blob: 4e7e2ec67078f783750eb43e8dc45db5600b1b94 [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>
27#include <linux/percpu.h>
28#include <linux/cpu.h>
29#include <linux/notifier.h>
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -070030#include <linux/backing-dev.h>
Balbir Singh66e17072008-02-07 00:13:56 -080031#include <linux/memcontrol.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Lee Schermerhorn64d65192008-10-18 20:26:52 -070034#include "internal.h"
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036/* How many pages do we try to swap or page in/out together? */
37int page_cluster;
38
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -070039static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
Vegard Nossumf84f95042008-07-23 21:28:14 -070040static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -070041static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070042
Adrian Bunkb2213852006-09-25 23:31:02 -070043/*
44 * This path almost never happens for VM activity - pages are normally
45 * freed via pagevecs. But it gets used by networking.
46 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080047static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070048{
49 if (PageLRU(page)) {
Adrian Bunkb2213852006-09-25 23:31:02 -070050 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070051 struct lruvec *lruvec;
52 unsigned long flags;
Adrian Bunkb2213852006-09-25 23:31:02 -070053
54 spin_lock_irqsave(&zone->lru_lock, flags);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070055 lruvec = mem_cgroup_page_lruvec(page, zone);
Adrian Bunkb2213852006-09-25 23:31:02 -070056 VM_BUG_ON(!PageLRU(page));
57 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070058 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070059 spin_unlock_irqrestore(&zone->lru_lock, flags);
60 }
Andrea Arcangeli91807062011-01-13 15:46:32 -080061}
62
63static void __put_single_page(struct page *page)
64{
65 __page_cache_release(page);
Li Hongfc916682010-03-05 13:41:54 -080066 free_hot_cold_page(page, 0);
Adrian Bunkb2213852006-09-25 23:31:02 -070067}
68
Andrea Arcangeli91807062011-01-13 15:46:32 -080069static void __put_compound_page(struct page *page)
70{
71 compound_page_dtor *dtor;
72
73 __page_cache_release(page);
74 dtor = get_compound_page_dtor(page);
75 (*dtor)(page);
76}
77
Nick Piggin8519fb32006-02-07 12:58:52 -080078static void put_compound_page(struct page *page)
79{
Andrea Arcangeli91807062011-01-13 15:46:32 -080080 if (unlikely(PageTail(page))) {
81 /* __split_huge_page_refcount can run under us */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070082 struct page *page_head = compound_trans_head(page);
83
84 if (likely(page != page_head &&
85 get_page_unless_zero(page_head))) {
Andrea Arcangeli91807062011-01-13 15:46:32 -080086 unsigned long flags;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -070087
88 /*
89 * THP can not break up slab pages so avoid taking
90 * compound_lock(). Slab performs non-atomic bit ops
91 * on page->flags for better performance. In particular
92 * slab_unlock() in slub used to be a hot path. It is
93 * still hot on arches that do not support
94 * this_cpu_cmpxchg_double().
95 */
96 if (PageSlab(page_head)) {
97 if (PageTail(page)) {
98 if (put_page_testzero(page_head))
99 VM_BUG_ON(1);
100
101 atomic_dec(&page->_mapcount);
102 goto skip_lock_tail;
103 } else
104 goto skip_lock;
105 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800106 /*
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700107 * page_head wasn't a dangling pointer but it
108 * may not be a head page anymore by the time
109 * we obtain the lock. That is ok as long as it
110 * can't be freed from under us.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800111 */
Andrea Arcangeli91807062011-01-13 15:46:32 -0800112 flags = compound_lock_irqsave(page_head);
113 if (unlikely(!PageTail(page))) {
114 /* __split_huge_page_refcount run before us */
115 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700116skip_lock:
Andrea Arcangeli91807062011-01-13 15:46:32 -0800117 if (put_page_testzero(page_head))
118 __put_single_page(page_head);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700119out_put_single:
Andrea Arcangeli91807062011-01-13 15:46:32 -0800120 if (put_page_testzero(page))
121 __put_single_page(page);
122 return;
123 }
124 VM_BUG_ON(page_head != page->first_page);
125 /*
126 * We can release the refcount taken by
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700127 * get_page_unless_zero() now that
128 * __split_huge_page_refcount() is blocked on
129 * the compound_lock.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800130 */
131 if (put_page_testzero(page_head))
132 VM_BUG_ON(1);
133 /* __split_huge_page_refcount will wait now */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700134 VM_BUG_ON(page_mapcount(page) <= 0);
135 atomic_dec(&page->_mapcount);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800136 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700137 VM_BUG_ON(atomic_read(&page->_count) != 0);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800138 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700139
140skip_lock_tail:
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800141 if (put_page_testzero(page_head)) {
142 if (PageHead(page_head))
143 __put_compound_page(page_head);
144 else
145 __put_single_page(page_head);
146 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800147 } else {
148 /* page_head is a dangling pointer */
149 VM_BUG_ON(PageTail(page));
150 goto out_put_single;
151 }
152 } else if (put_page_testzero(page)) {
153 if (PageHead(page))
154 __put_compound_page(page);
155 else
156 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800157 }
158}
159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160void put_page(struct page *page)
161{
Nick Piggin8519fb32006-02-07 12:58:52 -0800162 if (unlikely(PageCompound(page)))
163 put_compound_page(page);
164 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800165 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700169/*
170 * This function is exported but must not be called by anything other
171 * than get_page(). It implements the slow path of get_page().
172 */
173bool __get_page_tail(struct page *page)
174{
175 /*
176 * This takes care of get_page() if run on a tail page
177 * returned by one of the get_user_pages/follow_page variants.
178 * get_user_pages/follow_page itself doesn't need the compound
179 * lock because it runs __get_page_tail_foll() under the
180 * proper PT lock that already serializes against
181 * split_huge_page().
182 */
183 unsigned long flags;
184 bool got = false;
185 struct page *page_head = compound_trans_head(page);
186
187 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700188
189 /* Ref to put_compound_page() comment. */
190 if (PageSlab(page_head)) {
191 if (likely(PageTail(page))) {
192 __get_page_tail_foll(page, false);
193 return true;
194 } else {
195 put_page(page_head);
196 return false;
197 }
198 }
199
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700200 /*
201 * page_head wasn't a dangling pointer but it
202 * may not be a head page anymore by the time
203 * we obtain the lock. That is ok as long as it
204 * can't be freed from under us.
205 */
206 flags = compound_lock_irqsave(page_head);
207 /* here __split_huge_page_refcount won't run anymore */
208 if (likely(PageTail(page))) {
209 __get_page_tail_foll(page, false);
210 got = true;
211 }
212 compound_unlock_irqrestore(page_head, flags);
213 if (unlikely(!got))
214 put_page(page_head);
215 }
216 return got;
217}
218EXPORT_SYMBOL(__get_page_tail);
219
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700220/**
Randy Dunlap76824862008-03-19 17:00:40 -0700221 * put_pages_list() - release a list of pages
222 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700223 *
224 * Release a list of pages which are strung together on page.lru. Currently
225 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700226 */
227void put_pages_list(struct list_head *pages)
228{
229 while (!list_empty(pages)) {
230 struct page *victim;
231
232 victim = list_entry(pages->prev, struct page, lru);
233 list_del(&victim->lru);
234 page_cache_release(victim);
235 }
236}
237EXPORT_SYMBOL(put_pages_list);
238
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700239static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700240 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
241 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700242{
243 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700244 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700245 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700246 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700247
248 for (i = 0; i < pagevec_count(pvec); i++) {
249 struct page *page = pvec->pages[i];
250 struct zone *pagezone = page_zone(page);
251
252 if (pagezone != zone) {
253 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700254 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700255 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700256 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700257 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700258
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700259 lruvec = mem_cgroup_page_lruvec(page, zone);
260 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700261 }
262 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700263 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800264 release_pages(pvec->pages, pvec->nr, pvec->cold);
265 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800266}
267
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700268static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
269 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700270{
271 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700272
273 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
274 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800275 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700276 (*pgmoved)++;
277 }
278}
279
280/*
281 * pagevec_move_tail() must be called with IRQ disabled.
282 * Otherwise this may cause nasty races.
283 */
284static void pagevec_move_tail(struct pagevec *pvec)
285{
286 int pgmoved = 0;
287
288 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
289 __count_vm_events(PGROTATED, pgmoved);
290}
291
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700292/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 * Writeback is about to end against a page which has been marked for immediate
294 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700295 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700297void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700299 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700300 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700301 struct pagevec *pvec;
302 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700304 page_cache_get(page);
305 local_irq_save(flags);
306 pvec = &__get_cpu_var(lru_rotate_pvecs);
307 if (!pagevec_add(pvec, page))
308 pagevec_move_tail(pvec);
309 local_irq_restore(flags);
310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311}
312
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700313static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800314 int file, int rotated)
315{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700316 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800317
318 reclaim_stat->recent_scanned[file]++;
319 if (rotated)
320 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800321}
322
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700323static void __activate_page(struct page *page, struct lruvec *lruvec,
324 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
Linus Torvalds7a608572011-01-17 14:42:19 -0800326 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
327 int file = page_is_file_cache(page);
328 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800329
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700330 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800331 SetPageActive(page);
332 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700333 add_page_to_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800334
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700335 __count_vm_event(PGACTIVATE);
336 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800337 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700338}
339
340#ifdef CONFIG_SMP
341static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
342
343static void activate_page_drain(int cpu)
344{
345 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
346
347 if (pagevec_count(pvec))
348 pagevec_lru_move_fn(pvec, __activate_page, NULL);
349}
350
351void activate_page(struct page *page)
352{
353 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
354 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
355
356 page_cache_get(page);
357 if (!pagevec_add(pvec, page))
358 pagevec_lru_move_fn(pvec, __activate_page, NULL);
359 put_cpu_var(activate_page_pvecs);
360 }
361}
362
363#else
364static inline void activate_page_drain(int cpu)
365{
366}
367
368void activate_page(struct page *page)
369{
370 struct zone *zone = page_zone(page);
371
372 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700373 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 spin_unlock_irq(&zone->lru_lock);
375}
Shaohua Lieb709b02011-05-24 17:12:55 -0700376#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
378/*
379 * Mark a page as having seen activity.
380 *
381 * inactive,unreferenced -> inactive,referenced
382 * inactive,referenced -> active,unreferenced
383 * active,unreferenced -> active,referenced
384 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800385void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700387 if (!PageActive(page) && !PageUnevictable(page) &&
388 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 activate_page(page);
390 ClearPageReferenced(page);
391 } else if (!PageReferenced(page)) {
392 SetPageReferenced(page);
393 }
394}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395EXPORT_SYMBOL(mark_page_accessed);
396
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700397void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700399 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
401 page_cache_get(page);
402 if (!pagevec_add(pvec, page))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800403 __pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 put_cpu_var(lru_add_pvecs);
405}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200406EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700408/**
409 * lru_cache_add_lru - add a page to a page list
410 * @page: the page to be added to the LRU.
411 * @lru: the LRU list to which the page is added.
412 */
413void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700415 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700416 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700417 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700418 } else if (PageUnevictable(page)) {
419 VM_BUG_ON(PageActive(page));
420 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700423 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700424 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425}
426
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700427/**
428 * add_page_to_unevictable_list - add a page to the unevictable list
429 * @page: the page to be added to the unevictable list
430 *
431 * Add page directly to its zone's unevictable list. To avoid races with
432 * tasks that might be making the page evictable, through eg. munlock,
433 * munmap or exit, while it's not on the lru, we want to add the page
434 * while it's locked or otherwise "invisible" to other tasks. This is
435 * difficult to do when using the pagevec cache, so bypass that.
436 */
437void add_page_to_unevictable_list(struct page *page)
438{
439 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700440 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700441
442 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700443 lruvec = mem_cgroup_page_lruvec(page, zone);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700444 SetPageUnevictable(page);
445 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700446 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700447 spin_unlock_irq(&zone->lru_lock);
448}
449
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700450/*
Minchan Kim31560182011-03-22 16:32:52 -0700451 * If the page can not be invalidated, it is moved to the
452 * inactive list to speed up its reclaim. It is moved to the
453 * head of the list, rather than the tail, to give the flusher
454 * threads some time to write it out, as this is much more
455 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700456 *
457 * If the page isn't page_mapped and dirty/writeback, the page
458 * could reclaim asap using PG_reclaim.
459 *
460 * 1. active, mapped page -> none
461 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
462 * 3. inactive, mapped page -> none
463 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
464 * 5. inactive, clean -> inactive, tail
465 * 6. Others -> none
466 *
467 * In 4, why it moves inactive's head, the VM expects the page would
468 * be write it out by flusher threads as this is much more effective
469 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700470 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700471static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
472 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700473{
474 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700475 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700476
Minchan Kim278df9f2011-03-22 16:32:54 -0700477 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700478 return;
479
Minchan Kimbad49d92011-05-11 15:13:30 -0700480 if (PageUnevictable(page))
481 return;
482
Minchan Kim31560182011-03-22 16:32:52 -0700483 /* Some processes are using the page */
484 if (page_mapped(page))
485 return;
486
Minchan Kim278df9f2011-03-22 16:32:54 -0700487 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700488 file = page_is_file_cache(page);
489 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700490
491 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700492 ClearPageActive(page);
493 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700494 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700495
Minchan Kim278df9f2011-03-22 16:32:54 -0700496 if (PageWriteback(page) || PageDirty(page)) {
497 /*
498 * PG_reclaim could be raced with end_page_writeback
499 * It can make readahead confusing. But race window
500 * is _really_ small and it's non-critical problem.
501 */
502 SetPageReclaim(page);
503 } else {
504 /*
505 * The page's writeback ends up during pagevec
506 * We moves tha page into tail of inactive.
507 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800508 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700509 __count_vm_event(PGROTATED);
510 }
511
512 if (active)
513 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700514 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700515}
516
Minchan Kim31560182011-03-22 16:32:52 -0700517/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700518 * Drain pages out of the cpu's pagevecs.
519 * Either "cpu" is the current CPU, and preemption has already been
520 * disabled; or "cpu" is being hot-unplugged, and is already dead.
521 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700522void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700524 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700525 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700526 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700528 for_each_lru(lru) {
529 pvec = &pvecs[lru - LRU_BASE];
530 if (pagevec_count(pvec))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800531 __pagevec_lru_add(pvec, lru);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700532 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700533
534 pvec = &per_cpu(lru_rotate_pvecs, cpu);
535 if (pagevec_count(pvec)) {
536 unsigned long flags;
537
538 /* No harm done if a racing interrupt already did this */
539 local_irq_save(flags);
540 pagevec_move_tail(pvec);
541 local_irq_restore(flags);
542 }
Minchan Kim31560182011-03-22 16:32:52 -0700543
544 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
545 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700546 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700547
548 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700549}
550
551/**
552 * deactivate_page - forcefully deactivate a page
553 * @page: page to deactivate
554 *
555 * This function hints the VM that @page is a good reclaim candidate,
556 * for example if its invalidation fails due to the page being dirty
557 * or under writeback.
558 */
559void deactivate_page(struct page *page)
560{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700561 /*
562 * In a workload with many unevictable page such as mprotect, unevictable
563 * page deactivation for accelerating reclaim is pointless.
564 */
565 if (PageUnevictable(page))
566 return;
567
Minchan Kim31560182011-03-22 16:32:52 -0700568 if (likely(get_page_unless_zero(page))) {
569 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
570
571 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700572 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700573 put_cpu_var(lru_deactivate_pvecs);
574 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800575}
576
577void lru_add_drain(void)
578{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700579 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800580 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581}
582
David Howellsc4028952006-11-22 14:57:56 +0000583static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800584{
585 lru_add_drain();
586}
587
588/*
589 * Returns 0 for success
590 */
591int lru_add_drain_all(void)
592{
David Howellsc4028952006-11-22 14:57:56 +0000593 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800594}
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 * Batched page_cache_release(). Decrement the reference count on all the
598 * passed pages. If it fell to zero then remove the page from the LRU and
599 * free it.
600 *
601 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
602 * for the remainder of the operation.
603 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700604 * The locking in this function is against shrink_inactive_list(): we recheck
605 * the page count inside the lock to see whether shrink_inactive_list()
606 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
607 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 */
609void release_pages(struct page **pages, int nr, int cold)
610{
611 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800612 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700614 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700615 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 for (i = 0; i < nr; i++) {
618 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Nick Piggin8519fb32006-02-07 12:58:52 -0800620 if (unlikely(PageCompound(page))) {
621 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700622 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800623 zone = NULL;
624 }
625 put_compound_page(page);
626 continue;
627 }
628
Nick Pigginb5810032005-10-29 18:16:12 -0700629 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 continue;
631
Nick Piggin46453a62006-03-22 00:07:58 -0800632 if (PageLRU(page)) {
633 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700634
Nick Piggin46453a62006-03-22 00:07:58 -0800635 if (pagezone != zone) {
636 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700637 spin_unlock_irqrestore(&zone->lru_lock,
638 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800639 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700640 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700642
643 lruvec = mem_cgroup_page_lruvec(page, zone);
Nick Piggin725d7042006-09-25 23:30:55 -0700644 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800645 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700646 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 }
Nick Piggin46453a62006-03-22 00:07:58 -0800648
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800649 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 }
651 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700652 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800654 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700656EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658/*
659 * The pages which we're about to release may be in the deferred lru-addition
660 * queues. That would prevent them from really being freed right now. That's
661 * OK from a correctness point of view but is inefficient - those pages may be
662 * cache-warm and we want to give them back to the page allocator ASAP.
663 *
664 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
665 * and __pagevec_lru_add_active() call release_pages() directly to avoid
666 * mutual recursion.
667 */
668void __pagevec_release(struct pagevec *pvec)
669{
670 lru_add_drain();
671 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
672 pagevec_reinit(pvec);
673}
Steve French7f285702005-11-01 10:22:55 -0800674EXPORT_SYMBOL(__pagevec_release);
675
Hugh Dickins12d27102012-01-12 17:19:52 -0800676#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800677/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700678void lru_add_page_tail(struct page *page, struct page *page_tail,
679 struct lruvec *lruvec)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800680{
Hugh Dickins75121022012-03-05 14:59:18 -0800681 int uninitialized_var(active);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800682 enum lru_list lru;
683 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800684
685 VM_BUG_ON(!PageHead(page));
686 VM_BUG_ON(PageCompound(page_tail));
687 VM_BUG_ON(PageLRU(page_tail));
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700688 VM_BUG_ON(NR_CPUS != 1 &&
689 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800690
691 SetPageLRU(page_tail);
692
693 if (page_evictable(page_tail, NULL)) {
694 if (PageActive(page)) {
695 SetPageActive(page_tail);
696 active = 1;
697 lru = LRU_ACTIVE_ANON;
698 } else {
699 active = 0;
700 lru = LRU_INACTIVE_ANON;
701 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800702 } else {
703 SetPageUnevictable(page_tail);
Hugh Dickins12d27102012-01-12 17:19:52 -0800704 lru = LRU_UNEVICTABLE;
705 }
706
707 if (likely(PageLRU(page)))
708 list_add_tail(&page_tail->lru, &page->lru);
709 else {
710 struct list_head *list_head;
711 /*
712 * Head page has not yet been counted, as an hpage,
713 * so we must account for each subpage individually.
714 *
715 * Use the standard add function to put page_tail on the list,
716 * but then correct its position so they all end up in order.
717 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700718 add_page_to_lru_list(page_tail, lruvec, lru);
Hugh Dickins12d27102012-01-12 17:19:52 -0800719 list_head = page_tail->lru.prev;
720 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800721 }
Hugh Dickins75121022012-03-05 14:59:18 -0800722
723 if (!PageUnevictable(page))
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700724 update_page_reclaim_stat(lruvec, file, active);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800725}
Hugh Dickins12d27102012-01-12 17:19:52 -0800726#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800727
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700728static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
729 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700730{
731 enum lru_list lru = (enum lru_list)arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700732 int file = is_file_lru(lru);
733 int active = is_active_lru(lru);
734
735 VM_BUG_ON(PageActive(page));
736 VM_BUG_ON(PageUnevictable(page));
737 VM_BUG_ON(PageLRU(page));
738
739 SetPageLRU(page);
740 if (active)
741 SetPageActive(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700742 add_page_to_lru_list(page, lruvec, lru);
743 update_page_reclaim_stat(lruvec, file, active);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700744}
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 * Add the passed pages to the LRU, then drop the caller's refcount
748 * on them. Reinitialises the caller's pagevec.
749 */
Hugh Dickins5095ae832012-01-12 17:19:58 -0800750void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700752 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Hugh Dickins5095ae832012-01-12 17:19:58 -0800754 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800756EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758/**
759 * pagevec_lookup - gang pagecache lookup
760 * @pvec: Where the resulting pages are placed
761 * @mapping: The address_space to search
762 * @start: The starting page index
763 * @nr_pages: The maximum number of pages
764 *
765 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
766 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
767 * reference against the pages in @pvec.
768 *
769 * The search returns a group of mapping-contiguous pages with ascending
770 * indexes. There may be holes in the indices due to not-present pages.
771 *
772 * pagevec_lookup() returns the number of pages which were found.
773 */
774unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
775 pgoff_t start, unsigned nr_pages)
776{
777 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
778 return pagevec_count(pvec);
779}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100780EXPORT_SYMBOL(pagevec_lookup);
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
783 pgoff_t *index, int tag, unsigned nr_pages)
784{
785 pvec->nr = find_get_pages_tag(mapping, index, tag,
786 nr_pages, pvec->pages);
787 return pagevec_count(pvec);
788}
Steve French7f285702005-11-01 10:22:55 -0800789EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791/*
792 * Perform any setup for the swap system
793 */
794void __init swap_setup(void)
795{
Jan Beulich44813742009-09-21 17:03:05 -0700796 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700798#ifdef CONFIG_SWAP
799 bdi_init(swapper_space.backing_dev_info);
800#endif
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 /* Use a smaller cluster for small-memory machines */
803 if (megs < 16)
804 page_cluster = 2;
805 else
806 page_cluster = 3;
807 /*
808 * Right now other parts of the system means that we
809 * _really_ don't want to cluster much more
810 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}