished_update_free = true; | |
ff9543fd | 707 | high_pfn = max(high_pfn, pfn); |
c89511ab | 708 | } |
ff9543fd MN |
709 | } |
710 | ||
711 | /* split_free_page does not map the pages */ | |
712 | map_pages(freelist); | |
713 | ||
714 | cc->free_pfn = high_pfn; | |
715 | cc->nr_freepages = nr_freepages; | |
748446bb MG |
716 | } |
717 | ||
718 | /* | |
719 | * This is a migrate-callback that "allocates" freepages by taking pages | |
720 | * from the isolated freelists in the block we are migrating to. | |
721 | */ | |
722 | static struct page *compaction_alloc(struct page *migratepage, | |
723 | unsigned long data, | |
724 | int **result) | |
725 | { | |
726 | struct compact_control *cc = (struct compact_control *)data; | |
727 | struct page *freepage; | |
728 | ||
729 | /* Isolate free pages if necessary */ | |
730 | if (list_empty(&cc->freepages)) { | |
731 | isolate_freepages(cc->zone, cc); | |
732 | ||
733 | if (list_empty(&cc->freepages)) | |
734 | return NULL; | |
735 | } | |
736 | ||
737 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
738 | list_del(&freepage->lru); | |
739 | cc->nr_freepages--; | |
740 | ||
741 | return freepage; | |
742 | } | |
743 | ||
744 | /* | |
745 | * We cannot control nr_migratepages and nr_freepages fully when migration is | |
746 | * running as migrate_pages() has no knowledge of compact_control. When | |
747 | * migration is complete, we count the number of pages on the lists by hand. | |
748 | */ | |
749 | static void update_nr_listpages(struct compact_control *cc) | |
750 | { | |
751 | int nr_migratepages = 0; | |
752 | int nr_freepages = 0; | |
753 | struct page *page; | |
754 | ||
755 | list_for_each_entry(page, &cc->migratepages, lru) | |
756 | nr_migratepages++; | |
757 | list_for_each_entry(page, &cc->freepages, lru) | |
758 | nr_freepages++; | |
759 | ||
760 | cc->nr_migratepages = nr_migratepages; | |
761 | cc->nr_freepages = nr_freepages; | |
762 | } | |
763 | ||
ff9543fd MN |
764 | /* possible outcome of isolate_migratepages */ |
765 | typedef enum { | |
766 | ISOLATE_ABORT, /* Abort compaction now */ | |
767 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
768 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
769 | } isolate_migrate_t; | |
770 | ||
771 | /* | |
772 | * Isolate all pages that can be migrated from the block pointed to by | |
773 | * the migrate scanner within compact_control. | |
774 | */ | |
775 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
776 | struct compact_control *cc) | |
777 | { | |
778 | unsigned long low_pfn, end_pfn; | |
779 | ||
780 | /* Do not scan outside zone boundaries */ | |
781 | low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn); | |
782 | ||
783 | /* Only scan within a pageblock boundary */ | |
784 | end_pfn = ALIGN(low_pfn + pageblock_nr_pages, pageblock_nr_pages); | |
785 | ||
786 | /* Do not cross the free scanner or scan within a memory hole */ | |
787 | if (end_pfn > cc->free_pfn || !pfn_valid(low_pfn)) { | |
788 | cc->migrate_pfn = end_pfn; | |
789 | return ISOLATE_NONE; | |
790 | } | |
791 | ||
792 | /* Perform the isolation */ | |
e46a2879 | 793 | low_pfn = isolate_migratepages_range(zone, cc, low_pfn, end_pfn, false); |
e64c5237 | 794 | if (!low_pfn || cc->contended) |
ff9543fd MN |
795 | return ISOLATE_ABORT; |
796 | ||
797 | cc->migrate_pfn = low_pfn; | |
798 | ||
799 | return ISOLATE_SUCCESS; | |
800 | } | |
801 | ||
748446bb | 802 | static int compact_finished(struct zone *zone, |
5a03b051 | 803 | struct compact_control *cc) |
748446bb | 804 | { |
5a03b051 | 805 | unsigned long watermark; |
56de7263 | 806 | |
748446bb MG |
807 | if (fatal_signal_pending(current)) |
808 | return COMPACT_PARTIAL; | |
809 | ||
753341a4 | 810 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 811 | if (cc->free_pfn <= cc->migrate_pfn) { |
62997027 MG |
812 | /* |
813 | * Mark that the PG_migrate_skip information should be cleared | |
814 | * by kswapd when it goes to sleep. kswapd does not set the | |
815 | * flag itself as the decision to be clear should be directly | |
816 | * based on an allocation request. | |
817 | */ | |
818 | if (!current_is_kswapd()) | |
819 | zone->compact_blockskip_flush = true; | |
820 | ||
748446bb | 821 | return COMPACT_COMPLETE; |
bb13ffeb | 822 | } |
748446bb | 823 | |
82478fb7 JW |
824 | /* |
825 | * order == -1 is expected when compacting via | |
826 | * /proc/sys/vm/compact_memory | |
827 | */ | |
56de7263 MG |
828 | if (cc->order == -1) |
829 | return COMPACT_CONTINUE; | |
830 | ||
3957c776 MH |
831 | /* Compaction run is not finished if the watermark is not met */ |
832 | watermark = low_wmark_pages(zone); | |
833 | watermark += (1 << cc->order); | |
834 | ||
835 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
836 | return COMPACT_CONTINUE; | |
837 | ||
56de7263 | 838 | /* Direct compactor: Is a suitable page free? */ |
1fb3f8ca MG |
839 | if (cc->page) { |
840 | /* Was a suitable page captured? */ | |
841 | if (*cc->page) | |
56de7263 | 842 | return COMPACT_PARTIAL; |
1fb3f8ca MG |
843 | } else { |
844 | unsigned int order; | |
845 | for (order = cc->order; order < MAX_ORDER; order++) { | |
846 | struct free_area *area = &zone->free_area[cc->order]; | |
847 | /* Job done if page is free of the right migratetype */ | |
848 | if (!list_empty(&area->free_list[cc->migratetype])) | |
849 | return COMPACT_PARTIAL; | |
850 | ||
851 | /* Job done if allocation would set block type */ | |
852 | if (cc->order >= pageblock_order && area->nr_free) | |
853 | return COMPACT_PARTIAL; | |
854 | } | |
56de7263 MG |
855 | } |
856 | ||
748446bb MG |
857 | return COMPACT_CONTINUE; |
858 | } | |
859 | ||
3e7d3449 MG |
860 | /* |
861 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
862 | * Returns | |
863 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
864 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
865 | * COMPACT_CONTINUE - If compaction should run now | |
866 | */ | |
867 | unsigned long compaction_suitable(struct zone *zone, int order) | |
868 | { | |
869 | int fragindex; | |
870 | unsigned long watermark; | |
871 | ||
3957c776 MH |
872 | /* |
873 | * order == -1 is expected when compacting via | |
874 | * /proc/sys/vm/compact_memory | |
875 | */ | |
876 | if (order == -1) | |
877 | return COMPACT_CONTINUE; | |
878 | ||
3e7d3449 MG |
879 | /* |
880 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
881 | * This is because during migration, copies of pages need to be | |
882 | * allocated and for a short time, the footprint is higher | |
883 | */ | |
884 | watermark = low_wmark_pages(zone) + (2UL << order); | |
885 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
886 | return COMPACT_SKIPPED; | |
887 | ||
888 | /* | |
889 | * fragmentation index determines if allocation failures are due to | |
890 | * low memory or external fragmentation | |
891 | * | |
a582a738 SL |
892 | * index of -1000 implies allocations might succeed depending on |
893 | * watermarks | |
3e7d3449 MG |
894 | * index towards 0 implies failure is due to lack of memory |
895 | * index towards 1000 implies failure is due to fragmentation | |
896 | * | |
897 | * Only compact if a failure would be due to fragmentation. | |
898 | */ | |
899 | fragindex = fragmentation_index(zone, order); | |
900 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
901 | return COMPACT_SKIPPED; | |
902 | ||
a582a738 SL |
903 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
904 | 0, 0)) | |
3e7d3449 MG |
905 | return COMPACT_PARTIAL; |
906 | ||
907 | return COMPACT_CONTINUE; | |
908 | } | |
909 | ||
c8bf2d8b TR |
910 | static void compact_capture_page(struct compact_control *cc) |
911 | { | |
912 | unsigned long flags; | |
913 | int mtype, mtype_low, mtype_high; | |
914 | ||
915 | if (!cc->page || *cc->page) | |
916 | return; | |
917 | ||
918 | /* | |
919 | * For MIGRATE_MOVABLE allocations we capture a suitable page ASAP | |
920 | * regardless of the migratetype of the freelist is is captured from. | |
921 | * This is fine because the order for a high-order MIGRATE_MOVABLE | |
922 | * allocation is typically at least a pageblock size and overall | |
923 | * fragmentation is not impaired. Other allocation types must | |
924 | * capture pages from their own migratelist because otherwise they | |
925 | * could pollute other pageblocks like MIGRATE_MOVABLE with | |
926 | * difficult to move pages and making fragmentation worse overall. | |
927 | */ | |
928 | if (cc->migratetype == MIGRATE_MOVABLE) { | |
929 | mtype_low = 0; | |
930 | mtype_high = MIGRATE_PCPTYPES; | |
931 | } else { | |
932 | mtype_low = cc->migratetype; | |
933 | mtype_high = cc->migratetype + 1; | |
934 | } | |
935 | ||
936 | /* Speculatively examine the free lists without zone lock */ | |
937 | for (mtype = mtype_low; mtype < mtype_high; mtype++) { | |
938 | int order; | |
939 | for (order = cc->order; order < MAX_ORDER; order++) { | |
940 | struct page *page; | |
941 | struct free_area *area; | |
942 | area = &(cc->zone->free_area[order]); | |
943 | if (list_empty(&area->free_list[mtype])) | |
944 | continue; | |
945 | ||
946 | /* Take the lock and attempt capture of the page */ | |
947 | if (!compact_trylock_irqsave(&cc->zone->lock, &flags, cc)) | |
948 | return; | |
949 | if (!list_empty(&area->free_list[mtype])) { | |
950 | page = list_entry(area->free_list[mtype].next, | |
951 | struct page, lru); | |
952 | if (capture_free_page(page, cc->order, mtype)) { | |
953 | spin_unlock_irqrestore(&cc->zone->lock, | |
954 | flags); | |
955 | *cc->page = page; | |
956 | return; | |
957 | } | |
958 | } | |
959 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
960 | } | |
961 | } | |
962 | } | |
963 | ||
748446bb MG |
964 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
965 | { | |
966 | int ret; | |
c89511ab MG |
967 | unsigned long start_pfn = zone->zone_start_pfn; |
968 | unsigned long end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
748446bb | 969 | |
3e7d3449 MG |
970 | ret = compaction_suitable(zone, cc->order); |
971 | switch (ret) { | |
972 | case COMPACT_PARTIAL: | |
973 | case COMPACT_SKIPPED: | |
974 | /* Compaction is likely to fail */ | |
975 | return ret; | |
976 | case COMPACT_CONTINUE: | |
977 | /* Fall through to compaction */ | |
978 | ; | |
979 | } | |
980 | ||
c89511ab MG |
981 | /* |
982 | * Setup to move all movable pages to the end of the zone. Used cached | |
983 | * information on where the scanners should start but check that it | |
984 | * is initialised by ensuring the values are within zone boundaries. | |
985 | */ | |
986 | cc->migrate_pfn = zone->compact_cached_migrate_pfn; | |
987 | cc->free_pfn = zone->compact_cached_free_pfn; | |
988 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
989 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
990 | zone->compact_cached_free_pfn = cc->free_pfn; | |
991 | } | |
992 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
993 | cc->migrate_pfn = start_pfn; | |
994 | zone->compact_cached_migrate_pfn = cc->migrate_pfn; | |
995 | } | |
748446bb | 996 | |
62997027 MG |
997 | /* |
998 | * Clear pageblock skip if there were failures recently and compaction | |
999 | * is about to be retried after being deferred. kswapd does not do | |
1000 | * this reset as it'll reset the cached information when going to sleep. | |
1001 | */ | |
1002 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1003 | __reset_isolation_suitable(zone); | |
bb13ffeb | 1004 | |
748446bb MG |
1005 | migrate_prep_local(); |
1006 | ||
1007 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
1008 | unsigned long nr_migrate, nr_remaining; | |
9d502c1c | 1009 | int err; |
748446bb | 1010 | |
f9e35b3b MG |
1011 | switch (isolate_migratepages(zone, cc)) { |
1012 | case ISOLATE_ABORT: | |
1013 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1014 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1015 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1016 | goto out; |
1017 | case ISOLATE_NONE: | |
748446bb | 1018 | continue; |
f9e35b3b MG |
1019 | case ISOLATE_SUCCESS: |
1020 | ; | |
1021 | } | |
748446bb MG |
1022 | |
1023 | nr_migrate = cc->nr_migratepages; | |
9d502c1c | 1024 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
68e3e926 | 1025 | (unsigned long)cc, false, |
7b2a2d4a MG |
1026 | cc->sync ? MIGRATE_SYNC_LIGHT : MIGRATE_ASYNC, |
1027 | MR_COMPACTION); | |
748446bb MG |
1028 | update_nr_listpages(cc); |
1029 | nr_remaining = cc->nr_migratepages; | |
1030 | ||
b7aba698 MG |
1031 | trace_mm_compaction_migratepages(nr_migrate - nr_remaining, |
1032 | nr_remaining); | |
748446bb | 1033 | |
5733c7d1 | 1034 | /* Release isolated pages not migrated */ |
9d502c1c | 1035 | if (err) { |
5733c7d1 | 1036 | putback_movable_pages(&cc->migratepages); |
748446bb | 1037 | cc->nr_migratepages = 0; |
4bf2bba3 DR |
1038 | if (err == -ENOMEM) { |
1039 | ret = COMPACT_PARTIAL; | |
1040 | goto out; | |
1041 | } | |
748446bb | 1042 | } |
1fb3f8ca MG |
1043 | |
1044 | /* Capture a page now if it is a suitable size */ | |
1045 | compact_capture_page(cc); | |
748446bb MG |
1046 | } |
1047 | ||
f9e35b3b | 1048 | out: |
748446bb MG |
1049 | /* Release free pages and check accounting */ |
1050 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1051 | VM_BUG_ON(cc->nr_freepages != 0); | |
1052 | ||
1053 | return ret; | |
1054 | } | |
76ab0f53 | 1055 | |
d43a87e6 | 1056 | static unsigned long compact_zone_order(struct zone *zone, |
5a03b051 | 1057 | int order, gfp_t gfp_mask, |
1fb3f8ca MG |
1058 | bool sync, bool *contended, |
1059 | struct page **page) | |
56de7263 | 1060 | { |
e64c5237 | 1061 | unsigned long ret; |
56de7263 MG |
1062 | struct compact_control cc = { |
1063 | .nr_freepages = 0, | |
1064 | .nr_migratepages = 0, | |
1065 | .order = order, | |
1066 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1067 | .zone = zone, | |
68e3e926 | 1068 | .sync = sync, |
1fb3f8ca | 1069 | .page = page, |
56de7263 MG |
1070 | }; |
1071 | INIT_LIST_HEAD(&cc.freepages); | |
1072 | INIT_LIST_HEAD(&cc.migratepages); | |
1073 | ||
e64c5237 SL |
1074 | ret = compact_zone(zone, &cc); |
1075 | ||
1076 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1077 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1078 | ||
1079 | *contended = cc.contended; | |
1080 | return ret; | |
56de7263 MG |
1081 | } |
1082 | ||
5e771905 MG |
1083 | int sysctl_extfrag_threshold = 500; |
1084 | ||
56de7263 MG |
1085 | /** |
1086 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1087 | * @zonelist: The zonelist used for the current allocation | |
1088 | * @order: The order of the current allocation | |
1089 | * @gfp_mask: The GFP mask of the current allocation | |
1090 | * @nodemask: The allowed nodes to allocate from | |
77f1fe6b | 1091 | * @sync: Whether migration is synchronous or not |
661c4cb9 MG |
1092 | * @contended: Return value that is true if compaction was aborted due to lock contention |
1093 | * @page: Optionally capture a free page of the requested order during compaction | |
56de7263 MG |
1094 | * |
1095 | * This is the main entry point for direct page compaction. | |
1096 | */ | |
1097 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1098 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
1fb3f8ca | 1099 | bool sync, bool *contended, struct page **page) |
56de7263 MG |
1100 | { |
1101 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1102 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1103 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1104 | struct zoneref *z; |
1105 | struct zone *zone; | |
1106 | int rc = COMPACT_SKIPPED; | |
d95ea5d1 | 1107 | int alloc_flags = 0; |
56de7263 | 1108 | |
4ffb6335 | 1109 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1110 | if (!order || !may_enter_fs || !may_perform_io) |
56de7263 MG |
1111 | return rc; |
1112 | ||
1113 | count_vm_event(COMPACTSTALL); | |
1114 | ||
d95ea5d1 BZ |
1115 | #ifdef CONFIG_CMA |
1116 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1117 | alloc_flags |= ALLOC_CMA; | |
1118 | #endif | |
56de7263 MG |
1119 | /* Compact each zone in the list */ |
1120 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1121 | nodemask) { | |
56de7263 MG |
1122 | int status; |
1123 | ||
c67fe375 | 1124 | status = compact_zone_order(zone, order, gfp_mask, sync, |
1fb3f8ca | 1125 | contended, page); |
56de7263 MG |
1126 | rc = max(status, rc); |
1127 | ||
3e7d3449 | 1128 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 BZ |
1129 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
1130 | alloc_flags)) | |
56de7263 MG |
1131 | break; |
1132 | } | |
1133 | ||
1134 | return rc; | |
1135 | } | |
1136 | ||
1137 | ||
76ab0f53 | 1138 | /* Compact all zones within a node */ |
7be62de9 | 1139 | static int __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1140 | { |
1141 | int zoneid; | |
76ab0f53 MG |
1142 | struct zone *zone; |
1143 | ||
76ab0f53 | 1144 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1145 | |
1146 | zone = &pgdat->node_zones[zoneid]; | |
1147 | if (!populated_zone(zone)) | |
1148 | continue; | |
1149 | ||
7be62de9 RR |
1150 | cc->nr_freepages = 0; |
1151 | cc->nr_migratepages = 0; | |
1152 | cc->zone = zone; | |
1153 | INIT_LIST_HEAD(&cc->freepages); | |
1154 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1155 | |
aad6ec37 | 1156 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1157 | compact_zone(zone, cc); |
76ab0f53 | 1158 | |
aff62249 RR |
1159 | if (cc->order > 0) { |
1160 | int ok = zone_watermark_ok(zone, cc->order, | |
1161 | low_wmark_pages(zone), 0, 0); | |
c81758fb | 1162 | if (ok && cc->order >= zone->compact_order_failed) |
aff62249 RR |
1163 | zone->compact_order_failed = cc->order + 1; |
1164 | /* Currently async compaction is never deferred. */ | |
68e3e926 | 1165 | else if (!ok && cc->sync) |
aff62249 RR |
1166 | defer_compaction(zone, cc->order); |
1167 | } | |
1168 | ||
7be62de9 RR |
1169 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1170 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 MG |
1171 | } |
1172 | ||
1173 | return 0; | |
1174 | } | |
1175 | ||
7be62de9 RR |
1176 | int compact_pgdat(pg_data_t *pgdat, int order) |
1177 | { | |
1178 | struct compact_control cc = { | |
1179 | .order = order, | |
68e3e926 | 1180 | .sync = false, |
1fb3f8ca | 1181 | .page = NULL, |
7be62de9 RR |
1182 | }; |
1183 | ||
1184 | return __compact_pgdat(pgdat, &cc); | |
1185 | } | |
1186 | ||
1187 | static int compact_node(int nid) | |
1188 | { | |
7be62de9 RR |
1189 | struct compact_control cc = { |
1190 | .order = -1, | |
68e3e926 | 1191 | .sync = true, |
1fb3f8ca | 1192 | .page = NULL, |
7be62de9 RR |
1193 | }; |
1194 | ||
8575ec29 | 1195 | return __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1196 | } |
1197 | ||
76ab0f53 MG |
1198 | /* Compact all nodes in the system */ |
1199 | static int compact_nodes(void) | |
1200 | { | |
1201 | int nid; | |
1202 | ||
8575ec29 HD |
1203 | /* Flush pending updates to the LRU lists */ |
1204 | lru_add_drain_all(); | |
1205 | ||
76ab0f53 MG |
1206 | for_each_online_node(nid) |
1207 | compact_node(nid); | |
1208 | ||
1209 | return COMPACT_COMPLETE; | |
1210 | } | |
1211 | ||
1212 | /* The written value is actually unused, all memory is compacted */ | |
1213 | int sysctl_compact_memory; | |
1214 | ||
1215 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1216 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1217 | void __user *buffer, size_t *length, loff_t *ppos) | |
1218 | { | |
1219 | if (write) | |
1220 | return compact_nodes(); | |
1221 | ||
1222 | return 0; | |
1223 | } | |
ed4a6d7f | 1224 | |
5e771905 MG |
1225 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1226 | void __user *buffer, size_t *length, loff_t *ppos) | |
1227 | { | |
1228 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1229 | ||
1230 | return 0; | |
1231 | } | |
1232 | ||
ed4a6d7f | 1233 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
10fbcf4c KS |
1234 | ssize_t sysfs_compact_node(struct device *dev, |
1235 | struct device_attribute *attr, | |
ed4a6d7f MG |
1236 | const char *buf, size_t count) |
1237 | { | |
8575ec29 HD |
1238 | int nid = dev->id; |
1239 | ||
1240 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1241 | /* Flush pending updates to the LRU lists */ | |
1242 | lru_add_drain_all(); | |
1243 | ||
1244 | compact_node(nid); | |
1245 | } | |
ed4a6d7f MG |
1246 | |
1247 | return count; | |
1248 | } | |
10fbcf4c | 1249 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1250 | |
1251 | int compaction_register_node(struct node *node) | |
1252 | { | |
10fbcf4c | 1253 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1254 | } |
1255 | ||
1256 | void compaction_unregister_node(struct node *node) | |
1257 | { | |
10fbcf4c | 1258 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1259 | } |
1260 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1261 | |
1262 | #endif /* CONFIG_COMPACTION */ |