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> |
1da177e4 | 39 | |
64d6519d LS |
40 | #include "internal.h" |
41 | ||
c6286c98 MG |
42 | #define CREATE_TRACE_POINTS |
43 | #include <trace/events/pagemap.h> | |
44 | ||
1da177e4 LT |
45 | /* How many pages do we try to swap or page in/out together? */ |
46 | int page_cluster; | |
47 | ||
b01b2141 IM |
48 | /* Protecting only lru_rotate.pvec which requires disabling interrupts */ |
49 | struct lru_rotate { | |
50 | local_lock_t lock; | |
51 | struct pagevec pvec; | |
52 | }; | |
53 | static DEFINE_PER_CPU(struct lru_rotate, lru_rotate) = { | |
54 | .lock = INIT_LOCAL_LOCK(lock), | |
55 | }; | |
56 | ||
57 | /* | |
58 | * The following struct pagevec are grouped together because they are protected | |
59 | * by disabling preemption (and interrupts remain enabled). | |
60 | */ | |
61 | struct lru_pvecs { | |
62 | local_lock_t lock; | |
63 | struct pagevec lru_add; | |
64 | struct pagevec lru_deactivate_file; | |
65 | struct pagevec lru_deactivate; | |
66 | struct pagevec lru_lazyfree; | |
a4a921aa | 67 | #ifdef CONFIG_SMP |
b01b2141 | 68 | struct pagevec activate_page; |
a4a921aa | 69 | #endif |
b01b2141 IM |
70 | }; |
71 | static DEFINE_PER_CPU(struct lru_pvecs, lru_pvecs) = { | |
72 | .lock = INIT_LOCAL_LOCK(lock), | |
73 | }; | |
902aaed0 | 74 | |
b221385b AB |
75 | /* |
76 | * This path almost never happens for VM activity - pages are normally | |
77 | * freed via pagevecs. But it gets used by networking. | |
78 | */ | |
920c7a5d | 79 | static void __page_cache_release(struct page *page) |
b221385b AB |
80 | { |
81 | if (PageLRU(page)) { | |
fa9add64 HD |
82 | struct lruvec *lruvec; |
83 | unsigned long flags; | |
b221385b | 84 | |
6168d0da | 85 | lruvec = lock_page_lruvec_irqsave(page, &flags); |
46ae6b2c | 86 | del_page_from_lru_list(page, lruvec); |
87560179 | 87 | __clear_page_lru_flags(page); |
6168d0da | 88 | unlock_page_lruvec_irqrestore(lruvec, flags); |
b221385b | 89 | } |
62906027 | 90 | __ClearPageWaiters(page); |
91807063 AA |
91 | } |
92 | ||
93 | static void __put_single_page(struct page *page) | |
94 | { | |
95 | __page_cache_release(page); | |
7ae88534 | 96 | mem_cgroup_uncharge(page); |
2d4894b5 | 97 | free_unref_page(page); |
b221385b AB |
98 | } |
99 | ||
91807063 | 100 | static void __put_compound_page(struct page *page) |
1da177e4 | 101 | { |
822fc613 NH |
102 | /* |
103 | * __page_cache_release() is supposed to be called for thp, not for | |
104 | * hugetlb. This is because hugetlb page does never have PageLRU set | |
105 | * (it's never listed to any LRU lists) and no memcg routines should | |
106 | * be called for hugetlb (it has a separate hugetlb_cgroup.) | |
107 | */ | |
108 | if (!PageHuge(page)) | |
109 | __page_cache_release(page); | |
ff45fc3c | 110 | destroy_compound_page(page); |
91807063 AA |
111 | } |
112 | ||
ddc58f27 | 113 | void __put_page(struct page *page) |
8519fb30 | 114 | { |
71389703 DW |
115 | if (is_zone_device_page(page)) { |
116 | put_dev_pagemap(page->pgmap); | |
117 | ||
118 | /* | |
119 | * The page belongs to the device that created pgmap. Do | |
120 | * not return it to page allocator. | |
121 | */ | |
122 | return; | |
123 | } | |
124 | ||
8519fb30 | 125 | if (unlikely(PageCompound(page))) |
ddc58f27 KS |
126 | __put_compound_page(page); |
127 | else | |
91807063 | 128 | __put_single_page(page); |
1da177e4 | 129 | } |
ddc58f27 | 130 | EXPORT_SYMBOL(__put_page); |
70b50f94 | 131 | |
1d7ea732 | 132 | /** |
7682486b RD |
133 | * put_pages_list() - release a list of pages |
134 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
135 | * |
136 | * Release a list of pages which are strung together on page.lru. Currently | |
137 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
138 | */ |
139 | void put_pages_list(struct list_head *pages) | |
140 | { | |
141 | while (!list_empty(pages)) { | |
142 | struct page *victim; | |
143 | ||
f86196ea | 144 | victim = lru_to_page(pages); |
1d7ea732 | 145 | list_del(&victim->lru); |
09cbfeaf | 146 | put_page(victim); |
1d7ea732 AZ |
147 | } |
148 | } | |
149 | EXPORT_SYMBOL(put_pages_list); | |
150 | ||
18022c5d MG |
151 | /* |
152 | * get_kernel_pages() - pin kernel pages in memory | |
153 | * @kiov: An array of struct kvec structures | |
154 | * @nr_segs: number of segments to pin | |
155 | * @write: pinning for read/write, currently ignored | |
156 | * @pages: array that receives pointers to the pages pinned. | |
157 | * Should be at least nr_segs long. | |
158 | * | |
159 | * Returns number of pages pinned. This may be fewer than the number | |
160 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
161 | * were pinned, returns -errno. Each page returned must be released | |
162 | * with a put_page() call when it is finished with. | |
163 | */ | |
164 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
165 | struct page **pages) | |
166 | { | |
167 | int seg; | |
168 | ||
169 | for (seg = 0; seg < nr_segs; seg++) { | |
170 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
171 | return seg; | |
172 | ||
5a178119 | 173 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
09cbfeaf | 174 | get_page(pages[seg]); |
18022c5d MG |
175 | } |
176 | ||
177 | return seg; | |
178 | } | |
179 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
180 | ||
181 | /* | |
182 | * get_kernel_page() - pin a kernel page in memory | |
183 | * @start: starting kernel address | |
184 | * @write: pinning for read/write, currently ignored | |
185 | * @pages: array that receives pointer to the page pinned. | |
186 | * Must be at least nr_segs long. | |
187 | * | |
188 | * Returns 1 if page is pinned. If the page was not pinned, returns | |
189 | * -errno. The page returned must be released with a put_page() call | |
190 | * when it is finished with. | |
191 | */ | |
192 | int get_kernel_page(unsigned long start, int write, struct page **pages) | |
193 | { | |
194 | const struct kvec kiov = { | |
195 | .iov_base = (void *)start, | |
196 | .iov_len = PAGE_SIZE | |
197 | }; | |
198 | ||
199 | return get_kernel_pages(&kiov, 1, write, pages); | |
200 | } | |
201 | EXPORT_SYMBOL_GPL(get_kernel_page); | |
202 | ||
3dd7ae8e | 203 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
c7c7b80c | 204 | void (*move_fn)(struct page *page, struct lruvec *lruvec)) |
902aaed0 HH |
205 | { |
206 | int i; | |
6168d0da | 207 | struct lruvec *lruvec = NULL; |
3dd7ae8e | 208 | unsigned long flags = 0; |
902aaed0 HH |
209 | |
210 | for (i = 0; i < pagevec_count(pvec); i++) { | |
211 | struct page *page = pvec->pages[i]; | |
3dd7ae8e | 212 | |
fc574c23 AS |
213 | /* block memcg migration during page moving between lru */ |
214 | if (!TestClearPageLRU(page)) | |
215 | continue; | |
216 | ||
2a5e4e34 | 217 | lruvec = relock_page_lruvec_irqsave(page, lruvec, &flags); |
c7c7b80c | 218 | (*move_fn)(page, lruvec); |
fc574c23 AS |
219 | |
220 | SetPageLRU(page); | |
902aaed0 | 221 | } |
6168d0da AS |
222 | if (lruvec) |
223 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
c6f92f9f | 224 | release_pages(pvec->pages, pvec->nr); |
83896fb5 | 225 | pagevec_reinit(pvec); |
d8505dee SL |
226 | } |
227 | ||
c7c7b80c | 228 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec) |
3dd7ae8e | 229 | { |
fc574c23 | 230 | if (!PageUnevictable(page)) { |
46ae6b2c | 231 | del_page_from_lru_list(page, lruvec); |
c55e8d03 | 232 | ClearPageActive(page); |
3a9c9788 | 233 | add_page_to_lru_list_tail(page, lruvec); |
c7c7b80c | 234 | __count_vm_events(PGROTATED, thp_nr_pages(page)); |
3dd7ae8e SL |
235 | } |
236 | } | |
237 | ||
d479960e MK |
238 | /* return true if pagevec needs to drain */ |
239 | static bool pagevec_add_and_need_flush(struct pagevec *pvec, struct page *page) | |
240 | { | |
241 | bool ret = false; | |
242 | ||
243 | if (!pagevec_add(pvec, page) || PageCompound(page) || | |
244 | lru_cache_disabled()) | |
245 | ret = true; | |
246 | ||
247 | return ret; | |
248 | } | |
249 | ||
1da177e4 LT |
250 | /* |
251 | * Writeback is about to end against a page which has been marked for immediate | |
252 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 253 | * inactive list. |
c7c7b80c AS |
254 | * |
255 | * rotate_reclaimable_page() must disable IRQs, to prevent nasty races. | |
1da177e4 | 256 | */ |
3dd7ae8e | 257 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 258 | { |
c55e8d03 | 259 | if (!PageLocked(page) && !PageDirty(page) && |
894bc310 | 260 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
261 | struct pagevec *pvec; |
262 | unsigned long flags; | |
263 | ||
09cbfeaf | 264 | get_page(page); |
b01b2141 IM |
265 | local_lock_irqsave(&lru_rotate.lock, flags); |
266 | pvec = this_cpu_ptr(&lru_rotate.pvec); | |
d479960e | 267 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 268 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); |
b01b2141 | 269 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
ac6aadb2 | 270 | } |
1da177e4 LT |
271 | } |
272 | ||
96f8bf4f | 273 | void lru_note_cost(struct lruvec *lruvec, bool file, unsigned int nr_pages) |
3e2f41f1 | 274 | { |
7cf111bc JW |
275 | do { |
276 | unsigned long lrusize; | |
277 | ||
6168d0da AS |
278 | /* |
279 | * Hold lruvec->lru_lock is safe here, since | |
280 | * 1) The pinned lruvec in reclaim, or | |
281 | * 2) From a pre-LRU page during refault (which also holds the | |
282 | * rcu lock, so would be safe even if the page was on the LRU | |
283 | * and could move simultaneously to a new lruvec). | |
284 | */ | |
285 | spin_lock_irq(&lruvec->lru_lock); | |
7cf111bc | 286 | /* Record cost event */ |
96f8bf4f JW |
287 | if (file) |
288 | lruvec->file_cost += nr_pages; | |
7cf111bc | 289 | else |
96f8bf4f | 290 | lruvec->anon_cost += nr_pages; |
7cf111bc JW |
291 | |
292 | /* | |
293 | * Decay previous events | |
294 | * | |
295 | * Because workloads change over time (and to avoid | |
296 | * overflow) we keep these statistics as a floating | |
297 | * average, which ends up weighing recent refaults | |
298 | * more than old ones. | |
299 | */ | |
300 | lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + | |
301 | lruvec_page_state(lruvec, NR_ACTIVE_ANON) + | |
302 | lruvec_page_state(lruvec, NR_INACTIVE_FILE) + | |
303 | lruvec_page_state(lruvec, NR_ACTIVE_FILE); | |
304 | ||
305 | if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { | |
306 | lruvec->file_cost /= 2; | |
307 | lruvec->anon_cost /= 2; | |
308 | } | |
6168d0da | 309 | spin_unlock_irq(&lruvec->lru_lock); |
7cf111bc | 310 | } while ((lruvec = parent_lruvec(lruvec))); |
3e2f41f1 KM |
311 | } |
312 | ||
96f8bf4f JW |
313 | void lru_note_cost_page(struct page *page) |
314 | { | |
315 | lru_note_cost(mem_cgroup_page_lruvec(page, page_pgdat(page)), | |
6c357848 | 316 | page_is_file_lru(page), thp_nr_pages(page)); |
96f8bf4f JW |
317 | } |
318 | ||
c7c7b80c | 319 | static void __activate_page(struct page *page, struct lruvec *lruvec) |
1da177e4 | 320 | { |
fc574c23 | 321 | if (!PageActive(page) && !PageUnevictable(page)) { |
6c357848 | 322 | int nr_pages = thp_nr_pages(page); |
744ed144 | 323 | |
46ae6b2c | 324 | del_page_from_lru_list(page, lruvec); |
7a608572 | 325 | SetPageActive(page); |
3a9c9788 | 326 | add_page_to_lru_list(page, lruvec); |
24b7e581 | 327 | trace_mm_lru_activate(page); |
4f98a2fe | 328 | |
21e330fc SB |
329 | __count_vm_events(PGACTIVATE, nr_pages); |
330 | __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, | |
331 | nr_pages); | |
1da177e4 | 332 | } |
eb709b0d SL |
333 | } |
334 | ||
335 | #ifdef CONFIG_SMP | |
eb709b0d SL |
336 | static void activate_page_drain(int cpu) |
337 | { | |
b01b2141 | 338 | struct pagevec *pvec = &per_cpu(lru_pvecs.activate_page, cpu); |
eb709b0d SL |
339 | |
340 | if (pagevec_count(pvec)) | |
c7c7b80c | 341 | pagevec_lru_move_fn(pvec, __activate_page); |
eb709b0d SL |
342 | } |
343 | ||
5fbc4616 CM |
344 | static bool need_activate_page_drain(int cpu) |
345 | { | |
b01b2141 | 346 | return pagevec_count(&per_cpu(lru_pvecs.activate_page, cpu)) != 0; |
5fbc4616 CM |
347 | } |
348 | ||
cc2828b2 | 349 | static void activate_page(struct page *page) |
eb709b0d | 350 | { |
800d8c63 | 351 | page = compound_head(page); |
eb709b0d | 352 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
b01b2141 | 353 | struct pagevec *pvec; |
eb709b0d | 354 | |
b01b2141 IM |
355 | local_lock(&lru_pvecs.lock); |
356 | pvec = this_cpu_ptr(&lru_pvecs.activate_page); | |
09cbfeaf | 357 | get_page(page); |
d479960e | 358 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 359 | pagevec_lru_move_fn(pvec, __activate_page); |
b01b2141 | 360 | local_unlock(&lru_pvecs.lock); |
eb709b0d SL |
361 | } |
362 | } | |
363 | ||
364 | #else | |
365 | static inline void activate_page_drain(int cpu) | |
366 | { | |
367 | } | |
368 | ||
cc2828b2 | 369 | static void activate_page(struct page *page) |
eb709b0d | 370 | { |
6168d0da | 371 | struct lruvec *lruvec; |
eb709b0d | 372 | |
800d8c63 | 373 | page = compound_head(page); |
6168d0da AS |
374 | if (TestClearPageLRU(page)) { |
375 | lruvec = lock_page_lruvec_irq(page); | |
376 | __activate_page(page, lruvec); | |
377 | unlock_page_lruvec_irq(lruvec); | |
378 | SetPageLRU(page); | |
379 | } | |
1da177e4 | 380 | } |
eb709b0d | 381 | #endif |
1da177e4 | 382 | |
059285a2 MG |
383 | static void __lru_cache_activate_page(struct page *page) |
384 | { | |
b01b2141 | 385 | struct pagevec *pvec; |
059285a2 MG |
386 | int i; |
387 | ||
b01b2141 IM |
388 | local_lock(&lru_pvecs.lock); |
389 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
390 | ||
059285a2 MG |
391 | /* |
392 | * Search backwards on the optimistic assumption that the page being | |
393 | * activated has just been added to this pagevec. Note that only | |
394 | * the local pagevec is examined as a !PageLRU page could be in the | |
395 | * process of being released, reclaimed, migrated or on a remote | |
396 | * pagevec that is currently being drained. Furthermore, marking | |
397 | * a remote pagevec's page PageActive potentially hits a race where | |
398 | * a page is marked PageActive just after it is added to the inactive | |
399 | * list causing accounting errors and BUG_ON checks to trigger. | |
400 | */ | |
401 | for (i = pagevec_count(pvec) - 1; i >= 0; i--) { | |
402 | struct page *pagevec_page = pvec->pages[i]; | |
403 | ||
404 | if (pagevec_page == page) { | |
405 | SetPageActive(page); | |
406 | break; | |
407 | } | |
408 | } | |
409 | ||
b01b2141 | 410 | local_unlock(&lru_pvecs.lock); |
059285a2 MG |
411 | } |
412 | ||
1da177e4 LT |
413 | /* |
414 | * Mark a page as having seen activity. | |
415 | * | |
416 | * inactive,unreferenced -> inactive,referenced | |
417 | * inactive,referenced -> active,unreferenced | |
418 | * active,unreferenced -> active,referenced | |
eb39d618 HD |
419 | * |
420 | * When a newly allocated page is not yet visible, so safe for non-atomic ops, | |
421 | * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). | |
1da177e4 | 422 | */ |
920c7a5d | 423 | void mark_page_accessed(struct page *page) |
1da177e4 | 424 | { |
e90309c9 | 425 | page = compound_head(page); |
059285a2 | 426 | |
a1100a74 FW |
427 | if (!PageReferenced(page)) { |
428 | SetPageReferenced(page); | |
429 | } else if (PageUnevictable(page)) { | |
430 | /* | |
431 | * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, | |
432 | * this list is never rotated or maintained, so marking an | |
433 | * evictable page accessed has no effect. | |
434 | */ | |
435 | } else if (!PageActive(page)) { | |
059285a2 MG |
436 | /* |
437 | * If the page is on the LRU, queue it for activation via | |
b01b2141 | 438 | * lru_pvecs.activate_page. Otherwise, assume the page is on a |
059285a2 MG |
439 | * pagevec, mark it active and it'll be moved to the active |
440 | * LRU on the next drain. | |
441 | */ | |
442 | if (PageLRU(page)) | |
443 | activate_page(page); | |
444 | else | |
445 | __lru_cache_activate_page(page); | |
1da177e4 | 446 | ClearPageReferenced(page); |
cb686883 | 447 | workingset_activation(page); |
1da177e4 | 448 | } |
33c3fc71 VD |
449 | if (page_is_idle(page)) |
450 | clear_page_idle(page); | |
1da177e4 | 451 | } |
1da177e4 LT |
452 | EXPORT_SYMBOL(mark_page_accessed); |
453 | ||
f04e9ebb | 454 | /** |
c53954a0 | 455 | * lru_cache_add - add a page to a page list |
f04e9ebb | 456 | * @page: the page to be added to the LRU. |
2329d375 JZ |
457 | * |
458 | * Queue the page for addition to the LRU via pagevec. The decision on whether | |
459 | * to add the page to the [in]active [file|anon] list is deferred until the | |
460 | * pagevec is drained. This gives a chance for the caller of lru_cache_add() | |
461 | * have the page added to the active list using mark_page_accessed(). | |
f04e9ebb | 462 | */ |
c53954a0 | 463 | void lru_cache_add(struct page *page) |
1da177e4 | 464 | { |
6058eaec JW |
465 | struct pagevec *pvec; |
466 | ||
309381fe SL |
467 | VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); |
468 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6058eaec JW |
469 | |
470 | get_page(page); | |
471 | local_lock(&lru_pvecs.lock); | |
472 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
d479960e | 473 | if (pagevec_add_and_need_flush(pvec, page)) |
6058eaec JW |
474 | __pagevec_lru_add(pvec); |
475 | local_unlock(&lru_pvecs.lock); | |
1da177e4 | 476 | } |
6058eaec | 477 | EXPORT_SYMBOL(lru_cache_add); |
1da177e4 | 478 | |
00501b53 | 479 | /** |
b518154e | 480 | * lru_cache_add_inactive_or_unevictable |
00501b53 JW |
481 | * @page: the page to be added to LRU |
482 | * @vma: vma in which page is mapped for determining reclaimability | |
483 | * | |
b518154e | 484 | * Place @page on the inactive or unevictable LRU list, depending on its |
12eab428 | 485 | * evictability. |
00501b53 | 486 | */ |
b518154e | 487 | void lru_cache_add_inactive_or_unevictable(struct page *page, |
00501b53 JW |
488 | struct vm_area_struct *vma) |
489 | { | |
b518154e JK |
490 | bool unevictable; |
491 | ||
00501b53 JW |
492 | VM_BUG_ON_PAGE(PageLRU(page), page); |
493 | ||
b518154e JK |
494 | unevictable = (vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED; |
495 | if (unlikely(unevictable) && !TestSetPageMlocked(page)) { | |
0964730b | 496 | int nr_pages = thp_nr_pages(page); |
00501b53 JW |
497 | /* |
498 | * We use the irq-unsafe __mod_zone_page_stat because this | |
499 | * counter is not modified from interrupt context, and the pte | |
500 | * lock is held(spinlock), which implies preemption disabled. | |
501 | */ | |
0964730b HD |
502 | __mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages); |
503 | count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages); | |
00501b53 | 504 | } |
9c4e6b1a | 505 | lru_cache_add(page); |
00501b53 JW |
506 | } |
507 | ||
31560180 MK |
508 | /* |
509 | * If the page can not be invalidated, it is moved to the | |
510 | * inactive list to speed up its reclaim. It is moved to the | |
511 | * head of the list, rather than the tail, to give the flusher | |
512 | * threads some time to write it out, as this is much more | |
513 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
514 | * |
515 | * If the page isn't page_mapped and dirty/writeback, the page | |
516 | * could reclaim asap using PG_reclaim. | |
517 | * | |
518 | * 1. active, mapped page -> none | |
519 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
520 | * 3. inactive, mapped page -> none | |
521 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
522 | * 5. inactive, clean -> inactive, tail | |
523 | * 6. Others -> none | |
524 | * | |
525 | * In 4, why it moves inactive's head, the VM expects the page would | |
526 | * be write it out by flusher threads as this is much more effective | |
527 | * than the single-page writeout from reclaim. | |
31560180 | 528 | */ |
c7c7b80c | 529 | static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec) |
31560180 | 530 | { |
46ae6b2c | 531 | bool active = PageActive(page); |
6c357848 | 532 | int nr_pages = thp_nr_pages(page); |
31560180 | 533 | |
bad49d9c MK |
534 | if (PageUnevictable(page)) |
535 | return; | |
536 | ||
31560180 MK |
537 | /* Some processes are using the page */ |
538 | if (page_mapped(page)) | |
539 | return; | |
540 | ||
46ae6b2c | 541 | del_page_from_lru_list(page, lruvec); |
31560180 MK |
542 | ClearPageActive(page); |
543 | ClearPageReferenced(page); | |
31560180 | 544 | |
278df9f4 MK |
545 | if (PageWriteback(page) || PageDirty(page)) { |
546 | /* | |
547 | * PG_reclaim could be raced with end_page_writeback | |
548 | * It can make readahead confusing. But race window | |
549 | * is _really_ small and it's non-critical problem. | |
550 | */ | |
3a9c9788 | 551 | add_page_to_lru_list(page, lruvec); |
278df9f4 MK |
552 | SetPageReclaim(page); |
553 | } else { | |
554 | /* | |
555 | * The page's writeback ends up during pagevec | |
556 | * We moves tha page into tail of inactive. | |
557 | */ | |
3a9c9788 | 558 | add_page_to_lru_list_tail(page, lruvec); |
5d91f31f | 559 | __count_vm_events(PGROTATED, nr_pages); |
278df9f4 MK |
560 | } |
561 | ||
21e330fc | 562 | if (active) { |
5d91f31f | 563 | __count_vm_events(PGDEACTIVATE, nr_pages); |
21e330fc SB |
564 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, |
565 | nr_pages); | |
566 | } | |
31560180 MK |
567 | } |
568 | ||
c7c7b80c | 569 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec) |
9c276cc6 | 570 | { |
fc574c23 | 571 | if (PageActive(page) && !PageUnevictable(page)) { |
6c357848 | 572 | int nr_pages = thp_nr_pages(page); |
9c276cc6 | 573 | |
46ae6b2c | 574 | del_page_from_lru_list(page, lruvec); |
9c276cc6 MK |
575 | ClearPageActive(page); |
576 | ClearPageReferenced(page); | |
3a9c9788 | 577 | add_page_to_lru_list(page, lruvec); |
9c276cc6 | 578 | |
21e330fc SB |
579 | __count_vm_events(PGDEACTIVATE, nr_pages); |
580 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, | |
581 | nr_pages); | |
9c276cc6 MK |
582 | } |
583 | } | |
10853a03 | 584 | |
c7c7b80c | 585 | static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec) |
10853a03 | 586 | { |
fc574c23 | 587 | if (PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 588 | !PageSwapCache(page) && !PageUnevictable(page)) { |
6c357848 | 589 | int nr_pages = thp_nr_pages(page); |
10853a03 | 590 | |
46ae6b2c | 591 | del_page_from_lru_list(page, lruvec); |
10853a03 MK |
592 | ClearPageActive(page); |
593 | ClearPageReferenced(page); | |
f7ad2a6c | 594 | /* |
9de4f22a HY |
595 | * Lazyfree pages are clean anonymous pages. They have |
596 | * PG_swapbacked flag cleared, to distinguish them from normal | |
597 | * anonymous pages | |
f7ad2a6c SL |
598 | */ |
599 | ClearPageSwapBacked(page); | |
3a9c9788 | 600 | add_page_to_lru_list(page, lruvec); |
10853a03 | 601 | |
21e330fc SB |
602 | __count_vm_events(PGLAZYFREE, nr_pages); |
603 | __count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, | |
604 | nr_pages); | |
10853a03 MK |
605 | } |
606 | } | |
607 | ||
902aaed0 HH |
608 | /* |
609 | * Drain pages out of the cpu's pagevecs. | |
610 | * Either "cpu" is the current CPU, and preemption has already been | |
611 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
612 | */ | |
f0cb3c76 | 613 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 614 | { |
b01b2141 | 615 | struct pagevec *pvec = &per_cpu(lru_pvecs.lru_add, cpu); |
1da177e4 | 616 | |
13f7f789 | 617 | if (pagevec_count(pvec)) |
a0b8cab3 | 618 | __pagevec_lru_add(pvec); |
902aaed0 | 619 | |
b01b2141 | 620 | pvec = &per_cpu(lru_rotate.pvec, cpu); |
7e0cc01e QC |
621 | /* Disabling interrupts below acts as a compiler barrier. */ |
622 | if (data_race(pagevec_count(pvec))) { | |
902aaed0 HH |
623 | unsigned long flags; |
624 | ||
625 | /* No harm done if a racing interrupt already did this */ | |
b01b2141 | 626 | local_lock_irqsave(&lru_rotate.lock, flags); |
c7c7b80c | 627 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); |
b01b2141 | 628 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
902aaed0 | 629 | } |
31560180 | 630 | |
b01b2141 | 631 | pvec = &per_cpu(lru_pvecs.lru_deactivate_file, cpu); |
31560180 | 632 | if (pagevec_count(pvec)) |
c7c7b80c | 633 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); |
eb709b0d | 634 | |
b01b2141 | 635 | pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu); |
9c276cc6 | 636 | if (pagevec_count(pvec)) |
c7c7b80c | 637 | pagevec_lru_move_fn(pvec, lru_deactivate_fn); |
9c276cc6 | 638 | |
b01b2141 | 639 | pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu); |
10853a03 | 640 | if (pagevec_count(pvec)) |
c7c7b80c | 641 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn); |
10853a03 | 642 | |
eb709b0d | 643 | activate_page_drain(cpu); |
31560180 MK |
644 | } |
645 | ||
646 | /** | |
cc5993bd | 647 | * deactivate_file_page - forcefully deactivate a file page |
31560180 MK |
648 | * @page: page to deactivate |
649 | * | |
650 | * This function hints the VM that @page is a good reclaim candidate, | |
651 | * for example if its invalidation fails due to the page being dirty | |
652 | * or under writeback. | |
653 | */ | |
cc5993bd | 654 | void deactivate_file_page(struct page *page) |
31560180 | 655 | { |
821ed6bb | 656 | /* |
cc5993bd MK |
657 | * In a workload with many unevictable page such as mprotect, |
658 | * unevictable page deactivation for accelerating reclaim is pointless. | |
821ed6bb MK |
659 | */ |
660 | if (PageUnevictable(page)) | |
661 | return; | |
662 | ||
31560180 | 663 | if (likely(get_page_unless_zero(page))) { |
b01b2141 IM |
664 | struct pagevec *pvec; |
665 | ||
666 | local_lock(&lru_pvecs.lock); | |
667 | pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate_file); | |
31560180 | 668 | |
d479960e | 669 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 670 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); |
b01b2141 | 671 | local_unlock(&lru_pvecs.lock); |
31560180 | 672 | } |
80bfed90 AM |
673 | } |
674 | ||
9c276cc6 MK |
675 | /* |
676 | * deactivate_page - deactivate a page | |
677 | * @page: page to deactivate | |
678 | * | |
679 | * deactivate_page() moves @page to the inactive list if @page was on the active | |
680 | * list and was not an unevictable page. This is done to accelerate the reclaim | |
681 | * of @page. | |
682 | */ | |
683 | void deactivate_page(struct page *page) | |
684 | { | |
685 | if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { | |
b01b2141 | 686 | struct pagevec *pvec; |
9c276cc6 | 687 | |
b01b2141 IM |
688 | local_lock(&lru_pvecs.lock); |
689 | pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate); | |
9c276cc6 | 690 | get_page(page); |
d479960e | 691 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 692 | pagevec_lru_move_fn(pvec, lru_deactivate_fn); |
b01b2141 | 693 | local_unlock(&lru_pvecs.lock); |
9c276cc6 MK |
694 | } |
695 | } | |
696 | ||
10853a03 | 697 | /** |
f7ad2a6c | 698 | * mark_page_lazyfree - make an anon page lazyfree |
10853a03 MK |
699 | * @page: page to deactivate |
700 | * | |
f7ad2a6c SL |
701 | * mark_page_lazyfree() moves @page to the inactive file list. |
702 | * This is done to accelerate the reclaim of @page. | |
10853a03 | 703 | */ |
f7ad2a6c | 704 | void mark_page_lazyfree(struct page *page) |
10853a03 | 705 | { |
f7ad2a6c | 706 | if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 707 | !PageSwapCache(page) && !PageUnevictable(page)) { |
b01b2141 | 708 | struct pagevec *pvec; |
10853a03 | 709 | |
b01b2141 IM |
710 | local_lock(&lru_pvecs.lock); |
711 | pvec = this_cpu_ptr(&lru_pvecs.lru_lazyfree); | |
09cbfeaf | 712 | get_page(page); |
d479960e | 713 | if (pagevec_add_and_need_flush(pvec, page)) |
c7c7b80c | 714 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn); |
b01b2141 | 715 | local_unlock(&lru_pvecs.lock); |
10853a03 MK |
716 | } |
717 | } | |
718 | ||
80bfed90 AM |
719 | void lru_add_drain(void) |
720 | { | |
b01b2141 IM |
721 | local_lock(&lru_pvecs.lock); |
722 | lru_add_drain_cpu(smp_processor_id()); | |
723 | local_unlock(&lru_pvecs.lock); | |
724 | } | |
725 | ||
726 | void lru_add_drain_cpu_zone(struct zone *zone) | |
727 | { | |
728 | local_lock(&lru_pvecs.lock); | |
729 | lru_add_drain_cpu(smp_processor_id()); | |
730 | drain_local_pages(zone); | |
731 | local_unlock(&lru_pvecs.lock); | |
1da177e4 LT |
732 | } |
733 | ||
6ea183d6 MH |
734 | #ifdef CONFIG_SMP |
735 | ||
736 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); | |
737 | ||
c4028958 | 738 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
739 | { |
740 | lru_add_drain(); | |
741 | } | |
742 | ||
9852a721 MH |
743 | /* |
744 | * Doesn't need any cpu hotplug locking because we do rely on per-cpu | |
745 | * kworkers being shut down before our page_alloc_cpu_dead callback is | |
746 | * executed on the offlined cpu. | |
747 | * Calling this function with cpu hotplug locks held can actually lead | |
748 | * to obscure indirect dependencies via WQ context. | |
749 | */ | |
d479960e | 750 | inline void __lru_add_drain_all(bool force_all_cpus) |
053837fc | 751 | { |
6446a513 AD |
752 | /* |
753 | * lru_drain_gen - Global pages generation number | |
754 | * | |
755 | * (A) Definition: global lru_drain_gen = x implies that all generations | |
756 | * 0 < n <= x are already *scheduled* for draining. | |
757 | * | |
758 | * This is an optimization for the highly-contended use case where a | |
759 | * user space workload keeps constantly generating a flow of pages for | |
760 | * each CPU. | |
761 | */ | |
762 | static unsigned int lru_drain_gen; | |
5fbc4616 | 763 | static struct cpumask has_work; |
6446a513 AD |
764 | static DEFINE_MUTEX(lock); |
765 | unsigned cpu, this_gen; | |
5fbc4616 | 766 | |
ce612879 MH |
767 | /* |
768 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
769 | * initialized. | |
770 | */ | |
771 | if (WARN_ON(!mm_percpu_wq)) | |
772 | return; | |
773 | ||
6446a513 AD |
774 | /* |
775 | * Guarantee pagevec counter stores visible by this CPU are visible to | |
776 | * other CPUs before loading the current drain generation. | |
777 | */ | |
778 | smp_mb(); | |
779 | ||
780 | /* | |
781 | * (B) Locally cache global LRU draining generation number | |
782 | * | |
783 | * The read barrier ensures that the counter is loaded before the mutex | |
784 | * is taken. It pairs with smp_mb() inside the mutex critical section | |
785 | * at (D). | |
786 | */ | |
787 | this_gen = smp_load_acquire(&lru_drain_gen); | |
eef1a429 | 788 | |
5fbc4616 | 789 | mutex_lock(&lock); |
eef1a429 KK |
790 | |
791 | /* | |
6446a513 AD |
792 | * (C) Exit the draining operation if a newer generation, from another |
793 | * lru_add_drain_all(), was already scheduled for draining. Check (A). | |
eef1a429 | 794 | */ |
d479960e | 795 | if (unlikely(this_gen != lru_drain_gen && !force_all_cpus)) |
eef1a429 KK |
796 | goto done; |
797 | ||
6446a513 AD |
798 | /* |
799 | * (D) Increment global generation number | |
800 | * | |
801 | * Pairs with smp_load_acquire() at (B), outside of the critical | |
802 | * section. Use a full memory barrier to guarantee that the new global | |
803 | * drain generation number is stored before loading pagevec counters. | |
804 | * | |
805 | * This pairing must be done here, before the for_each_online_cpu loop | |
806 | * below which drains the page vectors. | |
807 | * | |
808 | * Let x, y, and z represent some system CPU numbers, where x < y < z. | |
809 | * Assume CPU #z is is in the middle of the for_each_online_cpu loop | |
810 | * below and has already reached CPU #y's per-cpu data. CPU #x comes | |
811 | * along, adds some pages to its per-cpu vectors, then calls | |
812 | * lru_add_drain_all(). | |
813 | * | |
814 | * If the paired barrier is done at any later step, e.g. after the | |
815 | * loop, CPU #x will just exit at (C) and miss flushing out all of its | |
816 | * added pages. | |
817 | */ | |
818 | WRITE_ONCE(lru_drain_gen, lru_drain_gen + 1); | |
819 | smp_mb(); | |
eef1a429 | 820 | |
5fbc4616 | 821 | cpumask_clear(&has_work); |
5fbc4616 CM |
822 | for_each_online_cpu(cpu) { |
823 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); | |
824 | ||
d479960e MK |
825 | if (force_all_cpus || |
826 | pagevec_count(&per_cpu(lru_pvecs.lru_add, cpu)) || | |
7e0cc01e | 827 | data_race(pagevec_count(&per_cpu(lru_rotate.pvec, cpu))) || |
b01b2141 IM |
828 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) || |
829 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) || | |
830 | pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) || | |
5fbc4616 CM |
831 | need_activate_page_drain(cpu)) { |
832 | INIT_WORK(work, lru_add_drain_per_cpu); | |
ce612879 | 833 | queue_work_on(cpu, mm_percpu_wq, work); |
6446a513 | 834 | __cpumask_set_cpu(cpu, &has_work); |
5fbc4616 CM |
835 | } |
836 | } | |
837 | ||
838 | for_each_cpu(cpu, &has_work) | |
839 | flush_work(&per_cpu(lru_add_drain_work, cpu)); | |
840 | ||
eef1a429 | 841 | done: |
5fbc4616 | 842 | mutex_unlock(&lock); |
053837fc | 843 | } |
d479960e MK |
844 | |
845 | void lru_add_drain_all(void) | |
846 | { | |
847 | __lru_add_drain_all(false); | |
848 | } | |
6ea183d6 MH |
849 | #else |
850 | void lru_add_drain_all(void) | |
851 | { | |
852 | lru_add_drain(); | |
853 | } | |
6446a513 | 854 | #endif /* CONFIG_SMP */ |
053837fc | 855 | |
d479960e MK |
856 | atomic_t lru_disable_count = ATOMIC_INIT(0); |
857 | ||
858 | /* | |
859 | * lru_cache_disable() needs to be called before we start compiling | |
860 | * a list of pages to be migrated using isolate_lru_page(). | |
861 | * It drains pages on LRU cache and then disable on all cpus until | |
862 | * lru_cache_enable is called. | |
863 | * | |
864 | * Must be paired with a call to lru_cache_enable(). | |
865 | */ | |
866 | void lru_cache_disable(void) | |
867 | { | |
868 | atomic_inc(&lru_disable_count); | |
869 | #ifdef CONFIG_SMP | |
870 | /* | |
871 | * lru_add_drain_all in the force mode will schedule draining on | |
872 | * all online CPUs so any calls of lru_cache_disabled wrapped by | |
873 | * local_lock or preemption disabled would be ordered by that. | |
874 | * The atomic operation doesn't need to have stronger ordering | |
875 | * requirements because that is enforeced by the scheduling | |
876 | * guarantees. | |
877 | */ | |
878 | __lru_add_drain_all(true); | |
879 | #else | |
880 | lru_add_drain(); | |
881 | #endif | |
882 | } | |
883 | ||
aabfb572 | 884 | /** |
ea1754a0 | 885 | * release_pages - batched put_page() |
aabfb572 MH |
886 | * @pages: array of pages to release |
887 | * @nr: number of pages | |
1da177e4 | 888 | * |
aabfb572 MH |
889 | * Decrement the reference count on all the pages in @pages. If it |
890 | * fell to zero, remove the page from the LRU and free it. | |
1da177e4 | 891 | */ |
c6f92f9f | 892 | void release_pages(struct page **pages, int nr) |
1da177e4 LT |
893 | { |
894 | int i; | |
cc59850e | 895 | LIST_HEAD(pages_to_free); |
6168d0da | 896 | struct lruvec *lruvec = NULL; |
3f649ab7 KC |
897 | unsigned long flags; |
898 | unsigned int lock_batch; | |
1da177e4 | 899 | |
1da177e4 LT |
900 | for (i = 0; i < nr; i++) { |
901 | struct page *page = pages[i]; | |
1da177e4 | 902 | |
aabfb572 MH |
903 | /* |
904 | * Make sure the IRQ-safe lock-holding time does not get | |
905 | * excessive with a continuous string of pages from the | |
6168d0da | 906 | * same lruvec. The lock is held only if lruvec != NULL. |
aabfb572 | 907 | */ |
6168d0da AS |
908 | if (lruvec && ++lock_batch == SWAP_CLUSTER_MAX) { |
909 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
910 | lruvec = NULL; | |
aabfb572 MH |
911 | } |
912 | ||
a9b576f7 | 913 | page = compound_head(page); |
6fcb52a5 | 914 | if (is_huge_zero_page(page)) |
aa88b68c | 915 | continue; |
aa88b68c | 916 | |
c5d6c45e | 917 | if (is_zone_device_page(page)) { |
6168d0da AS |
918 | if (lruvec) { |
919 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
920 | lruvec = NULL; | |
df6ad698 | 921 | } |
c5d6c45e IW |
922 | /* |
923 | * ZONE_DEVICE pages that return 'false' from | |
a3e7bea0 | 924 | * page_is_devmap_managed() do not require special |
c5d6c45e IW |
925 | * processing, and instead, expect a call to |
926 | * put_page_testzero(). | |
927 | */ | |
07d80269 JH |
928 | if (page_is_devmap_managed(page)) { |
929 | put_devmap_managed_page(page); | |
c5d6c45e | 930 | continue; |
07d80269 | 931 | } |
43fbdeb3 RC |
932 | if (put_page_testzero(page)) |
933 | put_dev_pagemap(page->pgmap); | |
934 | continue; | |
df6ad698 JG |
935 | } |
936 | ||
b5810039 | 937 | if (!put_page_testzero(page)) |
1da177e4 LT |
938 | continue; |
939 | ||
ddc58f27 | 940 | if (PageCompound(page)) { |
6168d0da AS |
941 | if (lruvec) { |
942 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
943 | lruvec = NULL; | |
ddc58f27 KS |
944 | } |
945 | __put_compound_page(page); | |
946 | continue; | |
947 | } | |
948 | ||
46453a6e | 949 | if (PageLRU(page)) { |
2a5e4e34 AD |
950 | struct lruvec *prev_lruvec = lruvec; |
951 | ||
952 | lruvec = relock_page_lruvec_irqsave(page, lruvec, | |
953 | &flags); | |
954 | if (prev_lruvec != lruvec) | |
aabfb572 | 955 | lock_batch = 0; |
fa9add64 | 956 | |
46ae6b2c | 957 | del_page_from_lru_list(page, lruvec); |
87560179 | 958 | __clear_page_lru_flags(page); |
46453a6e NP |
959 | } |
960 | ||
62906027 | 961 | __ClearPageWaiters(page); |
c53954a0 | 962 | |
cc59850e | 963 | list_add(&page->lru, &pages_to_free); |
1da177e4 | 964 | } |
6168d0da AS |
965 | if (lruvec) |
966 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
1da177e4 | 967 | |
747db954 | 968 | mem_cgroup_uncharge_list(&pages_to_free); |
2d4894b5 | 969 | free_unref_page_list(&pages_to_free); |
1da177e4 | 970 | } |
0be8557b | 971 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
972 | |
973 | /* | |
974 | * The pages which we're about to release may be in the deferred lru-addition | |
975 | * queues. That would prevent them from really being freed right now. That's | |
976 | * OK from a correctness point of view but is inefficient - those pages may be | |
977 | * cache-warm and we want to give them back to the page allocator ASAP. | |
978 | * | |
979 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
980 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
981 | * mutual recursion. | |
982 | */ | |
983 | void __pagevec_release(struct pagevec *pvec) | |
984 | { | |
7f0b5fb9 | 985 | if (!pvec->percpu_pvec_drained) { |
d9ed0d08 | 986 | lru_add_drain(); |
7f0b5fb9 | 987 | pvec->percpu_pvec_drained = true; |
d9ed0d08 | 988 | } |
c6f92f9f | 989 | release_pages(pvec->pages, pagevec_count(pvec)); |
1da177e4 LT |
990 | pagevec_reinit(pvec); |
991 | } | |
7f285701 SF |
992 | EXPORT_SYMBOL(__pagevec_release); |
993 | ||
c7c7b80c | 994 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec) |
3dd7ae8e | 995 | { |
9c4e6b1a | 996 | int was_unevictable = TestClearPageUnevictable(page); |
6c357848 | 997 | int nr_pages = thp_nr_pages(page); |
3dd7ae8e | 998 | |
309381fe | 999 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3dd7ae8e | 1000 | |
9c4e6b1a SB |
1001 | /* |
1002 | * Page becomes evictable in two ways: | |
dae966dc | 1003 | * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()]. |
9c4e6b1a SB |
1004 | * 2) Before acquiring LRU lock to put the page to correct LRU and then |
1005 | * a) do PageLRU check with lock [check_move_unevictable_pages] | |
1006 | * b) do PageLRU check before lock [clear_page_mlock] | |
1007 | * | |
1008 | * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need | |
1009 | * following strict ordering: | |
1010 | * | |
1011 | * #0: __pagevec_lru_add_fn #1: clear_page_mlock | |
1012 | * | |
1013 | * SetPageLRU() TestClearPageMlocked() | |
1014 | * smp_mb() // explicit ordering // above provides strict | |
1015 | * // ordering | |
1016 | * PageMlocked() PageLRU() | |
1017 | * | |
1018 | * | |
1019 | * if '#1' does not observe setting of PG_lru by '#0' and fails | |
1020 | * isolation, the explicit barrier will make sure that page_evictable | |
1021 | * check will put the page in correct LRU. Without smp_mb(), SetPageLRU | |
1022 | * can be reordered after PageMlocked check and can make '#1' to fail | |
1023 | * the isolation of the page whose Mlocked bit is cleared (#0 is also | |
1024 | * looking at the same page) and the evictable page will be stranded | |
1025 | * in an unevictable LRU. | |
1026 | */ | |
9a9b6cce YS |
1027 | SetPageLRU(page); |
1028 | smp_mb__after_atomic(); | |
9c4e6b1a SB |
1029 | |
1030 | if (page_evictable(page)) { | |
9c4e6b1a | 1031 | if (was_unevictable) |
5d91f31f | 1032 | __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); |
9c4e6b1a | 1033 | } else { |
9c4e6b1a SB |
1034 | ClearPageActive(page); |
1035 | SetPageUnevictable(page); | |
1036 | if (!was_unevictable) | |
5d91f31f | 1037 | __count_vm_events(UNEVICTABLE_PGCULLED, nr_pages); |
9c4e6b1a SB |
1038 | } |
1039 | ||
3a9c9788 | 1040 | add_page_to_lru_list(page, lruvec); |
86140453 | 1041 | trace_mm_lru_insertion(page); |
3dd7ae8e SL |
1042 | } |
1043 | ||
1da177e4 LT |
1044 | /* |
1045 | * Add the passed pages to the LRU, then drop the caller's refcount | |
1046 | * on them. Reinitialises the caller's pagevec. | |
1047 | */ | |
a0b8cab3 | 1048 | void __pagevec_lru_add(struct pagevec *pvec) |
1da177e4 | 1049 | { |
fc574c23 | 1050 | int i; |
6168d0da | 1051 | struct lruvec *lruvec = NULL; |
fc574c23 AS |
1052 | unsigned long flags = 0; |
1053 | ||
1054 | for (i = 0; i < pagevec_count(pvec); i++) { | |
1055 | struct page *page = pvec->pages[i]; | |
fc574c23 | 1056 | |
2a5e4e34 | 1057 | lruvec = relock_page_lruvec_irqsave(page, lruvec, &flags); |
fc574c23 AS |
1058 | __pagevec_lru_add_fn(page, lruvec); |
1059 | } | |
6168d0da AS |
1060 | if (lruvec) |
1061 | unlock_page_lruvec_irqrestore(lruvec, flags); | |
fc574c23 AS |
1062 | release_pages(pvec->pages, pvec->nr); |
1063 | pagevec_reinit(pvec); | |
1da177e4 | 1064 | } |
1da177e4 | 1065 | |
0cd6144a JW |
1066 | /** |
1067 | * pagevec_remove_exceptionals - pagevec exceptionals pruning | |
1068 | * @pvec: The pagevec to prune | |
1069 | * | |
a656a202 MWO |
1070 | * find_get_entries() fills both pages and XArray value entries (aka |
1071 | * exceptional entries) into the pagevec. This function prunes all | |
0cd6144a JW |
1072 | * exceptionals from @pvec without leaving holes, so that it can be |
1073 | * passed on to page-only pagevec operations. | |
1074 | */ | |
1075 | void pagevec_remove_exceptionals(struct pagevec *pvec) | |
1076 | { | |
1077 | int i, j; | |
1078 | ||
1079 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
1080 | struct page *page = pvec->pages[i]; | |
3159f943 | 1081 | if (!xa_is_value(page)) |
0cd6144a JW |
1082 | pvec->pages[j++] = page; |
1083 | } | |
1084 | pvec->nr = j; | |
1085 | } | |
1086 | ||
1da177e4 | 1087 | /** |
b947cee4 | 1088 | * pagevec_lookup_range - gang pagecache lookup |
1da177e4 LT |
1089 | * @pvec: Where the resulting pages are placed |
1090 | * @mapping: The address_space to search | |
1091 | * @start: The starting page index | |
b947cee4 | 1092 | * @end: The final page index |
1da177e4 | 1093 | * |
e02a9f04 | 1094 | * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE |
b947cee4 JK |
1095 | * pages in the mapping starting from index @start and upto index @end |
1096 | * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a | |
1da177e4 LT |
1097 | * reference against the pages in @pvec. |
1098 | * | |
1099 | * The search returns a group of mapping-contiguous pages with ascending | |
d72dc8a2 JK |
1100 | * indexes. There may be holes in the indices due to not-present pages. We |
1101 | * also update @start to index the next page for the traversal. | |
1da177e4 | 1102 | * |
b947cee4 | 1103 | * pagevec_lookup_range() returns the number of pages which were found. If this |
e02a9f04 | 1104 | * number is smaller than PAGEVEC_SIZE, the end of specified range has been |
b947cee4 | 1105 | * reached. |
1da177e4 | 1106 | */ |
b947cee4 | 1107 | unsigned pagevec_lookup_range(struct pagevec *pvec, |
397162ff | 1108 | struct address_space *mapping, pgoff_t *start, pgoff_t end) |
1da177e4 | 1109 | { |
397162ff | 1110 | pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, |
b947cee4 | 1111 | pvec->pages); |
1da177e4 LT |
1112 | return pagevec_count(pvec); |
1113 | } | |
b947cee4 | 1114 | EXPORT_SYMBOL(pagevec_lookup_range); |
78539fdf | 1115 | |
72b045ae JK |
1116 | unsigned pagevec_lookup_range_tag(struct pagevec *pvec, |
1117 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1118 | xa_mark_t tag) |
1da177e4 | 1119 | { |
72b045ae | 1120 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
67fd707f | 1121 | PAGEVEC_SIZE, pvec->pages); |
1da177e4 LT |
1122 | return pagevec_count(pvec); |
1123 | } | |
72b045ae | 1124 | EXPORT_SYMBOL(pagevec_lookup_range_tag); |
1da177e4 | 1125 | |
1da177e4 LT |
1126 | /* |
1127 | * Perform any setup for the swap system | |
1128 | */ | |
1129 | void __init swap_setup(void) | |
1130 | { | |
ca79b0c2 | 1131 | unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); |
e0bf68dd | 1132 | |
1da177e4 LT |
1133 | /* Use a smaller cluster for small-memory machines */ |
1134 | if (megs < 16) | |
1135 | page_cluster = 2; | |
1136 | else | |
1137 | page_cluster = 3; | |
1138 | /* | |
1139 | * Right now other parts of the system means that we | |
1140 | * _really_ don't want to cluster much more | |
1141 | */ | |
1da177e4 | 1142 | } |
07d80269 JH |
1143 | |
1144 | #ifdef CONFIG_DEV_PAGEMAP_OPS | |
1145 | void put_devmap_managed_page(struct page *page) | |
1146 | { | |
1147 | int count; | |
1148 | ||
1149 | if (WARN_ON_ONCE(!page_is_devmap_managed(page))) | |
1150 | return; | |
1151 | ||
1152 | count = page_ref_dec_return(page); | |
1153 | ||
1154 | /* | |
1155 | * devmap page refcounts are 1-based, rather than 0-based: if | |
1156 | * refcount is 1, then the page is free and the refcount is | |
1157 | * stable because nobody holds a reference on the page. | |
1158 | */ | |
1159 | if (count == 1) | |
1160 | free_devmap_managed_page(page); | |
1161 | else if (!count) | |
1162 | __put_page(page); | |
1163 | } | |
1164 | EXPORT_SYMBOL(put_devmap_managed_page); | |
1165 | #endif |