]>
Commit | Line | Data |
---|---|---|
748446bb MG |
1 | /* |
2 | * linux/mm/compaction.c | |
3 | * | |
4 | * Memory compaction for the reduction of external fragmentation. Note that | |
5 | * this heavily depends upon page migration to do all the real heavy | |
6 | * lifting | |
7 | * | |
8 | * Copyright IBM Corp. 2007-2010 Mel Gorman <mel@csn.ul.ie> | |
9 | */ | |
10 | #include <linux/swap.h> | |
11 | #include <linux/migrate.h> | |
12 | #include <linux/compaction.h> | |
13 | #include <linux/mm_inline.h> | |
14 | #include <linux/backing-dev.h> | |
76ab0f53 | 15 | #include <linux/sysctl.h> |
ed4a6d7f | 16 | #include <linux/sysfs.h> |
bf6bddf1 | 17 | #include <linux/balloon_compaction.h> |
194159fb | 18 | #include <linux/page-isolation.h> |
748446bb MG |
19 | #include "internal.h" |
20 | ||
010fc29a MK |
21 | #ifdef CONFIG_COMPACTION |
22 | static inline void count_compact_event(enum vm_event_item item) | |
23 | { | |
24 | count_vm_event(item); | |
25 | } | |
26 | ||
27 | static inline void count_compact_events(enum vm_event_item item, long delta) | |
28 | { | |
29 | count_vm_events(item, delta); | |
30 | } | |
31 | #else | |
32 | #define count_compact_event(item) do { } while (0) | |
33 | #define count_compact_events(item, delta) do { } while (0) | |
34 | #endif | |
35 | ||
ff9543fd MN |
36 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
37 | ||
b7aba698 MG |
38 | #define CREATE_TRACE_POINTS |
39 | #include <trace/events/compaction.h> | |
40 | ||
748446bb MG |
41 | static unsigned long release_freepages(struct list_head *freelist) |
42 | { | |
43 | struct page *page, *next; | |
44 | unsigned long count = 0; | |
45 | ||
46 | list_for_each_entry_safe(page, next, freelist, lru) { | |
47 | list_del(&page->lru); | |
48 | __free_page(page); | |
49 | count++; | |
50 | } | |
51 | ||
52 | return count; | |
53 | } | |
54 | ||
ff9543fd MN |
55 | static void map_pages(struct list_head *list) |
56 | { | |
57 | struct page *page; | |
58 | ||
59 | list_for_each_entry(page, list, lru) { | |
60 | arch_alloc_page(page, 0); | |
61 | kernel_map_pages(page, 1, 1); | |
62 | } | |
63 | } | |
64 | ||
47118af0 MN |
65 | static inline bool migrate_async_suitable(int migratetype) |
66 | { | |
67 | return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE; | |
68 | } | |
69 | ||
7d49d886 VB |
70 | /* |
71 | * Check that the whole (or subset of) a pageblock given by the interval of | |
72 | * [start_pfn, end_pfn) is valid and within the same zone, before scanning it | |
73 | * with the migration of free compaction scanner. The scanners then need to | |
74 | * use only pfn_valid_within() check for arches that allow holes within | |
75 | * pageblocks. | |
76 | * | |
77 | * Return struct page pointer of start_pfn, or NULL if checks were not passed. | |
78 | * | |
79 | * It's possible on some configurations to have a setup like node0 node1 node0 | |
80 | * i.e. it's possible that all pages within a zones range of pages do not | |
81 | * belong to a single zone. We assume that a border between node0 and node1 | |
82 | * can occur within a single pageblock, but not a node0 node1 node0 | |
83 | * interleaving within a single pageblock. It is therefore sufficient to check | |
84 | * the first and last page of a pageblock and avoid checking each individual | |
85 | * page in a pageblock. | |
86 | */ | |
87 | static struct page *pageblock_pfn_to_page(unsigned long start_pfn, | |
88 | unsigned long end_pfn, struct zone *zone) | |
89 | { | |
90 | struct page *start_page; | |
91 | struct page *end_page; | |
92 | ||
93 | /* end_pfn is one past the range we are checking */ | |
94 | end_pfn--; | |
95 | ||
96 | if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn)) | |
97 | return NULL; | |
98 | ||
99 | start_page = pfn_to_page(start_pfn); | |
100 | ||
101 | if (page_zone(start_page) != zone) | |
102 | return NULL; | |
103 | ||
104 | end_page = pfn_to_page(end_pfn); | |
105 | ||
106 | /* This gives a shorter code than deriving page_zone(end_page) */ | |
107 | if (page_zone_id(start_page) != page_zone_id(end_page)) | |
108 | return NULL; | |
109 | ||
110 | return start_page; | |
111 | } | |
112 | ||
bb13ffeb MG |
113 | #ifdef CONFIG_COMPACTION |
114 | /* Returns true if the pageblock should be scanned for pages to isolate. */ | |
115 | static inline bool isolation_suitable(struct compact_control *cc, | |
116 | struct page *page) | |
117 | { | |
118 | if (cc->ignore_skip_hint) | |
119 | return true; | |
120 | ||
121 | return !get_pageblock_skip(page); | |
122 | } | |
123 | ||
124 | /* | |
125 | * This function is called to clear all cached information on pageblocks that | |
126 | * should be skipped for page isolation when the migrate and free page scanner | |
127 | * meet. | |
128 | */ | |
62997027 | 129 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb MG |
130 | { |
131 | unsigned long start_pfn = zone->zone_start_pfn; | |
108bcc96 | 132 | unsigned long end_pfn = zone_end_pfn(zone); |
bb13ffeb MG |
133 | unsigned long pfn; |
134 | ||
35979ef3 DR |
135 | zone->compact_cached_migrate_pfn[0] = start_pfn; |
136 | zone->compact_cached_migrate_pfn[1] = start_pfn; | |
c89511ab | 137 | zone->compact_cached_free_pfn = end_pfn; |
62997027 | 138 | zone->compact_blockskip_flush = false; |
bb13ffeb MG |
139 | |
140 | /* Walk the zone and mark every pageblock as suitable for isolation */ | |
141 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
142 | struct page *page; | |
143 | ||
144 | cond_resched(); | |
145 | ||
146 | if (!pfn_valid(pfn)) | |
147 | continue; | |
148 | ||
149 | page = pfn_to_page(pfn); | |
150 | if (zone != page_zone(page)) | |
151 | continue; | |
152 | ||
153 | clear_pageblock_skip(page); | |
154 | } | |
155 | } | |
156 | ||
62997027 MG |
157 | void reset_isolation_suitable(pg_data_t *pgdat) |
158 | { | |
159 | int zoneid; | |
160 | ||
161 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
162 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
163 | if (!populated_zone(zone)) | |
164 | continue; | |
165 | ||
166 | /* Only flush if a full compaction finished recently */ | |
167 | if (zone->compact_blockskip_flush) | |
168 | __reset_isolation_suitable(zone); | |
169 | } | |
170 | } | |
171 | ||
bb13ffeb MG |
172 | /* |
173 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 174 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 175 | */ |
c89511ab MG |
176 | static void update_pageblock_skip(struct compact_control *cc, |
177 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 178 | bool migrate_scanner) |
bb13ffeb | 179 | { |
c89511ab | 180 | struct zone *zone = cc->zone; |
35979ef3 | 181 | unsigned long pfn; |
6815bf3f JK |
182 | |
183 | if (cc->ignore_skip_hint) | |
184 | return; | |
185 | ||
bb13ffeb MG |
186 | if (!page) |
187 | return; | |
188 | ||
35979ef3 DR |
189 | if (nr_isolated) |
190 | return; | |
191 | ||
edc2ca61 | 192 | set_pageblock_skip(page); |
c89511ab | 193 | |
35979ef3 DR |
194 | pfn = page_to_pfn(page); |
195 | ||
196 | /* Update where async and sync compaction should restart */ | |
197 | if (migrate_scanner) { | |
198 | if (cc->finished_update_migrate) | |
199 | return; | |
200 | if (pfn > zone->compact_cached_migrate_pfn[0]) | |
201 | zone->compact_cached_migrate_pfn[0] = pfn; | |
e0b9daeb DR |
202 | if (cc->mode != MIGRATE_ASYNC && |
203 | pfn > zone->compact_cached_migrate_pfn[1]) | |
35979ef3 DR |
204 | zone->compact_cached_migrate_pfn[1] = pfn; |
205 | } else { | |
206 | if (cc->finished_update_free) | |
207 | return; | |
208 | if (pfn < zone->compact_cached_free_pfn) | |
209 | zone->compact_cached_free_pfn = pfn; | |
c89511ab | 210 | } |
bb13ffeb MG |
211 | } |
212 | #else | |
213 | static inline bool isolation_suitable(struct compact_control *cc, | |
214 | struct page *page) | |
215 | { | |
216 | return true; | |
217 | } | |
218 | ||
c89511ab MG |
219 | static void update_pageblock_skip(struct compact_control *cc, |
220 | struct page *page, unsigned long nr_isolated, | |
edc2ca61 | 221 | bool migrate_scanner) |
bb13ffeb MG |
222 | { |
223 | } | |
224 | #endif /* CONFIG_COMPACTION */ | |
225 | ||
8b44d279 VB |
226 | /* |
227 | * Compaction requires the taking of some coarse locks that are potentially | |
228 | * very heavily contended. For async compaction, back out if the lock cannot | |
229 | * be taken immediately. For sync compaction, spin on the lock if needed. | |
230 | * | |
231 | * Returns true if the lock is held | |
232 | * Returns false if the lock is not held and compaction should abort | |
233 | */ | |
234 | static bool compact_trylock_irqsave(spinlock_t *lock, unsigned long *flags, | |
235 | struct compact_control *cc) | |
2a1402aa | 236 | { |
8b44d279 VB |
237 | if (cc->mode == MIGRATE_ASYNC) { |
238 | if (!spin_trylock_irqsave(lock, *flags)) { | |
239 | cc->contended = COMPACT_CONTENDED_LOCK; | |
240 | return false; | |
241 | } | |
242 | } else { | |
243 | spin_lock_irqsave(lock, *flags); | |
244 | } | |
1f9efdef | 245 | |
8b44d279 | 246 | return true; |
2a1402aa MG |
247 | } |
248 | ||
c67fe375 MG |
249 | /* |
250 | * Compaction requires the taking of some coarse locks that are potentially | |
8b44d279 VB |
251 | * very heavily contended. The lock should be periodically unlocked to avoid |
252 | * having disabled IRQs for a long time, even when there is nobody waiting on | |
253 | * the lock. It might also be that allowing the IRQs will result in | |
254 | * need_resched() becoming true. If scheduling is needed, async compaction | |
255 | * aborts. Sync compaction schedules. | |
256 | * Either compaction type will also abort if a fatal signal is pending. | |
257 | * In either case if the lock was locked, it is dropped and not regained. | |
c67fe375 | 258 | * |
8b44d279 VB |
259 | * Returns true if compaction should abort due to fatal signal pending, or |
260 | * async compaction due to need_resched() | |
261 | * Returns false when compaction can continue (sync compaction might have | |
262 | * scheduled) | |
c67fe375 | 263 | */ |
8b44d279 VB |
264 | static bool compact_unlock_should_abort(spinlock_t *lock, |
265 | unsigned long flags, bool *locked, struct compact_control *cc) | |
c67fe375 | 266 | { |
8b44d279 VB |
267 | if (*locked) { |
268 | spin_unlock_irqrestore(lock, flags); | |
269 | *locked = false; | |
270 | } | |
1f9efdef | 271 | |
8b44d279 VB |
272 | if (fatal_signal_pending(current)) { |
273 | cc->contended = COMPACT_CONTENDED_SCHED; | |
274 | return true; | |
275 | } | |
c67fe375 | 276 | |
8b44d279 | 277 | if (need_resched()) { |
e0b9daeb | 278 | if (cc->mode == MIGRATE_ASYNC) { |
8b44d279 VB |
279 | cc->contended = COMPACT_CONTENDED_SCHED; |
280 | return true; | |
c67fe375 | 281 | } |
c67fe375 | 282 | cond_resched(); |
c67fe375 MG |
283 | } |
284 | ||
8b44d279 | 285 | return false; |
c67fe375 MG |
286 | } |
287 | ||
be976572 VB |
288 | /* |
289 | * Aside from avoiding lock contention, compaction also periodically checks | |
290 | * need_resched() and either schedules in sync compaction or aborts async | |
8b44d279 | 291 | * compaction. This is similar to what compact_unlock_should_abort() does, but |
be976572 VB |
292 | * is used where no lock is concerned. |
293 | * | |
294 | * Returns false when no scheduling was needed, or sync compaction scheduled. | |
295 | * Returns true when async compaction should abort. | |
296 | */ | |
297 | static inline bool compact_should_abort(struct compact_control *cc) | |
298 | { | |
299 | /* async compaction aborts if contended */ | |
300 | if (need_resched()) { | |
301 | if (cc->mode == MIGRATE_ASYNC) { | |
1f9efdef | 302 | cc->contended = COMPACT_CONTENDED_SCHED; |
be976572 VB |
303 | return true; |
304 | } | |
305 | ||
306 | cond_resched(); | |
307 | } | |
308 | ||
309 | return false; | |
310 | } | |
311 | ||
f40d1e42 MG |
312 | /* Returns true if the page is within a block suitable for migration to */ |
313 | static bool suitable_migration_target(struct page *page) | |
314 | { | |
7d348b9e | 315 | /* If the page is a large free page, then disallow migration */ |
f40d1e42 | 316 | if (PageBuddy(page) && page_order(page) >= pageblock_order) |
7d348b9e | 317 | return false; |
f40d1e42 MG |
318 | |
319 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ | |
7d348b9e | 320 | if (migrate_async_suitable(get_pageblock_migratetype(page))) |
f40d1e42 MG |
321 | return true; |
322 | ||
323 | /* Otherwise skip the block */ | |
324 | return false; | |
325 | } | |
326 | ||
85aa125f | 327 | /* |
9e4be470 JM |
328 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
329 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
330 | * (even though it may still end up isolating some pages). | |
85aa125f | 331 | */ |
f40d1e42 MG |
332 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
333 | unsigned long blockpfn, | |
85aa125f MN |
334 | unsigned long end_pfn, |
335 | struct list_head *freelist, | |
336 | bool strict) | |
748446bb | 337 | { |
b7aba698 | 338 | int nr_scanned = 0, total_isolated = 0; |
bb13ffeb | 339 | struct page *cursor, *valid_page = NULL; |
f40d1e42 MG |
340 | unsigned long flags; |
341 | bool locked = false; | |
748446bb | 342 | |
748446bb MG |
343 | cursor = pfn_to_page(blockpfn); |
344 | ||
f40d1e42 | 345 | /* Isolate free pages. */ |
748446bb MG |
346 | for (; blockpfn < end_pfn; blockpfn++, cursor++) { |
347 | int isolated, i; | |
348 | struct page *page = cursor; | |
349 | ||
8b44d279 VB |
350 | /* |
351 | * Periodically drop the lock (if held) regardless of its | |
352 | * contention, to give chance to IRQs. Abort if fatal signal | |
353 | * pending or async compaction detects need_resched() | |
354 | */ | |
355 | if (!(blockpfn % SWAP_CLUSTER_MAX) | |
356 | && compact_unlock_should_abort(&cc->zone->lock, flags, | |
357 | &locked, cc)) | |
358 | break; | |
359 | ||
b7aba698 | 360 | nr_scanned++; |
f40d1e42 | 361 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
362 | goto isolate_fail; |
363 | ||
bb13ffeb MG |
364 | if (!valid_page) |
365 | valid_page = page; | |
f40d1e42 | 366 | if (!PageBuddy(page)) |
2af120bc | 367 | goto isolate_fail; |
f40d1e42 MG |
368 | |
369 | /* | |
69b7189f VB |
370 | * If we already hold the lock, we can skip some rechecking. |
371 | * Note that if we hold the lock now, checked_pageblock was | |
372 | * already set in some previous iteration (or strict is true), | |
373 | * so it is correct to skip the suitable migration target | |
374 | * recheck as well. | |
f40d1e42 | 375 | */ |
69b7189f VB |
376 | if (!locked) { |
377 | /* | |
378 | * The zone lock must be held to isolate freepages. | |
379 | * Unfortunately this is a very coarse lock and can be | |
380 | * heavily contended if there are parallel allocations | |
381 | * or parallel compactions. For async compaction do not | |
382 | * spin on the lock and we acquire the lock as late as | |
383 | * possible. | |
384 | */ | |
8b44d279 VB |
385 | locked = compact_trylock_irqsave(&cc->zone->lock, |
386 | &flags, cc); | |
69b7189f VB |
387 | if (!locked) |
388 | break; | |
f40d1e42 | 389 | |
69b7189f VB |
390 | /* Recheck this is a buddy page under lock */ |
391 | if (!PageBuddy(page)) | |
392 | goto isolate_fail; | |
393 | } | |
748446bb MG |
394 | |
395 | /* Found a free page, break it into order-0 pages */ | |
396 | isolated = split_free_page(page); | |
397 | total_isolated += isolated; | |
398 | for (i = 0; i < isolated; i++) { | |
399 | list_add(&page->lru, freelist); | |
400 | page++; | |
401 | } | |
402 | ||
403 | /* If a page was split, advance to the end of it */ | |
404 | if (isolated) { | |
405 | blockpfn += isolated - 1; | |
406 | cursor += isolated - 1; | |
2af120bc | 407 | continue; |
748446bb | 408 | } |
2af120bc LA |
409 | |
410 | isolate_fail: | |
411 | if (strict) | |
412 | break; | |
413 | else | |
414 | continue; | |
415 | ||
748446bb MG |
416 | } |
417 | ||
b7aba698 | 418 | trace_mm_compaction_isolate_freepages(nr_scanned, total_isolated); |
f40d1e42 MG |
419 | |
420 | /* | |
421 | * If strict isolation is requested by CMA then check that all the | |
422 | * pages requested were isolated. If there were any failures, 0 is | |
423 | * returned and CMA will fail. | |
424 | */ | |
2af120bc | 425 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
426 | total_isolated = 0; |
427 | ||
428 | if (locked) | |
429 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
430 | ||
bb13ffeb MG |
431 | /* Update the pageblock-skip if the whole pageblock was scanned */ |
432 | if (blockpfn == end_pfn) | |
edc2ca61 | 433 | update_pageblock_skip(cc, valid_page, total_isolated, false); |
bb13ffeb | 434 | |
010fc29a | 435 | count_compact_events(COMPACTFREE_SCANNED, nr_scanned); |
397487db | 436 | if (total_isolated) |
010fc29a | 437 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
438 | return total_isolated; |
439 | } | |
440 | ||
85aa125f MN |
441 | /** |
442 | * isolate_freepages_range() - isolate free pages. | |
443 | * @start_pfn: The first PFN to start isolating. | |
444 | * @end_pfn: The one-past-last PFN. | |
445 | * | |
446 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
447 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
448 | * undo its actions and return zero. | |
449 | * | |
450 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
451 | * (which may be greater then end_pfn if end fell in a middle of | |
452 | * a free page). | |
453 | */ | |
ff9543fd | 454 | unsigned long |
bb13ffeb MG |
455 | isolate_freepages_range(struct compact_control *cc, |
456 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 457 | { |
f40d1e42 | 458 | unsigned long isolated, pfn, block_end_pfn; |
85aa125f MN |
459 | LIST_HEAD(freelist); |
460 | ||
7d49d886 VB |
461 | pfn = start_pfn; |
462 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
463 | ||
464 | for (; pfn < end_pfn; pfn += isolated, | |
465 | block_end_pfn += pageblock_nr_pages) { | |
85aa125f | 466 | |
85aa125f MN |
467 | block_end_pfn = min(block_end_pfn, end_pfn); |
468 | ||
7d49d886 VB |
469 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
470 | break; | |
471 | ||
bb13ffeb | 472 | isolated = isolate_freepages_block(cc, pfn, block_end_pfn, |
85aa125f | 473 | &freelist, true); |
85aa125f MN |
474 | |
475 | /* | |
476 | * In strict mode, isolate_freepages_block() returns 0 if | |
477 | * there are any holes in the block (ie. invalid PFNs or | |
478 | * non-free pages). | |
479 | */ | |
480 | if (!isolated) | |
481 | break; | |
482 | ||
483 | /* | |
484 | * If we managed to isolate pages, it is always (1 << n) * | |
485 | * pageblock_nr_pages for some non-negative n. (Max order | |
486 | * page may span two pageblocks). | |
487 | */ | |
488 | } | |
489 | ||
490 | /* split_free_page does not map the pages */ | |
491 | map_pages(&freelist); | |
492 | ||
493 | if (pfn < end_pfn) { | |
494 | /* Loop terminated early, cleanup. */ | |
495 | release_freepages(&freelist); | |
496 | return 0; | |
497 | } | |
498 | ||
499 | /* We don't use freelists for anything. */ | |
500 | return pfn; | |
501 | } | |
502 | ||
748446bb | 503 | /* Update the number of anon and file isolated pages in the zone */ |
edc2ca61 | 504 | static void acct_isolated(struct zone *zone, struct compact_control *cc) |
748446bb MG |
505 | { |
506 | struct page *page; | |
b9e84ac1 | 507 | unsigned int count[2] = { 0, }; |
748446bb | 508 | |
edc2ca61 VB |
509 | if (list_empty(&cc->migratepages)) |
510 | return; | |
511 | ||
b9e84ac1 MK |
512 | list_for_each_entry(page, &cc->migratepages, lru) |
513 | count[!!page_is_file_cache(page)]++; | |
748446bb | 514 | |
edc2ca61 VB |
515 | mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]); |
516 | mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]); | |
748446bb MG |
517 | } |
518 | ||
519 | /* Similar to reclaim, but different enough that they don't share logic */ | |
520 | static bool too_many_isolated(struct zone *zone) | |
521 | { | |
bc693045 | 522 | unsigned long active, inactive, isolated; |
748446bb MG |
523 | |
524 | inactive = zone_page_state(zone, NR_INACTIVE_FILE) + | |
525 | zone_page_state(zone, NR_INACTIVE_ANON); | |
bc693045 MK |
526 | active = zone_page_state(zone, NR_ACTIVE_FILE) + |
527 | zone_page_state(zone, NR_ACTIVE_ANON); | |
748446bb MG |
528 | isolated = zone_page_state(zone, NR_ISOLATED_FILE) + |
529 | zone_page_state(zone, NR_ISOLATED_ANON); | |
530 | ||
bc693045 | 531 | return isolated > (inactive + active) / 2; |
748446bb MG |
532 | } |
533 | ||
2fe86e00 | 534 | /** |
edc2ca61 VB |
535 | * isolate_migratepages_block() - isolate all migrate-able pages within |
536 | * a single pageblock | |
2fe86e00 | 537 | * @cc: Compaction control structure. |
edc2ca61 VB |
538 | * @low_pfn: The first PFN to isolate |
539 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock | |
540 | * @isolate_mode: Isolation mode to be used. | |
2fe86e00 MN |
541 | * |
542 | * Isolate all pages that can be migrated from the range specified by | |
edc2ca61 VB |
543 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
544 | * Returns zero if there is a fatal signal pending, otherwise PFN of the | |
545 | * first page that was not scanned (which may be both less, equal to or more | |
546 | * than end_pfn). | |
2fe86e00 | 547 | * |
edc2ca61 VB |
548 | * The pages are isolated on cc->migratepages list (not required to be empty), |
549 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field | |
550 | * is neither read nor updated. | |
748446bb | 551 | */ |
edc2ca61 VB |
552 | static unsigned long |
553 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, | |
554 | unsigned long end_pfn, isolate_mode_t isolate_mode) | |
748446bb | 555 | { |
edc2ca61 | 556 | struct zone *zone = cc->zone; |
b7aba698 | 557 | unsigned long nr_scanned = 0, nr_isolated = 0; |
748446bb | 558 | struct list_head *migratelist = &cc->migratepages; |
fa9add64 | 559 | struct lruvec *lruvec; |
c67fe375 | 560 | unsigned long flags; |
2a1402aa | 561 | bool locked = false; |
bb13ffeb | 562 | struct page *page = NULL, *valid_page = NULL; |
748446bb | 563 | |
748446bb MG |
564 | /* |
565 | * Ensure that there are not too many pages isolated from the LRU | |
566 | * list by either parallel reclaimers or compaction. If there are, | |
567 | * delay for some time until fewer pages are isolated | |
568 | */ | |
569 | while (unlikely(too_many_isolated(zone))) { | |
f9e35b3b | 570 | /* async migration should just abort */ |
e0b9daeb | 571 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 572 | return 0; |
f9e35b3b | 573 | |
748446bb MG |
574 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
575 | ||
576 | if (fatal_signal_pending(current)) | |
2fe86e00 | 577 | return 0; |
748446bb MG |
578 | } |
579 | ||
be976572 VB |
580 | if (compact_should_abort(cc)) |
581 | return 0; | |
aeef4b83 | 582 | |
748446bb | 583 | /* Time to isolate some pages for migration */ |
748446bb | 584 | for (; low_pfn < end_pfn; low_pfn++) { |
8b44d279 VB |
585 | /* |
586 | * Periodically drop the lock (if held) regardless of its | |
587 | * contention, to give chance to IRQs. Abort async compaction | |
588 | * if contended. | |
589 | */ | |
590 | if (!(low_pfn % SWAP_CLUSTER_MAX) | |
591 | && compact_unlock_should_abort(&zone->lru_lock, flags, | |
592 | &locked, cc)) | |
593 | break; | |
c67fe375 | 594 | |
748446bb MG |
595 | if (!pfn_valid_within(low_pfn)) |
596 | continue; | |
b7aba698 | 597 | nr_scanned++; |
748446bb | 598 | |
748446bb | 599 | page = pfn_to_page(low_pfn); |
dc908600 | 600 | |
bb13ffeb MG |
601 | if (!valid_page) |
602 | valid_page = page; | |
603 | ||
6c14466c MG |
604 | /* |
605 | * Skip if free. page_order cannot be used without zone->lock | |
606 | * as nothing prevents parallel allocations or buddy merging. | |
607 | */ | |
748446bb MG |
608 | if (PageBuddy(page)) |
609 | continue; | |
610 | ||
bf6bddf1 RA |
611 | /* |
612 | * Check may be lockless but that's ok as we recheck later. | |
613 | * It's possible to migrate LRU pages and balloon pages | |
614 | * Skip any other type of page | |
615 | */ | |
616 | if (!PageLRU(page)) { | |
617 | if (unlikely(balloon_page_movable(page))) { | |
618 | if (locked && balloon_page_isolate(page)) { | |
619 | /* Successfully isolated */ | |
b6c75016 | 620 | goto isolate_success; |
bf6bddf1 RA |
621 | } |
622 | } | |
bc835011 | 623 | continue; |
bf6bddf1 | 624 | } |
bc835011 AA |
625 | |
626 | /* | |
2a1402aa MG |
627 | * PageLRU is set. lru_lock normally excludes isolation |
628 | * splitting and collapsing (collapsing has already happened | |
629 | * if PageLRU is set) but the lock is not necessarily taken | |
630 | * here and it is wasteful to take it just to check transhuge. | |
631 | * Check TransHuge without lock and skip the whole pageblock if | |
632 | * it's either a transhuge or hugetlbfs page, as calling | |
633 | * compound_order() without preventing THP from splitting the | |
634 | * page underneath us may return surprising results. | |
bc835011 | 635 | */ |
2a1402aa MG |
636 | if (PageTransHuge(page)) { |
637 | if (!locked) | |
edc2ca61 VB |
638 | low_pfn = ALIGN(low_pfn + 1, |
639 | pageblock_nr_pages) - 1; | |
640 | else | |
641 | low_pfn += (1 << compound_order(page)) - 1; | |
642 | ||
2a1402aa MG |
643 | continue; |
644 | } | |
645 | ||
119d6d59 DR |
646 | /* |
647 | * Migration will fail if an anonymous page is pinned in memory, | |
648 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
649 | * admittedly racy check. | |
650 | */ | |
651 | if (!page_mapping(page) && | |
652 | page_count(page) > page_mapcount(page)) | |
653 | continue; | |
654 | ||
69b7189f VB |
655 | /* If we already hold the lock, we can skip some rechecking */ |
656 | if (!locked) { | |
8b44d279 VB |
657 | locked = compact_trylock_irqsave(&zone->lru_lock, |
658 | &flags, cc); | |
69b7189f VB |
659 | if (!locked) |
660 | break; | |
2a1402aa | 661 | |
69b7189f VB |
662 | /* Recheck PageLRU and PageTransHuge under lock */ |
663 | if (!PageLRU(page)) | |
664 | continue; | |
665 | if (PageTransHuge(page)) { | |
666 | low_pfn += (1 << compound_order(page)) - 1; | |
667 | continue; | |
668 | } | |
bc835011 AA |
669 | } |
670 | ||
fa9add64 HD |
671 | lruvec = mem_cgroup_page_lruvec(page, zone); |
672 | ||
748446bb | 673 | /* Try isolate the page */ |
edc2ca61 | 674 | if (__isolate_lru_page(page, isolate_mode) != 0) |
748446bb MG |
675 | continue; |
676 | ||
309381fe | 677 | VM_BUG_ON_PAGE(PageTransCompound(page), page); |
bc835011 | 678 | |
748446bb | 679 | /* Successfully isolated */ |
fa9add64 | 680 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
b6c75016 JK |
681 | |
682 | isolate_success: | |
683 | cc->finished_update_migrate = true; | |
748446bb | 684 | list_add(&page->lru, migratelist); |
748446bb | 685 | cc->nr_migratepages++; |
b7aba698 | 686 | nr_isolated++; |
748446bb MG |
687 | |
688 | /* Avoid isolating too much */ | |
31b8384a HD |
689 | if (cc->nr_migratepages == COMPACT_CLUSTER_MAX) { |
690 | ++low_pfn; | |
748446bb | 691 | break; |
31b8384a | 692 | } |
748446bb MG |
693 | } |
694 | ||
c67fe375 MG |
695 | if (locked) |
696 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
748446bb | 697 | |
50b5b094 VB |
698 | /* |
699 | * Update the pageblock-skip information and cached scanner pfn, | |
700 | * if the whole pageblock was scanned without isolating any page. | |
50b5b094 | 701 | */ |
35979ef3 | 702 | if (low_pfn == end_pfn) |
edc2ca61 | 703 | update_pageblock_skip(cc, valid_page, nr_isolated, true); |
bb13ffeb | 704 | |
b7aba698 MG |
705 | trace_mm_compaction_isolate_migratepages(nr_scanned, nr_isolated); |
706 | ||
010fc29a | 707 | count_compact_events(COMPACTMIGRATE_SCANNED, nr_scanned); |
397487db | 708 | if (nr_isolated) |
010fc29a | 709 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 710 | |
2fe86e00 MN |
711 | return low_pfn; |
712 | } | |
713 | ||
edc2ca61 VB |
714 | /** |
715 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range | |
716 | * @cc: Compaction control structure. | |
717 | * @start_pfn: The first PFN to start isolating. | |
718 | * @end_pfn: The one-past-last PFN. | |
719 | * | |
720 | * Returns zero if isolation fails fatally due to e.g. pending signal. | |
721 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
722 | * (which may be greater than end_pfn if end fell in a middle of a THP page). | |
723 | */ | |
724 | unsigned long | |
725 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, | |
726 | unsigned long end_pfn) | |
727 | { | |
728 | unsigned long pfn, block_end_pfn; | |
729 | ||
730 | /* Scan block by block. First and last block may be incomplete */ | |
731 | pfn = start_pfn; | |
732 | block_end_pfn = ALIGN(pfn + 1, pageblock_nr_pages); | |
733 | ||
734 | for (; pfn < end_pfn; pfn = block_end_pfn, | |
735 | block_end_pfn += pageblock_nr_pages) { | |
736 | ||
737 | block_end_pfn = min(block_end_pfn, end_pfn); | |
738 | ||
7d49d886 | 739 | if (!pageblock_pfn_to_page(pfn, block_end_pfn, cc->zone)) |
edc2ca61 VB |
740 | continue; |
741 | ||
742 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, | |
743 | ISOLATE_UNEVICTABLE); | |
744 | ||
745 | /* | |
746 | * In case of fatal failure, release everything that might | |
747 | * have been isolated in the previous iteration, and signal | |
748 | * the failure back to caller. | |
749 | */ | |
750 | if (!pfn) { | |
751 | putback_movable_pages(&cc->migratepages); | |
752 | cc->nr_migratepages = 0; | |
753 | break; | |
754 | } | |
755 | } | |
756 | acct_isolated(cc->zone, cc); | |
757 | ||
758 | return pfn; | |
759 | } | |
760 | ||
ff9543fd MN |
761 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
762 | #ifdef CONFIG_COMPACTION | |
2fe86e00 | 763 | /* |
ff9543fd MN |
764 | * Based on information in the current compact_control, find blocks |
765 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 766 | */ |
edc2ca61 | 767 | static void isolate_freepages(struct compact_control *cc) |
2fe86e00 | 768 | { |
edc2ca61 | 769 | struct zone *zone = cc->zone; |
ff9543fd | 770 | struct page *page; |
c96b9e50 VB |
771 | unsigned long block_start_pfn; /* start of current pageblock */ |
772 | unsigned long block_end_pfn; /* end of current pageblock */ | |
773 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd MN |
774 | int nr_freepages = cc->nr_freepages; |
775 | struct list_head *freelist = &cc->freepages; | |
2fe86e00 | 776 | |
ff9543fd MN |
777 | /* |
778 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 VB |
779 | * successfully isolated from, zone-cached value, or the end of the |
780 | * zone when isolating for the first time. We need this aligned to | |
c96b9e50 VB |
781 | * the pageblock boundary, because we do |
782 | * block_start_pfn -= pageblock_nr_pages in the for loop. | |
783 | * For ending point, take care when isolating in last pageblock of a | |
784 | * a zone which ends in the middle of a pageblock. | |
49e068f0 VB |
785 | * The low boundary is the end of the pageblock the migration scanner |
786 | * is using. | |
ff9543fd | 787 | */ |
c96b9e50 VB |
788 | block_start_pfn = cc->free_pfn & ~(pageblock_nr_pages-1); |
789 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, | |
790 | zone_end_pfn(zone)); | |
7ed695e0 | 791 | low_pfn = ALIGN(cc->migrate_pfn + 1, pageblock_nr_pages); |
2fe86e00 | 792 | |
ff9543fd MN |
793 | /* |
794 | * Isolate free pages until enough are available to migrate the | |
795 | * pages on cc->migratepages. We stop searching if the migrate | |
796 | * and free page scanners meet or enough free pages are isolated. | |
797 | */ | |
c96b9e50 VB |
798 | for (; block_start_pfn >= low_pfn && cc->nr_migratepages > nr_freepages; |
799 | block_end_pfn = block_start_pfn, | |
800 | block_start_pfn -= pageblock_nr_pages) { | |
ff9543fd | 801 | unsigned long isolated; |
2fe86e00 | 802 | |
f6ea3adb DR |
803 | /* |
804 | * This can iterate a massively long zone without finding any | |
805 | * suitable migration targets, so periodically check if we need | |
be976572 | 806 | * to schedule, or even abort async compaction. |
f6ea3adb | 807 | */ |
be976572 VB |
808 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) |
809 | && compact_should_abort(cc)) | |
810 | break; | |
f6ea3adb | 811 | |
7d49d886 VB |
812 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
813 | zone); | |
814 | if (!page) | |
ff9543fd MN |
815 | continue; |
816 | ||
817 | /* Check the block is suitable for migration */ | |
68e3e926 | 818 | if (!suitable_migration_target(page)) |
ff9543fd | 819 | continue; |
68e3e926 | 820 | |
bb13ffeb MG |
821 | /* If isolation recently failed, do not retry */ |
822 | if (!isolation_suitable(cc, page)) | |
823 | continue; | |
824 | ||
f40d1e42 | 825 | /* Found a block suitable for isolating free pages from */ |
e9ade569 | 826 | cc->free_pfn = block_start_pfn; |
c96b9e50 VB |
827 | isolated = isolate_freepages_block(cc, block_start_pfn, |
828 | block_end_pfn, freelist, false); | |
f40d1e42 | 829 | nr_freepages += isolated; |
ff9543fd MN |
830 | |
831 | /* | |
e9ade569 VB |
832 | * Set a flag that we successfully isolated in this pageblock. |
833 | * In the next loop iteration, zone->compact_cached_free_pfn | |
834 | * will not be updated and thus it will effectively contain the | |
835 | * highest pageblock we isolated pages from. | |
ff9543fd | 836 | */ |
e9ade569 | 837 | if (isolated) |
c89511ab | 838 | cc->finished_update_free = true; |
be976572 VB |
839 | |
840 | /* | |
841 | * isolate_freepages_block() might have aborted due to async | |
842 | * compaction being contended | |
843 | */ | |
844 | if (cc->contended) | |
845 | break; | |
ff9543fd MN |
846 | } |
847 | ||
848 | /* split_free_page does not map the pages */ | |
849 | map_pages(freelist); | |
850 | ||
7ed695e0 VB |
851 | /* |
852 | * If we crossed the migrate scanner, we want to keep it that way | |
853 | * so that compact_finished() may detect this | |
854 | */ | |
c96b9e50 | 855 | if (block_start_pfn < low_pfn) |
e9ade569 | 856 | cc->free_pfn = cc->migrate_pfn; |
c96b9e50 | 857 | |
ff9543fd | 858 | cc->nr_freepages = nr_freepages; |
748446bb MG |
859 | } |
860 | ||
861 | /* | |
862 | * This is a migrate-callback that "allocates" freepages by taking pages | |
863 | * from the isolated freelists in the block we are migrating to. | |
864 | */ | |
865 | static struct page *compaction_alloc(struct page *migratepage, | |
866 | unsigned long data, | |
867 | int **result) | |
868 | { | |
869 | struct compact_control *cc = (struct compact_control *)data; | |
870 | struct page *freepage; | |
871 | ||
be976572 VB |
872 | /* |
873 | * Isolate free pages if necessary, and if we are not aborting due to | |
874 | * contention. | |
875 | */ | |
748446bb | 876 | if (list_empty(&cc->freepages)) { |
be976572 | 877 | if (!cc->contended) |
edc2ca61 | 878 | isolate_freepages(cc); |
748446bb MG |
879 | |
880 | if (list_empty(&cc->freepages)) | |
881 | return NULL; | |
882 | } | |
883 | ||
884 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
885 | list_del(&freepage->lru); | |
886 | cc->nr_freepages--; | |
887 | ||
888 | return freepage; | |
889 | } | |
890 | ||
891 | /* | |
d53aea3d DR |
892 | * This is a migrate-callback that "frees" freepages back to the isolated |
893 | * freelist. All pages on the freelist are from the same zone, so there is no | |
894 | * special handling needed for NUMA. | |
895 | */ | |
896 | static void compaction_free(struct page *page, unsigned long data) | |
897 | { | |
898 | struct compact_control *cc = (struct compact_control *)data; | |
899 | ||
900 | list_add(&page->lru, &cc->freepages); | |
901 | cc->nr_freepages++; | |
902 | } | |
903 | ||
ff9543fd MN |
904 | /* possible outcome of isolate_migratepages */ |
905 | typedef enum { | |
906 | ISOLATE_ABORT, /* Abort compaction now */ | |
907 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
908 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
909 | } isolate_migrate_t; | |
910 | ||
911 | /* | |
edc2ca61 VB |
912 | * Isolate all pages that can be migrated from the first suitable block, |
913 | * starting at the block pointed to by the migrate scanner pfn within | |
914 | * compact_control. | |
ff9543fd MN |
915 | */ |
916 | static isolate_migrate_t isolate_migratepages(struct zone *zone, | |
917 | struct compact_control *cc) | |
918 | { | |
919 | unsigned long low_pfn, end_pfn; | |
edc2ca61 VB |
920 | struct page *page; |
921 | const isolate_mode_t isolate_mode = | |
922 | (cc->mode == MIGRATE_ASYNC ? ISOLATE_ASYNC_MIGRATE : 0); | |
ff9543fd | 923 | |
edc2ca61 VB |
924 | /* |
925 | * Start at where we last stopped, or beginning of the zone as | |
926 | * initialized by compact_zone() | |
927 | */ | |
928 | low_pfn = cc->migrate_pfn; | |
ff9543fd MN |
929 | |
930 | /* Only scan within a pageblock boundary */ | |
a9aacbcc | 931 | end_pfn = ALIGN(low_pfn + 1, pageblock_nr_pages); |
ff9543fd | 932 | |
edc2ca61 VB |
933 | /* |
934 | * Iterate over whole pageblocks until we find the first suitable. | |
935 | * Do not cross the free scanner. | |
936 | */ | |
937 | for (; end_pfn <= cc->free_pfn; | |
938 | low_pfn = end_pfn, end_pfn += pageblock_nr_pages) { | |
ff9543fd | 939 | |
edc2ca61 VB |
940 | /* |
941 | * This can potentially iterate a massively long zone with | |
942 | * many pageblocks unsuitable, so periodically check if we | |
943 | * need to schedule, or even abort async compaction. | |
944 | */ | |
945 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages)) | |
946 | && compact_should_abort(cc)) | |
947 | break; | |
ff9543fd | 948 | |
7d49d886 VB |
949 | page = pageblock_pfn_to_page(low_pfn, end_pfn, zone); |
950 | if (!page) | |
edc2ca61 VB |
951 | continue; |
952 | ||
edc2ca61 VB |
953 | /* If isolation recently failed, do not retry */ |
954 | if (!isolation_suitable(cc, page)) | |
955 | continue; | |
956 | ||
957 | /* | |
958 | * For async compaction, also only scan in MOVABLE blocks. | |
959 | * Async compaction is optimistic to see if the minimum amount | |
960 | * of work satisfies the allocation. | |
961 | */ | |
962 | if (cc->mode == MIGRATE_ASYNC && | |
963 | !migrate_async_suitable(get_pageblock_migratetype(page))) | |
964 | continue; | |
965 | ||
966 | /* Perform the isolation */ | |
967 | low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn, | |
968 | isolate_mode); | |
969 | ||
970 | if (!low_pfn || cc->contended) | |
971 | return ISOLATE_ABORT; | |
972 | ||
973 | /* | |
974 | * Either we isolated something and proceed with migration. Or | |
975 | * we failed and compact_zone should decide if we should | |
976 | * continue or not. | |
977 | */ | |
978 | break; | |
979 | } | |
980 | ||
981 | acct_isolated(zone, cc); | |
982 | /* Record where migration scanner will be restarted */ | |
ff9543fd MN |
983 | cc->migrate_pfn = low_pfn; |
984 | ||
edc2ca61 | 985 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
ff9543fd MN |
986 | } |
987 | ||
748446bb | 988 | static int compact_finished(struct zone *zone, |
5a03b051 | 989 | struct compact_control *cc) |
748446bb | 990 | { |
8fb74b9f | 991 | unsigned int order; |
5a03b051 | 992 | unsigned long watermark; |
56de7263 | 993 | |
be976572 | 994 | if (cc->contended || fatal_signal_pending(current)) |
748446bb MG |
995 | return COMPACT_PARTIAL; |
996 | ||
753341a4 | 997 | /* Compaction run completes if the migrate and free scanner meet */ |
bb13ffeb | 998 | if (cc->free_pfn <= cc->migrate_pfn) { |
55b7c4c9 | 999 | /* Let the next compaction start anew. */ |
35979ef3 DR |
1000 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; |
1001 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
55b7c4c9 VB |
1002 | zone->compact_cached_free_pfn = zone_end_pfn(zone); |
1003 | ||
62997027 MG |
1004 | /* |
1005 | * Mark that the PG_migrate_skip information should be cleared | |
1006 | * by kswapd when it goes to sleep. kswapd does not set the | |
1007 | * flag itself as the decision to be clear should be directly | |
1008 | * based on an allocation request. | |
1009 | */ | |
1010 | if (!current_is_kswapd()) | |
1011 | zone->compact_blockskip_flush = true; | |
1012 | ||
748446bb | 1013 | return COMPACT_COMPLETE; |
bb13ffeb | 1014 | } |
748446bb | 1015 | |
82478fb7 JW |
1016 | /* |
1017 | * order == -1 is expected when compacting via | |
1018 | * /proc/sys/vm/compact_memory | |
1019 | */ | |
56de7263 MG |
1020 | if (cc->order == -1) |
1021 | return COMPACT_CONTINUE; | |
1022 | ||
3957c776 MH |
1023 | /* Compaction run is not finished if the watermark is not met */ |
1024 | watermark = low_wmark_pages(zone); | |
1025 | watermark += (1 << cc->order); | |
1026 | ||
1027 | if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0)) | |
1028 | return COMPACT_CONTINUE; | |
1029 | ||
56de7263 | 1030 | /* Direct compactor: Is a suitable page free? */ |
8fb74b9f MG |
1031 | for (order = cc->order; order < MAX_ORDER; order++) { |
1032 | struct free_area *area = &zone->free_area[order]; | |
1033 | ||
1034 | /* Job done if page is free of the right migratetype */ | |
1035 | if (!list_empty(&area->free_list[cc->migratetype])) | |
1036 | return COMPACT_PARTIAL; | |
1037 | ||
1038 | /* Job done if allocation would set block type */ | |
1039 | if (cc->order >= pageblock_order && area->nr_free) | |
56de7263 MG |
1040 | return COMPACT_PARTIAL; |
1041 | } | |
1042 | ||
748446bb MG |
1043 | return COMPACT_CONTINUE; |
1044 | } | |
1045 | ||
3e7d3449 MG |
1046 | /* |
1047 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
1048 | * Returns | |
1049 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
1050 | * COMPACT_PARTIAL - If the allocation would succeed without compaction | |
1051 | * COMPACT_CONTINUE - If compaction should run now | |
1052 | */ | |
1053 | unsigned long compaction_suitable(struct zone *zone, int order) | |
1054 | { | |
1055 | int fragindex; | |
1056 | unsigned long watermark; | |
1057 | ||
3957c776 MH |
1058 | /* |
1059 | * order == -1 is expected when compacting via | |
1060 | * /proc/sys/vm/compact_memory | |
1061 | */ | |
1062 | if (order == -1) | |
1063 | return COMPACT_CONTINUE; | |
1064 | ||
3e7d3449 MG |
1065 | /* |
1066 | * Watermarks for order-0 must be met for compaction. Note the 2UL. | |
1067 | * This is because during migration, copies of pages need to be | |
1068 | * allocated and for a short time, the footprint is higher | |
1069 | */ | |
1070 | watermark = low_wmark_pages(zone) + (2UL << order); | |
1071 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) | |
1072 | return COMPACT_SKIPPED; | |
1073 | ||
1074 | /* | |
1075 | * fragmentation index determines if allocation failures are due to | |
1076 | * low memory or external fragmentation | |
1077 | * | |
a582a738 SL |
1078 | * index of -1000 implies allocations might succeed depending on |
1079 | * watermarks | |
3e7d3449 MG |
1080 | * index towards 0 implies failure is due to lack of memory |
1081 | * index towards 1000 implies failure is due to fragmentation | |
1082 | * | |
1083 | * Only compact if a failure would be due to fragmentation. | |
1084 | */ | |
1085 | fragindex = fragmentation_index(zone, order); | |
1086 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
1087 | return COMPACT_SKIPPED; | |
1088 | ||
a582a738 SL |
1089 | if (fragindex == -1000 && zone_watermark_ok(zone, order, watermark, |
1090 | 0, 0)) | |
3e7d3449 MG |
1091 | return COMPACT_PARTIAL; |
1092 | ||
1093 | return COMPACT_CONTINUE; | |
1094 | } | |
1095 | ||
748446bb MG |
1096 | static int compact_zone(struct zone *zone, struct compact_control *cc) |
1097 | { | |
1098 | int ret; | |
c89511ab | 1099 | unsigned long start_pfn = zone->zone_start_pfn; |
108bcc96 | 1100 | unsigned long end_pfn = zone_end_pfn(zone); |
e0b9daeb | 1101 | const bool sync = cc->mode != MIGRATE_ASYNC; |
748446bb | 1102 | |
3e7d3449 MG |
1103 | ret = compaction_suitable(zone, cc->order); |
1104 | switch (ret) { | |
1105 | case COMPACT_PARTIAL: | |
1106 | case COMPACT_SKIPPED: | |
1107 | /* Compaction is likely to fail */ | |
1108 | return ret; | |
1109 | case COMPACT_CONTINUE: | |
1110 | /* Fall through to compaction */ | |
1111 | ; | |
1112 | } | |
1113 | ||
d3132e4b VB |
1114 | /* |
1115 | * Clear pageblock skip if there were failures recently and compaction | |
1116 | * is about to be retried after being deferred. kswapd does not do | |
1117 | * this reset as it'll reset the cached information when going to sleep. | |
1118 | */ | |
1119 | if (compaction_restarting(zone, cc->order) && !current_is_kswapd()) | |
1120 | __reset_isolation_suitable(zone); | |
1121 | ||
c89511ab MG |
1122 | /* |
1123 | * Setup to move all movable pages to the end of the zone. Used cached | |
1124 | * information on where the scanners should start but check that it | |
1125 | * is initialised by ensuring the values are within zone boundaries. | |
1126 | */ | |
e0b9daeb | 1127 | cc->migrate_pfn = zone->compact_cached_migrate_pfn[sync]; |
c89511ab MG |
1128 | cc->free_pfn = zone->compact_cached_free_pfn; |
1129 | if (cc->free_pfn < start_pfn || cc->free_pfn > end_pfn) { | |
1130 | cc->free_pfn = end_pfn & ~(pageblock_nr_pages-1); | |
1131 | zone->compact_cached_free_pfn = cc->free_pfn; | |
1132 | } | |
1133 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn > end_pfn) { | |
1134 | cc->migrate_pfn = start_pfn; | |
35979ef3 DR |
1135 | zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
1136 | zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
c89511ab | 1137 | } |
748446bb | 1138 | |
0eb927c0 MG |
1139 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, cc->free_pfn, end_pfn); |
1140 | ||
748446bb MG |
1141 | migrate_prep_local(); |
1142 | ||
1143 | while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) { | |
9d502c1c | 1144 | int err; |
748446bb | 1145 | |
f9e35b3b MG |
1146 | switch (isolate_migratepages(zone, cc)) { |
1147 | case ISOLATE_ABORT: | |
1148 | ret = COMPACT_PARTIAL; | |
5733c7d1 | 1149 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 1150 | cc->nr_migratepages = 0; |
f9e35b3b MG |
1151 | goto out; |
1152 | case ISOLATE_NONE: | |
748446bb | 1153 | continue; |
f9e35b3b MG |
1154 | case ISOLATE_SUCCESS: |
1155 | ; | |
1156 | } | |
748446bb | 1157 | |
d53aea3d | 1158 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 1159 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 1160 | MR_COMPACTION); |
748446bb | 1161 | |
f8c9301f VB |
1162 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
1163 | &cc->migratepages); | |
748446bb | 1164 | |
f8c9301f VB |
1165 | /* All pages were either migrated or will be released */ |
1166 | cc->nr_migratepages = 0; | |
9d502c1c | 1167 | if (err) { |
5733c7d1 | 1168 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
1169 | /* |
1170 | * migrate_pages() may return -ENOMEM when scanners meet | |
1171 | * and we want compact_finished() to detect it | |
1172 | */ | |
1173 | if (err == -ENOMEM && cc->free_pfn > cc->migrate_pfn) { | |
4bf2bba3 DR |
1174 | ret = COMPACT_PARTIAL; |
1175 | goto out; | |
1176 | } | |
748446bb | 1177 | } |
748446bb MG |
1178 | } |
1179 | ||
f9e35b3b | 1180 | out: |
748446bb MG |
1181 | /* Release free pages and check accounting */ |
1182 | cc->nr_freepages -= release_freepages(&cc->freepages); | |
1183 | VM_BUG_ON(cc->nr_freepages != 0); | |
1184 | ||
0eb927c0 MG |
1185 | trace_mm_compaction_end(ret); |
1186 | ||
748446bb MG |
1187 | return ret; |
1188 | } | |
76ab0f53 | 1189 | |
e0b9daeb | 1190 | static unsigned long compact_zone_order(struct zone *zone, int order, |
1f9efdef | 1191 | gfp_t gfp_mask, enum migrate_mode mode, int *contended) |
56de7263 | 1192 | { |
e64c5237 | 1193 | unsigned long ret; |
56de7263 MG |
1194 | struct compact_control cc = { |
1195 | .nr_freepages = 0, | |
1196 | .nr_migratepages = 0, | |
1197 | .order = order, | |
1198 | .migratetype = allocflags_to_migratetype(gfp_mask), | |
1199 | .zone = zone, | |
e0b9daeb | 1200 | .mode = mode, |
56de7263 MG |
1201 | }; |
1202 | INIT_LIST_HEAD(&cc.freepages); | |
1203 | INIT_LIST_HEAD(&cc.migratepages); | |
1204 | ||
e64c5237 SL |
1205 | ret = compact_zone(zone, &cc); |
1206 | ||
1207 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
1208 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
1209 | ||
1210 | *contended = cc.contended; | |
1211 | return ret; | |
56de7263 MG |
1212 | } |
1213 | ||
5e771905 MG |
1214 | int sysctl_extfrag_threshold = 500; |
1215 | ||
56de7263 MG |
1216 | /** |
1217 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
1218 | * @zonelist: The zonelist used for the current allocation | |
1219 | * @order: The order of the current allocation | |
1220 | * @gfp_mask: The GFP mask of the current allocation | |
1221 | * @nodemask: The allowed nodes to allocate from | |
e0b9daeb | 1222 | * @mode: The migration mode for async, sync light, or sync migration |
1f9efdef VB |
1223 | * @contended: Return value that determines if compaction was aborted due to |
1224 | * need_resched() or lock contention | |
53853e2d | 1225 | * @candidate_zone: Return the zone where we think allocation should succeed |
56de7263 MG |
1226 | * |
1227 | * This is the main entry point for direct page compaction. | |
1228 | */ | |
1229 | unsigned long try_to_compact_pages(struct zonelist *zonelist, | |
77f1fe6b | 1230 | int order, gfp_t gfp_mask, nodemask_t *nodemask, |
1f9efdef | 1231 | enum migrate_mode mode, int *contended, |
53853e2d | 1232 | struct zone **candidate_zone) |
56de7263 MG |
1233 | { |
1234 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1235 | int may_enter_fs = gfp_mask & __GFP_FS; | |
1236 | int may_perform_io = gfp_mask & __GFP_IO; | |
56de7263 MG |
1237 | struct zoneref *z; |
1238 | struct zone *zone; | |
53853e2d | 1239 | int rc = COMPACT_DEFERRED; |
d95ea5d1 | 1240 | int alloc_flags = 0; |
1f9efdef VB |
1241 | int all_zones_contended = COMPACT_CONTENDED_LOCK; /* init for &= op */ |
1242 | ||
1243 | *contended = COMPACT_CONTENDED_NONE; | |
56de7263 | 1244 | |
4ffb6335 | 1245 | /* Check if the GFP flags allow compaction */ |
c5a73c3d | 1246 | if (!order || !may_enter_fs || !may_perform_io) |
53853e2d | 1247 | return COMPACT_SKIPPED; |
56de7263 | 1248 | |
d95ea5d1 BZ |
1249 | #ifdef CONFIG_CMA |
1250 | if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE) | |
1251 | alloc_flags |= ALLOC_CMA; | |
1252 | #endif | |
56de7263 MG |
1253 | /* Compact each zone in the list */ |
1254 | for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx, | |
1255 | nodemask) { | |
56de7263 | 1256 | int status; |
1f9efdef | 1257 | int zone_contended; |
56de7263 | 1258 | |
53853e2d VB |
1259 | if (compaction_deferred(zone, order)) |
1260 | continue; | |
1261 | ||
e0b9daeb | 1262 | status = compact_zone_order(zone, order, gfp_mask, mode, |
1f9efdef | 1263 | &zone_contended); |
56de7263 | 1264 | rc = max(status, rc); |
1f9efdef VB |
1265 | /* |
1266 | * It takes at least one zone that wasn't lock contended | |
1267 | * to clear all_zones_contended. | |
1268 | */ | |
1269 | all_zones_contended &= zone_contended; | |
56de7263 | 1270 | |
3e7d3449 | 1271 | /* If a normal allocation would succeed, stop compacting */ |
d95ea5d1 | 1272 | if (zone_watermark_ok(zone, order, low_wmark_pages(zone), 0, |
53853e2d VB |
1273 | alloc_flags)) { |
1274 | *candidate_zone = zone; | |
1275 | /* | |
1276 | * We think the allocation will succeed in this zone, | |
1277 | * but it is not certain, hence the false. The caller | |
1278 | * will repeat this with true if allocation indeed | |
1279 | * succeeds in this zone. | |
1280 | */ | |
1281 | compaction_defer_reset(zone, order, false); | |
1f9efdef VB |
1282 | /* |
1283 | * It is possible that async compaction aborted due to | |
1284 | * need_resched() and the watermarks were ok thanks to | |
1285 | * somebody else freeing memory. The allocation can | |
1286 | * however still fail so we better signal the | |
1287 | * need_resched() contention anyway (this will not | |
1288 | * prevent the allocation attempt). | |
1289 | */ | |
1290 | if (zone_contended == COMPACT_CONTENDED_SCHED) | |
1291 | *contended = COMPACT_CONTENDED_SCHED; | |
1292 | ||
1293 | goto break_loop; | |
1294 | } | |
1295 | ||
1296 | if (mode != MIGRATE_ASYNC) { | |
53853e2d VB |
1297 | /* |
1298 | * We think that allocation won't succeed in this zone | |
1299 | * so we defer compaction there. If it ends up | |
1300 | * succeeding after all, it will be reset. | |
1301 | */ | |
1302 | defer_compaction(zone, order); | |
1303 | } | |
1f9efdef VB |
1304 | |
1305 | /* | |
1306 | * We might have stopped compacting due to need_resched() in | |
1307 | * async compaction, or due to a fatal signal detected. In that | |
1308 | * case do not try further zones and signal need_resched() | |
1309 | * contention. | |
1310 | */ | |
1311 | if ((zone_contended == COMPACT_CONTENDED_SCHED) | |
1312 | || fatal_signal_pending(current)) { | |
1313 | *contended = COMPACT_CONTENDED_SCHED; | |
1314 | goto break_loop; | |
1315 | } | |
1316 | ||
1317 | continue; | |
1318 | break_loop: | |
1319 | /* | |
1320 | * We might not have tried all the zones, so be conservative | |
1321 | * and assume they are not all lock contended. | |
1322 | */ | |
1323 | all_zones_contended = 0; | |
1324 | break; | |
56de7263 MG |
1325 | } |
1326 | ||
1f9efdef VB |
1327 | /* |
1328 | * If at least one zone wasn't deferred or skipped, we report if all | |
1329 | * zones that were tried were lock contended. | |
1330 | */ | |
1331 | if (rc > COMPACT_SKIPPED && all_zones_contended) | |
1332 | *contended = COMPACT_CONTENDED_LOCK; | |
1333 | ||
56de7263 MG |
1334 | return rc; |
1335 | } | |
1336 | ||
1337 | ||
76ab0f53 | 1338 | /* Compact all zones within a node */ |
7103f16d | 1339 | static void __compact_pgdat(pg_data_t *pgdat, struct compact_control *cc) |
76ab0f53 MG |
1340 | { |
1341 | int zoneid; | |
76ab0f53 MG |
1342 | struct zone *zone; |
1343 | ||
76ab0f53 | 1344 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
1345 | |
1346 | zone = &pgdat->node_zones[zoneid]; | |
1347 | if (!populated_zone(zone)) | |
1348 | continue; | |
1349 | ||
7be62de9 RR |
1350 | cc->nr_freepages = 0; |
1351 | cc->nr_migratepages = 0; | |
1352 | cc->zone = zone; | |
1353 | INIT_LIST_HEAD(&cc->freepages); | |
1354 | INIT_LIST_HEAD(&cc->migratepages); | |
76ab0f53 | 1355 | |
aad6ec37 | 1356 | if (cc->order == -1 || !compaction_deferred(zone, cc->order)) |
7be62de9 | 1357 | compact_zone(zone, cc); |
76ab0f53 | 1358 | |
aff62249 | 1359 | if (cc->order > 0) { |
de6c60a6 VB |
1360 | if (zone_watermark_ok(zone, cc->order, |
1361 | low_wmark_pages(zone), 0, 0)) | |
1362 | compaction_defer_reset(zone, cc->order, false); | |
aff62249 RR |
1363 | } |
1364 | ||
7be62de9 RR |
1365 | VM_BUG_ON(!list_empty(&cc->freepages)); |
1366 | VM_BUG_ON(!list_empty(&cc->migratepages)); | |
76ab0f53 | 1367 | } |
76ab0f53 MG |
1368 | } |
1369 | ||
7103f16d | 1370 | void compact_pgdat(pg_data_t *pgdat, int order) |
7be62de9 RR |
1371 | { |
1372 | struct compact_control cc = { | |
1373 | .order = order, | |
e0b9daeb | 1374 | .mode = MIGRATE_ASYNC, |
7be62de9 RR |
1375 | }; |
1376 | ||
3a7200af MG |
1377 | if (!order) |
1378 | return; | |
1379 | ||
7103f16d | 1380 | __compact_pgdat(pgdat, &cc); |
7be62de9 RR |
1381 | } |
1382 | ||
7103f16d | 1383 | static void compact_node(int nid) |
7be62de9 | 1384 | { |
7be62de9 RR |
1385 | struct compact_control cc = { |
1386 | .order = -1, | |
e0b9daeb | 1387 | .mode = MIGRATE_SYNC, |
91ca9186 | 1388 | .ignore_skip_hint = true, |
7be62de9 RR |
1389 | }; |
1390 | ||
7103f16d | 1391 | __compact_pgdat(NODE_DATA(nid), &cc); |
7be62de9 RR |
1392 | } |
1393 | ||
76ab0f53 | 1394 | /* Compact all nodes in the system */ |
7964c06d | 1395 | static void compact_nodes(void) |
76ab0f53 MG |
1396 | { |
1397 | int nid; | |
1398 | ||
8575ec29 HD |
1399 | /* Flush pending updates to the LRU lists */ |
1400 | lru_add_drain_all(); | |
1401 | ||
76ab0f53 MG |
1402 | for_each_online_node(nid) |
1403 | compact_node(nid); | |
76ab0f53 MG |
1404 | } |
1405 | ||
1406 | /* The written value is actually unused, all memory is compacted */ | |
1407 | int sysctl_compact_memory; | |
1408 | ||
1409 | /* This is the entry point for compacting all nodes via /proc/sys/vm */ | |
1410 | int sysctl_compaction_handler(struct ctl_table *table, int write, | |
1411 | void __user *buffer, size_t *length, loff_t *ppos) | |
1412 | { | |
1413 | if (write) | |
7964c06d | 1414 | compact_nodes(); |
76ab0f53 MG |
1415 | |
1416 | return 0; | |
1417 | } | |
ed4a6d7f | 1418 | |
5e771905 MG |
1419 | int sysctl_extfrag_handler(struct ctl_table *table, int write, |
1420 | void __user *buffer, size_t *length, loff_t *ppos) | |
1421 | { | |
1422 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
1423 | ||
1424 | return 0; | |
1425 | } | |
1426 | ||
ed4a6d7f | 1427 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) |
74e77fb9 | 1428 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 1429 | struct device_attribute *attr, |
ed4a6d7f MG |
1430 | const char *buf, size_t count) |
1431 | { | |
8575ec29 HD |
1432 | int nid = dev->id; |
1433 | ||
1434 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
1435 | /* Flush pending updates to the LRU lists */ | |
1436 | lru_add_drain_all(); | |
1437 | ||
1438 | compact_node(nid); | |
1439 | } | |
ed4a6d7f MG |
1440 | |
1441 | return count; | |
1442 | } | |
10fbcf4c | 1443 | static DEVICE_ATTR(compact, S_IWUSR, NULL, sysfs_compact_node); |
ed4a6d7f MG |
1444 | |
1445 | int compaction_register_node(struct node *node) | |
1446 | { | |
10fbcf4c | 1447 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1448 | } |
1449 | ||
1450 | void compaction_unregister_node(struct node *node) | |
1451 | { | |
10fbcf4c | 1452 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
1453 | } |
1454 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd MN |
1455 | |
1456 | #endif /* CONFIG_COMPACTION */ |