blob: 6fdd72ec15b01b4552480e02ad639de0a73d79de [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)) {
50 unsigned long flags;
51 struct zone *zone = page_zone(page);
52
53 spin_lock_irqsave(&zone->lru_lock, flags);
54 VM_BUG_ON(!PageLRU(page));
55 __ClearPageLRU(page);
Hugh Dickins1c1c53d2012-01-12 17:20:04 -080056 del_page_from_lru_list(zone, page, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070057 spin_unlock_irqrestore(&zone->lru_lock, flags);
58 }
Andrea Arcangeli91807062011-01-13 15:46:32 -080059}
60
61static void __put_single_page(struct page *page)
62{
63 __page_cache_release(page);
Li Hongfc916682010-03-05 13:41:54 -080064 free_hot_cold_page(page, 0);
Adrian Bunkb2213852006-09-25 23:31:02 -070065}
66
Andrea Arcangeli91807062011-01-13 15:46:32 -080067static void __put_compound_page(struct page *page)
68{
69 compound_page_dtor *dtor;
70
71 __page_cache_release(page);
72 dtor = get_compound_page_dtor(page);
73 (*dtor)(page);
74}
75
Nick Piggin8519fb32006-02-07 12:58:52 -080076static void put_compound_page(struct page *page)
77{
Andrea Arcangeli91807062011-01-13 15:46:32 -080078 if (unlikely(PageTail(page))) {
79 /* __split_huge_page_refcount can run under us */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070080 struct page *page_head = compound_trans_head(page);
81
82 if (likely(page != page_head &&
83 get_page_unless_zero(page_head))) {
Andrea Arcangeli91807062011-01-13 15:46:32 -080084 unsigned long flags;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -070085
86 /*
87 * THP can not break up slab pages so avoid taking
88 * compound_lock(). Slab performs non-atomic bit ops
89 * on page->flags for better performance. In particular
90 * slab_unlock() in slub used to be a hot path. It is
91 * still hot on arches that do not support
92 * this_cpu_cmpxchg_double().
93 */
94 if (PageSlab(page_head)) {
95 if (PageTail(page)) {
96 if (put_page_testzero(page_head))
97 VM_BUG_ON(1);
98
99 atomic_dec(&page->_mapcount);
100 goto skip_lock_tail;
101 } else
102 goto skip_lock;
103 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800104 /*
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700105 * page_head wasn't a dangling pointer but it
106 * may not be a head page anymore by the time
107 * we obtain the lock. That is ok as long as it
108 * can't be freed from under us.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800109 */
Andrea Arcangeli91807062011-01-13 15:46:32 -0800110 flags = compound_lock_irqsave(page_head);
111 if (unlikely(!PageTail(page))) {
112 /* __split_huge_page_refcount run before us */
113 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700114skip_lock:
Andrea Arcangeli91807062011-01-13 15:46:32 -0800115 if (put_page_testzero(page_head))
116 __put_single_page(page_head);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700117out_put_single:
Andrea Arcangeli91807062011-01-13 15:46:32 -0800118 if (put_page_testzero(page))
119 __put_single_page(page);
120 return;
121 }
122 VM_BUG_ON(page_head != page->first_page);
123 /*
124 * We can release the refcount taken by
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700125 * get_page_unless_zero() now that
126 * __split_huge_page_refcount() is blocked on
127 * the compound_lock.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800128 */
129 if (put_page_testzero(page_head))
130 VM_BUG_ON(1);
131 /* __split_huge_page_refcount will wait now */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700132 VM_BUG_ON(page_mapcount(page) <= 0);
133 atomic_dec(&page->_mapcount);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800134 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700135 VM_BUG_ON(atomic_read(&page->_count) != 0);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800136 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700137
138skip_lock_tail:
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800139 if (put_page_testzero(page_head)) {
140 if (PageHead(page_head))
141 __put_compound_page(page_head);
142 else
143 __put_single_page(page_head);
144 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800145 } else {
146 /* page_head is a dangling pointer */
147 VM_BUG_ON(PageTail(page));
148 goto out_put_single;
149 }
150 } else if (put_page_testzero(page)) {
151 if (PageHead(page))
152 __put_compound_page(page);
153 else
154 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800155 }
156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158void put_page(struct page *page)
159{
Nick Piggin8519fb32006-02-07 12:58:52 -0800160 if (unlikely(PageCompound(page)))
161 put_compound_page(page);
162 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800163 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
165EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700167/*
168 * This function is exported but must not be called by anything other
169 * than get_page(). It implements the slow path of get_page().
170 */
171bool __get_page_tail(struct page *page)
172{
173 /*
174 * This takes care of get_page() if run on a tail page
175 * returned by one of the get_user_pages/follow_page variants.
176 * get_user_pages/follow_page itself doesn't need the compound
177 * lock because it runs __get_page_tail_foll() under the
178 * proper PT lock that already serializes against
179 * split_huge_page().
180 */
181 unsigned long flags;
182 bool got = false;
183 struct page *page_head = compound_trans_head(page);
184
185 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700186
187 /* Ref to put_compound_page() comment. */
188 if (PageSlab(page_head)) {
189 if (likely(PageTail(page))) {
190 __get_page_tail_foll(page, false);
191 return true;
192 } else {
193 put_page(page_head);
194 return false;
195 }
196 }
197
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700198 /*
199 * page_head wasn't a dangling pointer but it
200 * may not be a head page anymore by the time
201 * we obtain the lock. That is ok as long as it
202 * can't be freed from under us.
203 */
204 flags = compound_lock_irqsave(page_head);
205 /* here __split_huge_page_refcount won't run anymore */
206 if (likely(PageTail(page))) {
207 __get_page_tail_foll(page, false);
208 got = true;
209 }
210 compound_unlock_irqrestore(page_head, flags);
211 if (unlikely(!got))
212 put_page(page_head);
213 }
214 return got;
215}
216EXPORT_SYMBOL(__get_page_tail);
217
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700218/**
Randy Dunlap76824862008-03-19 17:00:40 -0700219 * put_pages_list() - release a list of pages
220 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700221 *
222 * Release a list of pages which are strung together on page.lru. Currently
223 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700224 */
225void put_pages_list(struct list_head *pages)
226{
227 while (!list_empty(pages)) {
228 struct page *victim;
229
230 victim = list_entry(pages->prev, struct page, lru);
231 list_del(&victim->lru);
232 page_cache_release(victim);
233 }
234}
235EXPORT_SYMBOL(put_pages_list);
236
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700237static void pagevec_lru_move_fn(struct pagevec *pvec,
238 void (*move_fn)(struct page *page, void *arg),
239 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700240{
241 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700242 struct zone *zone = NULL;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700243 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700244
245 for (i = 0; i < pagevec_count(pvec); i++) {
246 struct page *page = pvec->pages[i];
247 struct zone *pagezone = page_zone(page);
248
249 if (pagezone != zone) {
250 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700251 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700252 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700253 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700254 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700255
256 (*move_fn)(page, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700257 }
258 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700259 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800260 release_pages(pvec->pages, pvec->nr, pvec->cold);
261 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800262}
263
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700264static void pagevec_move_tail_fn(struct page *page, void *arg)
265{
266 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700267
268 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
269 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800270 struct lruvec *lruvec;
271
272 lruvec = mem_cgroup_lru_move_lists(page_zone(page),
273 page, lru, lru);
274 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700275 (*pgmoved)++;
276 }
277}
278
279/*
280 * pagevec_move_tail() must be called with IRQ disabled.
281 * Otherwise this may cause nasty races.
282 */
283static void pagevec_move_tail(struct pagevec *pvec)
284{
285 int pgmoved = 0;
286
287 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
288 __count_vm_events(PGROTATED, pgmoved);
289}
290
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700291/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 * Writeback is about to end against a page which has been marked for immediate
293 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700294 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700296void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700298 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700299 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700300 struct pagevec *pvec;
301 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700303 page_cache_get(page);
304 local_irq_save(flags);
305 pvec = &__get_cpu_var(lru_rotate_pvecs);
306 if (!pagevec_add(pvec, page))
307 pagevec_move_tail(pvec);
308 local_irq_restore(flags);
309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800312static void update_page_reclaim_stat(struct zone *zone, struct page *page,
313 int file, int rotated)
314{
315 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
316 struct zone_reclaim_stat *memcg_reclaim_stat;
317
318 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
319
320 reclaim_stat->recent_scanned[file]++;
321 if (rotated)
322 reclaim_stat->recent_rotated[file]++;
323
324 if (!memcg_reclaim_stat)
325 return;
326
327 memcg_reclaim_stat->recent_scanned[file]++;
328 if (rotated)
329 memcg_reclaim_stat->recent_rotated[file]++;
330}
331
Shaohua Lieb709b02011-05-24 17:12:55 -0700332static void __activate_page(struct page *page, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
334 struct zone *zone = page_zone(page);
335
Linus Torvalds7a608572011-01-17 14:42:19 -0800336 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
337 int file = page_is_file_cache(page);
338 int lru = page_lru_base_type(page);
339 del_page_from_lru_list(zone, page, lru);
340
341 SetPageActive(page);
342 lru += LRU_ACTIVE;
343 add_page_to_lru_list(zone, page, lru);
344 __count_vm_event(PGACTIVATE);
345
346 update_page_reclaim_stat(zone, page, file, 1);
347 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700348}
349
350#ifdef CONFIG_SMP
351static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
352
353static void activate_page_drain(int cpu)
354{
355 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
356
357 if (pagevec_count(pvec))
358 pagevec_lru_move_fn(pvec, __activate_page, NULL);
359}
360
361void activate_page(struct page *page)
362{
363 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
364 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
365
366 page_cache_get(page);
367 if (!pagevec_add(pvec, page))
368 pagevec_lru_move_fn(pvec, __activate_page, NULL);
369 put_cpu_var(activate_page_pvecs);
370 }
371}
372
373#else
374static inline void activate_page_drain(int cpu)
375{
376}
377
378void activate_page(struct page *page)
379{
380 struct zone *zone = page_zone(page);
381
382 spin_lock_irq(&zone->lru_lock);
383 __activate_page(page, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 spin_unlock_irq(&zone->lru_lock);
385}
Shaohua Lieb709b02011-05-24 17:12:55 -0700386#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
388/*
389 * Mark a page as having seen activity.
390 *
391 * inactive,unreferenced -> inactive,referenced
392 * inactive,referenced -> active,unreferenced
393 * active,unreferenced -> active,referenced
394 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800395void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700397 if (!PageActive(page) && !PageUnevictable(page) &&
398 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 activate_page(page);
400 ClearPageReferenced(page);
401 } else if (!PageReferenced(page)) {
402 SetPageReferenced(page);
403 }
404}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405EXPORT_SYMBOL(mark_page_accessed);
406
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700407void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700409 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411 page_cache_get(page);
412 if (!pagevec_add(pvec, page))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800413 __pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 put_cpu_var(lru_add_pvecs);
415}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200416EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700418/**
419 * lru_cache_add_lru - add a page to a page list
420 * @page: the page to be added to the LRU.
421 * @lru: the LRU list to which the page is added.
422 */
423void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700425 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700426 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700427 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700428 } else if (PageUnevictable(page)) {
429 VM_BUG_ON(PageActive(page));
430 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700433 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700434 __lru_cache_add(page, lru);
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{
449 struct zone *zone = page_zone(page);
450
451 spin_lock_irq(&zone->lru_lock);
452 SetPageUnevictable(page);
453 SetPageLRU(page);
454 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
455 spin_unlock_irq(&zone->lru_lock);
456}
457
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700458/*
Minchan Kim31560182011-03-22 16:32:52 -0700459 * If the page can not be invalidated, it is moved to the
460 * inactive list to speed up its reclaim. It is moved to the
461 * head of the list, rather than the tail, to give the flusher
462 * threads some time to write it out, as this is much more
463 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700464 *
465 * If the page isn't page_mapped and dirty/writeback, the page
466 * could reclaim asap using PG_reclaim.
467 *
468 * 1. active, mapped page -> none
469 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
470 * 3. inactive, mapped page -> none
471 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
472 * 5. inactive, clean -> inactive, tail
473 * 6. Others -> none
474 *
475 * In 4, why it moves inactive's head, the VM expects the page would
476 * be write it out by flusher threads as this is much more effective
477 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700478 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700479static void lru_deactivate_fn(struct page *page, void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700480{
481 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700482 bool active;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700483 struct zone *zone = page_zone(page);
Minchan Kim31560182011-03-22 16:32:52 -0700484
Minchan Kim278df9f2011-03-22 16:32:54 -0700485 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700486 return;
487
Minchan Kimbad49d92011-05-11 15:13:30 -0700488 if (PageUnevictable(page))
489 return;
490
Minchan Kim31560182011-03-22 16:32:52 -0700491 /* Some processes are using the page */
492 if (page_mapped(page))
493 return;
494
Minchan Kim278df9f2011-03-22 16:32:54 -0700495 active = PageActive(page);
496
Minchan Kim31560182011-03-22 16:32:52 -0700497 file = page_is_file_cache(page);
498 lru = page_lru_base_type(page);
Minchan Kim278df9f2011-03-22 16:32:54 -0700499 del_page_from_lru_list(zone, page, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700500 ClearPageActive(page);
501 ClearPageReferenced(page);
502 add_page_to_lru_list(zone, page, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700503
Minchan Kim278df9f2011-03-22 16:32:54 -0700504 if (PageWriteback(page) || PageDirty(page)) {
505 /*
506 * PG_reclaim could be raced with end_page_writeback
507 * It can make readahead confusing. But race window
508 * is _really_ small and it's non-critical problem.
509 */
510 SetPageReclaim(page);
511 } else {
Johannes Weiner925b7672012-01-12 17:18:15 -0800512 struct lruvec *lruvec;
Minchan Kim278df9f2011-03-22 16:32:54 -0700513 /*
514 * The page's writeback ends up during pagevec
515 * We moves tha page into tail of inactive.
516 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800517 lruvec = mem_cgroup_lru_move_lists(zone, page, lru, lru);
518 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700519 __count_vm_event(PGROTATED);
520 }
521
522 if (active)
523 __count_vm_event(PGDEACTIVATE);
Minchan Kim31560182011-03-22 16:32:52 -0700524 update_page_reclaim_stat(zone, page, file, 0);
525}
526
Minchan Kim31560182011-03-22 16:32:52 -0700527/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700528 * Drain pages out of the cpu's pagevecs.
529 * Either "cpu" is the current CPU, and preemption has already been
530 * disabled; or "cpu" is being hot-unplugged, and is already dead.
531 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700532void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700534 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700535 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700536 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700538 for_each_lru(lru) {
539 pvec = &pvecs[lru - LRU_BASE];
540 if (pagevec_count(pvec))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800541 __pagevec_lru_add(pvec, lru);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700542 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700543
544 pvec = &per_cpu(lru_rotate_pvecs, cpu);
545 if (pagevec_count(pvec)) {
546 unsigned long flags;
547
548 /* No harm done if a racing interrupt already did this */
549 local_irq_save(flags);
550 pagevec_move_tail(pvec);
551 local_irq_restore(flags);
552 }
Minchan Kim31560182011-03-22 16:32:52 -0700553
554 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
555 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700556 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700557
558 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700559}
560
561/**
562 * deactivate_page - forcefully deactivate a page
563 * @page: page to deactivate
564 *
565 * This function hints the VM that @page is a good reclaim candidate,
566 * for example if its invalidation fails due to the page being dirty
567 * or under writeback.
568 */
569void deactivate_page(struct page *page)
570{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700571 /*
572 * In a workload with many unevictable page such as mprotect, unevictable
573 * page deactivation for accelerating reclaim is pointless.
574 */
575 if (PageUnevictable(page))
576 return;
577
Minchan Kim31560182011-03-22 16:32:52 -0700578 if (likely(get_page_unless_zero(page))) {
579 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
580
581 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700582 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700583 put_cpu_var(lru_deactivate_pvecs);
584 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800585}
586
587void lru_add_drain(void)
588{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700589 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800590 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591}
592
David Howellsc4028952006-11-22 14:57:56 +0000593static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800594{
595 lru_add_drain();
596}
597
598/*
599 * Returns 0 for success
600 */
601int lru_add_drain_all(void)
602{
David Howellsc4028952006-11-22 14:57:56 +0000603 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800604}
605
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 * Batched page_cache_release(). Decrement the reference count on all the
608 * passed pages. If it fell to zero then remove the page from the LRU and
609 * free it.
610 *
611 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
612 * for the remainder of the operation.
613 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700614 * The locking in this function is against shrink_inactive_list(): we recheck
615 * the page count inside the lock to see whether shrink_inactive_list()
616 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
617 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 */
619void release_pages(struct page **pages, int nr, int cold)
620{
621 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800622 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700624 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 for (i = 0; i < nr; i++) {
627 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Nick Piggin8519fb32006-02-07 12:58:52 -0800629 if (unlikely(PageCompound(page))) {
630 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700631 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800632 zone = NULL;
633 }
634 put_compound_page(page);
635 continue;
636 }
637
Nick Pigginb5810032005-10-29 18:16:12 -0700638 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 continue;
640
Nick Piggin46453a62006-03-22 00:07:58 -0800641 if (PageLRU(page)) {
642 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700643
Nick Piggin46453a62006-03-22 00:07:58 -0800644 if (pagezone != zone) {
645 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700646 spin_unlock_irqrestore(&zone->lru_lock,
647 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800648 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700649 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 }
Nick Piggin725d7042006-09-25 23:30:55 -0700651 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800652 __ClearPageLRU(page);
Hugh Dickins1c1c53d2012-01-12 17:20:04 -0800653 del_page_from_lru_list(zone, page, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 }
Nick Piggin46453a62006-03-22 00:07:58 -0800655
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800656 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 }
658 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700659 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800661 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700663EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665/*
666 * The pages which we're about to release may be in the deferred lru-addition
667 * queues. That would prevent them from really being freed right now. That's
668 * OK from a correctness point of view but is inefficient - those pages may be
669 * cache-warm and we want to give them back to the page allocator ASAP.
670 *
671 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
672 * and __pagevec_lru_add_active() call release_pages() directly to avoid
673 * mutual recursion.
674 */
675void __pagevec_release(struct pagevec *pvec)
676{
677 lru_add_drain();
678 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
679 pagevec_reinit(pvec);
680}
Steve French7f285702005-11-01 10:22:55 -0800681EXPORT_SYMBOL(__pagevec_release);
682
Hugh Dickins12d27102012-01-12 17:19:52 -0800683#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800684/* used by __split_huge_page_refcount() */
685void lru_add_page_tail(struct zone* zone,
686 struct page *page, struct page *page_tail)
687{
Hugh Dickins75121022012-03-05 14:59:18 -0800688 int uninitialized_var(active);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800689 enum lru_list lru;
690 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800691
692 VM_BUG_ON(!PageHead(page));
693 VM_BUG_ON(PageCompound(page_tail));
694 VM_BUG_ON(PageLRU(page_tail));
Hugh Dickinsb9980cd2012-02-08 17:13:40 -0800695 VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&zone->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800696
697 SetPageLRU(page_tail);
698
699 if (page_evictable(page_tail, NULL)) {
700 if (PageActive(page)) {
701 SetPageActive(page_tail);
702 active = 1;
703 lru = LRU_ACTIVE_ANON;
704 } else {
705 active = 0;
706 lru = LRU_INACTIVE_ANON;
707 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800708 } else {
709 SetPageUnevictable(page_tail);
Hugh Dickins12d27102012-01-12 17:19:52 -0800710 lru = LRU_UNEVICTABLE;
711 }
712
713 if (likely(PageLRU(page)))
714 list_add_tail(&page_tail->lru, &page->lru);
715 else {
716 struct list_head *list_head;
717 /*
718 * Head page has not yet been counted, as an hpage,
719 * so we must account for each subpage individually.
720 *
721 * Use the standard add function to put page_tail on the list,
722 * but then correct its position so they all end up in order.
723 */
724 add_page_to_lru_list(zone, page_tail, lru);
725 list_head = page_tail->lru.prev;
726 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800727 }
Hugh Dickins75121022012-03-05 14:59:18 -0800728
729 if (!PageUnevictable(page))
730 update_page_reclaim_stat(zone, page_tail, file, active);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800731}
Hugh Dickins12d27102012-01-12 17:19:52 -0800732#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800733
Hugh Dickins5095ae832012-01-12 17:19:58 -0800734static void __pagevec_lru_add_fn(struct page *page, void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700735{
736 enum lru_list lru = (enum lru_list)arg;
737 struct zone *zone = page_zone(page);
738 int file = is_file_lru(lru);
739 int active = is_active_lru(lru);
740
741 VM_BUG_ON(PageActive(page));
742 VM_BUG_ON(PageUnevictable(page));
743 VM_BUG_ON(PageLRU(page));
744
745 SetPageLRU(page);
746 if (active)
747 SetPageActive(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700748 add_page_to_lru_list(zone, page, lru);
Hugh Dickins75121022012-03-05 14:59:18 -0800749 update_page_reclaim_stat(zone, page, file, active);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700750}
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 * Add the passed pages to the LRU, then drop the caller's refcount
754 * on them. Reinitialises the caller's pagevec.
755 */
Hugh Dickins5095ae832012-01-12 17:19:58 -0800756void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700758 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Hugh Dickins5095ae832012-01-12 17:19:58 -0800760 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800762EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764/**
765 * pagevec_lookup - gang pagecache lookup
766 * @pvec: Where the resulting pages are placed
767 * @mapping: The address_space to search
768 * @start: The starting page index
769 * @nr_pages: The maximum number of pages
770 *
771 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
772 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
773 * reference against the pages in @pvec.
774 *
775 * The search returns a group of mapping-contiguous pages with ascending
776 * indexes. There may be holes in the indices due to not-present pages.
777 *
778 * pagevec_lookup() returns the number of pages which were found.
779 */
780unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
781 pgoff_t start, unsigned nr_pages)
782{
783 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
784 return pagevec_count(pvec);
785}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100786EXPORT_SYMBOL(pagevec_lookup);
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
789 pgoff_t *index, int tag, unsigned nr_pages)
790{
791 pvec->nr = find_get_pages_tag(mapping, index, tag,
792 nr_pages, pvec->pages);
793 return pagevec_count(pvec);
794}
Steve French7f285702005-11-01 10:22:55 -0800795EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797/*
798 * Perform any setup for the swap system
799 */
800void __init swap_setup(void)
801{
Jan Beulich44813742009-09-21 17:03:05 -0700802 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700804#ifdef CONFIG_SWAP
805 bdi_init(swapper_space.backing_dev_info);
806#endif
807
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 /* Use a smaller cluster for small-memory machines */
809 if (megs < 16)
810 page_cluster = 2;
811 else
812 page_cluster = 3;
813 /*
814 * Right now other parts of the system means that we
815 * _really_ don't want to cluster much more
816 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}