blob: 4cffcb48fd98b4595aa63b1df323ed7156f87c9b [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 }
Nicholas Piggin49e19db2016-12-25 13:00:30 +100072 __ClearPageWaiters(page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -070073 mem_cgroup_uncharge(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080074}
75
76static void __put_single_page(struct page *page)
77{
78 __page_cache_release(page);
Mel Gormanb745bc82014-06-04 16:10:22 -070079 free_hot_cold_page(page, false);
Adrian Bunkb2213852006-09-25 23:31:02 -070080}
81
Andrea Arcangeli91807062011-01-13 15:46:32 -080082static void __put_compound_page(struct page *page)
83{
84 compound_page_dtor *dtor;
85
Naoya Horiguchi822fc612015-04-15 16:14:35 -070086 /*
87 * __page_cache_release() is supposed to be called for thp, not for
88 * hugetlb. This is because hugetlb page does never have PageLRU set
89 * (it's never listed to any LRU lists) and no memcg routines should
90 * be called for hugetlb (it has a separate hugetlb_cgroup.)
91 */
92 if (!PageHuge(page))
93 __page_cache_release(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080094 dtor = get_compound_page_dtor(page);
95 (*dtor)(page);
96}
97
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -080098void __put_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099{
Nick Piggin8519fb32006-02-07 12:58:52 -0800100 if (unlikely(PageCompound(page)))
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800101 __put_compound_page(page);
102 else
Andrea Arcangeli91807062011-01-13 15:46:32 -0800103 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104}
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800105EXPORT_SYMBOL(__put_page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700106
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700107/**
Randy Dunlap76824862008-03-19 17:00:40 -0700108 * put_pages_list() - release a list of pages
109 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700110 *
111 * Release a list of pages which are strung together on page.lru. Currently
112 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700113 */
114void put_pages_list(struct list_head *pages)
115{
116 while (!list_empty(pages)) {
117 struct page *victim;
118
119 victim = list_entry(pages->prev, struct page, lru);
120 list_del(&victim->lru);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300121 put_page(victim);
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700122 }
123}
124EXPORT_SYMBOL(put_pages_list);
125
Mel Gorman18022c52012-07-31 16:44:51 -0700126/*
127 * get_kernel_pages() - pin kernel pages in memory
128 * @kiov: An array of struct kvec structures
129 * @nr_segs: number of segments to pin
130 * @write: pinning for read/write, currently ignored
131 * @pages: array that receives pointers to the pages pinned.
132 * Should be at least nr_segs long.
133 *
134 * Returns number of pages pinned. This may be fewer than the number
135 * requested. If nr_pages is 0 or negative, returns 0. If no pages
136 * were pinned, returns -errno. Each page returned must be released
137 * with a put_page() call when it is finished with.
138 */
139int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
140 struct page **pages)
141{
142 int seg;
143
144 for (seg = 0; seg < nr_segs; seg++) {
145 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
146 return seg;
147
Mel Gorman5a178112012-07-31 16:45:02 -0700148 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300149 get_page(pages[seg]);
Mel Gorman18022c52012-07-31 16:44:51 -0700150 }
151
152 return seg;
153}
154EXPORT_SYMBOL_GPL(get_kernel_pages);
155
156/*
157 * get_kernel_page() - pin a kernel page in memory
158 * @start: starting kernel address
159 * @write: pinning for read/write, currently ignored
160 * @pages: array that receives pointer to the page pinned.
161 * Must be at least nr_segs long.
162 *
163 * Returns 1 if page is pinned. If the page was not pinned, returns
164 * -errno. The page returned must be released with a put_page() call
165 * when it is finished with.
166 */
167int get_kernel_page(unsigned long start, int write, struct page **pages)
168{
169 const struct kvec kiov = {
170 .iov_base = (void *)start,
171 .iov_len = PAGE_SIZE
172 };
173
174 return get_kernel_pages(&kiov, 1, write, pages);
175}
176EXPORT_SYMBOL_GPL(get_kernel_page);
177
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700178static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700179 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
180 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700181{
182 int i;
Mel Gorman68eb0732016-07-28 15:47:11 -0700183 struct pglist_data *pgdat = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700184 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700185 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700186
187 for (i = 0; i < pagevec_count(pvec); i++) {
188 struct page *page = pvec->pages[i];
Mel Gorman68eb0732016-07-28 15:47:11 -0700189 struct pglist_data *pagepgdat = page_pgdat(page);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700190
Mel Gorman68eb0732016-07-28 15:47:11 -0700191 if (pagepgdat != pgdat) {
192 if (pgdat)
193 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
194 pgdat = pagepgdat;
195 spin_lock_irqsave(&pgdat->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700196 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700197
Mel Gorman68eb0732016-07-28 15:47:11 -0700198 lruvec = mem_cgroup_page_lruvec(page, pgdat);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700199 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700200 }
Mel Gorman68eb0732016-07-28 15:47:11 -0700201 if (pgdat)
202 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800203 release_pages(pvec->pages, pvec->nr, pvec->cold);
204 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800205}
206
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700207static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
208 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700209{
210 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700211
Johannes Weiner6b25f412017-02-24 14:56:23 -0800212 if (PageLRU(page) && !PageUnevictable(page)) {
213 del_page_from_lru_list(page, lruvec, page_lru(page));
214 ClearPageActive(page);
215 add_page_to_lru_list_tail(page, lruvec, page_lru(page));
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700216 (*pgmoved)++;
217 }
218}
219
220/*
221 * pagevec_move_tail() must be called with IRQ disabled.
222 * Otherwise this may cause nasty races.
223 */
224static void pagevec_move_tail(struct pagevec *pvec)
225{
226 int pgmoved = 0;
227
228 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
229 __count_vm_events(PGROTATED, pgmoved);
230}
231
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700232/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 * Writeback is about to end against a page which has been marked for immediate
234 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700235 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700237void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Johannes Weiner6b25f412017-02-24 14:56:23 -0800239 if (!PageLocked(page) && !PageDirty(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700240 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700241 struct pagevec *pvec;
242 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300244 get_page(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700245 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700246 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700247 if (!pagevec_add(pvec, page) || PageCompound(page))
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700248 pagevec_move_tail(pvec);
249 local_irq_restore(flags);
250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700253static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800254 int file, int rotated)
255{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700256 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800257
258 reclaim_stat->recent_scanned[file]++;
259 if (rotated)
260 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800261}
262
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700263static void __activate_page(struct page *page, struct lruvec *lruvec,
264 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Linus Torvalds7a608572011-01-17 14:42:19 -0800266 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
267 int file = page_is_file_cache(page);
268 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800269
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700270 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800271 SetPageActive(page);
272 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700273 add_page_to_lru_list(page, lruvec, lru);
Mel Gorman24b7e582014-08-06 16:07:11 -0700274 trace_mm_lru_activate(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800275
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700276 __count_vm_event(PGACTIVATE);
277 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800278 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700279}
280
281#ifdef CONFIG_SMP
Shaohua Lieb709b02011-05-24 17:12:55 -0700282static void activate_page_drain(int cpu)
283{
284 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
285
286 if (pagevec_count(pvec))
287 pagevec_lru_move_fn(pvec, __activate_page, NULL);
288}
289
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700290static bool need_activate_page_drain(int cpu)
291{
292 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
293}
294
Shaohua Lieb709b02011-05-24 17:12:55 -0700295void activate_page(struct page *page)
296{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700297 page = compound_head(page);
Shaohua Lieb709b02011-05-24 17:12:55 -0700298 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
299 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
300
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300301 get_page(page);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700302 if (!pagevec_add(pvec, page) || PageCompound(page))
Shaohua Lieb709b02011-05-24 17:12:55 -0700303 pagevec_lru_move_fn(pvec, __activate_page, NULL);
304 put_cpu_var(activate_page_pvecs);
305 }
306}
307
308#else
309static inline void activate_page_drain(int cpu)
310{
311}
312
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700313static bool need_activate_page_drain(int cpu)
314{
315 return false;
316}
317
Shaohua Lieb709b02011-05-24 17:12:55 -0700318void activate_page(struct page *page)
319{
320 struct zone *zone = page_zone(page);
321
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700322 page = compound_head(page);
Mel Gormana52633d2016-07-28 15:45:28 -0700323 spin_lock_irq(zone_lru_lock(zone));
Mel Gorman599d0c92016-07-28 15:45:31 -0700324 __activate_page(page, mem_cgroup_page_lruvec(page, zone->zone_pgdat), NULL);
Mel Gormana52633d2016-07-28 15:45:28 -0700325 spin_unlock_irq(zone_lru_lock(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
Shaohua Lieb709b02011-05-24 17:12:55 -0700327#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Mel Gorman059285a2013-07-03 15:02:30 -0700329static void __lru_cache_activate_page(struct page *page)
330{
331 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
332 int i;
333
334 /*
335 * Search backwards on the optimistic assumption that the page being
336 * activated has just been added to this pagevec. Note that only
337 * the local pagevec is examined as a !PageLRU page could be in the
338 * process of being released, reclaimed, migrated or on a remote
339 * pagevec that is currently being drained. Furthermore, marking
340 * a remote pagevec's page PageActive potentially hits a race where
341 * a page is marked PageActive just after it is added to the inactive
342 * list causing accounting errors and BUG_ON checks to trigger.
343 */
344 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
345 struct page *pagevec_page = pvec->pages[i];
346
347 if (pagevec_page == page) {
348 SetPageActive(page);
349 break;
350 }
351 }
352
353 put_cpu_var(lru_add_pvec);
354}
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356/*
357 * Mark a page as having seen activity.
358 *
359 * inactive,unreferenced -> inactive,referenced
360 * inactive,referenced -> active,unreferenced
361 * active,unreferenced -> active,referenced
Hugh Dickinseb39d612014-08-06 16:06:43 -0700362 *
363 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
364 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800366void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800368 page = compound_head(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700369 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700370 PageReferenced(page)) {
371
372 /*
373 * If the page is on the LRU, queue it for activation via
374 * activate_page_pvecs. Otherwise, assume the page is on a
375 * pagevec, mark it active and it'll be moved to the active
376 * LRU on the next drain.
377 */
378 if (PageLRU(page))
379 activate_page(page);
380 else
381 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700383 if (page_is_file_cache(page))
384 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 } else if (!PageReferenced(page)) {
386 SetPageReferenced(page);
387 }
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700388 if (page_is_idle(page))
389 clear_page_idle(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391EXPORT_SYMBOL(mark_page_accessed);
392
Jianyu Zhan2329d372014-06-04 16:07:31 -0700393static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
Mel Gorman13f7f782013-07-03 15:02:28 -0700395 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
396
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300397 get_page(page);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700398 if (!pagevec_add(pvec, page) || PageCompound(page))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700399 __pagevec_lru_add(pvec);
Mel Gorman13f7f782013-07-03 15:02:28 -0700400 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700402
403/**
404 * lru_cache_add: add a page to the page lists
405 * @page: the page to add
406 */
407void lru_cache_add_anon(struct page *page)
408{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700409 if (PageActive(page))
410 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700411 __lru_cache_add(page);
412}
413
414void lru_cache_add_file(struct page *page)
415{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700416 if (PageActive(page))
417 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700418 __lru_cache_add(page);
419}
420EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700422/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700423 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700424 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700425 *
426 * Queue the page for addition to the LRU via pagevec. The decision on whether
427 * to add the page to the [in]active [file|anon] list is deferred until the
428 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
429 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700430 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700431void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
Sasha Levin309381fea2014-01-23 15:52:54 -0800433 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
434 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700435 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700438/**
439 * add_page_to_unevictable_list - add a page to the unevictable list
440 * @page: the page to be added to the unevictable list
441 *
442 * Add page directly to its zone's unevictable list. To avoid races with
443 * tasks that might be making the page evictable, through eg. munlock,
444 * munmap or exit, while it's not on the lru, we want to add the page
445 * while it's locked or otherwise "invisible" to other tasks. This is
446 * difficult to do when using the pagevec cache, so bypass that.
447 */
448void add_page_to_unevictable_list(struct page *page)
449{
Mel Gorman599d0c92016-07-28 15:45:31 -0700450 struct pglist_data *pgdat = page_pgdat(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700451 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700452
Mel Gorman599d0c92016-07-28 15:45:31 -0700453 spin_lock_irq(&pgdat->lru_lock);
454 lruvec = mem_cgroup_page_lruvec(page, pgdat);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700455 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700456 SetPageUnevictable(page);
457 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700458 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Mel Gorman599d0c92016-07-28 15:45:31 -0700459 spin_unlock_irq(&pgdat->lru_lock);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700460}
461
Johannes Weiner00501b52014-08-08 14:19:20 -0700462/**
463 * lru_cache_add_active_or_unevictable
464 * @page: the page to be added to LRU
465 * @vma: vma in which page is mapped for determining reclaimability
466 *
467 * Place @page on the active or unevictable LRU list, depending on its
468 * evictability. Note that if the page is not evictable, it goes
469 * directly back onto it's zone's unevictable list, it does NOT use a
470 * per cpu pagevec.
471 */
Laurent Dufour62cd1d92018-04-17 16:33:20 +0200472void __lru_cache_add_active_or_unevictable(struct page *page,
473 unsigned long vma_flags)
Johannes Weiner00501b52014-08-08 14:19:20 -0700474{
475 VM_BUG_ON_PAGE(PageLRU(page), page);
476
Laurent Dufour62cd1d92018-04-17 16:33:20 +0200477 if (likely((vma_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700478 SetPageActive(page);
479 lru_cache_add(page);
480 return;
481 }
482
483 if (!TestSetPageMlocked(page)) {
484 /*
485 * We use the irq-unsafe __mod_zone_page_stat because this
486 * counter is not modified from interrupt context, and the pte
487 * lock is held(spinlock), which implies preemption disabled.
488 */
489 __mod_zone_page_state(page_zone(page), NR_MLOCK,
490 hpage_nr_pages(page));
491 count_vm_event(UNEVICTABLE_PGMLOCKED);
492 }
493 add_page_to_unevictable_list(page);
494}
495
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700496/*
Minchan Kim31560182011-03-22 16:32:52 -0700497 * If the page can not be invalidated, it is moved to the
498 * inactive list to speed up its reclaim. It is moved to the
499 * head of the list, rather than the tail, to give the flusher
500 * threads some time to write it out, as this is much more
501 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700502 *
503 * If the page isn't page_mapped and dirty/writeback, the page
504 * could reclaim asap using PG_reclaim.
505 *
506 * 1. active, mapped page -> none
507 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
508 * 3. inactive, mapped page -> none
509 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
510 * 5. inactive, clean -> inactive, tail
511 * 6. Others -> none
512 *
513 * In 4, why it moves inactive's head, the VM expects the page would
514 * be write it out by flusher threads as this is much more effective
515 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700516 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700517static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700518 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700519{
520 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700521 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700522
Minchan Kim278df9f2011-03-22 16:32:54 -0700523 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700524 return;
525
Minchan Kimbad49d92011-05-11 15:13:30 -0700526 if (PageUnevictable(page))
527 return;
528
Minchan Kim31560182011-03-22 16:32:52 -0700529 /* Some processes are using the page */
530 if (page_mapped(page))
531 return;
532
Minchan Kim278df9f2011-03-22 16:32:54 -0700533 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700534 file = page_is_file_cache(page);
535 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700536
537 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700538 ClearPageActive(page);
539 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700540 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700541
Minchan Kim278df9f2011-03-22 16:32:54 -0700542 if (PageWriteback(page) || PageDirty(page)) {
543 /*
544 * PG_reclaim could be raced with end_page_writeback
545 * It can make readahead confusing. But race window
546 * is _really_ small and it's non-critical problem.
547 */
548 SetPageReclaim(page);
549 } else {
550 /*
551 * The page's writeback ends up during pagevec
552 * We moves tha page into tail of inactive.
553 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800554 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700555 __count_vm_event(PGROTATED);
556 }
557
558 if (active)
559 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700560 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700561}
562
Minchan Kim10853a02016-01-15 16:55:11 -0800563
564static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
565 void *arg)
566{
567 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
568 int file = page_is_file_cache(page);
569 int lru = page_lru_base_type(page);
570
571 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
572 ClearPageActive(page);
573 ClearPageReferenced(page);
574 add_page_to_lru_list(page, lruvec, lru);
575
576 __count_vm_event(PGDEACTIVATE);
577 update_page_reclaim_stat(lruvec, file, 0);
578 }
579}
580
Minchan Kim31560182011-03-22 16:32:52 -0700581/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700582 * Drain pages out of the cpu's pagevecs.
583 * Either "cpu" is the current CPU, and preemption has already been
584 * disabled; or "cpu" is being hot-unplugged, and is already dead.
585 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700586void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
Mel Gorman13f7f782013-07-03 15:02:28 -0700588 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Mel Gorman13f7f782013-07-03 15:02:28 -0700590 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700591 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700592
593 pvec = &per_cpu(lru_rotate_pvecs, cpu);
594 if (pagevec_count(pvec)) {
595 unsigned long flags;
596
597 /* No harm done if a racing interrupt already did this */
598 local_irq_save(flags);
599 pagevec_move_tail(pvec);
600 local_irq_restore(flags);
601 }
Minchan Kim31560182011-03-22 16:32:52 -0700602
Minchan Kimcc5993b2015-04-15 16:13:26 -0700603 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700604 if (pagevec_count(pvec))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700605 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700606
Minchan Kim10853a02016-01-15 16:55:11 -0800607 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
608 if (pagevec_count(pvec))
609 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
610
Shaohua Lieb709b02011-05-24 17:12:55 -0700611 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700612}
613
614/**
Minchan Kimcc5993b2015-04-15 16:13:26 -0700615 * deactivate_file_page - forcefully deactivate a file page
Minchan Kim31560182011-03-22 16:32:52 -0700616 * @page: page to deactivate
617 *
618 * This function hints the VM that @page is a good reclaim candidate,
619 * for example if its invalidation fails due to the page being dirty
620 * or under writeback.
621 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700622void deactivate_file_page(struct page *page)
Minchan Kim31560182011-03-22 16:32:52 -0700623{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700624 /*
Minchan Kimcc5993b2015-04-15 16:13:26 -0700625 * In a workload with many unevictable page such as mprotect,
626 * unevictable page deactivation for accelerating reclaim is pointless.
Minchan Kim821ed6b2011-05-24 17:12:31 -0700627 */
628 if (PageUnevictable(page))
629 return;
630
Minchan Kim31560182011-03-22 16:32:52 -0700631 if (likely(get_page_unless_zero(page))) {
Minchan Kimcc5993b2015-04-15 16:13:26 -0700632 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700633
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700634 if (!pagevec_add(pvec, page) || PageCompound(page))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700635 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
636 put_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700637 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800638}
639
Minchan Kim10853a02016-01-15 16:55:11 -0800640/**
641 * deactivate_page - deactivate a page
642 * @page: page to deactivate
643 *
644 * deactivate_page() moves @page to the inactive list if @page was on the active
645 * list and was not an unevictable page. This is done to accelerate the reclaim
646 * of @page.
647 */
648void deactivate_page(struct page *page)
649{
650 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
651 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
652
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300653 get_page(page);
Lukasz Odzioba8f182272016-06-24 14:50:01 -0700654 if (!pagevec_add(pvec, page) || PageCompound(page))
Minchan Kim10853a02016-01-15 16:55:11 -0800655 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
656 put_cpu_var(lru_deactivate_pvecs);
657 }
658}
659
Andrew Morton80bfed92006-01-06 00:11:14 -0800660void lru_add_drain(void)
661{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700662 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800663 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
David Howellsc4028952006-11-22 14:57:56 +0000666static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800667{
668 lru_add_drain();
669}
670
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700671static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
672
Wang Sheng-Huif3a932ba2016-06-08 15:33:56 -0700673/*
674 * lru_add_drain_wq is used to do lru_add_drain_all() from a WQ_MEM_RECLAIM
675 * workqueue, aiding in getting memory freed.
676 */
677static struct workqueue_struct *lru_add_drain_wq;
678
679static int __init lru_init(void)
680{
681 lru_add_drain_wq = alloc_workqueue("lru-add-drain", WQ_MEM_RECLAIM, 0);
682
683 if (WARN(!lru_add_drain_wq,
684 "Failed to create workqueue lru_add_drain_wq"))
685 return -ENOMEM;
686
687 return 0;
688}
689early_initcall(lru_init);
690
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700691void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800692{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700693 static DEFINE_MUTEX(lock);
694 static struct cpumask has_work;
695 int cpu;
696
697 mutex_lock(&lock);
698 get_online_cpus();
699 cpumask_clear(&has_work);
700
701 for_each_online_cpu(cpu) {
702 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
703
704 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
705 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
Minchan Kimcc5993b2015-04-15 16:13:26 -0700706 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
Minchan Kim10853a02016-01-15 16:55:11 -0800707 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700708 need_activate_page_drain(cpu)) {
709 INIT_WORK(work, lru_add_drain_per_cpu);
Wang Sheng-Huif3a932ba2016-06-08 15:33:56 -0700710 queue_work_on(cpu, lru_add_drain_wq, work);
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700711 cpumask_set_cpu(cpu, &has_work);
712 }
713 }
714
715 for_each_cpu(cpu, &has_work)
716 flush_work(&per_cpu(lru_add_drain_work, cpu));
717
718 put_online_cpus();
719 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800720}
721
Michal Hockoaabfb572014-10-09 15:28:52 -0700722/**
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300723 * release_pages - batched put_page()
Michal Hockoaabfb572014-10-09 15:28:52 -0700724 * @pages: array of pages to release
725 * @nr: number of pages
726 * @cold: whether the pages are cache cold
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 *
Michal Hockoaabfb572014-10-09 15:28:52 -0700728 * Decrement the reference count on all the pages in @pages. If it
729 * fell to zero, remove the page from the LRU and free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 */
Mel Gormanb745bc82014-06-04 16:10:22 -0700731void release_pages(struct page **pages, int nr, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800734 LIST_HEAD(pages_to_free);
Mel Gorman599d0c92016-07-28 15:45:31 -0700735 struct pglist_data *locked_pgdat = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700736 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700737 unsigned long uninitialized_var(flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700738 unsigned int uninitialized_var(lock_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 for (i = 0; i < nr; i++) {
741 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
Michal Hockoaabfb572014-10-09 15:28:52 -0700743 /*
744 * Make sure the IRQ-safe lock-holding time does not get
745 * excessive with a continuous string of pages from the
Mel Gorman599d0c92016-07-28 15:45:31 -0700746 * same pgdat. The lock is held only if pgdat != NULL.
Michal Hockoaabfb572014-10-09 15:28:52 -0700747 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700748 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
749 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
750 locked_pgdat = NULL;
Michal Hockoaabfb572014-10-09 15:28:52 -0700751 }
752
Aaron Lu6fcb52a2016-10-07 17:00:08 -0700753 if (is_huge_zero_page(page))
Kirill A. Shutemovaa88b682016-04-28 16:18:27 -0700754 continue;
Kirill A. Shutemovaa88b682016-04-28 16:18:27 -0700755
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800756 page = compound_head(page);
Nick Pigginb5810032005-10-29 18:16:12 -0700757 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 continue;
759
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800760 if (PageCompound(page)) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700761 if (locked_pgdat) {
762 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
763 locked_pgdat = NULL;
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800764 }
765 __put_compound_page(page);
766 continue;
767 }
768
Nick Piggin46453a62006-03-22 00:07:58 -0800769 if (PageLRU(page)) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700770 struct pglist_data *pgdat = page_pgdat(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700771
Mel Gorman599d0c92016-07-28 15:45:31 -0700772 if (pgdat != locked_pgdat) {
773 if (locked_pgdat)
774 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700775 flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700776 lock_batch = 0;
Mel Gorman599d0c92016-07-28 15:45:31 -0700777 locked_pgdat = pgdat;
778 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700780
Mel Gorman599d0c92016-07-28 15:45:31 -0700781 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
Sasha Levin309381fea2014-01-23 15:52:54 -0800782 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800783 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700784 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
Nick Piggin46453a62006-03-22 00:07:58 -0800786
Mel Gormanc53954a2013-07-03 15:02:34 -0700787 /* Clear Active bit in case of parallel mark_page_accessed */
Mel Gormane3741b52014-06-04 16:10:26 -0700788 __ClearPageActive(page);
Nicholas Piggin49e19db2016-12-25 13:00:30 +1000789 __ClearPageWaiters(page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700790
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800791 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
Mel Gorman599d0c92016-07-28 15:45:31 -0700793 if (locked_pgdat)
794 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Johannes Weiner747db952014-08-08 14:19:24 -0700796 mem_cgroup_uncharge_list(&pages_to_free);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800797 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700799EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801/*
802 * The pages which we're about to release may be in the deferred lru-addition
803 * queues. That would prevent them from really being freed right now. That's
804 * OK from a correctness point of view but is inefficient - those pages may be
805 * cache-warm and we want to give them back to the page allocator ASAP.
806 *
807 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
808 * and __pagevec_lru_add_active() call release_pages() directly to avoid
809 * mutual recursion.
810 */
811void __pagevec_release(struct pagevec *pvec)
812{
813 lru_add_drain();
814 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
815 pagevec_reinit(pvec);
816}
Steve French7f285702005-11-01 10:22:55 -0800817EXPORT_SYMBOL(__pagevec_release);
818
Hugh Dickins12d27102012-01-12 17:19:52 -0800819#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800820/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700821void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700822 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800823{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800824 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800825
Sasha Levin309381fea2014-01-23 15:52:54 -0800826 VM_BUG_ON_PAGE(!PageHead(page), page);
827 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
828 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700829 VM_BUG_ON(NR_CPUS != 1 &&
Mel Gorman599d0c92016-07-28 15:45:31 -0700830 !spin_is_locked(&lruvec_pgdat(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800831
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700832 if (!list)
833 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800834
Hugh Dickins12d27102012-01-12 17:19:52 -0800835 if (likely(PageLRU(page)))
836 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700837 else if (list) {
838 /* page reclaim is reclaiming a huge page */
839 get_page(page_tail);
840 list_add_tail(&page_tail->lru, list);
841 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800842 struct list_head *list_head;
843 /*
844 * Head page has not yet been counted, as an hpage,
845 * so we must account for each subpage individually.
846 *
847 * Use the standard add function to put page_tail on the list,
848 * but then correct its position so they all end up in order.
849 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700850 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -0800851 list_head = page_tail->lru.prev;
852 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800853 }
Hugh Dickins75121022012-03-05 14:59:18 -0800854
855 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700856 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800857}
Hugh Dickins12d27102012-01-12 17:19:52 -0800858#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800859
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700860static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
861 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700862{
Mel Gorman13f7f782013-07-03 15:02:28 -0700863 int file = page_is_file_cache(page);
864 int active = PageActive(page);
865 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700866
Sasha Levin309381fea2014-01-23 15:52:54 -0800867 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700868
869 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700870 add_page_to_lru_list(page, lruvec, lru);
871 update_page_reclaim_stat(lruvec, file, active);
Mel Gorman24b7e582014-08-06 16:07:11 -0700872 trace_mm_lru_insertion(page, lru);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700873}
874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 * Add the passed pages to the LRU, then drop the caller's refcount
877 * on them. Reinitialises the caller's pagevec.
878 */
Mel Gormana0b8cab32013-07-03 15:02:32 -0700879void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880{
Mel Gormana0b8cab32013-07-03 15:02:32 -0700881 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800883EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700886 * pagevec_lookup_entries - gang pagecache lookup
887 * @pvec: Where the resulting entries are placed
888 * @mapping: The address_space to search
889 * @start: The starting entry index
890 * @nr_entries: The maximum number of entries
891 * @indices: The cache indices corresponding to the entries in @pvec
892 *
893 * pagevec_lookup_entries() will search for and return a group of up
894 * to @nr_entries pages and shadow entries in the mapping. All
895 * entries are placed in @pvec. pagevec_lookup_entries() takes a
896 * reference against actual pages in @pvec.
897 *
898 * The search returns a group of mapping-contiguous entries with
899 * ascending indexes. There may be holes in the indices due to
900 * not-present entries.
901 *
902 * pagevec_lookup_entries() returns the number of entries which were
903 * found.
904 */
905unsigned pagevec_lookup_entries(struct pagevec *pvec,
906 struct address_space *mapping,
907 pgoff_t start, unsigned nr_pages,
908 pgoff_t *indices)
909{
910 pvec->nr = find_get_entries(mapping, start, nr_pages,
911 pvec->pages, indices);
912 return pagevec_count(pvec);
913}
914
915/**
916 * pagevec_remove_exceptionals - pagevec exceptionals pruning
917 * @pvec: The pagevec to prune
918 *
919 * pagevec_lookup_entries() fills both pages and exceptional radix
920 * tree entries into the pagevec. This function prunes all
921 * exceptionals from @pvec without leaving holes, so that it can be
922 * passed on to page-only pagevec operations.
923 */
924void pagevec_remove_exceptionals(struct pagevec *pvec)
925{
926 int i, j;
927
928 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
929 struct page *page = pvec->pages[i];
930 if (!radix_tree_exceptional_entry(page))
931 pvec->pages[j++] = page;
932 }
933 pvec->nr = j;
934}
935
936/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 * pagevec_lookup - gang pagecache lookup
938 * @pvec: Where the resulting pages are placed
939 * @mapping: The address_space to search
940 * @start: The starting page index
941 * @nr_pages: The maximum number of pages
942 *
943 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
944 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
945 * reference against the pages in @pvec.
946 *
947 * The search returns a group of mapping-contiguous pages with ascending
948 * indexes. There may be holes in the indices due to not-present pages.
949 *
950 * pagevec_lookup() returns the number of pages which were found.
951 */
952unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
953 pgoff_t start, unsigned nr_pages)
954{
955 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
956 return pagevec_count(pvec);
957}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100958EXPORT_SYMBOL(pagevec_lookup);
959
Jan Karafb296a22017-11-15 17:34:33 -0800960unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
961 struct address_space *mapping, pgoff_t *index, pgoff_t end,
Jan Kara4d93a432017-11-15 17:35:19 -0800962 int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
Jan Karafb296a22017-11-15 17:34:33 -0800964 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
Jan Kara4d93a432017-11-15 17:35:19 -0800965 PAGEVEC_SIZE, pvec->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 return pagevec_count(pvec);
967}
Jan Karafb296a22017-11-15 17:34:33 -0800968EXPORT_SYMBOL(pagevec_lookup_range_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Jan Kara565d7442017-11-15 17:35:12 -0800970unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
971 struct address_space *mapping, pgoff_t *index, pgoff_t end,
972 int tag, unsigned max_pages)
973{
974 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
975 min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
976 return pagevec_count(pvec);
977}
978EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979/*
980 * Perform any setup for the swap system
981 */
982void __init swap_setup(void)
983{
Jan Beulich44813742009-09-21 17:03:05 -0700984 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700985#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -0800986 int i;
987
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -0800988 for (i = 0; i < MAX_SWAPFILES; i++)
Shaohua Li33806f02013-02-22 16:34:37 -0800989 spin_lock_init(&swapper_spaces[i].tree_lock);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700990#endif
991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 /* Use a smaller cluster for small-memory machines */
993 if (megs < 16)
994 page_cluster = 2;
995 else
996 page_cluster = 3;
997 /*
998 * Right now other parts of the system means that we
999 * _really_ don't want to cluster much more
1000 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001}