blob: 582722507484032067446aeaf766d52802954987 [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
Mel Gormana52633d2016-07-28 15:45:28 -070065 spin_lock_irqsave(zone_lru_lock(zone), flags);
Mel Gorman599d0c92016-07-28 15:45:31 -070066 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
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));
Mel Gormana52633d2016-07-28 15:45:28 -070070 spin_unlock_irqrestore(zone_lru_lock(zone), flags);
Adrian Bunkb2213852006-09-25 23:31:02 -070071 }
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;
Mel Gorman68eb0732016-07-28 15:47:11 -0700182 struct pglist_data *pgdat = 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];
Mel Gorman68eb0732016-07-28 15:47:11 -0700188 struct pglist_data *pagepgdat = page_pgdat(page);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700189
Mel Gorman68eb0732016-07-28 15:47:11 -0700190 if (pagepgdat != pgdat) {
191 if (pgdat)
192 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
193 pgdat = pagepgdat;
194 spin_lock_irqsave(&pgdat->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700195 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700196
Mel Gorman68eb0732016-07-28 15:47:11 -0700197 lruvec = mem_cgroup_page_lruvec(page, pgdat);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700198 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700199 }
Mel Gorman68eb0732016-07-28 15:47:11 -0700200 if (pgdat)
201 spin_unlock_irqrestore(&pgdat->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
Johannes Weiner6b25f412017-02-24 14:56:23 -0800211 if (PageLRU(page) && !PageUnevictable(page)) {
212 del_page_from_lru_list(page, lruvec, page_lru(page));
213 ClearPageActive(page);
214 add_page_to_lru_list_tail(page, lruvec, page_lru(page));
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700215 (*pgmoved)++;
216 }
217}
218
219/*
220 * pagevec_move_tail() must be called with IRQ disabled.
221 * Otherwise this may cause nasty races.
222 */
223static void pagevec_move_tail(struct pagevec *pvec)
224{
225 int pgmoved = 0;
226
227 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
228 __count_vm_events(PGROTATED, pgmoved);
229}
230
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700231/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 * Writeback is about to end against a page which has been marked for immediate
233 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700234 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700236void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Johannes Weiner6b25f412017-02-24 14:56:23 -0800238 if (!PageLocked(page) && !PageDirty(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700239 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700240 struct pagevec *pvec;
241 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300243 get_page(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700244 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700245 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700246 if (!pagevec_add(pvec, page) || PageCompound(page))
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700247 pagevec_move_tail(pvec);
248 local_irq_restore(flags);
249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
251
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700252static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800253 int file, int rotated)
254{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700255 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800256
257 reclaim_stat->recent_scanned[file]++;
258 if (rotated)
259 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800260}
261
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700262static void __activate_page(struct page *page, struct lruvec *lruvec,
263 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264{
Linus Torvalds7a608572011-01-17 14:42:19 -0800265 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
266 int file = page_is_file_cache(page);
267 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800268
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700269 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800270 SetPageActive(page);
271 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700272 add_page_to_lru_list(page, lruvec, lru);
Mel Gorman24b7e582014-08-06 16:07:11 -0700273 trace_mm_lru_activate(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800274
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700275 __count_vm_event(PGACTIVATE);
276 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800277 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700278}
279
280#ifdef CONFIG_SMP
Shaohua Lieb709b02011-05-24 17:12:55 -0700281static void activate_page_drain(int cpu)
282{
283 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
284
285 if (pagevec_count(pvec))
286 pagevec_lru_move_fn(pvec, __activate_page, NULL);
287}
288
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700289static bool need_activate_page_drain(int cpu)
290{
291 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
292}
293
Shaohua Lieb709b02011-05-24 17:12:55 -0700294void activate_page(struct page *page)
295{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700296 page = compound_head(page);
Shaohua Lieb709b02011-05-24 17:12:55 -0700297 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
298 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
299
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300300 get_page(page);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700301 if (!pagevec_add(pvec, page) || PageCompound(page))
Shaohua Lieb709b02011-05-24 17:12:55 -0700302 pagevec_lru_move_fn(pvec, __activate_page, NULL);
303 put_cpu_var(activate_page_pvecs);
304 }
305}
306
307#else
308static inline void activate_page_drain(int cpu)
309{
310}
311
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700312static bool need_activate_page_drain(int cpu)
313{
314 return false;
315}
316
Shaohua Lieb709b02011-05-24 17:12:55 -0700317void activate_page(struct page *page)
318{
319 struct zone *zone = page_zone(page);
320
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700321 page = compound_head(page);
Mel Gormana52633d2016-07-28 15:45:28 -0700322 spin_lock_irq(zone_lru_lock(zone));
Mel Gorman599d0c92016-07-28 15:45:31 -0700323 __activate_page(page, mem_cgroup_page_lruvec(page, zone->zone_pgdat), NULL);
Mel Gormana52633d2016-07-28 15:45:28 -0700324 spin_unlock_irq(zone_lru_lock(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
Shaohua Lieb709b02011-05-24 17:12:55 -0700326#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Mel Gorman059285a2013-07-03 15:02:30 -0700328static void __lru_cache_activate_page(struct page *page)
329{
330 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
331 int i;
332
333 /*
334 * Search backwards on the optimistic assumption that the page being
335 * activated has just been added to this pagevec. Note that only
336 * the local pagevec is examined as a !PageLRU page could be in the
337 * process of being released, reclaimed, migrated or on a remote
338 * pagevec that is currently being drained. Furthermore, marking
339 * a remote pagevec's page PageActive potentially hits a race where
340 * a page is marked PageActive just after it is added to the inactive
341 * list causing accounting errors and BUG_ON checks to trigger.
342 */
343 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
344 struct page *pagevec_page = pvec->pages[i];
345
346 if (pagevec_page == page) {
347 SetPageActive(page);
348 break;
349 }
350 }
351
352 put_cpu_var(lru_add_pvec);
353}
354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355/*
356 * Mark a page as having seen activity.
357 *
358 * inactive,unreferenced -> inactive,referenced
359 * inactive,referenced -> active,unreferenced
360 * active,unreferenced -> active,referenced
Hugh Dickinseb39d612014-08-06 16:06:43 -0700361 *
362 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
363 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800365void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800367 page = compound_head(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700368 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700369 PageReferenced(page)) {
370
371 /*
372 * If the page is on the LRU, queue it for activation via
373 * activate_page_pvecs. Otherwise, assume the page is on a
374 * pagevec, mark it active and it'll be moved to the active
375 * LRU on the next drain.
376 */
377 if (PageLRU(page))
378 activate_page(page);
379 else
380 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700382 if (page_is_file_cache(page))
383 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 } else if (!PageReferenced(page)) {
385 SetPageReferenced(page);
386 }
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700387 if (page_is_idle(page))
388 clear_page_idle(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390EXPORT_SYMBOL(mark_page_accessed);
391
Jianyu Zhan2329d372014-06-04 16:07:31 -0700392static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Mel Gorman13f7f782013-07-03 15:02:28 -0700394 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
395
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300396 get_page(page);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700397 if (!pagevec_add(pvec, page) || PageCompound(page))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700398 __pagevec_lru_add(pvec);
Mel Gorman13f7f782013-07-03 15:02:28 -0700399 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700401
402/**
403 * lru_cache_add: add a page to the page lists
404 * @page: the page to add
405 */
406void lru_cache_add_anon(struct page *page)
407{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700408 if (PageActive(page))
409 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700410 __lru_cache_add(page);
411}
412
413void lru_cache_add_file(struct page *page)
414{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700415 if (PageActive(page))
416 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700417 __lru_cache_add(page);
418}
419EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700421/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700422 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700423 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700424 *
425 * Queue the page for addition to the LRU via pagevec. The decision on whether
426 * to add the page to the [in]active [file|anon] list is deferred until the
427 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
428 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700429 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700430void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Sasha Levin309381fea2014-01-23 15:52:54 -0800432 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
433 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700434 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700437/**
438 * add_page_to_unevictable_list - add a page to the unevictable list
439 * @page: the page to be added to the unevictable list
440 *
441 * Add page directly to its zone's unevictable list. To avoid races with
442 * tasks that might be making the page evictable, through eg. munlock,
443 * munmap or exit, while it's not on the lru, we want to add the page
444 * while it's locked or otherwise "invisible" to other tasks. This is
445 * difficult to do when using the pagevec cache, so bypass that.
446 */
447void add_page_to_unevictable_list(struct page *page)
448{
Mel Gorman599d0c92016-07-28 15:45:31 -0700449 struct pglist_data *pgdat = page_pgdat(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700450 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700451
Mel Gorman599d0c92016-07-28 15:45:31 -0700452 spin_lock_irq(&pgdat->lru_lock);
453 lruvec = mem_cgroup_page_lruvec(page, pgdat);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700454 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700455 SetPageUnevictable(page);
456 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700457 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Mel Gorman599d0c92016-07-28 15:45:31 -0700458 spin_unlock_irq(&pgdat->lru_lock);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700459}
460
Johannes Weiner00501b52014-08-08 14:19:20 -0700461/**
462 * lru_cache_add_active_or_unevictable
463 * @page: the page to be added to LRU
464 * @vma: vma in which page is mapped for determining reclaimability
465 *
466 * Place @page on the active or unevictable LRU list, depending on its
467 * evictability. Note that if the page is not evictable, it goes
468 * directly back onto it's zone's unevictable list, it does NOT use a
469 * per cpu pagevec.
470 */
Laurent Dufour62cd1d92018-04-17 16:33:20 +0200471void __lru_cache_add_active_or_unevictable(struct page *page,
472 unsigned long vma_flags)
Johannes Weiner00501b52014-08-08 14:19:20 -0700473{
474 VM_BUG_ON_PAGE(PageLRU(page), page);
475
Laurent Dufour62cd1d92018-04-17 16:33:20 +0200476 if (likely((vma_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700477 SetPageActive(page);
478 lru_cache_add(page);
479 return;
480 }
481
482 if (!TestSetPageMlocked(page)) {
483 /*
484 * We use the irq-unsafe __mod_zone_page_stat because this
485 * counter is not modified from interrupt context, and the pte
486 * lock is held(spinlock), which implies preemption disabled.
487 */
488 __mod_zone_page_state(page_zone(page), NR_MLOCK,
489 hpage_nr_pages(page));
490 count_vm_event(UNEVICTABLE_PGMLOCKED);
491 }
492 add_page_to_unevictable_list(page);
493}
494
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700495/*
Minchan Kim31560182011-03-22 16:32:52 -0700496 * 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.
Minchan Kim278df9f2011-03-22 16:32:54 -0700501 *
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.
Minchan Kim31560182011-03-22 16:32:52 -0700515 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700516static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700517 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700518{
519 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700520 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700521
Minchan Kim278df9f2011-03-22 16:32:54 -0700522 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700523 return;
524
Minchan Kimbad49d92011-05-11 15:13:30 -0700525 if (PageUnevictable(page))
526 return;
527
Minchan Kim31560182011-03-22 16:32:52 -0700528 /* Some processes are using the page */
529 if (page_mapped(page))
530 return;
531
Minchan Kim278df9f2011-03-22 16:32:54 -0700532 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700533 file = page_is_file_cache(page);
534 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700535
536 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700537 ClearPageActive(page);
538 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700539 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700540
Minchan Kim278df9f2011-03-22 16:32:54 -0700541 if (PageWriteback(page) || PageDirty(page)) {
542 /*
543 * PG_reclaim could be raced with end_page_writeback
544 * It can make readahead confusing. But race window
545 * is _really_ small and it's non-critical problem.
546 */
547 SetPageReclaim(page);
548 } else {
549 /*
550 * The page's writeback ends up during pagevec
551 * We moves tha page into tail of inactive.
552 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800553 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700554 __count_vm_event(PGROTATED);
555 }
556
557 if (active)
558 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700559 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700560}
561
Minchan Kim10853a02016-01-15 16:55:11 -0800562
563static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
564 void *arg)
565{
566 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
567 int file = page_is_file_cache(page);
568 int lru = page_lru_base_type(page);
569
570 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
571 ClearPageActive(page);
572 ClearPageReferenced(page);
573 add_page_to_lru_list(page, lruvec, lru);
574
575 __count_vm_event(PGDEACTIVATE);
576 update_page_reclaim_stat(lruvec, file, 0);
577 }
578}
579
Minchan Kim31560182011-03-22 16:32:52 -0700580/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700581 * Drain pages out of the cpu's pagevecs.
582 * Either "cpu" is the current CPU, and preemption has already been
583 * disabled; or "cpu" is being hot-unplugged, and is already dead.
584 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700585void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586{
Mel Gorman13f7f782013-07-03 15:02:28 -0700587 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Mel Gorman13f7f782013-07-03 15:02:28 -0700589 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700590 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700591
592 pvec = &per_cpu(lru_rotate_pvecs, cpu);
593 if (pagevec_count(pvec)) {
594 unsigned long flags;
595
596 /* No harm done if a racing interrupt already did this */
597 local_irq_save(flags);
598 pagevec_move_tail(pvec);
599 local_irq_restore(flags);
600 }
Minchan Kim31560182011-03-22 16:32:52 -0700601
Minchan Kimcc5993b2015-04-15 16:13:26 -0700602 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700603 if (pagevec_count(pvec))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700604 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700605
Minchan Kim10853a02016-01-15 16:55:11 -0800606 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
607 if (pagevec_count(pvec))
608 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
609
Shaohua Lieb709b02011-05-24 17:12:55 -0700610 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700611}
612
613/**
Minchan Kimcc5993b2015-04-15 16:13:26 -0700614 * deactivate_file_page - forcefully deactivate a file page
Minchan Kim31560182011-03-22 16:32:52 -0700615 * @page: page to deactivate
616 *
617 * This function hints the VM that @page is a good reclaim candidate,
618 * for example if its invalidation fails due to the page being dirty
619 * or under writeback.
620 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700621void deactivate_file_page(struct page *page)
Minchan Kim31560182011-03-22 16:32:52 -0700622{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700623 /*
Minchan Kimcc5993b2015-04-15 16:13:26 -0700624 * In a workload with many unevictable page such as mprotect,
625 * unevictable page deactivation for accelerating reclaim is pointless.
Minchan Kim821ed6b2011-05-24 17:12:31 -0700626 */
627 if (PageUnevictable(page))
628 return;
629
Minchan Kim31560182011-03-22 16:32:52 -0700630 if (likely(get_page_unless_zero(page))) {
Minchan Kimcc5993b2015-04-15 16:13:26 -0700631 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700632
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700633 if (!pagevec_add(pvec, page) || PageCompound(page))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700634 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
635 put_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700636 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800637}
638
Minchan Kim10853a02016-01-15 16:55:11 -0800639/**
640 * deactivate_page - deactivate a page
641 * @page: page to deactivate
642 *
643 * deactivate_page() moves @page to the inactive list if @page was on the active
644 * list and was not an unevictable page. This is done to accelerate the reclaim
645 * of @page.
646 */
647void deactivate_page(struct page *page)
648{
649 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
650 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
651
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300652 get_page(page);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700653 if (!pagevec_add(pvec, page) || PageCompound(page))
Minchan Kim10853a02016-01-15 16:55:11 -0800654 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
655 put_cpu_var(lru_deactivate_pvecs);
656 }
657}
658
Andrew Morton80bfed92006-01-06 00:11:14 -0800659void lru_add_drain(void)
660{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700661 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800662 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
David Howellsc4028952006-11-22 14:57:56 +0000665static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800666{
667 lru_add_drain();
668}
669
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700670static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
671
Wang Sheng-Huif3a932ba2016-06-08 15:33:56 -0700672/*
673 * lru_add_drain_wq is used to do lru_add_drain_all() from a WQ_MEM_RECLAIM
674 * workqueue, aiding in getting memory freed.
675 */
676static struct workqueue_struct *lru_add_drain_wq;
677
678static int __init lru_init(void)
679{
680 lru_add_drain_wq = alloc_workqueue("lru-add-drain", WQ_MEM_RECLAIM, 0);
681
682 if (WARN(!lru_add_drain_wq,
683 "Failed to create workqueue lru_add_drain_wq"))
684 return -ENOMEM;
685
686 return 0;
687}
688early_initcall(lru_init);
689
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700690void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800691{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700692 static DEFINE_MUTEX(lock);
693 static struct cpumask has_work;
694 int cpu;
695
696 mutex_lock(&lock);
697 get_online_cpus();
698 cpumask_clear(&has_work);
699
700 for_each_online_cpu(cpu) {
701 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
702
703 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
704 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
Minchan Kimcc5993b2015-04-15 16:13:26 -0700705 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
Minchan Kim10853a02016-01-15 16:55:11 -0800706 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700707 need_activate_page_drain(cpu)) {
708 INIT_WORK(work, lru_add_drain_per_cpu);
Wang Sheng-Huif3a932ba2016-06-08 15:33:56 -0700709 queue_work_on(cpu, lru_add_drain_wq, work);
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700710 cpumask_set_cpu(cpu, &has_work);
711 }
712 }
713
714 for_each_cpu(cpu, &has_work)
715 flush_work(&per_cpu(lru_add_drain_work, cpu));
716
717 put_online_cpus();
718 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800719}
720
Michal Hockoaabfb572014-10-09 15:28:52 -0700721/**
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300722 * release_pages - batched put_page()
Michal Hockoaabfb572014-10-09 15:28:52 -0700723 * @pages: array of pages to release
724 * @nr: number of pages
725 * @cold: whether the pages are cache cold
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 *
Michal Hockoaabfb572014-10-09 15:28:52 -0700727 * Decrement the reference count on all the pages in @pages. If it
728 * fell to zero, remove the page from the LRU and free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 */
Mel Gormanb745bc82014-06-04 16:10:22 -0700730void release_pages(struct page **pages, int nr, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
732 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800733 LIST_HEAD(pages_to_free);
Mel Gorman599d0c92016-07-28 15:45:31 -0700734 struct pglist_data *locked_pgdat = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700735 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700736 unsigned long uninitialized_var(flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700737 unsigned int uninitialized_var(lock_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 for (i = 0; i < nr; i++) {
740 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Michal Hockoaabfb572014-10-09 15:28:52 -0700742 /*
743 * Make sure the IRQ-safe lock-holding time does not get
744 * excessive with a continuous string of pages from the
Mel Gorman599d0c92016-07-28 15:45:31 -0700745 * same pgdat. The lock is held only if pgdat != NULL.
Michal Hockoaabfb572014-10-09 15:28:52 -0700746 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700747 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
748 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
749 locked_pgdat = NULL;
Michal Hockoaabfb572014-10-09 15:28:52 -0700750 }
751
Aaron Lu6fcb52a2016-10-07 17:00:08 -0700752 if (is_huge_zero_page(page))
Kirill A. Shutemovaa88b682016-04-28 16:18:27 -0700753 continue;
Kirill A. Shutemovaa88b682016-04-28 16:18:27 -0700754
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800755 page = compound_head(page);
Nick Pigginb5810032005-10-29 18:16:12 -0700756 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 continue;
758
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800759 if (PageCompound(page)) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700760 if (locked_pgdat) {
761 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
762 locked_pgdat = NULL;
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800763 }
764 __put_compound_page(page);
765 continue;
766 }
767
Nick Piggin46453a62006-03-22 00:07:58 -0800768 if (PageLRU(page)) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700769 struct pglist_data *pgdat = page_pgdat(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700770
Mel Gorman599d0c92016-07-28 15:45:31 -0700771 if (pgdat != locked_pgdat) {
772 if (locked_pgdat)
773 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700774 flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700775 lock_batch = 0;
Mel Gorman599d0c92016-07-28 15:45:31 -0700776 locked_pgdat = pgdat;
777 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700779
Mel Gorman599d0c92016-07-28 15:45:31 -0700780 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
Sasha Levin309381fea2014-01-23 15:52:54 -0800781 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800782 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700783 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 }
Nick Piggin46453a62006-03-22 00:07:58 -0800785
Mel Gormanc53954a2013-07-03 15:02:34 -0700786 /* Clear Active bit in case of parallel mark_page_accessed */
Mel Gormane3741b52014-06-04 16:10:26 -0700787 __ClearPageActive(page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700788
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800789 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
Mel Gorman599d0c92016-07-28 15:45:31 -0700791 if (locked_pgdat)
792 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Johannes Weiner747db952014-08-08 14:19:24 -0700794 mem_cgroup_uncharge_list(&pages_to_free);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800795 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700797EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
799/*
800 * The pages which we're about to release may be in the deferred lru-addition
801 * queues. That would prevent them from really being freed right now. That's
802 * OK from a correctness point of view but is inefficient - those pages may be
803 * cache-warm and we want to give them back to the page allocator ASAP.
804 *
805 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
806 * and __pagevec_lru_add_active() call release_pages() directly to avoid
807 * mutual recursion.
808 */
809void __pagevec_release(struct pagevec *pvec)
810{
811 lru_add_drain();
812 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
813 pagevec_reinit(pvec);
814}
Steve French7f285702005-11-01 10:22:55 -0800815EXPORT_SYMBOL(__pagevec_release);
816
Hugh Dickins12d27102012-01-12 17:19:52 -0800817#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800818/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700819void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700820 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800821{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800822 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800823
Sasha Levin309381fea2014-01-23 15:52:54 -0800824 VM_BUG_ON_PAGE(!PageHead(page), page);
825 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
826 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700827 VM_BUG_ON(NR_CPUS != 1 &&
Mel Gorman599d0c92016-07-28 15:45:31 -0700828 !spin_is_locked(&lruvec_pgdat(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800829
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700830 if (!list)
831 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800832
Hugh Dickins12d27102012-01-12 17:19:52 -0800833 if (likely(PageLRU(page)))
834 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700835 else if (list) {
836 /* page reclaim is reclaiming a huge page */
837 get_page(page_tail);
838 list_add_tail(&page_tail->lru, list);
839 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800840 struct list_head *list_head;
841 /*
842 * Head page has not yet been counted, as an hpage,
843 * so we must account for each subpage individually.
844 *
845 * Use the standard add function to put page_tail on the list,
846 * but then correct its position so they all end up in order.
847 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700848 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -0800849 list_head = page_tail->lru.prev;
850 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800851 }
Hugh Dickins75121022012-03-05 14:59:18 -0800852
853 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700854 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800855}
Hugh Dickins12d27102012-01-12 17:19:52 -0800856#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800857
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700858static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
859 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700860{
Mel Gorman13f7f782013-07-03 15:02:28 -0700861 int file = page_is_file_cache(page);
862 int active = PageActive(page);
863 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700864
Sasha Levin309381fea2014-01-23 15:52:54 -0800865 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700866
867 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700868 add_page_to_lru_list(page, lruvec, lru);
869 update_page_reclaim_stat(lruvec, file, active);
Mel Gorman24b7e582014-08-06 16:07:11 -0700870 trace_mm_lru_insertion(page, lru);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700871}
872
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 * Add the passed pages to the LRU, then drop the caller's refcount
875 * on them. Reinitialises the caller's pagevec.
876 */
Mel Gormana0b8cab32013-07-03 15:02:32 -0700877void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
Mel Gormana0b8cab32013-07-03 15:02:32 -0700879 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800881EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700884 * pagevec_lookup_entries - gang pagecache lookup
885 * @pvec: Where the resulting entries are placed
886 * @mapping: The address_space to search
887 * @start: The starting entry index
888 * @nr_entries: The maximum number of entries
889 * @indices: The cache indices corresponding to the entries in @pvec
890 *
891 * pagevec_lookup_entries() will search for and return a group of up
892 * to @nr_entries pages and shadow entries in the mapping. All
893 * entries are placed in @pvec. pagevec_lookup_entries() takes a
894 * reference against actual pages in @pvec.
895 *
896 * The search returns a group of mapping-contiguous entries with
897 * ascending indexes. There may be holes in the indices due to
898 * not-present entries.
899 *
900 * pagevec_lookup_entries() returns the number of entries which were
901 * found.
902 */
903unsigned pagevec_lookup_entries(struct pagevec *pvec,
904 struct address_space *mapping,
905 pgoff_t start, unsigned nr_pages,
906 pgoff_t *indices)
907{
908 pvec->nr = find_get_entries(mapping, start, nr_pages,
909 pvec->pages, indices);
910 return pagevec_count(pvec);
911}
912
913/**
914 * pagevec_remove_exceptionals - pagevec exceptionals pruning
915 * @pvec: The pagevec to prune
916 *
917 * pagevec_lookup_entries() fills both pages and exceptional radix
918 * tree entries into the pagevec. This function prunes all
919 * exceptionals from @pvec without leaving holes, so that it can be
920 * passed on to page-only pagevec operations.
921 */
922void pagevec_remove_exceptionals(struct pagevec *pvec)
923{
924 int i, j;
925
926 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
927 struct page *page = pvec->pages[i];
928 if (!radix_tree_exceptional_entry(page))
929 pvec->pages[j++] = page;
930 }
931 pvec->nr = j;
932}
933
934/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 * pagevec_lookup - gang pagecache lookup
936 * @pvec: Where the resulting pages are placed
937 * @mapping: The address_space to search
938 * @start: The starting page index
939 * @nr_pages: The maximum number of pages
940 *
941 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
942 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
943 * reference against the pages in @pvec.
944 *
945 * The search returns a group of mapping-contiguous pages with ascending
946 * indexes. There may be holes in the indices due to not-present pages.
947 *
948 * pagevec_lookup() returns the number of pages which were found.
949 */
950unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
951 pgoff_t start, unsigned nr_pages)
952{
953 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
954 return pagevec_count(pvec);
955}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100956EXPORT_SYMBOL(pagevec_lookup);
957
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
959 pgoff_t *index, int tag, unsigned nr_pages)
960{
961 pvec->nr = find_get_pages_tag(mapping, index, tag,
962 nr_pages, pvec->pages);
963 return pagevec_count(pvec);
964}
Steve French7f285702005-11-01 10:22:55 -0800965EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967/*
968 * Perform any setup for the swap system
969 */
970void __init swap_setup(void)
971{
Jan Beulich44813742009-09-21 17:03:05 -0700972 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700973#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -0800974 int i;
975
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800976 for (i = 0; i < MAX_SWAPFILES; i++)
Shaohua Li33806f02013-02-22 16:34:37 -0800977 spin_lock_init(&swapper_spaces[i].tree_lock);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700978#endif
979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 /* Use a smaller cluster for small-memory machines */
981 if (megs < 16)
982 page_cluster = 2;
983 else
984 page_cluster = 3;
985 /*
986 * Right now other parts of the system means that we
987 * _really_ don't want to cluster much more
988 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989}