]>
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 | 36 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
16c4a097 JK |
37 | #ifdef CONFIG_TRACEPOINTS |
38 | static const char *const compaction_status_string[] = { | |
39 | "deferred", | |
40 | "skipped", | |
41 | "continue", | |
42 | "partial", | |
43 | "complete", | |
44 | }; | |
45 | #endif | |
ff9543fd | 46 | |
b7aba698 MG |
47 | #define CREATE_TRACE_POINTS |
48 | #include <trace/events/compaction.h> | |
49 | ||
748446bb MG |
50 | static unsigned long release_freepages(struct list_head *freelist) |
51 | { | |
52 | struct page *page, *next; | |
6bace090 | 53 | unsigned long high_pfn = 0; |
748446bb MG |
54 | |
55 | list_for_each_entry_safe(page, next, freelist, lru) { | |
6bace090 | 56 | unsigned long pfn = page_to_pfn(page); |
748446bb MG |
57 | list_del(&page->lru); |
58 | __free_page(page); | |
6bace090 VB |
59 | if (pfn > high_pfn) |
60 | high_pfn = pfn; | |
748446bb MG |
61 | } |
62 | ||
6bace090 | 63 | return high_pfn; |
748446bb MG |
64 | } |
65 | ||
ff9543fd MN |
66 | static void map_pages(struct list_head *list) |
67 | { | |
68 | struct page *page; | |
69 | ||
70 | list_for_each_entry(page, list, lru) { | |
71 | arch_alloc_page(page, 0); | |
72 | kernel_map_pages(page, 1, 1); | |
73 | } | |
74 | } | |
75 | ||
47118af0 MN |
76 | static inline bool migrate_async_suitable(int migratetype) |
77 | { | |
78 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; | |
79 | } | |
80 | ||
7d49d886 VB |
81 | /* |
82 | * Check that the whole (or subset of) a pageblock given by the interval of | |
83 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
84 | * with the migration of free compaction scanner. The scanners then need to | |
85 | * use only pfn_valid_within() check for arches that allow holes within | |
86 | * pageblocks. | |
87 | * | |
88 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
89 | * | |
90 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
91 | * i.e. it's possible that all pages within a zones range of pages do not | |
92 | * belong to a single zone. We assume that a border between node0 and node1 | |
93 | * can occur within a single pageblock, but not a node0 node1 node0 | |
94 | * interleaving within a single pageblock. It is therefore sufficient to check | |
95 | * the first and last page of a pageblock and avoid checking each individual | |
96 | * page in a pageblock. | |
97 | */ | |
98 | static struct page *pageblock_pfn_to_page(unsigned long start_pfn, | |
99 | unsigned long end_pfn, struct zone *zone) | |
100 | { | |
101 | struct page *start_page; | |
102 | struct page *end_page; | |
103 | ||
104 | /* end_pfn is one past the range we are checking */ | |
105 | end_pfn--; | |
106 | ||
107 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
108 | return NULL; | |
109 | ||
110 | start_page = pfn_to_page(start_pfn); | |
111 | ||
112 | if (page_zone(start_page) != zone) | |
113 | return NULL; | |
114 | ||
115 | end_page = pfn_to_page(end_pfn); | |
116 | ||
117 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
118 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
119 | return NULL; | |
120 | ||
121 | return start_page; | |
122 | } | |
123 | ||
bb13ffeb MG |
124 | #ifdef CONFIG_COMPACTION |
125 | /* Returns true if the pageblock should be scanned for pages to isolate. */ | |
126 | static inline bool isolation_suitable(struct compact_control *cc, | |
127 | struct page *page) | |
128 | { | |
129 | if (cc->ignore_skip_hint) | |
130 | return true; | |
131 | ||
132 | return !get_pageblock_skip(page); | |
133 | } | |
134 | ||
135 | /* | |
136 | * This function is called to clear all cached information on pageblocks that | |
137 | * should be skipped for page isolation when the migrate and free page scanner | |
138 | * meet. | |
139 | */ | |
62997027 | 140 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
141 | { |
142 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 143 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
144 | unsigned long pfn; |
145 | ||
35979ef3 DR |
146 | zone->compact_cached_migrate_pfn[0] = start_pfn; |
147 | zone->compact_cached_migrate_pfn[1] = start_pfn; | |
c89511ab | 148 | zone->compact_cached_free_pfn = end_pfn; |
62997027 | 149 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
150 | |
151 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
152 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
153 | struct page *page; | |
154 | ||
155 | cond_resched(); | |
156 | ||
157 | if (!pfn_valid(pfn)) | |
158 | continue; | |
159 | ||
160 | page = pfn_to_page(pfn); | |
161 | if (zone != page_zone(page)) | |
162 | continue; | |
163 | ||
164 | clear_pageblock_skip(page); | |
165 | } | |
166 | } | |
167 | ||
62997027 MG |
168 | void reset_isolation_suitable(pg_data_t *pgdat) |
169 | { | |
170 | int zoneid; | |
171 | ||
172 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
173 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
174 | if (!populated_zone(zone)) | |
175 | continue; | |
176 | ||
177 | /* Only flush if a full compaction finished recently */ | |
178 | if (zone->compact_blockskip_flush) | |
179 | __reset_isolation_suitable(zone); | |
180 | } | |
181 | } | |
182 | ||
bb13ffeb MG |
183 | /* |
184 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 185 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 186 | */ |
c89511ab MG |
187 | static void update_pageblock_skip(struct compact_control *cc, |
188 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 189 | bool migrate_scanner) |
bb13ffeb | 190 | { |
c89511ab | 191 | struct zone *zone = cc->zone; |
35979ef3 | 192 | unsigned long pfn; |
6815bf3f JK |
193 | |
194 | if (cc->ignore_skip_hint) | |
195 | return; | |
196 | ||
bb13ffeb MG |
197 | if (!page) |
198 | return; | |
199 | ||
35979ef3 DR |
200 | if (nr_isolated) |
201 | return; | |
202 | ||
edc2ca61 | 203 | set_pageblock_skip(page); |
c89511ab | 204 | |
35979ef3 DR |
205 | pfn = page_to_pfn(page); |
206 | ||
207 | /* Update where async and sync compaction should restart */ | |
208 | if (migrate_scanner) { | |
35979ef3 DR |
209 | if (pfn > zone->compact_cached_migrate_pfn[0]) |
210 | zone->compact_cached_migrate_pfn[0] = pfn; | |
e0b9daeb DR |
211 | if (cc->mode != MIGRATE_ASYNC && |
212 | pfn > zone->compact_cached_migrate_pfn[1]) | |
35979ef3 DR |
213 | zone->compact_cached_migrate_pfn[1] = pfn; |
214 | } else { | |
35979ef3 DR |
215 | if (pfn < zone->compact_cached_free_pfn) |
216 | zone->compact_cached_free_pfn = pfn; | |
c89511ab | 217 | } |
bb13ffeb MG |
218 | } |
219 | #else | |
220 | static inline bool isolation_suitable(struct compact_control *cc, | |
221 | struct page *page) | |
222 | { | |
223 | return true; | |
224 | } | |
225 | ||
c89511ab MG |
226 | static void update_pageblock_skip(struct compact_control *cc, |
227 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 228 | bool migrate_scanner) |
bb13ffeb MG |
229 | { |
230 | } | |
231 | #endif /* CONFIG_COMPACTION */ | |
232 | ||
8b44d279 VB |
233 | /* |
234 | * Compaction requires the taking of some coarse locks that are potentially | |
235 | * very heavily contended. For async compaction, back out if the lock cannot | |
236 | * be taken immediately. For sync compaction, spin on the lock if needed. | |
237 | * | |
238 | * Returns true if the lock is held | |
239 | * Returns false if the lock is not held and compaction should abort | |
240 | */ | |
241 | static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags, | |
242 | struct compact_control *cc) | |
2a1402aa | 243 | { |
8b44d279 VB |
244 | if (cc->mode == MIGRATE_ASYNC) { |
245 | if (!spin_trylock_irqsave(lock, *flags)) { | |
246 | cc->contended = COMPACT_CONTENDED_LOCK; | |
247 | return false; | |
248 | } | |
249 | } else { | |
250 | spin_lock_irqsave(lock, *flags); | |
251 | } | |
1f9efdef | 252 | |
8b44d279 | 253 | return true; |
2a1402aa MG |
254 | } |
255 | ||
c67fe375 MG |
256 | /* |
257 | * Compaction requires the taking of some coarse locks that are potentially | |
8b44d279 VB |
258 | * very heavily contended. The lock should be periodically unlocked to avoid |
259 | * having disabled IRQs for a long time, even when there is nobody waiting on | |
260 | * the lock. It might also be that allowing the IRQs will result in | |
261 | * need_resched() becoming true. If scheduling is needed, async compaction | |
262 | * aborts. Sync compaction schedules. | |
263 | * Either compaction type will also abort if a fatal signal is pending. | |
264 | * In either case if the lock was locked, it is dropped and not regained. | |
c67fe375 | 265 | * |
8b44d279 VB |
266 | * Returns true if compaction should abort due to fatal signal pending, or |
267 | * async compaction due to need_resched() | |
268 | * Returns false when compaction can continue (sync compaction might have | |
269 | * scheduled) | |
c67fe375 | 270 | */ |
8b44d279 VB |
271 | static bool compact_unlock_should_abort(spinlock_t *lock, |
272 | unsigned long flags, bool *locked, struct compact_control *cc) | |
c67fe375 | 273 | { |
8b44d279 VB |
274 | if (*locked) { |
275 | spin_unlock_irqrestore(lock, flags); | |
276 | *locked = false; | |
277 | } | |
1f9efdef | 278 | |
8b44d279 VB |
279 | if (fatal_signal_pending(current)) { |
280 | cc->contended = COMPACT_CONTENDED_SCHED; | |
281 | return true; | |
282 | } | |
c67fe375 | 283 | |
8b44d279 | 284 | if (need_resched()) { |
e0b9daeb | 285 | if (cc->mode == MIGRATE_ASYNC) { |
8b44d279 VB |
286 | cc->contended = COMPACT_CONTENDED_SCHED; |
287 | return true; | |
c67fe375 | 288 | } |
c67fe375 | 289 | cond_resched(); |
c67fe375 MG |
290 | } |
291 | ||
8b44d279 | 292 | return false; |
c67fe375 MG |
293 | } |
294 | ||
be976572 VB |
295 | /* |
296 | * Aside from avoiding lock contention, compaction also periodically checks | |
297 | * need_resched() and either schedules in sync compaction or aborts async | |
8b44d279 | 298 | * compaction. This is similar to what compact_unlock_should_abort() does, but |
be976572 VB |
299 | * is used where no lock is concerned. |
300 | * | |
301 | * Returns false when no scheduling was needed, or sync compaction scheduled. | |
302 | * Returns true when async compaction should abort. | |
303 | */ | |
304 | static inline bool compact_should_abort(struct compact_control *cc) | |
305 | { | |
306 | /* async compaction aborts if contended */ | |
307 | if (need_resched()) { | |
308 | if (cc->mode == MIGRATE_ASYNC) { | |
1f9efdef | 309 | cc->contended = COMPACT_CONTENDED_SCHED; |
be976572 VB |
310 | return true; |
311 | } | |
312 | ||
313 | cond_resched(); | |
314 | } | |
315 | ||
316 | return false; | |
317 | } | |
318 | ||
f40d1e42 MG |
319 | /* Returns true if the page is within a block suitable for migration to */ |
320 | static bool suitable_migration_target(struct page *page) | |
321 | { | |
7d348b9e | 322 | /* If the page is a large free page, then disallow migration */ |
99c0fd5e VB |
323 | if (PageBuddy(page)) { |
324 | /* | |
325 | * We are checking page_order without zone->lock taken. But | |
326 | * the only small danger is that we skip a potentially suitable | |
327 | * pageblock, so it's not worth to check order for valid range. | |
328 | */ | |
329 | if (page_order_unsafe(page) >= pageblock_order) | |
330 | return false; | |
331 | } | |
f40d1e42 MG |
332 | |
333 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
7d348b9e | 334 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
f40d1e42 MG |
335 | return true; |
336 | ||
337 | /* Otherwise skip the block */ | |
338 | return false; | |
339 | } | |
340 | ||
85aa125f | 341 | /* |
9e4be470 JM |
342 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
343 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
344 | * (even though it may still end up isolating some pages). | |
85aa125f | 345 | */ |
f40d1e42 | 346 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
e14c720e | 347 | unsigned long *start_pfn, |
85aa125f MN |
348 | unsigned long end_pfn, |
349 | struct list_head *freelist, | |
350 | bool strict) | |
748446bb | 351 | { |
b7aba698 | 352 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 353 | struct page *cursor, *valid_page = NULL; |
b8b2d825 | 354 | unsigned long flags = 0; |
f40d1e42 | 355 | bool locked = false; |
e14c720e | 356 | unsigned long blockpfn = *start_pfn; |
748446bb | 357 | |
748446bb MG |
358 | cursor = pfn_to_page(blockpfn); |
359 | ||
f40d1e42 | 360 | /* Isolate free pages. */ |
748446bb MG |
361 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
362 | int isolated, i; | |
363 | struct page *page = cursor; | |
364 | ||
8b44d279 VB |
365 | /* |
366 | * Periodically drop the lock (if held) regardless of its | |
367 | * contention, to give chance to IRQs. Abort if fatal signal | |
368 | * pending or async compaction detects need_resched() | |
369 | */ | |
370 | if (!(blockpfn % SWAP_CLUSTER_MAX) | |
371 | && compact_unlock_should_abort(&cc->zone->lock, flags, | |
372 | &locked, cc)) | |
373 | break; | |
374 | ||
b7aba698 | 375 | nr_scanned++; |
f40d1e42 | 376 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
377 | goto isolate_fail; |
378 | ||
bb13ffeb MG |
379 | if (!valid_page) |
380 | valid_page = page; | |
f40d1e42 | 381 | if (!PageBuddy(page)) |
2af120bc | 382 | goto isolate_fail; |
f40d1e42 MG |
383 | |
384 | /* | |
69b7189f VB |
385 | * If we already hold the lock, we can skip some rechecking. |
386 | * Note that if we hold the lock now, checked_pageblock was | |
387 | * already set in some previous iteration (or strict is true), | |
388 | * so it is correct to skip the suitable migration target | |
389 | * recheck as well. | |
f40d1e42 | 390 | */ |
69b7189f VB |
391 | if (!locked) { |
392 | /* | |
393 | * The zone lock must be held to isolate freepages. | |
394 | * Unfortunately this is a very coarse lock and can be | |
395 | * heavily contended if there are parallel allocations | |
396 | * or parallel compactions. For async compaction do not | |
397 | * spin on the lock and we acquire the lock as late as | |
398 | * possible. | |
399 | */ | |
8b44d279 VB |
400 | locked = compact_trylock_irqsave(&cc->zone->lock, |
401 | &flags, cc); | |
69b7189f VB |
402 | if (!locked) |
403 | break; | |
f40d1e42 | 404 | |
69b7189f VB |
405 | /* Recheck this is a buddy page under lock */ |
406 | if (!PageBuddy(page)) | |
407 | goto isolate_fail; | |
408 | } | |
748446bb MG |
409 | |
410 | /* Found a free page, break it into order-0 pages */ | |
411 | isolated = split_free_page(page); | |
412 | total_isolated += isolated; | |
413 | for (i = 0; i < isolated; i++) { | |
414 | list_add(&page->lru, freelist); | |
415 | page++; | |
416 | } | |
417 | ||
418 | /* If a page was split, advance to the end of it */ | |
419 | if (isolated) { | |
420 | blockpfn += isolated - 1; | |
421 | cursor += isolated - 1; | |
2af120bc | 422 | continue; |
748446bb | 423 | } |
2af120bc LA |
424 | |
425 | isolate_fail: | |
426 | if (strict) | |
427 | break; | |
428 | else | |
429 | continue; | |
430 | ||
748446bb MG |
431 | } |
432 | ||
e14c720e VB |
433 | /* Record how far we have got within the block */ |
434 | *start_pfn = blockpfn; | |
435 | ||
b7aba698 | 436 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
437 | |
438 | /* | |
439 | * If strict isolation is requested by CMA then check that all the | |
440 | * pages requested were isolated. If there were any failures, 0 is | |
441 | * returned and CMA will fail. | |
442 | */ | |
2af120bc | 443 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
444 | total_isolated = 0; |
445 | ||
446 | if (locked) | |
447 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
448 | ||
bb13ffeb MG |
449 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
450 | if (blockpfn == end_pfn) | |
edc2ca61 | 451 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 452 | |
010fc29a | 453 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 454 | if (total_isolated) |
010fc29a | 455 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
456 | return total_isolated; |
457 | } | |
458 | ||
85aa125f MN |
459 | /** |
460 | * isolate_freepages_range() - isolate free pages. | |
461 | * @start_pfn: The first PFN to start isolating. | |
462 | * @end_pfn: The one-past-last PFN. | |
463 | * | |
464 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
465 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
466 | * undo its actions and return zero. | |
467 | * | |
468 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
469 | * (which may be greater then end_pfn if end fell in a middle of | |
470 | * a free page). | |
471 | */ | |
ff9543fd | 472 | unsigned long |
bb13ffeb MG |
473 | isolate_freepages_range(struct compact_control *cc, |
474 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 475 | { |
f40d1e42 | 476 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
477 | LIST_HEAD(freelist); |
478 | ||
7d49d886 VB |
479 | pfn = start_pfn; |
480 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
481 | ||
482 | for (; pfn < end_pfn; pfn += isolated, | |
483 | block_end_pfn += pageblock_nr_pages) { | |
e14c720e VB |
484 | /* Protect pfn from changing by isolate_freepages_block */ |
485 | unsigned long isolate_start_pfn = pfn; | |
85aa125f | 486 | |
85aa125f MN |
487 | block_end_pfn = min(block_end_pfn, end_pfn); |
488 | ||
58420016 JK |
489 | /* |
490 | * pfn could pass the block_end_pfn if isolated freepage | |
491 | * is more than pageblock order. In this case, we adjust | |
492 | * scanning range to right one. | |
493 | */ | |
494 | if (pfn >= block_end_pfn) { | |
495 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
496 | block_end_pfn = min(block_end_pfn, end_pfn); | |
497 | } | |
498 | ||
7d49d886 VB |
499 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
500 | break; | |
501 | ||
e14c720e VB |
502 | isolated = isolate_freepages_block(cc, &isolate_start_pfn, |
503 | block_end_pfn, &freelist, true); | |
85aa125f MN |
504 | |
505 | /* | |
506 | * In strict mode, isolate_freepages_block() returns 0 if | |
507 | * there are any holes in the block (ie. invalid PFNs or | |
508 | * non-free pages). | |
509 | */ | |
510 | if (!isolated) | |
511 | break; | |
512 | ||
513 | /* | |
514 | * If we managed to isolate pages, it is always (1 << n) * | |
515 | * pageblock_nr_pages for some non-negative n. (Max order | |
516 | * page may span two pageblocks). | |
517 | */ | |
518 | } | |
519 | ||
520 | /* split_free_page does not map the pages */ | |
521 | map_pages(&freelist); | |
522 | ||
523 | if (pfn < end_pfn) { | |
524 | /* Loop terminated early, cleanup. */ | |
525 | release_freepages(&freelist); | |
526 | return 0; | |
527 | } | |
528 | ||
529 | /* We don't use freelists for anything. */ | |
530 | return pfn; | |
531 | } | |
532 | ||
748446bb | 533 | /* Update the number of anon and file isolated pages in the zone */ |
edc2ca61 | 534 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
748446bb MG |
535 | { |
536 | struct page *page; | |
b9e84ac1 | 537 | unsigned int count[2] = { 0, }; |
748446bb | 538 | |
edc2ca61 VB |
539 | if (list_empty(&cc->migratepages)) |
540 | return; | |
541 | ||
b9e84ac1 MK |
542 | list_for_each_entry(page, &cc->migratepages, lru) |
543 | count[!!page_is_file_cache(page)]++; | |
748446bb | 544 | |
edc2ca61 VB |
545 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
546 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
748446bb MG |
547 | } |
548 | ||
549 | /* Similar to reclaim, but different enough that they don't share logic */ | |
550 | static bool too_many_isolated(struct zone *zone) | |
551 | { | |
bc693045 | 552 | unsigned long active, inactive, isolated; |
748446bb MG |
553 | |
554 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
555 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
556 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
557 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
558 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
559 | zone_page_state(zone, NR_ISOLATED_ANON); | |
560 | ||
bc693045 | 561 | return isolated > (inactive + active) / 2; |
748446bb MG |
562 | } |
563 | ||
2fe86e00 | 564 | /** |
edc2ca61 VB |
565 | * isolate_migratepages_block() - isolate all migrate-able pages within |
566 | * a single pageblock | |
2fe86e00 | 567 | * @cc: Compaction control structure. |
edc2ca61 VB |
568 | * @low_pfn: The first PFN to isolate |
569 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock | |
570 | * @isolate_mode: Isolation mode to be used. | |
2fe86e00 MN |
571 | * |
572 | * Isolate all pages that can be migrated from the range specified by | |
edc2ca61 VB |
573 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
574 | * Returns zero if there is a fatal signal pending, otherwise PFN of the | |
575 | * first page that was not scanned (which may be both less, equal to or more | |
576 | * than end_pfn). | |
2fe86e00 | 577 | * |
edc2ca61 VB |
578 | * The pages are isolated on cc->migratepages list (not required to be empty), |
579 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field | |
580 | * is neither read nor updated. | |
748446bb | 581 | */ |
edc2ca61 VB |
582 | static unsigned long |
583 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, | |
584 | unsigned long end_pfn, isolate_mode_t isolate_mode) | |
748446bb | 585 | { |
edc2ca61 | 586 | struct zone *zone = cc->zone; |
b7aba698 | 587 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 588 | struct list_head *migratelist = &cc->migratepages; |
fa9add64 | 589 | struct lruvec *lruvec; |
b8b2d825 | 590 | unsigned long flags = 0; |
2a1402aa | 591 | bool locked = false; |
bb13ffeb | 592 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 593 | |
748446bb MG |
594 | /* |
595 | * Ensure that there are not too many pages isolated from the LRU | |
596 | * list by either parallel reclaimers or compaction. If there are, | |
597 | * delay for some time until fewer pages are isolated | |
598 | */ | |
599 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 600 | /* async migration should just abort */ |
e0b9daeb | 601 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 602 | return 0; |
f9e35b3b | 603 | |
748446bb MG |
604 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
605 | ||
606 | if (fatal_signal_pending(current)) | |
2fe86e00 | 607 | return 0; |
748446bb MG |
608 | } |
609 | ||
be976572 VB |
610 | if (compact_should_abort(cc)) |
611 | return 0; | |
aeef4b83 | 612 | |
748446bb | 613 | /* Time to isolate some pages for migration */ |
748446bb | 614 | for (; low_pfn < end_pfn; low_pfn++) { |
8b44d279 VB |
615 | /* |
616 | * Periodically drop the lock (if held) regardless of its | |
617 | * contention, to give chance to IRQs. Abort async compaction | |
618 | * if contended. | |
619 | */ | |
620 | if (!(low_pfn % SWAP_CLUSTER_MAX) | |
621 | && compact_unlock_should_abort(&zone->lru_lock, flags, | |
622 | &locked, cc)) | |
623 | break; | |
c67fe375 | 624 | |
748446bb MG |
625 | if (!pfn_valid_within(low_pfn)) |
626 | continue; | |
b7aba698 | 627 | nr_scanned++; |
748446bb | 628 | |
748446bb | 629 | page = pfn_to_page(low_pfn); |
dc908600 | 630 | |
bb13ffeb MG |
631 | if (!valid_page) |
632 | valid_page = page; | |
633 | ||
6c14466c | 634 | /* |
99c0fd5e VB |
635 | * Skip if free. We read page order here without zone lock |
636 | * which is generally unsafe, but the race window is small and | |
637 | * the worst thing that can happen is that we skip some | |
638 | * potential isolation targets. | |
6c14466c | 639 | */ |
99c0fd5e VB |
640 | if (PageBuddy(page)) { |
641 | unsigned long freepage_order = page_order_unsafe(page); | |
642 | ||
643 | /* | |
644 | * Without lock, we cannot be sure that what we got is | |
645 | * a valid page order. Consider only values in the | |
646 | * valid order range to prevent low_pfn overflow. | |
647 | */ | |
648 | if (freepage_order > 0 && freepage_order < MAX_ORDER) | |
649 | low_pfn += (1UL << freepage_order) - 1; | |
748446bb | 650 | continue; |
99c0fd5e | 651 | } |
748446bb | 652 | |
bf6bddf1 RA |
653 | /* |
654 | * Check may be lockless but that's ok as we recheck later. | |
655 | * It's possible to migrate LRU pages and balloon pages | |
656 | * Skip any other type of page | |
657 | */ | |
658 | if (!PageLRU(page)) { | |
659 | if (unlikely(balloon_page_movable(page))) { | |
d6d86c0a | 660 | if (balloon_page_isolate(page)) { |
bf6bddf1 | 661 | /* Successfully isolated */ |
b6c75016 | 662 | goto isolate_success; |
bf6bddf1 RA |
663 | } |
664 | } | |
bc835011 | 665 | continue; |
bf6bddf1 | 666 | } |
bc835011 AA |
667 | |
668 | /* | |
2a1402aa MG |
669 | * PageLRU is set. lru_lock normally excludes isolation |
670 | * splitting and collapsing (collapsing has already happened | |
671 | * if PageLRU is set) but the lock is not necessarily taken | |
672 | * here and it is wasteful to take it just to check transhuge. | |
673 | * Check TransHuge without lock and skip the whole pageblock if | |
674 | * it's either a transhuge or hugetlbfs page, as calling | |
675 | * compound_order() without preventing THP from splitting the | |
676 | * page underneath us may return surprising results. | |
bc835011 | 677 | */ |
2a1402aa MG |
678 | if (PageTransHuge(page)) { |
679 | if (!locked) | |
edc2ca61 VB |
680 | low_pfn = ALIGN(low_pfn + 1, |
681 | pageblock_nr_pages) - 1; | |
682 | else | |
683 | low_pfn += (1 << compound_order(page)) - 1; | |
684 | ||
2a1402aa MG |
685 | continue; |
686 | } | |
687 | ||
119d6d59 DR |
688 | /* |
689 | * Migration will fail if an anonymous page is pinned in memory, | |
690 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
691 | * admittedly racy check. | |
692 | */ | |
693 | if (!page_mapping(page) && | |
694 | page_count(page) > page_mapcount(page)) | |
695 | continue; | |
696 | ||
69b7189f VB |
697 | /* If we already hold the lock, we can skip some rechecking */ |
698 | if (!locked) { | |
8b44d279 VB |
699 | locked = compact_trylock_irqsave(&zone->lru_lock, |
700 | &flags, cc); | |
69b7189f VB |
701 | if (!locked) |
702 | break; | |
2a1402aa | 703 | |
69b7189f VB |
704 | /* Recheck PageLRU and PageTransHuge under lock */ |
705 | if (!PageLRU(page)) | |
706 | continue; | |
707 | if (PageTransHuge(page)) { | |
708 | low_pfn += (1 << compound_order(page)) - 1; | |
709 | continue; | |
710 | } | |
bc835011 AA |
711 | } |
712 | ||
fa9add64 HD |
713 | lruvec = mem_cgroup_page_lruvec(page, zone); |
714 | ||
748446bb | 715 | /* Try isolate the page */ |
edc2ca61 | 716 | if (__isolate_lru_page(page, isolate_mode) != 0) |
748446bb MG |
717 | continue; |
718 | ||
309381fe | 719 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 720 | |
748446bb | 721 | /* Successfully isolated */ |
fa9add64 | 722 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
723 | |
724 | isolate_success: | |
748446bb | 725 | list_add(&page->lru, migratelist); |
748446bb | 726 | cc->nr_migratepages++; |
b7aba698 | 727 | nr_isolated++; |
748446bb MG |
728 | |
729 | /* Avoid isolating too much */ | |
31b8384a HD |
730 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
731 | ++low_pfn; | |
748446bb | 732 | break; |
31b8384a | 733 | } |
748446bb MG |
734 | } |
735 | ||
99c0fd5e VB |
736 | /* |
737 | * The PageBuddy() check could have potentially brought us outside | |
738 | * the range to be scanned. | |
739 | */ | |
740 | if (unlikely(low_pfn > end_pfn)) | |
741 | low_pfn = end_pfn; | |
742 | ||
c67fe375 MG |
743 | if (locked) |
744 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 745 | |
50b5b094 VB |
746 | /* |
747 | * Update the pageblock-skip information and cached scanner pfn, | |
748 | * if the whole pageblock was scanned without isolating any page. | |
50b5b094 | 749 | */ |
35979ef3 | 750 | if (low_pfn == end_pfn) |
edc2ca61 | 751 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 752 | |
b7aba698 MG |
753 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
754 | ||
010fc29a | 755 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 756 | if (nr_isolated) |
010fc29a | 757 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 758 | |
2fe86e00 MN |
759 | return low_pfn; |
760 | } | |
761 | ||
edc2ca61 VB |
762 | /** |
763 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range | |
764 | * @cc: Compaction control structure. | |
765 | * @start_pfn: The first PFN to start isolating. | |
766 | * @end_pfn: The one-past-last PFN. | |
767 | * | |
768 | * Returns zero if isolation fails fatally due to e.g. pending signal. | |
769 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
770 | * (which may be greater than end_pfn if end fell in a middle of a THP page). | |
771 | */ | |
772 | unsigned long | |
773 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, | |
774 | unsigned long end_pfn) | |
775 | { | |
776 | unsigned long pfn, block_end_pfn; | |
777 | ||
778 | /* Scan block by block. First and last block may be incomplete */ | |
779 | pfn = start_pfn; | |
780 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
781 | ||
782 | for (; pfn < end_pfn; pfn = block_end_pfn, | |
783 | block_end_pfn += pageblock_nr_pages) { | |
784 | ||
785 | block_end_pfn = min(block_end_pfn, end_pfn); | |
786 | ||
7d49d886 | 787 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
edc2ca61 VB |
788 | continue; |
789 | ||
790 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, | |
791 | ISOLATE_UNEVICTABLE); | |
792 | ||
793 | /* | |
794 | * In case of fatal failure, release everything that might | |
795 | * have been isolated in the previous iteration, and signal | |
796 | * the failure back to caller. | |
797 | */ | |
798 | if (!pfn) { | |
799 | putback_movable_pages(&cc->migratepages); | |
800 | cc->nr_migratepages = 0; | |
801 | break; | |
802 | } | |
6ea41c0c JK |
803 | |
804 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) | |
805 | break; | |
edc2ca61 VB |
806 | } |
807 | acct_isolated(cc->zone, cc); | |
808 | ||
809 | return pfn; | |
810 | } | |
811 | ||
ff9543fd MN |
812 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
813 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 814 | /* |
ff9543fd MN |
815 | * Based on information in the current compact_control, find blocks |
816 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 817 | */ |
edc2ca61 | 818 | static void isolate_freepages(struct compact_control *cc) |
2fe86e00 | 819 | { |
edc2ca61 | 820 | struct zone *zone = cc->zone; |
ff9543fd | 821 | struct page *page; |
c96b9e50 | 822 | unsigned long block_start_pfn; /* start of current pageblock */ |
e14c720e | 823 | unsigned long isolate_start_pfn; /* exact pfn we start at */ |
c96b9e50 VB |
824 | unsigned long block_end_pfn; /* end of current pageblock */ |
825 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd MN |
826 | int nr_freepages = cc->nr_freepages; |
827 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 828 | |
ff9543fd MN |
829 | /* |
830 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 | 831 | * successfully isolated from, zone-cached value, or the end of the |
e14c720e VB |
832 | * zone when isolating for the first time. For looping we also need |
833 | * this pfn aligned down to the pageblock boundary, because we do | |
c96b9e50 VB |
834 | * block_start_pfn -= pageblock_nr_pages in the for loop. |
835 | * For ending point, take care when isolating in last pageblock of a | |
836 | * a zone which ends in the middle of a pageblock. | |
49e068f0 VB |
837 | * The low boundary is the end of the pageblock the migration scanner |
838 | * is using. | |
ff9543fd | 839 | */ |
e14c720e | 840 | isolate_start_pfn = cc->free_pfn; |
c96b9e50 VB |
841 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
842 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, | |
843 | zone_end_pfn(zone)); | |
7ed695e0 | 844 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 845 | |
ff9543fd MN |
846 | /* |
847 | * Isolate free pages until enough are available to migrate the | |
848 | * pages on cc->migratepages. We stop searching if the migrate | |
849 | * and free page scanners meet or enough free pages are isolated. | |
850 | */ | |
c96b9e50 VB |
851 | for (; block_start_pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
852 | block_end_pfn = block_start_pfn, | |
e14c720e VB |
853 | block_start_pfn -= pageblock_nr_pages, |
854 | isolate_start_pfn = block_start_pfn) { | |
ff9543fd | 855 | unsigned long isolated; |
2fe86e00 | 856 | |
f6ea3adb DR |
857 | /* |
858 | * This can iterate a massively long zone without finding any | |
859 | * suitable migration targets, so periodically check if we need | |
be976572 | 860 | * to schedule, or even abort async compaction. |
f6ea3adb | 861 | */ |
be976572 VB |
862 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
863 | && compact_should_abort(cc)) | |
864 | break; | |
f6ea3adb | 865 | |
7d49d886 VB |
866 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
867 | zone); | |
868 | if (!page) | |
ff9543fd MN |
869 | continue; |
870 | ||
871 | /* Check the block is suitable for migration */ | |
68e3e926 | 872 | if (!suitable_migration_target(page)) |
ff9543fd | 873 | continue; |
68e3e926 | 874 | |
bb13ffeb MG |
875 | /* If isolation recently failed, do not retry */ |
876 | if (!isolation_suitable(cc, page)) | |
877 | continue; | |
878 | ||
e14c720e VB |
879 | /* Found a block suitable for isolating free pages from. */ |
880 | isolated = isolate_freepages_block(cc, &isolate_start_pfn, | |
c96b9e50 | 881 | block_end_pfn, freelist, false); |
f40d1e42 | 882 | nr_freepages += isolated; |
ff9543fd | 883 | |
e14c720e VB |
884 | /* |
885 | * Remember where the free scanner should restart next time, | |
886 | * which is where isolate_freepages_block() left off. | |
887 | * But if it scanned the whole pageblock, isolate_start_pfn | |
888 | * now points at block_end_pfn, which is the start of the next | |
889 | * pageblock. | |
890 | * In that case we will however want to restart at the start | |
891 | * of the previous pageblock. | |
892 | */ | |
893 | cc->free_pfn = (isolate_start_pfn < block_end_pfn) ? | |
894 | isolate_start_pfn : | |
895 | block_start_pfn - pageblock_nr_pages; | |
896 | ||
be976572 VB |
897 | /* |
898 | * isolate_freepages_block() might have aborted due to async | |
899 | * compaction being contended | |
900 | */ | |
901 | if (cc->contended) | |
902 | break; | |
ff9543fd MN |
903 | } |
904 | ||
905 | /* split_free_page does not map the pages */ | |
906 | map_pages(freelist); | |
907 | ||
7ed695e0 VB |
908 | /* |
909 | * If we crossed the migrate scanner, we want to keep it that way | |
910 | * so that compact_finished() may detect this | |
911 | */ | |
c96b9e50 | 912 | if (block_start_pfn < low_pfn) |
e9ade569 | 913 | cc->free_pfn = cc->migrate_pfn; |
c96b9e50 | 914 | |
ff9543fd | 915 | cc->nr_freepages = nr_freepages; |
748446bb MG |
916 | } |
917 | ||
918 | /* | |
919 | * This is a migrate-callback that "allocates" freepages by taking pages | |
920 | * from the isolated freelists in the block we are migrating to. | |
921 | */ | |
922 | static struct page *compaction_alloc(struct page *migratepage, | |
923 | unsigned long data, | |
924 | int **result) | |
925 | { | |
926 | struct compact_control *cc = (struct compact_control *)data; | |
927 | struct page *freepage; | |
928 | ||
be976572 VB |
929 | /* |
930 | * Isolate free pages if necessary, and if we are not aborting due to | |
931 | * contention. | |
932 | */ | |
748446bb | 933 | if (list_empty(&cc->freepages)) { |
be976572 | 934 | if (!cc->contended) |
edc2ca61 | 935 | isolate_freepages(cc); |
748446bb MG |
936 | |
937 | if (list_empty(&cc->freepages)) | |
938 | return NULL; | |
939 | } | |
940 | ||
941 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
942 | list_del(&freepage->lru); | |
943 | cc->nr_freepages--; | |
944 | ||
945 | return freepage; | |
946 | } | |
947 | ||
948 | /* | |
d53aea3d DR |
949 | * This is a migrate-callback that "frees" freepages back to the isolated |
950 | * freelist. All pages on the freelist are from the same zone, so there is no | |
951 | * special handling needed for NUMA. | |
952 | */ | |
953 | static void compaction_free(struct page *page, unsigned long data) | |
954 | { | |
955 | struct compact_control *cc = (struct compact_control *)data; | |
956 | ||
957 | list_add(&page->lru, &cc->freepages); | |
958 | cc->nr_freepages++; | |
959 | } | |
960 | ||
ff9543fd MN |
961 | /* possible outcome of isolate_migratepages */ |
962 | typedef enum { | |
963 | ISOLATE_ABORT, /* Abort compaction now */ | |
964 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
965 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
966 | } isolate_migrate_t; | |
967 | ||
968 | /* | |
edc2ca61 VB |
969 | * Isolate all pages that can be migrated from the first suitable block, |
970 | * starting at the block pointed to by the migrate scanner pfn within | |
971 | * compact_control. | |
ff9543fd MN |
972 | */ |
973 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
974 | struct compact_control *cc) | |
975 | { | |
976 | unsigned long low_pfn, end_pfn; | |
edc2ca61 VB |
977 | struct page *page; |
978 | const isolate_mode_t isolate_mode = | |
979 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); | |
ff9543fd | 980 | |
edc2ca61 VB |
981 | /* |
982 | * Start at where we last stopped, or beginning of the zone as | |
983 | * initialized by compact_zone() | |
984 | */ | |
985 | low_pfn = cc->migrate_pfn; | |
ff9543fd MN |
986 | |
987 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 988 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd | 989 | |
edc2ca61 VB |
990 | /* |
991 | * Iterate over whole pageblocks until we find the first suitable. | |
992 | * Do not cross the free scanner. | |
993 | */ | |
994 | for (; end_pfn <= cc->free_pfn; | |
995 | low_pfn = end_pfn, end_pfn += pageblock_nr_pages) { | |
ff9543fd | 996 | |
edc2ca61 VB |
997 | /* |
998 | * This can potentially iterate a massively long zone with | |
999 | * many pageblocks unsuitable, so periodically check if we | |
1000 | * need to schedule, or even abort async compaction. | |
1001 | */ | |
1002 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) | |
1003 | && compact_should_abort(cc)) | |
1004 | break; | |
ff9543fd | 1005 | |
7d49d886 VB |
1006 | page = pageblock_pfn_to_page(low_pfn, end_pfn, zone); |
1007 | if (!page) | |
edc2ca61 VB |
1008 | continue; |
1009 | ||
edc2ca61 VB |
1010 | /* If isolation recently failed, do not retry */ |
1011 | if (!isolation_suitable(cc, page)) | |
1012 | continue; | |
1013 | ||
1014 | /* | |
1015 | * For async compaction, also only scan in MOVABLE blocks. | |
1016 | * Async compaction is optimistic to see if the minimum amount | |
1017 | * of work satisfies the allocation. | |
1018 | */ | |
1019 | if (cc->mode == MIGRATE_ASYNC && | |
1020 | !migrate_async_suitable(get_pageblock_migratetype(page))) | |
1021 | continue; | |
1022 | ||
1023 | /* Perform the isolation */ | |
1024 | low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn, | |
1025 | isolate_mode); | |
1026 | ||
1027 | if (!low_pfn || cc->contended) | |
1028 | return ISOLATE_ABORT; | |
1029 | ||
1030 | /* | |
1031 | * Either we isolated something and proceed with migration. Or | |
1032 | * we failed and compact_zone should decide if we should | |
1033 | * continue or not. | |
1034 | */ | |
1035 | break; | |
1036 | } | |
1037 | ||
1038 | acct_isolated(zone, cc); | |
1d5bfe1f VB |
1039 | /* |
1040 | * Record where migration scanner will be restarted. If we end up in | |
1041 | * the same pageblock as the free scanner, make the scanners fully | |
1042 | * meet so that compact_finished() terminates compaction. | |
1043 | */ | |
1044 | cc->migrate_pfn = (end_pfn <= cc->free_pfn) ? low_pfn : cc->free_pfn; | |
ff9543fd | 1045 | |
edc2ca61 | 1046 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
ff9543fd MN |
1047 | } |
1048 | ||
6d7ce559 DR |
1049 | static int compact_finished(struct zone *zone, struct compact_control *cc, |
1050 | const int migratetype) | |
748446bb | 1051 | { |
8fb74b9f | 1052 | unsigned int order; |
5a03b051 | 1053 | unsigned long watermark; |
56de7263 | 1054 | |
be976572 | 1055 | if (cc->contended || fatal_signal_pending(current)) |
748446bb MG |
1056 | return COMPACT_PARTIAL; |
1057 | ||
753341a4 | 1058 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 1059 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 | 1060 | /* Let the next compaction start anew. */ |
35979ef3 DR |
1061 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
1062 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
55b7c4c9 VB |
1063 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
1064 | ||
62997027 MG |
1065 | /* |
1066 | * Mark that the PG_migrate_skip information should be cleared | |
1067 | * by kswapd when it goes to sleep. kswapd does not set the | |
1068 | * flag itself as the decision to be clear should be directly | |
1069 | * based on an allocation request. | |
1070 | */ | |
1071 | if (!current_is_kswapd()) | |
1072 | zone->compact_blockskip_flush = true; | |
1073 | ||
748446bb | 1074 | return COMPACT_COMPLETE; |
bb13ffeb | 1075 | } |
748446bb | 1076 | |
82478fb7 JW |
1077 | /* |
1078 | * order == -1 is expected when compacting via | |
1079 | * /proc/sys/vm/compact_memory | |
1080 | */ | |
56de7263 MG |
1081 | if (cc->order == -1) |
1082 | return COMPACT_CONTINUE; | |
1083 | ||
3957c776 MH |
1084 | /* Compaction run is not finished if the watermark is not met */ |
1085 | watermark = low_wmark_pages(zone); | |
3957c776 | 1086 | |
ebff3980 VB |
1087 | if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx, |
1088 | cc->alloc_flags)) | |
3957c776 MH |
1089 | return COMPACT_CONTINUE; |
1090 | ||
56de7263 | 1091 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
1092 | for (order = cc->order; order < MAX_ORDER; order++) { |
1093 | struct free_area *area = &zone->free_area[order]; | |
1094 | ||
1095 | /* Job done if page is free of the right migratetype */ | |
6d7ce559 | 1096 | if (!list_empty(&area->free_list[migratetype])) |
8fb74b9f MG |
1097 | return COMPACT_PARTIAL; |
1098 | ||
1099 | /* Job done if allocation would set block type */ | |
1100 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
1101 | return COMPACT_PARTIAL; |
1102 | } | |
1103 | ||
748446bb MG |
1104 | return COMPACT_CONTINUE; |
1105 | } | |
1106 | ||
3e7d3449 MG |
1107 | /* |
1108 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
1109 | * Returns | |
1110 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
1111 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
1112 | * COMPACT_CONTINUE - If compaction should run now | |
1113 | */ | |
ebff3980 VB |
1114 | unsigned long compaction_suitable(struct zone *zone, int order, |
1115 | int alloc_flags, int classzone_idx) | |
3e7d3449 MG |
1116 | { |
1117 | int fragindex; | |
1118 | unsigned long watermark; | |
1119 | ||
3957c776 MH |
1120 | /* |
1121 | * order == -1 is expected when compacting via | |
1122 | * /proc/sys/vm/compact_memory | |
1123 | */ | |
1124 | if (order == -1) | |
1125 | return COMPACT_CONTINUE; | |
1126 | ||
ebff3980 VB |
1127 | watermark = low_wmark_pages(zone); |
1128 | /* | |
1129 | * If watermarks for high-order allocation are already met, there | |
1130 | * should be no need for compaction at all. | |
1131 | */ | |
1132 | if (zone_watermark_ok(zone, order, watermark, classzone_idx, | |
1133 | alloc_flags)) | |
1134 | return COMPACT_PARTIAL; | |
1135 | ||
3e7d3449 MG |
1136 | /* |
1137 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
1138 | * This is because during migration, copies of pages need to be | |
1139 | * allocated and for a short time, the footprint is higher | |
1140 | */ | |
ebff3980 VB |
1141 | watermark += (2UL << order); |
1142 | if (!zone_watermark_ok(zone, 0, watermark, classzone_idx, alloc_flags)) | |
3e7d3449 MG |
1143 | return COMPACT_SKIPPED; |
1144 | ||
1145 | /* | |
1146 | * fragmentation index determines if allocation failures are due to | |
1147 | * low memory or external fragmentation | |
1148 | * | |
ebff3980 VB |
1149 | * index of -1000 would imply allocations might succeed depending on |
1150 | * watermarks, but we already failed the high-order watermark check | |
3e7d3449 MG |
1151 | * index towards 0 implies failure is due to lack of memory |
1152 | * index towards 1000 implies failure is due to fragmentation | |
1153 | * | |
1154 | * Only compact if a failure would be due to fragmentation. | |
1155 | */ | |
1156 | fragindex = fragmentation_index(zone, order); | |
1157 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
1158 | return COMPACT_SKIPPED; | |
1159 | ||
3e7d3449 MG |
1160 | return COMPACT_CONTINUE; |
1161 | } | |
1162 | ||
748446bb MG |
1163 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
1164 | { | |
1165 | int ret; | |
c89511ab | 1166 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 1167 | unsigned long end_pfn = zone_end_pfn(zone); |
6d7ce559 | 1168 | const int migratetype = gfpflags_to_migratetype(cc->gfp_mask); |
e0b9daeb | 1169 | const bool sync = cc->mode != MIGRATE_ASYNC; |
fdaf7f5c | 1170 | unsigned long last_migrated_pfn = 0; |
748446bb | 1171 | |
ebff3980 VB |
1172 | ret = compaction_suitable(zone, cc->order, cc->alloc_flags, |
1173 | cc->classzone_idx); | |
3e7d3449 MG |
1174 | switch (ret) { |
1175 | case COMPACT_PARTIAL: | |
1176 | case COMPACT_SKIPPED: | |
1177 | /* Compaction is likely to fail */ | |
1178 | return ret; | |
1179 | case COMPACT_CONTINUE: | |
1180 | /* Fall through to compaction */ | |
1181 | ; | |
1182 | } | |
1183 | ||
d3132e4b VB |
1184 | /* |
1185 | * Clear pageblock skip if there were failures recently and compaction | |
1186 | * is about to be retried after being deferred. kswapd does not do | |
1187 | * this reset as it'll reset the cached information when going to sleep. | |
1188 | */ | |
1189 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1190 | __reset_isolation_suitable(zone); | |
1191 | ||
c89511ab MG |
1192 | /* |
1193 | * Setup to move all movable pages to the end of the zone. Used cached | |
1194 | * information on where the scanners should start but check that it | |
1195 | * is initialised by ensuring the values are within zone boundaries. | |
1196 | */ | |
e0b9daeb | 1197 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
c89511ab MG |
1198 | cc->free_pfn = zone->compact_cached_free_pfn; |
1199 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
1200 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
1201 | zone->compact_cached_free_pfn = cc->free_pfn; | |
1202 | } | |
1203 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1204 | cc->migrate_pfn = start_pfn; | |
35979ef3 DR |
1205 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
1206 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
c89511ab | 1207 | } |
748446bb | 1208 | |
16c4a097 JK |
1209 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, |
1210 | cc->free_pfn, end_pfn, sync); | |
0eb927c0 | 1211 | |
748446bb MG |
1212 | migrate_prep_local(); |
1213 | ||
6d7ce559 DR |
1214 | while ((ret = compact_finished(zone, cc, migratetype)) == |
1215 | COMPACT_CONTINUE) { | |
9d502c1c | 1216 | int err; |
fdaf7f5c | 1217 | unsigned long isolate_start_pfn = cc->migrate_pfn; |
748446bb | 1218 | |
f9e35b3b MG |
1219 | switch (isolate_migratepages(zone, cc)) { |
1220 | case ISOLATE_ABORT: | |
1221 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1222 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1223 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1224 | goto out; |
1225 | case ISOLATE_NONE: | |
fdaf7f5c VB |
1226 | /* |
1227 | * We haven't isolated and migrated anything, but | |
1228 | * there might still be unflushed migrations from | |
1229 | * previous cc->order aligned block. | |
1230 | */ | |
1231 | goto check_drain; | |
f9e35b3b MG |
1232 | case ISOLATE_SUCCESS: |
1233 | ; | |
1234 | } | |
748446bb | 1235 | |
d53aea3d | 1236 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 1237 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 1238 | MR_COMPACTION); |
748446bb | 1239 | |
f8c9301f VB |
1240 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
1241 | &cc->migratepages); | |
748446bb | 1242 | |
f8c9301f VB |
1243 | /* All pages were either migrated or will be released */ |
1244 | cc->nr_migratepages = 0; | |
9d502c1c | 1245 | if (err) { |
5733c7d1 | 1246 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
1247 | /* |
1248 | * migrate_pages() may return -ENOMEM when scanners meet | |
1249 | * and we want compact_finished() to detect it | |
1250 | */ | |
1251 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1252 | ret = COMPACT_PARTIAL; |
1253 | goto out; | |
1254 | } | |
748446bb | 1255 | } |
fdaf7f5c VB |
1256 | |
1257 | /* | |
1258 | * Record where we could have freed pages by migration and not | |
1259 | * yet flushed them to buddy allocator. We use the pfn that | |
1260 | * isolate_migratepages() started from in this loop iteration | |
1261 | * - this is the lowest page that could have been isolated and | |
1262 | * then freed by migration. | |
1263 | */ | |
1264 | if (!last_migrated_pfn) | |
1265 | last_migrated_pfn = isolate_start_pfn; | |
1266 | ||
1267 | check_drain: | |
1268 | /* | |
1269 | * Has the migration scanner moved away from the previous | |
1270 | * cc->order aligned block where we migrated from? If yes, | |
1271 | * flush the pages that were freed, so that they can merge and | |
1272 | * compact_finished() can detect immediately if allocation | |
1273 | * would succeed. | |
1274 | */ | |
1275 | if (cc->order > 0 && last_migrated_pfn) { | |
1276 | int cpu; | |
1277 | unsigned long current_block_start = | |
1278 | cc->migrate_pfn & ~((1UL << cc->order) - 1); | |
1279 | ||
1280 | if (last_migrated_pfn < current_block_start) { | |
1281 | cpu = get_cpu(); | |
1282 | lru_add_drain_cpu(cpu); | |
1283 | drain_local_pages(zone); | |
1284 | put_cpu(); | |
1285 | /* No more flushing until we migrate again */ | |
1286 | last_migrated_pfn = 0; | |
1287 | } | |
1288 | } | |
1289 | ||
748446bb MG |
1290 | } |
1291 | ||
f9e35b3b | 1292 | out: |
6bace090 VB |
1293 | /* |
1294 | * Release free pages and update where the free scanner should restart, | |
1295 | * so we don't leave any returned pages behind in the next attempt. | |
1296 | */ | |
1297 | if (cc->nr_freepages > 0) { | |
1298 | unsigned long free_pfn = release_freepages(&cc->freepages); | |
1299 | ||
1300 | cc->nr_freepages = 0; | |
1301 | VM_BUG_ON(free_pfn == 0); | |
1302 | /* The cached pfn is always the first in a pageblock */ | |
1303 | free_pfn &= ~(pageblock_nr_pages-1); | |
1304 | /* | |
1305 | * Only go back, not forward. The cached pfn might have been | |
1306 | * already reset to zone end in compact_finished() | |
1307 | */ | |
1308 | if (free_pfn > zone->compact_cached_free_pfn) | |
1309 | zone->compact_cached_free_pfn = free_pfn; | |
1310 | } | |
748446bb | 1311 | |
16c4a097 JK |
1312 | trace_mm_compaction_end(start_pfn, cc->migrate_pfn, |
1313 | cc->free_pfn, end_pfn, sync, ret); | |
0eb927c0 | 1314 | |
748446bb MG |
1315 | return ret; |
1316 | } | |
76ab0f53 | 1317 | |
e0b9daeb | 1318 | static unsigned long compact_zone_order(struct zone *zone, int order, |
ebff3980 VB |
1319 | gfp_t gfp_mask, enum migrate_mode mode, int *contended, |
1320 | int alloc_flags, int classzone_idx) | |
56de7263 | 1321 | { |
e64c5237 | 1322 | unsigned long ret; |
56de7263 MG |
1323 | struct compact_control cc = { |
1324 | .nr_freepages = 0, | |
1325 | .nr_migratepages = 0, | |
1326 | .order = order, | |
6d7ce559 | 1327 | .gfp_mask = gfp_mask, |
56de7263 | 1328 | .zone = zone, |
e0b9daeb | 1329 | .mode = mode, |
ebff3980 VB |
1330 | .alloc_flags = alloc_flags, |
1331 | .classzone_idx = classzone_idx, | |
56de7263 MG |
1332 | }; |
1333 | INIT_LIST_HEAD(&cc.freepages); | |
1334 | INIT_LIST_HEAD(&cc.migratepages); | |
1335 | ||
e64c5237 SL |
1336 | ret = compact_zone(zone, &cc); |
1337 | ||
1338 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1339 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1340 | ||
1341 | *contended = cc.contended; | |
1342 | return ret; | |
56de7263 MG |
1343 | } |
1344 | ||
5e771905 MG |
1345 | int sysctl_extfrag_threshold = 500; |
1346 | ||
56de7263 MG |
1347 | /** |
1348 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
56de7263 | 1349 | * @gfp_mask: The GFP mask of the current allocation |
1a6d53a1 VB |
1350 | * @order: The order of the current allocation |
1351 | * @alloc_flags: The allocation flags of the current allocation | |
1352 | * @ac: The context of current allocation | |
e0b9daeb | 1353 | * @mode: The migration mode for async, sync light, or sync migration |
1f9efdef VB |
1354 | * @contended: Return value that determines if compaction was aborted due to |
1355 | * need_resched() or lock contention | |
56de7263 MG |
1356 | * |
1357 | * This is the main entry point for direct page compaction. | |
1358 | */ | |
1a6d53a1 VB |
1359 | unsigned long try_to_compact_pages(gfp_t gfp_mask, unsigned int order, |
1360 | int alloc_flags, const struct alloc_context *ac, | |
1361 | enum migrate_mode mode, int *contended) | |
56de7263 | 1362 | { |
56de7263 MG |
1363 | int may_enter_fs = gfp_mask & __GFP_FS; |
1364 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1365 | struct zoneref *z; |
1366 | struct zone *zone; | |
53853e2d | 1367 | int rc = COMPACT_DEFERRED; |
1f9efdef VB |
1368 | int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ |
1369 | ||
1370 | *contended = COMPACT_CONTENDED_NONE; | |
56de7263 | 1371 | |
4ffb6335 | 1372 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1373 | if (!order || !may_enter_fs || !may_perform_io) |
53853e2d | 1374 | return COMPACT_SKIPPED; |
56de7263 MG |
1375 | |
1376 | /* Compact each zone in the list */ | |
1a6d53a1 VB |
1377 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
1378 | ac->nodemask) { | |
56de7263 | 1379 | int status; |
1f9efdef | 1380 | int zone_contended; |
56de7263 | 1381 | |
53853e2d VB |
1382 | if (compaction_deferred(zone, order)) |
1383 | continue; | |
1384 | ||
e0b9daeb | 1385 | status = compact_zone_order(zone, order, gfp_mask, mode, |
1a6d53a1 VB |
1386 | &zone_contended, alloc_flags, |
1387 | ac->classzone_idx); | |
56de7263 | 1388 | rc = max(status, rc); |
1f9efdef VB |
1389 | /* |
1390 | * It takes at least one zone that wasn't lock contended | |
1391 | * to clear all_zones_contended. | |
1392 | */ | |
1393 | all_zones_contended &= zone_contended; | |
56de7263 | 1394 | |
3e7d3449 | 1395 | /* If a normal allocation would succeed, stop compacting */ |
ebff3980 | 1396 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), |
1a6d53a1 | 1397 | ac->classzone_idx, alloc_flags)) { |
53853e2d VB |
1398 | /* |
1399 | * We think the allocation will succeed in this zone, | |
1400 | * but it is not certain, hence the false. The caller | |
1401 | * will repeat this with true if allocation indeed | |
1402 | * succeeds in this zone. | |
1403 | */ | |
1404 | compaction_defer_reset(zone, order, false); | |
1f9efdef VB |
1405 | /* |
1406 | * It is possible that async compaction aborted due to | |
1407 | * need_resched() and the watermarks were ok thanks to | |
1408 | * somebody else freeing memory. The allocation can | |
1409 | * however still fail so we better signal the | |
1410 | * need_resched() contention anyway (this will not | |
1411 | * prevent the allocation attempt). | |
1412 | */ | |
1413 | if (zone_contended == COMPACT_CONTENDED_SCHED) | |
1414 | *contended = COMPACT_CONTENDED_SCHED; | |
1415 | ||
1416 | goto break_loop; | |
1417 | } | |
1418 | ||
f8669795 | 1419 | if (mode != MIGRATE_ASYNC && status == COMPACT_COMPLETE) { |
53853e2d VB |
1420 | /* |
1421 | * We think that allocation won't succeed in this zone | |
1422 | * so we defer compaction there. If it ends up | |
1423 | * succeeding after all, it will be reset. | |
1424 | */ | |
1425 | defer_compaction(zone, order); | |
1426 | } | |
1f9efdef VB |
1427 | |
1428 | /* | |
1429 | * We might have stopped compacting due to need_resched() in | |
1430 | * async compaction, or due to a fatal signal detected. In that | |
1431 | * case do not try further zones and signal need_resched() | |
1432 | * contention. | |
1433 | */ | |
1434 | if ((zone_contended == COMPACT_CONTENDED_SCHED) | |
1435 | || fatal_signal_pending(current)) { | |
1436 | *contended = COMPACT_CONTENDED_SCHED; | |
1437 | goto break_loop; | |
1438 | } | |
1439 | ||
1440 | continue; | |
1441 | break_loop: | |
1442 | /* | |
1443 | * We might not have tried all the zones, so be conservative | |
1444 | * and assume they are not all lock contended. | |
1445 | */ | |
1446 | all_zones_contended = 0; | |
1447 | break; | |
56de7263 MG |
1448 | } |
1449 | ||
1f9efdef VB |
1450 | /* |
1451 | * If at least one zone wasn't deferred or skipped, we report if all | |
1452 | * zones that were tried were lock contended. | |
1453 | */ | |
1454 | if (rc > COMPACT_SKIPPED && all_zones_contended) | |
1455 | *contended = COMPACT_CONTENDED_LOCK; | |
1456 | ||
56de7263 MG |
1457 | return rc; |
1458 | } | |
1459 | ||
1460 | ||
76ab0f53 | 1461 | /* Compact all zones within a node */ |
7103f16d | 1462 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1463 | { |
1464 | int zoneid; | |
76ab0f53 MG |
1465 | struct zone *zone; |
1466 | ||
76ab0f53 | 1467 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1468 | |
1469 | zone = &pgdat->node_zones[zoneid]; | |
1470 | if (!populated_zone(zone)) | |
1471 | continue; | |
1472 | ||
7be62de9 RR |
1473 | cc->nr_freepages = 0; |
1474 | cc->nr_migratepages = 0; | |
1475 | cc->zone = zone; | |
1476 | INIT_LIST_HEAD(&cc->freepages); | |
1477 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1478 | |
aad6ec37 | 1479 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1480 | compact_zone(zone, cc); |
76ab0f53 | 1481 | |
aff62249 | 1482 | if (cc->order > 0) { |
de6c60a6 VB |
1483 | if (zone_watermark_ok(zone, cc->order, |
1484 | low_wmark_pages(zone), 0, 0)) | |
1485 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 RR |
1486 | } |
1487 | ||
7be62de9 RR |
1488 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1489 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1490 | } |
76ab0f53 MG |
1491 | } |
1492 | ||
7103f16d | 1493 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1494 | { |
1495 | struct compact_control cc = { | |
1496 | .order = order, | |
e0b9daeb | 1497 | .mode = MIGRATE_ASYNC, |
7be62de9 RR |
1498 | }; |
1499 | ||
3a7200af MG |
1500 | if (!order) |
1501 | return; | |
1502 | ||
7103f16d | 1503 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1504 | } |
1505 | ||
7103f16d | 1506 | static void compact_node(int nid) |
7be62de9 | 1507 | { |
7be62de9 RR |
1508 | struct compact_control cc = { |
1509 | .order = -1, | |
e0b9daeb | 1510 | .mode = MIGRATE_SYNC, |
91ca9186 | 1511 | .ignore_skip_hint = true, |
7be62de9 RR |
1512 | }; |
1513 | ||
7103f16d | 1514 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1515 | } |
1516 | ||
76ab0f53 | 1517 | /* Compact all nodes in the system */ |
7964c06d | 1518 | static void compact_nodes(void) |
76ab0f53 MG |
1519 | { |
1520 | int nid; | |
1521 | ||
8575ec29 HD |
1522 | /* Flush pending updates to the LRU lists */ |
1523 | lru_add_drain_all(); | |
1524 | ||
76ab0f53 MG |
1525 | for_each_online_node(nid) |
1526 | compact_node(nid); | |
76ab0f53 MG |
1527 | } |
1528 | ||
1529 | /* The written value is actually unused, all memory is compacted */ | |
1530 | int sysctl_compact_memory; | |
1531 | ||
1532 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1533 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1534 | void __user *buffer, size_t *length, loff_t *ppos) | |
1535 | { | |
1536 | if (write) | |
7964c06d | 1537 | compact_nodes(); |
76ab0f53 MG |
1538 | |
1539 | return 0; | |
1540 | } | |
ed4a6d7f | 1541 | |
5e771905 MG |
1542 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1543 | void __user *buffer, size_t *length, loff_t *ppos) | |
1544 | { | |
1545 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1546 | ||
1547 | return 0; | |
1548 | } | |
1549 | ||
ed4a6d7f | 1550 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1551 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1552 | struct device_attribute *attr, |
ed4a6d7f MG |
1553 | const char *buf, size_t count) |
1554 | { | |
8575ec29 HD |
1555 | int nid = dev->id; |
1556 | ||
1557 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1558 | /* Flush pending updates to the LRU lists */ | |
1559 | lru_add_drain_all(); | |
1560 | ||
1561 | compact_node(nid); | |
1562 | } | |
ed4a6d7f MG |
1563 | |
1564 | return count; | |
1565 | } | |
10fbcf4c | 1566 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1567 | |
1568 | int compaction_register_node(struct node *node) | |
1569 | { | |
10fbcf4c | 1570 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1571 | } |
1572 | ||
1573 | void compaction_unregister_node(struct node *node) | |
1574 | { | |
10fbcf4c | 1575 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1576 | } |
1577 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1578 | |
1579 | #endif /* CONFIG_COMPACTION */ |