blob: 95916142fc46d3a80a6ddfa54218e1de641f6045 [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);
Ming Lia4a921a2016-05-20 16:57:56 -070050#ifdef CONFIG_SMP
51static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
52#endif
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070053
Adrian Bunkb2213852006-09-25 23:31:02 -070054/*
55 * This path almost never happens for VM activity - pages are normally
56 * freed via pagevecs. But it gets used by networking.
57 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080058static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070059{
60 if (PageLRU(page)) {
Adrian Bunkb2213852006-09-25 23:31:02 -070061 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070062 struct lruvec *lruvec;
63 unsigned long flags;
Adrian Bunkb2213852006-09-25 23:31:02 -070064
65 spin_lock_irqsave(&zone->lru_lock, flags);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070066 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -080067 VM_BUG_ON_PAGE(!PageLRU(page), page);
Adrian Bunkb2213852006-09-25 23:31:02 -070068 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070069 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070070 spin_unlock_irqrestore(&zone->lru_lock, flags);
71 }
Johannes Weiner0a31bc92014-08-08 14:19:22 -070072 mem_cgroup_uncharge(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080073}
74
75static void __put_single_page(struct page *page)
76{
77 __page_cache_release(page);
Mel Gormanb745bc82014-06-04 16:10:22 -070078 free_hot_cold_page(page, false);
Adrian Bunkb2213852006-09-25 23:31:02 -070079}
80
Andrea Arcangeli91807062011-01-13 15:46:32 -080081static void __put_compound_page(struct page *page)
82{
83 compound_page_dtor *dtor;
84
Naoya Horiguchi822fc612015-04-15 16:14:35 -070085 /*
86 * __page_cache_release() is supposed to be called for thp, not for
87 * hugetlb. This is because hugetlb page does never have PageLRU set
88 * (it's never listed to any LRU lists) and no memcg routines should
89 * be called for hugetlb (it has a separate hugetlb_cgroup.)
90 */
91 if (!PageHuge(page))
92 __page_cache_release(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080093 dtor = get_compound_page_dtor(page);
94 (*dtor)(page);
95}
96
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -080097void __put_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
Nick Piggin8519fb32006-02-07 12:58:52 -080099 if (unlikely(PageCompound(page)))
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800100 __put_compound_page(page);
101 else
Andrea Arcangeli91807062011-01-13 15:46:32 -0800102 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103}
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800104EXPORT_SYMBOL(__put_page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700105
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700106/**
Randy Dunlap76824862008-03-19 17:00:40 -0700107 * put_pages_list() - release a list of pages
108 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700109 *
110 * Release a list of pages which are strung together on page.lru. Currently
111 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700112 */
113void put_pages_list(struct list_head *pages)
114{
115 while (!list_empty(pages)) {
116 struct page *victim;
117
118 victim = list_entry(pages->prev, struct page, lru);
119 list_del(&victim->lru);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300120 put_page(victim);
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700121 }
122}
123EXPORT_SYMBOL(put_pages_list);
124
Mel Gorman18022c52012-07-31 16:44:51 -0700125/*
126 * get_kernel_pages() - pin kernel pages in memory
127 * @kiov: An array of struct kvec structures
128 * @nr_segs: number of segments to pin
129 * @write: pinning for read/write, currently ignored
130 * @pages: array that receives pointers to the pages pinned.
131 * Should be at least nr_segs long.
132 *
133 * Returns number of pages pinned. This may be fewer than the number
134 * requested. If nr_pages is 0 or negative, returns 0. If no pages
135 * were pinned, returns -errno. Each page returned must be released
136 * with a put_page() call when it is finished with.
137 */
138int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
139 struct page **pages)
140{
141 int seg;
142
143 for (seg = 0; seg < nr_segs; seg++) {
144 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
145 return seg;
146
Mel Gorman5a178112012-07-31 16:45:02 -0700147 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300148 get_page(pages[seg]);
Mel Gorman18022c52012-07-31 16:44:51 -0700149 }
150
151 return seg;
152}
153EXPORT_SYMBOL_GPL(get_kernel_pages);
154
155/*
156 * get_kernel_page() - pin a kernel page in memory
157 * @start: starting kernel address
158 * @write: pinning for read/write, currently ignored
159 * @pages: array that receives pointer to the page pinned.
160 * Must be at least nr_segs long.
161 *
162 * Returns 1 if page is pinned. If the page was not pinned, returns
163 * -errno. The page returned must be released with a put_page() call
164 * when it is finished with.
165 */
166int get_kernel_page(unsigned long start, int write, struct page **pages)
167{
168 const struct kvec kiov = {
169 .iov_base = (void *)start,
170 .iov_len = PAGE_SIZE
171 };
172
173 return get_kernel_pages(&kiov, 1, write, pages);
174}
175EXPORT_SYMBOL_GPL(get_kernel_page);
176
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700177static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700178 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
179 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700180{
181 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700182 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700183 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700184 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700185
186 for (i = 0; i < pagevec_count(pvec); i++) {
187 struct page *page = pvec->pages[i];
188 struct zone *pagezone = page_zone(page);
189
190 if (pagezone != zone) {
191 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700192 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700193 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700194 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700195 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700196
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700197 lruvec = mem_cgroup_page_lruvec(page, zone);
198 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700199 }
200 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700201 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800202 release_pages(pvec->pages, pvec->nr, pvec->cold);
203 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800204}
205
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700206static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
207 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700208{
209 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700210
211 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
212 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800213 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700214 (*pgmoved)++;
215 }
216}
217
218/*
219 * pagevec_move_tail() must be called with IRQ disabled.
220 * Otherwise this may cause nasty races.
221 */
222static void pagevec_move_tail(struct pagevec *pvec)
223{
224 int pgmoved = 0;
225
226 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
227 __count_vm_events(PGROTATED, pgmoved);
228}
229
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700230/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 * Writeback is about to end against a page which has been marked for immediate
232 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700233 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700235void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700237 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700238 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700239 struct pagevec *pvec;
240 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300242 get_page(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700243 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700244 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700245 if (!pagevec_add(pvec, page))
246 pagevec_move_tail(pvec);
247 local_irq_restore(flags);
248 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700251static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800252 int file, int rotated)
253{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700254 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800255
256 reclaim_stat->recent_scanned[file]++;
257 if (rotated)
258 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800259}
260
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700261static void __activate_page(struct page *page, struct lruvec *lruvec,
262 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Linus Torvalds7a608572011-01-17 14:42:19 -0800264 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
265 int file = page_is_file_cache(page);
266 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800267
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700268 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800269 SetPageActive(page);
270 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700271 add_page_to_lru_list(page, lruvec, lru);
Mel Gorman24b7e582014-08-06 16:07:11 -0700272 trace_mm_lru_activate(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800273
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700274 __count_vm_event(PGACTIVATE);
275 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800276 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700277}
278
279#ifdef CONFIG_SMP
Shaohua Lieb709b02011-05-24 17:12:55 -0700280static void activate_page_drain(int cpu)
281{
282 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
283
284 if (pagevec_count(pvec))
285 pagevec_lru_move_fn(pvec, __activate_page, NULL);
286}
287
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700288static bool need_activate_page_drain(int cpu)
289{
290 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
291}
292
Shaohua Lieb709b02011-05-24 17:12:55 -0700293void activate_page(struct page *page)
294{
295 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
296 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
297
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300298 get_page(page);
Shaohua Lieb709b02011-05-24 17:12:55 -0700299 if (!pagevec_add(pvec, page))
300 pagevec_lru_move_fn(pvec, __activate_page, NULL);
301 put_cpu_var(activate_page_pvecs);
302 }
303}
304
305#else
306static inline void activate_page_drain(int cpu)
307{
308}
309
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700310static bool need_activate_page_drain(int cpu)
311{
312 return false;
313}
314
Shaohua Lieb709b02011-05-24 17:12:55 -0700315void activate_page(struct page *page)
316{
317 struct zone *zone = page_zone(page);
318
319 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700320 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 spin_unlock_irq(&zone->lru_lock);
322}
Shaohua Lieb709b02011-05-24 17:12:55 -0700323#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Mel Gorman059285a2013-07-03 15:02:30 -0700325static void __lru_cache_activate_page(struct page *page)
326{
327 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
328 int i;
329
330 /*
331 * Search backwards on the optimistic assumption that the page being
332 * activated has just been added to this pagevec. Note that only
333 * the local pagevec is examined as a !PageLRU page could be in the
334 * process of being released, reclaimed, migrated or on a remote
335 * pagevec that is currently being drained. Furthermore, marking
336 * a remote pagevec's page PageActive potentially hits a race where
337 * a page is marked PageActive just after it is added to the inactive
338 * list causing accounting errors and BUG_ON checks to trigger.
339 */
340 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
341 struct page *pagevec_page = pvec->pages[i];
342
343 if (pagevec_page == page) {
344 SetPageActive(page);
345 break;
346 }
347 }
348
349 put_cpu_var(lru_add_pvec);
350}
351
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352/*
353 * Mark a page as having seen activity.
354 *
355 * inactive,unreferenced -> inactive,referenced
356 * inactive,referenced -> active,unreferenced
357 * active,unreferenced -> active,referenced
Hugh Dickinseb39d612014-08-06 16:06:43 -0700358 *
359 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
360 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800362void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800364 page = compound_head(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700365 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700366 PageReferenced(page)) {
367
368 /*
369 * If the page is on the LRU, queue it for activation via
370 * activate_page_pvecs. Otherwise, assume the page is on a
371 * pagevec, mark it active and it'll be moved to the active
372 * LRU on the next drain.
373 */
374 if (PageLRU(page))
375 activate_page(page);
376 else
377 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700379 if (page_is_file_cache(page))
380 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 } else if (!PageReferenced(page)) {
382 SetPageReferenced(page);
383 }
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700384 if (page_is_idle(page))
385 clear_page_idle(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387EXPORT_SYMBOL(mark_page_accessed);
388
Jianyu Zhan2329d372014-06-04 16:07:31 -0700389static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Mel Gorman13f7f782013-07-03 15:02:28 -0700391 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
392
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300393 get_page(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700394 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700395 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700396 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700397 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700399
400/**
401 * lru_cache_add: add a page to the page lists
402 * @page: the page to add
403 */
404void lru_cache_add_anon(struct page *page)
405{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700406 if (PageActive(page))
407 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700408 __lru_cache_add(page);
409}
410
411void lru_cache_add_file(struct page *page)
412{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700413 if (PageActive(page))
414 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700415 __lru_cache_add(page);
416}
417EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700419/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700420 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700421 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700422 *
423 * Queue the page for addition to the LRU via pagevec. The decision on whether
424 * to add the page to the [in]active [file|anon] list is deferred until the
425 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
426 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700427 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700428void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Sasha Levin309381fea2014-01-23 15:52:54 -0800430 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
431 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700432 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700435/**
436 * add_page_to_unevictable_list - add a page to the unevictable list
437 * @page: the page to be added to the unevictable list
438 *
439 * Add page directly to its zone's unevictable list. To avoid races with
440 * tasks that might be making the page evictable, through eg. munlock,
441 * munmap or exit, while it's not on the lru, we want to add the page
442 * while it's locked or otherwise "invisible" to other tasks. This is
443 * difficult to do when using the pagevec cache, so bypass that.
444 */
445void add_page_to_unevictable_list(struct page *page)
446{
447 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700448 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700449
450 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700451 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700452 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700453 SetPageUnevictable(page);
454 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700455 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700456 spin_unlock_irq(&zone->lru_lock);
457}
458
Johannes Weiner00501b52014-08-08 14:19:20 -0700459/**
460 * lru_cache_add_active_or_unevictable
461 * @page: the page to be added to LRU
462 * @vma: vma in which page is mapped for determining reclaimability
463 *
464 * Place @page on the active or unevictable LRU list, depending on its
465 * evictability. Note that if the page is not evictable, it goes
466 * directly back onto it's zone's unevictable list, it does NOT use a
467 * per cpu pagevec.
468 */
469void lru_cache_add_active_or_unevictable(struct page *page,
470 struct vm_area_struct *vma)
471{
472 VM_BUG_ON_PAGE(PageLRU(page), page);
473
474 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
475 SetPageActive(page);
476 lru_cache_add(page);
477 return;
478 }
479
480 if (!TestSetPageMlocked(page)) {
481 /*
482 * We use the irq-unsafe __mod_zone_page_stat because this
483 * counter is not modified from interrupt context, and the pte
484 * lock is held(spinlock), which implies preemption disabled.
485 */
486 __mod_zone_page_state(page_zone(page), NR_MLOCK,
487 hpage_nr_pages(page));
488 count_vm_event(UNEVICTABLE_PGMLOCKED);
489 }
490 add_page_to_unevictable_list(page);
491}
492
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700493/*
Minchan Kim31560182011-03-22 16:32:52 -0700494 * If the page can not be invalidated, it is moved to the
495 * inactive list to speed up its reclaim. It is moved to the
496 * head of the list, rather than the tail, to give the flusher
497 * threads some time to write it out, as this is much more
498 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700499 *
500 * If the page isn't page_mapped and dirty/writeback, the page
501 * could reclaim asap using PG_reclaim.
502 *
503 * 1. active, mapped page -> none
504 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
505 * 3. inactive, mapped page -> none
506 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
507 * 5. inactive, clean -> inactive, tail
508 * 6. Others -> none
509 *
510 * In 4, why it moves inactive's head, the VM expects the page would
511 * be write it out by flusher threads as this is much more effective
512 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700513 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700514static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700515 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700516{
517 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700518 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700519
Minchan Kim278df9f2011-03-22 16:32:54 -0700520 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700521 return;
522
Minchan Kimbad49d92011-05-11 15:13:30 -0700523 if (PageUnevictable(page))
524 return;
525
Minchan Kim31560182011-03-22 16:32:52 -0700526 /* Some processes are using the page */
527 if (page_mapped(page))
528 return;
529
Minchan Kim278df9f2011-03-22 16:32:54 -0700530 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700531 file = page_is_file_cache(page);
532 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700533
534 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700535 ClearPageActive(page);
536 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700537 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700538
Minchan Kim278df9f2011-03-22 16:32:54 -0700539 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 */
545 SetPageReclaim(page);
546 } else {
547 /*
548 * The page's writeback ends up during pagevec
549 * We moves tha page into tail of inactive.
550 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800551 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700552 __count_vm_event(PGROTATED);
553 }
554
555 if (active)
556 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700557 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700558}
559
Minchan Kim10853a02016-01-15 16:55:11 -0800560
561static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
562 void *arg)
563{
564 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
565 int file = page_is_file_cache(page);
566 int lru = page_lru_base_type(page);
567
568 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
569 ClearPageActive(page);
570 ClearPageReferenced(page);
571 add_page_to_lru_list(page, lruvec, lru);
572
573 __count_vm_event(PGDEACTIVATE);
574 update_page_reclaim_stat(lruvec, file, 0);
575 }
576}
577
Minchan Kim31560182011-03-22 16:32:52 -0700578/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700579 * Drain pages out of the cpu's pagevecs.
580 * Either "cpu" is the current CPU, and preemption has already been
581 * disabled; or "cpu" is being hot-unplugged, and is already dead.
582 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700583void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
Mel Gorman13f7f782013-07-03 15:02:28 -0700585 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Mel Gorman13f7f782013-07-03 15:02:28 -0700587 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700588 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700589
590 pvec = &per_cpu(lru_rotate_pvecs, cpu);
591 if (pagevec_count(pvec)) {
592 unsigned long flags;
593
594 /* No harm done if a racing interrupt already did this */
595 local_irq_save(flags);
596 pagevec_move_tail(pvec);
597 local_irq_restore(flags);
598 }
Minchan Kim31560182011-03-22 16:32:52 -0700599
Minchan Kimcc5993b2015-04-15 16:13:26 -0700600 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700601 if (pagevec_count(pvec))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700602 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700603
Minchan Kim10853a02016-01-15 16:55:11 -0800604 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
605 if (pagevec_count(pvec))
606 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
607
Shaohua Lieb709b02011-05-24 17:12:55 -0700608 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700609}
610
611/**
Minchan Kimcc5993b2015-04-15 16:13:26 -0700612 * deactivate_file_page - forcefully deactivate a file page
Minchan Kim31560182011-03-22 16:32:52 -0700613 * @page: page to deactivate
614 *
615 * This function hints the VM that @page is a good reclaim candidate,
616 * for example if its invalidation fails due to the page being dirty
617 * or under writeback.
618 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700619void deactivate_file_page(struct page *page)
Minchan Kim31560182011-03-22 16:32:52 -0700620{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700621 /*
Minchan Kimcc5993b2015-04-15 16:13:26 -0700622 * In a workload with many unevictable page such as mprotect,
623 * unevictable page deactivation for accelerating reclaim is pointless.
Minchan Kim821ed6b2011-05-24 17:12:31 -0700624 */
625 if (PageUnevictable(page))
626 return;
627
Minchan Kim31560182011-03-22 16:32:52 -0700628 if (likely(get_page_unless_zero(page))) {
Minchan Kimcc5993b2015-04-15 16:13:26 -0700629 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700630
631 if (!pagevec_add(pvec, page))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700632 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
633 put_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700634 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800635}
636
Minchan Kim10853a02016-01-15 16:55:11 -0800637/**
638 * deactivate_page - deactivate a page
639 * @page: page to deactivate
640 *
641 * deactivate_page() moves @page to the inactive list if @page was on the active
642 * list and was not an unevictable page. This is done to accelerate the reclaim
643 * of @page.
644 */
645void deactivate_page(struct page *page)
646{
647 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
648 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
649
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300650 get_page(page);
Minchan Kim10853a02016-01-15 16:55:11 -0800651 if (!pagevec_add(pvec, page))
652 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
653 put_cpu_var(lru_deactivate_pvecs);
654 }
655}
656
Andrew Morton80bfed92006-01-06 00:11:14 -0800657void lru_add_drain(void)
658{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700659 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800660 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661}
662
David Howellsc4028952006-11-22 14:57:56 +0000663static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800664{
665 lru_add_drain();
666}
667
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700668static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
669
670void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800671{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700672 static DEFINE_MUTEX(lock);
673 static struct cpumask has_work;
674 int cpu;
675
676 mutex_lock(&lock);
677 get_online_cpus();
678 cpumask_clear(&has_work);
679
680 for_each_online_cpu(cpu) {
681 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
682
683 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
684 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
Minchan Kimcc5993b2015-04-15 16:13:26 -0700685 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
Minchan Kim10853a02016-01-15 16:55:11 -0800686 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700687 need_activate_page_drain(cpu)) {
688 INIT_WORK(work, lru_add_drain_per_cpu);
689 schedule_work_on(cpu, work);
690 cpumask_set_cpu(cpu, &has_work);
691 }
692 }
693
694 for_each_cpu(cpu, &has_work)
695 flush_work(&per_cpu(lru_add_drain_work, cpu));
696
697 put_online_cpus();
698 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800699}
700
Michal Hockoaabfb572014-10-09 15:28:52 -0700701/**
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300702 * release_pages - batched put_page()
Michal Hockoaabfb572014-10-09 15:28:52 -0700703 * @pages: array of pages to release
704 * @nr: number of pages
705 * @cold: whether the pages are cache cold
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 *
Michal Hockoaabfb572014-10-09 15:28:52 -0700707 * Decrement the reference count on all the pages in @pages. If it
708 * fell to zero, remove the page from the LRU and free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 */
Mel Gormanb745bc82014-06-04 16:10:22 -0700710void release_pages(struct page **pages, int nr, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
712 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800713 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700715 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700716 unsigned long uninitialized_var(flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700717 unsigned int uninitialized_var(lock_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 for (i = 0; i < nr; i++) {
720 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Michal Hockoaabfb572014-10-09 15:28:52 -0700722 /*
723 * Make sure the IRQ-safe lock-holding time does not get
724 * excessive with a continuous string of pages from the
725 * same zone. The lock is held only if zone != NULL.
726 */
727 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
728 spin_unlock_irqrestore(&zone->lru_lock, flags);
729 zone = NULL;
730 }
731
Kirill A. Shutemovaa88b682016-04-28 16:18:27 -0700732 if (is_huge_zero_page(page)) {
733 put_huge_zero_page();
734 continue;
735 }
736
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800737 page = compound_head(page);
Nick Pigginb5810032005-10-29 18:16:12 -0700738 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 continue;
740
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800741 if (PageCompound(page)) {
742 if (zone) {
743 spin_unlock_irqrestore(&zone->lru_lock, flags);
744 zone = NULL;
745 }
746 __put_compound_page(page);
747 continue;
748 }
749
Nick Piggin46453a62006-03-22 00:07:58 -0800750 if (PageLRU(page)) {
751 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700752
Nick Piggin46453a62006-03-22 00:07:58 -0800753 if (pagezone != zone) {
754 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700755 spin_unlock_irqrestore(&zone->lru_lock,
756 flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700757 lock_batch = 0;
Nick Piggin46453a62006-03-22 00:07:58 -0800758 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700759 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700761
762 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -0800763 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800764 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700765 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 }
Nick Piggin46453a62006-03-22 00:07:58 -0800767
Mel Gormanc53954a2013-07-03 15:02:34 -0700768 /* Clear Active bit in case of parallel mark_page_accessed */
Mel Gormane3741b52014-06-04 16:10:26 -0700769 __ClearPageActive(page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700770
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800771 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 }
773 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700774 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Johannes Weiner747db952014-08-08 14:19:24 -0700776 mem_cgroup_uncharge_list(&pages_to_free);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800777 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700779EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781/*
782 * The pages which we're about to release may be in the deferred lru-addition
783 * queues. That would prevent them from really being freed right now. That's
784 * OK from a correctness point of view but is inefficient - those pages may be
785 * cache-warm and we want to give them back to the page allocator ASAP.
786 *
787 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
788 * and __pagevec_lru_add_active() call release_pages() directly to avoid
789 * mutual recursion.
790 */
791void __pagevec_release(struct pagevec *pvec)
792{
793 lru_add_drain();
794 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
795 pagevec_reinit(pvec);
796}
Steve French7f285702005-11-01 10:22:55 -0800797EXPORT_SYMBOL(__pagevec_release);
798
Hugh Dickins12d27102012-01-12 17:19:52 -0800799#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800800/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700801void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700802 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800803{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800804 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800805
Sasha Levin309381fea2014-01-23 15:52:54 -0800806 VM_BUG_ON_PAGE(!PageHead(page), page);
807 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
808 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700809 VM_BUG_ON(NR_CPUS != 1 &&
810 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800811
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700812 if (!list)
813 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800814
Hugh Dickins12d27102012-01-12 17:19:52 -0800815 if (likely(PageLRU(page)))
816 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700817 else if (list) {
818 /* page reclaim is reclaiming a huge page */
819 get_page(page_tail);
820 list_add_tail(&page_tail->lru, list);
821 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800822 struct list_head *list_head;
823 /*
824 * Head page has not yet been counted, as an hpage,
825 * so we must account for each subpage individually.
826 *
827 * Use the standard add function to put page_tail on the list,
828 * but then correct its position so they all end up in order.
829 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700830 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -0800831 list_head = page_tail->lru.prev;
832 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800833 }
Hugh Dickins75121022012-03-05 14:59:18 -0800834
835 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700836 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800837}
Hugh Dickins12d27102012-01-12 17:19:52 -0800838#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800839
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700840static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
841 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700842{
Mel Gorman13f7f782013-07-03 15:02:28 -0700843 int file = page_is_file_cache(page);
844 int active = PageActive(page);
845 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700846
Sasha Levin309381fea2014-01-23 15:52:54 -0800847 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700848
849 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700850 add_page_to_lru_list(page, lruvec, lru);
851 update_page_reclaim_stat(lruvec, file, active);
Mel Gorman24b7e582014-08-06 16:07:11 -0700852 trace_mm_lru_insertion(page, lru);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700853}
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 * Add the passed pages to the LRU, then drop the caller's refcount
857 * on them. Reinitialises the caller's pagevec.
858 */
Mel Gormana0b8cab32013-07-03 15:02:32 -0700859void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
Mel Gormana0b8cab32013-07-03 15:02:32 -0700861 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800863EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700866 * pagevec_lookup_entries - gang pagecache lookup
867 * @pvec: Where the resulting entries are placed
868 * @mapping: The address_space to search
869 * @start: The starting entry index
870 * @nr_entries: The maximum number of entries
871 * @indices: The cache indices corresponding to the entries in @pvec
872 *
873 * pagevec_lookup_entries() will search for and return a group of up
874 * to @nr_entries pages and shadow entries in the mapping. All
875 * entries are placed in @pvec. pagevec_lookup_entries() takes a
876 * reference against actual pages in @pvec.
877 *
878 * The search returns a group of mapping-contiguous entries with
879 * ascending indexes. There may be holes in the indices due to
880 * not-present entries.
881 *
882 * pagevec_lookup_entries() returns the number of entries which were
883 * found.
884 */
885unsigned pagevec_lookup_entries(struct pagevec *pvec,
886 struct address_space *mapping,
887 pgoff_t start, unsigned nr_pages,
888 pgoff_t *indices)
889{
890 pvec->nr = find_get_entries(mapping, start, nr_pages,
891 pvec->pages, indices);
892 return pagevec_count(pvec);
893}
894
895/**
896 * pagevec_remove_exceptionals - pagevec exceptionals pruning
897 * @pvec: The pagevec to prune
898 *
899 * pagevec_lookup_entries() fills both pages and exceptional radix
900 * tree entries into the pagevec. This function prunes all
901 * exceptionals from @pvec without leaving holes, so that it can be
902 * passed on to page-only pagevec operations.
903 */
904void pagevec_remove_exceptionals(struct pagevec *pvec)
905{
906 int i, j;
907
908 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
909 struct page *page = pvec->pages[i];
910 if (!radix_tree_exceptional_entry(page))
911 pvec->pages[j++] = page;
912 }
913 pvec->nr = j;
914}
915
916/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 * pagevec_lookup - gang pagecache lookup
918 * @pvec: Where the resulting pages are placed
919 * @mapping: The address_space to search
920 * @start: The starting page index
921 * @nr_pages: The maximum number of pages
922 *
923 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
924 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
925 * reference against the pages in @pvec.
926 *
927 * The search returns a group of mapping-contiguous pages with ascending
928 * indexes. There may be holes in the indices due to not-present pages.
929 *
930 * pagevec_lookup() returns the number of pages which were found.
931 */
932unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
933 pgoff_t start, unsigned nr_pages)
934{
935 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
936 return pagevec_count(pvec);
937}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100938EXPORT_SYMBOL(pagevec_lookup);
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
941 pgoff_t *index, int tag, unsigned nr_pages)
942{
943 pvec->nr = find_get_pages_tag(mapping, index, tag,
944 nr_pages, pvec->pages);
945 return pagevec_count(pvec);
946}
Steve French7f285702005-11-01 10:22:55 -0800947EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949/*
950 * Perform any setup for the swap system
951 */
952void __init swap_setup(void)
953{
Jan Beulich44813742009-09-21 17:03:05 -0700954 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700955#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -0800956 int i;
957
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800958 for (i = 0; i < MAX_SWAPFILES; i++)
Shaohua Li33806f02013-02-22 16:34:37 -0800959 spin_lock_init(&swapper_spaces[i].tree_lock);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700960#endif
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 /* Use a smaller cluster for small-memory machines */
963 if (megs < 16)
964 page_cluster = 2;
965 else
966 page_cluster = 3;
967 /*
968 * Right now other parts of the system means that we
969 * _really_ don't want to cluster much more
970 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}