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, |
c7c7b80c | 207 | void (*move_fn)(struct page *page, struct lruvec *lruvec)) |
902aaed0 HH |
208 | { |
209 | int i; | |
68eb0731 | 210 | struct pglist_data *pgdat = NULL; |
fa9add64 | 211 | struct lruvec *lruvec; |
3dd7ae8e | 212 | unsigned long flags = 0; |
902aaed0 HH |
213 | |
214 | for (i = 0; i < pagevec_count(pvec); i++) { | |
215 | struct page *page = pvec->pages[i]; | |
68eb0731 | 216 | struct pglist_data *pagepgdat = page_pgdat(page); |
902aaed0 | 217 | |
68eb0731 MG |
218 | if (pagepgdat != pgdat) { |
219 | if (pgdat) | |
220 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); | |
221 | pgdat = pagepgdat; | |
222 | spin_lock_irqsave(&pgdat->lru_lock, flags); | |
902aaed0 | 223 | } |
3dd7ae8e | 224 | |
68eb0731 | 225 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
c7c7b80c | 226 | (*move_fn)(page, lruvec); |
902aaed0 | 227 | } |
68eb0731 MG |
228 | if (pgdat) |
229 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); | |
c6f92f9f | 230 | release_pages(pvec->pages, pvec->nr); |
83896fb5 | 231 | pagevec_reinit(pvec); |
d8505dee SL |
232 | } |
233 | ||
c7c7b80c | 234 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec) |
3dd7ae8e | 235 | { |
c55e8d03 JW |
236 | if (PageLRU(page) && !PageUnevictable(page)) { |
237 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
238 | ClearPageActive(page); | |
239 | add_page_to_lru_list_tail(page, lruvec, page_lru(page)); | |
c7c7b80c | 240 | __count_vm_events(PGROTATED, thp_nr_pages(page)); |
3dd7ae8e SL |
241 | } |
242 | } | |
243 | ||
1da177e4 LT |
244 | /* |
245 | * Writeback is about to end against a page which has been marked for immediate | |
246 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 247 | * inactive list. |
c7c7b80c AS |
248 | * |
249 | * rotate_reclaimable_page() must disable IRQs, to prevent nasty races. | |
1da177e4 | 250 | */ |
3dd7ae8e | 251 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 252 | { |
c55e8d03 | 253 | if (!PageLocked(page) && !PageDirty(page) && |
894bc310 | 254 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
255 | struct pagevec *pvec; |
256 | unsigned long flags; | |
257 | ||
09cbfeaf | 258 | get_page(page); |
b01b2141 IM |
259 | local_lock_irqsave(&lru_rotate.lock, flags); |
260 | pvec = this_cpu_ptr(&lru_rotate.pvec); | |
8f182270 | 261 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
c7c7b80c | 262 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); |
b01b2141 | 263 | local_unlock_irqrestore(&lru_rotate.lock, flags); |
ac6aadb2 | 264 | } |
1da177e4 LT |
265 | } |
266 | ||
96f8bf4f | 267 | void lru_note_cost(struct lruvec *lruvec, bool file, unsigned int nr_pages) |
3e2f41f1 | 268 | { |
7cf111bc JW |
269 | do { |
270 | unsigned long lrusize; | |
75cc3c91 | 271 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
7cf111bc | 272 | |
75cc3c91 | 273 | spin_lock_irq(&pgdat->lru_lock); |
7cf111bc | 274 | /* Record cost event */ |
96f8bf4f JW |
275 | if (file) |
276 | lruvec->file_cost += nr_pages; | |
7cf111bc | 277 | else |
96f8bf4f | 278 | lruvec->anon_cost += nr_pages; |
7cf111bc JW |
279 | |
280 | /* | |
281 | * Decay previous events | |
282 | * | |
283 | * Because workloads change over time (and to avoid | |
284 | * overflow) we keep these statistics as a floating | |
285 | * average, which ends up weighing recent refaults | |
286 | * more than old ones. | |
287 | */ | |
288 | lrusize = lruvec_page_state(lruvec, NR_INACTIVE_ANON) + | |
289 | lruvec_page_state(lruvec, NR_ACTIVE_ANON) + | |
290 | lruvec_page_state(lruvec, NR_INACTIVE_FILE) + | |
291 | lruvec_page_state(lruvec, NR_ACTIVE_FILE); | |
292 | ||
293 | if (lruvec->file_cost + lruvec->anon_cost > lrusize / 4) { | |
294 | lruvec->file_cost /= 2; | |
295 | lruvec->anon_cost /= 2; | |
296 | } | |
75cc3c91 | 297 | spin_unlock_irq(&pgdat->lru_lock); |
7cf111bc | 298 | } while ((lruvec = parent_lruvec(lruvec))); |
3e2f41f1 KM |
299 | } |
300 | ||
96f8bf4f JW |
301 | void lru_note_cost_page(struct page *page) |
302 | { | |
303 | lru_note_cost(mem_cgroup_page_lruvec(page, page_pgdat(page)), | |
6c357848 | 304 | page_is_file_lru(page), thp_nr_pages(page)); |
96f8bf4f JW |
305 | } |
306 | ||
c7c7b80c | 307 | static void __activate_page(struct page *page, struct lruvec *lruvec) |
1da177e4 | 308 | { |
744ed144 | 309 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 | 310 | int lru = page_lru_base_type(page); |
6c357848 | 311 | int nr_pages = thp_nr_pages(page); |
744ed144 | 312 | |
fa9add64 | 313 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
314 | SetPageActive(page); |
315 | lru += LRU_ACTIVE; | |
fa9add64 | 316 | add_page_to_lru_list(page, lruvec, lru); |
24b7e581 | 317 | trace_mm_lru_activate(page); |
4f98a2fe | 318 | |
21e330fc SB |
319 | __count_vm_events(PGACTIVATE, nr_pages); |
320 | __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, | |
321 | nr_pages); | |
1da177e4 | 322 | } |
eb709b0d SL |
323 | } |
324 | ||
325 | #ifdef CONFIG_SMP | |
eb709b0d SL |
326 | static void activate_page_drain(int cpu) |
327 | { | |
b01b2141 | 328 | struct pagevec *pvec = &per_cpu(lru_pvecs.activate_page, cpu); |
eb709b0d SL |
329 | |
330 | if (pagevec_count(pvec)) | |
c7c7b80c | 331 | pagevec_lru_move_fn(pvec, __activate_page); |
eb709b0d SL |
332 | } |
333 | ||
5fbc4616 CM |
334 | static bool need_activate_page_drain(int cpu) |
335 | { | |
b01b2141 | 336 | return pagevec_count(&per_cpu(lru_pvecs.activate_page, cpu)) != 0; |
5fbc4616 CM |
337 | } |
338 | ||
cc2828b2 | 339 | static void activate_page(struct page *page) |
eb709b0d | 340 | { |
800d8c63 | 341 | page = compound_head(page); |
eb709b0d | 342 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
b01b2141 | 343 | struct pagevec *pvec; |
eb709b0d | 344 | |
b01b2141 IM |
345 | local_lock(&lru_pvecs.lock); |
346 | pvec = this_cpu_ptr(&lru_pvecs.activate_page); | |
09cbfeaf | 347 | get_page(page); |
8f182270 | 348 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
c7c7b80c | 349 | pagevec_lru_move_fn(pvec, __activate_page); |
b01b2141 | 350 | local_unlock(&lru_pvecs.lock); |
eb709b0d SL |
351 | } |
352 | } | |
353 | ||
354 | #else | |
355 | static inline void activate_page_drain(int cpu) | |
356 | { | |
357 | } | |
358 | ||
cc2828b2 | 359 | static void activate_page(struct page *page) |
eb709b0d | 360 | { |
f4b7e272 | 361 | pg_data_t *pgdat = page_pgdat(page); |
eb709b0d | 362 | |
800d8c63 | 363 | page = compound_head(page); |
f4b7e272 | 364 | spin_lock_irq(&pgdat->lru_lock); |
c7c7b80c | 365 | __activate_page(page, mem_cgroup_page_lruvec(page, pgdat)); |
f4b7e272 | 366 | spin_unlock_irq(&pgdat->lru_lock); |
1da177e4 | 367 | } |
eb709b0d | 368 | #endif |
1da177e4 | 369 | |
059285a2 MG |
370 | static void __lru_cache_activate_page(struct page *page) |
371 | { | |
b01b2141 | 372 | struct pagevec *pvec; |
059285a2 MG |
373 | int i; |
374 | ||
b01b2141 IM |
375 | local_lock(&lru_pvecs.lock); |
376 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
377 | ||
059285a2 MG |
378 | /* |
379 | * Search backwards on the optimistic assumption that the page being | |
380 | * activated has just been added to this pagevec. Note that only | |
381 | * the local pagevec is examined as a !PageLRU page could be in the | |
382 | * process of being released, reclaimed, migrated or on a remote | |
383 | * pagevec that is currently being drained. Furthermore, marking | |
384 | * a remote pagevec's page PageActive potentially hits a race where | |
385 | * a page is marked PageActive just after it is added to the inactive | |
386 | * list causing accounting errors and BUG_ON checks to trigger. | |
387 | */ | |
388 | for (i = pagevec_count(pvec) - 1; i >= 0; i--) { | |
389 | struct page *pagevec_page = pvec->pages[i]; | |
390 | ||
391 | if (pagevec_page == page) { | |
392 | SetPageActive(page); | |
393 | break; | |
394 | } | |
395 | } | |
396 | ||
b01b2141 | 397 | local_unlock(&lru_pvecs.lock); |
059285a2 MG |
398 | } |
399 | ||
1da177e4 LT |
400 | /* |
401 | * Mark a page as having seen activity. | |
402 | * | |
403 | * inactive,unreferenced -> inactive,referenced | |
404 | * inactive,referenced -> active,unreferenced | |
405 | * active,unreferenced -> active,referenced | |
eb39d618 HD |
406 | * |
407 | * When a newly allocated page is not yet visible, so safe for non-atomic ops, | |
408 | * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). | |
1da177e4 | 409 | */ |
920c7a5d | 410 | void mark_page_accessed(struct page *page) |
1da177e4 | 411 | { |
e90309c9 | 412 | page = compound_head(page); |
059285a2 | 413 | |
a1100a74 FW |
414 | if (!PageReferenced(page)) { |
415 | SetPageReferenced(page); | |
416 | } else if (PageUnevictable(page)) { | |
417 | /* | |
418 | * Unevictable pages are on the "LRU_UNEVICTABLE" list. But, | |
419 | * this list is never rotated or maintained, so marking an | |
420 | * evictable page accessed has no effect. | |
421 | */ | |
422 | } else if (!PageActive(page)) { | |
059285a2 MG |
423 | /* |
424 | * If the page is on the LRU, queue it for activation via | |
b01b2141 | 425 | * lru_pvecs.activate_page. Otherwise, assume the page is on a |
059285a2 MG |
426 | * pagevec, mark it active and it'll be moved to the active |
427 | * LRU on the next drain. | |
428 | */ | |
429 | if (PageLRU(page)) | |
430 | activate_page(page); | |
431 | else | |
432 | __lru_cache_activate_page(page); | |
1da177e4 | 433 | ClearPageReferenced(page); |
cb686883 | 434 | workingset_activation(page); |
1da177e4 | 435 | } |
33c3fc71 VD |
436 | if (page_is_idle(page)) |
437 | clear_page_idle(page); | |
1da177e4 | 438 | } |
1da177e4 LT |
439 | EXPORT_SYMBOL(mark_page_accessed); |
440 | ||
f04e9ebb | 441 | /** |
c53954a0 | 442 | * lru_cache_add - add a page to a page list |
f04e9ebb | 443 | * @page: the page to be added to the LRU. |
2329d375 JZ |
444 | * |
445 | * Queue the page for addition to the LRU via pagevec. The decision on whether | |
446 | * to add the page to the [in]active [file|anon] list is deferred until the | |
447 | * pagevec is drained. This gives a chance for the caller of lru_cache_add() | |
448 | * have the page added to the active list using mark_page_accessed(). | |
f04e9ebb | 449 | */ |
c53954a0 | 450 | void lru_cache_add(struct page *page) |
1da177e4 | 451 | { |
6058eaec JW |
452 | struct pagevec *pvec; |
453 | ||
309381fe SL |
454 | VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); |
455 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6058eaec JW |
456 | |
457 | get_page(page); | |
458 | local_lock(&lru_pvecs.lock); | |
459 | pvec = this_cpu_ptr(&lru_pvecs.lru_add); | |
460 | if (!pagevec_add(pvec, page) || PageCompound(page)) | |
461 | __pagevec_lru_add(pvec); | |
462 | local_unlock(&lru_pvecs.lock); | |
1da177e4 | 463 | } |
6058eaec | 464 | EXPORT_SYMBOL(lru_cache_add); |
1da177e4 | 465 | |
00501b53 | 466 | /** |
b518154e | 467 | * lru_cache_add_inactive_or_unevictable |
00501b53 JW |
468 | * @page: the page to be added to LRU |
469 | * @vma: vma in which page is mapped for determining reclaimability | |
470 | * | |
b518154e | 471 | * Place @page on the inactive or unevictable LRU list, depending on its |
12eab428 | 472 | * evictability. |
00501b53 | 473 | */ |
b518154e | 474 | void lru_cache_add_inactive_or_unevictable(struct page *page, |
00501b53 JW |
475 | struct vm_area_struct *vma) |
476 | { | |
b518154e JK |
477 | bool unevictable; |
478 | ||
00501b53 JW |
479 | VM_BUG_ON_PAGE(PageLRU(page), page); |
480 | ||
b518154e JK |
481 | unevictable = (vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) == VM_LOCKED; |
482 | if (unlikely(unevictable) && !TestSetPageMlocked(page)) { | |
0964730b | 483 | int nr_pages = thp_nr_pages(page); |
00501b53 JW |
484 | /* |
485 | * We use the irq-unsafe __mod_zone_page_stat because this | |
486 | * counter is not modified from interrupt context, and the pte | |
487 | * lock is held(spinlock), which implies preemption disabled. | |
488 | */ | |
0964730b HD |
489 | __mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages); |
490 | count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages); | |
00501b53 | 491 | } |
9c4e6b1a | 492 | lru_cache_add(page); |
00501b53 JW |
493 | } |
494 | ||
31560180 MK |
495 | /* |
496 | * If the page can not be invalidated, it is moved to the | |
497 | * inactive list to speed up its reclaim. It is moved to the | |
498 | * head of the list, rather than the tail, to give the flusher | |
499 | * threads some time to write it out, as this is much more | |
500 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
501 | * |
502 | * If the page isn't page_mapped and dirty/writeback, the page | |
503 | * could reclaim asap using PG_reclaim. | |
504 | * | |
505 | * 1. active, mapped page -> none | |
506 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
507 | * 3. inactive, mapped page -> none | |
508 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
509 | * 5. inactive, clean -> inactive, tail | |
510 | * 6. Others -> none | |
511 | * | |
512 | * In 4, why it moves inactive's head, the VM expects the page would | |
513 | * be write it out by flusher threads as this is much more effective | |
514 | * than the single-page writeout from reclaim. | |
31560180 | 515 | */ |
c7c7b80c | 516 | static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec) |
31560180 | 517 | { |
fbbb602e | 518 | int lru; |
278df9f4 | 519 | bool active; |
6c357848 | 520 | int nr_pages = thp_nr_pages(page); |
31560180 | 521 | |
278df9f4 | 522 | if (!PageLRU(page)) |
31560180 MK |
523 | return; |
524 | ||
bad49d9c MK |
525 | if (PageUnevictable(page)) |
526 | return; | |
527 | ||
31560180 MK |
528 | /* Some processes are using the page */ |
529 | if (page_mapped(page)) | |
530 | return; | |
531 | ||
278df9f4 | 532 | active = PageActive(page); |
31560180 | 533 | lru = page_lru_base_type(page); |
fa9add64 HD |
534 | |
535 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
536 | ClearPageActive(page); |
537 | ClearPageReferenced(page); | |
31560180 | 538 | |
278df9f4 MK |
539 | if (PageWriteback(page) || PageDirty(page)) { |
540 | /* | |
541 | * PG_reclaim could be raced with end_page_writeback | |
542 | * It can make readahead confusing. But race window | |
543 | * is _really_ small and it's non-critical problem. | |
544 | */ | |
e7a1aaf2 | 545 | add_page_to_lru_list(page, lruvec, lru); |
278df9f4 MK |
546 | SetPageReclaim(page); |
547 | } else { | |
548 | /* | |
549 | * The page's writeback ends up during pagevec | |
550 | * We moves tha page into tail of inactive. | |
551 | */ | |
e7a1aaf2 | 552 | add_page_to_lru_list_tail(page, lruvec, lru); |
5d91f31f | 553 | __count_vm_events(PGROTATED, nr_pages); |
278df9f4 MK |
554 | } |
555 | ||
21e330fc | 556 | if (active) { |
5d91f31f | 557 | __count_vm_events(PGDEACTIVATE, nr_pages); |
21e330fc SB |
558 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, |
559 | nr_pages); | |
560 | } | |
31560180 MK |
561 | } |
562 | ||
c7c7b80c | 563 | static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec) |
9c276cc6 MK |
564 | { |
565 | if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { | |
9c276cc6 | 566 | int lru = page_lru_base_type(page); |
6c357848 | 567 | int nr_pages = thp_nr_pages(page); |
9c276cc6 MK |
568 | |
569 | del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE); | |
570 | ClearPageActive(page); | |
571 | ClearPageReferenced(page); | |
572 | add_page_to_lru_list(page, lruvec, lru); | |
573 | ||
21e330fc SB |
574 | __count_vm_events(PGDEACTIVATE, nr_pages); |
575 | __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, | |
576 | nr_pages); | |
9c276cc6 MK |
577 | } |
578 | } | |
10853a03 | 579 | |
c7c7b80c | 580 | static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec) |
10853a03 | 581 | { |
f7ad2a6c | 582 | if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 583 | !PageSwapCache(page) && !PageUnevictable(page)) { |
f7ad2a6c | 584 | bool active = PageActive(page); |
6c357848 | 585 | int nr_pages = thp_nr_pages(page); |
10853a03 | 586 | |
f7ad2a6c SL |
587 | del_page_from_lru_list(page, lruvec, |
588 | LRU_INACTIVE_ANON + active); | |
10853a03 MK |
589 | ClearPageActive(page); |
590 | ClearPageReferenced(page); | |
f7ad2a6c | 591 | /* |
9de4f22a HY |
592 | * Lazyfree pages are clean anonymous pages. They have |
593 | * PG_swapbacked flag cleared, to distinguish them from normal | |
594 | * anonymous pages | |
f7ad2a6c SL |
595 | */ |
596 | ClearPageSwapBacked(page); | |
597 | add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE); | |
10853a03 | 598 | |
21e330fc SB |
599 | __count_vm_events(PGLAZYFREE, nr_pages); |
600 | __count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, | |
601 | nr_pages); | |
10853a03 MK |
602 | } |
603 | } | |
604 | ||
902aaed0 HH |
605 | /* |
606 | * Drain pages out of the cpu's pagevecs. | |
607 | * Either "cpu" is the current CPU, and preemption has already been | |
608 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
609 | */ | |
f0cb3c76 | 610 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 611 | { |
b01b2141 | 612 | struct pagevec *pvec = &per_cpu(lru_pvecs.lru_add, cpu); |
1da177e4 | 613 | |
13f7f789 | 614 | if (pagevec_count(pvec)) |
a0b8cab3 | 615 | __pagevec_lru_add(pvec); |
902aaed0 | 616 | |
b01b2141 | 617 | pvec = &per_cpu(lru_rotate.pvec, cpu); |
7e0cc01e QC |
618 | /* Disabling interrupts below acts as a compiler barrier. */ |
619 | if (data_race(pagevec_count(pvec))) { | |
902aaed0 HH |
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); |
c7c7b80c | 624 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn); |
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)) |
c7c7b80c | 630 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); |
eb709b0d | 631 | |
b01b2141 | 632 | pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu); |
9c276cc6 | 633 | if (pagevec_count(pvec)) |
c7c7b80c | 634 | pagevec_lru_move_fn(pvec, lru_deactivate_fn); |
9c276cc6 | 635 | |
b01b2141 | 636 | pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu); |
10853a03 | 637 | if (pagevec_count(pvec)) |
c7c7b80c | 638 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn); |
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)) |
c7c7b80c | 667 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn); |
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)) | |
c7c7b80c | 689 | pagevec_lru_move_fn(pvec, lru_deactivate_fn); |
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)) |
c7c7b80c | 711 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn); |
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 | { |
6446a513 AD |
749 | /* |
750 | * lru_drain_gen - Global pages generation number | |
751 | * | |
752 | * (A) Definition: global lru_drain_gen = x implies that all generations | |
753 | * 0 < n <= x are already *scheduled* for draining. | |
754 | * | |
755 | * This is an optimization for the highly-contended use case where a | |
756 | * user space workload keeps constantly generating a flow of pages for | |
757 | * each CPU. | |
758 | */ | |
759 | static unsigned int lru_drain_gen; | |
5fbc4616 | 760 | static struct cpumask has_work; |
6446a513 AD |
761 | static DEFINE_MUTEX(lock); |
762 | unsigned cpu, this_gen; | |
5fbc4616 | 763 | |
ce612879 MH |
764 | /* |
765 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
766 | * initialized. | |
767 | */ | |
768 | if (WARN_ON(!mm_percpu_wq)) | |
769 | return; | |
770 | ||
6446a513 AD |
771 | /* |
772 | * Guarantee pagevec counter stores visible by this CPU are visible to | |
773 | * other CPUs before loading the current drain generation. | |
774 | */ | |
775 | smp_mb(); | |
776 | ||
777 | /* | |
778 | * (B) Locally cache global LRU draining generation number | |
779 | * | |
780 | * The read barrier ensures that the counter is loaded before the mutex | |
781 | * is taken. It pairs with smp_mb() inside the mutex critical section | |
782 | * at (D). | |
783 | */ | |
784 | this_gen = smp_load_acquire(&lru_drain_gen); | |
eef1a429 | 785 | |
5fbc4616 | 786 | mutex_lock(&lock); |
eef1a429 KK |
787 | |
788 | /* | |
6446a513 AD |
789 | * (C) Exit the draining operation if a newer generation, from another |
790 | * lru_add_drain_all(), was already scheduled for draining. Check (A). | |
eef1a429 | 791 | */ |
6446a513 | 792 | if (unlikely(this_gen != lru_drain_gen)) |
eef1a429 KK |
793 | goto done; |
794 | ||
6446a513 AD |
795 | /* |
796 | * (D) Increment global generation number | |
797 | * | |
798 | * Pairs with smp_load_acquire() at (B), outside of the critical | |
799 | * section. Use a full memory barrier to guarantee that the new global | |
800 | * drain generation number is stored before loading pagevec counters. | |
801 | * | |
802 | * This pairing must be done here, before the for_each_online_cpu loop | |
803 | * below which drains the page vectors. | |
804 | * | |
805 | * Let x, y, and z represent some system CPU numbers, where x < y < z. | |
806 | * Assume CPU #z is is in the middle of the for_each_online_cpu loop | |
807 | * below and has already reached CPU #y's per-cpu data. CPU #x comes | |
808 | * along, adds some pages to its per-cpu vectors, then calls | |
809 | * lru_add_drain_all(). | |
810 | * | |
811 | * If the paired barrier is done at any later step, e.g. after the | |
812 | * loop, CPU #x will just exit at (C) and miss flushing out all of its | |
813 | * added pages. | |
814 | */ | |
815 | WRITE_ONCE(lru_drain_gen, lru_drain_gen + 1); | |
816 | smp_mb(); | |
eef1a429 | 817 | |
5fbc4616 | 818 | cpumask_clear(&has_work); |
5fbc4616 CM |
819 | for_each_online_cpu(cpu) { |
820 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); | |
821 | ||
b01b2141 | 822 | if (pagevec_count(&per_cpu(lru_pvecs.lru_add, cpu)) || |
7e0cc01e | 823 | data_race(pagevec_count(&per_cpu(lru_rotate.pvec, cpu))) || |
b01b2141 IM |
824 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) || |
825 | pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) || | |
826 | pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) || | |
5fbc4616 CM |
827 | need_activate_page_drain(cpu)) { |
828 | INIT_WORK(work, lru_add_drain_per_cpu); | |
ce612879 | 829 | queue_work_on(cpu, mm_percpu_wq, work); |
6446a513 | 830 | __cpumask_set_cpu(cpu, &has_work); |
5fbc4616 CM |
831 | } |
832 | } | |
833 | ||
834 | for_each_cpu(cpu, &has_work) | |
835 | flush_work(&per_cpu(lru_add_drain_work, cpu)); | |
836 | ||
eef1a429 | 837 | done: |
5fbc4616 | 838 | mutex_unlock(&lock); |
053837fc | 839 | } |
6ea183d6 MH |
840 | #else |
841 | void lru_add_drain_all(void) | |
842 | { | |
843 | lru_add_drain(); | |
844 | } | |
6446a513 | 845 | #endif /* CONFIG_SMP */ |
053837fc | 846 | |
aabfb572 | 847 | /** |
ea1754a0 | 848 | * release_pages - batched put_page() |
aabfb572 MH |
849 | * @pages: array of pages to release |
850 | * @nr: number of pages | |
1da177e4 | 851 | * |
aabfb572 MH |
852 | * Decrement the reference count on all the pages in @pages. If it |
853 | * fell to zero, remove the page from the LRU and free it. | |
1da177e4 | 854 | */ |
c6f92f9f | 855 | void release_pages(struct page **pages, int nr) |
1da177e4 LT |
856 | { |
857 | int i; | |
cc59850e | 858 | LIST_HEAD(pages_to_free); |
599d0c95 | 859 | struct pglist_data *locked_pgdat = NULL; |
fa9add64 | 860 | struct lruvec *lruvec; |
3f649ab7 KC |
861 | unsigned long flags; |
862 | unsigned int lock_batch; | |
1da177e4 | 863 | |
1da177e4 LT |
864 | for (i = 0; i < nr; i++) { |
865 | struct page *page = pages[i]; | |
1da177e4 | 866 | |
aabfb572 MH |
867 | /* |
868 | * Make sure the IRQ-safe lock-holding time does not get | |
869 | * excessive with a continuous string of pages from the | |
599d0c95 | 870 | * same pgdat. The lock is held only if pgdat != NULL. |
aabfb572 | 871 | */ |
599d0c95 MG |
872 | if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) { |
873 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
874 | locked_pgdat = NULL; | |
aabfb572 MH |
875 | } |
876 | ||
a9b576f7 | 877 | page = compound_head(page); |
6fcb52a5 | 878 | if (is_huge_zero_page(page)) |
aa88b68c | 879 | continue; |
aa88b68c | 880 | |
c5d6c45e | 881 | if (is_zone_device_page(page)) { |
df6ad698 JG |
882 | if (locked_pgdat) { |
883 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, | |
884 | flags); | |
885 | locked_pgdat = NULL; | |
886 | } | |
c5d6c45e IW |
887 | /* |
888 | * ZONE_DEVICE pages that return 'false' from | |
a3e7bea0 | 889 | * page_is_devmap_managed() do not require special |
c5d6c45e IW |
890 | * processing, and instead, expect a call to |
891 | * put_page_testzero(). | |
892 | */ | |
07d80269 JH |
893 | if (page_is_devmap_managed(page)) { |
894 | put_devmap_managed_page(page); | |
c5d6c45e | 895 | continue; |
07d80269 | 896 | } |
43fbdeb3 RC |
897 | if (put_page_testzero(page)) |
898 | put_dev_pagemap(page->pgmap); | |
899 | continue; | |
df6ad698 JG |
900 | } |
901 | ||
b5810039 | 902 | if (!put_page_testzero(page)) |
1da177e4 LT |
903 | continue; |
904 | ||
ddc58f27 | 905 | if (PageCompound(page)) { |
599d0c95 MG |
906 | if (locked_pgdat) { |
907 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
908 | locked_pgdat = NULL; | |
ddc58f27 KS |
909 | } |
910 | __put_compound_page(page); | |
911 | continue; | |
912 | } | |
913 | ||
46453a6e | 914 | if (PageLRU(page)) { |
599d0c95 | 915 | struct pglist_data *pgdat = page_pgdat(page); |
894bc310 | 916 | |
599d0c95 MG |
917 | if (pgdat != locked_pgdat) { |
918 | if (locked_pgdat) | |
919 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, | |
902aaed0 | 920 | flags); |
aabfb572 | 921 | lock_batch = 0; |
599d0c95 MG |
922 | locked_pgdat = pgdat; |
923 | spin_lock_irqsave(&locked_pgdat->lru_lock, flags); | |
46453a6e | 924 | } |
fa9add64 | 925 | |
599d0c95 | 926 | lruvec = mem_cgroup_page_lruvec(page, locked_pgdat); |
309381fe | 927 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
67453911 | 928 | __ClearPageLRU(page); |
fa9add64 | 929 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
930 | } |
931 | ||
62906027 | 932 | __ClearPageWaiters(page); |
c53954a0 | 933 | |
cc59850e | 934 | list_add(&page->lru, &pages_to_free); |
1da177e4 | 935 | } |
599d0c95 MG |
936 | if (locked_pgdat) |
937 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
1da177e4 | 938 | |
747db954 | 939 | mem_cgroup_uncharge_list(&pages_to_free); |
2d4894b5 | 940 | free_unref_page_list(&pages_to_free); |
1da177e4 | 941 | } |
0be8557b | 942 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
943 | |
944 | /* | |
945 | * The pages which we're about to release may be in the deferred lru-addition | |
946 | * queues. That would prevent them from really being freed right now. That's | |
947 | * OK from a correctness point of view but is inefficient - those pages may be | |
948 | * cache-warm and we want to give them back to the page allocator ASAP. | |
949 | * | |
950 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
951 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
952 | * mutual recursion. | |
953 | */ | |
954 | void __pagevec_release(struct pagevec *pvec) | |
955 | { | |
7f0b5fb9 | 956 | if (!pvec->percpu_pvec_drained) { |
d9ed0d08 | 957 | lru_add_drain(); |
7f0b5fb9 | 958 | pvec->percpu_pvec_drained = true; |
d9ed0d08 | 959 | } |
c6f92f9f | 960 | release_pages(pvec->pages, pagevec_count(pvec)); |
1da177e4 LT |
961 | pagevec_reinit(pvec); |
962 | } | |
7f285701 SF |
963 | EXPORT_SYMBOL(__pagevec_release); |
964 | ||
c7c7b80c | 965 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec) |
3dd7ae8e | 966 | { |
9c4e6b1a SB |
967 | enum lru_list lru; |
968 | int was_unevictable = TestClearPageUnevictable(page); | |
6c357848 | 969 | int nr_pages = thp_nr_pages(page); |
3dd7ae8e | 970 | |
309381fe | 971 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3dd7ae8e | 972 | |
9c4e6b1a SB |
973 | /* |
974 | * Page becomes evictable in two ways: | |
dae966dc | 975 | * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()]. |
9c4e6b1a SB |
976 | * 2) Before acquiring LRU lock to put the page to correct LRU and then |
977 | * a) do PageLRU check with lock [check_move_unevictable_pages] | |
978 | * b) do PageLRU check before lock [clear_page_mlock] | |
979 | * | |
980 | * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need | |
981 | * following strict ordering: | |
982 | * | |
983 | * #0: __pagevec_lru_add_fn #1: clear_page_mlock | |
984 | * | |
985 | * SetPageLRU() TestClearPageMlocked() | |
986 | * smp_mb() // explicit ordering // above provides strict | |
987 | * // ordering | |
988 | * PageMlocked() PageLRU() | |
989 | * | |
990 | * | |
991 | * if '#1' does not observe setting of PG_lru by '#0' and fails | |
992 | * isolation, the explicit barrier will make sure that page_evictable | |
993 | * check will put the page in correct LRU. Without smp_mb(), SetPageLRU | |
994 | * can be reordered after PageMlocked check and can make '#1' to fail | |
995 | * the isolation of the page whose Mlocked bit is cleared (#0 is also | |
996 | * looking at the same page) and the evictable page will be stranded | |
997 | * in an unevictable LRU. | |
998 | */ | |
9a9b6cce YS |
999 | SetPageLRU(page); |
1000 | smp_mb__after_atomic(); | |
9c4e6b1a SB |
1001 | |
1002 | if (page_evictable(page)) { | |
1003 | lru = page_lru(page); | |
9c4e6b1a | 1004 | if (was_unevictable) |
5d91f31f | 1005 | __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages); |
9c4e6b1a SB |
1006 | } else { |
1007 | lru = LRU_UNEVICTABLE; | |
1008 | ClearPageActive(page); | |
1009 | SetPageUnevictable(page); | |
1010 | if (!was_unevictable) | |
5d91f31f | 1011 | __count_vm_events(UNEVICTABLE_PGCULLED, nr_pages); |
9c4e6b1a SB |
1012 | } |
1013 | ||
fa9add64 | 1014 | add_page_to_lru_list(page, lruvec, lru); |
24b7e581 | 1015 | trace_mm_lru_insertion(page, lru); |
3dd7ae8e SL |
1016 | } |
1017 | ||
1da177e4 LT |
1018 | /* |
1019 | * Add the passed pages to the LRU, then drop the caller's refcount | |
1020 | * on them. Reinitialises the caller's pagevec. | |
1021 | */ | |
a0b8cab3 | 1022 | void __pagevec_lru_add(struct pagevec *pvec) |
1da177e4 | 1023 | { |
c7c7b80c | 1024 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn); |
1da177e4 | 1025 | } |
1da177e4 | 1026 | |
0cd6144a JW |
1027 | /** |
1028 | * pagevec_lookup_entries - gang pagecache lookup | |
1029 | * @pvec: Where the resulting entries are placed | |
1030 | * @mapping: The address_space to search | |
1031 | * @start: The starting entry index | |
cb6f0f34 | 1032 | * @nr_entries: The maximum number of pages |
0cd6144a JW |
1033 | * @indices: The cache indices corresponding to the entries in @pvec |
1034 | * | |
1035 | * pagevec_lookup_entries() will search for and return a group of up | |
f144c390 | 1036 | * to @nr_pages pages and shadow entries in the mapping. All |
0cd6144a JW |
1037 | * entries are placed in @pvec. pagevec_lookup_entries() takes a |
1038 | * reference against actual pages in @pvec. | |
1039 | * | |
1040 | * The search returns a group of mapping-contiguous entries with | |
1041 | * ascending indexes. There may be holes in the indices due to | |
1042 | * not-present entries. | |
1043 | * | |
71725ed1 HD |
1044 | * Only one subpage of a Transparent Huge Page is returned in one call: |
1045 | * allowing truncate_inode_pages_range() to evict the whole THP without | |
1046 | * cycling through a pagevec of extra references. | |
1047 | * | |
0cd6144a JW |
1048 | * pagevec_lookup_entries() returns the number of entries which were |
1049 | * found. | |
1050 | */ | |
1051 | unsigned pagevec_lookup_entries(struct pagevec *pvec, | |
1052 | struct address_space *mapping, | |
e02a9f04 | 1053 | pgoff_t start, unsigned nr_entries, |
0cd6144a JW |
1054 | pgoff_t *indices) |
1055 | { | |
e02a9f04 | 1056 | pvec->nr = find_get_entries(mapping, start, nr_entries, |
0cd6144a JW |
1057 | pvec->pages, indices); |
1058 | return pagevec_count(pvec); | |
1059 | } | |
1060 | ||
1061 | /** | |
1062 | * pagevec_remove_exceptionals - pagevec exceptionals pruning | |
1063 | * @pvec: The pagevec to prune | |
1064 | * | |
1065 | * pagevec_lookup_entries() fills both pages and exceptional radix | |
1066 | * tree entries into the pagevec. This function prunes all | |
1067 | * exceptionals from @pvec without leaving holes, so that it can be | |
1068 | * passed on to page-only pagevec operations. | |
1069 | */ | |
1070 | void pagevec_remove_exceptionals(struct pagevec *pvec) | |
1071 | { | |
1072 | int i, j; | |
1073 | ||
1074 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
1075 | struct page *page = pvec->pages[i]; | |
3159f943 | 1076 | if (!xa_is_value(page)) |
0cd6144a JW |
1077 | pvec->pages[j++] = page; |
1078 | } | |
1079 | pvec->nr = j; | |
1080 | } | |
1081 | ||
1da177e4 | 1082 | /** |
b947cee4 | 1083 | * pagevec_lookup_range - gang pagecache lookup |
1da177e4 LT |
1084 | * @pvec: Where the resulting pages are placed |
1085 | * @mapping: The address_space to search | |
1086 | * @start: The starting page index | |
b947cee4 | 1087 | * @end: The final page index |
1da177e4 | 1088 | * |
e02a9f04 | 1089 | * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE |
b947cee4 JK |
1090 | * pages in the mapping starting from index @start and upto index @end |
1091 | * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a | |
1da177e4 LT |
1092 | * reference against the pages in @pvec. |
1093 | * | |
1094 | * The search returns a group of mapping-contiguous pages with ascending | |
d72dc8a2 JK |
1095 | * indexes. There may be holes in the indices due to not-present pages. We |
1096 | * also update @start to index the next page for the traversal. | |
1da177e4 | 1097 | * |
b947cee4 | 1098 | * pagevec_lookup_range() returns the number of pages which were found. If this |
e02a9f04 | 1099 | * number is smaller than PAGEVEC_SIZE, the end of specified range has been |
b947cee4 | 1100 | * reached. |
1da177e4 | 1101 | */ |
b947cee4 | 1102 | unsigned pagevec_lookup_range(struct pagevec *pvec, |
397162ff | 1103 | struct address_space *mapping, pgoff_t *start, pgoff_t end) |
1da177e4 | 1104 | { |
397162ff | 1105 | pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, |
b947cee4 | 1106 | pvec->pages); |
1da177e4 LT |
1107 | return pagevec_count(pvec); |
1108 | } | |
b947cee4 | 1109 | EXPORT_SYMBOL(pagevec_lookup_range); |
78539fdf | 1110 | |
72b045ae JK |
1111 | unsigned pagevec_lookup_range_tag(struct pagevec *pvec, |
1112 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1113 | xa_mark_t tag) |
1da177e4 | 1114 | { |
72b045ae | 1115 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
67fd707f | 1116 | PAGEVEC_SIZE, pvec->pages); |
1da177e4 LT |
1117 | return pagevec_count(pvec); |
1118 | } | |
72b045ae | 1119 | EXPORT_SYMBOL(pagevec_lookup_range_tag); |
1da177e4 | 1120 | |
1da177e4 LT |
1121 | /* |
1122 | * Perform any setup for the swap system | |
1123 | */ | |
1124 | void __init swap_setup(void) | |
1125 | { | |
ca79b0c2 | 1126 | unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT); |
e0bf68dd | 1127 | |
1da177e4 LT |
1128 | /* Use a smaller cluster for small-memory machines */ |
1129 | if (megs < 16) | |
1130 | page_cluster = 2; | |
1131 | else | |
1132 | page_cluster = 3; | |
1133 | /* | |
1134 | * Right now other parts of the system means that we | |
1135 | * _really_ don't want to cluster much more | |
1136 | */ | |
1da177e4 | 1137 | } |
07d80269 JH |
1138 | |
1139 | #ifdef CONFIG_DEV_PAGEMAP_OPS | |
1140 | void put_devmap_managed_page(struct page *page) | |
1141 | { | |
1142 | int count; | |
1143 | ||
1144 | if (WARN_ON_ONCE(!page_is_devmap_managed(page))) | |
1145 | return; | |
1146 | ||
1147 | count = page_ref_dec_return(page); | |
1148 | ||
1149 | /* | |
1150 | * devmap page refcounts are 1-based, rather than 0-based: if | |
1151 | * refcount is 1, then the page is free and the refcount is | |
1152 | * stable because nobody holds a reference on the page. | |
1153 | */ | |
1154 | if (count == 1) | |
1155 | free_devmap_managed_page(page); | |
1156 | else if (!count) | |
1157 | __put_page(page); | |
1158 | } | |
1159 | EXPORT_SYMBOL(put_devmap_managed_page); | |
1160 | #endif |