blob: 09fe5e97714a2ac9756a537c78b617d9ed7a1ffc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swap.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
Simon Arlott183ff222007-10-20 01:27:18 +02008 * This file contains the default values for the operation of the
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Linux VM subsystem. Fine-tuning documentation can be found in
10 * Documentation/sysctl/vm.txt.
11 * Started 18.12.91
12 * Swap aging added 23.2.95, Stephen Tweedie.
13 * Buffermem limits added 12.3.98, Rik van Riel.
14 */
15
16#include <linux/mm.h>
17#include <linux/sched.h>
18#include <linux/kernel_stat.h>
19#include <linux/swap.h>
20#include <linux/mman.h>
21#include <linux/pagemap.h>
22#include <linux/pagevec.h>
23#include <linux/init.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040024#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/mm_inline.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/percpu_counter.h>
Dan Williams3565fce2016-01-15 16:56:55 -080027#include <linux/memremap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/percpu.h>
29#include <linux/cpu.h>
30#include <linux/notifier.h>
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -070031#include <linux/backing-dev.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/memcontrol.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090033#include <linux/gfp.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070034#include <linux/uio.h>
Naoya Horiguchi822fc612015-04-15 16:14:35 -070035#include <linux/hugetlb.h>
Vladimir Davydov33c3fc72015-09-09 15:35:45 -070036#include <linux/page_idle.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Lee Schermerhorn64d65192008-10-18 20:26:52 -070038#include "internal.h"
39
Mel Gormanc6286c92013-07-03 15:02:26 -070040#define CREATE_TRACE_POINTS
41#include <trace/events/pagemap.h>
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* How many pages do we try to swap or page in/out together? */
44int page_cluster;
45
Mel Gorman13f7f782013-07-03 15:02:28 -070046static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
Vegard Nossumf84f95042008-07-23 21:28:14 -070047static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
Minchan Kimcc5993b2015-04-15 16:13:26 -070048static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
Minchan Kim10853a02016-01-15 16:55:11 -080049static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070050
Adrian Bunkb2213852006-09-25 23:31:02 -070051/*
52 * This path almost never happens for VM activity - pages are normally
53 * freed via pagevecs. But it gets used by networking.
54 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080055static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070056{
57 if (PageLRU(page)) {
Adrian Bunkb2213852006-09-25 23:31:02 -070058 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070059 struct lruvec *lruvec;
60 unsigned long flags;
Adrian Bunkb2213852006-09-25 23:31:02 -070061
62 spin_lock_irqsave(&zone->lru_lock, flags);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070063 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -080064 VM_BUG_ON_PAGE(!PageLRU(page), page);
Adrian Bunkb2213852006-09-25 23:31:02 -070065 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070066 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070067 spin_unlock_irqrestore(&zone->lru_lock, flags);
68 }
Johannes Weiner0a31bc92014-08-08 14:19:22 -070069 mem_cgroup_uncharge(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080070}
71
72static void __put_single_page(struct page *page)
73{
74 __page_cache_release(page);
Mel Gormanb745bc82014-06-04 16:10:22 -070075 free_hot_cold_page(page, false);
Adrian Bunkb2213852006-09-25 23:31:02 -070076}
77
Andrea Arcangeli91807062011-01-13 15:46:32 -080078static void __put_compound_page(struct page *page)
79{
80 compound_page_dtor *dtor;
81
Naoya Horiguchi822fc612015-04-15 16:14:35 -070082 /*
83 * __page_cache_release() is supposed to be called for thp, not for
84 * hugetlb. This is because hugetlb page does never have PageLRU set
85 * (it's never listed to any LRU lists) and no memcg routines should
86 * be called for hugetlb (it has a separate hugetlb_cgroup.)
87 */
88 if (!PageHuge(page))
89 __page_cache_release(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080090 dtor = get_compound_page_dtor(page);
91 (*dtor)(page);
92}
93
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -080094void __put_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095{
Nick Piggin8519fb32006-02-07 12:58:52 -080096 if (unlikely(PageCompound(page)))
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -080097 __put_compound_page(page);
98 else
Andrea Arcangeli91807062011-01-13 15:46:32 -080099 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100}
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800101EXPORT_SYMBOL(__put_page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700102
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700103/**
Randy Dunlap76824862008-03-19 17:00:40 -0700104 * put_pages_list() - release a list of pages
105 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700106 *
107 * Release a list of pages which are strung together on page.lru. Currently
108 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700109 */
110void put_pages_list(struct list_head *pages)
111{
112 while (!list_empty(pages)) {
113 struct page *victim;
114
115 victim = list_entry(pages->prev, struct page, lru);
116 list_del(&victim->lru);
117 page_cache_release(victim);
118 }
119}
120EXPORT_SYMBOL(put_pages_list);
121
Mel Gorman18022c52012-07-31 16:44:51 -0700122/*
123 * get_kernel_pages() - pin kernel pages in memory
124 * @kiov: An array of struct kvec structures
125 * @nr_segs: number of segments to pin
126 * @write: pinning for read/write, currently ignored
127 * @pages: array that receives pointers to the pages pinned.
128 * Should be at least nr_segs long.
129 *
130 * Returns number of pages pinned. This may be fewer than the number
131 * requested. If nr_pages is 0 or negative, returns 0. If no pages
132 * were pinned, returns -errno. Each page returned must be released
133 * with a put_page() call when it is finished with.
134 */
135int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
136 struct page **pages)
137{
138 int seg;
139
140 for (seg = 0; seg < nr_segs; seg++) {
141 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
142 return seg;
143
Mel Gorman5a178112012-07-31 16:45:02 -0700144 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700145 page_cache_get(pages[seg]);
146 }
147
148 return seg;
149}
150EXPORT_SYMBOL_GPL(get_kernel_pages);
151
152/*
153 * get_kernel_page() - pin a kernel page in memory
154 * @start: starting kernel address
155 * @write: pinning for read/write, currently ignored
156 * @pages: array that receives pointer to the page pinned.
157 * Must be at least nr_segs long.
158 *
159 * Returns 1 if page is pinned. If the page was not pinned, returns
160 * -errno. The page returned must be released with a put_page() call
161 * when it is finished with.
162 */
163int get_kernel_page(unsigned long start, int write, struct page **pages)
164{
165 const struct kvec kiov = {
166 .iov_base = (void *)start,
167 .iov_len = PAGE_SIZE
168 };
169
170 return get_kernel_pages(&kiov, 1, write, pages);
171}
172EXPORT_SYMBOL_GPL(get_kernel_page);
173
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700174static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700175 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
176 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700177{
178 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700179 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700180 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700181 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700182
183 for (i = 0; i < pagevec_count(pvec); i++) {
184 struct page *page = pvec->pages[i];
185 struct zone *pagezone = page_zone(page);
186
187 if (pagezone != zone) {
188 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700189 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700190 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700191 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700192 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700193
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700194 lruvec = mem_cgroup_page_lruvec(page, zone);
195 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700196 }
197 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700198 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800199 release_pages(pvec->pages, pvec->nr, pvec->cold);
200 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800201}
202
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700203static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
204 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700205{
206 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700207
208 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
209 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800210 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700211 (*pgmoved)++;
212 }
213}
214
215/*
216 * pagevec_move_tail() must be called with IRQ disabled.
217 * Otherwise this may cause nasty races.
218 */
219static void pagevec_move_tail(struct pagevec *pvec)
220{
221 int pgmoved = 0;
222
223 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
224 __count_vm_events(PGROTATED, pgmoved);
225}
226
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700227/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 * Writeback is about to end against a page which has been marked for immediate
229 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700230 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700232void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700234 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700235 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700236 struct pagevec *pvec;
237 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700239 page_cache_get(page);
240 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700241 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700242 if (!pagevec_add(pvec, page))
243 pagevec_move_tail(pvec);
244 local_irq_restore(flags);
245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700248static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800249 int file, int rotated)
250{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700251 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800252
253 reclaim_stat->recent_scanned[file]++;
254 if (rotated)
255 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800256}
257
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700258static void __activate_page(struct page *page, struct lruvec *lruvec,
259 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Linus Torvalds7a608572011-01-17 14:42:19 -0800261 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
262 int file = page_is_file_cache(page);
263 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800264
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700265 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800266 SetPageActive(page);
267 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700268 add_page_to_lru_list(page, lruvec, lru);
Mel Gorman24b7e582014-08-06 16:07:11 -0700269 trace_mm_lru_activate(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800270
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700271 __count_vm_event(PGACTIVATE);
272 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800273 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700274}
275
276#ifdef CONFIG_SMP
277static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
278
279static void activate_page_drain(int cpu)
280{
281 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
282
283 if (pagevec_count(pvec))
284 pagevec_lru_move_fn(pvec, __activate_page, NULL);
285}
286
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700287static bool need_activate_page_drain(int cpu)
288{
289 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
290}
291
Shaohua Lieb709b02011-05-24 17:12:55 -0700292void activate_page(struct page *page)
293{
294 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
295 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
296
297 page_cache_get(page);
298 if (!pagevec_add(pvec, page))
299 pagevec_lru_move_fn(pvec, __activate_page, NULL);
300 put_cpu_var(activate_page_pvecs);
301 }
302}
303
304#else
305static inline void activate_page_drain(int cpu)
306{
307}
308
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700309static bool need_activate_page_drain(int cpu)
310{
311 return false;
312}
313
Shaohua Lieb709b02011-05-24 17:12:55 -0700314void activate_page(struct page *page)
315{
316 struct zone *zone = page_zone(page);
317
318 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700319 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 spin_unlock_irq(&zone->lru_lock);
321}
Shaohua Lieb709b02011-05-24 17:12:55 -0700322#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Mel Gorman059285a2013-07-03 15:02:30 -0700324static void __lru_cache_activate_page(struct page *page)
325{
326 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
327 int i;
328
329 /*
330 * Search backwards on the optimistic assumption that the page being
331 * activated has just been added to this pagevec. Note that only
332 * the local pagevec is examined as a !PageLRU page could be in the
333 * process of being released, reclaimed, migrated or on a remote
334 * pagevec that is currently being drained. Furthermore, marking
335 * a remote pagevec's page PageActive potentially hits a race where
336 * a page is marked PageActive just after it is added to the inactive
337 * list causing accounting errors and BUG_ON checks to trigger.
338 */
339 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
340 struct page *pagevec_page = pvec->pages[i];
341
342 if (pagevec_page == page) {
343 SetPageActive(page);
344 break;
345 }
346 }
347
348 put_cpu_var(lru_add_pvec);
349}
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351/*
352 * Mark a page as having seen activity.
353 *
354 * inactive,unreferenced -> inactive,referenced
355 * inactive,referenced -> active,unreferenced
356 * active,unreferenced -> active,referenced
Hugh Dickinseb39d612014-08-06 16:06:43 -0700357 *
358 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
359 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800361void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800363 page = compound_head(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700364 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700365 PageReferenced(page)) {
366
367 /*
368 * If the page is on the LRU, queue it for activation via
369 * activate_page_pvecs. Otherwise, assume the page is on a
370 * pagevec, mark it active and it'll be moved to the active
371 * LRU on the next drain.
372 */
373 if (PageLRU(page))
374 activate_page(page);
375 else
376 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700378 if (page_is_file_cache(page))
379 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 } else if (!PageReferenced(page)) {
381 SetPageReferenced(page);
382 }
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700383 if (page_is_idle(page))
384 clear_page_idle(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386EXPORT_SYMBOL(mark_page_accessed);
387
Jianyu Zhan2329d372014-06-04 16:07:31 -0700388static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Mel Gorman13f7f782013-07-03 15:02:28 -0700390 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700393 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700394 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700395 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700396 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700398
399/**
400 * lru_cache_add: add a page to the page lists
401 * @page: the page to add
402 */
403void lru_cache_add_anon(struct page *page)
404{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700405 if (PageActive(page))
406 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700407 __lru_cache_add(page);
408}
409
410void lru_cache_add_file(struct page *page)
411{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700412 if (PageActive(page))
413 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700414 __lru_cache_add(page);
415}
416EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700418/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700419 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700420 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700421 *
422 * Queue the page for addition to the LRU via pagevec. The decision on whether
423 * to add the page to the [in]active [file|anon] list is deferred until the
424 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
425 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700426 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700427void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
Sasha Levin309381fea2014-01-23 15:52:54 -0800429 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
430 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700431 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700434/**
435 * add_page_to_unevictable_list - add a page to the unevictable list
436 * @page: the page to be added to the unevictable list
437 *
438 * Add page directly to its zone's unevictable list. To avoid races with
439 * tasks that might be making the page evictable, through eg. munlock,
440 * munmap or exit, while it's not on the lru, we want to add the page
441 * while it's locked or otherwise "invisible" to other tasks. This is
442 * difficult to do when using the pagevec cache, so bypass that.
443 */
444void add_page_to_unevictable_list(struct page *page)
445{
446 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700447 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700448
449 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700450 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700451 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700452 SetPageUnevictable(page);
453 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700454 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700455 spin_unlock_irq(&zone->lru_lock);
456}
457
Johannes Weiner00501b52014-08-08 14:19:20 -0700458/**
459 * lru_cache_add_active_or_unevictable
460 * @page: the page to be added to LRU
461 * @vma: vma in which page is mapped for determining reclaimability
462 *
463 * Place @page on the active or unevictable LRU list, depending on its
464 * evictability. Note that if the page is not evictable, it goes
465 * directly back onto it's zone's unevictable list, it does NOT use a
466 * per cpu pagevec.
467 */
468void lru_cache_add_active_or_unevictable(struct page *page,
469 struct vm_area_struct *vma)
470{
471 VM_BUG_ON_PAGE(PageLRU(page), page);
472
473 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
474 SetPageActive(page);
475 lru_cache_add(page);
476 return;
477 }
478
479 if (!TestSetPageMlocked(page)) {
480 /*
481 * We use the irq-unsafe __mod_zone_page_stat because this
482 * counter is not modified from interrupt context, and the pte
483 * lock is held(spinlock), which implies preemption disabled.
484 */
485 __mod_zone_page_state(page_zone(page), NR_MLOCK,
486 hpage_nr_pages(page));
487 count_vm_event(UNEVICTABLE_PGMLOCKED);
488 }
489 add_page_to_unevictable_list(page);
490}
491
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700492/*
Minchan Kim31560182011-03-22 16:32:52 -0700493 * If the page can not be invalidated, it is moved to the
494 * inactive list to speed up its reclaim. It is moved to the
495 * head of the list, rather than the tail, to give the flusher
496 * threads some time to write it out, as this is much more
497 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700498 *
499 * If the page isn't page_mapped and dirty/writeback, the page
500 * could reclaim asap using PG_reclaim.
501 *
502 * 1. active, mapped page -> none
503 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
504 * 3. inactive, mapped page -> none
505 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
506 * 5. inactive, clean -> inactive, tail
507 * 6. Others -> none
508 *
509 * In 4, why it moves inactive's head, the VM expects the page would
510 * be write it out by flusher threads as this is much more effective
511 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700512 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700513static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700514 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700515{
516 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700517 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700518
Minchan Kim278df9f2011-03-22 16:32:54 -0700519 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700520 return;
521
Minchan Kimbad49d92011-05-11 15:13:30 -0700522 if (PageUnevictable(page))
523 return;
524
Minchan Kim31560182011-03-22 16:32:52 -0700525 /* Some processes are using the page */
526 if (page_mapped(page))
527 return;
528
Minchan Kim278df9f2011-03-22 16:32:54 -0700529 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700530 file = page_is_file_cache(page);
531 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700532
533 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700534 ClearPageActive(page);
535 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700536 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700537
Minchan Kim278df9f2011-03-22 16:32:54 -0700538 if (PageWriteback(page) || PageDirty(page)) {
539 /*
540 * PG_reclaim could be raced with end_page_writeback
541 * It can make readahead confusing. But race window
542 * is _really_ small and it's non-critical problem.
543 */
544 SetPageReclaim(page);
545 } else {
546 /*
547 * The page's writeback ends up during pagevec
548 * We moves tha page into tail of inactive.
549 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800550 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700551 __count_vm_event(PGROTATED);
552 }
553
554 if (active)
555 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700556 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700557}
558
Minchan Kim10853a02016-01-15 16:55:11 -0800559
560static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
561 void *arg)
562{
563 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
564 int file = page_is_file_cache(page);
565 int lru = page_lru_base_type(page);
566
567 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
568 ClearPageActive(page);
569 ClearPageReferenced(page);
570 add_page_to_lru_list(page, lruvec, lru);
571
572 __count_vm_event(PGDEACTIVATE);
573 update_page_reclaim_stat(lruvec, file, 0);
574 }
575}
576
Minchan Kim31560182011-03-22 16:32:52 -0700577/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700578 * Drain pages out of the cpu's pagevecs.
579 * Either "cpu" is the current CPU, and preemption has already been
580 * disabled; or "cpu" is being hot-unplugged, and is already dead.
581 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700582void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Mel Gorman13f7f782013-07-03 15:02:28 -0700584 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Mel Gorman13f7f782013-07-03 15:02:28 -0700586 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700587 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700588
589 pvec = &per_cpu(lru_rotate_pvecs, cpu);
590 if (pagevec_count(pvec)) {
591 unsigned long flags;
592
593 /* No harm done if a racing interrupt already did this */
594 local_irq_save(flags);
595 pagevec_move_tail(pvec);
596 local_irq_restore(flags);
597 }
Minchan Kim31560182011-03-22 16:32:52 -0700598
Minchan Kimcc5993b2015-04-15 16:13:26 -0700599 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700600 if (pagevec_count(pvec))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700601 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700602
Minchan Kim10853a02016-01-15 16:55:11 -0800603 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
604 if (pagevec_count(pvec))
605 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
606
Shaohua Lieb709b02011-05-24 17:12:55 -0700607 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700608}
609
610/**
Minchan Kimcc5993b2015-04-15 16:13:26 -0700611 * deactivate_file_page - forcefully deactivate a file page
Minchan Kim31560182011-03-22 16:32:52 -0700612 * @page: page to deactivate
613 *
614 * This function hints the VM that @page is a good reclaim candidate,
615 * for example if its invalidation fails due to the page being dirty
616 * or under writeback.
617 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700618void deactivate_file_page(struct page *page)
Minchan Kim31560182011-03-22 16:32:52 -0700619{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700620 /*
Minchan Kimcc5993b2015-04-15 16:13:26 -0700621 * In a workload with many unevictable page such as mprotect,
622 * unevictable page deactivation for accelerating reclaim is pointless.
Minchan Kim821ed6b2011-05-24 17:12:31 -0700623 */
624 if (PageUnevictable(page))
625 return;
626
Minchan Kim31560182011-03-22 16:32:52 -0700627 if (likely(get_page_unless_zero(page))) {
Minchan Kimcc5993b2015-04-15 16:13:26 -0700628 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700629
630 if (!pagevec_add(pvec, page))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700631 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
632 put_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700633 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800634}
635
Minchan Kim10853a02016-01-15 16:55:11 -0800636/**
637 * deactivate_page - deactivate a page
638 * @page: page to deactivate
639 *
640 * deactivate_page() moves @page to the inactive list if @page was on the active
641 * list and was not an unevictable page. This is done to accelerate the reclaim
642 * of @page.
643 */
644void deactivate_page(struct page *page)
645{
646 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
647 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
648
649 page_cache_get(page);
650 if (!pagevec_add(pvec, page))
651 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
652 put_cpu_var(lru_deactivate_pvecs);
653 }
654}
655
Andrew Morton80bfed92006-01-06 00:11:14 -0800656void lru_add_drain(void)
657{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700658 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800659 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
David Howellsc4028952006-11-22 14:57:56 +0000662static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800663{
664 lru_add_drain();
665}
666
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700667static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
668
669void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800670{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700671 static DEFINE_MUTEX(lock);
672 static struct cpumask has_work;
673 int cpu;
674
675 mutex_lock(&lock);
676 get_online_cpus();
677 cpumask_clear(&has_work);
678
679 for_each_online_cpu(cpu) {
680 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
681
682 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
683 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
Minchan Kimcc5993b2015-04-15 16:13:26 -0700684 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
Minchan Kim10853a02016-01-15 16:55:11 -0800685 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700686 need_activate_page_drain(cpu)) {
687 INIT_WORK(work, lru_add_drain_per_cpu);
688 schedule_work_on(cpu, work);
689 cpumask_set_cpu(cpu, &has_work);
690 }
691 }
692
693 for_each_cpu(cpu, &has_work)
694 flush_work(&per_cpu(lru_add_drain_work, cpu));
695
696 put_online_cpus();
697 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800698}
699
Michal Hockoaabfb572014-10-09 15:28:52 -0700700/**
701 * release_pages - batched page_cache_release()
702 * @pages: array of pages to release
703 * @nr: number of pages
704 * @cold: whether the pages are cache cold
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 *
Michal Hockoaabfb572014-10-09 15:28:52 -0700706 * Decrement the reference count on all the pages in @pages. If it
707 * fell to zero, remove the page from the LRU and free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 */
Mel Gormanb745bc82014-06-04 16:10:22 -0700709void release_pages(struct page **pages, int nr, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800712 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700714 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700715 unsigned long uninitialized_var(flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700716 unsigned int uninitialized_var(lock_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 for (i = 0; i < nr; i++) {
719 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
Michal Hockoaabfb572014-10-09 15:28:52 -0700721 /*
722 * Make sure the IRQ-safe lock-holding time does not get
723 * excessive with a continuous string of pages from the
724 * same zone. The lock is held only if zone != NULL.
725 */
726 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
727 spin_unlock_irqrestore(&zone->lru_lock, flags);
728 zone = NULL;
729 }
730
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800731 page = compound_head(page);
Nick Pigginb5810032005-10-29 18:16:12 -0700732 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 continue;
734
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800735 if (PageCompound(page)) {
736 if (zone) {
737 spin_unlock_irqrestore(&zone->lru_lock, flags);
738 zone = NULL;
739 }
740 __put_compound_page(page);
741 continue;
742 }
743
Nick Piggin46453a62006-03-22 00:07:58 -0800744 if (PageLRU(page)) {
745 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700746
Nick Piggin46453a62006-03-22 00:07:58 -0800747 if (pagezone != zone) {
748 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700749 spin_unlock_irqrestore(&zone->lru_lock,
750 flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700751 lock_batch = 0;
Nick Piggin46453a62006-03-22 00:07:58 -0800752 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700753 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700755
756 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -0800757 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800758 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700759 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
Nick Piggin46453a62006-03-22 00:07:58 -0800761
Mel Gormanc53954a2013-07-03 15:02:34 -0700762 /* Clear Active bit in case of parallel mark_page_accessed */
Mel Gormane3741b52014-06-04 16:10:26 -0700763 __ClearPageActive(page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700764
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800765 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 }
767 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700768 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Johannes Weiner747db952014-08-08 14:19:24 -0700770 mem_cgroup_uncharge_list(&pages_to_free);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800771 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700773EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
775/*
776 * The pages which we're about to release may be in the deferred lru-addition
777 * queues. That would prevent them from really being freed right now. That's
778 * OK from a correctness point of view but is inefficient - those pages may be
779 * cache-warm and we want to give them back to the page allocator ASAP.
780 *
781 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
782 * and __pagevec_lru_add_active() call release_pages() directly to avoid
783 * mutual recursion.
784 */
785void __pagevec_release(struct pagevec *pvec)
786{
787 lru_add_drain();
788 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
789 pagevec_reinit(pvec);
790}
Steve French7f285702005-11-01 10:22:55 -0800791EXPORT_SYMBOL(__pagevec_release);
792
Hugh Dickins12d27102012-01-12 17:19:52 -0800793#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800794/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700795void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700796 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800797{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800798 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800799
Sasha Levin309381fea2014-01-23 15:52:54 -0800800 VM_BUG_ON_PAGE(!PageHead(page), page);
801 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
802 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700803 VM_BUG_ON(NR_CPUS != 1 &&
804 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800805
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700806 if (!list)
807 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800808
Hugh Dickins12d27102012-01-12 17:19:52 -0800809 if (likely(PageLRU(page)))
810 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700811 else if (list) {
812 /* page reclaim is reclaiming a huge page */
813 get_page(page_tail);
814 list_add_tail(&page_tail->lru, list);
815 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800816 struct list_head *list_head;
817 /*
818 * Head page has not yet been counted, as an hpage,
819 * so we must account for each subpage individually.
820 *
821 * Use the standard add function to put page_tail on the list,
822 * but then correct its position so they all end up in order.
823 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700824 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -0800825 list_head = page_tail->lru.prev;
826 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800827 }
Hugh Dickins75121022012-03-05 14:59:18 -0800828
829 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700830 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800831}
Hugh Dickins12d27102012-01-12 17:19:52 -0800832#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800833
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700834static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
835 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700836{
Mel Gorman13f7f782013-07-03 15:02:28 -0700837 int file = page_is_file_cache(page);
838 int active = PageActive(page);
839 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700840
Sasha Levin309381fea2014-01-23 15:52:54 -0800841 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700842
843 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700844 add_page_to_lru_list(page, lruvec, lru);
845 update_page_reclaim_stat(lruvec, file, active);
Mel Gorman24b7e582014-08-06 16:07:11 -0700846 trace_mm_lru_insertion(page, lru);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700847}
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 * Add the passed pages to the LRU, then drop the caller's refcount
851 * on them. Reinitialises the caller's pagevec.
852 */
Mel Gormana0b8cab32013-07-03 15:02:32 -0700853void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
Mel Gormana0b8cab32013-07-03 15:02:32 -0700855 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800857EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700860 * pagevec_lookup_entries - gang pagecache lookup
861 * @pvec: Where the resulting entries are placed
862 * @mapping: The address_space to search
863 * @start: The starting entry index
864 * @nr_entries: The maximum number of entries
865 * @indices: The cache indices corresponding to the entries in @pvec
866 *
867 * pagevec_lookup_entries() will search for and return a group of up
868 * to @nr_entries pages and shadow entries in the mapping. All
869 * entries are placed in @pvec. pagevec_lookup_entries() takes a
870 * reference against actual pages in @pvec.
871 *
872 * The search returns a group of mapping-contiguous entries with
873 * ascending indexes. There may be holes in the indices due to
874 * not-present entries.
875 *
876 * pagevec_lookup_entries() returns the number of entries which were
877 * found.
878 */
879unsigned pagevec_lookup_entries(struct pagevec *pvec,
880 struct address_space *mapping,
881 pgoff_t start, unsigned nr_pages,
882 pgoff_t *indices)
883{
884 pvec->nr = find_get_entries(mapping, start, nr_pages,
885 pvec->pages, indices);
886 return pagevec_count(pvec);
887}
888
889/**
890 * pagevec_remove_exceptionals - pagevec exceptionals pruning
891 * @pvec: The pagevec to prune
892 *
893 * pagevec_lookup_entries() fills both pages and exceptional radix
894 * tree entries into the pagevec. This function prunes all
895 * exceptionals from @pvec without leaving holes, so that it can be
896 * passed on to page-only pagevec operations.
897 */
898void pagevec_remove_exceptionals(struct pagevec *pvec)
899{
900 int i, j;
901
902 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
903 struct page *page = pvec->pages[i];
904 if (!radix_tree_exceptional_entry(page))
905 pvec->pages[j++] = page;
906 }
907 pvec->nr = j;
908}
909
910/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 * pagevec_lookup - gang pagecache lookup
912 * @pvec: Where the resulting pages are placed
913 * @mapping: The address_space to search
914 * @start: The starting page index
915 * @nr_pages: The maximum number of pages
916 *
917 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
918 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
919 * reference against the pages in @pvec.
920 *
921 * The search returns a group of mapping-contiguous pages with ascending
922 * indexes. There may be holes in the indices due to not-present pages.
923 *
924 * pagevec_lookup() returns the number of pages which were found.
925 */
926unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
927 pgoff_t start, unsigned nr_pages)
928{
929 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
930 return pagevec_count(pvec);
931}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100932EXPORT_SYMBOL(pagevec_lookup);
933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
935 pgoff_t *index, int tag, unsigned nr_pages)
936{
937 pvec->nr = find_get_pages_tag(mapping, index, tag,
938 nr_pages, pvec->pages);
939 return pagevec_count(pvec);
940}
Steve French7f285702005-11-01 10:22:55 -0800941EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943/*
944 * Perform any setup for the swap system
945 */
946void __init swap_setup(void)
947{
Jan Beulich44813742009-09-21 17:03:05 -0700948 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700949#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -0800950 int i;
951
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800952 for (i = 0; i < MAX_SWAPFILES; i++)
Shaohua Li33806f02013-02-22 16:34:37 -0800953 spin_lock_init(&swapper_spaces[i].tree_lock);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700954#endif
955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 /* Use a smaller cluster for small-memory machines */
957 if (megs < 16)
958 page_cluster = 2;
959 else
960 page_cluster = 3;
961 /*
962 * Right now other parts of the system means that we
963 * _really_ don't want to cluster much more
964 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}