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)) { | |
f4b7e272 | 82 | pg_data_t *pgdat = page_pgdat(page); |
fa9add64 HD |
83 | struct lruvec *lruvec; |
84 | unsigned long flags; | |
b221385b | 85 | |
f4b7e272 AR |
86 | spin_lock_irqsave(&pgdat->lru_lock, flags); |
87 | lruvec = mem_cgroup_page_lruvec(page, pgdat); | |
309381fe | 88 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
b221385b | 89 | __ClearPageLRU(page); |
fa9add64 | 90 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
f4b7e272 | 91 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); |
b221385b | 92 | } |
62906027 | 93 | __ClearPageWaiters(page); |
91807063 AA |
94 | } |
95 | ||
96 | static void __put_single_page(struct page *page) | |
97 | { | |
98 | __page_cache_release(page); | |
7ae88534 | 99 | mem_cgroup_uncharge(page); |
2d4894b5 | 100 | free_unref_page(page); |
b221385b AB |
101 | } |
102 | ||
91807063 | 103 | static void __put_compound_page(struct page *page) |
1da177e4 | 104 | { |
822fc613 NH |
105 | /* |
106 | * __page_cache_release() is supposed to be called for thp, not for | |
107 | * hugetlb. This is because hugetlb page does never have PageLRU set | |
108 | * (it's never listed to any LRU lists) and no memcg routines should | |
109 | * be called for hugetlb (it has a separate hugetlb_cgroup.) | |
110 | */ | |
111 | if (!PageHuge(page)) | |
112 | __page_cache_release(page); | |
ff45fc3c | 113 | destroy_compound_page(page); |
91807063 AA |
114 | } |
115 | ||
ddc58f27 | 116 | void __put_page(struct page *page) |
8519fb30 | 117 | { |
71389703 DW |
118 | if (is_zone_device_page(page)) { |
119 | put_dev_pagemap(page->pgmap); | |
120 | ||
121 | /* | |
122 | * The page belongs to the device that created pgmap. Do | |
123 | * not return it to page allocator. | |
124 | */ | |
125 | return; | |
126 | } | |
127 | ||
8519fb30 | 128 | if (unlikely(PageCompound(page))) |
ddc58f27 KS |
129 | __put_compound_page(page); |
130 | else | |
91807063 | 131 | __put_single_page(page); |
1da177e4 | 132 | } |
ddc58f27 | 133 | EXPORT_SYMBOL(__put_page); |
70b50f94 | 134 | |
1d7ea732 | 135 | /** |
7682486b RD |
136 | * put_pages_list() - release a list of pages |
137 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
138 | * |
139 | * Release a list of pages which are strung together on page.lru. Currently | |
140 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
141 | */ |
142 | void put_pages_list(struct list_head *pages) | |
143 | { | |
144 | while (!list_empty(pages)) { | |
145 | struct page *victim; | |
146 | ||
f86196ea | 147 | victim = lru_to_page(pages); |
1d7ea732 | 148 | list_del(&victim->lru); |
09cbfeaf | 149 | put_page(victim); |
1d7ea732 AZ |
150 | } |
151 | } | |
152 | EXPORT_SYMBOL(put_pages_list); | |
153 | ||
18022c5d MG |
154 | /* |
155 | * get_kernel_pages() - pin kernel pages in memory | |
156 | * @kiov: An array of struct kvec structures | |
157 | * @nr_segs: number of segments to pin | |
158 | * @write: pinning for read/write, currently ignored | |
159 | * @pages: array that receives pointers to the pages pinned. | |
160 | * Should be at least nr_segs long. | |
161 | * | |
162 | * Returns number of pages pinned. This may be fewer than the number | |
163 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
164 | * were pinned, returns -errno. Each page returned must be released | |
165 | * with a put_page() call when it is finished with. | |
166 | */ | |
167 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
168 | struct page **pages) | |
169 | { | |
170 | int seg; | |
171 | ||
172 | for (seg = 0; seg < nr_segs; seg++) { | |
173 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
174 | return seg; | |
175 | ||
5a178119 | 176 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
09cbfeaf | 177 | get_page(pages[seg]); |
18022c5d MG |
178 | } |
179 | ||
180 | return seg; | |
181 | } | |
182 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
183 | ||
184 | /* | |
185 | * get_kernel_page() - pin a kernel page in memory | |
186 | * @start: starting kernel address | |
187 | * @write: pinning for read/write, currently ignored | |
188 | * @pages: array that receives pointer to the page pinned. | |
189 | * Must be at least nr_segs long. | |
190 | * | |
191 | * Returns 1 if page is pinned. If the page was not pinned, returns | |
192 | * -errno. The page returned must be released with a put_page() call | |
193 | * when it is finished with. | |
194 | */ | |
195 | int get_kernel_page(unsigned long start, int write, struct page **pages) | |
196 | { | |
197 | const struct kvec kiov = { | |
198 | .iov_base = (void *)start, | |
199 | .iov_len = PAGE_SIZE | |
200 | }; | |
201 | ||
202 | return get_kernel_pages(&kiov, 1, write, pages); | |
203 | } | |
204 | EXPORT_SYMBOL_GPL(get_kernel_page); | |
205 | ||
3dd7ae8e | 206 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
fa9add64 HD |
207 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
208 | void *arg) | |
902aaed0 HH |
209 | { |
210 | int i; | |
68eb0731 | 211 | struct pglist_data *pgdat = NULL; |
fa9add64 | 212 | struct lruvec *lruvec; |
3dd7ae8e | 213 | unsigned long flags = 0; |
902aaed0 HH |
214 | |
215 | for (i = 0; i < pagevec_count(pvec); i++) { | |
216 | struct page *page = pvec->pages[i]; | |
68eb0731 | 217 | struct pglist_data *pagepgdat = page_pgdat(page); |
902aaed0 | 218 | |
68eb0731 MG |
219 | if (pagepgdat != pgdat) { |
220 | if (pgdat) | |
221 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); | |
222 | pgdat = pagepgdat; | |
223 | spin_lock_irqsave(&pgdat->lru_lock, flags); | |
902aaed0 | 224 | } |
3dd7ae8e | 225 | |
68eb0731 | 226 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
fa9add64 | 227 | (*move_fn)(page, lruvec, arg); |
902aaed0 | 228 | } |
68eb0731 MG |
229 | if (pgdat) |
230 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); | |
c6f92f9f | 231 | release_pages(pvec->pages, pvec->nr); |
83896fb5 | 232 | pagevec_reinit(pvec); |
d8505dee SL |
233 | } |
234 | ||
fa9add64 HD |
235 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
236 | void *arg) | |
3dd7ae8e SL |
237 | { |
238 | int *pgmoved = arg; | |
3dd7ae8e | 239 | |
c55e8d03 JW |
240 | if (PageLRU(page) && !PageUnevictable(page)) { |
241 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
242 | ClearPageActive(page); | |
243 | add_page_to_lru_list_tail(page, lruvec, page_lru(page)); | |
3dd7ae8e SL |
244 | (*pgmoved)++; |
245 | } | |
246 | } | |
247 | ||
248 | /* | |
249 | * pagevec_move_tail() must be called with IRQ disabled. | |
250 | * Otherwise this may cause nasty races. | |
251 | */ | |
252 | static void pagevec_move_tail(struct pagevec *pvec) | |
253 | { | |
254 | int pgmoved = 0; | |
255 | ||
256 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
257 | __count_vm_events(PGROTATED, pgmoved); | |
258 | } | |
259 | ||
1da177e4 LT |
260 | /* |
261 | * Writeback is about to end against a page which has been marked for immediate | |
262 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 263 | * inactive list. |
1da177e4 | 264 | */ |
3dd7ae8e | 265 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 266 | { |
c55e8d03 | 267 | if (!PageLocked(page) && !PageDirty(page) && |
894bc310 | 268 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
269 | struct pagevec *pvec; |
270 | unsigned long flags; | |
271 | ||
09cbfeaf | 272 | get_page(page); |
b01b2141 IM |
273 | local_lock_irqsave(&lru_rotate.lock, flags); |
274 | pvec = this_cpu_ptr(&lru_rotate.pvec); | |
8f182270 | 275 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
ac6aadb2 | 276 | pagevec_move_tail(pvec); |
b01b2141 | 277 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
ac6aadb2 | 278 | } |
1da177e4 LT |
279 | } |
280 | ||
314b57fb | 281 | void lru_note_cost(struct page *page) |
3e2f41f1 | 282 | { |
314b57fb JW |
283 | struct lruvec *lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page)); |
284 | ||
7cf111bc JW |
285 | do { |
286 | unsigned long lrusize; | |
287 | ||
288 | /* Record cost event */ | |
289 | if (page_is_file_lru(page)) | |
290 | lruvec->file_cost++; | |
291 | else | |
292 | lruvec->anon_cost++; | |
293 | ||
294 | /* | |
295 | * Decay previous events | |
296 | * | |
297 | * Because workloads change over time (and to avoid | |
298 | * overflow) we keep these statistics as a floating | |
299 | * average, which ends up weighing recent refaults | |
300 | * more than old ones. | |
301 | */ | |
302 | lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + | |
303 | lruvec_page_state(lruvec, NR_ACTIVE_ANON) + | |
304 | lruvec_page_state(lruvec, NR_INACTIVE_FILE) + | |
305 | lruvec_page_state(lruvec, NR_ACTIVE_FILE); | |
306 | ||
307 | if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { | |
308 | lruvec->file_cost /= 2; | |
309 | lruvec->anon_cost /= 2; | |
310 | } | |
311 | } while ((lruvec = parent_lruvec(lruvec))); | |
3e2f41f1 KM |
312 | } |
313 | ||
fa9add64 HD |
314 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
315 | void *arg) | |
1da177e4 | 316 | { |
744ed144 | 317 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 | 318 | int lru = page_lru_base_type(page); |
744ed144 | 319 | |
fa9add64 | 320 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
321 | SetPageActive(page); |
322 | lru += LRU_ACTIVE; | |
fa9add64 | 323 | add_page_to_lru_list(page, lruvec, lru); |
24b7e581 | 324 | trace_mm_lru_activate(page); |
4f98a2fe | 325 | |
fa9add64 | 326 | __count_vm_event(PGACTIVATE); |
1da177e4 | 327 | } |
eb709b0d SL |
328 | } |
329 | ||
330 | #ifdef CONFIG_SMP | |
eb709b0d SL |
331 | static void activate_page_drain(int cpu) |
332 | { | |
b01b2141 | 333 | struct pagevec *pvec = &per_cpu(lru_pvecs.activate_page, cpu); |
eb709b0d SL |
334 | |
335 | if (pagevec_count(pvec)) | |
336 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
337 | } | |
338 | ||
5fbc4616 CM |
339 | static bool need_activate_page_drain(int cpu) |
340 | { | |
b01b2141 | 341 | return pagevec_count(&per_cpu(lru_pvecs.activate_page, cpu)) != 0; |
5fbc4616 CM |
342 | } |
343 | ||
eb709b0d SL |
344 | void activate_page(struct page *page) |
345 | { | |
800d8c63 | 346 | page = compound_head(page); |
eb709b0d | 347 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
b01b2141 | 348 | struct pagevec *pvec; |
eb709b0d | 349 | |
b01b2141 IM |
350 | local_lock(&lru_pvecs.lock); |
351 | pvec = this_cpu_ptr(&lru_pvecs.activate_page); | |
09cbfeaf | 352 | get_page(page); |
8f182270 | 353 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
eb709b0d | 354 | pagevec_lru_move_fn(pvec, __activate_page, NULL); |
b01b2141 | 355 | local_unlock(&lru_pvecs.lock); |
eb709b0d SL |
356 | } |
357 | } | |
358 | ||
359 | #else | |
360 | static inline void activate_page_drain(int cpu) | |
361 | { | |
362 | } | |
363 | ||
364 | void activate_page(struct page *page) | |
365 | { | |
f4b7e272 | 366 | pg_data_t *pgdat = page_pgdat(page); |
eb709b0d | 367 | |
800d8c63 | 368 | page = compound_head(page); |
f4b7e272 AR |
369 | spin_lock_irq(&pgdat->lru_lock); |
370 | __activate_page(page, mem_cgroup_page_lruvec(page, pgdat), NULL); | |
371 | spin_unlock_irq(&pgdat->lru_lock); | |
1da177e4 | 372 | } |
eb709b0d | 373 | #endif |
1da177e4 | 374 | |
059285a2 MG |
375 | static void __lru_cache_activate_page(struct page *page) |
376 | { | |
b01b2141 | 377 | struct pagevec *pvec; |
059285a2 MG |
378 | int i; |
379 | ||
b01b2141 IM |
380 | local_lock(&lru_pvecs.lock); |
381 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
382 | ||
059285a2 MG |
383 | /* |
384 | * Search backwards on the optimistic assumption that the page being | |
385 | * activated has just been added to this pagevec. Note that only | |
386 | * the local pagevec is examined as a !PageLRU page could be in the | |
387 | * process of being released, reclaimed, migrated or on a remote | |
388 | * pagevec that is currently being drained. Furthermore, marking | |
389 | * a remote pagevec's page PageActive potentially hits a race where | |
390 | * a page is marked PageActive just after it is added to the inactive | |
391 | * list causing accounting errors and BUG_ON checks to trigger. | |
392 | */ | |
393 | for (i = pagevec_count(pvec) - 1; i >= 0; i--) { | |
394 | struct page *pagevec_page = pvec->pages[i]; | |
395 | ||
396 | if (pagevec_page == page) { | |
397 | SetPageActive(page); | |
398 | break; | |
399 | } | |
400 | } | |
401 | ||
b01b2141 | 402 | local_unlock(&lru_pvecs.lock); |
059285a2 MG |
403 | } |
404 | ||
1da177e4 LT |
405 | /* |
406 | * Mark a page as having seen activity. | |
407 | * | |
408 | * inactive,unreferenced -> inactive,referenced | |
409 | * inactive,referenced -> active,unreferenced | |
410 | * active,unreferenced -> active,referenced | |
eb39d618 HD |
411 | * |
412 | * When a newly allocated page is not yet visible, so safe for non-atomic ops, | |
413 | * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). | |
1da177e4 | 414 | */ |
920c7a5d | 415 | void mark_page_accessed(struct page *page) |
1da177e4 | 416 | { |
e90309c9 | 417 | page = compound_head(page); |
059285a2 | 418 | |
a1100a74 FW |
419 | if (!PageReferenced(page)) { |
420 | SetPageReferenced(page); | |
421 | } else if (PageUnevictable(page)) { | |
422 | /* | |
423 | * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, | |
424 | * this list is never rotated or maintained, so marking an | |
425 | * evictable page accessed has no effect. | |
426 | */ | |
427 | } else if (!PageActive(page)) { | |
059285a2 MG |
428 | /* |
429 | * If the page is on the LRU, queue it for activation via | |
b01b2141 | 430 | * lru_pvecs.activate_page. Otherwise, assume the page is on a |
059285a2 MG |
431 | * pagevec, mark it active and it'll be moved to the active |
432 | * LRU on the next drain. | |
433 | */ | |
434 | if (PageLRU(page)) | |
435 | activate_page(page); | |
436 | else | |
437 | __lru_cache_activate_page(page); | |
1da177e4 | 438 | ClearPageReferenced(page); |
9de4f22a | 439 | if (page_is_file_lru(page)) |
a528910e | 440 | workingset_activation(page); |
1da177e4 | 441 | } |
33c3fc71 VD |
442 | if (page_is_idle(page)) |
443 | clear_page_idle(page); | |
1da177e4 | 444 | } |
1da177e4 LT |
445 | EXPORT_SYMBOL(mark_page_accessed); |
446 | ||
f04e9ebb | 447 | /** |
c53954a0 | 448 | * lru_cache_add - add a page to a page list |
f04e9ebb | 449 | * @page: the page to be added to the LRU. |
2329d375 JZ |
450 | * |
451 | * Queue the page for addition to the LRU via pagevec. The decision on whether | |
452 | * to add the page to the [in]active [file|anon] list is deferred until the | |
453 | * pagevec is drained. This gives a chance for the caller of lru_cache_add() | |
454 | * have the page added to the active list using mark_page_accessed(). | |
f04e9ebb | 455 | */ |
c53954a0 | 456 | void lru_cache_add(struct page *page) |
1da177e4 | 457 | { |
6058eaec JW |
458 | struct pagevec *pvec; |
459 | ||
309381fe SL |
460 | VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); |
461 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6058eaec JW |
462 | |
463 | get_page(page); | |
464 | local_lock(&lru_pvecs.lock); | |
465 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
466 | if (!pagevec_add(pvec, page) || PageCompound(page)) | |
467 | __pagevec_lru_add(pvec); | |
468 | local_unlock(&lru_pvecs.lock); | |
1da177e4 | 469 | } |
6058eaec | 470 | EXPORT_SYMBOL(lru_cache_add); |
1da177e4 | 471 | |
00501b53 JW |
472 | /** |
473 | * lru_cache_add_active_or_unevictable | |
474 | * @page: the page to be added to LRU | |
475 | * @vma: vma in which page is mapped for determining reclaimability | |
476 | * | |
477 | * Place @page on the active or unevictable LRU list, depending on its | |
478 | * evictability. Note that if the page is not evictable, it goes | |
479 | * directly back onto it's zone's unevictable list, it does NOT use a | |
480 | * per cpu pagevec. | |
481 | */ | |
482 | void lru_cache_add_active_or_unevictable(struct page *page, | |
483 | struct vm_area_struct *vma) | |
484 | { | |
485 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
486 | ||
9c4e6b1a | 487 | if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) |
00501b53 | 488 | SetPageActive(page); |
9c4e6b1a | 489 | else if (!TestSetPageMlocked(page)) { |
00501b53 JW |
490 | /* |
491 | * We use the irq-unsafe __mod_zone_page_stat because this | |
492 | * counter is not modified from interrupt context, and the pte | |
493 | * lock is held(spinlock), which implies preemption disabled. | |
494 | */ | |
495 | __mod_zone_page_state(page_zone(page), NR_MLOCK, | |
496 | hpage_nr_pages(page)); | |
497 | count_vm_event(UNEVICTABLE_PGMLOCKED); | |
498 | } | |
9c4e6b1a | 499 | lru_cache_add(page); |
00501b53 JW |
500 | } |
501 | ||
31560180 MK |
502 | /* |
503 | * If the page can not be invalidated, it is moved to the | |
504 | * inactive list to speed up its reclaim. It is moved to the | |
505 | * head of the list, rather than the tail, to give the flusher | |
506 | * threads some time to write it out, as this is much more | |
507 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
508 | * |
509 | * If the page isn't page_mapped and dirty/writeback, the page | |
510 | * could reclaim asap using PG_reclaim. | |
511 | * | |
512 | * 1. active, mapped page -> none | |
513 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
514 | * 3. inactive, mapped page -> none | |
515 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
516 | * 5. inactive, clean -> inactive, tail | |
517 | * 6. Others -> none | |
518 | * | |
519 | * In 4, why it moves inactive's head, the VM expects the page would | |
520 | * be write it out by flusher threads as this is much more effective | |
521 | * than the single-page writeout from reclaim. | |
31560180 | 522 | */ |
cc5993bd | 523 | static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec, |
fa9add64 | 524 | void *arg) |
31560180 | 525 | { |
fbbb602e | 526 | int lru; |
278df9f4 | 527 | bool active; |
31560180 | 528 | |
278df9f4 | 529 | if (!PageLRU(page)) |
31560180 MK |
530 | return; |
531 | ||
bad49d9c MK |
532 | if (PageUnevictable(page)) |
533 | return; | |
534 | ||
31560180 MK |
535 | /* Some processes are using the page */ |
536 | if (page_mapped(page)) | |
537 | return; | |
538 | ||
278df9f4 | 539 | active = PageActive(page); |
31560180 | 540 | lru = page_lru_base_type(page); |
fa9add64 HD |
541 | |
542 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
543 | ClearPageActive(page); |
544 | ClearPageReferenced(page); | |
31560180 | 545 | |
278df9f4 MK |
546 | if (PageWriteback(page) || PageDirty(page)) { |
547 | /* | |
548 | * PG_reclaim could be raced with end_page_writeback | |
549 | * It can make readahead confusing. But race window | |
550 | * is _really_ small and it's non-critical problem. | |
551 | */ | |
e7a1aaf2 | 552 | add_page_to_lru_list(page, lruvec, lru); |
278df9f4 MK |
553 | SetPageReclaim(page); |
554 | } else { | |
555 | /* | |
556 | * The page's writeback ends up during pagevec | |
557 | * We moves tha page into tail of inactive. | |
558 | */ | |
e7a1aaf2 | 559 | add_page_to_lru_list_tail(page, lruvec, lru); |
278df9f4 MK |
560 | __count_vm_event(PGROTATED); |
561 | } | |
562 | ||
563 | if (active) | |
564 | __count_vm_event(PGDEACTIVATE); | |
31560180 MK |
565 | } |
566 | ||
9c276cc6 MK |
567 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, |
568 | void *arg) | |
569 | { | |
570 | if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { | |
9c276cc6 MK |
571 | int lru = page_lru_base_type(page); |
572 | ||
573 | del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE); | |
574 | ClearPageActive(page); | |
575 | ClearPageReferenced(page); | |
576 | add_page_to_lru_list(page, lruvec, lru); | |
577 | ||
578 | __count_vm_events(PGDEACTIVATE, hpage_nr_pages(page)); | |
9c276cc6 MK |
579 | } |
580 | } | |
10853a03 | 581 | |
f7ad2a6c | 582 | static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec, |
10853a03 MK |
583 | void *arg) |
584 | { | |
f7ad2a6c | 585 | if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 586 | !PageSwapCache(page) && !PageUnevictable(page)) { |
f7ad2a6c | 587 | bool active = PageActive(page); |
10853a03 | 588 | |
f7ad2a6c SL |
589 | del_page_from_lru_list(page, lruvec, |
590 | LRU_INACTIVE_ANON + active); | |
10853a03 MK |
591 | ClearPageActive(page); |
592 | ClearPageReferenced(page); | |
f7ad2a6c | 593 | /* |
9de4f22a HY |
594 | * Lazyfree pages are clean anonymous pages. They have |
595 | * PG_swapbacked flag cleared, to distinguish them from normal | |
596 | * anonymous pages | |
f7ad2a6c SL |
597 | */ |
598 | ClearPageSwapBacked(page); | |
599 | add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE); | |
10853a03 | 600 | |
f7ad2a6c | 601 | __count_vm_events(PGLAZYFREE, hpage_nr_pages(page)); |
2262185c | 602 | count_memcg_page_event(page, PGLAZYFREE); |
10853a03 MK |
603 | } |
604 | } | |
605 | ||
902aaed0 HH |
606 | /* |
607 | * Drain pages out of the cpu's pagevecs. | |
608 | * Either "cpu" is the current CPU, and preemption has already been | |
609 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
610 | */ | |
f0cb3c76 | 611 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 612 | { |
b01b2141 | 613 | struct pagevec *pvec = &per_cpu(lru_pvecs.lru_add, cpu); |
1da177e4 | 614 | |
13f7f789 | 615 | if (pagevec_count(pvec)) |
a0b8cab3 | 616 | __pagevec_lru_add(pvec); |
902aaed0 | 617 | |
b01b2141 | 618 | pvec = &per_cpu(lru_rotate.pvec, cpu); |
902aaed0 HH |
619 | if (pagevec_count(pvec)) { |
620 | unsigned long flags; | |
621 | ||
622 | /* No harm done if a racing interrupt already did this */ | |
b01b2141 | 623 | local_lock_irqsave(&lru_rotate.lock, flags); |
902aaed0 | 624 | pagevec_move_tail(pvec); |
b01b2141 | 625 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
902aaed0 | 626 | } |
31560180 | 627 | |
b01b2141 | 628 | pvec = &per_cpu(lru_pvecs.lru_deactivate_file, cpu); |
31560180 | 629 | if (pagevec_count(pvec)) |
cc5993bd | 630 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); |
eb709b0d | 631 | |
b01b2141 | 632 | pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu); |
9c276cc6 MK |
633 | if (pagevec_count(pvec)) |
634 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); | |
635 | ||
b01b2141 | 636 | pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu); |
10853a03 | 637 | if (pagevec_count(pvec)) |
f7ad2a6c | 638 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL); |
10853a03 | 639 | |
eb709b0d | 640 | activate_page_drain(cpu); |
31560180 MK |
641 | } |
642 | ||
643 | /** | |
cc5993bd | 644 | * deactivate_file_page - forcefully deactivate a file page |
31560180 MK |
645 | * @page: page to deactivate |
646 | * | |
647 | * This function hints the VM that @page is a good reclaim candidate, | |
648 | * for example if its invalidation fails due to the page being dirty | |
649 | * or under writeback. | |
650 | */ | |
cc5993bd | 651 | void deactivate_file_page(struct page *page) |
31560180 | 652 | { |
821ed6bb | 653 | /* |
cc5993bd MK |
654 | * In a workload with many unevictable page such as mprotect, |
655 | * unevictable page deactivation for accelerating reclaim is pointless. | |
821ed6bb MK |
656 | */ |
657 | if (PageUnevictable(page)) | |
658 | return; | |
659 | ||
31560180 | 660 | if (likely(get_page_unless_zero(page))) { |
b01b2141 IM |
661 | struct pagevec *pvec; |
662 | ||
663 | local_lock(&lru_pvecs.lock); | |
664 | pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate_file); | |
31560180 | 665 | |
8f182270 | 666 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
cc5993bd | 667 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); |
b01b2141 | 668 | local_unlock(&lru_pvecs.lock); |
31560180 | 669 | } |
80bfed90 AM |
670 | } |
671 | ||
9c276cc6 MK |
672 | /* |
673 | * deactivate_page - deactivate a page | |
674 | * @page: page to deactivate | |
675 | * | |
676 | * deactivate_page() moves @page to the inactive list if @page was on the active | |
677 | * list and was not an unevictable page. This is done to accelerate the reclaim | |
678 | * of @page. | |
679 | */ | |
680 | void deactivate_page(struct page *page) | |
681 | { | |
682 | if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { | |
b01b2141 | 683 | struct pagevec *pvec; |
9c276cc6 | 684 | |
b01b2141 IM |
685 | local_lock(&lru_pvecs.lock); |
686 | pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate); | |
9c276cc6 MK |
687 | get_page(page); |
688 | if (!pagevec_add(pvec, page) || PageCompound(page)) | |
689 | pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); | |
b01b2141 | 690 | local_unlock(&lru_pvecs.lock); |
9c276cc6 MK |
691 | } |
692 | } | |
693 | ||
10853a03 | 694 | /** |
f7ad2a6c | 695 | * mark_page_lazyfree - make an anon page lazyfree |
10853a03 MK |
696 | * @page: page to deactivate |
697 | * | |
f7ad2a6c SL |
698 | * mark_page_lazyfree() moves @page to the inactive file list. |
699 | * This is done to accelerate the reclaim of @page. | |
10853a03 | 700 | */ |
f7ad2a6c | 701 | void mark_page_lazyfree(struct page *page) |
10853a03 | 702 | { |
f7ad2a6c | 703 | if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 704 | !PageSwapCache(page) && !PageUnevictable(page)) { |
b01b2141 | 705 | struct pagevec *pvec; |
10853a03 | 706 | |
b01b2141 IM |
707 | local_lock(&lru_pvecs.lock); |
708 | pvec = this_cpu_ptr(&lru_pvecs.lru_lazyfree); | |
09cbfeaf | 709 | get_page(page); |
8f182270 | 710 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
f7ad2a6c | 711 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL); |
b01b2141 | 712 | local_unlock(&lru_pvecs.lock); |
10853a03 MK |
713 | } |
714 | } | |
715 | ||
80bfed90 AM |
716 | void lru_add_drain(void) |
717 | { | |
b01b2141 IM |
718 | local_lock(&lru_pvecs.lock); |
719 | lru_add_drain_cpu(smp_processor_id()); | |
720 | local_unlock(&lru_pvecs.lock); | |
721 | } | |
722 | ||
723 | void lru_add_drain_cpu_zone(struct zone *zone) | |
724 | { | |
725 | local_lock(&lru_pvecs.lock); | |
726 | lru_add_drain_cpu(smp_processor_id()); | |
727 | drain_local_pages(zone); | |
728 | local_unlock(&lru_pvecs.lock); | |
1da177e4 LT |
729 | } |
730 | ||
6ea183d6 MH |
731 | #ifdef CONFIG_SMP |
732 | ||
733 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); | |
734 | ||
c4028958 | 735 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
736 | { |
737 | lru_add_drain(); | |
738 | } | |
739 | ||
9852a721 MH |
740 | /* |
741 | * Doesn't need any cpu hotplug locking because we do rely on per-cpu | |
742 | * kworkers being shut down before our page_alloc_cpu_dead callback is | |
743 | * executed on the offlined cpu. | |
744 | * Calling this function with cpu hotplug locks held can actually lead | |
745 | * to obscure indirect dependencies via WQ context. | |
746 | */ | |
747 | void lru_add_drain_all(void) | |
053837fc | 748 | { |
eef1a429 | 749 | static seqcount_t seqcount = SEQCNT_ZERO(seqcount); |
5fbc4616 CM |
750 | static DEFINE_MUTEX(lock); |
751 | static struct cpumask has_work; | |
eef1a429 | 752 | int cpu, seq; |
5fbc4616 | 753 | |
ce612879 MH |
754 | /* |
755 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
756 | * initialized. | |
757 | */ | |
758 | if (WARN_ON(!mm_percpu_wq)) | |
759 | return; | |
760 | ||
eef1a429 KK |
761 | seq = raw_read_seqcount_latch(&seqcount); |
762 | ||
5fbc4616 | 763 | mutex_lock(&lock); |
eef1a429 KK |
764 | |
765 | /* | |
766 | * Piggyback on drain started and finished while we waited for lock: | |
767 | * all pages pended at the time of our enter were drained from vectors. | |
768 | */ | |
769 | if (__read_seqcount_retry(&seqcount, seq)) | |
770 | goto done; | |
771 | ||
772 | raw_write_seqcount_latch(&seqcount); | |
773 | ||
5fbc4616 CM |
774 | cpumask_clear(&has_work); |
775 | ||
776 | for_each_online_cpu(cpu) { | |
777 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); | |
778 | ||
b01b2141 IM |
779 | if (pagevec_count(&per_cpu(lru_pvecs.lru_add, cpu)) || |
780 | pagevec_count(&per_cpu(lru_rotate.pvec, cpu)) || | |
781 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) || | |
782 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) || | |
783 | pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) || | |
5fbc4616 CM |
784 | need_activate_page_drain(cpu)) { |
785 | INIT_WORK(work, lru_add_drain_per_cpu); | |
ce612879 | 786 | queue_work_on(cpu, mm_percpu_wq, work); |
5fbc4616 CM |
787 | cpumask_set_cpu(cpu, &has_work); |
788 | } | |
789 | } | |
790 | ||
791 | for_each_cpu(cpu, &has_work) | |
792 | flush_work(&per_cpu(lru_add_drain_work, cpu)); | |
793 | ||
eef1a429 | 794 | done: |
5fbc4616 | 795 | mutex_unlock(&lock); |
053837fc | 796 | } |
6ea183d6 MH |
797 | #else |
798 | void lru_add_drain_all(void) | |
799 | { | |
800 | lru_add_drain(); | |
801 | } | |
802 | #endif | |
053837fc | 803 | |
aabfb572 | 804 | /** |
ea1754a0 | 805 | * release_pages - batched put_page() |
aabfb572 MH |
806 | * @pages: array of pages to release |
807 | * @nr: number of pages | |
1da177e4 | 808 | * |
aabfb572 MH |
809 | * Decrement the reference count on all the pages in @pages. If it |
810 | * fell to zero, remove the page from the LRU and free it. | |
1da177e4 | 811 | */ |
c6f92f9f | 812 | void release_pages(struct page **pages, int nr) |
1da177e4 LT |
813 | { |
814 | int i; | |
cc59850e | 815 | LIST_HEAD(pages_to_free); |
599d0c95 | 816 | struct pglist_data *locked_pgdat = NULL; |
fa9add64 | 817 | struct lruvec *lruvec; |
902aaed0 | 818 | unsigned long uninitialized_var(flags); |
aabfb572 | 819 | unsigned int uninitialized_var(lock_batch); |
1da177e4 | 820 | |
1da177e4 LT |
821 | for (i = 0; i < nr; i++) { |
822 | struct page *page = pages[i]; | |
1da177e4 | 823 | |
aabfb572 MH |
824 | /* |
825 | * Make sure the IRQ-safe lock-holding time does not get | |
826 | * excessive with a continuous string of pages from the | |
599d0c95 | 827 | * same pgdat. The lock is held only if pgdat != NULL. |
aabfb572 | 828 | */ |
599d0c95 MG |
829 | if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) { |
830 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
831 | locked_pgdat = NULL; | |
aabfb572 MH |
832 | } |
833 | ||
6fcb52a5 | 834 | if (is_huge_zero_page(page)) |
aa88b68c | 835 | continue; |
aa88b68c | 836 | |
c5d6c45e | 837 | if (is_zone_device_page(page)) { |
df6ad698 JG |
838 | if (locked_pgdat) { |
839 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, | |
840 | flags); | |
841 | locked_pgdat = NULL; | |
842 | } | |
c5d6c45e IW |
843 | /* |
844 | * ZONE_DEVICE pages that return 'false' from | |
845 | * put_devmap_managed_page() do not require special | |
846 | * processing, and instead, expect a call to | |
847 | * put_page_testzero(). | |
848 | */ | |
07d80269 JH |
849 | if (page_is_devmap_managed(page)) { |
850 | put_devmap_managed_page(page); | |
c5d6c45e | 851 | continue; |
07d80269 | 852 | } |
df6ad698 JG |
853 | } |
854 | ||
ddc58f27 | 855 | page = compound_head(page); |
b5810039 | 856 | if (!put_page_testzero(page)) |
1da177e4 LT |
857 | continue; |
858 | ||
ddc58f27 | 859 | if (PageCompound(page)) { |
599d0c95 MG |
860 | if (locked_pgdat) { |
861 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
862 | locked_pgdat = NULL; | |
ddc58f27 KS |
863 | } |
864 | __put_compound_page(page); | |
865 | continue; | |
866 | } | |
867 | ||
46453a6e | 868 | if (PageLRU(page)) { |
599d0c95 | 869 | struct pglist_data *pgdat = page_pgdat(page); |
894bc310 | 870 | |
599d0c95 MG |
871 | if (pgdat != locked_pgdat) { |
872 | if (locked_pgdat) | |
873 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, | |
902aaed0 | 874 | flags); |
aabfb572 | 875 | lock_batch = 0; |
599d0c95 MG |
876 | locked_pgdat = pgdat; |
877 | spin_lock_irqsave(&locked_pgdat->lru_lock, flags); | |
46453a6e | 878 | } |
fa9add64 | 879 | |
599d0c95 | 880 | lruvec = mem_cgroup_page_lruvec(page, locked_pgdat); |
309381fe | 881 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
67453911 | 882 | __ClearPageLRU(page); |
fa9add64 | 883 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
884 | } |
885 | ||
c53954a0 | 886 | /* Clear Active bit in case of parallel mark_page_accessed */ |
e3741b50 | 887 | __ClearPageActive(page); |
62906027 | 888 | __ClearPageWaiters(page); |
c53954a0 | 889 | |
cc59850e | 890 | list_add(&page->lru, &pages_to_free); |
1da177e4 | 891 | } |
599d0c95 MG |
892 | if (locked_pgdat) |
893 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
1da177e4 | 894 | |
747db954 | 895 | mem_cgroup_uncharge_list(&pages_to_free); |
2d4894b5 | 896 | free_unref_page_list(&pages_to_free); |
1da177e4 | 897 | } |
0be8557b | 898 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
899 | |
900 | /* | |
901 | * The pages which we're about to release may be in the deferred lru-addition | |
902 | * queues. That would prevent them from really being freed right now. That's | |
903 | * OK from a correctness point of view but is inefficient - those pages may be | |
904 | * cache-warm and we want to give them back to the page allocator ASAP. | |
905 | * | |
906 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
907 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
908 | * mutual recursion. | |
909 | */ | |
910 | void __pagevec_release(struct pagevec *pvec) | |
911 | { | |
7f0b5fb9 | 912 | if (!pvec->percpu_pvec_drained) { |
d9ed0d08 | 913 | lru_add_drain(); |
7f0b5fb9 | 914 | pvec->percpu_pvec_drained = true; |
d9ed0d08 | 915 | } |
c6f92f9f | 916 | release_pages(pvec->pages, pagevec_count(pvec)); |
1da177e4 LT |
917 | pagevec_reinit(pvec); |
918 | } | |
7f285701 SF |
919 | EXPORT_SYMBOL(__pagevec_release); |
920 | ||
12d27107 | 921 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 | 922 | /* used by __split_huge_page_refcount() */ |
fa9add64 | 923 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
5bc7b8ac | 924 | struct lruvec *lruvec, struct list_head *list) |
71e3aac0 | 925 | { |
309381fe SL |
926 | VM_BUG_ON_PAGE(!PageHead(page), page); |
927 | VM_BUG_ON_PAGE(PageCompound(page_tail), page); | |
928 | VM_BUG_ON_PAGE(PageLRU(page_tail), page); | |
35f3aa39 | 929 | lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock); |
71e3aac0 | 930 | |
5bc7b8ac SL |
931 | if (!list) |
932 | SetPageLRU(page_tail); | |
71e3aac0 | 933 | |
12d27107 HD |
934 | if (likely(PageLRU(page))) |
935 | list_add_tail(&page_tail->lru, &page->lru); | |
5bc7b8ac SL |
936 | else if (list) { |
937 | /* page reclaim is reclaiming a huge page */ | |
938 | get_page(page_tail); | |
939 | list_add_tail(&page_tail->lru, list); | |
940 | } else { | |
12d27107 HD |
941 | /* |
942 | * Head page has not yet been counted, as an hpage, | |
943 | * so we must account for each subpage individually. | |
944 | * | |
e7a1aaf2 YZ |
945 | * Put page_tail on the list at the correct position |
946 | * so they all end up in order. | |
12d27107 | 947 | */ |
e7a1aaf2 YZ |
948 | add_page_to_lru_list_tail(page_tail, lruvec, |
949 | page_lru(page_tail)); | |
71e3aac0 AA |
950 | } |
951 | } | |
12d27107 | 952 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 953 | |
fa9add64 HD |
954 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
955 | void *arg) | |
3dd7ae8e | 956 | { |
9c4e6b1a SB |
957 | enum lru_list lru; |
958 | int was_unevictable = TestClearPageUnevictable(page); | |
3dd7ae8e | 959 | |
309381fe | 960 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3dd7ae8e | 961 | |
9c4e6b1a SB |
962 | /* |
963 | * Page becomes evictable in two ways: | |
dae966dc | 964 | * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()]. |
9c4e6b1a SB |
965 | * 2) Before acquiring LRU lock to put the page to correct LRU and then |
966 | * a) do PageLRU check with lock [check_move_unevictable_pages] | |
967 | * b) do PageLRU check before lock [clear_page_mlock] | |
968 | * | |
969 | * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need | |
970 | * following strict ordering: | |
971 | * | |
972 | * #0: __pagevec_lru_add_fn #1: clear_page_mlock | |
973 | * | |
974 | * SetPageLRU() TestClearPageMlocked() | |
975 | * smp_mb() // explicit ordering // above provides strict | |
976 | * // ordering | |
977 | * PageMlocked() PageLRU() | |
978 | * | |
979 | * | |
980 | * if '#1' does not observe setting of PG_lru by '#0' and fails | |
981 | * isolation, the explicit barrier will make sure that page_evictable | |
982 | * check will put the page in correct LRU. Without smp_mb(), SetPageLRU | |
983 | * can be reordered after PageMlocked check and can make '#1' to fail | |
984 | * the isolation of the page whose Mlocked bit is cleared (#0 is also | |
985 | * looking at the same page) and the evictable page will be stranded | |
986 | * in an unevictable LRU. | |
987 | */ | |
9a9b6cce YS |
988 | SetPageLRU(page); |
989 | smp_mb__after_atomic(); | |
9c4e6b1a SB |
990 | |
991 | if (page_evictable(page)) { | |
992 | lru = page_lru(page); | |
9c4e6b1a SB |
993 | if (was_unevictable) |
994 | count_vm_event(UNEVICTABLE_PGRESCUED); | |
995 | } else { | |
996 | lru = LRU_UNEVICTABLE; | |
997 | ClearPageActive(page); | |
998 | SetPageUnevictable(page); | |
999 | if (!was_unevictable) | |
1000 | count_vm_event(UNEVICTABLE_PGCULLED); | |
1001 | } | |
1002 | ||
fa9add64 | 1003 | add_page_to_lru_list(page, lruvec, lru); |
24b7e581 | 1004 | trace_mm_lru_insertion(page, lru); |
3dd7ae8e SL |
1005 | } |
1006 | ||
1da177e4 LT |
1007 | /* |
1008 | * Add the passed pages to the LRU, then drop the caller's refcount | |
1009 | * on them. Reinitialises the caller's pagevec. | |
1010 | */ | |
a0b8cab3 | 1011 | void __pagevec_lru_add(struct pagevec *pvec) |
1da177e4 | 1012 | { |
a0b8cab3 | 1013 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL); |
1da177e4 | 1014 | } |
1da177e4 | 1015 | |
0cd6144a JW |
1016 | /** |
1017 | * pagevec_lookup_entries - gang pagecache lookup | |
1018 | * @pvec: Where the resulting entries are placed | |
1019 | * @mapping: The address_space to search | |
1020 | * @start: The starting entry index | |
cb6f0f34 | 1021 | * @nr_entries: The maximum number of pages |
0cd6144a JW |
1022 | * @indices: The cache indices corresponding to the entries in @pvec |
1023 | * | |
1024 | * pagevec_lookup_entries() will search for and return a group of up | |
f144c390 | 1025 | * to @nr_pages pages and shadow entries in the mapping. All |
0cd6144a JW |
1026 | * entries are placed in @pvec. pagevec_lookup_entries() takes a |
1027 | * reference against actual pages in @pvec. | |
1028 | * | |
1029 | * The search returns a group of mapping-contiguous entries with | |
1030 | * ascending indexes. There may be holes in the indices due to | |
1031 | * not-present entries. | |
1032 | * | |
71725ed1 HD |
1033 | * Only one subpage of a Transparent Huge Page is returned in one call: |
1034 | * allowing truncate_inode_pages_range() to evict the whole THP without | |
1035 | * cycling through a pagevec of extra references. | |
1036 | * | |
0cd6144a JW |
1037 | * pagevec_lookup_entries() returns the number of entries which were |
1038 | * found. | |
1039 | */ | |
1040 | unsigned pagevec_lookup_entries(struct pagevec *pvec, | |
1041 | struct address_space *mapping, | |
e02a9f04 | 1042 | pgoff_t start, unsigned nr_entries, |
0cd6144a JW |
1043 | pgoff_t *indices) |
1044 | { | |
e02a9f04 | 1045 | pvec->nr = find_get_entries(mapping, start, nr_entries, |
0cd6144a JW |
1046 | pvec->pages, indices); |
1047 | return pagevec_count(pvec); | |
1048 | } | |
1049 | ||
1050 | /** | |
1051 | * pagevec_remove_exceptionals - pagevec exceptionals pruning | |
1052 | * @pvec: The pagevec to prune | |
1053 | * | |
1054 | * pagevec_lookup_entries() fills both pages and exceptional radix | |
1055 | * tree entries into the pagevec. This function prunes all | |
1056 | * exceptionals from @pvec without leaving holes, so that it can be | |
1057 | * passed on to page-only pagevec operations. | |
1058 | */ | |
1059 | void pagevec_remove_exceptionals(struct pagevec *pvec) | |
1060 | { | |
1061 | int i, j; | |
1062 | ||
1063 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
1064 | struct page *page = pvec->pages[i]; | |
3159f943 | 1065 | if (!xa_is_value(page)) |
0cd6144a JW |
1066 | pvec->pages[j++] = page; |
1067 | } | |
1068 | pvec->nr = j; | |
1069 | } | |
1070 | ||
1da177e4 | 1071 | /** |
b947cee4 | 1072 | * pagevec_lookup_range - gang pagecache lookup |
1da177e4 LT |
1073 | * @pvec: Where the resulting pages are placed |
1074 | * @mapping: The address_space to search | |
1075 | * @start: The starting page index | |
b947cee4 | 1076 | * @end: The final page index |
1da177e4 | 1077 | * |
e02a9f04 | 1078 | * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE |
b947cee4 JK |
1079 | * pages in the mapping starting from index @start and upto index @end |
1080 | * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a | |
1da177e4 LT |
1081 | * reference against the pages in @pvec. |
1082 | * | |
1083 | * The search returns a group of mapping-contiguous pages with ascending | |
d72dc8a2 JK |
1084 | * indexes. There may be holes in the indices due to not-present pages. We |
1085 | * also update @start to index the next page for the traversal. | |
1da177e4 | 1086 | * |
b947cee4 | 1087 | * pagevec_lookup_range() returns the number of pages which were found. If this |
e02a9f04 | 1088 | * number is smaller than PAGEVEC_SIZE, the end of specified range has been |
b947cee4 | 1089 | * reached. |
1da177e4 | 1090 | */ |
b947cee4 | 1091 | unsigned pagevec_lookup_range(struct pagevec *pvec, |
397162ff | 1092 | struct address_space *mapping, pgoff_t *start, pgoff_t end) |
1da177e4 | 1093 | { |
397162ff | 1094 | pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, |
b947cee4 | 1095 | pvec->pages); |
1da177e4 LT |
1096 | return pagevec_count(pvec); |
1097 | } | |
b947cee4 | 1098 | EXPORT_SYMBOL(pagevec_lookup_range); |
78539fdf | 1099 | |
72b045ae JK |
1100 | unsigned pagevec_lookup_range_tag(struct pagevec *pvec, |
1101 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1102 | xa_mark_t tag) |
1da177e4 | 1103 | { |
72b045ae | 1104 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
67fd707f | 1105 | PAGEVEC_SIZE, pvec->pages); |
1da177e4 LT |
1106 | return pagevec_count(pvec); |
1107 | } | |
72b045ae | 1108 | EXPORT_SYMBOL(pagevec_lookup_range_tag); |
1da177e4 | 1109 | |
93d3b714 JK |
1110 | unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec, |
1111 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1112 | xa_mark_t tag, unsigned max_pages) |
93d3b714 JK |
1113 | { |
1114 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, | |
1115 | min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages); | |
1116 | return pagevec_count(pvec); | |
1117 | } | |
1118 | EXPORT_SYMBOL(pagevec_lookup_range_nr_tag); | |
1da177e4 LT |
1119 | /* |
1120 | * Perform any setup for the swap system | |
1121 | */ | |
1122 | void __init swap_setup(void) | |
1123 | { | |
ca79b0c2 | 1124 | unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); |
e0bf68dd | 1125 | |
1da177e4 LT |
1126 | /* Use a smaller cluster for small-memory machines */ |
1127 | if (megs < 16) | |
1128 | page_cluster = 2; | |
1129 | else | |
1130 | page_cluster = 3; | |
1131 | /* | |
1132 | * Right now other parts of the system means that we | |
1133 | * _really_ don't want to cluster much more | |
1134 | */ | |
1da177e4 | 1135 | } |
07d80269 JH |
1136 | |
1137 | #ifdef CONFIG_DEV_PAGEMAP_OPS | |
1138 | void put_devmap_managed_page(struct page *page) | |
1139 | { | |
1140 | int count; | |
1141 | ||
1142 | if (WARN_ON_ONCE(!page_is_devmap_managed(page))) | |
1143 | return; | |
1144 | ||
1145 | count = page_ref_dec_return(page); | |
1146 | ||
1147 | /* | |
1148 | * devmap page refcounts are 1-based, rather than 0-based: if | |
1149 | * refcount is 1, then the page is free and the refcount is | |
1150 | * stable because nobody holds a reference on the page. | |
1151 | */ | |
1152 | if (count == 1) | |
1153 | free_devmap_managed_page(page); | |
1154 | else if (!count) | |
1155 | __put_page(page); | |
1156 | } | |
1157 | EXPORT_SYMBOL(put_devmap_managed_page); | |
1158 | #endif |