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> |
b8c73fc2 | 19 | #include <linux/kasan.h> |
748446bb MG |
20 | #include "internal.h" |
21 | ||
010fc29a MK |
22 | #ifdef CONFIG_COMPACTION |
23 | static inline void count_compact_event(enum vm_event_item item) | |
24 | { | |
25 | count_vm_event(item); | |
26 | } | |
27 | ||
28 | static inline void count_compact_events(enum vm_event_item item, long delta) | |
29 | { | |
30 | count_vm_events(item, delta); | |
31 | } | |
32 | #else | |
33 | #define count_compact_event(item) do { } while (0) | |
34 | #define count_compact_events(item, delta) do { } while (0) | |
35 | #endif | |
36 | ||
ff9543fd | 37 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
16c4a097 JK |
38 | #ifdef CONFIG_TRACEPOINTS |
39 | static const char *const compaction_status_string[] = { | |
40 | "deferred", | |
41 | "skipped", | |
42 | "continue", | |
43 | "partial", | |
44 | "complete", | |
837d026d JK |
45 | "no_suitable_page", |
46 | "not_suitable_zone", | |
16c4a097 JK |
47 | }; |
48 | #endif | |
ff9543fd | 49 | |
b7aba698 MG |
50 | #define CREATE_TRACE_POINTS |
51 | #include <trace/events/compaction.h> | |
52 | ||
748446bb MG |
53 | static unsigned long release_freepages(struct list_head *freelist) |
54 | { | |
55 | struct page *page, *next; | |
6bace090 | 56 | unsigned long high_pfn = 0; |
748446bb MG |
57 | |
58 | list_for_each_entry_safe(page, next, freelist, lru) { | |
6bace090 | 59 | unsigned long pfn = page_to_pfn(page); |
748446bb MG |
60 | list_del(&page->lru); |
61 | __free_page(page); | |
6bace090 VB |
62 | if (pfn > high_pfn) |
63 | high_pfn = pfn; | |
748446bb MG |
64 | } |
65 | ||
6bace090 | 66 | return high_pfn; |
748446bb MG |
67 | } |
68 | ||
ff9543fd MN |
69 | static void map_pages(struct list_head *list) |
70 | { | |
71 | struct page *page; | |
72 | ||
73 | list_for_each_entry(page, list, lru) { | |
74 | arch_alloc_page(page, 0); | |
75 | kernel_map_pages(page, 1, 1); | |
b8c73fc2 | 76 | kasan_alloc_pages(page, 0); |
ff9543fd MN |
77 | } |
78 | } | |
79 | ||
47118af0 MN |
80 | static inline bool migrate_async_suitable(int migratetype) |
81 | { | |
82 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; | |
83 | } | |
84 | ||
7d49d886 VB |
85 | /* |
86 | * Check that the whole (or subset of) a pageblock given by the interval of | |
87 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
88 | * with the migration of free compaction scanner. The scanners then need to | |
89 | * use only pfn_valid_within() check for arches that allow holes within | |
90 | * pageblocks. | |
91 | * | |
92 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
93 | * | |
94 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
95 | * i.e. it's possible that all pages within a zones range of pages do not | |
96 | * belong to a single zone. We assume that a border between node0 and node1 | |
97 | * can occur within a single pageblock, but not a node0 node1 node0 | |
98 | * interleaving within a single pageblock. It is therefore sufficient to check | |
99 | * the first and last page of a pageblock and avoid checking each individual | |
100 | * page in a pageblock. | |
101 | */ | |
102 | static struct page *pageblock_pfn_to_page(unsigned long start_pfn, | |
103 | unsigned long end_pfn, struct zone *zone) | |
104 | { | |
105 | struct page *start_page; | |
106 | struct page *end_page; | |
107 | ||
108 | /* end_pfn is one past the range we are checking */ | |
109 | end_pfn--; | |
110 | ||
111 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
112 | return NULL; | |
113 | ||
114 | start_page = pfn_to_page(start_pfn); | |
115 | ||
116 | if (page_zone(start_page) != zone) | |
117 | return NULL; | |
118 | ||
119 | end_page = pfn_to_page(end_pfn); | |
120 | ||
121 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
122 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
123 | return NULL; | |
124 | ||
125 | return start_page; | |
126 | } | |
127 | ||
bb13ffeb | 128 | #ifdef CONFIG_COMPACTION |
24e2716f JK |
129 | |
130 | /* Do not skip compaction more than 64 times */ | |
131 | #define COMPACT_MAX_DEFER_SHIFT 6 | |
132 | ||
133 | /* | |
134 | * Compaction is deferred when compaction fails to result in a page | |
135 | * allocation success. 1 << compact_defer_limit compactions are skipped up | |
136 | * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT | |
137 | */ | |
138 | void defer_compaction(struct zone *zone, int order) | |
139 | { | |
140 | zone->compact_considered = 0; | |
141 | zone->compact_defer_shift++; | |
142 | ||
143 | if (order < zone->compact_order_failed) | |
144 | zone->compact_order_failed = order; | |
145 | ||
146 | if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT) | |
147 | zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT; | |
148 | ||
149 | trace_mm_compaction_defer_compaction(zone, order); | |
150 | } | |
151 | ||
152 | /* Returns true if compaction should be skipped this time */ | |
153 | bool compaction_deferred(struct zone *zone, int order) | |
154 | { | |
155 | unsigned long defer_limit = 1UL << zone->compact_defer_shift; | |
156 | ||
157 | if (order < zone->compact_order_failed) | |
158 | return false; | |
159 | ||
160 | /* Avoid possible overflow */ | |
161 | if (++zone->compact_considered > defer_limit) | |
162 | zone->compact_considered = defer_limit; | |
163 | ||
164 | if (zone->compact_considered >= defer_limit) | |
165 | return false; | |
166 | ||
167 | trace_mm_compaction_deferred(zone, order); | |
168 | ||
169 | return true; | |
170 | } | |
171 | ||
172 | /* | |
173 | * Update defer tracking counters after successful compaction of given order, | |
174 | * which means an allocation either succeeded (alloc_success == true) or is | |
175 | * expected to succeed. | |
176 | */ | |
177 | void compaction_defer_reset(struct zone *zone, int order, | |
178 | bool alloc_success) | |
179 | { | |
180 | if (alloc_success) { | |
181 | zone->compact_considered = 0; | |
182 | zone->compact_defer_shift = 0; | |
183 | } | |
184 | if (order >= zone->compact_order_failed) | |
185 | zone->compact_order_failed = order + 1; | |
186 | ||
187 | trace_mm_compaction_defer_reset(zone, order); | |
188 | } | |
189 | ||
190 | /* Returns true if restarting compaction after many failures */ | |
191 | bool compaction_restarting(struct zone *zone, int order) | |
192 | { | |
193 | if (order < zone->compact_order_failed) | |
194 | return false; | |
195 | ||
196 | return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT && | |
197 | zone->compact_considered >= 1UL << zone->compact_defer_shift; | |
198 | } | |
199 | ||
bb13ffeb MG |
200 | /* Returns true if the pageblock should be scanned for pages to isolate. */ |
201 | static inline bool isolation_suitable(struct compact_control *cc, | |
202 | struct page *page) | |
203 | { | |
204 | if (cc->ignore_skip_hint) | |
205 | return true; | |
206 | ||
207 | return !get_pageblock_skip(page); | |
208 | } | |
209 | ||
210 | /* | |
211 | * This function is called to clear all cached information on pageblocks that | |
212 | * should be skipped for page isolation when the migrate and free page scanner | |
213 | * meet. | |
214 | */ | |
62997027 | 215 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
216 | { |
217 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 218 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
219 | unsigned long pfn; |
220 | ||
35979ef3 DR |
221 | zone->compact_cached_migrate_pfn[0] = start_pfn; |
222 | zone->compact_cached_migrate_pfn[1] = start_pfn; | |
c89511ab | 223 | zone->compact_cached_free_pfn = end_pfn; |
62997027 | 224 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
225 | |
226 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
227 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
228 | struct page *page; | |
229 | ||
230 | cond_resched(); | |
231 | ||
232 | if (!pfn_valid(pfn)) | |
233 | continue; | |
234 | ||
235 | page = pfn_to_page(pfn); | |
236 | if (zone != page_zone(page)) | |
237 | continue; | |
238 | ||
239 | clear_pageblock_skip(page); | |
240 | } | |
241 | } | |
242 | ||
62997027 MG |
243 | void reset_isolation_suitable(pg_data_t *pgdat) |
244 | { | |
245 | int zoneid; | |
246 | ||
247 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
248 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
249 | if (!populated_zone(zone)) | |
250 | continue; | |
251 | ||
252 | /* Only flush if a full compaction finished recently */ | |
253 | if (zone->compact_blockskip_flush) | |
254 | __reset_isolation_suitable(zone); | |
255 | } | |
256 | } | |
257 | ||
bb13ffeb MG |
258 | /* |
259 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 260 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 261 | */ |
c89511ab MG |
262 | static void update_pageblock_skip(struct compact_control *cc, |
263 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 264 | bool migrate_scanner) |
bb13ffeb | 265 | { |
c89511ab | 266 | struct zone *zone = cc->zone; |
35979ef3 | 267 | unsigned long pfn; |
6815bf3f JK |
268 | |
269 | if (cc->ignore_skip_hint) | |
270 | return; | |
271 | ||
bb13ffeb MG |
272 | if (!page) |
273 | return; | |
274 | ||
35979ef3 DR |
275 | if (nr_isolated) |
276 | return; | |
277 | ||
edc2ca61 | 278 | set_pageblock_skip(page); |
c89511ab | 279 | |
35979ef3 DR |
280 | pfn = page_to_pfn(page); |
281 | ||
282 | /* Update where async and sync compaction should restart */ | |
283 | if (migrate_scanner) { | |
35979ef3 DR |
284 | if (pfn > zone->compact_cached_migrate_pfn[0]) |
285 | zone->compact_cached_migrate_pfn[0] = pfn; | |
e0b9daeb DR |
286 | if (cc->mode != MIGRATE_ASYNC && |
287 | pfn > zone->compact_cached_migrate_pfn[1]) | |
35979ef3 DR |
288 | zone->compact_cached_migrate_pfn[1] = pfn; |
289 | } else { | |
35979ef3 DR |
290 | if (pfn < zone->compact_cached_free_pfn) |
291 | zone->compact_cached_free_pfn = pfn; | |
c89511ab | 292 | } |
bb13ffeb MG |
293 | } |
294 | #else | |
295 | static inline bool isolation_suitable(struct compact_control *cc, | |
296 | struct page *page) | |
297 | { | |
298 | return true; | |
299 | } | |
300 | ||
c89511ab MG |
301 | static void update_pageblock_skip(struct compact_control *cc, |
302 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 303 | bool migrate_scanner) |
bb13ffeb MG |
304 | { |
305 | } | |
306 | #endif /* CONFIG_COMPACTION */ | |
307 | ||
8b44d279 VB |
308 | /* |
309 | * Compaction requires the taking of some coarse locks that are potentially | |
310 | * very heavily contended. For async compaction, back out if the lock cannot | |
311 | * be taken immediately. For sync compaction, spin on the lock if needed. | |
312 | * | |
313 | * Returns true if the lock is held | |
314 | * Returns false if the lock is not held and compaction should abort | |
315 | */ | |
316 | static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags, | |
317 | struct compact_control *cc) | |
2a1402aa | 318 | { |
8b44d279 VB |
319 | if (cc->mode == MIGRATE_ASYNC) { |
320 | if (!spin_trylock_irqsave(lock, *flags)) { | |
321 | cc->contended = COMPACT_CONTENDED_LOCK; | |
322 | return false; | |
323 | } | |
324 | } else { | |
325 | spin_lock_irqsave(lock, *flags); | |
326 | } | |
1f9efdef | 327 | |
8b44d279 | 328 | return true; |
2a1402aa MG |
329 | } |
330 | ||
c67fe375 MG |
331 | /* |
332 | * Compaction requires the taking of some coarse locks that are potentially | |
8b44d279 VB |
333 | * very heavily contended. The lock should be periodically unlocked to avoid |
334 | * having disabled IRQs for a long time, even when there is nobody waiting on | |
335 | * the lock. It might also be that allowing the IRQs will result in | |
336 | * need_resched() becoming true. If scheduling is needed, async compaction | |
337 | * aborts. Sync compaction schedules. | |
338 | * Either compaction type will also abort if a fatal signal is pending. | |
339 | * In either case if the lock was locked, it is dropped and not regained. | |
c67fe375 | 340 | * |
8b44d279 VB |
341 | * Returns true if compaction should abort due to fatal signal pending, or |
342 | * async compaction due to need_resched() | |
343 | * Returns false when compaction can continue (sync compaction might have | |
344 | * scheduled) | |
c67fe375 | 345 | */ |
8b44d279 VB |
346 | static bool compact_unlock_should_abort(spinlock_t *lock, |
347 | unsigned long flags, bool *locked, struct compact_control *cc) | |
c67fe375 | 348 | { |
8b44d279 VB |
349 | if (*locked) { |
350 | spin_unlock_irqrestore(lock, flags); | |
351 | *locked = false; | |
352 | } | |
1f9efdef | 353 | |
8b44d279 VB |
354 | if (fatal_signal_pending(current)) { |
355 | cc->contended = COMPACT_CONTENDED_SCHED; | |
356 | return true; | |
357 | } | |
c67fe375 | 358 | |
8b44d279 | 359 | if (need_resched()) { |
e0b9daeb | 360 | if (cc->mode == MIGRATE_ASYNC) { |
8b44d279 VB |
361 | cc->contended = COMPACT_CONTENDED_SCHED; |
362 | return true; | |
c67fe375 | 363 | } |
c67fe375 | 364 | cond_resched(); |
c67fe375 MG |
365 | } |
366 | ||
8b44d279 | 367 | return false; |
c67fe375 MG |
368 | } |
369 | ||
be976572 VB |
370 | /* |
371 | * Aside from avoiding lock contention, compaction also periodically checks | |
372 | * need_resched() and either schedules in sync compaction or aborts async | |
8b44d279 | 373 | * compaction. This is similar to what compact_unlock_should_abort() does, but |
be976572 VB |
374 | * is used where no lock is concerned. |
375 | * | |
376 | * Returns false when no scheduling was needed, or sync compaction scheduled. | |
377 | * Returns true when async compaction should abort. | |
378 | */ | |
379 | static inline bool compact_should_abort(struct compact_control *cc) | |
380 | { | |
381 | /* async compaction aborts if contended */ | |
382 | if (need_resched()) { | |
383 | if (cc->mode == MIGRATE_ASYNC) { | |
1f9efdef | 384 | cc->contended = COMPACT_CONTENDED_SCHED; |
be976572 VB |
385 | return true; |
386 | } | |
387 | ||
388 | cond_resched(); | |
389 | } | |
390 | ||
391 | return false; | |
392 | } | |
393 | ||
85aa125f | 394 | /* |
9e4be470 JM |
395 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
396 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
397 | * (even though it may still end up isolating some pages). | |
85aa125f | 398 | */ |
f40d1e42 | 399 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
e14c720e | 400 | unsigned long *start_pfn, |
85aa125f MN |
401 | unsigned long end_pfn, |
402 | struct list_head *freelist, | |
403 | bool strict) | |
748446bb | 404 | { |
b7aba698 | 405 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 406 | struct page *cursor, *valid_page = NULL; |
b8b2d825 | 407 | unsigned long flags = 0; |
f40d1e42 | 408 | bool locked = false; |
e14c720e | 409 | unsigned long blockpfn = *start_pfn; |
748446bb | 410 | |
748446bb MG |
411 | cursor = pfn_to_page(blockpfn); |
412 | ||
f40d1e42 | 413 | /* Isolate free pages. */ |
748446bb MG |
414 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
415 | int isolated, i; | |
416 | struct page *page = cursor; | |
417 | ||
8b44d279 VB |
418 | /* |
419 | * Periodically drop the lock (if held) regardless of its | |
420 | * contention, to give chance to IRQs. Abort if fatal signal | |
421 | * pending or async compaction detects need_resched() | |
422 | */ | |
423 | if (!(blockpfn % SWAP_CLUSTER_MAX) | |
424 | && compact_unlock_should_abort(&cc->zone->lock, flags, | |
425 | &locked, cc)) | |
426 | break; | |
427 | ||
b7aba698 | 428 | nr_scanned++; |
f40d1e42 | 429 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
430 | goto isolate_fail; |
431 | ||
bb13ffeb MG |
432 | if (!valid_page) |
433 | valid_page = page; | |
f40d1e42 | 434 | if (!PageBuddy(page)) |
2af120bc | 435 | goto isolate_fail; |
f40d1e42 MG |
436 | |
437 | /* | |
69b7189f VB |
438 | * If we already hold the lock, we can skip some rechecking. |
439 | * Note that if we hold the lock now, checked_pageblock was | |
440 | * already set in some previous iteration (or strict is true), | |
441 | * so it is correct to skip the suitable migration target | |
442 | * recheck as well. | |
f40d1e42 | 443 | */ |
69b7189f VB |
444 | if (!locked) { |
445 | /* | |
446 | * The zone lock must be held to isolate freepages. | |
447 | * Unfortunately this is a very coarse lock and can be | |
448 | * heavily contended if there are parallel allocations | |
449 | * or parallel compactions. For async compaction do not | |
450 | * spin on the lock and we acquire the lock as late as | |
451 | * possible. | |
452 | */ | |
8b44d279 VB |
453 | locked = compact_trylock_irqsave(&cc->zone->lock, |
454 | &flags, cc); | |
69b7189f VB |
455 | if (!locked) |
456 | break; | |
f40d1e42 | 457 | |
69b7189f VB |
458 | /* Recheck this is a buddy page under lock */ |
459 | if (!PageBuddy(page)) | |
460 | goto isolate_fail; | |
461 | } | |
748446bb MG |
462 | |
463 | /* Found a free page, break it into order-0 pages */ | |
464 | isolated = split_free_page(page); | |
465 | total_isolated += isolated; | |
466 | for (i = 0; i < isolated; i++) { | |
467 | list_add(&page->lru, freelist); | |
468 | page++; | |
469 | } | |
470 | ||
471 | /* If a page was split, advance to the end of it */ | |
472 | if (isolated) { | |
932ff6bb JK |
473 | cc->nr_freepages += isolated; |
474 | if (!strict && | |
475 | cc->nr_migratepages <= cc->nr_freepages) { | |
476 | blockpfn += isolated; | |
477 | break; | |
478 | } | |
479 | ||
748446bb MG |
480 | blockpfn += isolated - 1; |
481 | cursor += isolated - 1; | |
2af120bc | 482 | continue; |
748446bb | 483 | } |
2af120bc LA |
484 | |
485 | isolate_fail: | |
486 | if (strict) | |
487 | break; | |
488 | else | |
489 | continue; | |
490 | ||
748446bb MG |
491 | } |
492 | ||
e34d85f0 JK |
493 | trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn, |
494 | nr_scanned, total_isolated); | |
495 | ||
e14c720e VB |
496 | /* Record how far we have got within the block */ |
497 | *start_pfn = blockpfn; | |
498 | ||
f40d1e42 MG |
499 | /* |
500 | * If strict isolation is requested by CMA then check that all the | |
501 | * pages requested were isolated. If there were any failures, 0 is | |
502 | * returned and CMA will fail. | |
503 | */ | |
2af120bc | 504 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
505 | total_isolated = 0; |
506 | ||
507 | if (locked) | |
508 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
509 | ||
bb13ffeb MG |
510 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
511 | if (blockpfn == end_pfn) | |
edc2ca61 | 512 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 513 | |
010fc29a | 514 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 515 | if (total_isolated) |
010fc29a | 516 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
517 | return total_isolated; |
518 | } | |
519 | ||
85aa125f MN |
520 | /** |
521 | * isolate_freepages_range() - isolate free pages. | |
522 | * @start_pfn: The first PFN to start isolating. | |
523 | * @end_pfn: The one-past-last PFN. | |
524 | * | |
525 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
526 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
527 | * undo its actions and return zero. | |
528 | * | |
529 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
530 | * (which may be greater then end_pfn if end fell in a middle of | |
531 | * a free page). | |
532 | */ | |
ff9543fd | 533 | unsigned long |
bb13ffeb MG |
534 | isolate_freepages_range(struct compact_control *cc, |
535 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 536 | { |
f40d1e42 | 537 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
538 | LIST_HEAD(freelist); |
539 | ||
7d49d886 VB |
540 | pfn = start_pfn; |
541 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
542 | ||
543 | for (; pfn < end_pfn; pfn += isolated, | |
544 | block_end_pfn += pageblock_nr_pages) { | |
e14c720e VB |
545 | /* Protect pfn from changing by isolate_freepages_block */ |
546 | unsigned long isolate_start_pfn = pfn; | |
85aa125f | 547 | |
85aa125f MN |
548 | block_end_pfn = min(block_end_pfn, end_pfn); |
549 | ||
58420016 JK |
550 | /* |
551 | * pfn could pass the block_end_pfn if isolated freepage | |
552 | * is more than pageblock order. In this case, we adjust | |
553 | * scanning range to right one. | |
554 | */ | |
555 | if (pfn >= block_end_pfn) { | |
556 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
557 | block_end_pfn = min(block_end_pfn, end_pfn); | |
558 | } | |
559 | ||
7d49d886 VB |
560 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
561 | break; | |
562 | ||
e14c720e VB |
563 | isolated = isolate_freepages_block(cc, &isolate_start_pfn, |
564 | block_end_pfn, &freelist, true); | |
85aa125f MN |
565 | |
566 | /* | |
567 | * In strict mode, isolate_freepages_block() returns 0 if | |
568 | * there are any holes in the block (ie. invalid PFNs or | |
569 | * non-free pages). | |
570 | */ | |
571 | if (!isolated) | |
572 | break; | |
573 | ||
574 | /* | |
575 | * If we managed to isolate pages, it is always (1 << n) * | |
576 | * pageblock_nr_pages for some non-negative n. (Max order | |
577 | * page may span two pageblocks). | |
578 | */ | |
579 | } | |
580 | ||
581 | /* split_free_page does not map the pages */ | |
582 | map_pages(&freelist); | |
583 | ||
584 | if (pfn < end_pfn) { | |
585 | /* Loop terminated early, cleanup. */ | |
586 | release_freepages(&freelist); | |
587 | return 0; | |
588 | } | |
589 | ||
590 | /* We don't use freelists for anything. */ | |
591 | return pfn; | |
592 | } | |
593 | ||
748446bb | 594 | /* Update the number of anon and file isolated pages in the zone */ |
edc2ca61 | 595 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
748446bb MG |
596 | { |
597 | struct page *page; | |
b9e84ac1 | 598 | unsigned int count[2] = { 0, }; |
748446bb | 599 | |
edc2ca61 VB |
600 | if (list_empty(&cc->migratepages)) |
601 | return; | |
602 | ||
b9e84ac1 MK |
603 | list_for_each_entry(page, &cc->migratepages, lru) |
604 | count[!!page_is_file_cache(page)]++; | |
748446bb | 605 | |
edc2ca61 VB |
606 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
607 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
748446bb MG |
608 | } |
609 | ||
610 | /* Similar to reclaim, but different enough that they don't share logic */ | |
611 | static bool too_many_isolated(struct zone *zone) | |
612 | { | |
bc693045 | 613 | unsigned long active, inactive, isolated; |
748446bb MG |
614 | |
615 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
616 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
617 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
618 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
619 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
620 | zone_page_state(zone, NR_ISOLATED_ANON); | |
621 | ||
bc693045 | 622 | return isolated > (inactive + active) / 2; |
748446bb MG |
623 | } |
624 | ||
2fe86e00 | 625 | /** |
edc2ca61 VB |
626 | * isolate_migratepages_block() - isolate all migrate-able pages within |
627 | * a single pageblock | |
2fe86e00 | 628 | * @cc: Compaction control structure. |
edc2ca61 VB |
629 | * @low_pfn: The first PFN to isolate |
630 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock | |
631 | * @isolate_mode: Isolation mode to be used. | |
2fe86e00 MN |
632 | * |
633 | * Isolate all pages that can be migrated from the range specified by | |
edc2ca61 VB |
634 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
635 | * Returns zero if there is a fatal signal pending, otherwise PFN of the | |
636 | * first page that was not scanned (which may be both less, equal to or more | |
637 | * than end_pfn). | |
2fe86e00 | 638 | * |
edc2ca61 VB |
639 | * The pages are isolated on cc->migratepages list (not required to be empty), |
640 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field | |
641 | * is neither read nor updated. | |
748446bb | 642 | */ |
edc2ca61 VB |
643 | static unsigned long |
644 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, | |
645 | unsigned long end_pfn, isolate_mode_t isolate_mode) | |
748446bb | 646 | { |
edc2ca61 | 647 | struct zone *zone = cc->zone; |
b7aba698 | 648 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 649 | struct list_head *migratelist = &cc->migratepages; |
fa9add64 | 650 | struct lruvec *lruvec; |
b8b2d825 | 651 | unsigned long flags = 0; |
2a1402aa | 652 | bool locked = false; |
bb13ffeb | 653 | struct page *page = NULL, *valid_page = NULL; |
e34d85f0 | 654 | unsigned long start_pfn = low_pfn; |
748446bb | 655 | |
748446bb MG |
656 | /* |
657 | * Ensure that there are not too many pages isolated from the LRU | |
658 | * list by either parallel reclaimers or compaction. If there are, | |
659 | * delay for some time until fewer pages are isolated | |
660 | */ | |
661 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 662 | /* async migration should just abort */ |
e0b9daeb | 663 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 664 | return 0; |
f9e35b3b | 665 | |
748446bb MG |
666 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
667 | ||
668 | if (fatal_signal_pending(current)) | |
2fe86e00 | 669 | return 0; |
748446bb MG |
670 | } |
671 | ||
be976572 VB |
672 | if (compact_should_abort(cc)) |
673 | return 0; | |
aeef4b83 | 674 | |
748446bb | 675 | /* Time to isolate some pages for migration */ |
748446bb | 676 | for (; low_pfn < end_pfn; low_pfn++) { |
8b44d279 VB |
677 | /* |
678 | * Periodically drop the lock (if held) regardless of its | |
679 | * contention, to give chance to IRQs. Abort async compaction | |
680 | * if contended. | |
681 | */ | |
682 | if (!(low_pfn % SWAP_CLUSTER_MAX) | |
683 | && compact_unlock_should_abort(&zone->lru_lock, flags, | |
684 | &locked, cc)) | |
685 | break; | |
c67fe375 | 686 | |
748446bb MG |
687 | if (!pfn_valid_within(low_pfn)) |
688 | continue; | |
b7aba698 | 689 | nr_scanned++; |
748446bb | 690 | |
748446bb | 691 | page = pfn_to_page(low_pfn); |
dc908600 | 692 | |
bb13ffeb MG |
693 | if (!valid_page) |
694 | valid_page = page; | |
695 | ||
6c14466c | 696 | /* |
99c0fd5e VB |
697 | * Skip if free. We read page order here without zone lock |
698 | * which is generally unsafe, but the race window is small and | |
699 | * the worst thing that can happen is that we skip some | |
700 | * potential isolation targets. | |
6c14466c | 701 | */ |
99c0fd5e VB |
702 | if (PageBuddy(page)) { |
703 | unsigned long freepage_order = page_order_unsafe(page); | |
704 | ||
705 | /* | |
706 | * Without lock, we cannot be sure that what we got is | |
707 | * a valid page order. Consider only values in the | |
708 | * valid order range to prevent low_pfn overflow. | |
709 | */ | |
710 | if (freepage_order > 0 && freepage_order < MAX_ORDER) | |
711 | low_pfn += (1UL << freepage_order) - 1; | |
748446bb | 712 | continue; |
99c0fd5e | 713 | } |
748446bb | 714 | |
bf6bddf1 RA |
715 | /* |
716 | * Check may be lockless but that's ok as we recheck later. | |
717 | * It's possible to migrate LRU pages and balloon pages | |
718 | * Skip any other type of page | |
719 | */ | |
720 | if (!PageLRU(page)) { | |
721 | if (unlikely(balloon_page_movable(page))) { | |
d6d86c0a | 722 | if (balloon_page_isolate(page)) { |
bf6bddf1 | 723 | /* Successfully isolated */ |
b6c75016 | 724 | goto isolate_success; |
bf6bddf1 RA |
725 | } |
726 | } | |
bc835011 | 727 | continue; |
bf6bddf1 | 728 | } |
bc835011 AA |
729 | |
730 | /* | |
2a1402aa MG |
731 | * PageLRU is set. lru_lock normally excludes isolation |
732 | * splitting and collapsing (collapsing has already happened | |
733 | * if PageLRU is set) but the lock is not necessarily taken | |
734 | * here and it is wasteful to take it just to check transhuge. | |
735 | * Check TransHuge without lock and skip the whole pageblock if | |
736 | * it's either a transhuge or hugetlbfs page, as calling | |
737 | * compound_order() without preventing THP from splitting the | |
738 | * page underneath us may return surprising results. | |
bc835011 | 739 | */ |
2a1402aa MG |
740 | if (PageTransHuge(page)) { |
741 | if (!locked) | |
edc2ca61 VB |
742 | low_pfn = ALIGN(low_pfn + 1, |
743 | pageblock_nr_pages) - 1; | |
744 | else | |
745 | low_pfn += (1 << compound_order(page)) - 1; | |
746 | ||
2a1402aa MG |
747 | continue; |
748 | } | |
749 | ||
119d6d59 DR |
750 | /* |
751 | * Migration will fail if an anonymous page is pinned in memory, | |
752 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
753 | * admittedly racy check. | |
754 | */ | |
755 | if (!page_mapping(page) && | |
756 | page_count(page) > page_mapcount(page)) | |
757 | continue; | |
758 | ||
69b7189f VB |
759 | /* If we already hold the lock, we can skip some rechecking */ |
760 | if (!locked) { | |
8b44d279 VB |
761 | locked = compact_trylock_irqsave(&zone->lru_lock, |
762 | &flags, cc); | |
69b7189f VB |
763 | if (!locked) |
764 | break; | |
2a1402aa | 765 | |
69b7189f VB |
766 | /* Recheck PageLRU and PageTransHuge under lock */ |
767 | if (!PageLRU(page)) | |
768 | continue; | |
769 | if (PageTransHuge(page)) { | |
770 | low_pfn += (1 << compound_order(page)) - 1; | |
771 | continue; | |
772 | } | |
bc835011 AA |
773 | } |
774 | ||
fa9add64 HD |
775 | lruvec = mem_cgroup_page_lruvec(page, zone); |
776 | ||
748446bb | 777 | /* Try isolate the page */ |
edc2ca61 | 778 | if (__isolate_lru_page(page, isolate_mode) != 0) |
748446bb MG |
779 | continue; |
780 | ||
309381fe | 781 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 782 | |
748446bb | 783 | /* Successfully isolated */ |
fa9add64 | 784 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
785 | |
786 | isolate_success: | |
748446bb | 787 | list_add(&page->lru, migratelist); |
748446bb | 788 | cc->nr_migratepages++; |
b7aba698 | 789 | nr_isolated++; |
748446bb MG |
790 | |
791 | /* Avoid isolating too much */ | |
31b8384a HD |
792 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
793 | ++low_pfn; | |
748446bb | 794 | break; |
31b8384a | 795 | } |
748446bb MG |
796 | } |
797 | ||
99c0fd5e VB |
798 | /* |
799 | * The PageBuddy() check could have potentially brought us outside | |
800 | * the range to be scanned. | |
801 | */ | |
802 | if (unlikely(low_pfn > end_pfn)) | |
803 | low_pfn = end_pfn; | |
804 | ||
c67fe375 MG |
805 | if (locked) |
806 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 807 | |
50b5b094 VB |
808 | /* |
809 | * Update the pageblock-skip information and cached scanner pfn, | |
810 | * if the whole pageblock was scanned without isolating any page. | |
50b5b094 | 811 | */ |
35979ef3 | 812 | if (low_pfn == end_pfn) |
edc2ca61 | 813 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 814 | |
e34d85f0 JK |
815 | trace_mm_compaction_isolate_migratepages(start_pfn, low_pfn, |
816 | nr_scanned, nr_isolated); | |
b7aba698 | 817 | |
010fc29a | 818 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 819 | if (nr_isolated) |
010fc29a | 820 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 821 | |
2fe86e00 MN |
822 | return low_pfn; |
823 | } | |
824 | ||
edc2ca61 VB |
825 | /** |
826 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range | |
827 | * @cc: Compaction control structure. | |
828 | * @start_pfn: The first PFN to start isolating. | |
829 | * @end_pfn: The one-past-last PFN. | |
830 | * | |
831 | * Returns zero if isolation fails fatally due to e.g. pending signal. | |
832 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
833 | * (which may be greater than end_pfn if end fell in a middle of a THP page). | |
834 | */ | |
835 | unsigned long | |
836 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, | |
837 | unsigned long end_pfn) | |
838 | { | |
839 | unsigned long pfn, block_end_pfn; | |
840 | ||
841 | /* Scan block by block. First and last block may be incomplete */ | |
842 | pfn = start_pfn; | |
843 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
844 | ||
845 | for (; pfn < end_pfn; pfn = block_end_pfn, | |
846 | block_end_pfn += pageblock_nr_pages) { | |
847 | ||
848 | block_end_pfn = min(block_end_pfn, end_pfn); | |
849 | ||
7d49d886 | 850 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
edc2ca61 VB |
851 | continue; |
852 | ||
853 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, | |
854 | ISOLATE_UNEVICTABLE); | |
855 | ||
856 | /* | |
857 | * In case of fatal failure, release everything that might | |
858 | * have been isolated in the previous iteration, and signal | |
859 | * the failure back to caller. | |
860 | */ | |
861 | if (!pfn) { | |
862 | putback_movable_pages(&cc->migratepages); | |
863 | cc->nr_migratepages = 0; | |
864 | break; | |
865 | } | |
6ea41c0c JK |
866 | |
867 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) | |
868 | break; | |
edc2ca61 VB |
869 | } |
870 | acct_isolated(cc->zone, cc); | |
871 | ||
872 | return pfn; | |
873 | } | |
874 | ||
ff9543fd MN |
875 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
876 | #ifdef CONFIG_COMPACTION | |
018e9a49 AM |
877 | |
878 | /* Returns true if the page is within a block suitable for migration to */ | |
879 | static bool suitable_migration_target(struct page *page) | |
880 | { | |
881 | /* If the page is a large free page, then disallow migration */ | |
882 | if (PageBuddy(page)) { | |
883 | /* | |
884 | * We are checking page_order without zone->lock taken. But | |
885 | * the only small danger is that we skip a potentially suitable | |
886 | * pageblock, so it's not worth to check order for valid range. | |
887 | */ | |
888 | if (page_order_unsafe(page) >= pageblock_order) | |
889 | return false; | |
890 | } | |
891 | ||
892 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
893 | if (migrate_async_suitable(get_pageblock_migratetype(page))) | |
894 | return true; | |
895 | ||
896 | /* Otherwise skip the block */ | |
897 | return false; | |
898 | } | |
899 | ||
f2849aa0 VB |
900 | /* |
901 | * Test whether the free scanner has reached the same or lower pageblock than | |
902 | * the migration scanner, and compaction should thus terminate. | |
903 | */ | |
904 | static inline bool compact_scanners_met(struct compact_control *cc) | |
905 | { | |
906 | return (cc->free_pfn >> pageblock_order) | |
907 | <= (cc->migrate_pfn >> pageblock_order); | |
908 | } | |
909 | ||
2fe86e00 | 910 | /* |
ff9543fd MN |
911 | * Based on information in the current compact_control, find blocks |
912 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 913 | */ |
edc2ca61 | 914 | static void isolate_freepages(struct compact_control *cc) |
2fe86e00 | 915 | { |
edc2ca61 | 916 | struct zone *zone = cc->zone; |
ff9543fd | 917 | struct page *page; |
c96b9e50 | 918 | unsigned long block_start_pfn; /* start of current pageblock */ |
e14c720e | 919 | unsigned long isolate_start_pfn; /* exact pfn we start at */ |
c96b9e50 VB |
920 | unsigned long block_end_pfn; /* end of current pageblock */ |
921 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd | 922 | struct list_head *freelist = &cc->freepages; |
2fe86e00 | 923 | |
ff9543fd MN |
924 | /* |
925 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 | 926 | * successfully isolated from, zone-cached value, or the end of the |
e14c720e VB |
927 | * zone when isolating for the first time. For looping we also need |
928 | * this pfn aligned down to the pageblock boundary, because we do | |
c96b9e50 VB |
929 | * block_start_pfn -= pageblock_nr_pages in the for loop. |
930 | * For ending point, take care when isolating in last pageblock of a | |
931 | * a zone which ends in the middle of a pageblock. | |
49e068f0 VB |
932 | * The low boundary is the end of the pageblock the migration scanner |
933 | * is using. | |
ff9543fd | 934 | */ |
e14c720e | 935 | isolate_start_pfn = cc->free_pfn; |
c96b9e50 VB |
936 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
937 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, | |
938 | zone_end_pfn(zone)); | |
7ed695e0 | 939 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 940 | |
ff9543fd MN |
941 | /* |
942 | * Isolate free pages until enough are available to migrate the | |
943 | * pages on cc->migratepages. We stop searching if the migrate | |
944 | * and free page scanners meet or enough free pages are isolated. | |
945 | */ | |
932ff6bb JK |
946 | for (; block_start_pfn >= low_pfn && |
947 | cc->nr_migratepages > cc->nr_freepages; | |
c96b9e50 | 948 | block_end_pfn = block_start_pfn, |
e14c720e VB |
949 | block_start_pfn -= pageblock_nr_pages, |
950 | isolate_start_pfn = block_start_pfn) { | |
2fe86e00 | 951 | |
f6ea3adb DR |
952 | /* |
953 | * This can iterate a massively long zone without finding any | |
954 | * suitable migration targets, so periodically check if we need | |
be976572 | 955 | * to schedule, or even abort async compaction. |
f6ea3adb | 956 | */ |
be976572 VB |
957 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
958 | && compact_should_abort(cc)) | |
959 | break; | |
f6ea3adb | 960 | |
7d49d886 VB |
961 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
962 | zone); | |
963 | if (!page) | |
ff9543fd MN |
964 | continue; |
965 | ||
966 | /* Check the block is suitable for migration */ | |
68e3e926 | 967 | if (!suitable_migration_target(page)) |
ff9543fd | 968 | continue; |
68e3e926 | 969 | |
bb13ffeb MG |
970 | /* If isolation recently failed, do not retry */ |
971 | if (!isolation_suitable(cc, page)) | |
972 | continue; | |
973 | ||
e14c720e | 974 | /* Found a block suitable for isolating free pages from. */ |
932ff6bb | 975 | isolate_freepages_block(cc, &isolate_start_pfn, |
c96b9e50 | 976 | block_end_pfn, freelist, false); |
ff9543fd | 977 | |
e14c720e VB |
978 | /* |
979 | * Remember where the free scanner should restart next time, | |
980 | * which is where isolate_freepages_block() left off. | |
981 | * But if it scanned the whole pageblock, isolate_start_pfn | |
982 | * now points at block_end_pfn, which is the start of the next | |
983 | * pageblock. | |
984 | * In that case we will however want to restart at the start | |
985 | * of the previous pageblock. | |
986 | */ | |
987 | cc->free_pfn = (isolate_start_pfn < block_end_pfn) ? | |
988 | isolate_start_pfn : | |
989 | block_start_pfn - pageblock_nr_pages; | |
990 | ||
be976572 VB |
991 | /* |
992 | * isolate_freepages_block() might have aborted due to async | |
993 | * compaction being contended | |
994 | */ | |
995 | if (cc->contended) | |
996 | break; | |
ff9543fd MN |
997 | } |
998 | ||
999 | /* split_free_page does not map the pages */ | |
1000 | map_pages(freelist); | |
1001 | ||
7ed695e0 VB |
1002 | /* |
1003 | * If we crossed the migrate scanner, we want to keep it that way | |
1004 | * so that compact_finished() may detect this | |
1005 | */ | |
c96b9e50 | 1006 | if (block_start_pfn < low_pfn) |
e9ade569 | 1007 | cc->free_pfn = cc->migrate_pfn; |
748446bb MG |
1008 | } |
1009 | ||
1010 | /* | |
1011 | * This is a migrate-callback that "allocates" freepages by taking pages | |
1012 | * from the isolated freelists in the block we are migrating to. | |
1013 | */ | |
1014 | static struct page *compaction_alloc(struct page *migratepage, | |
1015 | unsigned long data, | |
1016 | int **result) | |
1017 | { | |
1018 | struct compact_control *cc = (struct compact_control *)data; | |
1019 | struct page *freepage; | |
1020 | ||
be976572 VB |
1021 | /* |
1022 | * Isolate free pages if necessary, and if we are not aborting due to | |
1023 | * contention. | |
1024 | */ | |
748446bb | 1025 | if (list_empty(&cc->freepages)) { |
be976572 | 1026 | if (!cc->contended) |
edc2ca61 | 1027 | isolate_freepages(cc); |
748446bb MG |
1028 | |
1029 | if (list_empty(&cc->freepages)) | |
1030 | return NULL; | |
1031 | } | |
1032 | ||
1033 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
1034 | list_del(&freepage->lru); | |
1035 | cc->nr_freepages--; | |
1036 | ||
1037 | return freepage; | |
1038 | } | |
1039 | ||
1040 | /* | |
d53aea3d DR |
1041 | * This is a migrate-callback that "frees" freepages back to the isolated |
1042 | * freelist. All pages on the freelist are from the same zone, so there is no | |
1043 | * special handling needed for NUMA. | |
1044 | */ | |
1045 | static void compaction_free(struct page *page, unsigned long data) | |
1046 | { | |
1047 | struct compact_control *cc = (struct compact_control *)data; | |
1048 | ||
1049 | list_add(&page->lru, &cc->freepages); | |
1050 | cc->nr_freepages++; | |
1051 | } | |
1052 | ||
ff9543fd MN |
1053 | /* possible outcome of isolate_migratepages */ |
1054 | typedef enum { | |
1055 | ISOLATE_ABORT, /* Abort compaction now */ | |
1056 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
1057 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
1058 | } isolate_migrate_t; | |
1059 | ||
5bbe3547 EM |
1060 | /* |
1061 | * Allow userspace to control policy on scanning the unevictable LRU for | |
1062 | * compactable pages. | |
1063 | */ | |
1064 | int sysctl_compact_unevictable_allowed __read_mostly = 1; | |
1065 | ||
ff9543fd | 1066 | /* |
edc2ca61 VB |
1067 | * Isolate all pages that can be migrated from the first suitable block, |
1068 | * starting at the block pointed to by the migrate scanner pfn within | |
1069 | * compact_control. | |
ff9543fd MN |
1070 | */ |
1071 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
1072 | struct compact_control *cc) | |
1073 | { | |
1074 | unsigned long low_pfn, end_pfn; | |
edc2ca61 VB |
1075 | struct page *page; |
1076 | const isolate_mode_t isolate_mode = | |
5bbe3547 | 1077 | (sysctl_compact_unevictable_allowed ? ISOLATE_UNEVICTABLE : 0) | |
edc2ca61 | 1078 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); |
ff9543fd | 1079 | |
edc2ca61 VB |
1080 | /* |
1081 | * Start at where we last stopped, or beginning of the zone as | |
1082 | * initialized by compact_zone() | |
1083 | */ | |
1084 | low_pfn = cc->migrate_pfn; | |
ff9543fd MN |
1085 | |
1086 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 1087 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd | 1088 | |
edc2ca61 VB |
1089 | /* |
1090 | * Iterate over whole pageblocks until we find the first suitable. | |
1091 | * Do not cross the free scanner. | |
1092 | */ | |
1093 | for (; end_pfn <= cc->free_pfn; | |
1094 | low_pfn = end_pfn, end_pfn += pageblock_nr_pages) { | |
ff9543fd | 1095 | |
edc2ca61 VB |
1096 | /* |
1097 | * This can potentially iterate a massively long zone with | |
1098 | * many pageblocks unsuitable, so periodically check if we | |
1099 | * need to schedule, or even abort async compaction. | |
1100 | */ | |
1101 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) | |
1102 | && compact_should_abort(cc)) | |
1103 | break; | |
ff9543fd | 1104 | |
7d49d886 VB |
1105 | page = pageblock_pfn_to_page(low_pfn, end_pfn, zone); |
1106 | if (!page) | |
edc2ca61 VB |
1107 | continue; |
1108 | ||
edc2ca61 VB |
1109 | /* If isolation recently failed, do not retry */ |
1110 | if (!isolation_suitable(cc, page)) | |
1111 | continue; | |
1112 | ||
1113 | /* | |
1114 | * For async compaction, also only scan in MOVABLE blocks. | |
1115 | * Async compaction is optimistic to see if the minimum amount | |
1116 | * of work satisfies the allocation. | |
1117 | */ | |
1118 | if (cc->mode == MIGRATE_ASYNC && | |
1119 | !migrate_async_suitable(get_pageblock_migratetype(page))) | |
1120 | continue; | |
1121 | ||
1122 | /* Perform the isolation */ | |
1123 | low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn, | |
1124 | isolate_mode); | |
1125 | ||
ff59909a HD |
1126 | if (!low_pfn || cc->contended) { |
1127 | acct_isolated(zone, cc); | |
edc2ca61 | 1128 | return ISOLATE_ABORT; |
ff59909a | 1129 | } |
edc2ca61 VB |
1130 | |
1131 | /* | |
1132 | * Either we isolated something and proceed with migration. Or | |
1133 | * we failed and compact_zone should decide if we should | |
1134 | * continue or not. | |
1135 | */ | |
1136 | break; | |
1137 | } | |
1138 | ||
1139 | acct_isolated(zone, cc); | |
f2849aa0 VB |
1140 | /* Record where migration scanner will be restarted. */ |
1141 | cc->migrate_pfn = low_pfn; | |
ff9543fd | 1142 | |
edc2ca61 | 1143 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
ff9543fd MN |
1144 | } |
1145 | ||
837d026d | 1146 | static int __compact_finished(struct zone *zone, struct compact_control *cc, |
6d7ce559 | 1147 | const int migratetype) |
748446bb | 1148 | { |
8fb74b9f | 1149 | unsigned int order; |
5a03b051 | 1150 | unsigned long watermark; |
56de7263 | 1151 | |
be976572 | 1152 | if (cc->contended || fatal_signal_pending(current)) |
748446bb MG |
1153 | return COMPACT_PARTIAL; |
1154 | ||
753341a4 | 1155 | /* Compaction run completes if the migrate and free scanner meet */ |
f2849aa0 | 1156 | if (compact_scanners_met(cc)) { |
55b7c4c9 | 1157 | /* Let the next compaction start anew. */ |
35979ef3 DR |
1158 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
1159 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
55b7c4c9 VB |
1160 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
1161 | ||
62997027 MG |
1162 | /* |
1163 | * Mark that the PG_migrate_skip information should be cleared | |
1164 | * by kswapd when it goes to sleep. kswapd does not set the | |
1165 | * flag itself as the decision to be clear should be directly | |
1166 | * based on an allocation request. | |
1167 | */ | |
1168 | if (!current_is_kswapd()) | |
1169 | zone->compact_blockskip_flush = true; | |
1170 | ||
748446bb | 1171 | return COMPACT_COMPLETE; |
bb13ffeb | 1172 | } |
748446bb | 1173 | |
82478fb7 JW |
1174 | /* |
1175 | * order == -1 is expected when compacting via | |
1176 | * /proc/sys/vm/compact_memory | |
1177 | */ | |
56de7263 MG |
1178 | if (cc->order == -1) |
1179 | return COMPACT_CONTINUE; | |
1180 | ||
3957c776 MH |
1181 | /* Compaction run is not finished if the watermark is not met */ |
1182 | watermark = low_wmark_pages(zone); | |
3957c776 | 1183 | |
ebff3980 VB |
1184 | if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx, |
1185 | cc->alloc_flags)) | |
3957c776 MH |
1186 | return COMPACT_CONTINUE; |
1187 | ||
56de7263 | 1188 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
1189 | for (order = cc->order; order < MAX_ORDER; order++) { |
1190 | struct free_area *area = &zone->free_area[order]; | |
2149cdae | 1191 | bool can_steal; |
8fb74b9f MG |
1192 | |
1193 | /* Job done if page is free of the right migratetype */ | |
6d7ce559 | 1194 | if (!list_empty(&area->free_list[migratetype])) |
8fb74b9f MG |
1195 | return COMPACT_PARTIAL; |
1196 | ||
2149cdae JK |
1197 | #ifdef CONFIG_CMA |
1198 | /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */ | |
1199 | if (migratetype == MIGRATE_MOVABLE && | |
1200 | !list_empty(&area->free_list[MIGRATE_CMA])) | |
1201 | return COMPACT_PARTIAL; | |
1202 | #endif | |
1203 | /* | |
1204 | * Job done if allocation would steal freepages from | |
1205 | * other migratetype buddy lists. | |
1206 | */ | |
1207 | if (find_suitable_fallback(area, order, migratetype, | |
1208 | true, &can_steal) != -1) | |
56de7263 MG |
1209 | return COMPACT_PARTIAL; |
1210 | } | |
1211 | ||
837d026d JK |
1212 | return COMPACT_NO_SUITABLE_PAGE; |
1213 | } | |
1214 | ||
1215 | static int compact_finished(struct zone *zone, struct compact_control *cc, | |
1216 | const int migratetype) | |
1217 | { | |
1218 | int ret; | |
1219 | ||
1220 | ret = __compact_finished(zone, cc, migratetype); | |
1221 | trace_mm_compaction_finished(zone, cc->order, ret); | |
1222 | if (ret == COMPACT_NO_SUITABLE_PAGE) | |
1223 | ret = COMPACT_CONTINUE; | |
1224 | ||
1225 | return ret; | |
748446bb MG |
1226 | } |
1227 | ||
3e7d3449 MG |
1228 | /* |
1229 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
1230 | * Returns | |
1231 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
1232 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
1233 | * COMPACT_CONTINUE - If compaction should run now | |
1234 | */ | |
837d026d | 1235 | static unsigned long __compaction_suitable(struct zone *zone, int order, |
ebff3980 | 1236 | int alloc_flags, int classzone_idx) |
3e7d3449 MG |
1237 | { |
1238 | int fragindex; | |
1239 | unsigned long watermark; | |
1240 | ||
3957c776 MH |
1241 | /* |
1242 | * order == -1 is expected when compacting via | |
1243 | * /proc/sys/vm/compact_memory | |
1244 | */ | |
1245 | if (order == -1) | |
1246 | return COMPACT_CONTINUE; | |
1247 | ||
ebff3980 VB |
1248 | watermark = low_wmark_pages(zone); |
1249 | /* | |
1250 | * If watermarks for high-order allocation are already met, there | |
1251 | * should be no need for compaction at all. | |
1252 | */ | |
1253 | if (zone_watermark_ok(zone, order, watermark, classzone_idx, | |
1254 | alloc_flags)) | |
1255 | return COMPACT_PARTIAL; | |
1256 | ||
3e7d3449 MG |
1257 | /* |
1258 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
1259 | * This is because during migration, copies of pages need to be | |
1260 | * allocated and for a short time, the footprint is higher | |
1261 | */ | |
ebff3980 VB |
1262 | watermark += (2UL << order); |
1263 | if (!zone_watermark_ok(zone, 0, watermark, classzone_idx, alloc_flags)) | |
3e7d3449 MG |
1264 | return COMPACT_SKIPPED; |
1265 | ||
1266 | /* | |
1267 | * fragmentation index determines if allocation failures are due to | |
1268 | * low memory or external fragmentation | |
1269 | * | |
ebff3980 VB |
1270 | * index of -1000 would imply allocations might succeed depending on |
1271 | * watermarks, but we already failed the high-order watermark check | |
3e7d3449 MG |
1272 | * index towards 0 implies failure is due to lack of memory |
1273 | * index towards 1000 implies failure is due to fragmentation | |
1274 | * | |
1275 | * Only compact if a failure would be due to fragmentation. | |
1276 | */ | |
1277 | fragindex = fragmentation_index(zone, order); | |
1278 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
837d026d | 1279 | return COMPACT_NOT_SUITABLE_ZONE; |
3e7d3449 | 1280 | |
3e7d3449 MG |
1281 | return COMPACT_CONTINUE; |
1282 | } | |
1283 | ||
837d026d JK |
1284 | unsigned long compaction_suitable(struct zone *zone, int order, |
1285 | int alloc_flags, int classzone_idx) | |
1286 | { | |
1287 | unsigned long ret; | |
1288 | ||
1289 | ret = __compaction_suitable(zone, order, alloc_flags, classzone_idx); | |
1290 | trace_mm_compaction_suitable(zone, order, ret); | |
1291 | if (ret == COMPACT_NOT_SUITABLE_ZONE) | |
1292 | ret = COMPACT_SKIPPED; | |
1293 | ||
1294 | return ret; | |
1295 | } | |
1296 | ||
748446bb MG |
1297 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
1298 | { | |
1299 | int ret; | |
c89511ab | 1300 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 1301 | unsigned long end_pfn = zone_end_pfn(zone); |
6d7ce559 | 1302 | const int migratetype = gfpflags_to_migratetype(cc->gfp_mask); |
e0b9daeb | 1303 | const bool sync = cc->mode != MIGRATE_ASYNC; |
fdaf7f5c | 1304 | unsigned long last_migrated_pfn = 0; |
748446bb | 1305 | |
ebff3980 VB |
1306 | ret = compaction_suitable(zone, cc->order, cc->alloc_flags, |
1307 | cc->classzone_idx); | |
3e7d3449 MG |
1308 | switch (ret) { |
1309 | case COMPACT_PARTIAL: | |
1310 | case COMPACT_SKIPPED: | |
1311 | /* Compaction is likely to fail */ | |
1312 | return ret; | |
1313 | case COMPACT_CONTINUE: | |
1314 | /* Fall through to compaction */ | |
1315 | ; | |
1316 | } | |
1317 | ||
d3132e4b VB |
1318 | /* |
1319 | * Clear pageblock skip if there were failures recently and compaction | |
1320 | * is about to be retried after being deferred. kswapd does not do | |
1321 | * this reset as it'll reset the cached information when going to sleep. | |
1322 | */ | |
1323 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1324 | __reset_isolation_suitable(zone); | |
1325 | ||
c89511ab MG |
1326 | /* |
1327 | * Setup to move all movable pages to the end of the zone. Used cached | |
1328 | * information on where the scanners should start but check that it | |
1329 | * is initialised by ensuring the values are within zone boundaries. | |
1330 | */ | |
e0b9daeb | 1331 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
c89511ab MG |
1332 | cc->free_pfn = zone->compact_cached_free_pfn; |
1333 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
1334 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
1335 | zone->compact_cached_free_pfn = cc->free_pfn; | |
1336 | } | |
1337 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1338 | cc->migrate_pfn = start_pfn; | |
35979ef3 DR |
1339 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
1340 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
c89511ab | 1341 | } |
748446bb | 1342 | |
16c4a097 JK |
1343 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, |
1344 | cc->free_pfn, end_pfn, sync); | |
0eb927c0 | 1345 | |
748446bb MG |
1346 | migrate_prep_local(); |
1347 | ||
6d7ce559 DR |
1348 | while ((ret = compact_finished(zone, cc, migratetype)) == |
1349 | COMPACT_CONTINUE) { | |
9d502c1c | 1350 | int err; |
fdaf7f5c | 1351 | unsigned long isolate_start_pfn = cc->migrate_pfn; |
748446bb | 1352 | |
f9e35b3b MG |
1353 | switch (isolate_migratepages(zone, cc)) { |
1354 | case ISOLATE_ABORT: | |
1355 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1356 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1357 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1358 | goto out; |
1359 | case ISOLATE_NONE: | |
fdaf7f5c VB |
1360 | /* |
1361 | * We haven't isolated and migrated anything, but | |
1362 | * there might still be unflushed migrations from | |
1363 | * previous cc->order aligned block. | |
1364 | */ | |
1365 | goto check_drain; | |
f9e35b3b MG |
1366 | case ISOLATE_SUCCESS: |
1367 | ; | |
1368 | } | |
748446bb | 1369 | |
d53aea3d | 1370 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 1371 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 1372 | MR_COMPACTION); |
748446bb | 1373 | |
f8c9301f VB |
1374 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
1375 | &cc->migratepages); | |
748446bb | 1376 | |
f8c9301f VB |
1377 | /* All pages were either migrated or will be released */ |
1378 | cc->nr_migratepages = 0; | |
9d502c1c | 1379 | if (err) { |
5733c7d1 | 1380 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
1381 | /* |
1382 | * migrate_pages() may return -ENOMEM when scanners meet | |
1383 | * and we want compact_finished() to detect it | |
1384 | */ | |
f2849aa0 | 1385 | if (err == -ENOMEM && !compact_scanners_met(cc)) { |
4bf2bba3 DR |
1386 | ret = COMPACT_PARTIAL; |
1387 | goto out; | |
1388 | } | |
748446bb | 1389 | } |
fdaf7f5c VB |
1390 | |
1391 | /* | |
1392 | * Record where we could have freed pages by migration and not | |
1393 | * yet flushed them to buddy allocator. We use the pfn that | |
1394 | * isolate_migratepages() started from in this loop iteration | |
1395 | * - this is the lowest page that could have been isolated and | |
1396 | * then freed by migration. | |
1397 | */ | |
1398 | if (!last_migrated_pfn) | |
1399 | last_migrated_pfn = isolate_start_pfn; | |
1400 | ||
1401 | check_drain: | |
1402 | /* | |
1403 | * Has the migration scanner moved away from the previous | |
1404 | * cc->order aligned block where we migrated from? If yes, | |
1405 | * flush the pages that were freed, so that they can merge and | |
1406 | * compact_finished() can detect immediately if allocation | |
1407 | * would succeed. | |
1408 | */ | |
1409 | if (cc->order > 0 && last_migrated_pfn) { | |
1410 | int cpu; | |
1411 | unsigned long current_block_start = | |
1412 | cc->migrate_pfn & ~((1UL << cc->order) - 1); | |
1413 | ||
1414 | if (last_migrated_pfn < current_block_start) { | |
1415 | cpu = get_cpu(); | |
1416 | lru_add_drain_cpu(cpu); | |
1417 | drain_local_pages(zone); | |
1418 | put_cpu(); | |
1419 | /* No more flushing until we migrate again */ | |
1420 | last_migrated_pfn = 0; | |
1421 | } | |
1422 | } | |
1423 | ||
748446bb MG |
1424 | } |
1425 | ||
f9e35b3b | 1426 | out: |
6bace090 VB |
1427 | /* |
1428 | * Release free pages and update where the free scanner should restart, | |
1429 | * so we don't leave any returned pages behind in the next attempt. | |
1430 | */ | |
1431 | if (cc->nr_freepages > 0) { | |
1432 | unsigned long free_pfn = release_freepages(&cc->freepages); | |
1433 | ||
1434 | cc->nr_freepages = 0; | |
1435 | VM_BUG_ON(free_pfn == 0); | |
1436 | /* The cached pfn is always the first in a pageblock */ | |
1437 | free_pfn &= ~(pageblock_nr_pages-1); | |
1438 | /* | |
1439 | * Only go back, not forward. The cached pfn might have been | |
1440 | * already reset to zone end in compact_finished() | |
1441 | */ | |
1442 | if (free_pfn > zone->compact_cached_free_pfn) | |
1443 | zone->compact_cached_free_pfn = free_pfn; | |
1444 | } | |
748446bb | 1445 | |
16c4a097 JK |
1446 | trace_mm_compaction_end(start_pfn, cc->migrate_pfn, |
1447 | cc->free_pfn, end_pfn, sync, ret); | |
0eb927c0 | 1448 | |
748446bb MG |
1449 | return ret; |
1450 | } | |
76ab0f53 | 1451 | |
e0b9daeb | 1452 | static unsigned long compact_zone_order(struct zone *zone, int order, |
ebff3980 VB |
1453 | gfp_t gfp_mask, enum migrate_mode mode, int *contended, |
1454 | int alloc_flags, int classzone_idx) | |
56de7263 | 1455 | { |
e64c5237 | 1456 | unsigned long ret; |
56de7263 MG |
1457 | struct compact_control cc = { |
1458 | .nr_freepages = 0, | |
1459 | .nr_migratepages = 0, | |
1460 | .order = order, | |
6d7ce559 | 1461 | .gfp_mask = gfp_mask, |
56de7263 | 1462 | .zone = zone, |
e0b9daeb | 1463 | .mode = mode, |
ebff3980 VB |
1464 | .alloc_flags = alloc_flags, |
1465 | .classzone_idx = classzone_idx, | |
56de7263 MG |
1466 | }; |
1467 | INIT_LIST_HEAD(&cc.freepages); | |
1468 | INIT_LIST_HEAD(&cc.migratepages); | |
1469 | ||
e64c5237 SL |
1470 | ret = compact_zone(zone, &cc); |
1471 | ||
1472 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1473 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1474 | ||
1475 | *contended = cc.contended; | |
1476 | return ret; | |
56de7263 MG |
1477 | } |
1478 | ||
5e771905 MG |
1479 | int sysctl_extfrag_threshold = 500; |
1480 | ||
56de7263 MG |
1481 | /** |
1482 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
56de7263 | 1483 | * @gfp_mask: The GFP mask of the current allocation |
1a6d53a1 VB |
1484 | * @order: The order of the current allocation |
1485 | * @alloc_flags: The allocation flags of the current allocation | |
1486 | * @ac: The context of current allocation | |
e0b9daeb | 1487 | * @mode: The migration mode for async, sync light, or sync migration |
1f9efdef VB |
1488 | * @contended: Return value that determines if compaction was aborted due to |
1489 | * need_resched() or lock contention | |
56de7263 MG |
1490 | * |
1491 | * This is the main entry point for direct page compaction. | |
1492 | */ | |
1a6d53a1 VB |
1493 | unsigned long try_to_compact_pages(gfp_t gfp_mask, unsigned int order, |
1494 | int alloc_flags, const struct alloc_context *ac, | |
1495 | enum migrate_mode mode, int *contended) | |
56de7263 | 1496 | { |
56de7263 MG |
1497 | int may_enter_fs = gfp_mask & __GFP_FS; |
1498 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1499 | struct zoneref *z; |
1500 | struct zone *zone; | |
53853e2d | 1501 | int rc = COMPACT_DEFERRED; |
1f9efdef VB |
1502 | int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ |
1503 | ||
1504 | *contended = COMPACT_CONTENDED_NONE; | |
56de7263 | 1505 | |
4ffb6335 | 1506 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1507 | if (!order || !may_enter_fs || !may_perform_io) |
53853e2d | 1508 | return COMPACT_SKIPPED; |
56de7263 | 1509 | |
837d026d JK |
1510 | trace_mm_compaction_try_to_compact_pages(order, gfp_mask, mode); |
1511 | ||
56de7263 | 1512 | /* Compact each zone in the list */ |
1a6d53a1 VB |
1513 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, |
1514 | ac->nodemask) { | |
56de7263 | 1515 | int status; |
1f9efdef | 1516 | int zone_contended; |
56de7263 | 1517 | |
53853e2d VB |
1518 | if (compaction_deferred(zone, order)) |
1519 | continue; | |
1520 | ||
e0b9daeb | 1521 | status = compact_zone_order(zone, order, gfp_mask, mode, |
1a6d53a1 VB |
1522 | &zone_contended, alloc_flags, |
1523 | ac->classzone_idx); | |
56de7263 | 1524 | rc = max(status, rc); |
1f9efdef VB |
1525 | /* |
1526 | * It takes at least one zone that wasn't lock contended | |
1527 | * to clear all_zones_contended. | |
1528 | */ | |
1529 | all_zones_contended &= zone_contended; | |
56de7263 | 1530 | |
3e7d3449 | 1531 | /* If a normal allocation would succeed, stop compacting */ |
ebff3980 | 1532 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), |
1a6d53a1 | 1533 | ac->classzone_idx, alloc_flags)) { |
53853e2d VB |
1534 | /* |
1535 | * We think the allocation will succeed in this zone, | |
1536 | * but it is not certain, hence the false. The caller | |
1537 | * will repeat this with true if allocation indeed | |
1538 | * succeeds in this zone. | |
1539 | */ | |
1540 | compaction_defer_reset(zone, order, false); | |
1f9efdef VB |
1541 | /* |
1542 | * It is possible that async compaction aborted due to | |
1543 | * need_resched() and the watermarks were ok thanks to | |
1544 | * somebody else freeing memory. The allocation can | |
1545 | * however still fail so we better signal the | |
1546 | * need_resched() contention anyway (this will not | |
1547 | * prevent the allocation attempt). | |
1548 | */ | |
1549 | if (zone_contended == COMPACT_CONTENDED_SCHED) | |
1550 | *contended = COMPACT_CONTENDED_SCHED; | |
1551 | ||
1552 | goto break_loop; | |
1553 | } | |
1554 | ||
f8669795 | 1555 | if (mode != MIGRATE_ASYNC && status == COMPACT_COMPLETE) { |
53853e2d VB |
1556 | /* |
1557 | * We think that allocation won't succeed in this zone | |
1558 | * so we defer compaction there. If it ends up | |
1559 | * succeeding after all, it will be reset. | |
1560 | */ | |
1561 | defer_compaction(zone, order); | |
1562 | } | |
1f9efdef VB |
1563 | |
1564 | /* | |
1565 | * We might have stopped compacting due to need_resched() in | |
1566 | * async compaction, or due to a fatal signal detected. In that | |
1567 | * case do not try further zones and signal need_resched() | |
1568 | * contention. | |
1569 | */ | |
1570 | if ((zone_contended == COMPACT_CONTENDED_SCHED) | |
1571 | || fatal_signal_pending(current)) { | |
1572 | *contended = COMPACT_CONTENDED_SCHED; | |
1573 | goto break_loop; | |
1574 | } | |
1575 | ||
1576 | continue; | |
1577 | break_loop: | |
1578 | /* | |
1579 | * We might not have tried all the zones, so be conservative | |
1580 | * and assume they are not all lock contended. | |
1581 | */ | |
1582 | all_zones_contended = 0; | |
1583 | break; | |
56de7263 MG |
1584 | } |
1585 | ||
1f9efdef VB |
1586 | /* |
1587 | * If at least one zone wasn't deferred or skipped, we report if all | |
1588 | * zones that were tried were lock contended. | |
1589 | */ | |
1590 | if (rc > COMPACT_SKIPPED && all_zones_contended) | |
1591 | *contended = COMPACT_CONTENDED_LOCK; | |
1592 | ||
56de7263 MG |
1593 | return rc; |
1594 | } | |
1595 | ||
1596 | ||
76ab0f53 | 1597 | /* Compact all zones within a node */ |
7103f16d | 1598 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1599 | { |
1600 | int zoneid; | |
76ab0f53 MG |
1601 | struct zone *zone; |
1602 | ||
76ab0f53 | 1603 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1604 | |
1605 | zone = &pgdat->node_zones[zoneid]; | |
1606 | if (!populated_zone(zone)) | |
1607 | continue; | |
1608 | ||
7be62de9 RR |
1609 | cc->nr_freepages = 0; |
1610 | cc->nr_migratepages = 0; | |
1611 | cc->zone = zone; | |
1612 | INIT_LIST_HEAD(&cc->freepages); | |
1613 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1614 | |
195b0c60 GK |
1615 | /* |
1616 | * When called via /proc/sys/vm/compact_memory | |
1617 | * this makes sure we compact the whole zone regardless of | |
1618 | * cached scanner positions. | |
1619 | */ | |
1620 | if (cc->order == -1) | |
1621 | __reset_isolation_suitable(zone); | |
1622 | ||
aad6ec37 | 1623 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1624 | compact_zone(zone, cc); |
76ab0f53 | 1625 | |
aff62249 | 1626 | if (cc->order > 0) { |
de6c60a6 VB |
1627 | if (zone_watermark_ok(zone, cc->order, |
1628 | low_wmark_pages(zone), 0, 0)) | |
1629 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 RR |
1630 | } |
1631 | ||
7be62de9 RR |
1632 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1633 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1634 | } |
76ab0f53 MG |
1635 | } |
1636 | ||
7103f16d | 1637 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1638 | { |
1639 | struct compact_control cc = { | |
1640 | .order = order, | |
e0b9daeb | 1641 | .mode = MIGRATE_ASYNC, |
7be62de9 RR |
1642 | }; |
1643 | ||
3a7200af MG |
1644 | if (!order) |
1645 | return; | |
1646 | ||
7103f16d | 1647 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1648 | } |
1649 | ||
7103f16d | 1650 | static void compact_node(int nid) |
7be62de9 | 1651 | { |
7be62de9 RR |
1652 | struct compact_control cc = { |
1653 | .order = -1, | |
e0b9daeb | 1654 | .mode = MIGRATE_SYNC, |
91ca9186 | 1655 | .ignore_skip_hint = true, |
7be62de9 RR |
1656 | }; |
1657 | ||
7103f16d | 1658 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1659 | } |
1660 | ||
76ab0f53 | 1661 | /* Compact all nodes in the system */ |
7964c06d | 1662 | static void compact_nodes(void) |
76ab0f53 MG |
1663 | { |
1664 | int nid; | |
1665 | ||
8575ec29 HD |
1666 | /* Flush pending updates to the LRU lists */ |
1667 | lru_add_drain_all(); | |
1668 | ||
76ab0f53 MG |
1669 | for_each_online_node(nid) |
1670 | compact_node(nid); | |
76ab0f53 MG |
1671 | } |
1672 | ||
1673 | /* The written value is actually unused, all memory is compacted */ | |
1674 | int sysctl_compact_memory; | |
1675 | ||
1676 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1677 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1678 | void __user *buffer, size_t *length, loff_t *ppos) | |
1679 | { | |
1680 | if (write) | |
7964c06d | 1681 | compact_nodes(); |
76ab0f53 MG |
1682 | |
1683 | return 0; | |
1684 | } | |
ed4a6d7f | 1685 | |
5e771905 MG |
1686 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1687 | void __user *buffer, size_t *length, loff_t *ppos) | |
1688 | { | |
1689 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1690 | ||
1691 | return 0; | |
1692 | } | |
1693 | ||
ed4a6d7f | 1694 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1695 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1696 | struct device_attribute *attr, |
ed4a6d7f MG |
1697 | const char *buf, size_t count) |
1698 | { | |
8575ec29 HD |
1699 | int nid = dev->id; |
1700 | ||
1701 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1702 | /* Flush pending updates to the LRU lists */ | |
1703 | lru_add_drain_all(); | |
1704 | ||
1705 | compact_node(nid); | |
1706 | } | |
ed4a6d7f MG |
1707 | |
1708 | return count; | |
1709 | } | |
10fbcf4c | 1710 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1711 | |
1712 | int compaction_register_node(struct node *node) | |
1713 | { | |
10fbcf4c | 1714 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1715 | } |
1716 | ||
1717 | void compaction_unregister_node(struct node *node) | |
1718 | { | |
10fbcf4c | 1719 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1720 | } |
1721 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1722 | |
1723 | #endif /* CONFIG_COMPACTION */ |