Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
748446bb MG |
2 | /* |
3 | * linux/mm/compaction.c | |
4 | * | |
5 | * Memory compaction for the reduction of external fragmentation. Note that | |
6 | * this heavily depends upon page migration to do all the real heavy | |
7 | * lifting | |
8 | * | |
9 | * Copyright IBM Corp. 2007-2010 Mel Gorman <mel@csn.ul.ie> | |
10 | */ | |
698b1b30 | 11 | #include <linux/cpu.h> |
748446bb MG |
12 | #include <linux/swap.h> |
13 | #include <linux/migrate.h> | |
14 | #include <linux/compaction.h> | |
15 | #include <linux/mm_inline.h> | |
174cd4b1 | 16 | #include <linux/sched/signal.h> |
748446bb | 17 | #include <linux/backing-dev.h> |
76ab0f53 | 18 | #include <linux/sysctl.h> |
ed4a6d7f | 19 | #include <linux/sysfs.h> |
194159fb | 20 | #include <linux/page-isolation.h> |
b8c73fc2 | 21 | #include <linux/kasan.h> |
698b1b30 VB |
22 | #include <linux/kthread.h> |
23 | #include <linux/freezer.h> | |
83358ece | 24 | #include <linux/page_owner.h> |
eb414681 | 25 | #include <linux/psi.h> |
748446bb MG |
26 | #include "internal.h" |
27 | ||
010fc29a MK |
28 | #ifdef CONFIG_COMPACTION |
29 | static inline void count_compact_event(enum vm_event_item item) | |
30 | { | |
31 | count_vm_event(item); | |
32 | } | |
33 | ||
34 | static inline void count_compact_events(enum vm_event_item item, long delta) | |
35 | { | |
36 | count_vm_events(item, delta); | |
37 | } | |
38 | #else | |
39 | #define count_compact_event(item) do { } while (0) | |
40 | #define count_compact_events(item, delta) do { } while (0) | |
41 | #endif | |
42 | ||
ff9543fd MN |
43 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
44 | ||
b7aba698 MG |
45 | #define CREATE_TRACE_POINTS |
46 | #include <trace/events/compaction.h> | |
47 | ||
06b6640a VB |
48 | #define block_start_pfn(pfn, order) round_down(pfn, 1UL << (order)) |
49 | #define block_end_pfn(pfn, order) ALIGN((pfn) + 1, 1UL << (order)) | |
50 | #define pageblock_start_pfn(pfn) block_start_pfn(pfn, pageblock_order) | |
51 | #define pageblock_end_pfn(pfn) block_end_pfn(pfn, pageblock_order) | |
52 | ||
facdaa91 NG |
53 | /* |
54 | * Fragmentation score check interval for proactive compaction purposes. | |
55 | */ | |
d34c0a75 | 56 | static const unsigned int HPAGE_FRAG_CHECK_INTERVAL_MSEC = 500; |
facdaa91 NG |
57 | |
58 | /* | |
59 | * Page order with-respect-to which proactive compaction | |
60 | * calculates external fragmentation, which is used as | |
61 | * the "fragmentation score" of a node/zone. | |
62 | */ | |
63 | #if defined CONFIG_TRANSPARENT_HUGEPAGE | |
64 | #define COMPACTION_HPAGE_ORDER HPAGE_PMD_ORDER | |
25788738 | 65 | #elif defined CONFIG_HUGETLBFS |
facdaa91 NG |
66 | #define COMPACTION_HPAGE_ORDER HUGETLB_PAGE_ORDER |
67 | #else | |
68 | #define COMPACTION_HPAGE_ORDER (PMD_SHIFT - PAGE_SHIFT) | |
69 | #endif | |
70 | ||
748446bb MG |
71 | static unsigned long release_freepages(struct list_head *freelist) |
72 | { | |
73 | struct page *page, *next; | |
6bace090 | 74 | unsigned long high_pfn = 0; |
748446bb MG |
75 | |
76 | list_for_each_entry_safe(page, next, freelist, lru) { | |
6bace090 | 77 | unsigned long pfn = page_to_pfn(page); |
748446bb MG |
78 | list_del(&page->lru); |
79 | __free_page(page); | |
6bace090 VB |
80 | if (pfn > high_pfn) |
81 | high_pfn = pfn; | |
748446bb MG |
82 | } |
83 | ||
6bace090 | 84 | return high_pfn; |
748446bb MG |
85 | } |
86 | ||
4469ab98 | 87 | static void split_map_pages(struct list_head *list) |
ff9543fd | 88 | { |
66c64223 JK |
89 | unsigned int i, order, nr_pages; |
90 | struct page *page, *next; | |
91 | LIST_HEAD(tmp_list); | |
92 | ||
93 | list_for_each_entry_safe(page, next, list, lru) { | |
94 | list_del(&page->lru); | |
95 | ||
96 | order = page_private(page); | |
97 | nr_pages = 1 << order; | |
66c64223 | 98 | |
46f24fd8 | 99 | post_alloc_hook(page, order, __GFP_MOVABLE); |
66c64223 JK |
100 | if (order) |
101 | split_page(page, order); | |
ff9543fd | 102 | |
66c64223 JK |
103 | for (i = 0; i < nr_pages; i++) { |
104 | list_add(&page->lru, &tmp_list); | |
105 | page++; | |
106 | } | |
ff9543fd | 107 | } |
66c64223 JK |
108 | |
109 | list_splice(&tmp_list, list); | |
ff9543fd MN |
110 | } |
111 | ||
bb13ffeb | 112 | #ifdef CONFIG_COMPACTION |
24e2716f | 113 | |
bda807d4 MK |
114 | int PageMovable(struct page *page) |
115 | { | |
116 | struct address_space *mapping; | |
117 | ||
118 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
119 | if (!__PageMovable(page)) | |
120 | return 0; | |
121 | ||
122 | mapping = page_mapping(page); | |
123 | if (mapping && mapping->a_ops && mapping->a_ops->isolate_page) | |
124 | return 1; | |
125 | ||
126 | return 0; | |
127 | } | |
128 | EXPORT_SYMBOL(PageMovable); | |
129 | ||
130 | void __SetPageMovable(struct page *page, struct address_space *mapping) | |
131 | { | |
132 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
133 | VM_BUG_ON_PAGE((unsigned long)mapping & PAGE_MAPPING_MOVABLE, page); | |
134 | page->mapping = (void *)((unsigned long)mapping | PAGE_MAPPING_MOVABLE); | |
135 | } | |
136 | EXPORT_SYMBOL(__SetPageMovable); | |
137 | ||
138 | void __ClearPageMovable(struct page *page) | |
139 | { | |
140 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
141 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
142 | /* | |
143 | * Clear registered address_space val with keeping PAGE_MAPPING_MOVABLE | |
144 | * flag so that VM can catch up released page by driver after isolation. | |
145 | * With it, VM migration doesn't try to put it back. | |
146 | */ | |
147 | page->mapping = (void *)((unsigned long)page->mapping & | |
148 | PAGE_MAPPING_MOVABLE); | |
149 | } | |
150 | EXPORT_SYMBOL(__ClearPageMovable); | |
151 | ||
24e2716f JK |
152 | /* Do not skip compaction more than 64 times */ |
153 | #define COMPACT_MAX_DEFER_SHIFT 6 | |
154 | ||
155 | /* | |
156 | * Compaction is deferred when compaction fails to result in a page | |
860b3272 | 157 | * allocation success. 1 << compact_defer_shift, compactions are skipped up |
24e2716f JK |
158 | * to a limit of 1 << COMPACT_MAX_DEFER_SHIFT |
159 | */ | |
2271b016 | 160 | static void defer_compaction(struct zone *zone, int order) |
24e2716f JK |
161 | { |
162 | zone->compact_considered = 0; | |
163 | zone->compact_defer_shift++; | |
164 | ||
165 | if (order < zone->compact_order_failed) | |
166 | zone->compact_order_failed = order; | |
167 | ||
168 | if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT) | |
169 | zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT; | |
170 | ||
171 | trace_mm_compaction_defer_compaction(zone, order); | |
172 | } | |
173 | ||
174 | /* Returns true if compaction should be skipped this time */ | |
2271b016 | 175 | static bool compaction_deferred(struct zone *zone, int order) |
24e2716f JK |
176 | { |
177 | unsigned long defer_limit = 1UL << zone->compact_defer_shift; | |
178 | ||
179 | if (order < zone->compact_order_failed) | |
180 | return false; | |
181 | ||
182 | /* Avoid possible overflow */ | |
62b35fe0 | 183 | if (++zone->compact_considered >= defer_limit) { |
24e2716f | 184 | zone->compact_considered = defer_limit; |
24e2716f | 185 | return false; |
62b35fe0 | 186 | } |
24e2716f JK |
187 | |
188 | trace_mm_compaction_deferred(zone, order); | |
189 | ||
190 | return true; | |
191 | } | |
192 | ||
193 | /* | |
194 | * Update defer tracking counters after successful compaction of given order, | |
195 | * which means an allocation either succeeded (alloc_success == true) or is | |
196 | * expected to succeed. | |
197 | */ | |
198 | void compaction_defer_reset(struct zone *zone, int order, | |
199 | bool alloc_success) | |
200 | { | |
201 | if (alloc_success) { | |
202 | zone->compact_considered = 0; | |
203 | zone->compact_defer_shift = 0; | |
204 | } | |
205 | if (order >= zone->compact_order_failed) | |
206 | zone->compact_order_failed = order + 1; | |
207 | ||
208 | trace_mm_compaction_defer_reset(zone, order); | |
209 | } | |
210 | ||
211 | /* Returns true if restarting compaction after many failures */ | |
2271b016 | 212 | static bool compaction_restarting(struct zone *zone, int order) |
24e2716f JK |
213 | { |
214 | if (order < zone->compact_order_failed) | |
215 | return false; | |
216 | ||
217 | return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT && | |
218 | zone->compact_considered >= 1UL << zone->compact_defer_shift; | |
219 | } | |
220 | ||
bb13ffeb MG |
221 | /* Returns true if the pageblock should be scanned for pages to isolate. */ |
222 | static inline bool isolation_suitable(struct compact_control *cc, | |
223 | struct page *page) | |
224 | { | |
225 | if (cc->ignore_skip_hint) | |
226 | return true; | |
227 | ||
228 | return !get_pageblock_skip(page); | |
229 | } | |
230 | ||
02333641 VB |
231 | static void reset_cached_positions(struct zone *zone) |
232 | { | |
233 | zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn; | |
234 | zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn; | |
623446e4 | 235 | zone->compact_cached_free_pfn = |
06b6640a | 236 | pageblock_start_pfn(zone_end_pfn(zone) - 1); |
02333641 VB |
237 | } |
238 | ||
21dc7e02 | 239 | /* |
2271b016 | 240 | * Compound pages of >= pageblock_order should consistently be skipped until |
b527cfe5 VB |
241 | * released. It is always pointless to compact pages of such order (if they are |
242 | * migratable), and the pageblocks they occupy cannot contain any free pages. | |
21dc7e02 | 243 | */ |
b527cfe5 | 244 | static bool pageblock_skip_persistent(struct page *page) |
21dc7e02 | 245 | { |
b527cfe5 | 246 | if (!PageCompound(page)) |
21dc7e02 | 247 | return false; |
b527cfe5 VB |
248 | |
249 | page = compound_head(page); | |
250 | ||
251 | if (compound_order(page) >= pageblock_order) | |
252 | return true; | |
253 | ||
254 | return false; | |
21dc7e02 DR |
255 | } |
256 | ||
e332f741 MG |
257 | static bool |
258 | __reset_isolation_pfn(struct zone *zone, unsigned long pfn, bool check_source, | |
259 | bool check_target) | |
260 | { | |
261 | struct page *page = pfn_to_online_page(pfn); | |
6b0868c8 | 262 | struct page *block_page; |
e332f741 MG |
263 | struct page *end_page; |
264 | unsigned long block_pfn; | |
265 | ||
266 | if (!page) | |
267 | return false; | |
268 | if (zone != page_zone(page)) | |
269 | return false; | |
270 | if (pageblock_skip_persistent(page)) | |
271 | return false; | |
272 | ||
273 | /* | |
274 | * If skip is already cleared do no further checking once the | |
275 | * restart points have been set. | |
276 | */ | |
277 | if (check_source && check_target && !get_pageblock_skip(page)) | |
278 | return true; | |
279 | ||
280 | /* | |
281 | * If clearing skip for the target scanner, do not select a | |
282 | * non-movable pageblock as the starting point. | |
283 | */ | |
284 | if (!check_source && check_target && | |
285 | get_pageblock_migratetype(page) != MIGRATE_MOVABLE) | |
286 | return false; | |
287 | ||
6b0868c8 MG |
288 | /* Ensure the start of the pageblock or zone is online and valid */ |
289 | block_pfn = pageblock_start_pfn(pfn); | |
a2e9a5af VB |
290 | block_pfn = max(block_pfn, zone->zone_start_pfn); |
291 | block_page = pfn_to_online_page(block_pfn); | |
6b0868c8 MG |
292 | if (block_page) { |
293 | page = block_page; | |
294 | pfn = block_pfn; | |
295 | } | |
296 | ||
297 | /* Ensure the end of the pageblock or zone is online and valid */ | |
a2e9a5af | 298 | block_pfn = pageblock_end_pfn(pfn) - 1; |
6b0868c8 MG |
299 | block_pfn = min(block_pfn, zone_end_pfn(zone) - 1); |
300 | end_page = pfn_to_online_page(block_pfn); | |
301 | if (!end_page) | |
302 | return false; | |
303 | ||
e332f741 MG |
304 | /* |
305 | * Only clear the hint if a sample indicates there is either a | |
306 | * free page or an LRU page in the block. One or other condition | |
307 | * is necessary for the block to be a migration source/target. | |
308 | */ | |
e332f741 MG |
309 | do { |
310 | if (pfn_valid_within(pfn)) { | |
311 | if (check_source && PageLRU(page)) { | |
312 | clear_pageblock_skip(page); | |
313 | return true; | |
314 | } | |
315 | ||
316 | if (check_target && PageBuddy(page)) { | |
317 | clear_pageblock_skip(page); | |
318 | return true; | |
319 | } | |
320 | } | |
321 | ||
322 | page += (1 << PAGE_ALLOC_COSTLY_ORDER); | |
323 | pfn += (1 << PAGE_ALLOC_COSTLY_ORDER); | |
a2e9a5af | 324 | } while (page <= end_page); |
e332f741 MG |
325 | |
326 | return false; | |
327 | } | |
328 | ||
bb13ffeb MG |
329 | /* |
330 | * This function is called to clear all cached information on pageblocks that | |
331 | * should be skipped for page isolation when the migrate and free page scanner | |
332 | * meet. | |
333 | */ | |
62997027 | 334 | static void __reset_isolation_suitable(struct zone *zone) |
bb13ffeb | 335 | { |
e332f741 | 336 | unsigned long migrate_pfn = zone->zone_start_pfn; |
6b0868c8 | 337 | unsigned long free_pfn = zone_end_pfn(zone) - 1; |
e332f741 MG |
338 | unsigned long reset_migrate = free_pfn; |
339 | unsigned long reset_free = migrate_pfn; | |
340 | bool source_set = false; | |
341 | bool free_set = false; | |
342 | ||
343 | if (!zone->compact_blockskip_flush) | |
344 | return; | |
bb13ffeb | 345 | |
62997027 | 346 | zone->compact_blockskip_flush = false; |
bb13ffeb | 347 | |
e332f741 MG |
348 | /* |
349 | * Walk the zone and update pageblock skip information. Source looks | |
350 | * for PageLRU while target looks for PageBuddy. When the scanner | |
351 | * is found, both PageBuddy and PageLRU are checked as the pageblock | |
352 | * is suitable as both source and target. | |
353 | */ | |
354 | for (; migrate_pfn < free_pfn; migrate_pfn += pageblock_nr_pages, | |
355 | free_pfn -= pageblock_nr_pages) { | |
bb13ffeb MG |
356 | cond_resched(); |
357 | ||
e332f741 MG |
358 | /* Update the migrate PFN */ |
359 | if (__reset_isolation_pfn(zone, migrate_pfn, true, source_set) && | |
360 | migrate_pfn < reset_migrate) { | |
361 | source_set = true; | |
362 | reset_migrate = migrate_pfn; | |
363 | zone->compact_init_migrate_pfn = reset_migrate; | |
364 | zone->compact_cached_migrate_pfn[0] = reset_migrate; | |
365 | zone->compact_cached_migrate_pfn[1] = reset_migrate; | |
366 | } | |
bb13ffeb | 367 | |
e332f741 MG |
368 | /* Update the free PFN */ |
369 | if (__reset_isolation_pfn(zone, free_pfn, free_set, true) && | |
370 | free_pfn > reset_free) { | |
371 | free_set = true; | |
372 | reset_free = free_pfn; | |
373 | zone->compact_init_free_pfn = reset_free; | |
374 | zone->compact_cached_free_pfn = reset_free; | |
375 | } | |
bb13ffeb | 376 | } |
02333641 | 377 | |
e332f741 MG |
378 | /* Leave no distance if no suitable block was reset */ |
379 | if (reset_migrate >= reset_free) { | |
380 | zone->compact_cached_migrate_pfn[0] = migrate_pfn; | |
381 | zone->compact_cached_migrate_pfn[1] = migrate_pfn; | |
382 | zone->compact_cached_free_pfn = free_pfn; | |
383 | } | |
bb13ffeb MG |
384 | } |
385 | ||
62997027 MG |
386 | void reset_isolation_suitable(pg_data_t *pgdat) |
387 | { | |
388 | int zoneid; | |
389 | ||
390 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
391 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
392 | if (!populated_zone(zone)) | |
393 | continue; | |
394 | ||
395 | /* Only flush if a full compaction finished recently */ | |
396 | if (zone->compact_blockskip_flush) | |
397 | __reset_isolation_suitable(zone); | |
398 | } | |
399 | } | |
400 | ||
e380bebe MG |
401 | /* |
402 | * Sets the pageblock skip bit if it was clear. Note that this is a hint as | |
403 | * locks are not required for read/writers. Returns true if it was already set. | |
404 | */ | |
405 | static bool test_and_set_skip(struct compact_control *cc, struct page *page, | |
406 | unsigned long pfn) | |
407 | { | |
408 | bool skip; | |
409 | ||
410 | /* Do no update if skip hint is being ignored */ | |
411 | if (cc->ignore_skip_hint) | |
412 | return false; | |
413 | ||
414 | if (!IS_ALIGNED(pfn, pageblock_nr_pages)) | |
415 | return false; | |
416 | ||
417 | skip = get_pageblock_skip(page); | |
418 | if (!skip && !cc->no_set_skip_hint) | |
419 | set_pageblock_skip(page); | |
420 | ||
421 | return skip; | |
422 | } | |
423 | ||
424 | static void update_cached_migrate(struct compact_control *cc, unsigned long pfn) | |
425 | { | |
426 | struct zone *zone = cc->zone; | |
427 | ||
428 | pfn = pageblock_end_pfn(pfn); | |
429 | ||
430 | /* Set for isolation rather than compaction */ | |
431 | if (cc->no_set_skip_hint) | |
432 | return; | |
433 | ||
434 | if (pfn > zone->compact_cached_migrate_pfn[0]) | |
435 | zone->compact_cached_migrate_pfn[0] = pfn; | |
436 | if (cc->mode != MIGRATE_ASYNC && | |
437 | pfn > zone->compact_cached_migrate_pfn[1]) | |
438 | zone->compact_cached_migrate_pfn[1] = pfn; | |
439 | } | |
440 | ||
bb13ffeb MG |
441 | /* |
442 | * If no pages were isolated then mark this pageblock to be skipped in the | |
62997027 | 443 | * future. The information is later cleared by __reset_isolation_suitable(). |
bb13ffeb | 444 | */ |
c89511ab | 445 | static void update_pageblock_skip(struct compact_control *cc, |
d097a6f6 | 446 | struct page *page, unsigned long pfn) |
bb13ffeb | 447 | { |
c89511ab | 448 | struct zone *zone = cc->zone; |
6815bf3f | 449 | |
2583d671 | 450 | if (cc->no_set_skip_hint) |
6815bf3f JK |
451 | return; |
452 | ||
bb13ffeb MG |
453 | if (!page) |
454 | return; | |
455 | ||
edc2ca61 | 456 | set_pageblock_skip(page); |
c89511ab | 457 | |
35979ef3 | 458 | /* Update where async and sync compaction should restart */ |
e380bebe MG |
459 | if (pfn < zone->compact_cached_free_pfn) |
460 | zone->compact_cached_free_pfn = pfn; | |
bb13ffeb MG |
461 | } |
462 | #else | |
463 | static inline bool isolation_suitable(struct compact_control *cc, | |
464 | struct page *page) | |
465 | { | |
466 | return true; | |
467 | } | |
468 | ||
b527cfe5 | 469 | static inline bool pageblock_skip_persistent(struct page *page) |
21dc7e02 DR |
470 | { |
471 | return false; | |
472 | } | |
473 | ||
474 | static inline void update_pageblock_skip(struct compact_control *cc, | |
d097a6f6 | 475 | struct page *page, unsigned long pfn) |
bb13ffeb MG |
476 | { |
477 | } | |
e380bebe MG |
478 | |
479 | static void update_cached_migrate(struct compact_control *cc, unsigned long pfn) | |
480 | { | |
481 | } | |
482 | ||
483 | static bool test_and_set_skip(struct compact_control *cc, struct page *page, | |
484 | unsigned long pfn) | |
485 | { | |
486 | return false; | |
487 | } | |
bb13ffeb MG |
488 | #endif /* CONFIG_COMPACTION */ |
489 | ||
8b44d279 VB |
490 | /* |
491 | * Compaction requires the taking of some coarse locks that are potentially | |
cb2dcaf0 MG |
492 | * very heavily contended. For async compaction, trylock and record if the |
493 | * lock is contended. The lock will still be acquired but compaction will | |
494 | * abort when the current block is finished regardless of success rate. | |
495 | * Sync compaction acquires the lock. | |
8b44d279 | 496 | * |
cb2dcaf0 | 497 | * Always returns true which makes it easier to track lock state in callers. |
8b44d279 | 498 | */ |
cb2dcaf0 | 499 | static bool compact_lock_irqsave(spinlock_t *lock, unsigned long *flags, |
8b44d279 | 500 | struct compact_control *cc) |
77337ede | 501 | __acquires(lock) |
2a1402aa | 502 | { |
cb2dcaf0 MG |
503 | /* Track if the lock is contended in async mode */ |
504 | if (cc->mode == MIGRATE_ASYNC && !cc->contended) { | |
505 | if (spin_trylock_irqsave(lock, *flags)) | |
506 | return true; | |
507 | ||
508 | cc->contended = true; | |
8b44d279 | 509 | } |
1f9efdef | 510 | |
cb2dcaf0 | 511 | spin_lock_irqsave(lock, *flags); |
8b44d279 | 512 | return true; |
2a1402aa MG |
513 | } |
514 | ||
c67fe375 MG |
515 | /* |
516 | * Compaction requires the taking of some coarse locks that are potentially | |
8b44d279 VB |
517 | * very heavily contended. The lock should be periodically unlocked to avoid |
518 | * having disabled IRQs for a long time, even when there is nobody waiting on | |
519 | * the lock. It might also be that allowing the IRQs will result in | |
520 | * need_resched() becoming true. If scheduling is needed, async compaction | |
521 | * aborts. Sync compaction schedules. | |
522 | * Either compaction type will also abort if a fatal signal is pending. | |
523 | * In either case if the lock was locked, it is dropped and not regained. | |
c67fe375 | 524 | * |
8b44d279 VB |
525 | * Returns true if compaction should abort due to fatal signal pending, or |
526 | * async compaction due to need_resched() | |
527 | * Returns false when compaction can continue (sync compaction might have | |
528 | * scheduled) | |
c67fe375 | 529 | */ |
8b44d279 VB |
530 | static bool compact_unlock_should_abort(spinlock_t *lock, |
531 | unsigned long flags, bool *locked, struct compact_control *cc) | |
c67fe375 | 532 | { |
8b44d279 VB |
533 | if (*locked) { |
534 | spin_unlock_irqrestore(lock, flags); | |
535 | *locked = false; | |
536 | } | |
1f9efdef | 537 | |
8b44d279 | 538 | if (fatal_signal_pending(current)) { |
c3486f53 | 539 | cc->contended = true; |
8b44d279 VB |
540 | return true; |
541 | } | |
c67fe375 | 542 | |
cf66f070 | 543 | cond_resched(); |
be976572 VB |
544 | |
545 | return false; | |
546 | } | |
547 | ||
85aa125f | 548 | /* |
9e4be470 JM |
549 | * Isolate free pages onto a private freelist. If @strict is true, will abort |
550 | * returning 0 on any invalid PFNs or non-free pages inside of the pageblock | |
551 | * (even though it may still end up isolating some pages). | |
85aa125f | 552 | */ |
f40d1e42 | 553 | static unsigned long isolate_freepages_block(struct compact_control *cc, |
e14c720e | 554 | unsigned long *start_pfn, |
85aa125f MN |
555 | unsigned long end_pfn, |
556 | struct list_head *freelist, | |
4fca9730 | 557 | unsigned int stride, |
85aa125f | 558 | bool strict) |
748446bb | 559 | { |
b7aba698 | 560 | int nr_scanned = 0, total_isolated = 0; |
d097a6f6 | 561 | struct page *cursor; |
b8b2d825 | 562 | unsigned long flags = 0; |
f40d1e42 | 563 | bool locked = false; |
e14c720e | 564 | unsigned long blockpfn = *start_pfn; |
66c64223 | 565 | unsigned int order; |
748446bb | 566 | |
4fca9730 MG |
567 | /* Strict mode is for isolation, speed is secondary */ |
568 | if (strict) | |
569 | stride = 1; | |
570 | ||
748446bb MG |
571 | cursor = pfn_to_page(blockpfn); |
572 | ||
f40d1e42 | 573 | /* Isolate free pages. */ |
4fca9730 | 574 | for (; blockpfn < end_pfn; blockpfn += stride, cursor += stride) { |
66c64223 | 575 | int isolated; |
748446bb MG |
576 | struct page *page = cursor; |
577 | ||
8b44d279 VB |
578 | /* |
579 | * Periodically drop the lock (if held) regardless of its | |
580 | * contention, to give chance to IRQs. Abort if fatal signal | |
581 | * pending or async compaction detects need_resched() | |
582 | */ | |
583 | if (!(blockpfn % SWAP_CLUSTER_MAX) | |
584 | && compact_unlock_should_abort(&cc->zone->lock, flags, | |
585 | &locked, cc)) | |
586 | break; | |
587 | ||
b7aba698 | 588 | nr_scanned++; |
f40d1e42 | 589 | if (!pfn_valid_within(blockpfn)) |
2af120bc LA |
590 | goto isolate_fail; |
591 | ||
9fcd6d2e VB |
592 | /* |
593 | * For compound pages such as THP and hugetlbfs, we can save | |
594 | * potentially a lot of iterations if we skip them at once. | |
595 | * The check is racy, but we can consider only valid values | |
596 | * and the only danger is skipping too much. | |
597 | */ | |
598 | if (PageCompound(page)) { | |
21dc7e02 DR |
599 | const unsigned int order = compound_order(page); |
600 | ||
d3c85bad | 601 | if (likely(order < MAX_ORDER)) { |
21dc7e02 DR |
602 | blockpfn += (1UL << order) - 1; |
603 | cursor += (1UL << order) - 1; | |
9fcd6d2e | 604 | } |
9fcd6d2e VB |
605 | goto isolate_fail; |
606 | } | |
607 | ||
f40d1e42 | 608 | if (!PageBuddy(page)) |
2af120bc | 609 | goto isolate_fail; |
f40d1e42 MG |
610 | |
611 | /* | |
69b7189f VB |
612 | * If we already hold the lock, we can skip some rechecking. |
613 | * Note that if we hold the lock now, checked_pageblock was | |
614 | * already set in some previous iteration (or strict is true), | |
615 | * so it is correct to skip the suitable migration target | |
616 | * recheck as well. | |
f40d1e42 | 617 | */ |
69b7189f | 618 | if (!locked) { |
cb2dcaf0 | 619 | locked = compact_lock_irqsave(&cc->zone->lock, |
8b44d279 | 620 | &flags, cc); |
f40d1e42 | 621 | |
69b7189f VB |
622 | /* Recheck this is a buddy page under lock */ |
623 | if (!PageBuddy(page)) | |
624 | goto isolate_fail; | |
625 | } | |
748446bb | 626 | |
66c64223 | 627 | /* Found a free page, will break it into order-0 pages */ |
ab130f91 | 628 | order = buddy_order(page); |
66c64223 | 629 | isolated = __isolate_free_page(page, order); |
a4f04f2c DR |
630 | if (!isolated) |
631 | break; | |
66c64223 | 632 | set_page_private(page, order); |
a4f04f2c | 633 | |
748446bb | 634 | total_isolated += isolated; |
a4f04f2c | 635 | cc->nr_freepages += isolated; |
66c64223 JK |
636 | list_add_tail(&page->lru, freelist); |
637 | ||
a4f04f2c DR |
638 | if (!strict && cc->nr_migratepages <= cc->nr_freepages) { |
639 | blockpfn += isolated; | |
640 | break; | |
748446bb | 641 | } |
a4f04f2c DR |
642 | /* Advance to the end of split page */ |
643 | blockpfn += isolated - 1; | |
644 | cursor += isolated - 1; | |
645 | continue; | |
2af120bc LA |
646 | |
647 | isolate_fail: | |
648 | if (strict) | |
649 | break; | |
650 | else | |
651 | continue; | |
652 | ||
748446bb MG |
653 | } |
654 | ||
a4f04f2c DR |
655 | if (locked) |
656 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
657 | ||
9fcd6d2e VB |
658 | /* |
659 | * There is a tiny chance that we have read bogus compound_order(), | |
660 | * so be careful to not go outside of the pageblock. | |
661 | */ | |
662 | if (unlikely(blockpfn > end_pfn)) | |
663 | blockpfn = end_pfn; | |
664 | ||
e34d85f0 JK |
665 | trace_mm_compaction_isolate_freepages(*start_pfn, blockpfn, |
666 | nr_scanned, total_isolated); | |
667 | ||
e14c720e VB |
668 | /* Record how far we have got within the block */ |
669 | *start_pfn = blockpfn; | |
670 | ||
f40d1e42 MG |
671 | /* |
672 | * If strict isolation is requested by CMA then check that all the | |
673 | * pages requested were isolated. If there were any failures, 0 is | |
674 | * returned and CMA will fail. | |
675 | */ | |
2af120bc | 676 | if (strict && blockpfn < end_pfn) |
f40d1e42 MG |
677 | total_isolated = 0; |
678 | ||
7f354a54 | 679 | cc->total_free_scanned += nr_scanned; |
397487db | 680 | if (total_isolated) |
010fc29a | 681 | count_compact_events(COMPACTISOLATED, total_isolated); |
748446bb MG |
682 | return total_isolated; |
683 | } | |
684 | ||
85aa125f MN |
685 | /** |
686 | * isolate_freepages_range() - isolate free pages. | |
e8b098fc | 687 | * @cc: Compaction control structure. |
85aa125f MN |
688 | * @start_pfn: The first PFN to start isolating. |
689 | * @end_pfn: The one-past-last PFN. | |
690 | * | |
691 | * Non-free pages, invalid PFNs, or zone boundaries within the | |
692 | * [start_pfn, end_pfn) range are considered errors, cause function to | |
693 | * undo its actions and return zero. | |
694 | * | |
695 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
696 | * (which may be greater then end_pfn if end fell in a middle of | |
697 | * a free page). | |
698 | */ | |
ff9543fd | 699 | unsigned long |
bb13ffeb MG |
700 | isolate_freepages_range(struct compact_control *cc, |
701 | unsigned long start_pfn, unsigned long end_pfn) | |
85aa125f | 702 | { |
e1409c32 | 703 | unsigned long isolated, pfn, block_start_pfn, block_end_pfn; |
85aa125f MN |
704 | LIST_HEAD(freelist); |
705 | ||
7d49d886 | 706 | pfn = start_pfn; |
06b6640a | 707 | block_start_pfn = pageblock_start_pfn(pfn); |
e1409c32 JK |
708 | if (block_start_pfn < cc->zone->zone_start_pfn) |
709 | block_start_pfn = cc->zone->zone_start_pfn; | |
06b6640a | 710 | block_end_pfn = pageblock_end_pfn(pfn); |
7d49d886 VB |
711 | |
712 | for (; pfn < end_pfn; pfn += isolated, | |
e1409c32 | 713 | block_start_pfn = block_end_pfn, |
7d49d886 | 714 | block_end_pfn += pageblock_nr_pages) { |
e14c720e VB |
715 | /* Protect pfn from changing by isolate_freepages_block */ |
716 | unsigned long isolate_start_pfn = pfn; | |
85aa125f | 717 | |
85aa125f MN |
718 | block_end_pfn = min(block_end_pfn, end_pfn); |
719 | ||
58420016 JK |
720 | /* |
721 | * pfn could pass the block_end_pfn if isolated freepage | |
722 | * is more than pageblock order. In this case, we adjust | |
723 | * scanning range to right one. | |
724 | */ | |
725 | if (pfn >= block_end_pfn) { | |
06b6640a VB |
726 | block_start_pfn = pageblock_start_pfn(pfn); |
727 | block_end_pfn = pageblock_end_pfn(pfn); | |
58420016 JK |
728 | block_end_pfn = min(block_end_pfn, end_pfn); |
729 | } | |
730 | ||
e1409c32 JK |
731 | if (!pageblock_pfn_to_page(block_start_pfn, |
732 | block_end_pfn, cc->zone)) | |
7d49d886 VB |
733 | break; |
734 | ||
e14c720e | 735 | isolated = isolate_freepages_block(cc, &isolate_start_pfn, |
4fca9730 | 736 | block_end_pfn, &freelist, 0, true); |
85aa125f MN |
737 | |
738 | /* | |
739 | * In strict mode, isolate_freepages_block() returns 0 if | |
740 | * there are any holes in the block (ie. invalid PFNs or | |
741 | * non-free pages). | |
742 | */ | |
743 | if (!isolated) | |
744 | break; | |
745 | ||
746 | /* | |
747 | * If we managed to isolate pages, it is always (1 << n) * | |
748 | * pageblock_nr_pages for some non-negative n. (Max order | |
749 | * page may span two pageblocks). | |
750 | */ | |
751 | } | |
752 | ||
66c64223 | 753 | /* __isolate_free_page() does not map the pages */ |
4469ab98 | 754 | split_map_pages(&freelist); |
85aa125f MN |
755 | |
756 | if (pfn < end_pfn) { | |
757 | /* Loop terminated early, cleanup. */ | |
758 | release_freepages(&freelist); | |
759 | return 0; | |
760 | } | |
761 | ||
762 | /* We don't use freelists for anything. */ | |
763 | return pfn; | |
764 | } | |
765 | ||
748446bb | 766 | /* Similar to reclaim, but different enough that they don't share logic */ |
5f438eee | 767 | static bool too_many_isolated(pg_data_t *pgdat) |
748446bb | 768 | { |
bc693045 | 769 | unsigned long active, inactive, isolated; |
748446bb | 770 | |
5f438eee AR |
771 | inactive = node_page_state(pgdat, NR_INACTIVE_FILE) + |
772 | node_page_state(pgdat, NR_INACTIVE_ANON); | |
773 | active = node_page_state(pgdat, NR_ACTIVE_FILE) + | |
774 | node_page_state(pgdat, NR_ACTIVE_ANON); | |
775 | isolated = node_page_state(pgdat, NR_ISOLATED_FILE) + | |
776 | node_page_state(pgdat, NR_ISOLATED_ANON); | |
748446bb | 777 | |
bc693045 | 778 | return isolated > (inactive + active) / 2; |
748446bb MG |
779 | } |
780 | ||
2fe86e00 | 781 | /** |
edc2ca61 VB |
782 | * isolate_migratepages_block() - isolate all migrate-able pages within |
783 | * a single pageblock | |
2fe86e00 | 784 | * @cc: Compaction control structure. |
edc2ca61 VB |
785 | * @low_pfn: The first PFN to isolate |
786 | * @end_pfn: The one-past-the-last PFN to isolate, within same pageblock | |
787 | * @isolate_mode: Isolation mode to be used. | |
2fe86e00 MN |
788 | * |
789 | * Isolate all pages that can be migrated from the range specified by | |
edc2ca61 VB |
790 | * [low_pfn, end_pfn). The range is expected to be within same pageblock. |
791 | * Returns zero if there is a fatal signal pending, otherwise PFN of the | |
792 | * first page that was not scanned (which may be both less, equal to or more | |
793 | * than end_pfn). | |
2fe86e00 | 794 | * |
edc2ca61 VB |
795 | * The pages are isolated on cc->migratepages list (not required to be empty), |
796 | * and cc->nr_migratepages is updated accordingly. The cc->migrate_pfn field | |
797 | * is neither read nor updated. | |
748446bb | 798 | */ |
edc2ca61 VB |
799 | static unsigned long |
800 | isolate_migratepages_block(struct compact_control *cc, unsigned long low_pfn, | |
801 | unsigned long end_pfn, isolate_mode_t isolate_mode) | |
748446bb | 802 | { |
5f438eee | 803 | pg_data_t *pgdat = cc->zone->zone_pgdat; |
b7aba698 | 804 | unsigned long nr_scanned = 0, nr_isolated = 0; |
fa9add64 | 805 | struct lruvec *lruvec; |
b8b2d825 | 806 | unsigned long flags = 0; |
6168d0da | 807 | struct lruvec *locked = NULL; |
bb13ffeb | 808 | struct page *page = NULL, *valid_page = NULL; |
e34d85f0 | 809 | unsigned long start_pfn = low_pfn; |
fdd048e1 VB |
810 | bool skip_on_failure = false; |
811 | unsigned long next_skip_pfn = 0; | |
e380bebe | 812 | bool skip_updated = false; |
748446bb | 813 | |
748446bb MG |
814 | /* |
815 | * Ensure that there are not too many pages isolated from the LRU | |
816 | * list by either parallel reclaimers or compaction. If there are, | |
817 | * delay for some time until fewer pages are isolated | |
818 | */ | |
5f438eee | 819 | while (unlikely(too_many_isolated(pgdat))) { |
d20bdd57 ZY |
820 | /* stop isolation if there are still pages not migrated */ |
821 | if (cc->nr_migratepages) | |
822 | return 0; | |
823 | ||
f9e35b3b | 824 | /* async migration should just abort */ |
e0b9daeb | 825 | if (cc->mode == MIGRATE_ASYNC) |
2fe86e00 | 826 | return 0; |
f9e35b3b | 827 | |
748446bb MG |
828 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
829 | ||
830 | if (fatal_signal_pending(current)) | |
2fe86e00 | 831 | return 0; |
748446bb MG |
832 | } |
833 | ||
cf66f070 | 834 | cond_resched(); |
aeef4b83 | 835 | |
fdd048e1 VB |
836 | if (cc->direct_compaction && (cc->mode == MIGRATE_ASYNC)) { |
837 | skip_on_failure = true; | |
838 | next_skip_pfn = block_end_pfn(low_pfn, cc->order); | |
839 | } | |
840 | ||
748446bb | 841 | /* Time to isolate some pages for migration */ |
748446bb | 842 | for (; low_pfn < end_pfn; low_pfn++) { |
29c0dde8 | 843 | |
fdd048e1 VB |
844 | if (skip_on_failure && low_pfn >= next_skip_pfn) { |
845 | /* | |
846 | * We have isolated all migration candidates in the | |
847 | * previous order-aligned block, and did not skip it due | |
848 | * to failure. We should migrate the pages now and | |
849 | * hopefully succeed compaction. | |
850 | */ | |
851 | if (nr_isolated) | |
852 | break; | |
853 | ||
854 | /* | |
855 | * We failed to isolate in the previous order-aligned | |
856 | * block. Set the new boundary to the end of the | |
857 | * current block. Note we can't simply increase | |
858 | * next_skip_pfn by 1 << order, as low_pfn might have | |
859 | * been incremented by a higher number due to skipping | |
860 | * a compound or a high-order buddy page in the | |
861 | * previous loop iteration. | |
862 | */ | |
863 | next_skip_pfn = block_end_pfn(low_pfn, cc->order); | |
864 | } | |
865 | ||
8b44d279 VB |
866 | /* |
867 | * Periodically drop the lock (if held) regardless of its | |
670105a2 MG |
868 | * contention, to give chance to IRQs. Abort completely if |
869 | * a fatal signal is pending. | |
8b44d279 | 870 | */ |
6168d0da AS |
871 | if (!(low_pfn % SWAP_CLUSTER_MAX)) { |
872 | if (locked) { | |
873 | unlock_page_lruvec_irqrestore(locked, flags); | |
874 | locked = NULL; | |
875 | } | |
876 | ||
877 | if (fatal_signal_pending(current)) { | |
878 | cc->contended = true; | |
879 | ||
880 | low_pfn = 0; | |
881 | goto fatal_pending; | |
882 | } | |
883 | ||
884 | cond_resched(); | |
670105a2 | 885 | } |
c67fe375 | 886 | |
748446bb | 887 | if (!pfn_valid_within(low_pfn)) |
fdd048e1 | 888 | goto isolate_fail; |
b7aba698 | 889 | nr_scanned++; |
748446bb | 890 | |
748446bb | 891 | page = pfn_to_page(low_pfn); |
dc908600 | 892 | |
e380bebe MG |
893 | /* |
894 | * Check if the pageblock has already been marked skipped. | |
895 | * Only the aligned PFN is checked as the caller isolates | |
896 | * COMPACT_CLUSTER_MAX at a time so the second call must | |
897 | * not falsely conclude that the block should be skipped. | |
898 | */ | |
899 | if (!valid_page && IS_ALIGNED(low_pfn, pageblock_nr_pages)) { | |
900 | if (!cc->ignore_skip_hint && get_pageblock_skip(page)) { | |
901 | low_pfn = end_pfn; | |
9df41314 | 902 | page = NULL; |
e380bebe MG |
903 | goto isolate_abort; |
904 | } | |
bb13ffeb | 905 | valid_page = page; |
e380bebe | 906 | } |
bb13ffeb | 907 | |
6c14466c | 908 | /* |
99c0fd5e VB |
909 | * Skip if free. We read page order here without zone lock |
910 | * which is generally unsafe, but the race window is small and | |
911 | * the worst thing that can happen is that we skip some | |
912 | * potential isolation targets. | |
6c14466c | 913 | */ |
99c0fd5e | 914 | if (PageBuddy(page)) { |
ab130f91 | 915 | unsigned long freepage_order = buddy_order_unsafe(page); |
99c0fd5e VB |
916 | |
917 | /* | |
918 | * Without lock, we cannot be sure that what we got is | |
919 | * a valid page order. Consider only values in the | |
920 | * valid order range to prevent low_pfn overflow. | |
921 | */ | |
922 | if (freepage_order > 0 && freepage_order < MAX_ORDER) | |
923 | low_pfn += (1UL << freepage_order) - 1; | |
748446bb | 924 | continue; |
99c0fd5e | 925 | } |
748446bb | 926 | |
bc835011 | 927 | /* |
29c0dde8 | 928 | * Regardless of being on LRU, compound pages such as THP and |
1da2f328 RR |
929 | * hugetlbfs are not to be compacted unless we are attempting |
930 | * an allocation much larger than the huge page size (eg CMA). | |
931 | * We can potentially save a lot of iterations if we skip them | |
932 | * at once. The check is racy, but we can consider only valid | |
933 | * values and the only danger is skipping too much. | |
bc835011 | 934 | */ |
1da2f328 | 935 | if (PageCompound(page) && !cc->alloc_contig) { |
21dc7e02 | 936 | const unsigned int order = compound_order(page); |
edc2ca61 | 937 | |
d3c85bad | 938 | if (likely(order < MAX_ORDER)) |
21dc7e02 | 939 | low_pfn += (1UL << order) - 1; |
fdd048e1 | 940 | goto isolate_fail; |
2a1402aa MG |
941 | } |
942 | ||
bda807d4 MK |
943 | /* |
944 | * Check may be lockless but that's ok as we recheck later. | |
945 | * It's possible to migrate LRU and non-lru movable pages. | |
946 | * Skip any other type of page | |
947 | */ | |
948 | if (!PageLRU(page)) { | |
bda807d4 MK |
949 | /* |
950 | * __PageMovable can return false positive so we need | |
951 | * to verify it under page_lock. | |
952 | */ | |
953 | if (unlikely(__PageMovable(page)) && | |
954 | !PageIsolated(page)) { | |
955 | if (locked) { | |
6168d0da AS |
956 | unlock_page_lruvec_irqrestore(locked, flags); |
957 | locked = NULL; | |
bda807d4 MK |
958 | } |
959 | ||
9e5bcd61 | 960 | if (!isolate_movable_page(page, isolate_mode)) |
bda807d4 MK |
961 | goto isolate_success; |
962 | } | |
963 | ||
fdd048e1 | 964 | goto isolate_fail; |
bda807d4 | 965 | } |
29c0dde8 | 966 | |
119d6d59 DR |
967 | /* |
968 | * Migration will fail if an anonymous page is pinned in memory, | |
969 | * so avoid taking lru_lock and isolating it unnecessarily in an | |
970 | * admittedly racy check. | |
971 | */ | |
972 | if (!page_mapping(page) && | |
973 | page_count(page) > page_mapcount(page)) | |
fdd048e1 | 974 | goto isolate_fail; |
119d6d59 | 975 | |
73e64c51 MH |
976 | /* |
977 | * Only allow to migrate anonymous pages in GFP_NOFS context | |
978 | * because those do not depend on fs locks. | |
979 | */ | |
980 | if (!(cc->gfp_mask & __GFP_FS) && page_mapping(page)) | |
981 | goto isolate_fail; | |
982 | ||
9df41314 AS |
983 | /* |
984 | * Be careful not to clear PageLRU until after we're | |
985 | * sure the page is not being freed elsewhere -- the | |
986 | * page release code relies on it. | |
987 | */ | |
988 | if (unlikely(!get_page_unless_zero(page))) | |
989 | goto isolate_fail; | |
990 | ||
991 | if (__isolate_lru_page_prepare(page, isolate_mode) != 0) | |
992 | goto isolate_fail_put; | |
993 | ||
994 | /* Try isolate the page */ | |
995 | if (!TestClearPageLRU(page)) | |
996 | goto isolate_fail_put; | |
997 | ||
6168d0da AS |
998 | rcu_read_lock(); |
999 | lruvec = mem_cgroup_page_lruvec(page, pgdat); | |
1000 | ||
69b7189f | 1001 | /* If we already hold the lock, we can skip some rechecking */ |
6168d0da AS |
1002 | if (lruvec != locked) { |
1003 | if (locked) | |
1004 | unlock_page_lruvec_irqrestore(locked, flags); | |
1005 | ||
1006 | compact_lock_irqsave(&lruvec->lru_lock, &flags, cc); | |
1007 | locked = lruvec; | |
1008 | rcu_read_unlock(); | |
1009 | ||
1010 | lruvec_memcg_debug(lruvec, page); | |
e380bebe | 1011 | |
e380bebe MG |
1012 | /* Try get exclusive access under lock */ |
1013 | if (!skip_updated) { | |
1014 | skip_updated = true; | |
1015 | if (test_and_set_skip(cc, page, low_pfn)) | |
1016 | goto isolate_abort; | |
1017 | } | |
2a1402aa | 1018 | |
29c0dde8 VB |
1019 | /* |
1020 | * Page become compound since the non-locked check, | |
1021 | * and it's on LRU. It can only be a THP so the order | |
1022 | * is safe to read and it's 0 for tail pages. | |
1023 | */ | |
1da2f328 | 1024 | if (unlikely(PageCompound(page) && !cc->alloc_contig)) { |
d8c6546b | 1025 | low_pfn += compound_nr(page) - 1; |
9df41314 AS |
1026 | SetPageLRU(page); |
1027 | goto isolate_fail_put; | |
69b7189f | 1028 | } |
6168d0da AS |
1029 | } else |
1030 | rcu_read_unlock(); | |
fa9add64 | 1031 | |
1da2f328 RR |
1032 | /* The whole page is taken off the LRU; skip the tail pages. */ |
1033 | if (PageCompound(page)) | |
1034 | low_pfn += compound_nr(page) - 1; | |
bc835011 | 1035 | |
748446bb | 1036 | /* Successfully isolated */ |
fa9add64 | 1037 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
1da2f328 | 1038 | mod_node_page_state(page_pgdat(page), |
9de4f22a | 1039 | NR_ISOLATED_ANON + page_is_file_lru(page), |
6c357848 | 1040 | thp_nr_pages(page)); |
b6c75016 JK |
1041 | |
1042 | isolate_success: | |
fdd048e1 | 1043 | list_add(&page->lru, &cc->migratepages); |
38935861 ZY |
1044 | cc->nr_migratepages += compound_nr(page); |
1045 | nr_isolated += compound_nr(page); | |
748446bb | 1046 | |
804d3121 MG |
1047 | /* |
1048 | * Avoid isolating too much unless this block is being | |
cb2dcaf0 MG |
1049 | * rescanned (e.g. dirty/writeback pages, parallel allocation) |
1050 | * or a lock is contended. For contention, isolate quickly to | |
1051 | * potentially remove one source of contention. | |
804d3121 | 1052 | */ |
38935861 | 1053 | if (cc->nr_migratepages >= COMPACT_CLUSTER_MAX && |
cb2dcaf0 | 1054 | !cc->rescan && !cc->contended) { |
31b8384a | 1055 | ++low_pfn; |
748446bb | 1056 | break; |
31b8384a | 1057 | } |
fdd048e1 VB |
1058 | |
1059 | continue; | |
9df41314 AS |
1060 | |
1061 | isolate_fail_put: | |
1062 | /* Avoid potential deadlock in freeing page under lru_lock */ | |
1063 | if (locked) { | |
6168d0da AS |
1064 | unlock_page_lruvec_irqrestore(locked, flags); |
1065 | locked = NULL; | |
9df41314 AS |
1066 | } |
1067 | put_page(page); | |
1068 | ||
fdd048e1 VB |
1069 | isolate_fail: |
1070 | if (!skip_on_failure) | |
1071 | continue; | |
1072 | ||
1073 | /* | |
1074 | * We have isolated some pages, but then failed. Release them | |
1075 | * instead of migrating, as we cannot form the cc->order buddy | |
1076 | * page anyway. | |
1077 | */ | |
1078 | if (nr_isolated) { | |
1079 | if (locked) { | |
6168d0da AS |
1080 | unlock_page_lruvec_irqrestore(locked, flags); |
1081 | locked = NULL; | |
fdd048e1 | 1082 | } |
fdd048e1 VB |
1083 | putback_movable_pages(&cc->migratepages); |
1084 | cc->nr_migratepages = 0; | |
fdd048e1 VB |
1085 | nr_isolated = 0; |
1086 | } | |
1087 | ||
1088 | if (low_pfn < next_skip_pfn) { | |
1089 | low_pfn = next_skip_pfn - 1; | |
1090 | /* | |
1091 | * The check near the loop beginning would have updated | |
1092 | * next_skip_pfn too, but this is a bit simpler. | |
1093 | */ | |
1094 | next_skip_pfn += 1UL << cc->order; | |
1095 | } | |
748446bb MG |
1096 | } |
1097 | ||
99c0fd5e VB |
1098 | /* |
1099 | * The PageBuddy() check could have potentially brought us outside | |
1100 | * the range to be scanned. | |
1101 | */ | |
1102 | if (unlikely(low_pfn > end_pfn)) | |
1103 | low_pfn = end_pfn; | |
1104 | ||
9df41314 AS |
1105 | page = NULL; |
1106 | ||
e380bebe | 1107 | isolate_abort: |
c67fe375 | 1108 | if (locked) |
6168d0da | 1109 | unlock_page_lruvec_irqrestore(locked, flags); |
9df41314 AS |
1110 | if (page) { |
1111 | SetPageLRU(page); | |
1112 | put_page(page); | |
1113 | } | |
748446bb | 1114 | |
50b5b094 | 1115 | /* |
804d3121 MG |
1116 | * Updated the cached scanner pfn once the pageblock has been scanned |
1117 | * Pages will either be migrated in which case there is no point | |
1118 | * scanning in the near future or migration failed in which case the | |
1119 | * failure reason may persist. The block is marked for skipping if | |
1120 | * there were no pages isolated in the block or if the block is | |
1121 | * rescanned twice in a row. | |
50b5b094 | 1122 | */ |
804d3121 | 1123 | if (low_pfn == end_pfn && (!nr_isolated || cc->rescan)) { |
e380bebe MG |
1124 | if (valid_page && !skip_updated) |
1125 | set_pageblock_skip(valid_page); | |
1126 | update_cached_migrate(cc, low_pfn); | |
1127 | } | |
bb13ffeb | 1128 | |
e34d85f0 JK |
1129 | trace_mm_compaction_isolate_migratepages(start_pfn, low_pfn, |
1130 | nr_scanned, nr_isolated); | |
b7aba698 | 1131 | |
670105a2 | 1132 | fatal_pending: |
7f354a54 | 1133 | cc->total_migrate_scanned += nr_scanned; |
397487db | 1134 | if (nr_isolated) |
010fc29a | 1135 | count_compact_events(COMPACTISOLATED, nr_isolated); |
397487db | 1136 | |
2fe86e00 MN |
1137 | return low_pfn; |
1138 | } | |
1139 | ||
edc2ca61 VB |
1140 | /** |
1141 | * isolate_migratepages_range() - isolate migrate-able pages in a PFN range | |
1142 | * @cc: Compaction control structure. | |
1143 | * @start_pfn: The first PFN to start isolating. | |
1144 | * @end_pfn: The one-past-last PFN. | |
1145 | * | |
1146 | * Returns zero if isolation fails fatally due to e.g. pending signal. | |
1147 | * Otherwise, function returns one-past-the-last PFN of isolated page | |
1148 | * (which may be greater than end_pfn if end fell in a middle of a THP page). | |
1149 | */ | |
1150 | unsigned long | |
1151 | isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn, | |
1152 | unsigned long end_pfn) | |
1153 | { | |
e1409c32 | 1154 | unsigned long pfn, block_start_pfn, block_end_pfn; |
edc2ca61 VB |
1155 | |
1156 | /* Scan block by block. First and last block may be incomplete */ | |
1157 | pfn = start_pfn; | |
06b6640a | 1158 | block_start_pfn = pageblock_start_pfn(pfn); |
e1409c32 JK |
1159 | if (block_start_pfn < cc->zone->zone_start_pfn) |
1160 | block_start_pfn = cc->zone->zone_start_pfn; | |
06b6640a | 1161 | block_end_pfn = pageblock_end_pfn(pfn); |
edc2ca61 VB |
1162 | |
1163 | for (; pfn < end_pfn; pfn = block_end_pfn, | |
e1409c32 | 1164 | block_start_pfn = block_end_pfn, |
edc2ca61 VB |
1165 | block_end_pfn += pageblock_nr_pages) { |
1166 | ||
1167 | block_end_pfn = min(block_end_pfn, end_pfn); | |
1168 | ||
e1409c32 JK |
1169 | if (!pageblock_pfn_to_page(block_start_pfn, |
1170 | block_end_pfn, cc->zone)) | |
edc2ca61 VB |
1171 | continue; |
1172 | ||
1173 | pfn = isolate_migratepages_block(cc, pfn, block_end_pfn, | |
1174 | ISOLATE_UNEVICTABLE); | |
1175 | ||
14af4a5e | 1176 | if (!pfn) |
edc2ca61 | 1177 | break; |
6ea41c0c | 1178 | |
38935861 | 1179 | if (cc->nr_migratepages >= COMPACT_CLUSTER_MAX) |
6ea41c0c | 1180 | break; |
edc2ca61 | 1181 | } |
edc2ca61 VB |
1182 | |
1183 | return pfn; | |
1184 | } | |
1185 | ||
ff9543fd MN |
1186 | #endif /* CONFIG_COMPACTION || CONFIG_CMA */ |
1187 | #ifdef CONFIG_COMPACTION | |
018e9a49 | 1188 | |
b682debd VB |
1189 | static bool suitable_migration_source(struct compact_control *cc, |
1190 | struct page *page) | |
1191 | { | |
282722b0 VB |
1192 | int block_mt; |
1193 | ||
9bebefd5 MG |
1194 | if (pageblock_skip_persistent(page)) |
1195 | return false; | |
1196 | ||
282722b0 | 1197 | if ((cc->mode != MIGRATE_ASYNC) || !cc->direct_compaction) |
b682debd VB |
1198 | return true; |
1199 | ||
282722b0 VB |
1200 | block_mt = get_pageblock_migratetype(page); |
1201 | ||
1202 | if (cc->migratetype == MIGRATE_MOVABLE) | |
1203 | return is_migrate_movable(block_mt); | |
1204 | else | |
1205 | return block_mt == cc->migratetype; | |
b682debd VB |
1206 | } |
1207 | ||
018e9a49 | 1208 | /* Returns true if the page is within a block suitable for migration to */ |
9f7e3387 VB |
1209 | static bool suitable_migration_target(struct compact_control *cc, |
1210 | struct page *page) | |
018e9a49 AM |
1211 | { |
1212 | /* If the page is a large free page, then disallow migration */ | |
1213 | if (PageBuddy(page)) { | |
1214 | /* | |
1215 | * We are checking page_order without zone->lock taken. But | |
1216 | * the only small danger is that we skip a potentially suitable | |
1217 | * pageblock, so it's not worth to check order for valid range. | |
1218 | */ | |
ab130f91 | 1219 | if (buddy_order_unsafe(page) >= pageblock_order) |
018e9a49 AM |
1220 | return false; |
1221 | } | |
1222 | ||
1ef36db2 YX |
1223 | if (cc->ignore_block_suitable) |
1224 | return true; | |
1225 | ||
018e9a49 | 1226 | /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ |
b682debd | 1227 | if (is_migrate_movable(get_pageblock_migratetype(page))) |
018e9a49 AM |
1228 | return true; |
1229 | ||
1230 | /* Otherwise skip the block */ | |
1231 | return false; | |
1232 | } | |
1233 | ||
70b44595 MG |
1234 | static inline unsigned int |
1235 | freelist_scan_limit(struct compact_control *cc) | |
1236 | { | |
dd7ef7bd QC |
1237 | unsigned short shift = BITS_PER_LONG - 1; |
1238 | ||
1239 | return (COMPACT_CLUSTER_MAX >> min(shift, cc->fast_search_fail)) + 1; | |
70b44595 MG |
1240 | } |
1241 | ||
f2849aa0 VB |
1242 | /* |
1243 | * Test whether the free scanner has reached the same or lower pageblock than | |
1244 | * the migration scanner, and compaction should thus terminate. | |
1245 | */ | |
1246 | static inline bool compact_scanners_met(struct compact_control *cc) | |
1247 | { | |
1248 | return (cc->free_pfn >> pageblock_order) | |
1249 | <= (cc->migrate_pfn >> pageblock_order); | |
1250 | } | |
1251 | ||
5a811889 MG |
1252 | /* |
1253 | * Used when scanning for a suitable migration target which scans freelists | |
1254 | * in reverse. Reorders the list such as the unscanned pages are scanned | |
1255 | * first on the next iteration of the free scanner | |
1256 | */ | |
1257 | static void | |
1258 | move_freelist_head(struct list_head *freelist, struct page *freepage) | |
1259 | { | |
1260 | LIST_HEAD(sublist); | |
1261 | ||
1262 | if (!list_is_last(freelist, &freepage->lru)) { | |
1263 | list_cut_before(&sublist, freelist, &freepage->lru); | |
1264 | if (!list_empty(&sublist)) | |
1265 | list_splice_tail(&sublist, freelist); | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | /* | |
1270 | * Similar to move_freelist_head except used by the migration scanner | |
1271 | * when scanning forward. It's possible for these list operations to | |
1272 | * move against each other if they search the free list exactly in | |
1273 | * lockstep. | |
1274 | */ | |
70b44595 MG |
1275 | static void |
1276 | move_freelist_tail(struct list_head *freelist, struct page *freepage) | |
1277 | { | |
1278 | LIST_HEAD(sublist); | |
1279 | ||
1280 | if (!list_is_first(freelist, &freepage->lru)) { | |
1281 | list_cut_position(&sublist, freelist, &freepage->lru); | |
1282 | if (!list_empty(&sublist)) | |
1283 | list_splice_tail(&sublist, freelist); | |
1284 | } | |
1285 | } | |
1286 | ||
5a811889 MG |
1287 | static void |
1288 | fast_isolate_around(struct compact_control *cc, unsigned long pfn, unsigned long nr_isolated) | |
1289 | { | |
1290 | unsigned long start_pfn, end_pfn; | |
1291 | struct page *page = pfn_to_page(pfn); | |
1292 | ||
1293 | /* Do not search around if there are enough pages already */ | |
1294 | if (cc->nr_freepages >= cc->nr_migratepages) | |
1295 | return; | |
1296 | ||
1297 | /* Minimise scanning during async compaction */ | |
1298 | if (cc->direct_compaction && cc->mode == MIGRATE_ASYNC) | |
1299 | return; | |
1300 | ||
1301 | /* Pageblock boundaries */ | |
1302 | start_pfn = pageblock_start_pfn(pfn); | |
60fce36a | 1303 | end_pfn = min(pageblock_end_pfn(pfn), zone_end_pfn(cc->zone)) - 1; |
5a811889 MG |
1304 | |
1305 | /* Scan before */ | |
1306 | if (start_pfn != pfn) { | |
4fca9730 | 1307 | isolate_freepages_block(cc, &start_pfn, pfn, &cc->freepages, 1, false); |
5a811889 MG |
1308 | if (cc->nr_freepages >= cc->nr_migratepages) |
1309 | return; | |
1310 | } | |
1311 | ||
1312 | /* Scan after */ | |
1313 | start_pfn = pfn + nr_isolated; | |
60fce36a | 1314 | if (start_pfn < end_pfn) |
4fca9730 | 1315 | isolate_freepages_block(cc, &start_pfn, end_pfn, &cc->freepages, 1, false); |
5a811889 MG |
1316 | |
1317 | /* Skip this pageblock in the future as it's full or nearly full */ | |
1318 | if (cc->nr_freepages < cc->nr_migratepages) | |
1319 | set_pageblock_skip(page); | |
1320 | } | |
1321 | ||
dbe2d4e4 MG |
1322 | /* Search orders in round-robin fashion */ |
1323 | static int next_search_order(struct compact_control *cc, int order) | |
1324 | { | |
1325 | order--; | |
1326 | if (order < 0) | |
1327 | order = cc->order - 1; | |
1328 | ||
1329 | /* Search wrapped around? */ | |
1330 | if (order == cc->search_order) { | |
1331 | cc->search_order--; | |
1332 | if (cc->search_order < 0) | |
1333 | cc->search_order = cc->order - 1; | |
1334 | return -1; | |
1335 | } | |
1336 | ||
1337 | return order; | |
1338 | } | |
1339 | ||
5a811889 MG |
1340 | static unsigned long |
1341 | fast_isolate_freepages(struct compact_control *cc) | |
1342 | { | |
1343 | unsigned int limit = min(1U, freelist_scan_limit(cc) >> 1); | |
1344 | unsigned int nr_scanned = 0; | |
1345 | unsigned long low_pfn, min_pfn, high_pfn = 0, highest = 0; | |
1346 | unsigned long nr_isolated = 0; | |
1347 | unsigned long distance; | |
1348 | struct page *page = NULL; | |
1349 | bool scan_start = false; | |
1350 | int order; | |
1351 | ||
1352 | /* Full compaction passes in a negative order */ | |
1353 | if (cc->order <= 0) | |
1354 | return cc->free_pfn; | |
1355 | ||
1356 | /* | |
1357 | * If starting the scan, use a deeper search and use the highest | |
1358 | * PFN found if a suitable one is not found. | |
1359 | */ | |
e332f741 | 1360 | if (cc->free_pfn >= cc->zone->compact_init_free_pfn) { |
5a811889 MG |
1361 | limit = pageblock_nr_pages >> 1; |
1362 | scan_start = true; | |
1363 | } | |
1364 | ||
1365 | /* | |
1366 | * Preferred point is in the top quarter of the scan space but take | |
1367 | * a pfn from the top half if the search is problematic. | |
1368 | */ | |
1369 | distance = (cc->free_pfn - cc->migrate_pfn); | |
1370 | low_pfn = pageblock_start_pfn(cc->free_pfn - (distance >> 2)); | |
1371 | min_pfn = pageblock_start_pfn(cc->free_pfn - (distance >> 1)); | |
1372 | ||
1373 | if (WARN_ON_ONCE(min_pfn > low_pfn)) | |
1374 | low_pfn = min_pfn; | |
1375 | ||
dbe2d4e4 MG |
1376 | /* |
1377 | * Search starts from the last successful isolation order or the next | |
1378 | * order to search after a previous failure | |
1379 | */ | |
1380 | cc->search_order = min_t(unsigned int, cc->order - 1, cc->search_order); | |
1381 | ||
1382 | for (order = cc->search_order; | |
1383 | !page && order >= 0; | |
1384 | order = next_search_order(cc, order)) { | |
5a811889 MG |
1385 | struct free_area *area = &cc->zone->free_area[order]; |
1386 | struct list_head *freelist; | |
1387 | struct page *freepage; | |
1388 | unsigned long flags; | |
1389 | unsigned int order_scanned = 0; | |
1390 | ||
1391 | if (!area->nr_free) | |
1392 | continue; | |
1393 | ||
1394 | spin_lock_irqsave(&cc->zone->lock, flags); | |
1395 | freelist = &area->free_list[MIGRATE_MOVABLE]; | |
1396 | list_for_each_entry_reverse(freepage, freelist, lru) { | |
1397 | unsigned long pfn; | |
1398 | ||
1399 | order_scanned++; | |
1400 | nr_scanned++; | |
1401 | pfn = page_to_pfn(freepage); | |
1402 | ||
1403 | if (pfn >= highest) | |
1404 | highest = pageblock_start_pfn(pfn); | |
1405 | ||
1406 | if (pfn >= low_pfn) { | |
1407 | cc->fast_search_fail = 0; | |
dbe2d4e4 | 1408 | cc->search_order = order; |
5a811889 MG |
1409 | page = freepage; |
1410 | break; | |
1411 | } | |
1412 | ||
1413 | if (pfn >= min_pfn && pfn > high_pfn) { | |
1414 | high_pfn = pfn; | |
1415 | ||
1416 | /* Shorten the scan if a candidate is found */ | |
1417 | limit >>= 1; | |
1418 | } | |
1419 | ||
1420 | if (order_scanned >= limit) | |
1421 | break; | |
1422 | } | |
1423 | ||
1424 | /* Use a minimum pfn if a preferred one was not found */ | |
1425 | if (!page && high_pfn) { | |
1426 | page = pfn_to_page(high_pfn); | |
1427 | ||
1428 | /* Update freepage for the list reorder below */ | |
1429 | freepage = page; | |
1430 | } | |
1431 | ||
1432 | /* Reorder to so a future search skips recent pages */ | |
1433 | move_freelist_head(freelist, freepage); | |
1434 | ||
1435 | /* Isolate the page if available */ | |
1436 | if (page) { | |
1437 | if (__isolate_free_page(page, order)) { | |
1438 | set_page_private(page, order); | |
1439 | nr_isolated = 1 << order; | |
1440 | cc->nr_freepages += nr_isolated; | |
1441 | list_add_tail(&page->lru, &cc->freepages); | |
1442 | count_compact_events(COMPACTISOLATED, nr_isolated); | |
1443 | } else { | |
1444 | /* If isolation fails, abort the search */ | |
5b56d996 | 1445 | order = cc->search_order + 1; |
5a811889 MG |
1446 | page = NULL; |
1447 | } | |
1448 | } | |
1449 | ||
1450 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
1451 | ||
1452 | /* | |
1453 | * Smaller scan on next order so the total scan ig related | |
1454 | * to freelist_scan_limit. | |
1455 | */ | |
1456 | if (order_scanned >= limit) | |
1457 | limit = min(1U, limit >> 1); | |
1458 | } | |
1459 | ||
1460 | if (!page) { | |
1461 | cc->fast_search_fail++; | |
1462 | if (scan_start) { | |
1463 | /* | |
1464 | * Use the highest PFN found above min. If one was | |
f3867755 | 1465 | * not found, be pessimistic for direct compaction |
5a811889 MG |
1466 | * and use the min mark. |
1467 | */ | |
1468 | if (highest) { | |
1469 | page = pfn_to_page(highest); | |
1470 | cc->free_pfn = highest; | |
1471 | } else { | |
e577c8b6 | 1472 | if (cc->direct_compaction && pfn_valid(min_pfn)) { |
73a6e474 BH |
1473 | page = pageblock_pfn_to_page(min_pfn, |
1474 | pageblock_end_pfn(min_pfn), | |
1475 | cc->zone); | |
5a811889 MG |
1476 | cc->free_pfn = min_pfn; |
1477 | } | |
1478 | } | |
1479 | } | |
1480 | } | |
1481 | ||
d097a6f6 MG |
1482 | if (highest && highest >= cc->zone->compact_cached_free_pfn) { |
1483 | highest -= pageblock_nr_pages; | |
5a811889 | 1484 | cc->zone->compact_cached_free_pfn = highest; |
d097a6f6 | 1485 | } |
5a811889 MG |
1486 | |
1487 | cc->total_free_scanned += nr_scanned; | |
1488 | if (!page) | |
1489 | return cc->free_pfn; | |
1490 | ||
1491 | low_pfn = page_to_pfn(page); | |
1492 | fast_isolate_around(cc, low_pfn, nr_isolated); | |
1493 | return low_pfn; | |
1494 | } | |
1495 | ||
2fe86e00 | 1496 | /* |
ff9543fd MN |
1497 | * Based on information in the current compact_control, find blocks |
1498 | * suitable for isolating free pages from and then isolate them. | |
2fe86e00 | 1499 | */ |
edc2ca61 | 1500 | static void isolate_freepages(struct compact_control *cc) |
2fe86e00 | 1501 | { |
edc2ca61 | 1502 | struct zone *zone = cc->zone; |
ff9543fd | 1503 | struct page *page; |
c96b9e50 | 1504 | unsigned long block_start_pfn; /* start of current pageblock */ |
e14c720e | 1505 | unsigned long isolate_start_pfn; /* exact pfn we start at */ |
c96b9e50 VB |
1506 | unsigned long block_end_pfn; /* end of current pageblock */ |
1507 | unsigned long low_pfn; /* lowest pfn scanner is able to scan */ | |
ff9543fd | 1508 | struct list_head *freelist = &cc->freepages; |
4fca9730 | 1509 | unsigned int stride; |
2fe86e00 | 1510 | |
5a811889 MG |
1511 | /* Try a small search of the free lists for a candidate */ |
1512 | isolate_start_pfn = fast_isolate_freepages(cc); | |
1513 | if (cc->nr_freepages) | |
1514 | goto splitmap; | |
1515 | ||
ff9543fd MN |
1516 | /* |
1517 | * Initialise the free scanner. The starting point is where we last | |
49e068f0 | 1518 | * successfully isolated from, zone-cached value, or the end of the |
e14c720e VB |
1519 | * zone when isolating for the first time. For looping we also need |
1520 | * this pfn aligned down to the pageblock boundary, because we do | |
c96b9e50 VB |
1521 | * block_start_pfn -= pageblock_nr_pages in the for loop. |
1522 | * For ending point, take care when isolating in last pageblock of a | |
a1c1dbeb | 1523 | * zone which ends in the middle of a pageblock. |
49e068f0 VB |
1524 | * The low boundary is the end of the pageblock the migration scanner |
1525 | * is using. | |
ff9543fd | 1526 | */ |
e14c720e | 1527 | isolate_start_pfn = cc->free_pfn; |
5a811889 | 1528 | block_start_pfn = pageblock_start_pfn(isolate_start_pfn); |
c96b9e50 VB |
1529 | block_end_pfn = min(block_start_pfn + pageblock_nr_pages, |
1530 | zone_end_pfn(zone)); | |
06b6640a | 1531 | low_pfn = pageblock_end_pfn(cc->migrate_pfn); |
4fca9730 | 1532 | stride = cc->mode == MIGRATE_ASYNC ? COMPACT_CLUSTER_MAX : 1; |
2fe86e00 | 1533 | |
ff9543fd MN |
1534 | /* |
1535 | * Isolate free pages until enough are available to migrate the | |
1536 | * pages on cc->migratepages. We stop searching if the migrate | |
1537 | * and free page scanners meet or enough free pages are isolated. | |
1538 | */ | |
f5f61a32 | 1539 | for (; block_start_pfn >= low_pfn; |
c96b9e50 | 1540 | block_end_pfn = block_start_pfn, |
e14c720e VB |
1541 | block_start_pfn -= pageblock_nr_pages, |
1542 | isolate_start_pfn = block_start_pfn) { | |
4fca9730 MG |
1543 | unsigned long nr_isolated; |
1544 | ||
f6ea3adb DR |
1545 | /* |
1546 | * This can iterate a massively long zone without finding any | |
cb810ad2 | 1547 | * suitable migration targets, so periodically check resched. |
f6ea3adb | 1548 | */ |
cb810ad2 | 1549 | if (!(block_start_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages))) |
cf66f070 | 1550 | cond_resched(); |
f6ea3adb | 1551 | |
7d49d886 VB |
1552 | page = pageblock_pfn_to_page(block_start_pfn, block_end_pfn, |
1553 | zone); | |
1554 | if (!page) | |
ff9543fd MN |
1555 | continue; |
1556 | ||
1557 | /* Check the block is suitable for migration */ | |
9f7e3387 | 1558 | if (!suitable_migration_target(cc, page)) |
ff9543fd | 1559 | continue; |
68e3e926 | 1560 | |
bb13ffeb MG |
1561 | /* If isolation recently failed, do not retry */ |
1562 | if (!isolation_suitable(cc, page)) | |
1563 | continue; | |
1564 | ||
e14c720e | 1565 | /* Found a block suitable for isolating free pages from. */ |
4fca9730 MG |
1566 | nr_isolated = isolate_freepages_block(cc, &isolate_start_pfn, |
1567 | block_end_pfn, freelist, stride, false); | |
ff9543fd | 1568 | |
d097a6f6 MG |
1569 | /* Update the skip hint if the full pageblock was scanned */ |
1570 | if (isolate_start_pfn == block_end_pfn) | |
1571 | update_pageblock_skip(cc, page, block_start_pfn); | |
1572 | ||
cb2dcaf0 MG |
1573 | /* Are enough freepages isolated? */ |
1574 | if (cc->nr_freepages >= cc->nr_migratepages) { | |
a46cbf3b DR |
1575 | if (isolate_start_pfn >= block_end_pfn) { |
1576 | /* | |
1577 | * Restart at previous pageblock if more | |
1578 | * freepages can be isolated next time. | |
1579 | */ | |
f5f61a32 VB |
1580 | isolate_start_pfn = |
1581 | block_start_pfn - pageblock_nr_pages; | |
a46cbf3b | 1582 | } |
be976572 | 1583 | break; |
a46cbf3b | 1584 | } else if (isolate_start_pfn < block_end_pfn) { |
f5f61a32 | 1585 | /* |
a46cbf3b DR |
1586 | * If isolation failed early, do not continue |
1587 | * needlessly. | |
f5f61a32 | 1588 | */ |
a46cbf3b | 1589 | break; |
f5f61a32 | 1590 | } |
4fca9730 MG |
1591 | |
1592 | /* Adjust stride depending on isolation */ | |
1593 | if (nr_isolated) { | |
1594 | stride = 1; | |
1595 | continue; | |
1596 | } | |
1597 | stride = min_t(unsigned int, COMPACT_CLUSTER_MAX, stride << 1); | |
ff9543fd MN |
1598 | } |
1599 | ||
7ed695e0 | 1600 | /* |
f5f61a32 VB |
1601 | * Record where the free scanner will restart next time. Either we |
1602 | * broke from the loop and set isolate_start_pfn based on the last | |
1603 | * call to isolate_freepages_block(), or we met the migration scanner | |
1604 | * and the loop terminated due to isolate_start_pfn < low_pfn | |
7ed695e0 | 1605 | */ |
f5f61a32 | 1606 | cc->free_pfn = isolate_start_pfn; |
5a811889 MG |
1607 | |
1608 | splitmap: | |
1609 | /* __isolate_free_page() does not map the pages */ | |
1610 | split_map_pages(freelist); | |
748446bb MG |
1611 | } |
1612 | ||
1613 | /* | |
1614 | * This is a migrate-callback that "allocates" freepages by taking pages | |
1615 | * from the isolated freelists in the block we are migrating to. | |
1616 | */ | |
1617 | static struct page *compaction_alloc(struct page *migratepage, | |
666feb21 | 1618 | unsigned long data) |
748446bb MG |
1619 | { |
1620 | struct compact_control *cc = (struct compact_control *)data; | |
1621 | struct page *freepage; | |
1622 | ||
748446bb | 1623 | if (list_empty(&cc->freepages)) { |
cb2dcaf0 | 1624 | isolate_freepages(cc); |
748446bb MG |
1625 | |
1626 | if (list_empty(&cc->freepages)) | |
1627 | return NULL; | |
1628 | } | |
1629 | ||
1630 | freepage = list_entry(cc->freepages.next, struct page, lru); | |
1631 | list_del(&freepage->lru); | |
1632 | cc->nr_freepages--; | |
1633 | ||
1634 | return freepage; | |
1635 | } | |
1636 | ||
1637 | /* | |
d53aea3d DR |
1638 | * This is a migrate-callback that "frees" freepages back to the isolated |
1639 | * freelist. All pages on the freelist are from the same zone, so there is no | |
1640 | * special handling needed for NUMA. | |
1641 | */ | |
1642 | static void compaction_free(struct page *page, unsigned long data) | |
1643 | { | |
1644 | struct compact_control *cc = (struct compact_control *)data; | |
1645 | ||
1646 | list_add(&page->lru, &cc->freepages); | |
1647 | cc->nr_freepages++; | |
1648 | } | |
1649 | ||
ff9543fd MN |
1650 | /* possible outcome of isolate_migratepages */ |
1651 | typedef enum { | |
1652 | ISOLATE_ABORT, /* Abort compaction now */ | |
1653 | ISOLATE_NONE, /* No pages isolated, continue scanning */ | |
1654 | ISOLATE_SUCCESS, /* Pages isolated, migrate */ | |
1655 | } isolate_migrate_t; | |
1656 | ||
5bbe3547 EM |
1657 | /* |
1658 | * Allow userspace to control policy on scanning the unevictable LRU for | |
1659 | * compactable pages. | |
1660 | */ | |
6923aa0d SAS |
1661 | #ifdef CONFIG_PREEMPT_RT |
1662 | int sysctl_compact_unevictable_allowed __read_mostly = 0; | |
1663 | #else | |
5bbe3547 | 1664 | int sysctl_compact_unevictable_allowed __read_mostly = 1; |
6923aa0d | 1665 | #endif |
5bbe3547 | 1666 | |
70b44595 MG |
1667 | static inline void |
1668 | update_fast_start_pfn(struct compact_control *cc, unsigned long pfn) | |
1669 | { | |
1670 | if (cc->fast_start_pfn == ULONG_MAX) | |
1671 | return; | |
1672 | ||
1673 | if (!cc->fast_start_pfn) | |
1674 | cc->fast_start_pfn = pfn; | |
1675 | ||
1676 | cc->fast_start_pfn = min(cc->fast_start_pfn, pfn); | |
1677 | } | |
1678 | ||
1679 | static inline unsigned long | |
1680 | reinit_migrate_pfn(struct compact_control *cc) | |
1681 | { | |
1682 | if (!cc->fast_start_pfn || cc->fast_start_pfn == ULONG_MAX) | |
1683 | return cc->migrate_pfn; | |
1684 | ||
1685 | cc->migrate_pfn = cc->fast_start_pfn; | |
1686 | cc->fast_start_pfn = ULONG_MAX; | |
1687 | ||
1688 | return cc->migrate_pfn; | |
1689 | } | |
1690 | ||
1691 | /* | |
1692 | * Briefly search the free lists for a migration source that already has | |
1693 | * some free pages to reduce the number of pages that need migration | |
1694 | * before a pageblock is free. | |
1695 | */ | |
1696 | static unsigned long fast_find_migrateblock(struct compact_control *cc) | |
1697 | { | |
1698 | unsigned int limit = freelist_scan_limit(cc); | |
1699 | unsigned int nr_scanned = 0; | |
1700 | unsigned long distance; | |
1701 | unsigned long pfn = cc->migrate_pfn; | |
1702 | unsigned long high_pfn; | |
1703 | int order; | |
1704 | ||
1705 | /* Skip hints are relied on to avoid repeats on the fast search */ | |
1706 | if (cc->ignore_skip_hint) | |
1707 | return pfn; | |
1708 | ||
1709 | /* | |
1710 | * If the migrate_pfn is not at the start of a zone or the start | |
1711 | * of a pageblock then assume this is a continuation of a previous | |
1712 | * scan restarted due to COMPACT_CLUSTER_MAX. | |
1713 | */ | |
1714 | if (pfn != cc->zone->zone_start_pfn && pfn != pageblock_start_pfn(pfn)) | |
1715 | return pfn; | |
1716 | ||
1717 | /* | |
1718 | * For smaller orders, just linearly scan as the number of pages | |
1719 | * to migrate should be relatively small and does not necessarily | |
1720 | * justify freeing up a large block for a small allocation. | |
1721 | */ | |
1722 | if (cc->order <= PAGE_ALLOC_COSTLY_ORDER) | |
1723 | return pfn; | |
1724 | ||
1725 | /* | |
1726 | * Only allow kcompactd and direct requests for movable pages to | |
1727 | * quickly clear out a MOVABLE pageblock for allocation. This | |
1728 | * reduces the risk that a large movable pageblock is freed for | |
1729 | * an unmovable/reclaimable small allocation. | |
1730 | */ | |
1731 | if (cc->direct_compaction && cc->migratetype != MIGRATE_MOVABLE) | |
1732 | return pfn; | |
1733 | ||
1734 | /* | |
1735 | * When starting the migration scanner, pick any pageblock within the | |
1736 | * first half of the search space. Otherwise try and pick a pageblock | |
1737 | * within the first eighth to reduce the chances that a migration | |
1738 | * target later becomes a source. | |
1739 | */ | |
1740 | distance = (cc->free_pfn - cc->migrate_pfn) >> 1; | |
1741 | if (cc->migrate_pfn != cc->zone->zone_start_pfn) | |
1742 | distance >>= 2; | |
1743 | high_pfn = pageblock_start_pfn(cc->migrate_pfn + distance); | |
1744 | ||
1745 | for (order = cc->order - 1; | |
1746 | order >= PAGE_ALLOC_COSTLY_ORDER && pfn == cc->migrate_pfn && nr_scanned < limit; | |
1747 | order--) { | |
1748 | struct free_area *area = &cc->zone->free_area[order]; | |
1749 | struct list_head *freelist; | |
1750 | unsigned long flags; | |
1751 | struct page *freepage; | |
1752 | ||
1753 | if (!area->nr_free) | |
1754 | continue; | |
1755 | ||
1756 | spin_lock_irqsave(&cc->zone->lock, flags); | |
1757 | freelist = &area->free_list[MIGRATE_MOVABLE]; | |
1758 | list_for_each_entry(freepage, freelist, lru) { | |
1759 | unsigned long free_pfn; | |
1760 | ||
1761 | nr_scanned++; | |
1762 | free_pfn = page_to_pfn(freepage); | |
1763 | if (free_pfn < high_pfn) { | |
70b44595 MG |
1764 | /* |
1765 | * Avoid if skipped recently. Ideally it would | |
1766 | * move to the tail but even safe iteration of | |
1767 | * the list assumes an entry is deleted, not | |
1768 | * reordered. | |
1769 | */ | |
1770 | if (get_pageblock_skip(freepage)) { | |
1771 | if (list_is_last(freelist, &freepage->lru)) | |
1772 | break; | |
1773 | ||
1774 | continue; | |
1775 | } | |
1776 | ||
1777 | /* Reorder to so a future search skips recent pages */ | |
1778 | move_freelist_tail(freelist, freepage); | |
1779 | ||
e380bebe | 1780 | update_fast_start_pfn(cc, free_pfn); |
70b44595 MG |
1781 | pfn = pageblock_start_pfn(free_pfn); |
1782 | cc->fast_search_fail = 0; | |
1783 | set_pageblock_skip(freepage); | |
1784 | break; | |
1785 | } | |
1786 | ||
1787 | if (nr_scanned >= limit) { | |
1788 | cc->fast_search_fail++; | |
1789 | move_freelist_tail(freelist, freepage); | |
1790 | break; | |
1791 | } | |
1792 | } | |
1793 | spin_unlock_irqrestore(&cc->zone->lock, flags); | |
1794 | } | |
1795 | ||
1796 | cc->total_migrate_scanned += nr_scanned; | |
1797 | ||
1798 | /* | |
1799 | * If fast scanning failed then use a cached entry for a page block | |
1800 | * that had free pages as the basis for starting a linear scan. | |
1801 | */ | |
1802 | if (pfn == cc->migrate_pfn) | |
1803 | pfn = reinit_migrate_pfn(cc); | |
1804 | ||
1805 | return pfn; | |
1806 | } | |
1807 | ||
ff9543fd | 1808 | /* |
edc2ca61 VB |
1809 | * Isolate all pages that can be migrated from the first suitable block, |
1810 | * starting at the block pointed to by the migrate scanner pfn within | |
1811 | * compact_control. | |
ff9543fd | 1812 | */ |
32aaf055 | 1813 | static isolate_migrate_t isolate_migratepages(struct compact_control *cc) |
ff9543fd | 1814 | { |
e1409c32 JK |
1815 | unsigned long block_start_pfn; |
1816 | unsigned long block_end_pfn; | |
1817 | unsigned long low_pfn; | |
edc2ca61 VB |
1818 | struct page *page; |
1819 | const isolate_mode_t isolate_mode = | |
5bbe3547 | 1820 | (sysctl_compact_unevictable_allowed ? ISOLATE_UNEVICTABLE : 0) | |
1d2047fe | 1821 | (cc->mode != MIGRATE_SYNC ? ISOLATE_ASYNC_MIGRATE : 0); |
70b44595 | 1822 | bool fast_find_block; |
ff9543fd | 1823 | |
edc2ca61 VB |
1824 | /* |
1825 | * Start at where we last stopped, or beginning of the zone as | |
70b44595 MG |
1826 | * initialized by compact_zone(). The first failure will use |
1827 | * the lowest PFN as the starting point for linear scanning. | |
edc2ca61 | 1828 | */ |
70b44595 | 1829 | low_pfn = fast_find_migrateblock(cc); |
06b6640a | 1830 | block_start_pfn = pageblock_start_pfn(low_pfn); |
32aaf055 PL |
1831 | if (block_start_pfn < cc->zone->zone_start_pfn) |
1832 | block_start_pfn = cc->zone->zone_start_pfn; | |
ff9543fd | 1833 | |
70b44595 MG |
1834 | /* |
1835 | * fast_find_migrateblock marks a pageblock skipped so to avoid | |
1836 | * the isolation_suitable check below, check whether the fast | |
1837 | * search was successful. | |
1838 | */ | |
1839 | fast_find_block = low_pfn != cc->migrate_pfn && !cc->fast_search_fail; | |
1840 | ||
ff9543fd | 1841 | /* Only scan within a pageblock boundary */ |
06b6640a | 1842 | block_end_pfn = pageblock_end_pfn(low_pfn); |
ff9543fd | 1843 | |
edc2ca61 VB |
1844 | /* |
1845 | * Iterate over whole pageblocks until we find the first suitable. | |
1846 | * Do not cross the free scanner. | |
1847 | */ | |
e1409c32 | 1848 | for (; block_end_pfn <= cc->free_pfn; |
70b44595 | 1849 | fast_find_block = false, |
e1409c32 JK |
1850 | low_pfn = block_end_pfn, |
1851 | block_start_pfn = block_end_pfn, | |
1852 | block_end_pfn += pageblock_nr_pages) { | |
ff9543fd | 1853 | |
edc2ca61 VB |
1854 | /* |
1855 | * This can potentially iterate a massively long zone with | |
1856 | * many pageblocks unsuitable, so periodically check if we | |
cb810ad2 | 1857 | * need to schedule. |
edc2ca61 | 1858 | */ |
cb810ad2 | 1859 | if (!(low_pfn % (SWAP_CLUSTER_MAX * pageblock_nr_pages))) |
cf66f070 | 1860 | cond_resched(); |
ff9543fd | 1861 | |
32aaf055 PL |
1862 | page = pageblock_pfn_to_page(block_start_pfn, |
1863 | block_end_pfn, cc->zone); | |
7d49d886 | 1864 | if (!page) |
edc2ca61 VB |
1865 | continue; |
1866 | ||
e380bebe MG |
1867 | /* |
1868 | * If isolation recently failed, do not retry. Only check the | |
1869 | * pageblock once. COMPACT_CLUSTER_MAX causes a pageblock | |
1870 | * to be visited multiple times. Assume skip was checked | |
1871 | * before making it "skip" so other compaction instances do | |
1872 | * not scan the same block. | |
1873 | */ | |
1874 | if (IS_ALIGNED(low_pfn, pageblock_nr_pages) && | |
1875 | !fast_find_block && !isolation_suitable(cc, page)) | |
edc2ca61 VB |
1876 | continue; |
1877 | ||
1878 | /* | |
9bebefd5 MG |
1879 | * For async compaction, also only scan in MOVABLE blocks |
1880 | * without huge pages. Async compaction is optimistic to see | |
1881 | * if the minimum amount of work satisfies the allocation. | |
1882 | * The cached PFN is updated as it's possible that all | |
1883 | * remaining blocks between source and target are unsuitable | |
1884 | * and the compaction scanners fail to meet. | |
edc2ca61 | 1885 | */ |
9bebefd5 MG |
1886 | if (!suitable_migration_source(cc, page)) { |
1887 | update_cached_migrate(cc, block_end_pfn); | |
edc2ca61 | 1888 | continue; |
9bebefd5 | 1889 | } |
edc2ca61 VB |
1890 | |
1891 | /* Perform the isolation */ | |
e1409c32 JK |
1892 | low_pfn = isolate_migratepages_block(cc, low_pfn, |
1893 | block_end_pfn, isolate_mode); | |
edc2ca61 | 1894 | |
cb2dcaf0 | 1895 | if (!low_pfn) |
edc2ca61 VB |
1896 | return ISOLATE_ABORT; |
1897 | ||
1898 | /* | |
1899 | * Either we isolated something and proceed with migration. Or | |
1900 | * we failed and compact_zone should decide if we should | |
1901 | * continue or not. | |
1902 | */ | |
1903 | break; | |
1904 | } | |
1905 | ||
f2849aa0 VB |
1906 | /* Record where migration scanner will be restarted. */ |
1907 | cc->migrate_pfn = low_pfn; | |
ff9543fd | 1908 | |
edc2ca61 | 1909 | return cc->nr_migratepages ? ISOLATE_SUCCESS : ISOLATE_NONE; |
ff9543fd MN |
1910 | } |
1911 | ||
21c527a3 YB |
1912 | /* |
1913 | * order == -1 is expected when compacting via | |
1914 | * /proc/sys/vm/compact_memory | |
1915 | */ | |
1916 | static inline bool is_via_compact_memory(int order) | |
1917 | { | |
1918 | return order == -1; | |
1919 | } | |
1920 | ||
facdaa91 NG |
1921 | static bool kswapd_is_running(pg_data_t *pgdat) |
1922 | { | |
1923 | return pgdat->kswapd && (pgdat->kswapd->state == TASK_RUNNING); | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * A zone's fragmentation score is the external fragmentation wrt to the | |
1928 | * COMPACTION_HPAGE_ORDER scaled by the zone's size. It returns a value | |
1929 | * in the range [0, 100]. | |
1930 | * | |
1931 | * The scaling factor ensures that proactive compaction focuses on larger | |
1932 | * zones like ZONE_NORMAL, rather than smaller, specialized zones like | |
1933 | * ZONE_DMA32. For smaller zones, the score value remains close to zero, | |
1934 | * and thus never exceeds the high threshold for proactive compaction. | |
1935 | */ | |
d34c0a75 | 1936 | static unsigned int fragmentation_score_zone(struct zone *zone) |
facdaa91 NG |
1937 | { |
1938 | unsigned long score; | |
1939 | ||
1940 | score = zone->present_pages * | |
1941 | extfrag_for_order(zone, COMPACTION_HPAGE_ORDER); | |
1942 | return div64_ul(score, zone->zone_pgdat->node_present_pages + 1); | |
1943 | } | |
1944 | ||
1945 | /* | |
1946 | * The per-node proactive (background) compaction process is started by its | |
1947 | * corresponding kcompactd thread when the node's fragmentation score | |
1948 | * exceeds the high threshold. The compaction process remains active till | |
1949 | * the node's score falls below the low threshold, or one of the back-off | |
1950 | * conditions is met. | |
1951 | */ | |
d34c0a75 | 1952 | static unsigned int fragmentation_score_node(pg_data_t *pgdat) |
facdaa91 | 1953 | { |
d34c0a75 | 1954 | unsigned int score = 0; |
facdaa91 NG |
1955 | int zoneid; |
1956 | ||
1957 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
1958 | struct zone *zone; | |
1959 | ||
1960 | zone = &pgdat->node_zones[zoneid]; | |
1961 | score += fragmentation_score_zone(zone); | |
1962 | } | |
1963 | ||
1964 | return score; | |
1965 | } | |
1966 | ||
d34c0a75 | 1967 | static unsigned int fragmentation_score_wmark(pg_data_t *pgdat, bool low) |
facdaa91 | 1968 | { |
d34c0a75 | 1969 | unsigned int wmark_low; |
facdaa91 NG |
1970 | |
1971 | /* | |
1972 | * Cap the low watermak to avoid excessive compaction | |
1973 | * activity in case a user sets the proactivess tunable | |
1974 | * close to 100 (maximum). | |
1975 | */ | |
d34c0a75 NG |
1976 | wmark_low = max(100U - sysctl_compaction_proactiveness, 5U); |
1977 | return low ? wmark_low : min(wmark_low + 10, 100U); | |
facdaa91 NG |
1978 | } |
1979 | ||
1980 | static bool should_proactive_compact_node(pg_data_t *pgdat) | |
1981 | { | |
1982 | int wmark_high; | |
1983 | ||
1984 | if (!sysctl_compaction_proactiveness || kswapd_is_running(pgdat)) | |
1985 | return false; | |
1986 | ||
1987 | wmark_high = fragmentation_score_wmark(pgdat, false); | |
1988 | return fragmentation_score_node(pgdat) > wmark_high; | |
1989 | } | |
1990 | ||
40cacbcb | 1991 | static enum compact_result __compact_finished(struct compact_control *cc) |
748446bb | 1992 | { |
8fb74b9f | 1993 | unsigned int order; |
d39773a0 | 1994 | const int migratetype = cc->migratetype; |
cb2dcaf0 | 1995 | int ret; |
748446bb | 1996 | |
753341a4 | 1997 | /* Compaction run completes if the migrate and free scanner meet */ |
f2849aa0 | 1998 | if (compact_scanners_met(cc)) { |
55b7c4c9 | 1999 | /* Let the next compaction start anew. */ |
40cacbcb | 2000 | reset_cached_positions(cc->zone); |
55b7c4c9 | 2001 | |
62997027 MG |
2002 | /* |
2003 | * Mark that the PG_migrate_skip information should be cleared | |
accf6242 | 2004 | * by kswapd when it goes to sleep. kcompactd does not set the |
62997027 MG |
2005 | * flag itself as the decision to be clear should be directly |
2006 | * based on an allocation request. | |
2007 | */ | |
accf6242 | 2008 | if (cc->direct_compaction) |
40cacbcb | 2009 | cc->zone->compact_blockskip_flush = true; |
62997027 | 2010 | |
c8f7de0b MH |
2011 | if (cc->whole_zone) |
2012 | return COMPACT_COMPLETE; | |
2013 | else | |
2014 | return COMPACT_PARTIAL_SKIPPED; | |
bb13ffeb | 2015 | } |
748446bb | 2016 | |
facdaa91 NG |
2017 | if (cc->proactive_compaction) { |
2018 | int score, wmark_low; | |
2019 | pg_data_t *pgdat; | |
2020 | ||
2021 | pgdat = cc->zone->zone_pgdat; | |
2022 | if (kswapd_is_running(pgdat)) | |
2023 | return COMPACT_PARTIAL_SKIPPED; | |
2024 | ||
2025 | score = fragmentation_score_zone(cc->zone); | |
2026 | wmark_low = fragmentation_score_wmark(pgdat, true); | |
2027 | ||
2028 | if (score > wmark_low) | |
2029 | ret = COMPACT_CONTINUE; | |
2030 | else | |
2031 | ret = COMPACT_SUCCESS; | |
2032 | ||
2033 | goto out; | |
2034 | } | |
2035 | ||
21c527a3 | 2036 | if (is_via_compact_memory(cc->order)) |
56de7263 MG |
2037 | return COMPACT_CONTINUE; |
2038 | ||
efe771c7 MG |
2039 | /* |
2040 | * Always finish scanning a pageblock to reduce the possibility of | |
2041 | * fallbacks in the future. This is particularly important when | |
2042 | * migration source is unmovable/reclaimable but it's not worth | |
2043 | * special casing. | |
2044 | */ | |
2045 | if (!IS_ALIGNED(cc->migrate_pfn, pageblock_nr_pages)) | |
2046 | return COMPACT_CONTINUE; | |
baf6a9a1 | 2047 | |
56de7263 | 2048 | /* Direct compactor: Is a suitable page free? */ |
cb2dcaf0 | 2049 | ret = COMPACT_NO_SUITABLE_PAGE; |
8fb74b9f | 2050 | for (order = cc->order; order < MAX_ORDER; order++) { |
40cacbcb | 2051 | struct free_area *area = &cc->zone->free_area[order]; |
2149cdae | 2052 | bool can_steal; |
8fb74b9f MG |
2053 | |
2054 | /* Job done if page is free of the right migratetype */ | |
b03641af | 2055 | if (!free_area_empty(area, migratetype)) |
cf378319 | 2056 | return COMPACT_SUCCESS; |
8fb74b9f | 2057 | |
2149cdae JK |
2058 | #ifdef CONFIG_CMA |
2059 | /* MIGRATE_MOVABLE can fallback on MIGRATE_CMA */ | |
2060 | if (migratetype == MIGRATE_MOVABLE && | |
b03641af | 2061 | !free_area_empty(area, MIGRATE_CMA)) |
cf378319 | 2062 | return COMPACT_SUCCESS; |
2149cdae JK |
2063 | #endif |
2064 | /* | |
2065 | * Job done if allocation would steal freepages from | |
2066 | * other migratetype buddy lists. | |
2067 | */ | |
2068 | if (find_suitable_fallback(area, order, migratetype, | |
baf6a9a1 VB |
2069 | true, &can_steal) != -1) { |
2070 | ||
2071 | /* movable pages are OK in any pageblock */ | |
2072 | if (migratetype == MIGRATE_MOVABLE) | |
2073 | return COMPACT_SUCCESS; | |
2074 | ||
2075 | /* | |
2076 | * We are stealing for a non-movable allocation. Make | |
2077 | * sure we finish compacting the current pageblock | |
2078 | * first so it is as free as possible and we won't | |
2079 | * have to steal another one soon. This only applies | |
2080 | * to sync compaction, as async compaction operates | |
2081 | * on pageblocks of the same migratetype. | |
2082 | */ | |
2083 | if (cc->mode == MIGRATE_ASYNC || | |
2084 | IS_ALIGNED(cc->migrate_pfn, | |
2085 | pageblock_nr_pages)) { | |
2086 | return COMPACT_SUCCESS; | |
2087 | } | |
2088 | ||
cb2dcaf0 MG |
2089 | ret = COMPACT_CONTINUE; |
2090 | break; | |
baf6a9a1 | 2091 | } |
56de7263 MG |
2092 | } |
2093 | ||
facdaa91 | 2094 | out: |
cb2dcaf0 MG |
2095 | if (cc->contended || fatal_signal_pending(current)) |
2096 | ret = COMPACT_CONTENDED; | |
2097 | ||
2098 | return ret; | |
837d026d JK |
2099 | } |
2100 | ||
40cacbcb | 2101 | static enum compact_result compact_finished(struct compact_control *cc) |
837d026d JK |
2102 | { |
2103 | int ret; | |
2104 | ||
40cacbcb MG |
2105 | ret = __compact_finished(cc); |
2106 | trace_mm_compaction_finished(cc->zone, cc->order, ret); | |
837d026d JK |
2107 | if (ret == COMPACT_NO_SUITABLE_PAGE) |
2108 | ret = COMPACT_CONTINUE; | |
2109 | ||
2110 | return ret; | |
748446bb MG |
2111 | } |
2112 | ||
ea7ab982 | 2113 | static enum compact_result __compaction_suitable(struct zone *zone, int order, |
c603844b | 2114 | unsigned int alloc_flags, |
97a225e6 | 2115 | int highest_zoneidx, |
86a294a8 | 2116 | unsigned long wmark_target) |
3e7d3449 | 2117 | { |
3e7d3449 MG |
2118 | unsigned long watermark; |
2119 | ||
21c527a3 | 2120 | if (is_via_compact_memory(order)) |
3957c776 MH |
2121 | return COMPACT_CONTINUE; |
2122 | ||
a9214443 | 2123 | watermark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK); |
ebff3980 VB |
2124 | /* |
2125 | * If watermarks for high-order allocation are already met, there | |
2126 | * should be no need for compaction at all. | |
2127 | */ | |
97a225e6 | 2128 | if (zone_watermark_ok(zone, order, watermark, highest_zoneidx, |
ebff3980 | 2129 | alloc_flags)) |
cf378319 | 2130 | return COMPACT_SUCCESS; |
ebff3980 | 2131 | |
3e7d3449 | 2132 | /* |
9861a62c | 2133 | * Watermarks for order-0 must be met for compaction to be able to |
984fdba6 VB |
2134 | * isolate free pages for migration targets. This means that the |
2135 | * watermark and alloc_flags have to match, or be more pessimistic than | |
2136 | * the check in __isolate_free_page(). We don't use the direct | |
2137 | * compactor's alloc_flags, as they are not relevant for freepage | |
97a225e6 JK |
2138 | * isolation. We however do use the direct compactor's highest_zoneidx |
2139 | * to skip over zones where lowmem reserves would prevent allocation | |
2140 | * even if compaction succeeds. | |
8348faf9 VB |
2141 | * For costly orders, we require low watermark instead of min for |
2142 | * compaction to proceed to increase its chances. | |
d883c6cf JK |
2143 | * ALLOC_CMA is used, as pages in CMA pageblocks are considered |
2144 | * suitable migration targets | |
3e7d3449 | 2145 | */ |
8348faf9 VB |
2146 | watermark = (order > PAGE_ALLOC_COSTLY_ORDER) ? |
2147 | low_wmark_pages(zone) : min_wmark_pages(zone); | |
2148 | watermark += compact_gap(order); | |
97a225e6 | 2149 | if (!__zone_watermark_ok(zone, 0, watermark, highest_zoneidx, |
d883c6cf | 2150 | ALLOC_CMA, wmark_target)) |
3e7d3449 MG |
2151 | return COMPACT_SKIPPED; |
2152 | ||
cc5c9f09 VB |
2153 | return COMPACT_CONTINUE; |
2154 | } | |
2155 | ||
2b1a20c3 HS |
2156 | /* |
2157 | * compaction_suitable: Is this suitable to run compaction on this zone now? | |
2158 | * Returns | |
2159 | * COMPACT_SKIPPED - If there are too few free pages for compaction | |
2160 | * COMPACT_SUCCESS - If the allocation would succeed without compaction | |
2161 | * COMPACT_CONTINUE - If compaction should run now | |
2162 | */ | |
cc5c9f09 VB |
2163 | enum compact_result compaction_suitable(struct zone *zone, int order, |
2164 | unsigned int alloc_flags, | |
97a225e6 | 2165 | int highest_zoneidx) |
cc5c9f09 VB |
2166 | { |
2167 | enum compact_result ret; | |
2168 | int fragindex; | |
2169 | ||
97a225e6 | 2170 | ret = __compaction_suitable(zone, order, alloc_flags, highest_zoneidx, |
cc5c9f09 | 2171 | zone_page_state(zone, NR_FREE_PAGES)); |
3e7d3449 MG |
2172 | /* |
2173 | * fragmentation index determines if allocation failures are due to | |
2174 | * low memory or external fragmentation | |
2175 | * | |
ebff3980 VB |
2176 | * index of -1000 would imply allocations might succeed depending on |
2177 | * watermarks, but we already failed the high-order watermark check | |
3e7d3449 MG |
2178 | * index towards 0 implies failure is due to lack of memory |
2179 | * index towards 1000 implies failure is due to fragmentation | |
2180 | * | |
20311420 VB |
2181 | * Only compact if a failure would be due to fragmentation. Also |
2182 | * ignore fragindex for non-costly orders where the alternative to | |
2183 | * a successful reclaim/compaction is OOM. Fragindex and the | |
2184 | * vm.extfrag_threshold sysctl is meant as a heuristic to prevent | |
2185 | * excessive compaction for costly orders, but it should not be at the | |
2186 | * expense of system stability. | |
3e7d3449 | 2187 | */ |
20311420 | 2188 | if (ret == COMPACT_CONTINUE && (order > PAGE_ALLOC_COSTLY_ORDER)) { |
cc5c9f09 VB |
2189 | fragindex = fragmentation_index(zone, order); |
2190 | if (fragindex >= 0 && fragindex <= sysctl_extfrag_threshold) | |
2191 | ret = COMPACT_NOT_SUITABLE_ZONE; | |
2192 | } | |
837d026d | 2193 | |
837d026d JK |
2194 | trace_mm_compaction_suitable(zone, order, ret); |
2195 | if (ret == COMPACT_NOT_SUITABLE_ZONE) | |
2196 | ret = COMPACT_SKIPPED; | |
2197 | ||
2198 | return ret; | |
2199 | } | |
2200 | ||
86a294a8 MH |
2201 | bool compaction_zonelist_suitable(struct alloc_context *ac, int order, |
2202 | int alloc_flags) | |
2203 | { | |
2204 | struct zone *zone; | |
2205 | struct zoneref *z; | |
2206 | ||
2207 | /* | |
2208 | * Make sure at least one zone would pass __compaction_suitable if we continue | |
2209 | * retrying the reclaim. | |
2210 | */ | |
97a225e6 JK |
2211 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
2212 | ac->highest_zoneidx, ac->nodemask) { | |
86a294a8 MH |
2213 | unsigned long available; |
2214 | enum compact_result compact_result; | |
2215 | ||
2216 | /* | |
2217 | * Do not consider all the reclaimable memory because we do not | |
2218 | * want to trash just for a single high order allocation which | |
2219 | * is even not guaranteed to appear even if __compaction_suitable | |
2220 | * is happy about the watermark check. | |
2221 | */ | |
5a1c84b4 | 2222 | available = zone_reclaimable_pages(zone) / order; |
86a294a8 MH |
2223 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); |
2224 | compact_result = __compaction_suitable(zone, order, alloc_flags, | |
97a225e6 | 2225 | ac->highest_zoneidx, available); |
cc5c9f09 | 2226 | if (compact_result != COMPACT_SKIPPED) |
86a294a8 MH |
2227 | return true; |
2228 | } | |
2229 | ||
2230 | return false; | |
2231 | } | |
2232 | ||
5e1f0f09 MG |
2233 | static enum compact_result |
2234 | compact_zone(struct compact_control *cc, struct capture_control *capc) | |
748446bb | 2235 | { |
ea7ab982 | 2236 | enum compact_result ret; |
40cacbcb MG |
2237 | unsigned long start_pfn = cc->zone->zone_start_pfn; |
2238 | unsigned long end_pfn = zone_end_pfn(cc->zone); | |
566e54e1 | 2239 | unsigned long last_migrated_pfn; |
e0b9daeb | 2240 | const bool sync = cc->mode != MIGRATE_ASYNC; |
8854c55f | 2241 | bool update_cached; |
748446bb | 2242 | |
a94b5252 YS |
2243 | /* |
2244 | * These counters track activities during zone compaction. Initialize | |
2245 | * them before compacting a new zone. | |
2246 | */ | |
2247 | cc->total_migrate_scanned = 0; | |
2248 | cc->total_free_scanned = 0; | |
2249 | cc->nr_migratepages = 0; | |
2250 | cc->nr_freepages = 0; | |
2251 | INIT_LIST_HEAD(&cc->freepages); | |
2252 | INIT_LIST_HEAD(&cc->migratepages); | |
2253 | ||
01c0bfe0 | 2254 | cc->migratetype = gfp_migratetype(cc->gfp_mask); |
40cacbcb | 2255 | ret = compaction_suitable(cc->zone, cc->order, cc->alloc_flags, |
97a225e6 | 2256 | cc->highest_zoneidx); |
c46649de | 2257 | /* Compaction is likely to fail */ |
cf378319 | 2258 | if (ret == COMPACT_SUCCESS || ret == COMPACT_SKIPPED) |
3e7d3449 | 2259 | return ret; |
c46649de MH |
2260 | |
2261 | /* huh, compaction_suitable is returning something unexpected */ | |
2262 | VM_BUG_ON(ret != COMPACT_CONTINUE); | |
3e7d3449 | 2263 | |
d3132e4b VB |
2264 | /* |
2265 | * Clear pageblock skip if there were failures recently and compaction | |
accf6242 | 2266 | * is about to be retried after being deferred. |
d3132e4b | 2267 | */ |
40cacbcb MG |
2268 | if (compaction_restarting(cc->zone, cc->order)) |
2269 | __reset_isolation_suitable(cc->zone); | |
d3132e4b | 2270 | |
c89511ab MG |
2271 | /* |
2272 | * Setup to move all movable pages to the end of the zone. Used cached | |
06ed2998 VB |
2273 | * information on where the scanners should start (unless we explicitly |
2274 | * want to compact the whole zone), but check that it is initialised | |
2275 | * by ensuring the values are within zone boundaries. | |
c89511ab | 2276 | */ |
70b44595 | 2277 | cc->fast_start_pfn = 0; |
06ed2998 | 2278 | if (cc->whole_zone) { |
c89511ab | 2279 | cc->migrate_pfn = start_pfn; |
06ed2998 VB |
2280 | cc->free_pfn = pageblock_start_pfn(end_pfn - 1); |
2281 | } else { | |
40cacbcb MG |
2282 | cc->migrate_pfn = cc->zone->compact_cached_migrate_pfn[sync]; |
2283 | cc->free_pfn = cc->zone->compact_cached_free_pfn; | |
06ed2998 VB |
2284 | if (cc->free_pfn < start_pfn || cc->free_pfn >= end_pfn) { |
2285 | cc->free_pfn = pageblock_start_pfn(end_pfn - 1); | |
40cacbcb | 2286 | cc->zone->compact_cached_free_pfn = cc->free_pfn; |
06ed2998 VB |
2287 | } |
2288 | if (cc->migrate_pfn < start_pfn || cc->migrate_pfn >= end_pfn) { | |
2289 | cc->migrate_pfn = start_pfn; | |
40cacbcb MG |
2290 | cc->zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn; |
2291 | cc->zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn; | |
06ed2998 | 2292 | } |
c8f7de0b | 2293 | |
e332f741 | 2294 | if (cc->migrate_pfn <= cc->zone->compact_init_migrate_pfn) |
06ed2998 VB |
2295 | cc->whole_zone = true; |
2296 | } | |
c8f7de0b | 2297 | |
566e54e1 | 2298 | last_migrated_pfn = 0; |
748446bb | 2299 | |
8854c55f MG |
2300 | /* |
2301 | * Migrate has separate cached PFNs for ASYNC and SYNC* migration on | |
2302 | * the basis that some migrations will fail in ASYNC mode. However, | |
2303 | * if the cached PFNs match and pageblocks are skipped due to having | |
2304 | * no isolation candidates, then the sync state does not matter. | |
2305 | * Until a pageblock with isolation candidates is found, keep the | |
2306 | * cached PFNs in sync to avoid revisiting the same blocks. | |
2307 | */ | |
2308 | update_cached = !sync && | |
2309 | cc->zone->compact_cached_migrate_pfn[0] == cc->zone->compact_cached_migrate_pfn[1]; | |
2310 | ||
16c4a097 JK |
2311 | trace_mm_compaction_begin(start_pfn, cc->migrate_pfn, |
2312 | cc->free_pfn, end_pfn, sync); | |
0eb927c0 | 2313 | |
748446bb MG |
2314 | migrate_prep_local(); |
2315 | ||
40cacbcb | 2316 | while ((ret = compact_finished(cc)) == COMPACT_CONTINUE) { |
9d502c1c | 2317 | int err; |
19d3cf9d | 2318 | unsigned long iteration_start_pfn = cc->migrate_pfn; |
748446bb | 2319 | |
804d3121 MG |
2320 | /* |
2321 | * Avoid multiple rescans which can happen if a page cannot be | |
2322 | * isolated (dirty/writeback in async mode) or if the migrated | |
2323 | * pages are being allocated before the pageblock is cleared. | |
2324 | * The first rescan will capture the entire pageblock for | |
2325 | * migration. If it fails, it'll be marked skip and scanning | |
2326 | * will proceed as normal. | |
2327 | */ | |
2328 | cc->rescan = false; | |
2329 | if (pageblock_start_pfn(last_migrated_pfn) == | |
19d3cf9d | 2330 | pageblock_start_pfn(iteration_start_pfn)) { |
804d3121 MG |
2331 | cc->rescan = true; |
2332 | } | |
2333 | ||
32aaf055 | 2334 | switch (isolate_migratepages(cc)) { |
f9e35b3b | 2335 | case ISOLATE_ABORT: |
2d1e1041 | 2336 | ret = COMPACT_CONTENDED; |
5733c7d1 | 2337 | putback_movable_pages(&cc->migratepages); |
e64c5237 | 2338 | cc->nr_migratepages = 0; |
f9e35b3b MG |
2339 | goto out; |
2340 | case ISOLATE_NONE: | |
8854c55f MG |
2341 | if (update_cached) { |
2342 | cc->zone->compact_cached_migrate_pfn[1] = | |
2343 | cc->zone->compact_cached_migrate_pfn[0]; | |
2344 | } | |
2345 | ||
fdaf7f5c VB |
2346 | /* |
2347 | * We haven't isolated and migrated anything, but | |
2348 | * there might still be unflushed migrations from | |
2349 | * previous cc->order aligned block. | |
2350 | */ | |
2351 | goto check_drain; | |
f9e35b3b | 2352 | case ISOLATE_SUCCESS: |
8854c55f | 2353 | update_cached = false; |
19d3cf9d | 2354 | last_migrated_pfn = iteration_start_pfn; |
f9e35b3b | 2355 | } |
748446bb | 2356 | |
d53aea3d | 2357 | err = migrate_pages(&cc->migratepages, compaction_alloc, |
e0b9daeb | 2358 | compaction_free, (unsigned long)cc, cc->mode, |
7b2a2d4a | 2359 | MR_COMPACTION); |
748446bb | 2360 | |
f8c9301f VB |
2361 | trace_mm_compaction_migratepages(cc->nr_migratepages, err, |
2362 | &cc->migratepages); | |
748446bb | 2363 | |
f8c9301f VB |
2364 | /* All pages were either migrated or will be released */ |
2365 | cc->nr_migratepages = 0; | |
9d502c1c | 2366 | if (err) { |
5733c7d1 | 2367 | putback_movable_pages(&cc->migratepages); |
7ed695e0 VB |
2368 | /* |
2369 | * migrate_pages() may return -ENOMEM when scanners meet | |
2370 | * and we want compact_finished() to detect it | |
2371 | */ | |
f2849aa0 | 2372 | if (err == -ENOMEM && !compact_scanners_met(cc)) { |
2d1e1041 | 2373 | ret = COMPACT_CONTENDED; |
4bf2bba3 DR |
2374 | goto out; |
2375 | } | |
fdd048e1 VB |
2376 | /* |
2377 | * We failed to migrate at least one page in the current | |
2378 | * order-aligned block, so skip the rest of it. | |
2379 | */ | |
2380 | if (cc->direct_compaction && | |
2381 | (cc->mode == MIGRATE_ASYNC)) { | |
2382 | cc->migrate_pfn = block_end_pfn( | |
2383 | cc->migrate_pfn - 1, cc->order); | |
2384 | /* Draining pcplists is useless in this case */ | |
566e54e1 | 2385 | last_migrated_pfn = 0; |
fdd048e1 | 2386 | } |
748446bb | 2387 | } |
fdaf7f5c | 2388 | |
fdaf7f5c VB |
2389 | check_drain: |
2390 | /* | |
2391 | * Has the migration scanner moved away from the previous | |
2392 | * cc->order aligned block where we migrated from? If yes, | |
2393 | * flush the pages that were freed, so that they can merge and | |
2394 | * compact_finished() can detect immediately if allocation | |
2395 | * would succeed. | |
2396 | */ | |
566e54e1 | 2397 | if (cc->order > 0 && last_migrated_pfn) { |
fdaf7f5c | 2398 | unsigned long current_block_start = |
06b6640a | 2399 | block_start_pfn(cc->migrate_pfn, cc->order); |
fdaf7f5c | 2400 | |
566e54e1 | 2401 | if (last_migrated_pfn < current_block_start) { |
b01b2141 | 2402 | lru_add_drain_cpu_zone(cc->zone); |
fdaf7f5c | 2403 | /* No more flushing until we migrate again */ |
566e54e1 | 2404 | last_migrated_pfn = 0; |
fdaf7f5c VB |
2405 | } |
2406 | } | |
2407 | ||
5e1f0f09 MG |
2408 | /* Stop if a page has been captured */ |
2409 | if (capc && capc->page) { | |
2410 | ret = COMPACT_SUCCESS; | |
2411 | break; | |
2412 | } | |
748446bb MG |
2413 | } |
2414 | ||
f9e35b3b | 2415 | out: |
6bace090 VB |
2416 | /* |
2417 | * Release free pages and update where the free scanner should restart, | |
2418 | * so we don't leave any returned pages behind in the next attempt. | |
2419 | */ | |
2420 | if (cc->nr_freepages > 0) { | |
2421 | unsigned long free_pfn = release_freepages(&cc->freepages); | |
2422 | ||
2423 | cc->nr_freepages = 0; | |
2424 | VM_BUG_ON(free_pfn == 0); | |
2425 | /* The cached pfn is always the first in a pageblock */ | |
06b6640a | 2426 | free_pfn = pageblock_start_pfn(free_pfn); |
6bace090 VB |
2427 | /* |
2428 | * Only go back, not forward. The cached pfn might have been | |
2429 | * already reset to zone end in compact_finished() | |
2430 | */ | |
40cacbcb MG |
2431 | if (free_pfn > cc->zone->compact_cached_free_pfn) |
2432 | cc->zone->compact_cached_free_pfn = free_pfn; | |
6bace090 | 2433 | } |
748446bb | 2434 | |
7f354a54 DR |
2435 | count_compact_events(COMPACTMIGRATE_SCANNED, cc->total_migrate_scanned); |
2436 | count_compact_events(COMPACTFREE_SCANNED, cc->total_free_scanned); | |
2437 | ||
16c4a097 JK |
2438 | trace_mm_compaction_end(start_pfn, cc->migrate_pfn, |
2439 | cc->free_pfn, end_pfn, sync, ret); | |
0eb927c0 | 2440 | |
748446bb MG |
2441 | return ret; |
2442 | } | |
76ab0f53 | 2443 | |
ea7ab982 | 2444 | static enum compact_result compact_zone_order(struct zone *zone, int order, |
c3486f53 | 2445 | gfp_t gfp_mask, enum compact_priority prio, |
97a225e6 | 2446 | unsigned int alloc_flags, int highest_zoneidx, |
5e1f0f09 | 2447 | struct page **capture) |
56de7263 | 2448 | { |
ea7ab982 | 2449 | enum compact_result ret; |
56de7263 | 2450 | struct compact_control cc = { |
56de7263 | 2451 | .order = order, |
dbe2d4e4 | 2452 | .search_order = order, |
6d7ce559 | 2453 | .gfp_mask = gfp_mask, |
56de7263 | 2454 | .zone = zone, |
a5508cd8 VB |
2455 | .mode = (prio == COMPACT_PRIO_ASYNC) ? |
2456 | MIGRATE_ASYNC : MIGRATE_SYNC_LIGHT, | |
ebff3980 | 2457 | .alloc_flags = alloc_flags, |
97a225e6 | 2458 | .highest_zoneidx = highest_zoneidx, |
accf6242 | 2459 | .direct_compaction = true, |
a8e025e5 | 2460 | .whole_zone = (prio == MIN_COMPACT_PRIORITY), |
9f7e3387 VB |
2461 | .ignore_skip_hint = (prio == MIN_COMPACT_PRIORITY), |
2462 | .ignore_block_suitable = (prio == MIN_COMPACT_PRIORITY) | |
56de7263 | 2463 | }; |
5e1f0f09 MG |
2464 | struct capture_control capc = { |
2465 | .cc = &cc, | |
2466 | .page = NULL, | |
2467 | }; | |
2468 | ||
b9e20f0d VB |
2469 | /* |
2470 | * Make sure the structs are really initialized before we expose the | |
2471 | * capture control, in case we are interrupted and the interrupt handler | |
2472 | * frees a page. | |
2473 | */ | |
2474 | barrier(); | |
2475 | WRITE_ONCE(current->capture_control, &capc); | |
56de7263 | 2476 | |
5e1f0f09 | 2477 | ret = compact_zone(&cc, &capc); |
e64c5237 SL |
2478 | |
2479 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
2480 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
2481 | ||
b9e20f0d VB |
2482 | /* |
2483 | * Make sure we hide capture control first before we read the captured | |
2484 | * page pointer, otherwise an interrupt could free and capture a page | |
2485 | * and we would leak it. | |
2486 | */ | |
2487 | WRITE_ONCE(current->capture_control, NULL); | |
2488 | *capture = READ_ONCE(capc.page); | |
5e1f0f09 | 2489 | |
e64c5237 | 2490 | return ret; |
56de7263 MG |
2491 | } |
2492 | ||
5e771905 MG |
2493 | int sysctl_extfrag_threshold = 500; |
2494 | ||
56de7263 MG |
2495 | /** |
2496 | * try_to_compact_pages - Direct compact to satisfy a high-order allocation | |
56de7263 | 2497 | * @gfp_mask: The GFP mask of the current allocation |
1a6d53a1 VB |
2498 | * @order: The order of the current allocation |
2499 | * @alloc_flags: The allocation flags of the current allocation | |
2500 | * @ac: The context of current allocation | |
112d2d29 | 2501 | * @prio: Determines how hard direct compaction should try to succeed |
6467552c | 2502 | * @capture: Pointer to free page created by compaction will be stored here |
56de7263 MG |
2503 | * |
2504 | * This is the main entry point for direct page compaction. | |
2505 | */ | |
ea7ab982 | 2506 | enum compact_result try_to_compact_pages(gfp_t gfp_mask, unsigned int order, |
c603844b | 2507 | unsigned int alloc_flags, const struct alloc_context *ac, |
5e1f0f09 | 2508 | enum compact_priority prio, struct page **capture) |
56de7263 | 2509 | { |
56de7263 | 2510 | int may_perform_io = gfp_mask & __GFP_IO; |
56de7263 MG |
2511 | struct zoneref *z; |
2512 | struct zone *zone; | |
1d4746d3 | 2513 | enum compact_result rc = COMPACT_SKIPPED; |
56de7263 | 2514 | |
73e64c51 MH |
2515 | /* |
2516 | * Check if the GFP flags allow compaction - GFP_NOIO is really | |
2517 | * tricky context because the migration might require IO | |
2518 | */ | |
2519 | if (!may_perform_io) | |
53853e2d | 2520 | return COMPACT_SKIPPED; |
56de7263 | 2521 | |
a5508cd8 | 2522 | trace_mm_compaction_try_to_compact_pages(order, gfp_mask, prio); |
837d026d | 2523 | |
56de7263 | 2524 | /* Compact each zone in the list */ |
97a225e6 JK |
2525 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, |
2526 | ac->highest_zoneidx, ac->nodemask) { | |
ea7ab982 | 2527 | enum compact_result status; |
56de7263 | 2528 | |
a8e025e5 VB |
2529 | if (prio > MIN_COMPACT_PRIORITY |
2530 | && compaction_deferred(zone, order)) { | |
1d4746d3 | 2531 | rc = max_t(enum compact_result, COMPACT_DEFERRED, rc); |
53853e2d | 2532 | continue; |
1d4746d3 | 2533 | } |
53853e2d | 2534 | |
a5508cd8 | 2535 | status = compact_zone_order(zone, order, gfp_mask, prio, |
97a225e6 | 2536 | alloc_flags, ac->highest_zoneidx, capture); |
56de7263 MG |
2537 | rc = max(status, rc); |
2538 | ||
7ceb009a VB |
2539 | /* The allocation should succeed, stop compacting */ |
2540 | if (status == COMPACT_SUCCESS) { | |
53853e2d VB |
2541 | /* |
2542 | * We think the allocation will succeed in this zone, | |
2543 | * but it is not certain, hence the false. The caller | |
2544 | * will repeat this with true if allocation indeed | |
2545 | * succeeds in this zone. | |
2546 | */ | |
2547 | compaction_defer_reset(zone, order, false); | |
1f9efdef | 2548 | |
c3486f53 | 2549 | break; |
1f9efdef VB |
2550 | } |
2551 | ||
a5508cd8 | 2552 | if (prio != COMPACT_PRIO_ASYNC && (status == COMPACT_COMPLETE || |
c3486f53 | 2553 | status == COMPACT_PARTIAL_SKIPPED)) |
53853e2d VB |
2554 | /* |
2555 | * We think that allocation won't succeed in this zone | |
2556 | * so we defer compaction there. If it ends up | |
2557 | * succeeding after all, it will be reset. | |
2558 | */ | |
2559 | defer_compaction(zone, order); | |
1f9efdef VB |
2560 | |
2561 | /* | |
2562 | * We might have stopped compacting due to need_resched() in | |
2563 | * async compaction, or due to a fatal signal detected. In that | |
c3486f53 | 2564 | * case do not try further zones |
1f9efdef | 2565 | */ |
c3486f53 VB |
2566 | if ((prio == COMPACT_PRIO_ASYNC && need_resched()) |
2567 | || fatal_signal_pending(current)) | |
2568 | break; | |
56de7263 MG |
2569 | } |
2570 | ||
2571 | return rc; | |
2572 | } | |
2573 | ||
facdaa91 NG |
2574 | /* |
2575 | * Compact all zones within a node till each zone's fragmentation score | |
2576 | * reaches within proactive compaction thresholds (as determined by the | |
2577 | * proactiveness tunable). | |
2578 | * | |
2579 | * It is possible that the function returns before reaching score targets | |
2580 | * due to various back-off conditions, such as, contention on per-node or | |
2581 | * per-zone locks. | |
2582 | */ | |
2583 | static void proactive_compact_node(pg_data_t *pgdat) | |
2584 | { | |
2585 | int zoneid; | |
2586 | struct zone *zone; | |
2587 | struct compact_control cc = { | |
2588 | .order = -1, | |
2589 | .mode = MIGRATE_SYNC_LIGHT, | |
2590 | .ignore_skip_hint = true, | |
2591 | .whole_zone = true, | |
2592 | .gfp_mask = GFP_KERNEL, | |
2593 | .proactive_compaction = true, | |
2594 | }; | |
2595 | ||
2596 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
2597 | zone = &pgdat->node_zones[zoneid]; | |
2598 | if (!populated_zone(zone)) | |
2599 | continue; | |
2600 | ||
2601 | cc.zone = zone; | |
2602 | ||
2603 | compact_zone(&cc, NULL); | |
2604 | ||
2605 | VM_BUG_ON(!list_empty(&cc.freepages)); | |
2606 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
2607 | } | |
2608 | } | |
56de7263 | 2609 | |
76ab0f53 | 2610 | /* Compact all zones within a node */ |
791cae96 | 2611 | static void compact_node(int nid) |
76ab0f53 | 2612 | { |
791cae96 | 2613 | pg_data_t *pgdat = NODE_DATA(nid); |
76ab0f53 | 2614 | int zoneid; |
76ab0f53 | 2615 | struct zone *zone; |
791cae96 VB |
2616 | struct compact_control cc = { |
2617 | .order = -1, | |
2618 | .mode = MIGRATE_SYNC, | |
2619 | .ignore_skip_hint = true, | |
2620 | .whole_zone = true, | |
73e64c51 | 2621 | .gfp_mask = GFP_KERNEL, |
791cae96 VB |
2622 | }; |
2623 | ||
76ab0f53 | 2624 | |
76ab0f53 | 2625 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { |
76ab0f53 MG |
2626 | |
2627 | zone = &pgdat->node_zones[zoneid]; | |
2628 | if (!populated_zone(zone)) | |
2629 | continue; | |
2630 | ||
791cae96 | 2631 | cc.zone = zone; |
76ab0f53 | 2632 | |
5e1f0f09 | 2633 | compact_zone(&cc, NULL); |
75469345 | 2634 | |
791cae96 VB |
2635 | VM_BUG_ON(!list_empty(&cc.freepages)); |
2636 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
76ab0f53 | 2637 | } |
76ab0f53 MG |
2638 | } |
2639 | ||
2640 | /* Compact all nodes in the system */ | |
7964c06d | 2641 | static void compact_nodes(void) |
76ab0f53 MG |
2642 | { |
2643 | int nid; | |
2644 | ||
8575ec29 HD |
2645 | /* Flush pending updates to the LRU lists */ |
2646 | lru_add_drain_all(); | |
2647 | ||
76ab0f53 MG |
2648 | for_each_online_node(nid) |
2649 | compact_node(nid); | |
76ab0f53 MG |
2650 | } |
2651 | ||
2652 | /* The written value is actually unused, all memory is compacted */ | |
2653 | int sysctl_compact_memory; | |
2654 | ||
facdaa91 NG |
2655 | /* |
2656 | * Tunable for proactive compaction. It determines how | |
2657 | * aggressively the kernel should compact memory in the | |
2658 | * background. It takes values in the range [0, 100]. | |
2659 | */ | |
d34c0a75 | 2660 | unsigned int __read_mostly sysctl_compaction_proactiveness = 20; |
facdaa91 | 2661 | |
fec4eb2c YB |
2662 | /* |
2663 | * This is the entry point for compacting all nodes via | |
2664 | * /proc/sys/vm/compact_memory | |
2665 | */ | |
76ab0f53 | 2666 | int sysctl_compaction_handler(struct ctl_table *table, int write, |
32927393 | 2667 | void *buffer, size_t *length, loff_t *ppos) |
76ab0f53 MG |
2668 | { |
2669 | if (write) | |
7964c06d | 2670 | compact_nodes(); |
76ab0f53 MG |
2671 | |
2672 | return 0; | |
2673 | } | |
ed4a6d7f MG |
2674 | |
2675 | #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA) | |
74e77fb9 | 2676 | static ssize_t sysfs_compact_node(struct device *dev, |
10fbcf4c | 2677 | struct device_attribute *attr, |
ed4a6d7f MG |
2678 | const char *buf, size_t count) |
2679 | { | |
8575ec29 HD |
2680 | int nid = dev->id; |
2681 | ||
2682 | if (nid >= 0 && nid < nr_node_ids && node_online(nid)) { | |
2683 | /* Flush pending updates to the LRU lists */ | |
2684 | lru_add_drain_all(); | |
2685 | ||
2686 | compact_node(nid); | |
2687 | } | |
ed4a6d7f MG |
2688 | |
2689 | return count; | |
2690 | } | |
0825a6f9 | 2691 | static DEVICE_ATTR(compact, 0200, NULL, sysfs_compact_node); |
ed4a6d7f MG |
2692 | |
2693 | int compaction_register_node(struct node *node) | |
2694 | { | |
10fbcf4c | 2695 | return device_create_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
2696 | } |
2697 | ||
2698 | void compaction_unregister_node(struct node *node) | |
2699 | { | |
10fbcf4c | 2700 | return device_remove_file(&node->dev, &dev_attr_compact); |
ed4a6d7f MG |
2701 | } |
2702 | #endif /* CONFIG_SYSFS && CONFIG_NUMA */ | |
ff9543fd | 2703 | |
698b1b30 VB |
2704 | static inline bool kcompactd_work_requested(pg_data_t *pgdat) |
2705 | { | |
172400c6 | 2706 | return pgdat->kcompactd_max_order > 0 || kthread_should_stop(); |
698b1b30 VB |
2707 | } |
2708 | ||
2709 | static bool kcompactd_node_suitable(pg_data_t *pgdat) | |
2710 | { | |
2711 | int zoneid; | |
2712 | struct zone *zone; | |
97a225e6 | 2713 | enum zone_type highest_zoneidx = pgdat->kcompactd_highest_zoneidx; |
698b1b30 | 2714 | |
97a225e6 | 2715 | for (zoneid = 0; zoneid <= highest_zoneidx; zoneid++) { |
698b1b30 VB |
2716 | zone = &pgdat->node_zones[zoneid]; |
2717 | ||
2718 | if (!populated_zone(zone)) | |
2719 | continue; | |
2720 | ||
2721 | if (compaction_suitable(zone, pgdat->kcompactd_max_order, 0, | |
97a225e6 | 2722 | highest_zoneidx) == COMPACT_CONTINUE) |
698b1b30 VB |
2723 | return true; |
2724 | } | |
2725 | ||
2726 | return false; | |
2727 | } | |
2728 | ||
2729 | static void kcompactd_do_work(pg_data_t *pgdat) | |
2730 | { | |
2731 | /* | |
2732 | * With no special task, compact all zones so that a page of requested | |
2733 | * order is allocatable. | |
2734 | */ | |
2735 | int zoneid; | |
2736 | struct zone *zone; | |
2737 | struct compact_control cc = { | |
2738 | .order = pgdat->kcompactd_max_order, | |
dbe2d4e4 | 2739 | .search_order = pgdat->kcompactd_max_order, |
97a225e6 | 2740 | .highest_zoneidx = pgdat->kcompactd_highest_zoneidx, |
698b1b30 | 2741 | .mode = MIGRATE_SYNC_LIGHT, |
a0647dc9 | 2742 | .ignore_skip_hint = false, |
73e64c51 | 2743 | .gfp_mask = GFP_KERNEL, |
698b1b30 | 2744 | }; |
698b1b30 | 2745 | trace_mm_compaction_kcompactd_wake(pgdat->node_id, cc.order, |
97a225e6 | 2746 | cc.highest_zoneidx); |
7f354a54 | 2747 | count_compact_event(KCOMPACTD_WAKE); |
698b1b30 | 2748 | |
97a225e6 | 2749 | for (zoneid = 0; zoneid <= cc.highest_zoneidx; zoneid++) { |
698b1b30 VB |
2750 | int status; |
2751 | ||
2752 | zone = &pgdat->node_zones[zoneid]; | |
2753 | if (!populated_zone(zone)) | |
2754 | continue; | |
2755 | ||
2756 | if (compaction_deferred(zone, cc.order)) | |
2757 | continue; | |
2758 | ||
2759 | if (compaction_suitable(zone, cc.order, 0, zoneid) != | |
2760 | COMPACT_CONTINUE) | |
2761 | continue; | |
2762 | ||
172400c6 VB |
2763 | if (kthread_should_stop()) |
2764 | return; | |
a94b5252 YS |
2765 | |
2766 | cc.zone = zone; | |
5e1f0f09 | 2767 | status = compact_zone(&cc, NULL); |
698b1b30 | 2768 | |
7ceb009a | 2769 | if (status == COMPACT_SUCCESS) { |
698b1b30 | 2770 | compaction_defer_reset(zone, cc.order, false); |
c8f7de0b | 2771 | } else if (status == COMPACT_PARTIAL_SKIPPED || status == COMPACT_COMPLETE) { |
bc3106b2 DR |
2772 | /* |
2773 | * Buddy pages may become stranded on pcps that could | |
2774 | * otherwise coalesce on the zone's free area for | |
2775 | * order >= cc.order. This is ratelimited by the | |
2776 | * upcoming deferral. | |
2777 | */ | |
2778 | drain_all_pages(zone); | |
2779 | ||
698b1b30 VB |
2780 | /* |
2781 | * We use sync migration mode here, so we defer like | |
2782 | * sync direct compaction does. | |
2783 | */ | |
2784 | defer_compaction(zone, cc.order); | |
2785 | } | |
2786 | ||
7f354a54 DR |
2787 | count_compact_events(KCOMPACTD_MIGRATE_SCANNED, |
2788 | cc.total_migrate_scanned); | |
2789 | count_compact_events(KCOMPACTD_FREE_SCANNED, | |
2790 | cc.total_free_scanned); | |
2791 | ||
698b1b30 VB |
2792 | VM_BUG_ON(!list_empty(&cc.freepages)); |
2793 | VM_BUG_ON(!list_empty(&cc.migratepages)); | |
2794 | } | |
2795 | ||
2796 | /* | |
2797 | * Regardless of success, we are done until woken up next. But remember | |
97a225e6 JK |
2798 | * the requested order/highest_zoneidx in case it was higher/tighter |
2799 | * than our current ones | |
698b1b30 VB |
2800 | */ |
2801 | if (pgdat->kcompactd_max_order <= cc.order) | |
2802 | pgdat->kcompactd_max_order = 0; | |
97a225e6 JK |
2803 | if (pgdat->kcompactd_highest_zoneidx >= cc.highest_zoneidx) |
2804 | pgdat->kcompactd_highest_zoneidx = pgdat->nr_zones - 1; | |
698b1b30 VB |
2805 | } |
2806 | ||
97a225e6 | 2807 | void wakeup_kcompactd(pg_data_t *pgdat, int order, int highest_zoneidx) |
698b1b30 VB |
2808 | { |
2809 | if (!order) | |
2810 | return; | |
2811 | ||
2812 | if (pgdat->kcompactd_max_order < order) | |
2813 | pgdat->kcompactd_max_order = order; | |
2814 | ||
97a225e6 JK |
2815 | if (pgdat->kcompactd_highest_zoneidx > highest_zoneidx) |
2816 | pgdat->kcompactd_highest_zoneidx = highest_zoneidx; | |
698b1b30 | 2817 | |
6818600f DB |
2818 | /* |
2819 | * Pairs with implicit barrier in wait_event_freezable() | |
2820 | * such that wakeups are not missed. | |
2821 | */ | |
2822 | if (!wq_has_sleeper(&pgdat->kcompactd_wait)) | |
698b1b30 VB |
2823 | return; |
2824 | ||
2825 | if (!kcompactd_node_suitable(pgdat)) | |
2826 | return; | |
2827 | ||
2828 | trace_mm_compaction_wakeup_kcompactd(pgdat->node_id, order, | |
97a225e6 | 2829 | highest_zoneidx); |
698b1b30 VB |
2830 | wake_up_interruptible(&pgdat->kcompactd_wait); |
2831 | } | |
2832 | ||
2833 | /* | |
2834 | * The background compaction daemon, started as a kernel thread | |
2835 | * from the init process. | |
2836 | */ | |
2837 | static int kcompactd(void *p) | |
2838 | { | |
2839 | pg_data_t *pgdat = (pg_data_t*)p; | |
2840 | struct task_struct *tsk = current; | |
facdaa91 | 2841 | unsigned int proactive_defer = 0; |
698b1b30 VB |
2842 | |
2843 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); | |
2844 | ||
2845 | if (!cpumask_empty(cpumask)) | |
2846 | set_cpus_allowed_ptr(tsk, cpumask); | |
2847 | ||
2848 | set_freezable(); | |
2849 | ||
2850 | pgdat->kcompactd_max_order = 0; | |
97a225e6 | 2851 | pgdat->kcompactd_highest_zoneidx = pgdat->nr_zones - 1; |
698b1b30 VB |
2852 | |
2853 | while (!kthread_should_stop()) { | |
eb414681 JW |
2854 | unsigned long pflags; |
2855 | ||
698b1b30 | 2856 | trace_mm_compaction_kcompactd_sleep(pgdat->node_id); |
facdaa91 NG |
2857 | if (wait_event_freezable_timeout(pgdat->kcompactd_wait, |
2858 | kcompactd_work_requested(pgdat), | |
2859 | msecs_to_jiffies(HPAGE_FRAG_CHECK_INTERVAL_MSEC))) { | |
2860 | ||
2861 | psi_memstall_enter(&pflags); | |
2862 | kcompactd_do_work(pgdat); | |
2863 | psi_memstall_leave(&pflags); | |
2864 | continue; | |
2865 | } | |
698b1b30 | 2866 | |
facdaa91 NG |
2867 | /* kcompactd wait timeout */ |
2868 | if (should_proactive_compact_node(pgdat)) { | |
2869 | unsigned int prev_score, score; | |
2870 | ||
2871 | if (proactive_defer) { | |
2872 | proactive_defer--; | |
2873 | continue; | |
2874 | } | |
2875 | prev_score = fragmentation_score_node(pgdat); | |
2876 | proactive_compact_node(pgdat); | |
2877 | score = fragmentation_score_node(pgdat); | |
2878 | /* | |
2879 | * Defer proactive compaction if the fragmentation | |
2880 | * score did not go down i.e. no progress made. | |
2881 | */ | |
2882 | proactive_defer = score < prev_score ? | |
2883 | 0 : 1 << COMPACT_MAX_DEFER_SHIFT; | |
2884 | } | |
698b1b30 VB |
2885 | } |
2886 | ||
2887 | return 0; | |
2888 | } | |
2889 | ||
2890 | /* | |
2891 | * This kcompactd start function will be called by init and node-hot-add. | |
2892 | * On node-hot-add, kcompactd will moved to proper cpus if cpus are hot-added. | |
2893 | */ | |
2894 | int kcompactd_run(int nid) | |
2895 | { | |
2896 | pg_data_t *pgdat = NODE_DATA(nid); | |
2897 | int ret = 0; | |
2898 | ||
2899 | if (pgdat->kcompactd) | |
2900 | return 0; | |
2901 | ||
2902 | pgdat->kcompactd = kthread_run(kcompactd, pgdat, "kcompactd%d", nid); | |
2903 | if (IS_ERR(pgdat->kcompactd)) { | |
2904 | pr_err("Failed to start kcompactd on node %d\n", nid); | |
2905 | ret = PTR_ERR(pgdat->kcompactd); | |
2906 | pgdat->kcompactd = NULL; | |
2907 | } | |
2908 | return ret; | |
2909 | } | |
2910 | ||
2911 | /* | |
2912 | * Called by memory hotplug when all memory in a node is offlined. Caller must | |
2913 | * hold mem_hotplug_begin/end(). | |
2914 | */ | |
2915 | void kcompactd_stop(int nid) | |
2916 | { | |
2917 | struct task_struct *kcompactd = NODE_DATA(nid)->kcompactd; | |
2918 | ||
2919 | if (kcompactd) { | |
2920 | kthread_stop(kcompactd); | |
2921 | NODE_DATA(nid)->kcompactd = NULL; | |
2922 | } | |
2923 | } | |
2924 | ||
2925 | /* | |
2926 | * It's optimal to keep kcompactd on the same CPUs as their memory, but | |
2927 | * not required for correctness. So if the last cpu in a node goes | |
2928 | * away, we get changed to run anywhere: as the first one comes back, | |
2929 | * restore their cpu bindings. | |
2930 | */ | |
e46b1db2 | 2931 | static int kcompactd_cpu_online(unsigned int cpu) |
698b1b30 VB |
2932 | { |
2933 | int nid; | |
2934 | ||
e46b1db2 AMG |
2935 | for_each_node_state(nid, N_MEMORY) { |
2936 | pg_data_t *pgdat = NODE_DATA(nid); | |
2937 | const struct cpumask *mask; | |
698b1b30 | 2938 | |
e46b1db2 | 2939 | mask = cpumask_of_node(pgdat->node_id); |
698b1b30 | 2940 | |
e46b1db2 AMG |
2941 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
2942 | /* One of our CPUs online: restore mask */ | |
2943 | set_cpus_allowed_ptr(pgdat->kcompactd, mask); | |
698b1b30 | 2944 | } |
e46b1db2 | 2945 | return 0; |
698b1b30 VB |
2946 | } |
2947 | ||
2948 | static int __init kcompactd_init(void) | |
2949 | { | |
2950 | int nid; | |
e46b1db2 AMG |
2951 | int ret; |
2952 | ||
2953 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, | |
2954 | "mm/compaction:online", | |
2955 | kcompactd_cpu_online, NULL); | |
2956 | if (ret < 0) { | |
2957 | pr_err("kcompactd: failed to register hotplug callbacks.\n"); | |
2958 | return ret; | |
2959 | } | |
698b1b30 VB |
2960 | |
2961 | for_each_node_state(nid, N_MEMORY) | |
2962 | kcompactd_run(nid); | |
698b1b30 VB |
2963 | return 0; |
2964 | } | |
2965 | subsys_initcall(kcompactd_init) | |
2966 | ||
ff9543fd | 2967 | #endif /* CONFIG_COMPACTION */ |