]>
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", | |
837d026d JK |
44 | "no_suitable_page", |
45 | "not_suitable_zone", | |
16c4a097 JK |
46 | }; |
47 | #endif | |
ff9543fd | 48 | |
b7aba698 MG |
49 | #define CREATE_TRACE_POINTS |
50 | #include <trace/events/compaction.h> | |
51 | ||
748446bb MG |
52 | static unsigned long release_freepages(struct list_head *freelist) |
53 | { | |
54 | struct page *page, *next; | |
6bace090 | 55 | unsigned long high_pfn = 0; |
748446bb MG |
56 | |
57 | list_for_each_entry_safe(page, next, freelist, lru) { | |
6bace090 | 58 | unsigned long pfn = page_to_pfn(page); |
748446bb MG |
59 | list_del(&page->lru); |
60 | __free_page(page); | |
6bace090 VB |
61 | if (pfn > high_pfn) |
62 | high_pfn = pfn; | |
748446bb MG |
63 | } |
64 | ||
6bace090 | 65 | return high_pfn; |
748446bb MG |
66 | } |
67 | ||
ff9543fd MN |
68 | static void map_pages(struct list_head *list) |
69 | { | |
70 | struct page *page; | |
71 | ||
72 | list_for_each_entry(page, list, lru) { | |
73 | arch_alloc_page(page, 0); | |
74 | kernel_map_pages(page, 1, 1); | |
75 | } | |
76 | } | |
77 | ||
47118af0 MN |
78 | static inline bool migrate_async_suitable(int migratetype) |
79 | { | |
80 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; | |
81 | } | |
82 | ||
7d49d886 VB |
83 | /* |
84 | * Check that the whole (or subset of) a pageblock given by the interval of | |
85 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
86 | * with the migration of free compaction scanner. The scanners then need to | |
87 | * use only pfn_valid_within() check for arches that allow holes within | |
88 | * pageblocks. | |
89 | * | |
90 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
91 | * | |
92 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
93 | * i.e. it's possible that all pages within a zones range of pages do not | |
94 | * belong to a single zone. We assume that a border between node0 and node1 | |
95 | * can occur within a single pageblock, but not a node0 node1 node0 | |
96 | * interleaving within a single pageblock. It is therefore sufficient to check | |
97 | * the first and last page of a pageblock and avoid checking each individual | |
98 | * page in a pageblock. | |
99 | */ | |
100 | static struct page *pageblock_pfn_to_page(unsigned long start_pfn, | |
101 | unsigned long end_pfn, struct zone *zone) | |
102 | { | |
103 | struct page *start_page; | |
104 | struct page *end_page; | |
105 | ||
106 | /* end_pfn is one past the range we are checking */ | |
107 | end_pfn--; | |
108 | ||
109 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
110 | return NULL; | |
111 | ||
112 | start_page = pfn_to_page(start_pfn); | |
113 | ||
114 | if (page_zone(start_page) != zone) | |
115 | return NULL; | |
116 | ||
117 | end_page = pfn_to_page(end_pfn); | |
118 | ||
119 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
120 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
121 | return NULL; | |
122 | ||
123 | return start_page; | |
124 | } | |
125 | ||
bb13ffeb | 126 | #ifdef CONFIG_COMPACTION |
24e2716f JK |
127 | |
128 | /* Do not skip compaction more than 64 times */ | |
129 | #define COMPACT_MAX_DEFER_SHIFT 6 | |
130 | ||
131 | /* | |
132 | * Compaction is deferred when compaction fails to result in a page | |
133 | * allocation success. 1 << compact_defer_limit compactions are skipped up | |
134 | * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT | |
135 | */ | |
136 | void defer_compaction(struct zone *zone, int order) | |
137 | { | |
138 | zone->compact_considered = 0; | |
139 | zone->compact_defer_shift++; | |
140 | ||
141 | if (order < zone->compact_order_failed) | |
142 | zone->compact_order_failed = order; | |
143 | ||
144 | if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT) | |
145 | zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT; | |
146 | ||
147 | trace_mm_compaction_defer_compaction(zone, order); | |
148 | } | |
149 | ||
150 | /* Returns true if compaction should be skipped this time */ | |
151 | bool compaction_deferred(struct zone *zone, int order) | |
152 | { | |
153 | unsigned long defer_limit = 1UL << zone->compact_defer_shift; | |
154 | ||
155 | if (order < zone->compact_order_failed) | |
156 | return false; | |
157 | ||
158 | /* Avoid possible overflow */ | |
159 | if (++zone->compact_considered > defer_limit) | |
160 | zone->compact_considered = defer_limit; | |
161 | ||
162 | if (zone->compact_considered >= defer_limit) | |
163 | return false; | |
164 | ||
165 | trace_mm_compaction_deferred(zone, order); | |
166 | ||
167 | return true; | |
168 | } | |
169 | ||
170 | /* | |
171 | * Update defer tracking counters after successful compaction of given order, | |
172 | * which means an allocation either succeeded (alloc_success == true) or is | |
173 | * expected to succeed. | |
174 | */ | |
175 | void compaction_defer_reset(struct zone *zone, int order, | |
176 | bool alloc_success) | |
177 | { | |
178 | if (alloc_success) { | |
179 | zone->compact_considered = 0; | |
180 | zone->compact_defer_shift = 0; | |
181 | } | |
182 | if (order >= zone->compact_order_failed) | |
183 | zone->compact_order_failed = order + 1; | |
184 | ||
185 | trace_mm_compaction_defer_reset(zone, order); | |
186 | } | |
187 | ||
188 | /* Returns true if restarting compaction after many failures */ | |
189 | bool compaction_restarting(struct zone *zone, int order) | |
190 | { | |
191 | if (order < zone->compact_order_failed) | |
192 | return false; | |
193 | ||
194 | return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT && | |
195 | zone->compact_considered >= 1UL << zone->compact_defer_shift; | |
196 | } | |
197 | ||
bb13ffeb MG |
198 | /* Returns true if the pageblock should be scanned for pages to isolate. */ |
199 | static inline bool isolation_suitable(struct compact_control *cc, | |
200 | struct page *page) | |
201 | { | |
202 | if (cc->ignore_skip_hint) | |
203 | return true; | |
204 | ||
205 | return !get_pageblock_skip(page); | |
206 | } | |
207 | ||
208 | /* | |
209 | * This function is called to clear all cached information on pageblocks that | |
210 | * should be skipped for page isolation when the migrate and free page scanner | |
211 | * meet. | |
212 | */ | |
62997027 | 213 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
214 | { |
215 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 216 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
217 | unsigned long pfn; |
218 | ||
35979ef3 DR |
219 | zone->compact_cached_migrate_pfn[0] = start_pfn; |
220 | zone->compact_cached_migrate_pfn[1] = start_pfn; | |
c89511ab | 221 | zone->compact_cached_free_pfn = end_pfn; |
62997027 | 222 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
223 | |
224 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
225 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
226 | struct page *page; | |
227 | ||
228 | cond_resched(); | |
229 | ||
230 | if (!pfn_valid(pfn)) | |
231 | continue; | |
232 | ||
233 | page = pfn_to_page(pfn); | |
234 | if (zone != page_zone(page)) | |
235 | continue; | |
236 | ||
237 | clear_pageblock_skip(page); | |
238 | } | |
239 | } | |
240 | ||
62997027 MG |
241 | void reset_isolation_suitable(pg_data_t *pgdat) |
242 | { | |
243 | int zoneid; | |
244 | ||
245 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
246 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
247 | if (!populated_zone(zone)) | |
248 | continue; | |
249 | ||
250 | /* Only flush if a full compaction finished recently */ | |
251 | if (zone->compact_blockskip_flush) | |
252 | __reset_isolation_suitable(zone); | |
253 | } | |
254 | } | |
255 | ||
bb13ffeb MG |
256 | /* |
257 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 258 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 259 | */ |
c89511ab MG |
260 | static void update_pageblock_skip(struct compact_control *cc, |
261 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 262 | bool migrate_scanner) |
bb13ffeb | 263 | { |
c89511ab | 264 | struct zone *zone = cc->zone; |
35979ef3 | 265 | unsigned long pfn; |
6815bf3f JK |
266 | |
267 | if (cc->ignore_skip_hint) | |
268 | return; | |
269 | ||
bb13ffeb MG |
270 | if (!page) |
271 | return; | |
272 | ||
35979ef3 DR |
273 | if (nr_isolated) |
274 | return; | |
275 | ||
edc2ca61 | 276 | set_pageblock_skip(page); |
c89511ab | 277 | |
35979ef3 DR |
278 | pfn = page_to_pfn(page); |
279 | ||
280 | /* Update where async and sync compaction should restart */ | |
281 | if (migrate_scanner) { | |
35979ef3 DR |
282 | if (pfn > zone->compact_cached_migrate_pfn[0]) |
283 | zone->compact_cached_migrate_pfn[0] = pfn; | |
e0b9daeb DR |
284 | if (cc->mode != MIGRATE_ASYNC && |
285 | pfn > zone->compact_cached_migrate_pfn[1]) | |
35979ef3 DR |
286 | zone->compact_cached_migrate_pfn[1] = pfn; |
287 | } else { | |
35979ef3 DR |
288 | if (pfn < zone->compact_cached_free_pfn) |
289 | zone->compact_cached_free_pfn = pfn; | |
c89511ab | 290 | } |
bb13ffeb MG |
291 | } |
292 | #else | |
293 | static inline bool isolation_suitable(struct compact_control *cc, | |
294 | struct page *page) | |
295 | { | |
296 | return true; | |
297 | } | |
298 | ||
c89511ab MG |
299 | static void update_pageblock_skip(struct compact_control *cc, |
300 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 301 | bool migrate_scanner) |
bb13ffeb MG |
302 | { |
303 | } | |
304 | #endif /* CONFIG_COMPACTION */ | |
305 | ||
8b44d279 VB |
306 | /* |
307 | * Compaction requires the taking of some coarse locks that are potentially | |
308 | * very heavily contended. For async compaction, back out if the lock cannot | |
309 | * be taken immediately. For sync compaction, spin on the lock if needed. | |
310 | * | |
311 | * Returns true if the lock is held | |
312 | * Returns false if the lock is not held and compaction should abort | |
313 | */ | |
314 | static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags, | |
315 | struct compact_control *cc) | |
2a1402aa | 316 | { |
8b44d279 VB |
317 | if (cc->mode == MIGRATE_ASYNC) { |
318 | if (!spin_trylock_irqsave(lock, *flags)) { | |
319 | cc->contended = COMPACT_CONTENDED_LOCK; | |
320 | return false; | |
321 | } | |
322 | } else { | |
323 | spin_lock_irqsave(lock, *flags); | |
324 | } | |
1f9efdef | 325 | |
8b44d279 | 326 | return true; |
2a1402aa MG |
327 | } |
328 | ||
c67fe375 MG |
329 | /* |
330 | * Compaction requires the taking of some coarse locks that are potentially | |
8b44d279 VB |
331 | * very heavily contended. The lock should be periodically unlocked to avoid |
332 | * having disabled IRQs for a long time, even when there is nobody waiting on | |
333 | * the lock. It might also be that allowing the IRQs will result in | |
334 | * need_resched() becoming true. If scheduling is needed, async compaction | |
335 | * aborts. Sync compaction schedules. | |
336 | * Either compaction type will also abort if a fatal signal is pending. | |
337 | * In either case if the lock was locked, it is dropped and not regained. | |
c67fe375 | 338 | * |
8b44d279 VB |
339 | * Returns true if compaction should abort due to fatal signal pending, or |
340 | * async compaction due to need_resched() | |
341 | * Returns false when compaction can continue (sync compaction might have | |
342 | * scheduled) | |
c67fe375 | 343 | */ |
8b44d279 VB |
344 | static bool compact_unlock_should_abort(spinlock_t *lock, |
345 | unsigned long flags, bool *locked, struct compact_control *cc) | |
c67fe375 | 346 | { |
8b44d279 VB |
347 | if (*locked) { |
348 | spin_unlock_irqrestore(lock, flags); | |
349 | *locked = false; | |
350 | } | |
1f9efdef | 351 | |
8b44d279 VB |
352 | if (fatal_signal_pending(current)) { |
353 | cc->contended = COMPACT_CONTENDED_SCHED; | |
354 | return true; | |
355 | } | |
c67fe375 | 356 | |
8b44d279 | 357 | if (need_resched()) { |
e0b9daeb | 358 | if (cc->mode == MIGRATE_ASYNC) { |
8b44d279 VB |
359 | cc->contended = COMPACT_CONTENDED_SCHED; |
360 | return true; | |
c67fe375 | 361 | } |
c67fe375 | 362 | cond_resched(); |
c67fe375 MG |
363 | } |
364 | ||
8b44d279 | 365 | return false; |
c67fe375 MG |
366 | } |
367 | ||
be976572 VB |
368 | /* |
369 | * Aside from avoiding lock contention, compaction also periodically checks | |
370 | * need_resched() and either schedules in sync compaction or aborts async | |
8b44d279 | 371 | * compaction. This is similar to what compact_unlock_should_abort() does, but |
be976572 VB |
372 | * is used where no lock is concerned. |
373 | * | |
374 | * Returns false when no scheduling was needed, or sync compaction scheduled. | |
375 | * Returns true when async compaction should abort. | |
376 | */ | |
377 | static inline bool compact_should_abort(struct compact_control *cc) | |
378 | { | |
379 | /* async compaction aborts if contended */ | |
380 | if (need_resched()) { | |
381 | if (cc->mode == MIGRATE_ASYNC) { | |
1f9efdef | 382 | cc->contended = COMPACT_CONTENDED_SCHED; |
be976572 VB |
383 | return true; |
384 | } | |
385 | ||
386 | cond_resched(); | |
387 | } | |
388 | ||
389 | return false; | |
390 | } | |
391 | ||
f40d1e42 MG |
392 | /* Returns true if the page is within a block suitable for migration to */ |
393 | static bool suitable_migration_target(struct page *page) | |
394 | { | |
7d348b9e | 395 | /* If the page is a large free page, then disallow migration */ |
99c0fd5e VB |
396 | if (PageBuddy(page)) { |
397 | /* | |
398 | * We are checking page_order without zone->lock taken. But | |
399 | * the only small danger is that we skip a potentially suitable | |
400 | * pageblock, so it's not worth to check order for valid range. | |
401 | */ | |
402 | if (page_order_unsafe(page) >= pageblock_order) | |
403 | return false; | |
404 | } | |
f40d1e42 MG |
405 | |
406 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
7d348b9e | 407 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
f40d1e42 MG |
408 | return true; |
409 | ||
410 | /* Otherwise skip the block */ | |
411 | return false; | |
412 | } | |
413 | ||
85aa125f | 414 | /* |
9e4be470 JM |
415 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
416 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
417 | * (even though it may still end up isolating some pages). | |
85aa125f | 418 | */ |
f40d1e42 | 419 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
e14c720e | 420 | unsigned long *start_pfn, |
85aa125f MN |
421 | unsigned long end_pfn, |
422 | struct list_head *freelist, | |
423 | bool strict) | |
748446bb | 424 | { |
b7aba698 | 425 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 426 | struct page *cursor, *valid_page = NULL; |
b8b2d825 | 427 | unsigned long flags = 0; |
f40d1e42 | 428 | bool locked = false; |
e14c720e | 429 | unsigned long blockpfn = *start_pfn; |
748446bb | 430 | |
748446bb MG |
431 | cursor = pfn_to_page(blockpfn); |
432 | ||
f40d1e42 | 433 | /* Isolate free pages. */ |
748446bb MG |
434 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
435 | int isolated, i; | |
436 | struct page *page = cursor; | |
437 | ||
8b44d279 VB |
438 | /* |
439 | * Periodically drop the lock (if held) regardless of its | |
440 | * contention, to give chance to IRQs. Abort if fatal signal | |
441 | * pending or async compaction detects need_resched() | |
442 | */ | |
443 | if (!(blockpfn % SWAP_CLUSTER_MAX) | |
444 | && compact_unlock_should_abort(&cc->zone->lock, flags, | |
445 | &locked, cc)) | |
446 | break; | |
447 | ||
b7aba698 | 448 | nr_scanned++; |
f40d1e42 | 449 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
450 | goto isolate_fail; |
451 | ||
bb13ffeb MG |
452 | if (!valid_page) |
453 | valid_page = page; | |
f40d1e42 | 454 | if (!PageBuddy(page)) |
2af120bc | 455 | goto isolate_fail; |
f40d1e42 MG |
456 | |
457 | /* | |
69b7189f VB |
458 | * If we already hold the lock, we can skip some rechecking. |
459 | * Note that if we hold the lock now, checked_pageblock was | |
460 | * already set in some previous iteration (or strict is true), | |
461 | * so it is correct to skip the suitable migration target | |
462 | * recheck as well. | |
f40d1e42 | 463 | */ |
69b7189f VB |
464 | if (!locked) { |
465 | /* | |
466 | * The zone lock must be held to isolate freepages. | |
467 | * Unfortunately this is a very coarse lock and can be | |
468 | * heavily contended if there are parallel allocations | |
469 | * or parallel compactions. For async compaction do not | |
470 | * spin on the lock and we acquire the lock as late as | |
471 | * possible. | |
472 | */ | |
8b44d279 VB |
473 | locked = compact_trylock_irqsave(&cc->zone->lock, |
474 | &flags, cc); | |
69b7189f VB |
475 | if (!locked) |
476 | break; | |
f40d1e42 | 477 | |
69b7189f VB |
478 | /* Recheck this is a buddy page under lock */ |
479 | if (!PageBuddy(page)) | |
480 | goto isolate_fail; | |
481 | } | |
748446bb MG |
482 | |
483 | /* Found a free page, break it into order-0 pages */ | |
484 | isolated = split_free_page(page); | |
485 | total_isolated += isolated; | |
486 | for (i = 0; i < isolated; i++) { | |
487 | list_add(&page->lru, freelist); | |
488 | page++; | |
489 | } | |
490 | ||
491 | /* If a page was split, advance to the end of it */ | |
492 | if (isolated) { | |
932ff6bb JK |
493 | cc->nr_freepages += isolated; |
494 | if (!strict && | |
495 | cc->nr_migratepages <= cc->nr_freepages) { | |
496 | blockpfn += isolated; | |
497 | break; | |
498 | } | |
499 | ||
748446bb MG |
500 | blockpfn += isolated - 1; |
501 | cursor += isolated - 1; | |
2af120bc | 502 | continue; |
748446bb | 503 | } |
2af120bc LA |
504 | |
505 | isolate_fail: | |
506 | if (strict) | |
507 | break; | |
508 | else | |
509 | continue; | |
510 | ||
748446bb MG |
511 | } |
512 | ||
e34d85f0 JK |
513 | trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn, |
514 | nr_scanned, total_isolated); | |
515 | ||
e14c720e VB |
516 | /* Record how far we have got within the block */ |
517 | *start_pfn = blockpfn; | |
518 | ||
f40d1e42 MG |
519 | /* |
520 | * If strict isolation is requested by CMA then check that all the | |
521 | * pages requested were isolated. If there were any failures, 0 is | |
522 | * returned and CMA will fail. | |
523 | */ | |
2af120bc | 524 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
525 | total_isolated = 0; |
526 | ||
527 | if (locked) | |
528 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
529 | ||
bb13ffeb MG |
530 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
531 | if (blockpfn == end_pfn) | |
edc2ca61 | 532 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 533 | |
010fc29a | 534 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 535 | if (total_isolated) |
010fc29a | 536 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
537 | return total_isolated; |
538 | } | |
539 | ||
85aa125f MN |
540 | /** |
541 | * isolate_freepages_range() - isolate free pages. | |
542 | * @start_pfn: The first PFN to start isolating. | |
543 | * @end_pfn: The one-past-last PFN. | |
544 | * | |
545 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
546 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
547 | * undo its actions and return zero. | |
548 | * | |
549 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
550 | * (which may be greater then end_pfn if end fell in a middle of | |
551 | * a free page). | |
552 | */ | |
ff9543fd | 553 | unsigned long |
bb13ffeb MG |
554 | isolate_freepages_range(struct compact_control *cc, |
555 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 556 | { |
f40d1e42 | 557 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
558 | LIST_HEAD(freelist); |
559 | ||
7d49d886 VB |
560 | pfn = start_pfn; |
561 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
562 | ||
563 | for (; pfn < end_pfn; pfn += isolated, | |
564 | block_end_pfn += pageblock_nr_pages) { | |
e14c720e VB |
565 | /* Protect pfn from changing by isolate_freepages_block */ |
566 | unsigned long isolate_start_pfn = pfn; | |
85aa125f | 567 | |
85aa125f MN |
568 | block_end_pfn = min(block_end_pfn, end_pfn); |
569 | ||
58420016 JK |
570 | /* |
571 | * pfn could pass the block_end_pfn if isolated freepage | |
572 | * is more than pageblock order. In this case, we adjust | |
573 | * scanning range to right one. | |
574 | */ | |
575 | if (pfn >= block_end_pfn) { | |
576 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
577 | block_end_pfn = min(block_end_pfn, end_pfn); | |
578 | } | |
579 | ||
7d49d886 VB |
580 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
581 | break; | |
582 | ||
e14c720e VB |
583 | isolated = isolate_freepages_block(cc, &isolate_start_pfn, |
584 | block_end_pfn, &freelist, true); | |
85aa125f MN |
585 | |
586 | /* | |
587 | * In strict mode, isolate_freepages_block() returns 0 if | |
588 | * there are any holes in the block (ie. invalid PFNs or | |
589 | * non-free pages). | |
590 | */ | |
591 | if (!isolated) | |
592 | break; | |
593 | ||
594 | /* | |
595 | * If we managed to isolate pages, it is always (1 << n) * | |
596 | * pageblock_nr_pages for some non-negative n. (Max order | |
597 | * page may span two pageblocks). | |
598 | */ | |
599 | } | |
600 | ||
601 | /* split_free_page does not map the pages */ | |
602 | map_pages(&freelist); | |
603 | ||
604 | if (pfn < end_pfn) { | |
605 | /* Loop terminated early, cleanup. */ | |
606 | release_freepages(&freelist); | |
607 | return 0; | |
608 | } | |
609 | ||
610 | /* We don't use freelists for anything. */ | |
611 | return pfn; | |
612 | } | |
613 | ||
748446bb | 614 | /* Update the number of anon and file isolated pages in the zone */ |
edc2ca61 | 615 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
748446bb MG |
616 | { |
617 | struct page *page; | |
b9e84ac1 | 618 | unsigned int count[2] = { 0, }; |
748446bb | 619 | |
edc2ca61 VB |
620 | if (list_empty(&cc->migratepages)) |
621 | return; | |
622 | ||
b9e84ac1 MK |
623 | list_for_each_entry(page, &cc->migratepages, lru) |
624 | count[!!page_is_file_cache(page)]++; | |
748446bb | 625 | |
edc2ca61 VB |
626 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
627 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
748446bb MG |
628 | } |
629 | ||
630 | /* Similar to reclaim, but different enough that they don't share logic */ | |
631 | static bool too_many_isolated(struct zone *zone) | |
632 | { | |
bc693045 | 633 | unsigned long active, inactive, isolated; |
748446bb MG |
634 | |
635 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
636 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
637 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
638 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
639 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
640 | zone_page_state(zone, NR_ISOLATED_ANON); | |
641 | ||
bc693045 | 642 | return isolated > (inactive + active) / 2; |
748446bb MG |
643 | } |
644 | ||
2fe86e00 | 645 | /** |
edc2ca61 VB |
646 | * isolate_migratepages_block() - isolate all migrate-able pages within |
647 | * a single pageblock | |
2fe86e00 | 648 | * @cc: Compaction control structure. |
edc2ca61 VB |
649 | * @low_pfn: The first PFN to isolate |
650 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock | |
651 | * @isolate_mode: Isolation mode to be used. | |
2fe86e00 MN |
652 | * |
653 | * Isolate all pages that can be migrated from the range specified by | |
edc2ca61 VB |
654 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
655 | * Returns zero if there is a fatal signal pending, otherwise PFN of the | |
656 | * first page that was not scanned (which may be both less, equal to or more | |
657 | * than end_pfn). | |
2fe86e00 | 658 | * |
edc2ca61 VB |
659 | * The pages are isolated on cc->migratepages list (not required to be empty), |
660 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field | |
661 | * is neither read nor updated. | |
748446bb | 662 | */ |
edc2ca61 VB |
663 | static unsigned long |
664 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, | |
665 | unsigned long end_pfn, isolate_mode_t isolate_mode) | |
748446bb | 666 | { |
edc2ca61 | 667 | struct zone *zone = cc->zone; |
b7aba698 | 668 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 669 | struct list_head *migratelist = &cc->migratepages; |
fa9add64 | 670 | struct lruvec *lruvec; |
b8b2d825 | 671 | unsigned long flags = 0; |
2a1402aa | 672 | bool locked = false; |
bb13ffeb | 673 | struct page *page = NULL, *valid_page = NULL; |
e34d85f0 | 674 | unsigned long start_pfn = low_pfn; |
748446bb | 675 | |
748446bb MG |
676 | /* |
677 | * Ensure that there are not too many pages isolated from the LRU | |
678 | * list by either parallel reclaimers or compaction. If there are, | |
679 | * delay for some time until fewer pages are isolated | |
680 | */ | |
681 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 682 | /* async migration should just abort */ |
e0b9daeb | 683 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 684 | return 0; |
f9e35b3b | 685 | |
748446bb MG |
686 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
687 | ||
688 | if (fatal_signal_pending(current)) | |
2fe86e00 | 689 | return 0; |
748446bb MG |
690 | } |
691 | ||
be976572 VB |
692 | if (compact_should_abort(cc)) |
693 | return 0; | |
aeef4b83 | 694 | |
748446bb | 695 | /* Time to isolate some pages for migration */ |
748446bb | 696 | for (; low_pfn < end_pfn; low_pfn++) { |
8b44d279 VB |
697 | /* |
698 | * Periodically drop the lock (if held) regardless of its | |
699 | * contention, to give chance to IRQs. Abort async compaction | |
700 | * if contended. | |
701 | */ | |
702 | if (!(low_pfn % SWAP_CLUSTER_MAX) | |
703 | && compact_unlock_should_abort(&zone->lru_lock, flags, | |
704 | &locked, cc)) | |
705 | break; | |
c67fe375 | 706 | |
748446bb MG |
707 | if (!pfn_valid_within(low_pfn)) |
708 | continue; | |
b7aba698 | 709 | nr_scanned++; |
748446bb | 710 | |
748446bb | 711 | page = pfn_to_page(low_pfn); |
dc908600 | 712 | |
bb13ffeb MG |
713 | if (!valid_page) |
714 | valid_page = page; | |
715 | ||
6c14466c | 716 | /* |
99c0fd5e VB |
717 | * Skip if free. We read page order here without zone lock |
718 | * which is generally unsafe, but the race window is small and | |
719 | * the worst thing that can happen is that we skip some | |
720 | * potential isolation targets. | |
6c14466c | 721 | */ |
99c0fd5e VB |
722 | if (PageBuddy(page)) { |
723 | unsigned long freepage_order = page_order_unsafe(page); | |
724 | ||
725 | /* | |
726 | * Without lock, we cannot be sure that what we got is | |
727 | * a valid page order. Consider only values in the | |
728 | * valid order range to prevent low_pfn overflow. | |
729 | */ | |
730 | if (freepage_order > 0 && freepage_order < MAX_ORDER) | |
731 | low_pfn += (1UL << freepage_order) - 1; | |
748446bb | 732 | continue; |
99c0fd5e | 733 | } |
748446bb | 734 | |
bf6bddf1 RA |
735 | /* |
736 | * Check may be lockless but that's ok as we recheck later. | |
737 | * It's possible to migrate LRU pages and balloon pages | |
738 | * Skip any other type of page | |
739 | */ | |
740 | if (!PageLRU(page)) { | |
741 | if (unlikely(balloon_page_movable(page))) { | |
d6d86c0a | 742 | if (balloon_page_isolate(page)) { |
bf6bddf1 | 743 | /* Successfully isolated */ |
b6c75016 | 744 | goto isolate_success; |
bf6bddf1 RA |
745 | } |
746 | } | |
bc835011 | 747 | continue; |
bf6bddf1 | 748 | } |
bc835011 AA |
749 | |
750 | /* | |
2a1402aa MG |
751 | * PageLRU is set. lru_lock normally excludes isolation |
752 | * splitting and collapsing (collapsing has already happened | |
753 | * if PageLRU is set) but the lock is not necessarily taken | |
754 | * here and it is wasteful to take it just to check transhuge. | |
755 | * Check TransHuge without lock and skip the whole pageblock if | |
756 | * it's either a transhuge or hugetlbfs page, as calling | |
757 | * compound_order() without preventing THP from splitting the | |
758 | * page underneath us may return surprising results. | |
bc835011 | 759 | */ |
2a1402aa MG |
760 | if (PageTransHuge(page)) { |
761 | if (!locked) | |
edc2ca61 VB |
762 | low_pfn = ALIGN(low_pfn + 1, |
763 | pageblock_nr_pages) - 1; | |
764 | else | |
765 | low_pfn += (1 << compound_order(page)) - 1; | |
766 | ||
2a1402aa MG |
767 | continue; |
768 | } | |
769 | ||
119d6d59 DR |
770 | /* |
771 | * Migration will fail if an anonymous page is pinned in memory, | |
772 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
773 | * admittedly racy check. | |
774 | */ | |
775 | if (!page_mapping(page) && | |
776 | page_count(page) > page_mapcount(page)) | |
777 | continue; | |
778 | ||
69b7189f VB |
779 | /* If we already hold the lock, we can skip some rechecking */ |
780 | if (!locked) { | |
8b44d279 VB |
781 | locked = compact_trylock_irqsave(&zone->lru_lock, |
782 | &flags, cc); | |
69b7189f VB |
783 | if (!locked) |
784 | break; | |
2a1402aa | 785 | |
69b7189f VB |
786 | /* Recheck PageLRU and PageTransHuge under lock */ |
787 | if (!PageLRU(page)) | |
788 | continue; | |
789 | if (PageTransHuge(page)) { | |
790 | low_pfn += (1 << compound_order(page)) - 1; | |
791 | continue; | |
792 | } | |
bc835011 AA |
793 | } |
794 | ||
fa9add64 HD |
795 | lruvec = mem_cgroup_page_lruvec(page, zone); |
796 | ||
748446bb | 797 | /* Try isolate the page */ |
edc2ca61 | 798 | if (__isolate_lru_page(page, isolate_mode) != 0) |
748446bb MG |
799 | continue; |
800 | ||
309381fe | 801 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 802 | |
748446bb | 803 | /* Successfully isolated */ |
fa9add64 | 804 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
805 | |
806 | isolate_success: | |
748446bb | 807 | list_add(&page->lru, migratelist); |
748446bb | 808 | cc->nr_migratepages++; |
b7aba698 | 809 | nr_isolated++; |
748446bb MG |
810 | |
811 | /* Avoid isolating too much */ | |
31b8384a HD |
812 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
813 | ++low_pfn; | |
748446bb | 814 | break; |
31b8384a | 815 | } |
748446bb MG |
816 | } |
817 | ||
99c0fd5e VB |
818 | /* |
819 | * The PageBuddy() check could have potentially brought us outside | |
820 | * the range to be scanned. | |
821 | */ | |
822 | if (unlikely(low_pfn > end_pfn)) | |
823 | low_pfn = end_pfn; | |
824 | ||
c67fe375 MG |
825 | if (locked) |
826 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 827 | |
50b5b094 VB |
828 | /* |
829 | * Update the pageblock-skip information and cached scanner pfn, | |
830 | * if the whole pageblock was scanned without isolating any page. | |
50b5b094 | 831 | */ |
35979ef3 | 832 | if (low_pfn == end_pfn) |
edc2ca61 | 833 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 834 | |
e34d85f0 JK |
835 | trace_mm_compaction_isolate_migratepages(start_pfn, low_pfn, |
836 | nr_scanned, nr_isolated); | |
b7aba698 | 837 | |
010fc29a | 838 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 839 | if (nr_isolated) |
010fc29a | 840 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 841 | |
2fe86e00 MN |
842 | return low_pfn; |
843 | } | |
844 | ||
edc2ca61 VB |
845 | /** |
846 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range | |
847 | * @cc: Compaction control structure. | |
848 | * @start_pfn: The first PFN to start isolating. | |
849 | * @end_pfn: The one-past-last PFN. | |
850 | * | |
851 | * Returns zero if isolation fails fatally due to e.g. pending signal. | |
852 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
853 | * (which may be greater than end_pfn if end fell in a middle of a THP page). | |
854 | */ | |
855 | unsigned long | |
856 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, | |
857 | unsigned long end_pfn) | |
858 | { | |
859 | unsigned long pfn, block_end_pfn; | |
860 | ||
861 | /* Scan block by block. First and last block may be incomplete */ | |
862 | pfn = start_pfn; | |
863 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
864 | ||
865 | for (; pfn < end_pfn; pfn = block_end_pfn, | |
866 | block_end_pfn += pageblock_nr_pages) { | |
867 | ||
868 | block_end_pfn = min(block_end_pfn, end_pfn); | |
869 | ||
7d49d886 | 870 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
edc2ca61 VB |
871 | continue; |
872 | ||
873 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, | |
874 | ISOLATE_UNEVICTABLE); | |
875 | ||
876 | /* | |
877 | * In case of fatal failure, release everything that might | |
878 | * have been isolated in the previous iteration, and signal | |
879 | * the failure back to caller. | |
880 | */ | |
881 | if (!pfn) { | |
882 | putback_movable_pages(&cc->migratepages); | |
883 | cc->nr_migratepages = 0; | |
884 | break; | |
885 | } | |
6ea41c0c JK |
886 | |
887 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) | |
888 | break; | |
edc2ca61 VB |
889 | } |
890 | acct_isolated(cc->zone, cc); | |
891 | ||
892 | return pfn; | |
893 | } | |
894 | ||
ff9543fd MN |
895 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
896 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 897 | /* |
ff9543fd MN |
898 | * Based on information in the current compact_control, find blocks |
899 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 900 | */ |
edc2ca61 | 901 | static void isolate_freepages(struct compact_control *cc) |
2fe86e00 | 902 | { |
edc2ca61 | 903 | struct zone *zone = cc->zone; |
ff9543fd | 904 | struct page *page; |
c96b9e50 | 905 | unsigned long block_start_pfn; /* start of current pageblock */ |
e14c720e | 906 | unsigned long isolate_start_pfn; /* exact pfn we start at */ |
c96b9e50 VB |
907 | unsigned long block_end_pfn; /* end of current pageblock */ |
908 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd | 909 | struct list_head *freelist = &cc->freepages; |
2fe86e00 | 910 | |
ff9543fd MN |
911 | /* |
912 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 | 913 | * successfully isolated from, zone-cached value, or the end of the |
e14c720e VB |
914 | * zone when isolating for the first time. For looping we also need |
915 | * this pfn aligned down to the pageblock boundary, because we do | |
c96b9e50 VB |
916 | * block_start_pfn -= pageblock_nr_pages in the for loop. |
917 | * For ending point, take care when isolating in last pageblock of a | |
918 | * a zone which ends in the middle of a pageblock. | |
49e068f0 VB |
919 | * The low boundary is the end of the pageblock the migration scanner |
920 | * is using. | |
ff9543fd | 921 | */ |
e14c720e | 922 | isolate_start_pfn = cc->free_pfn; |
c96b9e50 VB |
923 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
924 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, | |
925 | zone_end_pfn(zone)); | |
7ed695e0 | 926 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 927 | |
ff9543fd MN |
928 | /* |
929 | * Isolate free pages until enough are available to migrate the | |
930 | * pages on cc->migratepages. We stop searching if the migrate | |
931 | * and free page scanners meet or enough free pages are isolated. | |
932 | */ | |
932ff6bb JK |
933 | for (; block_start_pfn >= low_pfn && |
934 | cc->nr_migratepages > cc->nr_freepages; | |
c96b9e50 | 935 | block_end_pfn = block_start_pfn, |
e14c720e VB |
936 | block_start_pfn -= pageblock_nr_pages, |
937 | isolate_start_pfn = block_start_pfn) { | |
2fe86e00 | 938 | |
f6ea3adb DR |
939 | /* |
940 | * This can iterate a massively long zone without finding any | |
941 | * suitable migration targets, so periodically check if we need | |
be976572 | 942 | * to schedule, or even abort async compaction. |
f6ea3adb | 943 | */ |
be976572 VB |
944 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
945 | && compact_should_abort(cc)) | |
946 | break; | |
f6ea3adb | 947 | |
7d49d886 VB |
948 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
949 | zone); | |
950 | if (!page) | |
ff9543fd MN |
951 | continue; |
952 | ||
953 | /* Check the block is suitable for migration */ | |
68e3e926 | 954 | if (!suitable_migration_target(page)) |
ff9543fd | 955 | continue; |
68e3e926 | 956 | |
bb13ffeb MG |
957 | /* If isolation recently failed, do not retry */ |
958 | if (!isolation_suitable(cc, page)) | |
959 | continue; | |
960 | ||
e14c720e | 961 | /* Found a block suitable for isolating free pages from. */ |
932ff6bb | 962 | isolate_freepages_block(cc, &isolate_start_pfn, |
c96b9e50 | 963 | block_end_pfn, freelist, false); |
ff9543fd | 964 | |
e14c720e VB |
965 | /* |
966 | * Remember where the free scanner should restart next time, | |
967 | * which is where isolate_freepages_block() left off. | |
968 | * But if it scanned the whole pageblock, isolate_start_pfn | |
969 | * now points at block_end_pfn, which is the start of the next | |
970 | * pageblock. | |
971 | * In that case we will however want to restart at the start | |
972 | * of the previous pageblock. | |
973 | */ | |
974 | cc->free_pfn = (isolate_start_pfn < block_end_pfn) ? | |
975 | isolate_start_pfn : | |
976 | block_start_pfn - pageblock_nr_pages; | |
977 | ||
be976572 VB |
978 | /* |
979 | * isolate_freepages_block() might have aborted due to async | |
980 | * compaction being contended | |
981 | */ | |
982 | if (cc->contended) | |
983 | break; | |
ff9543fd MN |
984 | } |
985 | ||
986 | /* split_free_page does not map the pages */ | |
987 | map_pages(freelist); | |
988 | ||
7ed695e0 VB |
989 | /* |
990 | * If we crossed the migrate scanner, we want to keep it that way | |
991 | * so that compact_finished() may detect this | |
992 | */ | |
c96b9e50 | 993 | if (block_start_pfn < low_pfn) |
e9ade569 | 994 | cc->free_pfn = cc->migrate_pfn; |
748446bb MG |
995 | } |
996 | ||
997 | /* | |
998 | * This is a migrate-callback that "allocates" freepages by taking pages | |
999 | * from the isolated freelists in the block we are migrating to. | |
1000 | */ | |
1001 | static struct page *compaction_alloc(struct page *migratepage, | |
1002 | unsigned long data, | |
1003 | int **result) | |
1004 | { | |
1005 | struct compact_control *cc = (struct compact_control *)data; | |
1006 | struct page *freepage; | |
1007 | ||
be976572 VB |
1008 | /* |
1009 | * Isolate free pages if necessary, and if we are not aborting due to | |
1010 | * contention. | |
1011 | */ | |
748446bb | 1012 | if (list_empty(&cc->freepages)) { |
be976572 | 1013 | if (!cc->contended) |
edc2ca61 | 1014 | isolate_freepages(cc); |
748446bb MG |
1015 | |
1016 | if (list_empty(&cc->freepages)) | |
1017 | return NULL; | |
1018 | } | |
1019 | ||
1020 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
1021 | list_del(&freepage->lru); | |
1022 | cc->nr_freepages--; | |
1023 | ||
1024 | return freepage; | |
1025 | } | |
1026 | ||
1027 | /* | |
d53aea3d DR |
1028 | * This is a migrate-callback that "frees" freepages back to the isolated |
1029 | * freelist. All pages on the freelist are from the same zone, so there is no | |
1030 | * special handling needed for NUMA. | |
1031 | */ | |
1032 | static void compaction_free(struct page *page, unsigned long data) | |
1033 | { | |
1034 | struct compact_control *cc = (struct compact_control *)data; | |
1035 | ||
1036 | list_add(&page->lru, &cc->freepages); | |
1037 | cc->nr_freepages++; | |
1038 | } | |
1039 | ||
ff9543fd MN |
1040 | /* possible outcome of isolate_migratepages */ |
1041 | typedef enum { | |
1042 | ISOLATE_ABORT, /* Abort compaction now */ | |
1043 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
1044 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
1045 | } isolate_migrate_t; | |
1046 | ||
1047 | /* | |
edc2ca61 VB |
1048 | * Isolate all pages that can be migrated from the first suitable block, |
1049 | * starting at the block pointed to by the migrate scanner pfn within | |
1050 | * compact_control. | |
ff9543fd MN |
1051 | */ |
1052 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
1053 | struct compact_control *cc) | |
1054 | { | |
1055 | unsigned long low_pfn, end_pfn; | |
edc2ca61 VB |
1056 | struct page *page; |
1057 | const isolate_mode_t isolate_mode = | |
1058 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); | |
ff9543fd | 1059 | |
edc2ca61 VB |
1060 | /* |
1061 | * Start at where we last stopped, or beginning of the zone as | |
1062 | * initialized by compact_zone() | |
1063 | */ | |
1064 | low_pfn = cc->migrate_pfn; | |
ff9543fd MN |
1065 | |
1066 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 1067 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd | 1068 | |
edc2ca61 VB |
1069 | /* |
1070 | * Iterate over whole pageblocks until we find the first suitable. | |
1071 | * Do not cross the free scanner. | |
1072 | */ | |
1073 | for (; end_pfn <= cc->free_pfn; | |
1074 | low_pfn = end_pfn, end_pfn += pageblock_nr_pages) { | |
ff9543fd | 1075 | |
edc2ca61 VB |
1076 | /* |
1077 | * This can potentially iterate a massively long zone with | |
1078 | * many pageblocks unsuitable, so periodically check if we | |
1079 | * need to schedule, or even abort async compaction. | |
1080 | */ | |
1081 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) | |
1082 | && compact_should_abort(cc)) | |
1083 | break; | |
ff9543fd | 1084 | |
7d49d886 VB |
1085 | page = pageblock_pfn_to_page(low_pfn, end_pfn, zone); |
1086 | if (!page) | |
edc2ca61 VB |
1087 | continue; |
1088 | ||
edc2ca61 VB |
1089 | /* If isolation recently failed, do not retry */ |
1090 | if (!isolation_suitable(cc, page)) | |
1091 | continue; | |
1092 | ||
1093 | /* | |
1094 | * For async compaction, also only scan in MOVABLE blocks. | |
1095 | * Async compaction is optimistic to see if the minimum amount | |
1096 | * of work satisfies the allocation. | |
1097 | */ | |
1098 | if (cc->mode == MIGRATE_ASYNC && | |
1099 | !migrate_async_suitable(get_pageblock_migratetype(page))) | |
1100 | continue; | |
1101 | ||
1102 | /* Perform the isolation */ | |
1103 | low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn, | |
1104 | isolate_mode); | |
1105 | ||
ff59909a HD |
1106 | if (!low_pfn || cc->contended) { |
1107 | acct_isolated(zone, cc); | |
edc2ca61 | 1108 | return ISOLATE_ABORT; |
ff59909a | 1109 | } |
edc2ca61 VB |
1110 | |
1111 | /* | |
1112 | * Either we isolated something and proceed with migration. Or | |
1113 | * we failed and compact_zone should decide if we should | |
1114 | * continue or not. | |
1115 | */ | |
1116 | break; | |
1117 | } | |
1118 | ||
1119 | acct_isolated(zone, cc); | |
1d5bfe1f VB |
1120 | /* |
1121 | * Record where migration scanner will be restarted. If we end up in | |
1122 | * the same pageblock as the free scanner, make the scanners fully | |
1123 | * meet so that compact_finished() terminates compaction. | |
1124 | */ | |
1125 | cc->migrate_pfn = (end_pfn <= cc->free_pfn) ? low_pfn : cc->free_pfn; | |
ff9543fd | 1126 | |
edc2ca61 | 1127 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
ff9543fd MN |
1128 | } |
1129 | ||
837d026d | 1130 | static int __compact_finished(struct zone *zone, struct compact_control *cc, |
6d7ce559 | 1131 | const int migratetype) |
748446bb | 1132 | { |
8fb74b9f | 1133 | unsigned int order; |
5a03b051 | 1134 | unsigned long watermark; |
56de7263 | 1135 | |
be976572 | 1136 | if (cc->contended || fatal_signal_pending(current)) |
748446bb MG |
1137 | return COMPACT_PARTIAL; |
1138 | ||
753341a4 | 1139 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 1140 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 | 1141 | /* Let the next compaction start anew. */ |
35979ef3 DR |
1142 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
1143 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
55b7c4c9 VB |
1144 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
1145 | ||
62997027 MG |
1146 | /* |
1147 | * Mark that the PG_migrate_skip information should be cleared | |
1148 | * by kswapd when it goes to sleep. kswapd does not set the | |
1149 | * flag itself as the decision to be clear should be directly | |
1150 | * based on an allocation request. | |
1151 | */ | |
1152 | if (!current_is_kswapd()) | |
1153 | zone->compact_blockskip_flush = true; | |
1154 | ||
748446bb | 1155 | return COMPACT_COMPLETE; |
bb13ffeb | 1156 | } |
748446bb | 1157 | |
82478fb7 JW |
1158 | /* |
1159 | * order == -1 is expected when compacting via | |
1160 | * /proc/sys/vm/compact_memory | |
1161 | */ | |
56de7263 MG |
1162 | if (cc->order == -1) |
1163 | return COMPACT_CONTINUE; | |
1164 | ||
3957c776 MH |
1165 | /* Compaction run is not finished if the watermark is not met */ |
1166 | watermark = low_wmark_pages(zone); | |
3957c776 | 1167 | |
ebff3980 VB |
1168 | if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx, |
1169 | cc->alloc_flags)) | |
3957c776 MH |
1170 | return COMPACT_CONTINUE; |
1171 | ||
56de7263 | 1172 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
1173 | for (order = cc->order; order < MAX_ORDER; order++) { |
1174 | struct free_area *area = &zone->free_area[order]; | |
1175 | ||
1176 | /* Job done if page is free of the right migratetype */ | |
6d7ce559 | 1177 | if (!list_empty(&area->free_list[migratetype])) |
8fb74b9f MG |
1178 | return COMPACT_PARTIAL; |
1179 | ||
1180 | /* Job done if allocation would set block type */ | |
372549c2 | 1181 | if (order >= pageblock_order && area->nr_free) |
56de7263 MG |
1182 | return COMPACT_PARTIAL; |
1183 | } | |
1184 | ||
837d026d JK |
1185 | return COMPACT_NO_SUITABLE_PAGE; |
1186 | } | |
1187 | ||
1188 | static int compact_finished(struct zone *zone, struct compact_control *cc, | |
1189 | const int migratetype) | |
1190 | { | |
1191 | int ret; | |
1192 | ||
1193 | ret = __compact_finished(zone, cc, migratetype); | |
1194 | trace_mm_compaction_finished(zone, cc->order, ret); | |
1195 | if (ret == COMPACT_NO_SUITABLE_PAGE) | |
1196 | ret = COMPACT_CONTINUE; | |
1197 | ||
1198 | return ret; | |
748446bb MG |
1199 | } |
1200 | ||
3e7d3449 MG |
1201 | /* |
1202 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
1203 | * Returns | |
1204 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
1205 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
1206 | * COMPACT_CONTINUE - If compaction should run now | |
1207 | */ | |
837d026d | 1208 | static unsigned long __compaction_suitable(struct zone *zone, int order, |
ebff3980 | 1209 | int alloc_flags, int classzone_idx) |
3e7d3449 MG |
1210 | { |
1211 | int fragindex; | |
1212 | unsigned long watermark; | |
1213 | ||
3957c776 MH |
1214 | /* |
1215 | * order == -1 is expected when compacting via | |
1216 | * /proc/sys/vm/compact_memory | |
1217 | */ | |
1218 | if (order == -1) | |
1219 | return COMPACT_CONTINUE; | |
1220 | ||
ebff3980 VB |
1221 | watermark = low_wmark_pages(zone); |
1222 | /* | |
1223 | * If watermarks for high-order allocation are already met, there | |
1224 | * should be no need for compaction at all. | |
1225 | */ | |
1226 | if (zone_watermark_ok(zone, order, watermark, classzone_idx, | |
1227 | alloc_flags)) | |
1228 | return COMPACT_PARTIAL; | |
1229 | ||
3e7d3449 MG |
1230 | /* |
1231 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
1232 | * This is because during migration, copies of pages need to be | |
1233 | * allocated and for a short time, the footprint is higher | |
1234 | */ | |
ebff3980 VB |
1235 | watermark += (2UL << order); |
1236 | if (!zone_watermark_ok(zone, 0, watermark, classzone_idx, alloc_flags)) | |
3e7d3449 MG |
1237 | return COMPACT_SKIPPED; |
1238 | ||
1239 | /* | |
1240 | * fragmentation index determines if allocation failures are due to | |
1241 | * low memory or external fragmentation | |
1242 | * | |
ebff3980 VB |
1243 | * index of -1000 would imply allocations might succeed depending on |
1244 | * watermarks, but we already failed the high-order watermark check | |
3e7d3449 MG |
1245 | * index towards 0 implies failure is due to lack of memory |
1246 | * index towards 1000 implies failure is due to fragmentation | |
1247 | * | |
1248 | * Only compact if a failure would be due to fragmentation. | |
1249 | */ | |
1250 | fragindex = fragmentation_index(zone, order); | |
1251 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
837d026d | 1252 | return COMPACT_NOT_SUITABLE_ZONE; |
3e7d3449 | 1253 | |
3e7d3449 MG |
1254 | return COMPACT_CONTINUE; |
1255 | } | |
1256 | ||
837d026d JK |
1257 | unsigned long compaction_suitable(struct zone *zone, int order, |
1258 | int alloc_flags, int classzone_idx) | |
1259 | { | |
1260 | unsigned long ret; | |
1261 | ||
1262 | ret = __compaction_suitable(zone, order, alloc_flags, classzone_idx); | |
1263 | trace_mm_compaction_suitable(zone, order, ret); | |
1264 | if (ret == COMPACT_NOT_SUITABLE_ZONE) | |
1265 | ret = COMPACT_SKIPPED; | |
1266 | ||
1267 | return ret; | |
1268 | } | |
1269 | ||
748446bb MG |
1270 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
1271 | { | |
1272 | int ret; | |
c89511ab | 1273 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 1274 | unsigned long end_pfn = zone_end_pfn(zone); |
6d7ce559 | 1275 | const int migratetype = gfpflags_to_migratetype(cc->gfp_mask); |
e0b9daeb | 1276 | const bool sync = cc->mode != MIGRATE_ASYNC; |
fdaf7f5c | 1277 | unsigned long last_migrated_pfn = 0; |
748446bb | 1278 | |
ebff3980 VB |
1279 | ret = compaction_suitable(zone, cc->order, cc->alloc_flags, |
1280 | cc->classzone_idx); | |
3e7d3449 MG |
1281 | switch (ret) { |
1282 | case COMPACT_PARTIAL: | |
1283 | case COMPACT_SKIPPED: | |
1284 | /* Compaction is likely to fail */ | |
1285 | return ret; | |
1286 | case COMPACT_CONTINUE: | |
1287 | /* Fall through to compaction */ | |
1288 | ; | |
1289 | } | |
1290 | ||
d3132e4b VB |
1291 | /* |
1292 | * Clear pageblock skip if there were failures recently and compaction | |
1293 | * is about to be retried after being deferred. kswapd does not do | |
1294 | * this reset as it'll reset the cached information when going to sleep. | |
1295 | */ | |
1296 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1297 | __reset_isolation_suitable(zone); | |
1298 | ||
c89511ab MG |
1299 | /* |
1300 | * Setup to move all movable pages to the end of the zone. Used cached | |
1301 | * information on where the scanners should start but check that it | |
1302 | * is initialised by ensuring the values are within zone boundaries. | |
1303 | */ | |
e0b9daeb | 1304 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
c89511ab MG |
1305 | cc->free_pfn = zone->compact_cached_free_pfn; |
1306 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
1307 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
1308 | zone->compact_cached_free_pfn = cc->free_pfn; | |
1309 | } | |
1310 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1311 | cc->migrate_pfn = start_pfn; | |
35979ef3 DR |
1312 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
1313 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
c89511ab | 1314 | } |
748446bb | 1315 | |
16c4a097 JK |
1316 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, |
1317 | cc->free_pfn, end_pfn, sync); | |
0eb927c0 | 1318 | |
748446bb MG |
1319 | migrate_prep_local(); |
1320 | ||
6d7ce559 DR |
1321 | while ((ret = compact_finished(zone, cc, migratetype)) == |
1322 | COMPACT_CONTINUE) { | |
9d502c1c | 1323 | int err; |
fdaf7f5c | 1324 | unsigned long isolate_start_pfn = cc->migrate_pfn; |
748446bb | 1325 | |
f9e35b3b MG |
1326 | switch (isolate_migratepages(zone, cc)) { |
1327 | case ISOLATE_ABORT: | |
1328 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1329 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1330 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1331 | goto out; |
1332 | case ISOLATE_NONE: | |
fdaf7f5c VB |
1333 | /* |
1334 | * We haven't isolated and migrated anything, but | |
1335 | * there might still be unflushed migrations from | |
1336 | * previous cc->order aligned block. | |
1337 | */ | |
1338 | goto check_drain; | |
f9e35b3b MG |
1339 | case ISOLATE_SUCCESS: |
1340 | ; | |
1341 | } | |
748446bb | 1342 | |
d53aea3d | 1343 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 1344 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 1345 | MR_COMPACTION); |
748446bb | 1346 | |
f8c9301f VB |
1347 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
1348 | &cc->migratepages); | |
748446bb | 1349 | |
f8c9301f VB |
1350 | /* All pages were either migrated or will be released */ |
1351 | cc->nr_migratepages = 0; | |
9d502c1c | 1352 | if (err) { |
5733c7d1 | 1353 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
1354 | /* |
1355 | * migrate_pages() may return -ENOMEM when scanners meet | |
1356 | * and we want compact_finished() to detect it | |
1357 | */ | |
1358 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1359 | ret = COMPACT_PARTIAL; |
1360 | goto out; | |
1361 | } | |
748446bb | 1362 | } |
fdaf7f5c VB |
1363 | |
1364 | /* | |
1365 | * Record where we could have freed pages by migration and not | |
1366 | * yet flushed them to buddy allocator. We use the pfn that | |
1367 | * isolate_migratepages() started from in this loop iteration | |
1368 | * - this is the lowest page that could have been isolated and | |
1369 | * then freed by migration. | |
1370 | */ | |
1371 | if (!last_migrated_pfn) | |
1372 | last_migrated_pfn = isolate_start_pfn; | |
1373 | ||
1374 | check_drain: | |
1375 | /* | |
1376 | * Has the migration scanner moved away from the previous | |
1377 | * cc->order aligned block where we migrated from? If yes, | |
1378 | * flush the pages that were freed, so that they can merge and | |
1379 | * compact_finished() can detect immediately if allocation | |
1380 | * would succeed. | |
1381 | */ | |
1382 | if (cc->order > 0 && last_migrated_pfn) { | |
1383 | int cpu; | |
1384 | unsigned long current_block_start = | |
1385 | cc->migrate_pfn & ~((1UL << cc->order) - 1); | |
1386 | ||
1387 | if (last_migrated_pfn < current_block_start) { | |
1388 | cpu = get_cpu(); | |
1389 | lru_add_drain_cpu(cpu); | |
1390 | drain_local_pages(zone); | |
1391 | put_cpu(); | |
1392 | /* No more flushing until we migrate again */ | |
1393 | last_migrated_pfn = 0; | |
1394 | } | |
1395 | } | |
1396 | ||
748446bb MG |
1397 | } |
1398 | ||
f9e35b3b | 1399 | out: |
6bace090 VB |
1400 | /* |
1401 | * Release free pages and update where the free scanner should restart, | |
1402 | * so we don't leave any returned pages behind in the next attempt. | |
1403 | */ | |
1404 | if (cc->nr_freepages > 0) { | |
1405 | unsigned long free_pfn = release_freepages(&cc->freepages); | |
1406 | ||
1407 | cc->nr_freepages = 0; | |
1408 | VM_BUG_ON(free_pfn == 0); | |
1409 | /* The cached pfn is always the first in a pageblock */ | |
1410 | free_pfn &= ~(pageblock_nr_pages-1); | |
1411 | /* | |
1412 | * Only go back, not forward. The cached pfn might have been | |
1413 | * already reset to zone end in compact_finished() | |
1414 | */ | |
1415 | if (free_pfn > zone->compact_cached_free_pfn) | |
1416 | zone->compact_cached_free_pfn = free_pfn; | |
1417 | } | |
748446bb | 1418 | |
16c4a097 JK |
1419 | trace_mm_compaction_end(start_pfn, cc->migrate_pfn, |
1420 | cc->free_pfn, end_pfn, sync, ret); | |
0eb927c0 | 1421 | |
748446bb MG |
1422 | return ret; |
1423 | } | |
76ab0f53 | 1424 | |
e0b9daeb | 1425 | static unsigned long compact_zone_order(struct zone *zone, int order, |
ebff3980 VB |
1426 | gfp_t gfp_mask, enum migrate_mode mode, int *contended, |
1427 | int alloc_flags, int classzone_idx) | |
56de7263 | 1428 | { |
e64c5237 | 1429 | unsigned long ret; |
56de7263 MG |
1430 | struct compact_control cc = { |
1431 | .nr_freepages = 0, | |
1432 | .nr_migratepages = 0, | |
1433 | .order = order, | |
6d7ce559 | 1434 | .gfp_mask = gfp_mask, |
56de7263 | 1435 | .zone = zone, |
e0b9daeb | 1436 | .mode = mode, |
ebff3980 VB |
1437 | .alloc_flags = alloc_flags, |
1438 | .classzone_idx = classzone_idx, | |
56de7263 MG |
1439 | }; |
1440 | INIT_LIST_HEAD(&cc.freepages); | |
1441 | INIT_LIST_HEAD(&cc.migratepages); | |
1442 | ||
e64c5237 SL |
1443 | ret = compact_zone(zone, &cc); |
1444 | ||
1445 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1446 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1447 | ||
1448 | *contended = cc.contended; | |
1449 | return ret; | |
56de7263 MG |
1450 | } |
1451 | ||
5e771905 MG |
1452 | int sysctl_extfrag_threshold = 500; |
1453 | ||
56de7263 MG |
1454 | /** |
1455 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
56de7263 | 1456 | * @gfp_mask: The GFP mask of the current allocation |
1a6d53a1 VB |
1457 | * @order: The order of the current allocation |
1458 | * @alloc_flags: The allocation flags of the current allocation | |
1459 | * @ac: The context of current allocation | |
e0b9daeb | 1460 | * @mode: The migration mode for async, sync light, or sync migration |
1f9efdef VB |
1461 | * @contended: Return value that determines if compaction was aborted due to |
1462 | * need_resched() or lock contention | |
56de7263 MG |
1463 | * |
1464 | * This is the main entry point for direct page compaction. | |
1465 | */ | |
1a6d53a1 VB |
1466 | unsigned long try_to_compact_pages(gfp_t gfp_mask, unsigned int order, |
1467 | int alloc_flags, const struct alloc_context *ac, | |
1468 | enum migrate_mode mode, int *contended) | |
56de7263 | 1469 | { |
56de7263 MG |
1470 | int may_enter_fs = gfp_mask & __GFP_FS; |
1471 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1472 | struct zoneref *z; |
1473 | struct zone *zone; | |
53853e2d | 1474 | int rc = COMPACT_DEFERRED; |
1f9efdef VB |
1475 | int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ |
1476 | ||
1477 | *contended = COMPACT_CONTENDED_NONE; | |
56de7263 | 1478 | |
4ffb6335 | 1479 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1480 | if (!order || !may_enter_fs || !may_perform_io) |
53853e2d | 1481 | return COMPACT_SKIPPED; |
56de7263 | 1482 | |
837d026d JK |
1483 | trace_mm_compaction_try_to_compact_pages(order, gfp_mask, mode); |
1484 | ||
56de7263 | 1485 | /* Compact each zone in the list */ |
1a6d53a1 VB |
1486 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
1487 | ac->nodemask) { | |
56de7263 | 1488 | int status; |
1f9efdef | 1489 | int zone_contended; |
56de7263 | 1490 | |
53853e2d VB |
1491 | if (compaction_deferred(zone, order)) |
1492 | continue; | |
1493 | ||
e0b9daeb | 1494 | status = compact_zone_order(zone, order, gfp_mask, mode, |
1a6d53a1 VB |
1495 | &zone_contended, alloc_flags, |
1496 | ac->classzone_idx); | |
56de7263 | 1497 | rc = max(status, rc); |
1f9efdef VB |
1498 | /* |
1499 | * It takes at least one zone that wasn't lock contended | |
1500 | * to clear all_zones_contended. | |
1501 | */ | |
1502 | all_zones_contended &= zone_contended; | |
56de7263 | 1503 | |
3e7d3449 | 1504 | /* If a normal allocation would succeed, stop compacting */ |
ebff3980 | 1505 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), |
1a6d53a1 | 1506 | ac->classzone_idx, alloc_flags)) { |
53853e2d VB |
1507 | /* |
1508 | * We think the allocation will succeed in this zone, | |
1509 | * but it is not certain, hence the false. The caller | |
1510 | * will repeat this with true if allocation indeed | |
1511 | * succeeds in this zone. | |
1512 | */ | |
1513 | compaction_defer_reset(zone, order, false); | |
1f9efdef VB |
1514 | /* |
1515 | * It is possible that async compaction aborted due to | |
1516 | * need_resched() and the watermarks were ok thanks to | |
1517 | * somebody else freeing memory. The allocation can | |
1518 | * however still fail so we better signal the | |
1519 | * need_resched() contention anyway (this will not | |
1520 | * prevent the allocation attempt). | |
1521 | */ | |
1522 | if (zone_contended == COMPACT_CONTENDED_SCHED) | |
1523 | *contended = COMPACT_CONTENDED_SCHED; | |
1524 | ||
1525 | goto break_loop; | |
1526 | } | |
1527 | ||
f8669795 | 1528 | if (mode != MIGRATE_ASYNC && status == COMPACT_COMPLETE) { |
53853e2d VB |
1529 | /* |
1530 | * We think that allocation won't succeed in this zone | |
1531 | * so we defer compaction there. If it ends up | |
1532 | * succeeding after all, it will be reset. | |
1533 | */ | |
1534 | defer_compaction(zone, order); | |
1535 | } | |
1f9efdef VB |
1536 | |
1537 | /* | |
1538 | * We might have stopped compacting due to need_resched() in | |
1539 | * async compaction, or due to a fatal signal detected. In that | |
1540 | * case do not try further zones and signal need_resched() | |
1541 | * contention. | |
1542 | */ | |
1543 | if ((zone_contended == COMPACT_CONTENDED_SCHED) | |
1544 | || fatal_signal_pending(current)) { | |
1545 | *contended = COMPACT_CONTENDED_SCHED; | |
1546 | goto break_loop; | |
1547 | } | |
1548 | ||
1549 | continue; | |
1550 | break_loop: | |
1551 | /* | |
1552 | * We might not have tried all the zones, so be conservative | |
1553 | * and assume they are not all lock contended. | |
1554 | */ | |
1555 | all_zones_contended = 0; | |
1556 | break; | |
56de7263 MG |
1557 | } |
1558 | ||
1f9efdef VB |
1559 | /* |
1560 | * If at least one zone wasn't deferred or skipped, we report if all | |
1561 | * zones that were tried were lock contended. | |
1562 | */ | |
1563 | if (rc > COMPACT_SKIPPED && all_zones_contended) | |
1564 | *contended = COMPACT_CONTENDED_LOCK; | |
1565 | ||
56de7263 MG |
1566 | return rc; |
1567 | } | |
1568 | ||
1569 | ||
76ab0f53 | 1570 | /* Compact all zones within a node */ |
7103f16d | 1571 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1572 | { |
1573 | int zoneid; | |
76ab0f53 MG |
1574 | struct zone *zone; |
1575 | ||
76ab0f53 | 1576 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1577 | |
1578 | zone = &pgdat->node_zones[zoneid]; | |
1579 | if (!populated_zone(zone)) | |
1580 | continue; | |
1581 | ||
7be62de9 RR |
1582 | cc->nr_freepages = 0; |
1583 | cc->nr_migratepages = 0; | |
1584 | cc->zone = zone; | |
1585 | INIT_LIST_HEAD(&cc->freepages); | |
1586 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1587 | |
aad6ec37 | 1588 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1589 | compact_zone(zone, cc); |
76ab0f53 | 1590 | |
aff62249 | 1591 | if (cc->order > 0) { |
de6c60a6 VB |
1592 | if (zone_watermark_ok(zone, cc->order, |
1593 | low_wmark_pages(zone), 0, 0)) | |
1594 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 RR |
1595 | } |
1596 | ||
7be62de9 RR |
1597 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1598 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1599 | } |
76ab0f53 MG |
1600 | } |
1601 | ||
7103f16d | 1602 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1603 | { |
1604 | struct compact_control cc = { | |
1605 | .order = order, | |
e0b9daeb | 1606 | .mode = MIGRATE_ASYNC, |
7be62de9 RR |
1607 | }; |
1608 | ||
3a7200af MG |
1609 | if (!order) |
1610 | return; | |
1611 | ||
7103f16d | 1612 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1613 | } |
1614 | ||
7103f16d | 1615 | static void compact_node(int nid) |
7be62de9 | 1616 | { |
7be62de9 RR |
1617 | struct compact_control cc = { |
1618 | .order = -1, | |
e0b9daeb | 1619 | .mode = MIGRATE_SYNC, |
91ca9186 | 1620 | .ignore_skip_hint = true, |
7be62de9 RR |
1621 | }; |
1622 | ||
7103f16d | 1623 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1624 | } |
1625 | ||
76ab0f53 | 1626 | /* Compact all nodes in the system */ |
7964c06d | 1627 | static void compact_nodes(void) |
76ab0f53 MG |
1628 | { |
1629 | int nid; | |
1630 | ||
8575ec29 HD |
1631 | /* Flush pending updates to the LRU lists */ |
1632 | lru_add_drain_all(); | |
1633 | ||
76ab0f53 MG |
1634 | for_each_online_node(nid) |
1635 | compact_node(nid); | |
76ab0f53 MG |
1636 | } |
1637 | ||
1638 | /* The written value is actually unused, all memory is compacted */ | |
1639 | int sysctl_compact_memory; | |
1640 | ||
1641 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1642 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1643 | void __user *buffer, size_t *length, loff_t *ppos) | |
1644 | { | |
1645 | if (write) | |
7964c06d | 1646 | compact_nodes(); |
76ab0f53 MG |
1647 | |
1648 | return 0; | |
1649 | } | |
ed4a6d7f | 1650 | |
5e771905 MG |
1651 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1652 | void __user *buffer, size_t *length, loff_t *ppos) | |
1653 | { | |
1654 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1655 | ||
1656 | return 0; | |
1657 | } | |
1658 | ||
ed4a6d7f | 1659 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1660 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1661 | struct device_attribute *attr, |
ed4a6d7f MG |
1662 | const char *buf, size_t count) |
1663 | { | |
8575ec29 HD |
1664 | int nid = dev->id; |
1665 | ||
1666 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1667 | /* Flush pending updates to the LRU lists */ | |
1668 | lru_add_drain_all(); | |
1669 | ||
1670 | compact_node(nid); | |
1671 | } | |
ed4a6d7f MG |
1672 | |
1673 | return count; | |
1674 | } | |
10fbcf4c | 1675 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1676 | |
1677 | int compaction_register_node(struct node *node) | |
1678 | { | |
10fbcf4c | 1679 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1680 | } |
1681 | ||
1682 | void compaction_unregister_node(struct node *node) | |
1683 | { | |
10fbcf4c | 1684 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1685 | } |
1686 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1687 | |
1688 | #endif /* CONFIG_COMPACTION */ |