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 | ||
b01b2141 IM |
49 | /* Protecting only lru_rotate.pvec which requires disabling interrupts */ |
50 | struct lru_rotate { | |
51 | local_lock_t lock; | |
52 | struct pagevec pvec; | |
53 | }; | |
54 | static DEFINE_PER_CPU(struct lru_rotate, lru_rotate) = { | |
55 | .lock = INIT_LOCAL_LOCK(lock), | |
56 | }; | |
57 | ||
58 | /* | |
59 | * The following struct pagevec are grouped together because they are protected | |
60 | * by disabling preemption (and interrupts remain enabled). | |
61 | */ | |
62 | struct lru_pvecs { | |
63 | local_lock_t lock; | |
64 | struct pagevec lru_add; | |
65 | struct pagevec lru_deactivate_file; | |
66 | struct pagevec lru_deactivate; | |
67 | struct pagevec lru_lazyfree; | |
a4a921aa | 68 | #ifdef CONFIG_SMP |
b01b2141 | 69 | struct pagevec activate_page; |
a4a921aa | 70 | #endif |
b01b2141 IM |
71 | }; |
72 | static DEFINE_PER_CPU(struct lru_pvecs, lru_pvecs) = { | |
73 | .lock = INIT_LOCAL_LOCK(lock), | |
74 | }; | |
902aaed0 | 75 | |
b221385b AB |
76 | /* |
77 | * This path almost never happens for VM activity - pages are normally | |
78 | * freed via pagevecs. But it gets used by networking. | |
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 | } |
62906027 | 92 | __ClearPageWaiters(page); |
91807063 AA |
93 | } |
94 | ||
95 | static void __put_single_page(struct page *page) | |
96 | { | |
97 | __page_cache_release(page); | |
bbc6b703 | 98 | mem_cgroup_uncharge(page_folio(page)); |
44042b44 | 99 | free_unref_page(page, 0); |
b221385b AB |
100 | } |
101 | ||
91807063 | 102 | static void __put_compound_page(struct page *page) |
1da177e4 | 103 | { |
822fc613 NH |
104 | /* |
105 | * __page_cache_release() is supposed to be called for thp, not for | |
106 | * hugetlb. This is because hugetlb page does never have PageLRU set | |
107 | * (it's never listed to any LRU lists) and no memcg routines should | |
108 | * be called for hugetlb (it has a separate hugetlb_cgroup.) | |
109 | */ | |
110 | if (!PageHuge(page)) | |
111 | __page_cache_release(page); | |
ff45fc3c | 112 | destroy_compound_page(page); |
91807063 AA |
113 | } |
114 | ||
ddc58f27 | 115 | void __put_page(struct page *page) |
8519fb30 | 116 | { |
71389703 DW |
117 | if (is_zone_device_page(page)) { |
118 | put_dev_pagemap(page->pgmap); | |
119 | ||
120 | /* | |
121 | * The page belongs to the device that created pgmap. Do | |
122 | * not return it to page allocator. | |
123 | */ | |
124 | return; | |
125 | } | |
126 | ||
8519fb30 | 127 | if (unlikely(PageCompound(page))) |
ddc58f27 KS |
128 | __put_compound_page(page); |
129 | else | |
91807063 | 130 | __put_single_page(page); |
1da177e4 | 131 | } |
ddc58f27 | 132 | EXPORT_SYMBOL(__put_page); |
70b50f94 | 133 | |
1d7ea732 | 134 | /** |
7682486b RD |
135 | * put_pages_list() - release a list of pages |
136 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
137 | * |
138 | * Release a list of pages which are strung together on page.lru. Currently | |
139 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
140 | */ |
141 | void put_pages_list(struct list_head *pages) | |
142 | { | |
143 | while (!list_empty(pages)) { | |
144 | struct page *victim; | |
145 | ||
f86196ea | 146 | victim = lru_to_page(pages); |
1d7ea732 | 147 | list_del(&victim->lru); |
09cbfeaf | 148 | put_page(victim); |
1d7ea732 AZ |
149 | } |
150 | } | |
151 | EXPORT_SYMBOL(put_pages_list); | |
152 | ||
18022c5d MG |
153 | /* |
154 | * get_kernel_pages() - pin kernel pages in memory | |
155 | * @kiov: An array of struct kvec structures | |
156 | * @nr_segs: number of segments to pin | |
157 | * @write: pinning for read/write, currently ignored | |
158 | * @pages: array that receives pointers to the pages pinned. | |
159 | * Should be at least nr_segs long. | |
160 | * | |
161 | * Returns number of pages pinned. This may be fewer than the number | |
162 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
163 | * were pinned, returns -errno. Each page returned must be released | |
164 | * with a put_page() call when it is finished with. | |
165 | */ | |
166 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
167 | struct page **pages) | |
168 | { | |
169 | int seg; | |
170 | ||
171 | for (seg = 0; seg < nr_segs; seg++) { | |
172 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
173 | return seg; | |
174 | ||
5a178119 | 175 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
09cbfeaf | 176 | get_page(pages[seg]); |
18022c5d MG |
177 | } |
178 | ||
179 | return seg; | |
180 | } | |
181 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
182 | ||
3dd7ae8e | 183 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
c7c7b80c | 184 | void (*move_fn)(struct page *page, struct lruvec *lruvec)) |
902aaed0 HH |
185 | { |
186 | int i; | |
6168d0da | 187 | struct lruvec *lruvec = NULL; |
3dd7ae8e | 188 | unsigned long flags = 0; |
902aaed0 HH |
189 | |
190 | for (i = 0; i < pagevec_count(pvec); i++) { | |
191 | struct page *page = pvec->pages[i]; | |
0de340cb | 192 | struct folio *folio = page_folio(page); |
3dd7ae8e | 193 | |
fc574c23 AS |
194 | /* block memcg migration during page moving between lru */ |
195 | if (!TestClearPageLRU(page)) | |
196 | continue; | |
197 | ||
0de340cb | 198 | lruvec = folio_lruvec_relock_irqsave(folio, lruvec, &flags); |
c7c7b80c | 199 | (*move_fn)(page, lruvec); |
fc574c23 AS |
200 | |
201 | SetPageLRU(page); | |
902aaed0 | 202 | } |
6168d0da AS |
203 | if (lruvec) |
204 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
c6f92f9f | 205 | release_pages(pvec->pages, pvec->nr); |
83896fb5 | 206 | pagevec_reinit(pvec); |
d8505dee SL |
207 | } |
208 | ||
c7c7b80c | 209 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec) |
3dd7ae8e | 210 | { |
575ced1c MWO |
211 | struct folio *folio = page_folio(page); |
212 | ||
213 | if (!folio_test_unevictable(folio)) { | |
214 | lruvec_del_folio(lruvec, folio); | |
215 | folio_clear_active(folio); | |
216 | lruvec_add_folio_tail(lruvec, folio); | |
217 | __count_vm_events(PGROTATED, folio_nr_pages(folio)); | |
3dd7ae8e SL |
218 | } |
219 | } | |
220 | ||
d479960e MK |
221 | /* return true if pagevec needs to drain */ |
222 | static bool pagevec_add_and_need_flush(struct pagevec *pvec, struct page *page) | |
223 | { | |
224 | bool ret = false; | |
225 | ||
226 | if (!pagevec_add(pvec, page) || PageCompound(page) || | |
227 | lru_cache_disabled()) | |
228 | ret = true; | |
229 | ||
230 | return ret; | |
231 | } | |
232 | ||
1da177e4 | 233 | /* |
575ced1c MWO |
234 | * Writeback is about to end against a folio which has been marked for |
235 | * immediate reclaim. If it still appears to be reclaimable, move it | |
236 | * to the tail of the inactive list. | |
c7c7b80c | 237 | * |
575ced1c | 238 | * folio_rotate_reclaimable() must disable IRQs, to prevent nasty races. |
1da177e4 | 239 | */ |
575ced1c | 240 | void folio_rotate_reclaimable(struct folio *folio) |
1da177e4 | 241 | { |
575ced1c MWO |
242 | if (!folio_test_locked(folio) && !folio_test_dirty(folio) && |
243 | !folio_test_unevictable(folio) && folio_test_lru(folio)) { | |
ac6aadb2 MS |
244 | struct pagevec *pvec; |
245 | unsigned long flags; | |
246 | ||
575ced1c | 247 | folio_get(folio); |
b01b2141 IM |
248 | local_lock_irqsave(&lru_rotate.lock, flags); |
249 | pvec = this_cpu_ptr(&lru_rotate.pvec); | |
575ced1c | 250 | if (pagevec_add_and_need_flush(pvec, &folio->page)) |
c7c7b80c | 251 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); |
b01b2141 | 252 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
ac6aadb2 | 253 | } |
1da177e4 LT |
254 | } |
255 | ||
96f8bf4f | 256 | void lru_note_cost(struct lruvec *lruvec, bool file, unsigned int nr_pages) |
3e2f41f1 | 257 | { |
7cf111bc JW |
258 | do { |
259 | unsigned long lrusize; | |
260 | ||
6168d0da AS |
261 | /* |
262 | * Hold lruvec->lru_lock is safe here, since | |
263 | * 1) The pinned lruvec in reclaim, or | |
264 | * 2) From a pre-LRU page during refault (which also holds the | |
265 | * rcu lock, so would be safe even if the page was on the LRU | |
266 | * and could move simultaneously to a new lruvec). | |
267 | */ | |
268 | spin_lock_irq(&lruvec->lru_lock); | |
7cf111bc | 269 | /* Record cost event */ |
96f8bf4f JW |
270 | if (file) |
271 | lruvec->file_cost += nr_pages; | |
7cf111bc | 272 | else |
96f8bf4f | 273 | lruvec->anon_cost += nr_pages; |
7cf111bc JW |
274 | |
275 | /* | |
276 | * Decay previous events | |
277 | * | |
278 | * Because workloads change over time (and to avoid | |
279 | * overflow) we keep these statistics as a floating | |
280 | * average, which ends up weighing recent refaults | |
281 | * more than old ones. | |
282 | */ | |
283 | lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + | |
284 | lruvec_page_state(lruvec, NR_ACTIVE_ANON) + | |
285 | lruvec_page_state(lruvec, NR_INACTIVE_FILE) + | |
286 | lruvec_page_state(lruvec, NR_ACTIVE_FILE); | |
287 | ||
288 | if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { | |
289 | lruvec->file_cost /= 2; | |
290 | lruvec->anon_cost /= 2; | |
291 | } | |
6168d0da | 292 | spin_unlock_irq(&lruvec->lru_lock); |
7cf111bc | 293 | } while ((lruvec = parent_lruvec(lruvec))); |
3e2f41f1 KM |
294 | } |
295 | ||
96f8bf4f JW |
296 | void lru_note_cost_page(struct page *page) |
297 | { | |
b1baabd9 MWO |
298 | struct folio *folio = page_folio(page); |
299 | lru_note_cost(folio_lruvec(folio), | |
6c357848 | 300 | page_is_file_lru(page), thp_nr_pages(page)); |
96f8bf4f JW |
301 | } |
302 | ||
f2d27392 | 303 | static void __folio_activate(struct folio *folio, struct lruvec *lruvec) |
1da177e4 | 304 | { |
f2d27392 MWO |
305 | if (!folio_test_active(folio) && !folio_test_unevictable(folio)) { |
306 | long nr_pages = folio_nr_pages(folio); | |
744ed144 | 307 | |
f2d27392 MWO |
308 | lruvec_del_folio(lruvec, folio); |
309 | folio_set_active(folio); | |
310 | lruvec_add_folio(lruvec, folio); | |
311 | trace_mm_lru_activate(folio); | |
4f98a2fe | 312 | |
21e330fc SB |
313 | __count_vm_events(PGACTIVATE, nr_pages); |
314 | __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, | |
315 | nr_pages); | |
1da177e4 | 316 | } |
eb709b0d SL |
317 | } |
318 | ||
319 | #ifdef CONFIG_SMP | |
f2d27392 MWO |
320 | static void __activate_page(struct page *page, struct lruvec *lruvec) |
321 | { | |
322 | return __folio_activate(page_folio(page), lruvec); | |
323 | } | |
324 | ||
eb709b0d SL |
325 | static void activate_page_drain(int cpu) |
326 | { | |
b01b2141 | 327 | struct pagevec *pvec = &per_cpu(lru_pvecs.activate_page, cpu); |
eb709b0d SL |
328 | |
329 | if (pagevec_count(pvec)) | |
c7c7b80c | 330 | pagevec_lru_move_fn(pvec, __activate_page); |
eb709b0d SL |
331 | } |
332 | ||
5fbc4616 CM |
333 | static bool need_activate_page_drain(int cpu) |
334 | { | |
b01b2141 | 335 | return pagevec_count(&per_cpu(lru_pvecs.activate_page, cpu)) != 0; |
5fbc4616 CM |
336 | } |
337 | ||
f2d27392 | 338 | static void folio_activate(struct folio *folio) |
eb709b0d | 339 | { |
f2d27392 MWO |
340 | if (folio_test_lru(folio) && !folio_test_active(folio) && |
341 | !folio_test_unevictable(folio)) { | |
b01b2141 | 342 | struct pagevec *pvec; |
eb709b0d | 343 | |
f2d27392 | 344 | folio_get(folio); |
b01b2141 IM |
345 | local_lock(&lru_pvecs.lock); |
346 | pvec = this_cpu_ptr(&lru_pvecs.activate_page); | |
f2d27392 | 347 | if (pagevec_add_and_need_flush(pvec, &folio->page)) |
c7c7b80c | 348 | pagevec_lru_move_fn(pvec, __activate_page); |
b01b2141 | 349 | local_unlock(&lru_pvecs.lock); |
eb709b0d SL |
350 | } |
351 | } | |
352 | ||
353 | #else | |
354 | static inline void activate_page_drain(int cpu) | |
355 | { | |
356 | } | |
357 | ||
f2d27392 | 358 | static void folio_activate(struct folio *folio) |
eb709b0d | 359 | { |
6168d0da | 360 | struct lruvec *lruvec; |
eb709b0d | 361 | |
f2d27392 | 362 | if (folio_test_clear_lru(folio)) { |
e809c3fe | 363 | lruvec = folio_lruvec_lock_irq(folio); |
f2d27392 | 364 | __folio_activate(folio, lruvec); |
6168d0da | 365 | unlock_page_lruvec_irq(lruvec); |
f2d27392 | 366 | folio_set_lru(folio); |
6168d0da | 367 | } |
1da177e4 | 368 | } |
eb709b0d | 369 | #endif |
1da177e4 | 370 | |
059285a2 MG |
371 | static void __lru_cache_activate_page(struct page *page) |
372 | { | |
b01b2141 | 373 | struct pagevec *pvec; |
059285a2 MG |
374 | int i; |
375 | ||
b01b2141 IM |
376 | local_lock(&lru_pvecs.lock); |
377 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
378 | ||
059285a2 MG |
379 | /* |
380 | * Search backwards on the optimistic assumption that the page being | |
381 | * activated has just been added to this pagevec. Note that only | |
382 | * the local pagevec is examined as a !PageLRU page could be in the | |
383 | * process of being released, reclaimed, migrated or on a remote | |
384 | * pagevec that is currently being drained. Furthermore, marking | |
385 | * a remote pagevec's page PageActive potentially hits a race where | |
386 | * a page is marked PageActive just after it is added to the inactive | |
387 | * list causing accounting errors and BUG_ON checks to trigger. | |
388 | */ | |
389 | for (i = pagevec_count(pvec) - 1; i >= 0; i--) { | |
390 | struct page *pagevec_page = pvec->pages[i]; | |
391 | ||
392 | if (pagevec_page == page) { | |
393 | SetPageActive(page); | |
394 | break; | |
395 | } | |
396 | } | |
397 | ||
b01b2141 | 398 | local_unlock(&lru_pvecs.lock); |
059285a2 MG |
399 | } |
400 | ||
1da177e4 LT |
401 | /* |
402 | * Mark a page as having seen activity. | |
403 | * | |
404 | * inactive,unreferenced -> inactive,referenced | |
405 | * inactive,referenced -> active,unreferenced | |
406 | * active,unreferenced -> active,referenced | |
eb39d618 HD |
407 | * |
408 | * When a newly allocated page is not yet visible, so safe for non-atomic ops, | |
409 | * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). | |
1da177e4 | 410 | */ |
920c7a5d | 411 | void mark_page_accessed(struct page *page) |
1da177e4 | 412 | { |
e90309c9 | 413 | page = compound_head(page); |
059285a2 | 414 | |
a1100a74 FW |
415 | if (!PageReferenced(page)) { |
416 | SetPageReferenced(page); | |
417 | } else if (PageUnevictable(page)) { | |
418 | /* | |
419 | * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, | |
420 | * this list is never rotated or maintained, so marking an | |
421 | * evictable page accessed has no effect. | |
422 | */ | |
423 | } else if (!PageActive(page)) { | |
059285a2 MG |
424 | /* |
425 | * If the page is on the LRU, queue it for activation via | |
b01b2141 | 426 | * lru_pvecs.activate_page. Otherwise, assume the page is on a |
059285a2 MG |
427 | * pagevec, mark it active and it'll be moved to the active |
428 | * LRU on the next drain. | |
429 | */ | |
430 | if (PageLRU(page)) | |
f2d27392 | 431 | folio_activate(page_folio(page)); |
059285a2 MG |
432 | else |
433 | __lru_cache_activate_page(page); | |
1da177e4 | 434 | ClearPageReferenced(page); |
c5ce619a | 435 | workingset_activation(page_folio(page)); |
1da177e4 | 436 | } |
33c3fc71 VD |
437 | if (page_is_idle(page)) |
438 | clear_page_idle(page); | |
1da177e4 | 439 | } |
1da177e4 LT |
440 | EXPORT_SYMBOL(mark_page_accessed); |
441 | ||
f04e9ebb | 442 | /** |
c53954a0 | 443 | * lru_cache_add - add a page to a page list |
f04e9ebb | 444 | * @page: the page to be added to the LRU. |
2329d375 JZ |
445 | * |
446 | * Queue the page for addition to the LRU via pagevec. The decision on whether | |
447 | * to add the page to the [in]active [file|anon] list is deferred until the | |
448 | * pagevec is drained. This gives a chance for the caller of lru_cache_add() | |
449 | * have the page added to the active list using mark_page_accessed(). | |
f04e9ebb | 450 | */ |
c53954a0 | 451 | void lru_cache_add(struct page *page) |
1da177e4 | 452 | { |
6058eaec JW |
453 | struct pagevec *pvec; |
454 | ||
309381fe SL |
455 | VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); |
456 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6058eaec JW |
457 | |
458 | get_page(page); | |
459 | local_lock(&lru_pvecs.lock); | |
460 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
d479960e | 461 | if (pagevec_add_and_need_flush(pvec, page)) |
6058eaec JW |
462 | __pagevec_lru_add(pvec); |
463 | local_unlock(&lru_pvecs.lock); | |
1da177e4 | 464 | } |
6058eaec | 465 | EXPORT_SYMBOL(lru_cache_add); |
1da177e4 | 466 | |
00501b53 | 467 | /** |
b518154e | 468 | * lru_cache_add_inactive_or_unevictable |
00501b53 JW |
469 | * @page: the page to be added to LRU |
470 | * @vma: vma in which page is mapped for determining reclaimability | |
471 | * | |
b518154e | 472 | * Place @page on the inactive or unevictable LRU list, depending on its |
12eab428 | 473 | * evictability. |
00501b53 | 474 | */ |
b518154e | 475 | void lru_cache_add_inactive_or_unevictable(struct page *page, |
00501b53 JW |
476 | struct vm_area_struct *vma) |
477 | { | |
b518154e JK |
478 | bool unevictable; |
479 | ||
00501b53 JW |
480 | VM_BUG_ON_PAGE(PageLRU(page), page); |
481 | ||
b518154e JK |
482 | unevictable = (vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED; |
483 | if (unlikely(unevictable) && !TestSetPageMlocked(page)) { | |
0964730b | 484 | int nr_pages = thp_nr_pages(page); |
00501b53 | 485 | /* |
cb152a1a | 486 | * We use the irq-unsafe __mod_zone_page_state because this |
00501b53 JW |
487 | * counter is not modified from interrupt context, and the pte |
488 | * lock is held(spinlock), which implies preemption disabled. | |
489 | */ | |
0964730b HD |
490 | __mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages); |
491 | count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages); | |
00501b53 | 492 | } |
9c4e6b1a | 493 | lru_cache_add(page); |
00501b53 JW |
494 | } |
495 | ||
31560180 MK |
496 | /* |
497 | * If the page can not be invalidated, it is moved to the | |
498 | * inactive list to speed up its reclaim. It is moved to the | |
499 | * head of the list, rather than the tail, to give the flusher | |
500 | * threads some time to write it out, as this is much more | |
501 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
502 | * |
503 | * If the page isn't page_mapped and dirty/writeback, the page | |
504 | * could reclaim asap using PG_reclaim. | |
505 | * | |
506 | * 1. active, mapped page -> none | |
507 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
508 | * 3. inactive, mapped page -> none | |
509 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
510 | * 5. inactive, clean -> inactive, tail | |
511 | * 6. Others -> none | |
512 | * | |
513 | * In 4, why it moves inactive's head, the VM expects the page would | |
514 | * be write it out by flusher threads as this is much more effective | |
515 | * than the single-page writeout from reclaim. | |
31560180 | 516 | */ |
c7c7b80c | 517 | static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec) |
31560180 | 518 | { |
46ae6b2c | 519 | bool active = PageActive(page); |
6c357848 | 520 | int nr_pages = thp_nr_pages(page); |
31560180 | 521 | |
bad49d9c MK |
522 | if (PageUnevictable(page)) |
523 | return; | |
524 | ||
31560180 MK |
525 | /* Some processes are using the page */ |
526 | if (page_mapped(page)) | |
527 | return; | |
528 | ||
46ae6b2c | 529 | del_page_from_lru_list(page, lruvec); |
31560180 MK |
530 | ClearPageActive(page); |
531 | ClearPageReferenced(page); | |
31560180 | 532 | |
278df9f4 MK |
533 | if (PageWriteback(page) || PageDirty(page)) { |
534 | /* | |
535 | * PG_reclaim could be raced with end_page_writeback | |
536 | * It can make readahead confusing. But race window | |
537 | * is _really_ small and it's non-critical problem. | |
538 | */ | |
3a9c9788 | 539 | add_page_to_lru_list(page, lruvec); |
278df9f4 MK |
540 | SetPageReclaim(page); |
541 | } else { | |
542 | /* | |
543 | * The page's writeback ends up during pagevec | |
c4ffefd1 | 544 | * We move that page into tail of inactive. |
278df9f4 | 545 | */ |
3a9c9788 | 546 | add_page_to_lru_list_tail(page, lruvec); |
5d91f31f | 547 | __count_vm_events(PGROTATED, nr_pages); |
278df9f4 MK |
548 | } |
549 | ||
21e330fc | 550 | if (active) { |
5d91f31f | 551 | __count_vm_events(PGDEACTIVATE, nr_pages); |
21e330fc SB |
552 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, |
553 | nr_pages); | |
554 | } | |
31560180 MK |
555 | } |
556 | ||
c7c7b80c | 557 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec) |
9c276cc6 | 558 | { |
fc574c23 | 559 | if (PageActive(page) && !PageUnevictable(page)) { |
6c357848 | 560 | int nr_pages = thp_nr_pages(page); |
9c276cc6 | 561 | |
46ae6b2c | 562 | del_page_from_lru_list(page, lruvec); |
9c276cc6 MK |
563 | ClearPageActive(page); |
564 | ClearPageReferenced(page); | |
3a9c9788 | 565 | add_page_to_lru_list(page, lruvec); |
9c276cc6 | 566 | |
21e330fc SB |
567 | __count_vm_events(PGDEACTIVATE, nr_pages); |
568 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, | |
569 | nr_pages); | |
9c276cc6 MK |
570 | } |
571 | } | |
10853a03 | 572 | |
c7c7b80c | 573 | static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec) |
10853a03 | 574 | { |
fc574c23 | 575 | if (PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 576 | !PageSwapCache(page) && !PageUnevictable(page)) { |
6c357848 | 577 | int nr_pages = thp_nr_pages(page); |
10853a03 | 578 | |
46ae6b2c | 579 | del_page_from_lru_list(page, lruvec); |
10853a03 MK |
580 | ClearPageActive(page); |
581 | ClearPageReferenced(page); | |
f7ad2a6c | 582 | /* |
9de4f22a HY |
583 | * Lazyfree pages are clean anonymous pages. They have |
584 | * PG_swapbacked flag cleared, to distinguish them from normal | |
585 | * anonymous pages | |
f7ad2a6c SL |
586 | */ |
587 | ClearPageSwapBacked(page); | |
3a9c9788 | 588 | add_page_to_lru_list(page, lruvec); |
10853a03 | 589 | |
21e330fc SB |
590 | __count_vm_events(PGLAZYFREE, nr_pages); |
591 | __count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, | |
592 | nr_pages); | |
10853a03 MK |
593 | } |
594 | } | |
595 | ||
902aaed0 HH |
596 | /* |
597 | * Drain pages out of the cpu's pagevecs. | |
598 | * Either "cpu" is the current CPU, and preemption has already been | |
599 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
600 | */ | |
f0cb3c76 | 601 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 602 | { |
b01b2141 | 603 | struct pagevec *pvec = &per_cpu(lru_pvecs.lru_add, cpu); |
1da177e4 | 604 | |
13f7f789 | 605 | if (pagevec_count(pvec)) |
a0b8cab3 | 606 | __pagevec_lru_add(pvec); |
902aaed0 | 607 | |
b01b2141 | 608 | pvec = &per_cpu(lru_rotate.pvec, cpu); |
7e0cc01e QC |
609 | /* Disabling interrupts below acts as a compiler barrier. */ |
610 | if (data_race(pagevec_count(pvec))) { | |
902aaed0 HH |
611 | unsigned long flags; |
612 | ||
613 | /* No harm done if a racing interrupt already did this */ | |
b01b2141 | 614 | local_lock_irqsave(&lru_rotate.lock, flags); |
c7c7b80c | 615 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); |
b01b2141 | 616 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
902aaed0 | 617 | } |
31560180 | 618 | |
b01b2141 | 619 | pvec = &per_cpu(lru_pvecs.lru_deactivate_file, cpu); |
31560180 | 620 | if (pagevec_count(pvec)) |
c7c7b80c | 621 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); |
eb709b0d | 622 | |
b01b2141 | 623 | pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu); |
9c276cc6 | 624 | if (pagevec_count(pvec)) |
c7c7b80c | 625 | pagevec_lru_move_fn(pvec, lru_deactivate_fn); |
9c276cc6 | 626 | |
b01b2141 | 627 | pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu); |
10853a03 | 628 | if (pagevec_count(pvec)) |
c7c7b80c | 629 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn); |
10853a03 | 630 | |
eb709b0d | 631 | activate_page_drain(cpu); |
31560180 MK |
632 | } |
633 | ||
634 | /** | |
cc5993bd | 635 | * deactivate_file_page - forcefully deactivate a file page |
31560180 MK |
636 | * @page: page to deactivate |
637 | * | |
638 | * This function hints the VM that @page is a good reclaim candidate, | |
639 | * for example if its invalidation fails due to the page being dirty | |
640 | * or under writeback. | |
641 | */ | |
cc5993bd | 642 | void deactivate_file_page(struct page *page) |
31560180 | 643 | { |
821ed6bb | 644 | /* |
cc5993bd MK |
645 | * In a workload with many unevictable page such as mprotect, |
646 | * unevictable page deactivation for accelerating reclaim is pointless. | |
821ed6bb MK |
647 | */ |
648 | if (PageUnevictable(page)) | |
649 | return; | |
650 | ||
31560180 | 651 | if (likely(get_page_unless_zero(page))) { |
b01b2141 IM |
652 | struct pagevec *pvec; |
653 | ||
654 | local_lock(&lru_pvecs.lock); | |
655 | pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate_file); | |
31560180 | 656 | |
d479960e | 657 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 658 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); |
b01b2141 | 659 | local_unlock(&lru_pvecs.lock); |
31560180 | 660 | } |
80bfed90 AM |
661 | } |
662 | ||
9c276cc6 MK |
663 | /* |
664 | * deactivate_page - deactivate a page | |
665 | * @page: page to deactivate | |
666 | * | |
667 | * deactivate_page() moves @page to the inactive list if @page was on the active | |
668 | * list and was not an unevictable page. This is done to accelerate the reclaim | |
669 | * of @page. | |
670 | */ | |
671 | void deactivate_page(struct page *page) | |
672 | { | |
673 | if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { | |
b01b2141 | 674 | struct pagevec *pvec; |
9c276cc6 | 675 | |
b01b2141 IM |
676 | local_lock(&lru_pvecs.lock); |
677 | pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate); | |
9c276cc6 | 678 | get_page(page); |
d479960e | 679 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 680 | pagevec_lru_move_fn(pvec, lru_deactivate_fn); |
b01b2141 | 681 | local_unlock(&lru_pvecs.lock); |
9c276cc6 MK |
682 | } |
683 | } | |
684 | ||
10853a03 | 685 | /** |
f7ad2a6c | 686 | * mark_page_lazyfree - make an anon page lazyfree |
10853a03 MK |
687 | * @page: page to deactivate |
688 | * | |
f7ad2a6c SL |
689 | * mark_page_lazyfree() moves @page to the inactive file list. |
690 | * This is done to accelerate the reclaim of @page. | |
10853a03 | 691 | */ |
f7ad2a6c | 692 | void mark_page_lazyfree(struct page *page) |
10853a03 | 693 | { |
f7ad2a6c | 694 | if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 695 | !PageSwapCache(page) && !PageUnevictable(page)) { |
b01b2141 | 696 | struct pagevec *pvec; |
10853a03 | 697 | |
b01b2141 IM |
698 | local_lock(&lru_pvecs.lock); |
699 | pvec = this_cpu_ptr(&lru_pvecs.lru_lazyfree); | |
09cbfeaf | 700 | get_page(page); |
d479960e | 701 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 702 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn); |
b01b2141 | 703 | local_unlock(&lru_pvecs.lock); |
10853a03 MK |
704 | } |
705 | } | |
706 | ||
80bfed90 AM |
707 | void lru_add_drain(void) |
708 | { | |
b01b2141 IM |
709 | local_lock(&lru_pvecs.lock); |
710 | lru_add_drain_cpu(smp_processor_id()); | |
711 | local_unlock(&lru_pvecs.lock); | |
712 | } | |
713 | ||
243418e3 MK |
714 | /* |
715 | * It's called from per-cpu workqueue context in SMP case so | |
716 | * lru_add_drain_cpu and invalidate_bh_lrus_cpu should run on | |
717 | * the same cpu. It shouldn't be a problem in !SMP case since | |
718 | * the core is only one and the locks will disable preemption. | |
719 | */ | |
720 | static void lru_add_and_bh_lrus_drain(void) | |
721 | { | |
722 | local_lock(&lru_pvecs.lock); | |
723 | lru_add_drain_cpu(smp_processor_id()); | |
724 | local_unlock(&lru_pvecs.lock); | |
725 | invalidate_bh_lrus_cpu(); | |
726 | } | |
727 | ||
b01b2141 IM |
728 | void lru_add_drain_cpu_zone(struct zone *zone) |
729 | { | |
730 | local_lock(&lru_pvecs.lock); | |
731 | lru_add_drain_cpu(smp_processor_id()); | |
732 | drain_local_pages(zone); | |
733 | local_unlock(&lru_pvecs.lock); | |
1da177e4 LT |
734 | } |
735 | ||
6ea183d6 MH |
736 | #ifdef CONFIG_SMP |
737 | ||
738 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); | |
739 | ||
c4028958 | 740 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc | 741 | { |
243418e3 | 742 | lru_add_and_bh_lrus_drain(); |
053837fc NP |
743 | } |
744 | ||
9852a721 MH |
745 | /* |
746 | * Doesn't need any cpu hotplug locking because we do rely on per-cpu | |
747 | * kworkers being shut down before our page_alloc_cpu_dead callback is | |
748 | * executed on the offlined cpu. | |
749 | * Calling this function with cpu hotplug locks held can actually lead | |
750 | * to obscure indirect dependencies via WQ context. | |
751 | */ | |
d479960e | 752 | inline void __lru_add_drain_all(bool force_all_cpus) |
053837fc | 753 | { |
6446a513 AD |
754 | /* |
755 | * lru_drain_gen - Global pages generation number | |
756 | * | |
757 | * (A) Definition: global lru_drain_gen = x implies that all generations | |
758 | * 0 < n <= x are already *scheduled* for draining. | |
759 | * | |
760 | * This is an optimization for the highly-contended use case where a | |
761 | * user space workload keeps constantly generating a flow of pages for | |
762 | * each CPU. | |
763 | */ | |
764 | static unsigned int lru_drain_gen; | |
5fbc4616 | 765 | static struct cpumask has_work; |
6446a513 AD |
766 | static DEFINE_MUTEX(lock); |
767 | unsigned cpu, this_gen; | |
5fbc4616 | 768 | |
ce612879 MH |
769 | /* |
770 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
771 | * initialized. | |
772 | */ | |
773 | if (WARN_ON(!mm_percpu_wq)) | |
774 | return; | |
775 | ||
6446a513 AD |
776 | /* |
777 | * Guarantee pagevec counter stores visible by this CPU are visible to | |
778 | * other CPUs before loading the current drain generation. | |
779 | */ | |
780 | smp_mb(); | |
781 | ||
782 | /* | |
783 | * (B) Locally cache global LRU draining generation number | |
784 | * | |
785 | * The read barrier ensures that the counter is loaded before the mutex | |
786 | * is taken. It pairs with smp_mb() inside the mutex critical section | |
787 | * at (D). | |
788 | */ | |
789 | this_gen = smp_load_acquire(&lru_drain_gen); | |
eef1a429 | 790 | |
5fbc4616 | 791 | mutex_lock(&lock); |
eef1a429 KK |
792 | |
793 | /* | |
6446a513 AD |
794 | * (C) Exit the draining operation if a newer generation, from another |
795 | * lru_add_drain_all(), was already scheduled for draining. Check (A). | |
eef1a429 | 796 | */ |
d479960e | 797 | if (unlikely(this_gen != lru_drain_gen && !force_all_cpus)) |
eef1a429 KK |
798 | goto done; |
799 | ||
6446a513 AD |
800 | /* |
801 | * (D) Increment global generation number | |
802 | * | |
803 | * Pairs with smp_load_acquire() at (B), outside of the critical | |
804 | * section. Use a full memory barrier to guarantee that the new global | |
805 | * drain generation number is stored before loading pagevec counters. | |
806 | * | |
807 | * This pairing must be done here, before the for_each_online_cpu loop | |
808 | * below which drains the page vectors. | |
809 | * | |
810 | * Let x, y, and z represent some system CPU numbers, where x < y < z. | |
cb152a1a | 811 | * Assume CPU #z is in the middle of the for_each_online_cpu loop |
6446a513 AD |
812 | * below and has already reached CPU #y's per-cpu data. CPU #x comes |
813 | * along, adds some pages to its per-cpu vectors, then calls | |
814 | * lru_add_drain_all(). | |
815 | * | |
816 | * If the paired barrier is done at any later step, e.g. after the | |
817 | * loop, CPU #x will just exit at (C) and miss flushing out all of its | |
818 | * added pages. | |
819 | */ | |
820 | WRITE_ONCE(lru_drain_gen, lru_drain_gen + 1); | |
821 | smp_mb(); | |
eef1a429 | 822 | |
5fbc4616 | 823 | cpumask_clear(&has_work); |
5fbc4616 CM |
824 | for_each_online_cpu(cpu) { |
825 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); | |
826 | ||
d479960e MK |
827 | if (force_all_cpus || |
828 | pagevec_count(&per_cpu(lru_pvecs.lru_add, cpu)) || | |
7e0cc01e | 829 | data_race(pagevec_count(&per_cpu(lru_rotate.pvec, cpu))) || |
b01b2141 IM |
830 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) || |
831 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) || | |
832 | pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) || | |
8cc621d2 MK |
833 | need_activate_page_drain(cpu) || |
834 | has_bh_in_lru(cpu, NULL)) { | |
5fbc4616 | 835 | INIT_WORK(work, lru_add_drain_per_cpu); |
ce612879 | 836 | queue_work_on(cpu, mm_percpu_wq, work); |
6446a513 | 837 | __cpumask_set_cpu(cpu, &has_work); |
5fbc4616 CM |
838 | } |
839 | } | |
840 | ||
841 | for_each_cpu(cpu, &has_work) | |
842 | flush_work(&per_cpu(lru_add_drain_work, cpu)); | |
843 | ||
eef1a429 | 844 | done: |
5fbc4616 | 845 | mutex_unlock(&lock); |
053837fc | 846 | } |
d479960e MK |
847 | |
848 | void lru_add_drain_all(void) | |
849 | { | |
850 | __lru_add_drain_all(false); | |
851 | } | |
6ea183d6 MH |
852 | #else |
853 | void lru_add_drain_all(void) | |
854 | { | |
855 | lru_add_drain(); | |
856 | } | |
6446a513 | 857 | #endif /* CONFIG_SMP */ |
053837fc | 858 | |
d479960e MK |
859 | atomic_t lru_disable_count = ATOMIC_INIT(0); |
860 | ||
861 | /* | |
862 | * lru_cache_disable() needs to be called before we start compiling | |
863 | * a list of pages to be migrated using isolate_lru_page(). | |
864 | * It drains pages on LRU cache and then disable on all cpus until | |
865 | * lru_cache_enable is called. | |
866 | * | |
867 | * Must be paired with a call to lru_cache_enable(). | |
868 | */ | |
869 | void lru_cache_disable(void) | |
870 | { | |
871 | atomic_inc(&lru_disable_count); | |
872 | #ifdef CONFIG_SMP | |
873 | /* | |
874 | * lru_add_drain_all in the force mode will schedule draining on | |
875 | * all online CPUs so any calls of lru_cache_disabled wrapped by | |
876 | * local_lock or preemption disabled would be ordered by that. | |
877 | * The atomic operation doesn't need to have stronger ordering | |
878 | * requirements because that is enforeced by the scheduling | |
879 | * guarantees. | |
880 | */ | |
881 | __lru_add_drain_all(true); | |
882 | #else | |
243418e3 | 883 | lru_add_and_bh_lrus_drain(); |
d479960e MK |
884 | #endif |
885 | } | |
886 | ||
aabfb572 | 887 | /** |
ea1754a0 | 888 | * release_pages - batched put_page() |
aabfb572 MH |
889 | * @pages: array of pages to release |
890 | * @nr: number of pages | |
1da177e4 | 891 | * |
aabfb572 MH |
892 | * Decrement the reference count on all the pages in @pages. If it |
893 | * fell to zero, remove the page from the LRU and free it. | |
1da177e4 | 894 | */ |
c6f92f9f | 895 | void release_pages(struct page **pages, int nr) |
1da177e4 LT |
896 | { |
897 | int i; | |
cc59850e | 898 | LIST_HEAD(pages_to_free); |
6168d0da | 899 | struct lruvec *lruvec = NULL; |
0de340cb | 900 | unsigned long flags = 0; |
3f649ab7 | 901 | unsigned int lock_batch; |
1da177e4 | 902 | |
1da177e4 LT |
903 | for (i = 0; i < nr; i++) { |
904 | struct page *page = pages[i]; | |
0de340cb | 905 | struct folio *folio = page_folio(page); |
1da177e4 | 906 | |
aabfb572 MH |
907 | /* |
908 | * Make sure the IRQ-safe lock-holding time does not get | |
909 | * excessive with a continuous string of pages from the | |
6168d0da | 910 | * same lruvec. The lock is held only if lruvec != NULL. |
aabfb572 | 911 | */ |
6168d0da AS |
912 | if (lruvec && ++lock_batch == SWAP_CLUSTER_MAX) { |
913 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
914 | lruvec = NULL; | |
aabfb572 MH |
915 | } |
916 | ||
0de340cb | 917 | page = &folio->page; |
6fcb52a5 | 918 | if (is_huge_zero_page(page)) |
aa88b68c | 919 | continue; |
aa88b68c | 920 | |
c5d6c45e | 921 | if (is_zone_device_page(page)) { |
6168d0da AS |
922 | if (lruvec) { |
923 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
924 | lruvec = NULL; | |
df6ad698 | 925 | } |
c5d6c45e IW |
926 | /* |
927 | * ZONE_DEVICE pages that return 'false' from | |
a3e7bea0 | 928 | * page_is_devmap_managed() do not require special |
c5d6c45e IW |
929 | * processing, and instead, expect a call to |
930 | * put_page_testzero(). | |
931 | */ | |
07d80269 JH |
932 | if (page_is_devmap_managed(page)) { |
933 | put_devmap_managed_page(page); | |
c5d6c45e | 934 | continue; |
07d80269 | 935 | } |
43fbdeb3 RC |
936 | if (put_page_testzero(page)) |
937 | put_dev_pagemap(page->pgmap); | |
938 | continue; | |
df6ad698 JG |
939 | } |
940 | ||
b5810039 | 941 | if (!put_page_testzero(page)) |
1da177e4 LT |
942 | continue; |
943 | ||
ddc58f27 | 944 | if (PageCompound(page)) { |
6168d0da AS |
945 | if (lruvec) { |
946 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
947 | lruvec = NULL; | |
ddc58f27 KS |
948 | } |
949 | __put_compound_page(page); | |
950 | continue; | |
951 | } | |
952 | ||
46453a6e | 953 | if (PageLRU(page)) { |
2a5e4e34 AD |
954 | struct lruvec *prev_lruvec = lruvec; |
955 | ||
0de340cb | 956 | lruvec = folio_lruvec_relock_irqsave(folio, lruvec, |
2a5e4e34 AD |
957 | &flags); |
958 | if (prev_lruvec != lruvec) | |
aabfb572 | 959 | lock_batch = 0; |
fa9add64 | 960 | |
46ae6b2c | 961 | del_page_from_lru_list(page, lruvec); |
87560179 | 962 | __clear_page_lru_flags(page); |
46453a6e NP |
963 | } |
964 | ||
62906027 | 965 | __ClearPageWaiters(page); |
c53954a0 | 966 | |
cc59850e | 967 | list_add(&page->lru, &pages_to_free); |
1da177e4 | 968 | } |
6168d0da AS |
969 | if (lruvec) |
970 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
1da177e4 | 971 | |
747db954 | 972 | mem_cgroup_uncharge_list(&pages_to_free); |
2d4894b5 | 973 | free_unref_page_list(&pages_to_free); |
1da177e4 | 974 | } |
0be8557b | 975 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
976 | |
977 | /* | |
978 | * The pages which we're about to release may be in the deferred lru-addition | |
979 | * queues. That would prevent them from really being freed right now. That's | |
980 | * OK from a correctness point of view but is inefficient - those pages may be | |
981 | * cache-warm and we want to give them back to the page allocator ASAP. | |
982 | * | |
983 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
984 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
985 | * mutual recursion. | |
986 | */ | |
987 | void __pagevec_release(struct pagevec *pvec) | |
988 | { | |
7f0b5fb9 | 989 | if (!pvec->percpu_pvec_drained) { |
d9ed0d08 | 990 | lru_add_drain(); |
7f0b5fb9 | 991 | pvec->percpu_pvec_drained = true; |
d9ed0d08 | 992 | } |
c6f92f9f | 993 | release_pages(pvec->pages, pagevec_count(pvec)); |
1da177e4 LT |
994 | pagevec_reinit(pvec); |
995 | } | |
7f285701 SF |
996 | EXPORT_SYMBOL(__pagevec_release); |
997 | ||
c7c7b80c | 998 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec) |
3dd7ae8e | 999 | { |
9c4e6b1a | 1000 | int was_unevictable = TestClearPageUnevictable(page); |
6c357848 | 1001 | int nr_pages = thp_nr_pages(page); |
3dd7ae8e | 1002 | |
309381fe | 1003 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3dd7ae8e | 1004 | |
9c4e6b1a SB |
1005 | /* |
1006 | * Page becomes evictable in two ways: | |
dae966dc | 1007 | * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()]. |
9c4e6b1a SB |
1008 | * 2) Before acquiring LRU lock to put the page to correct LRU and then |
1009 | * a) do PageLRU check with lock [check_move_unevictable_pages] | |
1010 | * b) do PageLRU check before lock [clear_page_mlock] | |
1011 | * | |
1012 | * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need | |
1013 | * following strict ordering: | |
1014 | * | |
1015 | * #0: __pagevec_lru_add_fn #1: clear_page_mlock | |
1016 | * | |
1017 | * SetPageLRU() TestClearPageMlocked() | |
1018 | * smp_mb() // explicit ordering // above provides strict | |
1019 | * // ordering | |
1020 | * PageMlocked() PageLRU() | |
1021 | * | |
1022 | * | |
1023 | * if '#1' does not observe setting of PG_lru by '#0' and fails | |
1024 | * isolation, the explicit barrier will make sure that page_evictable | |
1025 | * check will put the page in correct LRU. Without smp_mb(), SetPageLRU | |
1026 | * can be reordered after PageMlocked check and can make '#1' to fail | |
1027 | * the isolation of the page whose Mlocked bit is cleared (#0 is also | |
1028 | * looking at the same page) and the evictable page will be stranded | |
1029 | * in an unevictable LRU. | |
1030 | */ | |
9a9b6cce YS |
1031 | SetPageLRU(page); |
1032 | smp_mb__after_atomic(); | |
9c4e6b1a SB |
1033 | |
1034 | if (page_evictable(page)) { | |
9c4e6b1a | 1035 | if (was_unevictable) |
5d91f31f | 1036 | __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); |
9c4e6b1a | 1037 | } else { |
9c4e6b1a SB |
1038 | ClearPageActive(page); |
1039 | SetPageUnevictable(page); | |
1040 | if (!was_unevictable) | |
5d91f31f | 1041 | __count_vm_events(UNEVICTABLE_PGCULLED, nr_pages); |
9c4e6b1a SB |
1042 | } |
1043 | ||
3a9c9788 | 1044 | add_page_to_lru_list(page, lruvec); |
86140453 | 1045 | trace_mm_lru_insertion(page); |
3dd7ae8e SL |
1046 | } |
1047 | ||
1da177e4 LT |
1048 | /* |
1049 | * Add the passed pages to the LRU, then drop the caller's refcount | |
1050 | * on them. Reinitialises the caller's pagevec. | |
1051 | */ | |
a0b8cab3 | 1052 | void __pagevec_lru_add(struct pagevec *pvec) |
1da177e4 | 1053 | { |
fc574c23 | 1054 | int i; |
6168d0da | 1055 | struct lruvec *lruvec = NULL; |
fc574c23 AS |
1056 | unsigned long flags = 0; |
1057 | ||
1058 | for (i = 0; i < pagevec_count(pvec); i++) { | |
1059 | struct page *page = pvec->pages[i]; | |
0de340cb | 1060 | struct folio *folio = page_folio(page); |
fc574c23 | 1061 | |
0de340cb | 1062 | lruvec = folio_lruvec_relock_irqsave(folio, lruvec, &flags); |
fc574c23 AS |
1063 | __pagevec_lru_add_fn(page, lruvec); |
1064 | } | |
6168d0da AS |
1065 | if (lruvec) |
1066 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
fc574c23 AS |
1067 | release_pages(pvec->pages, pvec->nr); |
1068 | pagevec_reinit(pvec); | |
1da177e4 | 1069 | } |
1da177e4 | 1070 | |
0cd6144a JW |
1071 | /** |
1072 | * pagevec_remove_exceptionals - pagevec exceptionals pruning | |
1073 | * @pvec: The pagevec to prune | |
1074 | * | |
a656a202 MWO |
1075 | * find_get_entries() fills both pages and XArray value entries (aka |
1076 | * exceptional entries) into the pagevec. This function prunes all | |
0cd6144a JW |
1077 | * exceptionals from @pvec without leaving holes, so that it can be |
1078 | * passed on to page-only pagevec operations. | |
1079 | */ | |
1080 | void pagevec_remove_exceptionals(struct pagevec *pvec) | |
1081 | { | |
1082 | int i, j; | |
1083 | ||
1084 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
1085 | struct page *page = pvec->pages[i]; | |
3159f943 | 1086 | if (!xa_is_value(page)) |
0cd6144a JW |
1087 | pvec->pages[j++] = page; |
1088 | } | |
1089 | pvec->nr = j; | |
1090 | } | |
1091 | ||
1da177e4 | 1092 | /** |
b947cee4 | 1093 | * pagevec_lookup_range - gang pagecache lookup |
1da177e4 LT |
1094 | * @pvec: Where the resulting pages are placed |
1095 | * @mapping: The address_space to search | |
1096 | * @start: The starting page index | |
b947cee4 | 1097 | * @end: The final page index |
1da177e4 | 1098 | * |
e02a9f04 | 1099 | * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE |
b947cee4 JK |
1100 | * pages in the mapping starting from index @start and upto index @end |
1101 | * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a | |
1da177e4 LT |
1102 | * reference against the pages in @pvec. |
1103 | * | |
1104 | * The search returns a group of mapping-contiguous pages with ascending | |
d72dc8a2 JK |
1105 | * indexes. There may be holes in the indices due to not-present pages. We |
1106 | * also update @start to index the next page for the traversal. | |
1da177e4 | 1107 | * |
b947cee4 | 1108 | * pagevec_lookup_range() returns the number of pages which were found. If this |
e02a9f04 | 1109 | * number is smaller than PAGEVEC_SIZE, the end of specified range has been |
b947cee4 | 1110 | * reached. |
1da177e4 | 1111 | */ |
b947cee4 | 1112 | unsigned pagevec_lookup_range(struct pagevec *pvec, |
397162ff | 1113 | struct address_space *mapping, pgoff_t *start, pgoff_t end) |
1da177e4 | 1114 | { |
397162ff | 1115 | pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, |
b947cee4 | 1116 | pvec->pages); |
1da177e4 LT |
1117 | return pagevec_count(pvec); |
1118 | } | |
b947cee4 | 1119 | EXPORT_SYMBOL(pagevec_lookup_range); |
78539fdf | 1120 | |
72b045ae JK |
1121 | unsigned pagevec_lookup_range_tag(struct pagevec *pvec, |
1122 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1123 | xa_mark_t tag) |
1da177e4 | 1124 | { |
72b045ae | 1125 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
67fd707f | 1126 | PAGEVEC_SIZE, pvec->pages); |
1da177e4 LT |
1127 | return pagevec_count(pvec); |
1128 | } | |
72b045ae | 1129 | EXPORT_SYMBOL(pagevec_lookup_range_tag); |
1da177e4 | 1130 | |
1da177e4 LT |
1131 | /* |
1132 | * Perform any setup for the swap system | |
1133 | */ | |
1134 | void __init swap_setup(void) | |
1135 | { | |
ca79b0c2 | 1136 | unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); |
e0bf68dd | 1137 | |
1da177e4 LT |
1138 | /* Use a smaller cluster for small-memory machines */ |
1139 | if (megs < 16) | |
1140 | page_cluster = 2; | |
1141 | else | |
1142 | page_cluster = 3; | |
1143 | /* | |
1144 | * Right now other parts of the system means that we | |
1145 | * _really_ don't want to cluster much more | |
1146 | */ | |
1da177e4 | 1147 | } |
07d80269 JH |
1148 | |
1149 | #ifdef CONFIG_DEV_PAGEMAP_OPS | |
1150 | void put_devmap_managed_page(struct page *page) | |
1151 | { | |
1152 | int count; | |
1153 | ||
1154 | if (WARN_ON_ONCE(!page_is_devmap_managed(page))) | |
1155 | return; | |
1156 | ||
1157 | count = page_ref_dec_return(page); | |
1158 | ||
1159 | /* | |
1160 | * devmap page refcounts are 1-based, rather than 0-based: if | |
1161 | * refcount is 1, then the page is free and the refcount is | |
1162 | * stable because nobody holds a reference on the page. | |
1163 | */ | |
1164 | if (count == 1) | |
1165 | free_devmap_managed_page(page); | |
1166 | else if (!count) | |
1167 | __put_page(page); | |
1168 | } | |
1169 | EXPORT_SYMBOL(put_devmap_managed_page); | |
1170 | #endif |