Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swap.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
183ff22b | 9 | * This file contains the default values for the operation of the |
1da177e4 | 10 | * Linux VM subsystem. Fine-tuning documentation can be found in |
57043247 | 11 | * Documentation/admin-guide/sysctl/vm.rst. |
1da177e4 LT |
12 | * Started 18.12.91 |
13 | * Swap aging added 23.2.95, Stephen Tweedie. | |
14 | * Buffermem limits added 12.3.98, Rik van Riel. | |
15 | */ | |
16 | ||
17 | #include <linux/mm.h> | |
18 | #include <linux/sched.h> | |
19 | #include <linux/kernel_stat.h> | |
20 | #include <linux/swap.h> | |
21 | #include <linux/mman.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/pagevec.h> | |
24 | #include <linux/init.h> | |
b95f1b31 | 25 | #include <linux/export.h> |
1da177e4 | 26 | #include <linux/mm_inline.h> |
1da177e4 | 27 | #include <linux/percpu_counter.h> |
3565fce3 | 28 | #include <linux/memremap.h> |
1da177e4 LT |
29 | #include <linux/percpu.h> |
30 | #include <linux/cpu.h> | |
31 | #include <linux/notifier.h> | |
e0bf68dd | 32 | #include <linux/backing-dev.h> |
66e1707b | 33 | #include <linux/memcontrol.h> |
5a0e3ad6 | 34 | #include <linux/gfp.h> |
a27bb332 | 35 | #include <linux/uio.h> |
822fc613 | 36 | #include <linux/hugetlb.h> |
33c3fc71 | 37 | #include <linux/page_idle.h> |
b01b2141 | 38 | #include <linux/local_lock.h> |
8cc621d2 | 39 | #include <linux/buffer_head.h> |
1da177e4 | 40 | |
64d6519d LS |
41 | #include "internal.h" |
42 | ||
c6286c98 MG |
43 | #define CREATE_TRACE_POINTS |
44 | #include <trace/events/pagemap.h> | |
45 | ||
1da177e4 LT |
46 | /* How many pages do we try to swap or page in/out together? */ |
47 | int page_cluster; | |
48 | ||
c2bc1681 | 49 | /* Protecting only lru_rotate.fbatch which requires disabling interrupts */ |
b01b2141 IM |
50 | struct lru_rotate { |
51 | local_lock_t lock; | |
c2bc1681 | 52 | struct folio_batch fbatch; |
b01b2141 IM |
53 | }; |
54 | static DEFINE_PER_CPU(struct lru_rotate, lru_rotate) = { | |
55 | .lock = INIT_LOCAL_LOCK(lock), | |
56 | }; | |
57 | ||
58 | /* | |
82ac64d8 | 59 | * The following folio batches are grouped together because they are protected |
b01b2141 IM |
60 | * by disabling preemption (and interrupts remain enabled). |
61 | */ | |
82ac64d8 | 62 | struct cpu_fbatches { |
b01b2141 | 63 | local_lock_t lock; |
70dea534 | 64 | struct folio_batch lru_add; |
7a3dbfe8 | 65 | struct folio_batch lru_deactivate_file; |
85cd7791 | 66 | struct folio_batch lru_deactivate; |
cec394ba | 67 | struct folio_batch lru_lazyfree; |
a4a921aa | 68 | #ifdef CONFIG_SMP |
3a44610b | 69 | struct folio_batch activate; |
a4a921aa | 70 | #endif |
b01b2141 | 71 | }; |
82ac64d8 | 72 | static DEFINE_PER_CPU(struct cpu_fbatches, cpu_fbatches) = { |
b01b2141 IM |
73 | .lock = INIT_LOCAL_LOCK(lock), |
74 | }; | |
902aaed0 | 75 | |
b221385b | 76 | /* |
b109b870 HD |
77 | * This path almost never happens for VM activity - pages are normally freed |
78 | * via pagevecs. But it gets used by networking - and for compound pages. | |
b221385b | 79 | */ |
920c7a5d | 80 | static void __page_cache_release(struct page *page) |
b221385b AB |
81 | { |
82 | if (PageLRU(page)) { | |
e809c3fe | 83 | struct folio *folio = page_folio(page); |
fa9add64 HD |
84 | struct lruvec *lruvec; |
85 | unsigned long flags; | |
b221385b | 86 | |
e809c3fe | 87 | lruvec = folio_lruvec_lock_irqsave(folio, &flags); |
46ae6b2c | 88 | del_page_from_lru_list(page, lruvec); |
87560179 | 89 | __clear_page_lru_flags(page); |
6168d0da | 90 | unlock_page_lruvec_irqrestore(lruvec, flags); |
b221385b | 91 | } |
b109b870 HD |
92 | /* See comment on PageMlocked in release_pages() */ |
93 | if (unlikely(PageMlocked(page))) { | |
94 | int nr_pages = thp_nr_pages(page); | |
95 | ||
96 | __ClearPageMlocked(page); | |
97 | mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages); | |
98 | count_vm_events(UNEVICTABLE_PGCLEARED, nr_pages); | |
99 | } | |
91807063 AA |
100 | } |
101 | ||
83d99659 | 102 | static void __folio_put_small(struct folio *folio) |
91807063 | 103 | { |
83d99659 MWO |
104 | __page_cache_release(&folio->page); |
105 | mem_cgroup_uncharge(folio); | |
106 | free_unref_page(&folio->page, 0); | |
b221385b AB |
107 | } |
108 | ||
91807063 | 109 | static void __put_compound_page(struct page *page) |
1da177e4 | 110 | { |
822fc613 NH |
111 | /* |
112 | * __page_cache_release() is supposed to be called for thp, not for | |
113 | * hugetlb. This is because hugetlb page does never have PageLRU set | |
114 | * (it's never listed to any LRU lists) and no memcg routines should | |
115 | * be called for hugetlb (it has a separate hugetlb_cgroup.) | |
116 | */ | |
117 | if (!PageHuge(page)) | |
118 | __page_cache_release(page); | |
ff45fc3c | 119 | destroy_compound_page(page); |
91807063 AA |
120 | } |
121 | ||
8d29c703 | 122 | void __folio_put(struct folio *folio) |
8519fb30 | 123 | { |
8d29c703 MWO |
124 | if (unlikely(folio_is_zone_device(folio))) |
125 | free_zone_device_page(&folio->page); | |
126 | else if (unlikely(folio_test_large(folio))) | |
127 | __put_compound_page(&folio->page); | |
ddc58f27 | 128 | else |
83d99659 | 129 | __folio_put_small(folio); |
1da177e4 | 130 | } |
8d29c703 | 131 | EXPORT_SYMBOL(__folio_put); |
70b50f94 | 132 | |
1d7ea732 | 133 | /** |
7682486b RD |
134 | * put_pages_list() - release a list of pages |
135 | * @pages: list of pages threaded on page->lru | |
1d7ea732 | 136 | * |
988c69f1 | 137 | * Release a list of pages which are strung together on page.lru. |
1d7ea732 AZ |
138 | */ |
139 | void put_pages_list(struct list_head *pages) | |
140 | { | |
2f58e5de | 141 | struct folio *folio, *next; |
988c69f1 | 142 | |
2f58e5de MWO |
143 | list_for_each_entry_safe(folio, next, pages, lru) { |
144 | if (!folio_put_testzero(folio)) { | |
145 | list_del(&folio->lru); | |
988c69f1 MWO |
146 | continue; |
147 | } | |
2f58e5de MWO |
148 | if (folio_test_large(folio)) { |
149 | list_del(&folio->lru); | |
150 | __put_compound_page(&folio->page); | |
988c69f1 MWO |
151 | continue; |
152 | } | |
2f58e5de | 153 | /* LRU flag must be clear because it's passed using the lru */ |
1d7ea732 | 154 | } |
988c69f1 MWO |
155 | |
156 | free_unref_page_list(pages); | |
3cd018b4 | 157 | INIT_LIST_HEAD(pages); |
1d7ea732 AZ |
158 | } |
159 | EXPORT_SYMBOL(put_pages_list); | |
160 | ||
18022c5d MG |
161 | /* |
162 | * get_kernel_pages() - pin kernel pages in memory | |
163 | * @kiov: An array of struct kvec structures | |
164 | * @nr_segs: number of segments to pin | |
165 | * @write: pinning for read/write, currently ignored | |
166 | * @pages: array that receives pointers to the pages pinned. | |
167 | * Should be at least nr_segs long. | |
168 | * | |
133d2743 ML |
169 | * Returns number of pages pinned. This may be fewer than the number requested. |
170 | * If nr_segs is 0 or negative, returns 0. If no pages were pinned, returns 0. | |
171 | * Each page returned must be released with a put_page() call when it is | |
172 | * finished with. | |
18022c5d MG |
173 | */ |
174 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
175 | struct page **pages) | |
176 | { | |
177 | int seg; | |
178 | ||
179 | for (seg = 0; seg < nr_segs; seg++) { | |
180 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
181 | return seg; | |
182 | ||
5a178119 | 183 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
09cbfeaf | 184 | get_page(pages[seg]); |
18022c5d MG |
185 | } |
186 | ||
187 | return seg; | |
188 | } | |
189 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
190 | ||
c2bc1681 MWO |
191 | typedef void (*move_fn_t)(struct lruvec *lruvec, struct folio *folio); |
192 | ||
70dea534 | 193 | static void lru_add_fn(struct lruvec *lruvec, struct folio *folio) |
7d80dd09 MWO |
194 | { |
195 | int was_unevictable = folio_test_clear_unevictable(folio); | |
196 | long nr_pages = folio_nr_pages(folio); | |
197 | ||
198 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); | |
199 | ||
7d80dd09 MWO |
200 | /* |
201 | * Is an smp_mb__after_atomic() still required here, before | |
202 | * folio_evictable() tests PageMlocked, to rule out the possibility | |
203 | * of stranding an evictable folio on an unevictable LRU? I think | |
204 | * not, because __munlock_page() only clears PageMlocked while the LRU | |
205 | * lock is held. | |
206 | * | |
207 | * (That is not true of __page_cache_release(), and not necessarily | |
208 | * true of release_pages(): but those only clear PageMlocked after | |
209 | * put_page_testzero() has excluded any other users of the page.) | |
210 | */ | |
211 | if (folio_evictable(folio)) { | |
212 | if (was_unevictable) | |
213 | __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); | |
214 | } else { | |
215 | folio_clear_active(folio); | |
216 | folio_set_unevictable(folio); | |
217 | /* | |
218 | * folio->mlock_count = !!folio_test_mlocked(folio)? | |
219 | * But that leaves __mlock_page() in doubt whether another | |
220 | * actor has already counted the mlock or not. Err on the | |
221 | * safe side, underestimate, let page reclaim fix it, rather | |
222 | * than leaving a page on the unevictable LRU indefinitely. | |
223 | */ | |
224 | folio->mlock_count = 0; | |
225 | if (!was_unevictable) | |
226 | __count_vm_events(UNEVICTABLE_PGCULLED, nr_pages); | |
227 | } | |
228 | ||
229 | lruvec_add_folio(lruvec, folio); | |
230 | trace_mm_lru_insertion(folio); | |
231 | } | |
232 | ||
c2bc1681 MWO |
233 | static void folio_batch_move_lru(struct folio_batch *fbatch, move_fn_t move_fn) |
234 | { | |
235 | int i; | |
236 | struct lruvec *lruvec = NULL; | |
237 | unsigned long flags = 0; | |
238 | ||
239 | for (i = 0; i < folio_batch_count(fbatch); i++) { | |
240 | struct folio *folio = fbatch->folios[i]; | |
241 | ||
242 | /* block memcg migration while the folio moves between lru */ | |
70dea534 | 243 | if (move_fn != lru_add_fn && !folio_test_clear_lru(folio)) |
c2bc1681 MWO |
244 | continue; |
245 | ||
246 | lruvec = folio_lruvec_relock_irqsave(folio, lruvec, &flags); | |
247 | move_fn(lruvec, folio); | |
248 | ||
249 | folio_set_lru(folio); | |
250 | } | |
251 | ||
252 | if (lruvec) | |
253 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
254 | folios_put(fbatch->folios, folio_batch_count(fbatch)); | |
255 | folio_batch_init(fbatch); | |
256 | } | |
257 | ||
258 | static void folio_batch_add_and_move(struct folio_batch *fbatch, | |
259 | struct folio *folio, move_fn_t move_fn) | |
260 | { | |
261 | if (folio_batch_add(fbatch, folio) && !folio_test_large(folio) && | |
262 | !lru_cache_disabled()) | |
263 | return; | |
264 | folio_batch_move_lru(fbatch, move_fn); | |
265 | } | |
266 | ||
267 | static void lru_move_tail_fn(struct lruvec *lruvec, struct folio *folio) | |
268 | { | |
269 | if (!folio_test_unevictable(folio)) { | |
270 | lruvec_del_folio(lruvec, folio); | |
271 | folio_clear_active(folio); | |
272 | lruvec_add_folio_tail(lruvec, folio); | |
273 | __count_vm_events(PGROTATED, folio_nr_pages(folio)); | |
274 | } | |
275 | } | |
276 | ||
1da177e4 | 277 | /* |
575ced1c MWO |
278 | * Writeback is about to end against a folio which has been marked for |
279 | * immediate reclaim. If it still appears to be reclaimable, move it | |
280 | * to the tail of the inactive list. | |
c7c7b80c | 281 | * |
575ced1c | 282 | * folio_rotate_reclaimable() must disable IRQs, to prevent nasty races. |
1da177e4 | 283 | */ |
575ced1c | 284 | void folio_rotate_reclaimable(struct folio *folio) |
1da177e4 | 285 | { |
575ced1c MWO |
286 | if (!folio_test_locked(folio) && !folio_test_dirty(folio) && |
287 | !folio_test_unevictable(folio) && folio_test_lru(folio)) { | |
c2bc1681 | 288 | struct folio_batch *fbatch; |
ac6aadb2 MS |
289 | unsigned long flags; |
290 | ||
575ced1c | 291 | folio_get(folio); |
b01b2141 | 292 | local_lock_irqsave(&lru_rotate.lock, flags); |
c2bc1681 MWO |
293 | fbatch = this_cpu_ptr(&lru_rotate.fbatch); |
294 | folio_batch_add_and_move(fbatch, folio, lru_move_tail_fn); | |
b01b2141 | 295 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
ac6aadb2 | 296 | } |
1da177e4 LT |
297 | } |
298 | ||
96f8bf4f | 299 | void lru_note_cost(struct lruvec *lruvec, bool file, unsigned int nr_pages) |
3e2f41f1 | 300 | { |
7cf111bc JW |
301 | do { |
302 | unsigned long lrusize; | |
303 | ||
6168d0da AS |
304 | /* |
305 | * Hold lruvec->lru_lock is safe here, since | |
306 | * 1) The pinned lruvec in reclaim, or | |
307 | * 2) From a pre-LRU page during refault (which also holds the | |
308 | * rcu lock, so would be safe even if the page was on the LRU | |
309 | * and could move simultaneously to a new lruvec). | |
310 | */ | |
311 | spin_lock_irq(&lruvec->lru_lock); | |
7cf111bc | 312 | /* Record cost event */ |
96f8bf4f JW |
313 | if (file) |
314 | lruvec->file_cost += nr_pages; | |
7cf111bc | 315 | else |
96f8bf4f | 316 | lruvec->anon_cost += nr_pages; |
7cf111bc JW |
317 | |
318 | /* | |
319 | * Decay previous events | |
320 | * | |
321 | * Because workloads change over time (and to avoid | |
322 | * overflow) we keep these statistics as a floating | |
323 | * average, which ends up weighing recent refaults | |
324 | * more than old ones. | |
325 | */ | |
326 | lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + | |
327 | lruvec_page_state(lruvec, NR_ACTIVE_ANON) + | |
328 | lruvec_page_state(lruvec, NR_INACTIVE_FILE) + | |
329 | lruvec_page_state(lruvec, NR_ACTIVE_FILE); | |
330 | ||
331 | if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { | |
332 | lruvec->file_cost /= 2; | |
333 | lruvec->anon_cost /= 2; | |
334 | } | |
6168d0da | 335 | spin_unlock_irq(&lruvec->lru_lock); |
7cf111bc | 336 | } while ((lruvec = parent_lruvec(lruvec))); |
3e2f41f1 KM |
337 | } |
338 | ||
0995d7e5 | 339 | void lru_note_cost_folio(struct folio *folio) |
96f8bf4f | 340 | { |
0995d7e5 MWO |
341 | lru_note_cost(folio_lruvec(folio), folio_is_file_lru(folio), |
342 | folio_nr_pages(folio)); | |
96f8bf4f JW |
343 | } |
344 | ||
3a44610b | 345 | static void folio_activate_fn(struct lruvec *lruvec, struct folio *folio) |
1da177e4 | 346 | { |
f2d27392 MWO |
347 | if (!folio_test_active(folio) && !folio_test_unevictable(folio)) { |
348 | long nr_pages = folio_nr_pages(folio); | |
744ed144 | 349 | |
f2d27392 MWO |
350 | lruvec_del_folio(lruvec, folio); |
351 | folio_set_active(folio); | |
352 | lruvec_add_folio(lruvec, folio); | |
353 | trace_mm_lru_activate(folio); | |
4f98a2fe | 354 | |
21e330fc SB |
355 | __count_vm_events(PGACTIVATE, nr_pages); |
356 | __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, | |
357 | nr_pages); | |
1da177e4 | 358 | } |
eb709b0d SL |
359 | } |
360 | ||
361 | #ifdef CONFIG_SMP | |
3a44610b | 362 | static void folio_activate_drain(int cpu) |
eb709b0d | 363 | { |
82ac64d8 | 364 | struct folio_batch *fbatch = &per_cpu(cpu_fbatches.activate, cpu); |
eb709b0d | 365 | |
3a44610b MWO |
366 | if (folio_batch_count(fbatch)) |
367 | folio_batch_move_lru(fbatch, folio_activate_fn); | |
5fbc4616 CM |
368 | } |
369 | ||
f2d27392 | 370 | static void folio_activate(struct folio *folio) |
eb709b0d | 371 | { |
f2d27392 MWO |
372 | if (folio_test_lru(folio) && !folio_test_active(folio) && |
373 | !folio_test_unevictable(folio)) { | |
3a44610b | 374 | struct folio_batch *fbatch; |
eb709b0d | 375 | |
f2d27392 | 376 | folio_get(folio); |
82ac64d8 MWO |
377 | local_lock(&cpu_fbatches.lock); |
378 | fbatch = this_cpu_ptr(&cpu_fbatches.activate); | |
3a44610b | 379 | folio_batch_add_and_move(fbatch, folio, folio_activate_fn); |
82ac64d8 | 380 | local_unlock(&cpu_fbatches.lock); |
eb709b0d SL |
381 | } |
382 | } | |
383 | ||
384 | #else | |
3a44610b | 385 | static inline void folio_activate_drain(int cpu) |
eb709b0d SL |
386 | { |
387 | } | |
388 | ||
f2d27392 | 389 | static void folio_activate(struct folio *folio) |
eb709b0d | 390 | { |
6168d0da | 391 | struct lruvec *lruvec; |
eb709b0d | 392 | |
f2d27392 | 393 | if (folio_test_clear_lru(folio)) { |
e809c3fe | 394 | lruvec = folio_lruvec_lock_irq(folio); |
3a44610b | 395 | folio_activate_fn(lruvec, folio); |
6168d0da | 396 | unlock_page_lruvec_irq(lruvec); |
f2d27392 | 397 | folio_set_lru(folio); |
6168d0da | 398 | } |
1da177e4 | 399 | } |
eb709b0d | 400 | #endif |
1da177e4 | 401 | |
76580b65 | 402 | static void __lru_cache_activate_folio(struct folio *folio) |
059285a2 | 403 | { |
70dea534 | 404 | struct folio_batch *fbatch; |
059285a2 MG |
405 | int i; |
406 | ||
82ac64d8 MWO |
407 | local_lock(&cpu_fbatches.lock); |
408 | fbatch = this_cpu_ptr(&cpu_fbatches.lru_add); | |
b01b2141 | 409 | |
059285a2 | 410 | /* |
70dea534 MWO |
411 | * Search backwards on the optimistic assumption that the folio being |
412 | * activated has just been added to this batch. Note that only | |
413 | * the local batch is examined as a !LRU folio could be in the | |
059285a2 | 414 | * process of being released, reclaimed, migrated or on a remote |
70dea534 MWO |
415 | * batch that is currently being drained. Furthermore, marking |
416 | * a remote batch's folio active potentially hits a race where | |
417 | * a folio is marked active just after it is added to the inactive | |
059285a2 MG |
418 | * list causing accounting errors and BUG_ON checks to trigger. |
419 | */ | |
70dea534 MWO |
420 | for (i = folio_batch_count(fbatch) - 1; i >= 0; i--) { |
421 | struct folio *batch_folio = fbatch->folios[i]; | |
059285a2 | 422 | |
70dea534 | 423 | if (batch_folio == folio) { |
76580b65 | 424 | folio_set_active(folio); |
059285a2 MG |
425 | break; |
426 | } | |
427 | } | |
428 | ||
82ac64d8 | 429 | local_unlock(&cpu_fbatches.lock); |
059285a2 MG |
430 | } |
431 | ||
1da177e4 LT |
432 | /* |
433 | * Mark a page as having seen activity. | |
434 | * | |
435 | * inactive,unreferenced -> inactive,referenced | |
436 | * inactive,referenced -> active,unreferenced | |
437 | * active,unreferenced -> active,referenced | |
eb39d618 HD |
438 | * |
439 | * When a newly allocated page is not yet visible, so safe for non-atomic ops, | |
440 | * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). | |
1da177e4 | 441 | */ |
76580b65 | 442 | void folio_mark_accessed(struct folio *folio) |
1da177e4 | 443 | { |
76580b65 MWO |
444 | if (!folio_test_referenced(folio)) { |
445 | folio_set_referenced(folio); | |
446 | } else if (folio_test_unevictable(folio)) { | |
a1100a74 FW |
447 | /* |
448 | * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, | |
449 | * this list is never rotated or maintained, so marking an | |
914c32e4 | 450 | * unevictable page accessed has no effect. |
a1100a74 | 451 | */ |
76580b65 | 452 | } else if (!folio_test_active(folio)) { |
059285a2 | 453 | /* |
3a44610b | 454 | * If the folio is on the LRU, queue it for activation via |
82ac64d8 | 455 | * cpu_fbatches.activate. Otherwise, assume the folio is in a |
3a44610b | 456 | * folio_batch, mark it active and it'll be moved to the active |
059285a2 MG |
457 | * LRU on the next drain. |
458 | */ | |
76580b65 MWO |
459 | if (folio_test_lru(folio)) |
460 | folio_activate(folio); | |
059285a2 | 461 | else |
76580b65 MWO |
462 | __lru_cache_activate_folio(folio); |
463 | folio_clear_referenced(folio); | |
464 | workingset_activation(folio); | |
1da177e4 | 465 | } |
76580b65 MWO |
466 | if (folio_test_idle(folio)) |
467 | folio_clear_idle(folio); | |
1da177e4 | 468 | } |
76580b65 | 469 | EXPORT_SYMBOL(folio_mark_accessed); |
1da177e4 | 470 | |
f04e9ebb | 471 | /** |
0d31125d MWO |
472 | * folio_add_lru - Add a folio to an LRU list. |
473 | * @folio: The folio to be added to the LRU. | |
2329d375 | 474 | * |
0d31125d | 475 | * Queue the folio for addition to the LRU. The decision on whether |
2329d375 | 476 | * to add the page to the [in]active [file|anon] list is deferred until the |
82ac64d8 | 477 | * folio_batch is drained. This gives a chance for the caller of folio_add_lru() |
0d31125d | 478 | * have the folio added to the active list using folio_mark_accessed(). |
f04e9ebb | 479 | */ |
0d31125d | 480 | void folio_add_lru(struct folio *folio) |
1da177e4 | 481 | { |
70dea534 | 482 | struct folio_batch *fbatch; |
6058eaec | 483 | |
70dea534 MWO |
484 | VM_BUG_ON_FOLIO(folio_test_active(folio) && |
485 | folio_test_unevictable(folio), folio); | |
0d31125d | 486 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
6058eaec | 487 | |
0d31125d | 488 | folio_get(folio); |
82ac64d8 MWO |
489 | local_lock(&cpu_fbatches.lock); |
490 | fbatch = this_cpu_ptr(&cpu_fbatches.lru_add); | |
70dea534 | 491 | folio_batch_add_and_move(fbatch, folio, lru_add_fn); |
82ac64d8 | 492 | local_unlock(&cpu_fbatches.lock); |
1da177e4 | 493 | } |
0d31125d | 494 | EXPORT_SYMBOL(folio_add_lru); |
1da177e4 | 495 | |
00501b53 | 496 | /** |
b518154e | 497 | * lru_cache_add_inactive_or_unevictable |
00501b53 JW |
498 | * @page: the page to be added to LRU |
499 | * @vma: vma in which page is mapped for determining reclaimability | |
500 | * | |
b518154e | 501 | * Place @page on the inactive or unevictable LRU list, depending on its |
12eab428 | 502 | * evictability. |
00501b53 | 503 | */ |
b518154e | 504 | void lru_cache_add_inactive_or_unevictable(struct page *page, |
00501b53 JW |
505 | struct vm_area_struct *vma) |
506 | { | |
507 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
508 | ||
2fbb0c10 HD |
509 | if (unlikely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED)) |
510 | mlock_new_page(page); | |
511 | else | |
512 | lru_cache_add(page); | |
00501b53 JW |
513 | } |
514 | ||
31560180 | 515 | /* |
7a3dbfe8 | 516 | * If the folio cannot be invalidated, it is moved to the |
31560180 MK |
517 | * inactive list to speed up its reclaim. It is moved to the |
518 | * head of the list, rather than the tail, to give the flusher | |
519 | * threads some time to write it out, as this is much more | |
520 | * effective than the single-page writeout from reclaim. | |
278df9f4 | 521 | * |
7a3dbfe8 MWO |
522 | * If the folio isn't mapped and dirty/writeback, the folio |
523 | * could be reclaimed asap using the reclaim flag. | |
278df9f4 | 524 | * |
7a3dbfe8 MWO |
525 | * 1. active, mapped folio -> none |
526 | * 2. active, dirty/writeback folio -> inactive, head, reclaim | |
527 | * 3. inactive, mapped folio -> none | |
528 | * 4. inactive, dirty/writeback folio -> inactive, head, reclaim | |
278df9f4 MK |
529 | * 5. inactive, clean -> inactive, tail |
530 | * 6. Others -> none | |
531 | * | |
7a3dbfe8 MWO |
532 | * In 4, it moves to the head of the inactive list so the folio is |
533 | * written out by flusher threads as this is much more efficient | |
278df9f4 | 534 | * than the single-page writeout from reclaim. |
31560180 | 535 | */ |
7a3dbfe8 | 536 | static void lru_deactivate_file_fn(struct lruvec *lruvec, struct folio *folio) |
31560180 | 537 | { |
7a3dbfe8 MWO |
538 | bool active = folio_test_active(folio); |
539 | long nr_pages = folio_nr_pages(folio); | |
31560180 | 540 | |
7a3dbfe8 | 541 | if (folio_test_unevictable(folio)) |
bad49d9c MK |
542 | return; |
543 | ||
7a3dbfe8 MWO |
544 | /* Some processes are using the folio */ |
545 | if (folio_mapped(folio)) | |
31560180 MK |
546 | return; |
547 | ||
7a3dbfe8 MWO |
548 | lruvec_del_folio(lruvec, folio); |
549 | folio_clear_active(folio); | |
550 | folio_clear_referenced(folio); | |
31560180 | 551 | |
7a3dbfe8 | 552 | if (folio_test_writeback(folio) || folio_test_dirty(folio)) { |
278df9f4 | 553 | /* |
7a3dbfe8 MWO |
554 | * Setting the reclaim flag could race with |
555 | * folio_end_writeback() and confuse readahead. But the | |
556 | * race window is _really_ small and it's not a critical | |
557 | * problem. | |
278df9f4 | 558 | */ |
7a3dbfe8 MWO |
559 | lruvec_add_folio(lruvec, folio); |
560 | folio_set_reclaim(folio); | |
278df9f4 MK |
561 | } else { |
562 | /* | |
7a3dbfe8 MWO |
563 | * The folio's writeback ended while it was in the batch. |
564 | * We move that folio to the tail of the inactive list. | |
278df9f4 | 565 | */ |
7a3dbfe8 | 566 | lruvec_add_folio_tail(lruvec, folio); |
5d91f31f | 567 | __count_vm_events(PGROTATED, nr_pages); |
278df9f4 MK |
568 | } |
569 | ||
21e330fc | 570 | if (active) { |
5d91f31f | 571 | __count_vm_events(PGDEACTIVATE, nr_pages); |
21e330fc SB |
572 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, |
573 | nr_pages); | |
574 | } | |
31560180 MK |
575 | } |
576 | ||
85cd7791 | 577 | static void lru_deactivate_fn(struct lruvec *lruvec, struct folio *folio) |
9c276cc6 | 578 | { |
85cd7791 MWO |
579 | if (folio_test_active(folio) && !folio_test_unevictable(folio)) { |
580 | long nr_pages = folio_nr_pages(folio); | |
9c276cc6 | 581 | |
85cd7791 MWO |
582 | lruvec_del_folio(lruvec, folio); |
583 | folio_clear_active(folio); | |
584 | folio_clear_referenced(folio); | |
585 | lruvec_add_folio(lruvec, folio); | |
9c276cc6 | 586 | |
21e330fc SB |
587 | __count_vm_events(PGDEACTIVATE, nr_pages); |
588 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, | |
589 | nr_pages); | |
9c276cc6 MK |
590 | } |
591 | } | |
10853a03 | 592 | |
cec394ba | 593 | static void lru_lazyfree_fn(struct lruvec *lruvec, struct folio *folio) |
10853a03 | 594 | { |
cec394ba MWO |
595 | if (folio_test_anon(folio) && folio_test_swapbacked(folio) && |
596 | !folio_test_swapcache(folio) && !folio_test_unevictable(folio)) { | |
597 | long nr_pages = folio_nr_pages(folio); | |
10853a03 | 598 | |
cec394ba MWO |
599 | lruvec_del_folio(lruvec, folio); |
600 | folio_clear_active(folio); | |
601 | folio_clear_referenced(folio); | |
f7ad2a6c | 602 | /* |
cec394ba MWO |
603 | * Lazyfree folios are clean anonymous folios. They have |
604 | * the swapbacked flag cleared, to distinguish them from normal | |
605 | * anonymous folios | |
f7ad2a6c | 606 | */ |
cec394ba MWO |
607 | folio_clear_swapbacked(folio); |
608 | lruvec_add_folio(lruvec, folio); | |
10853a03 | 609 | |
21e330fc SB |
610 | __count_vm_events(PGLAZYFREE, nr_pages); |
611 | __count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, | |
612 | nr_pages); | |
10853a03 MK |
613 | } |
614 | } | |
615 | ||
902aaed0 | 616 | /* |
82ac64d8 | 617 | * Drain pages out of the cpu's folio_batch. |
902aaed0 HH |
618 | * Either "cpu" is the current CPU, and preemption has already been |
619 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
620 | */ | |
f0cb3c76 | 621 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 622 | { |
a2d33b5d MWO |
623 | struct cpu_fbatches *fbatches = &per_cpu(cpu_fbatches, cpu); |
624 | struct folio_batch *fbatch = &fbatches->lru_add; | |
1da177e4 | 625 | |
70dea534 MWO |
626 | if (folio_batch_count(fbatch)) |
627 | folio_batch_move_lru(fbatch, lru_add_fn); | |
902aaed0 | 628 | |
c2bc1681 | 629 | fbatch = &per_cpu(lru_rotate.fbatch, cpu); |
7e0cc01e | 630 | /* Disabling interrupts below acts as a compiler barrier. */ |
c2bc1681 | 631 | if (data_race(folio_batch_count(fbatch))) { |
902aaed0 HH |
632 | unsigned long flags; |
633 | ||
634 | /* No harm done if a racing interrupt already did this */ | |
b01b2141 | 635 | local_lock_irqsave(&lru_rotate.lock, flags); |
c2bc1681 | 636 | folio_batch_move_lru(fbatch, lru_move_tail_fn); |
b01b2141 | 637 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
902aaed0 | 638 | } |
31560180 | 639 | |
a2d33b5d | 640 | fbatch = &fbatches->lru_deactivate_file; |
7a3dbfe8 MWO |
641 | if (folio_batch_count(fbatch)) |
642 | folio_batch_move_lru(fbatch, lru_deactivate_file_fn); | |
eb709b0d | 643 | |
a2d33b5d | 644 | fbatch = &fbatches->lru_deactivate; |
85cd7791 MWO |
645 | if (folio_batch_count(fbatch)) |
646 | folio_batch_move_lru(fbatch, lru_deactivate_fn); | |
9c276cc6 | 647 | |
a2d33b5d | 648 | fbatch = &fbatches->lru_lazyfree; |
cec394ba MWO |
649 | if (folio_batch_count(fbatch)) |
650 | folio_batch_move_lru(fbatch, lru_lazyfree_fn); | |
10853a03 | 651 | |
3a44610b | 652 | folio_activate_drain(cpu); |
31560180 MK |
653 | } |
654 | ||
655 | /** | |
7a3dbfe8 | 656 | * deactivate_file_folio() - Deactivate a file folio. |
261b6840 | 657 | * @folio: Folio to deactivate. |
31560180 | 658 | * |
261b6840 MWO |
659 | * This function hints to the VM that @folio is a good reclaim candidate, |
660 | * for example if its invalidation fails due to the folio being dirty | |
31560180 | 661 | * or under writeback. |
261b6840 | 662 | * |
7a3dbfe8 | 663 | * Context: Caller holds a reference on the folio. |
31560180 | 664 | */ |
261b6840 | 665 | void deactivate_file_folio(struct folio *folio) |
31560180 | 666 | { |
7a3dbfe8 | 667 | struct folio_batch *fbatch; |
261b6840 | 668 | |
7a3dbfe8 | 669 | /* Deactivating an unevictable folio will not accelerate reclaim */ |
261b6840 | 670 | if (folio_test_unevictable(folio)) |
821ed6bb MK |
671 | return; |
672 | ||
261b6840 | 673 | folio_get(folio); |
82ac64d8 MWO |
674 | local_lock(&cpu_fbatches.lock); |
675 | fbatch = this_cpu_ptr(&cpu_fbatches.lru_deactivate_file); | |
7a3dbfe8 | 676 | folio_batch_add_and_move(fbatch, folio, lru_deactivate_file_fn); |
82ac64d8 | 677 | local_unlock(&cpu_fbatches.lock); |
80bfed90 AM |
678 | } |
679 | ||
9c276cc6 MK |
680 | /* |
681 | * deactivate_page - deactivate a page | |
682 | * @page: page to deactivate | |
683 | * | |
684 | * deactivate_page() moves @page to the inactive list if @page was on the active | |
685 | * list and was not an unevictable page. This is done to accelerate the reclaim | |
686 | * of @page. | |
687 | */ | |
688 | void deactivate_page(struct page *page) | |
689 | { | |
85cd7791 | 690 | struct folio *folio = page_folio(page); |
9c276cc6 | 691 | |
85cd7791 MWO |
692 | if (folio_test_lru(folio) && folio_test_active(folio) && |
693 | !folio_test_unevictable(folio)) { | |
694 | struct folio_batch *fbatch; | |
695 | ||
696 | folio_get(folio); | |
82ac64d8 MWO |
697 | local_lock(&cpu_fbatches.lock); |
698 | fbatch = this_cpu_ptr(&cpu_fbatches.lru_deactivate); | |
85cd7791 | 699 | folio_batch_add_and_move(fbatch, folio, lru_deactivate_fn); |
82ac64d8 | 700 | local_unlock(&cpu_fbatches.lock); |
9c276cc6 MK |
701 | } |
702 | } | |
703 | ||
10853a03 | 704 | /** |
f7ad2a6c | 705 | * mark_page_lazyfree - make an anon page lazyfree |
10853a03 MK |
706 | * @page: page to deactivate |
707 | * | |
f7ad2a6c SL |
708 | * mark_page_lazyfree() moves @page to the inactive file list. |
709 | * This is done to accelerate the reclaim of @page. | |
10853a03 | 710 | */ |
f7ad2a6c | 711 | void mark_page_lazyfree(struct page *page) |
10853a03 | 712 | { |
cec394ba | 713 | struct folio *folio = page_folio(page); |
10853a03 | 714 | |
cec394ba MWO |
715 | if (folio_test_lru(folio) && folio_test_anon(folio) && |
716 | folio_test_swapbacked(folio) && !folio_test_swapcache(folio) && | |
717 | !folio_test_unevictable(folio)) { | |
718 | struct folio_batch *fbatch; | |
719 | ||
720 | folio_get(folio); | |
82ac64d8 MWO |
721 | local_lock(&cpu_fbatches.lock); |
722 | fbatch = this_cpu_ptr(&cpu_fbatches.lru_lazyfree); | |
cec394ba | 723 | folio_batch_add_and_move(fbatch, folio, lru_lazyfree_fn); |
82ac64d8 | 724 | local_unlock(&cpu_fbatches.lock); |
10853a03 MK |
725 | } |
726 | } | |
727 | ||
80bfed90 AM |
728 | void lru_add_drain(void) |
729 | { | |
82ac64d8 | 730 | local_lock(&cpu_fbatches.lock); |
b01b2141 | 731 | lru_add_drain_cpu(smp_processor_id()); |
82ac64d8 | 732 | local_unlock(&cpu_fbatches.lock); |
adb11e78 | 733 | mlock_page_drain_local(); |
b01b2141 IM |
734 | } |
735 | ||
243418e3 MK |
736 | /* |
737 | * It's called from per-cpu workqueue context in SMP case so | |
738 | * lru_add_drain_cpu and invalidate_bh_lrus_cpu should run on | |
739 | * the same cpu. It shouldn't be a problem in !SMP case since | |
740 | * the core is only one and the locks will disable preemption. | |
741 | */ | |
742 | static void lru_add_and_bh_lrus_drain(void) | |
743 | { | |
82ac64d8 | 744 | local_lock(&cpu_fbatches.lock); |
243418e3 | 745 | lru_add_drain_cpu(smp_processor_id()); |
82ac64d8 | 746 | local_unlock(&cpu_fbatches.lock); |
243418e3 | 747 | invalidate_bh_lrus_cpu(); |
adb11e78 | 748 | mlock_page_drain_local(); |
243418e3 MK |
749 | } |
750 | ||
b01b2141 IM |
751 | void lru_add_drain_cpu_zone(struct zone *zone) |
752 | { | |
82ac64d8 | 753 | local_lock(&cpu_fbatches.lock); |
b01b2141 IM |
754 | lru_add_drain_cpu(smp_processor_id()); |
755 | drain_local_pages(zone); | |
82ac64d8 | 756 | local_unlock(&cpu_fbatches.lock); |
adb11e78 | 757 | mlock_page_drain_local(); |
1da177e4 LT |
758 | } |
759 | ||
6ea183d6 MH |
760 | #ifdef CONFIG_SMP |
761 | ||
762 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); | |
763 | ||
c4028958 | 764 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc | 765 | { |
243418e3 | 766 | lru_add_and_bh_lrus_drain(); |
053837fc NP |
767 | } |
768 | ||
4864545a MWO |
769 | static bool cpu_needs_drain(unsigned int cpu) |
770 | { | |
771 | struct cpu_fbatches *fbatches = &per_cpu(cpu_fbatches, cpu); | |
772 | ||
773 | /* Check these in order of likelihood that they're not zero */ | |
774 | return folio_batch_count(&fbatches->lru_add) || | |
775 | data_race(folio_batch_count(&per_cpu(lru_rotate.fbatch, cpu))) || | |
776 | folio_batch_count(&fbatches->lru_deactivate_file) || | |
777 | folio_batch_count(&fbatches->lru_deactivate) || | |
778 | folio_batch_count(&fbatches->lru_lazyfree) || | |
779 | folio_batch_count(&fbatches->activate) || | |
780 | need_mlock_page_drain(cpu) || | |
781 | has_bh_in_lru(cpu, NULL); | |
782 | } | |
783 | ||
9852a721 MH |
784 | /* |
785 | * Doesn't need any cpu hotplug locking because we do rely on per-cpu | |
786 | * kworkers being shut down before our page_alloc_cpu_dead callback is | |
787 | * executed on the offlined cpu. | |
788 | * Calling this function with cpu hotplug locks held can actually lead | |
789 | * to obscure indirect dependencies via WQ context. | |
790 | */ | |
3db3264d | 791 | static inline void __lru_add_drain_all(bool force_all_cpus) |
053837fc | 792 | { |
6446a513 AD |
793 | /* |
794 | * lru_drain_gen - Global pages generation number | |
795 | * | |
796 | * (A) Definition: global lru_drain_gen = x implies that all generations | |
797 | * 0 < n <= x are already *scheduled* for draining. | |
798 | * | |
799 | * This is an optimization for the highly-contended use case where a | |
800 | * user space workload keeps constantly generating a flow of pages for | |
801 | * each CPU. | |
802 | */ | |
803 | static unsigned int lru_drain_gen; | |
5fbc4616 | 804 | static struct cpumask has_work; |
6446a513 AD |
805 | static DEFINE_MUTEX(lock); |
806 | unsigned cpu, this_gen; | |
5fbc4616 | 807 | |
ce612879 MH |
808 | /* |
809 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
810 | * initialized. | |
811 | */ | |
812 | if (WARN_ON(!mm_percpu_wq)) | |
813 | return; | |
814 | ||
6446a513 | 815 | /* |
82ac64d8 MWO |
816 | * Guarantee folio_batch counter stores visible by this CPU |
817 | * are visible to other CPUs before loading the current drain | |
818 | * generation. | |
6446a513 AD |
819 | */ |
820 | smp_mb(); | |
821 | ||
822 | /* | |
823 | * (B) Locally cache global LRU draining generation number | |
824 | * | |
825 | * The read barrier ensures that the counter is loaded before the mutex | |
826 | * is taken. It pairs with smp_mb() inside the mutex critical section | |
827 | * at (D). | |
828 | */ | |
829 | this_gen = smp_load_acquire(&lru_drain_gen); | |
eef1a429 | 830 | |
5fbc4616 | 831 | mutex_lock(&lock); |
eef1a429 KK |
832 | |
833 | /* | |
6446a513 AD |
834 | * (C) Exit the draining operation if a newer generation, from another |
835 | * lru_add_drain_all(), was already scheduled for draining. Check (A). | |
eef1a429 | 836 | */ |
d479960e | 837 | if (unlikely(this_gen != lru_drain_gen && !force_all_cpus)) |
eef1a429 KK |
838 | goto done; |
839 | ||
6446a513 AD |
840 | /* |
841 | * (D) Increment global generation number | |
842 | * | |
843 | * Pairs with smp_load_acquire() at (B), outside of the critical | |
82ac64d8 MWO |
844 | * section. Use a full memory barrier to guarantee that the |
845 | * new global drain generation number is stored before loading | |
846 | * folio_batch counters. | |
6446a513 AD |
847 | * |
848 | * This pairing must be done here, before the for_each_online_cpu loop | |
849 | * below which drains the page vectors. | |
850 | * | |
851 | * Let x, y, and z represent some system CPU numbers, where x < y < z. | |
cb152a1a | 852 | * Assume CPU #z is in the middle of the for_each_online_cpu loop |
6446a513 AD |
853 | * below and has already reached CPU #y's per-cpu data. CPU #x comes |
854 | * along, adds some pages to its per-cpu vectors, then calls | |
855 | * lru_add_drain_all(). | |
856 | * | |
857 | * If the paired barrier is done at any later step, e.g. after the | |
858 | * loop, CPU #x will just exit at (C) and miss flushing out all of its | |
859 | * added pages. | |
860 | */ | |
861 | WRITE_ONCE(lru_drain_gen, lru_drain_gen + 1); | |
862 | smp_mb(); | |
eef1a429 | 863 | |
5fbc4616 | 864 | cpumask_clear(&has_work); |
5fbc4616 CM |
865 | for_each_online_cpu(cpu) { |
866 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); | |
867 | ||
4864545a | 868 | if (cpu_needs_drain(cpu)) { |
5fbc4616 | 869 | INIT_WORK(work, lru_add_drain_per_cpu); |
ce612879 | 870 | queue_work_on(cpu, mm_percpu_wq, work); |
6446a513 | 871 | __cpumask_set_cpu(cpu, &has_work); |
5fbc4616 CM |
872 | } |
873 | } | |
874 | ||
875 | for_each_cpu(cpu, &has_work) | |
876 | flush_work(&per_cpu(lru_add_drain_work, cpu)); | |
877 | ||
eef1a429 | 878 | done: |
5fbc4616 | 879 | mutex_unlock(&lock); |
053837fc | 880 | } |
d479960e MK |
881 | |
882 | void lru_add_drain_all(void) | |
883 | { | |
884 | __lru_add_drain_all(false); | |
885 | } | |
6ea183d6 MH |
886 | #else |
887 | void lru_add_drain_all(void) | |
888 | { | |
889 | lru_add_drain(); | |
890 | } | |
6446a513 | 891 | #endif /* CONFIG_SMP */ |
053837fc | 892 | |
d479960e MK |
893 | atomic_t lru_disable_count = ATOMIC_INIT(0); |
894 | ||
895 | /* | |
896 | * lru_cache_disable() needs to be called before we start compiling | |
897 | * a list of pages to be migrated using isolate_lru_page(). | |
898 | * It drains pages on LRU cache and then disable on all cpus until | |
899 | * lru_cache_enable is called. | |
900 | * | |
901 | * Must be paired with a call to lru_cache_enable(). | |
902 | */ | |
903 | void lru_cache_disable(void) | |
904 | { | |
905 | atomic_inc(&lru_disable_count); | |
d479960e | 906 | /* |
ff042f4a MT |
907 | * Readers of lru_disable_count are protected by either disabling |
908 | * preemption or rcu_read_lock: | |
909 | * | |
910 | * preempt_disable, local_irq_disable [bh_lru_lock()] | |
911 | * rcu_read_lock [rt_spin_lock CONFIG_PREEMPT_RT] | |
912 | * preempt_disable [local_lock !CONFIG_PREEMPT_RT] | |
913 | * | |
914 | * Since v5.1 kernel, synchronize_rcu() is guaranteed to wait on | |
915 | * preempt_disable() regions of code. So any CPU which sees | |
916 | * lru_disable_count = 0 will have exited the critical | |
917 | * section when synchronize_rcu() returns. | |
d479960e | 918 | */ |
31733463 | 919 | synchronize_rcu_expedited(); |
ff042f4a | 920 | #ifdef CONFIG_SMP |
d479960e MK |
921 | __lru_add_drain_all(true); |
922 | #else | |
243418e3 | 923 | lru_add_and_bh_lrus_drain(); |
d479960e MK |
924 | #endif |
925 | } | |
926 | ||
aabfb572 | 927 | /** |
ea1754a0 | 928 | * release_pages - batched put_page() |
aabfb572 MH |
929 | * @pages: array of pages to release |
930 | * @nr: number of pages | |
1da177e4 | 931 | * |
aabfb572 MH |
932 | * Decrement the reference count on all the pages in @pages. If it |
933 | * fell to zero, remove the page from the LRU and free it. | |
1da177e4 | 934 | */ |
c6f92f9f | 935 | void release_pages(struct page **pages, int nr) |
1da177e4 LT |
936 | { |
937 | int i; | |
cc59850e | 938 | LIST_HEAD(pages_to_free); |
6168d0da | 939 | struct lruvec *lruvec = NULL; |
0de340cb | 940 | unsigned long flags = 0; |
3f649ab7 | 941 | unsigned int lock_batch; |
1da177e4 | 942 | |
1da177e4 | 943 | for (i = 0; i < nr; i++) { |
ab5e653e | 944 | struct folio *folio = page_folio(pages[i]); |
1da177e4 | 945 | |
aabfb572 MH |
946 | /* |
947 | * Make sure the IRQ-safe lock-holding time does not get | |
948 | * excessive with a continuous string of pages from the | |
6168d0da | 949 | * same lruvec. The lock is held only if lruvec != NULL. |
aabfb572 | 950 | */ |
6168d0da AS |
951 | if (lruvec && ++lock_batch == SWAP_CLUSTER_MAX) { |
952 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
953 | lruvec = NULL; | |
aabfb572 MH |
954 | } |
955 | ||
ab5e653e | 956 | if (is_huge_zero_page(&folio->page)) |
aa88b68c | 957 | continue; |
aa88b68c | 958 | |
ab5e653e | 959 | if (folio_is_zone_device(folio)) { |
6168d0da AS |
960 | if (lruvec) { |
961 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
962 | lruvec = NULL; | |
df6ad698 | 963 | } |
ab5e653e | 964 | if (put_devmap_managed_page(&folio->page)) |
c5d6c45e | 965 | continue; |
ab5e653e MWO |
966 | if (folio_put_testzero(folio)) |
967 | free_zone_device_page(&folio->page); | |
43fbdeb3 | 968 | continue; |
df6ad698 JG |
969 | } |
970 | ||
ab5e653e | 971 | if (!folio_put_testzero(folio)) |
1da177e4 LT |
972 | continue; |
973 | ||
ab5e653e | 974 | if (folio_test_large(folio)) { |
6168d0da AS |
975 | if (lruvec) { |
976 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
977 | lruvec = NULL; | |
ddc58f27 | 978 | } |
ab5e653e | 979 | __put_compound_page(&folio->page); |
ddc58f27 KS |
980 | continue; |
981 | } | |
982 | ||
ab5e653e | 983 | if (folio_test_lru(folio)) { |
2a5e4e34 AD |
984 | struct lruvec *prev_lruvec = lruvec; |
985 | ||
0de340cb | 986 | lruvec = folio_lruvec_relock_irqsave(folio, lruvec, |
2a5e4e34 AD |
987 | &flags); |
988 | if (prev_lruvec != lruvec) | |
aabfb572 | 989 | lock_batch = 0; |
fa9add64 | 990 | |
ab5e653e MWO |
991 | lruvec_del_folio(lruvec, folio); |
992 | __folio_clear_lru_flags(folio); | |
46453a6e NP |
993 | } |
994 | ||
b109b870 HD |
995 | /* |
996 | * In rare cases, when truncation or holepunching raced with | |
997 | * munlock after VM_LOCKED was cleared, Mlocked may still be | |
998 | * found set here. This does not indicate a problem, unless | |
999 | * "unevictable_pgs_cleared" appears worryingly large. | |
1000 | */ | |
ab5e653e MWO |
1001 | if (unlikely(folio_test_mlocked(folio))) { |
1002 | __folio_clear_mlocked(folio); | |
1003 | zone_stat_sub_folio(folio, NR_MLOCK); | |
b109b870 HD |
1004 | count_vm_event(UNEVICTABLE_PGCLEARED); |
1005 | } | |
1006 | ||
ab5e653e | 1007 | list_add(&folio->lru, &pages_to_free); |
1da177e4 | 1008 | } |
6168d0da AS |
1009 | if (lruvec) |
1010 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
1da177e4 | 1011 | |
747db954 | 1012 | mem_cgroup_uncharge_list(&pages_to_free); |
2d4894b5 | 1013 | free_unref_page_list(&pages_to_free); |
1da177e4 | 1014 | } |
0be8557b | 1015 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
1016 | |
1017 | /* | |
1018 | * The pages which we're about to release may be in the deferred lru-addition | |
1019 | * queues. That would prevent them from really being freed right now. That's | |
1020 | * OK from a correctness point of view but is inefficient - those pages may be | |
1021 | * cache-warm and we want to give them back to the page allocator ASAP. | |
1022 | * | |
70dea534 MWO |
1023 | * So __pagevec_release() will drain those queues here. |
1024 | * folio_batch_move_lru() calls folios_put() directly to avoid | |
1da177e4 LT |
1025 | * mutual recursion. |
1026 | */ | |
1027 | void __pagevec_release(struct pagevec *pvec) | |
1028 | { | |
7f0b5fb9 | 1029 | if (!pvec->percpu_pvec_drained) { |
d9ed0d08 | 1030 | lru_add_drain(); |
7f0b5fb9 | 1031 | pvec->percpu_pvec_drained = true; |
d9ed0d08 | 1032 | } |
c6f92f9f | 1033 | release_pages(pvec->pages, pagevec_count(pvec)); |
1da177e4 LT |
1034 | pagevec_reinit(pvec); |
1035 | } | |
7f285701 SF |
1036 | EXPORT_SYMBOL(__pagevec_release); |
1037 | ||
0cd6144a | 1038 | /** |
1613fac9 MWO |
1039 | * folio_batch_remove_exceptionals() - Prune non-folios from a batch. |
1040 | * @fbatch: The batch to prune | |
0cd6144a | 1041 | * |
1613fac9 MWO |
1042 | * find_get_entries() fills a batch with both folios and shadow/swap/DAX |
1043 | * entries. This function prunes all the non-folio entries from @fbatch | |
1044 | * without leaving holes, so that it can be passed on to folio-only batch | |
1045 | * operations. | |
0cd6144a | 1046 | */ |
1613fac9 | 1047 | void folio_batch_remove_exceptionals(struct folio_batch *fbatch) |
0cd6144a | 1048 | { |
1613fac9 | 1049 | unsigned int i, j; |
0cd6144a | 1050 | |
1613fac9 MWO |
1051 | for (i = 0, j = 0; i < folio_batch_count(fbatch); i++) { |
1052 | struct folio *folio = fbatch->folios[i]; | |
1053 | if (!xa_is_value(folio)) | |
1054 | fbatch->folios[j++] = folio; | |
0cd6144a | 1055 | } |
1613fac9 | 1056 | fbatch->nr = j; |
0cd6144a JW |
1057 | } |
1058 | ||
1da177e4 | 1059 | /** |
b947cee4 | 1060 | * pagevec_lookup_range - gang pagecache lookup |
1da177e4 LT |
1061 | * @pvec: Where the resulting pages are placed |
1062 | * @mapping: The address_space to search | |
1063 | * @start: The starting page index | |
b947cee4 | 1064 | * @end: The final page index |
1da177e4 | 1065 | * |
e02a9f04 | 1066 | * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE |
b947cee4 JK |
1067 | * pages in the mapping starting from index @start and upto index @end |
1068 | * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a | |
1da177e4 LT |
1069 | * reference against the pages in @pvec. |
1070 | * | |
1071 | * The search returns a group of mapping-contiguous pages with ascending | |
d72dc8a2 JK |
1072 | * indexes. There may be holes in the indices due to not-present pages. We |
1073 | * also update @start to index the next page for the traversal. | |
1da177e4 | 1074 | * |
b947cee4 | 1075 | * pagevec_lookup_range() returns the number of pages which were found. If this |
e02a9f04 | 1076 | * number is smaller than PAGEVEC_SIZE, the end of specified range has been |
b947cee4 | 1077 | * reached. |
1da177e4 | 1078 | */ |
b947cee4 | 1079 | unsigned pagevec_lookup_range(struct pagevec *pvec, |
397162ff | 1080 | struct address_space *mapping, pgoff_t *start, pgoff_t end) |
1da177e4 | 1081 | { |
397162ff | 1082 | pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, |
b947cee4 | 1083 | pvec->pages); |
1da177e4 LT |
1084 | return pagevec_count(pvec); |
1085 | } | |
b947cee4 | 1086 | EXPORT_SYMBOL(pagevec_lookup_range); |
78539fdf | 1087 | |
72b045ae JK |
1088 | unsigned pagevec_lookup_range_tag(struct pagevec *pvec, |
1089 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1090 | xa_mark_t tag) |
1da177e4 | 1091 | { |
72b045ae | 1092 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
67fd707f | 1093 | PAGEVEC_SIZE, pvec->pages); |
1da177e4 LT |
1094 | return pagevec_count(pvec); |
1095 | } | |
72b045ae | 1096 | EXPORT_SYMBOL(pagevec_lookup_range_tag); |
1da177e4 | 1097 | |
1da177e4 LT |
1098 | /* |
1099 | * Perform any setup for the swap system | |
1100 | */ | |
1101 | void __init swap_setup(void) | |
1102 | { | |
ca79b0c2 | 1103 | unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); |
e0bf68dd | 1104 | |
1da177e4 LT |
1105 | /* Use a smaller cluster for small-memory machines */ |
1106 | if (megs < 16) | |
1107 | page_cluster = 2; | |
1108 | else | |
1109 | page_cluster = 3; | |
1110 | /* | |
1111 | * Right now other parts of the system means that we | |
1112 | * _really_ don't want to cluster much more | |
1113 | */ | |
1da177e4 | 1114 | } |