blob: fff1ff7fb9ada36fca1be10d33dec593486f50a6 [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;
85 /*
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070086 * page_head wasn't a dangling pointer but it
87 * may not be a head page anymore by the time
88 * we obtain the lock. That is ok as long as it
89 * can't be freed from under us.
Andrea Arcangeli91807062011-01-13 15:46:32 -080090 */
Andrea Arcangeli91807062011-01-13 15:46:32 -080091 flags = compound_lock_irqsave(page_head);
92 if (unlikely(!PageTail(page))) {
93 /* __split_huge_page_refcount run before us */
94 compound_unlock_irqrestore(page_head, flags);
95 VM_BUG_ON(PageHead(page_head));
Andrea Arcangeli91807062011-01-13 15:46:32 -080096 if (put_page_testzero(page_head))
97 __put_single_page(page_head);
98 out_put_single:
99 if (put_page_testzero(page))
100 __put_single_page(page);
101 return;
102 }
103 VM_BUG_ON(page_head != page->first_page);
104 /*
105 * We can release the refcount taken by
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700106 * get_page_unless_zero() now that
107 * __split_huge_page_refcount() is blocked on
108 * the compound_lock.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800109 */
110 if (put_page_testzero(page_head))
111 VM_BUG_ON(1);
112 /* __split_huge_page_refcount will wait now */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700113 VM_BUG_ON(page_mapcount(page) <= 0);
114 atomic_dec(&page->_mapcount);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800115 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700116 VM_BUG_ON(atomic_read(&page->_count) != 0);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800117 compound_unlock_irqrestore(page_head, flags);
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800118 if (put_page_testzero(page_head)) {
119 if (PageHead(page_head))
120 __put_compound_page(page_head);
121 else
122 __put_single_page(page_head);
123 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800124 } else {
125 /* page_head is a dangling pointer */
126 VM_BUG_ON(PageTail(page));
127 goto out_put_single;
128 }
129 } else if (put_page_testzero(page)) {
130 if (PageHead(page))
131 __put_compound_page(page);
132 else
133 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800134 }
135}
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137void put_page(struct page *page)
138{
Nick Piggin8519fb32006-02-07 12:58:52 -0800139 if (unlikely(PageCompound(page)))
140 put_compound_page(page);
141 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800142 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700146/*
147 * This function is exported but must not be called by anything other
148 * than get_page(). It implements the slow path of get_page().
149 */
150bool __get_page_tail(struct page *page)
151{
152 /*
153 * This takes care of get_page() if run on a tail page
154 * returned by one of the get_user_pages/follow_page variants.
155 * get_user_pages/follow_page itself doesn't need the compound
156 * lock because it runs __get_page_tail_foll() under the
157 * proper PT lock that already serializes against
158 * split_huge_page().
159 */
160 unsigned long flags;
161 bool got = false;
162 struct page *page_head = compound_trans_head(page);
163
164 if (likely(page != page_head && get_page_unless_zero(page_head))) {
165 /*
166 * page_head wasn't a dangling pointer but it
167 * may not be a head page anymore by the time
168 * we obtain the lock. That is ok as long as it
169 * can't be freed from under us.
170 */
171 flags = compound_lock_irqsave(page_head);
172 /* here __split_huge_page_refcount won't run anymore */
173 if (likely(PageTail(page))) {
174 __get_page_tail_foll(page, false);
175 got = true;
176 }
177 compound_unlock_irqrestore(page_head, flags);
178 if (unlikely(!got))
179 put_page(page_head);
180 }
181 return got;
182}
183EXPORT_SYMBOL(__get_page_tail);
184
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700185/**
Randy Dunlap76824862008-03-19 17:00:40 -0700186 * put_pages_list() - release a list of pages
187 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700188 *
189 * Release a list of pages which are strung together on page.lru. Currently
190 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700191 */
192void put_pages_list(struct list_head *pages)
193{
194 while (!list_empty(pages)) {
195 struct page *victim;
196
197 victim = list_entry(pages->prev, struct page, lru);
198 list_del(&victim->lru);
199 page_cache_release(victim);
200 }
201}
202EXPORT_SYMBOL(put_pages_list);
203
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700204static void pagevec_lru_move_fn(struct pagevec *pvec,
205 void (*move_fn)(struct page *page, void *arg),
206 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700207{
208 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700209 struct zone *zone = NULL;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700210 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700211
212 for (i = 0; i < pagevec_count(pvec); i++) {
213 struct page *page = pvec->pages[i];
214 struct zone *pagezone = page_zone(page);
215
216 if (pagezone != zone) {
217 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700218 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700219 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700220 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700221 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700222
223 (*move_fn)(page, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700224 }
225 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700226 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800227 release_pages(pvec->pages, pvec->nr, pvec->cold);
228 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800229}
230
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700231static void pagevec_move_tail_fn(struct page *page, void *arg)
232{
233 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700234
235 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
236 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800237 struct lruvec *lruvec;
238
239 lruvec = mem_cgroup_lru_move_lists(page_zone(page),
240 page, lru, lru);
241 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700242 (*pgmoved)++;
243 }
244}
245
246/*
247 * pagevec_move_tail() must be called with IRQ disabled.
248 * Otherwise this may cause nasty races.
249 */
250static void pagevec_move_tail(struct pagevec *pvec)
251{
252 int pgmoved = 0;
253
254 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
255 __count_vm_events(PGROTATED, pgmoved);
256}
257
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700258/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * Writeback is about to end against a page which has been marked for immediate
260 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700261 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700263void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700265 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700266 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700267 struct pagevec *pvec;
268 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700270 page_cache_get(page);
271 local_irq_save(flags);
272 pvec = &__get_cpu_var(lru_rotate_pvecs);
273 if (!pagevec_add(pvec, page))
274 pagevec_move_tail(pvec);
275 local_irq_restore(flags);
276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800279static void update_page_reclaim_stat(struct zone *zone, struct page *page,
280 int file, int rotated)
281{
282 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
283 struct zone_reclaim_stat *memcg_reclaim_stat;
284
285 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
286
287 reclaim_stat->recent_scanned[file]++;
288 if (rotated)
289 reclaim_stat->recent_rotated[file]++;
290
291 if (!memcg_reclaim_stat)
292 return;
293
294 memcg_reclaim_stat->recent_scanned[file]++;
295 if (rotated)
296 memcg_reclaim_stat->recent_rotated[file]++;
297}
298
Shaohua Lieb709b02011-05-24 17:12:55 -0700299static void __activate_page(struct page *page, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
301 struct zone *zone = page_zone(page);
302
Linus Torvalds7a608572011-01-17 14:42:19 -0800303 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
304 int file = page_is_file_cache(page);
305 int lru = page_lru_base_type(page);
306 del_page_from_lru_list(zone, page, lru);
307
308 SetPageActive(page);
309 lru += LRU_ACTIVE;
310 add_page_to_lru_list(zone, page, lru);
311 __count_vm_event(PGACTIVATE);
312
313 update_page_reclaim_stat(zone, page, file, 1);
314 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700315}
316
317#ifdef CONFIG_SMP
318static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
319
320static void activate_page_drain(int cpu)
321{
322 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
323
324 if (pagevec_count(pvec))
325 pagevec_lru_move_fn(pvec, __activate_page, NULL);
326}
327
328void activate_page(struct page *page)
329{
330 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
331 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
332
333 page_cache_get(page);
334 if (!pagevec_add(pvec, page))
335 pagevec_lru_move_fn(pvec, __activate_page, NULL);
336 put_cpu_var(activate_page_pvecs);
337 }
338}
339
340#else
341static inline void activate_page_drain(int cpu)
342{
343}
344
345void activate_page(struct page *page)
346{
347 struct zone *zone = page_zone(page);
348
349 spin_lock_irq(&zone->lru_lock);
350 __activate_page(page, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 spin_unlock_irq(&zone->lru_lock);
352}
Shaohua Lieb709b02011-05-24 17:12:55 -0700353#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355/*
356 * Mark a page as having seen activity.
357 *
358 * inactive,unreferenced -> inactive,referenced
359 * inactive,referenced -> active,unreferenced
360 * active,unreferenced -> active,referenced
361 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800362void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700364 if (!PageActive(page) && !PageUnevictable(page) &&
365 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 activate_page(page);
367 ClearPageReferenced(page);
368 } else if (!PageReferenced(page)) {
369 SetPageReferenced(page);
370 }
371}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372EXPORT_SYMBOL(mark_page_accessed);
373
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700374void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700376 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
378 page_cache_get(page);
379 if (!pagevec_add(pvec, page))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800380 __pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 put_cpu_var(lru_add_pvecs);
382}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200383EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700385/**
386 * lru_cache_add_lru - add a page to a page list
387 * @page: the page to be added to the LRU.
388 * @lru: the LRU list to which the page is added.
389 */
390void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700392 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700393 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700394 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700395 } else if (PageUnevictable(page)) {
396 VM_BUG_ON(PageActive(page));
397 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700400 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700401 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700404/**
405 * add_page_to_unevictable_list - add a page to the unevictable list
406 * @page: the page to be added to the unevictable list
407 *
408 * Add page directly to its zone's unevictable list. To avoid races with
409 * tasks that might be making the page evictable, through eg. munlock,
410 * munmap or exit, while it's not on the lru, we want to add the page
411 * while it's locked or otherwise "invisible" to other tasks. This is
412 * difficult to do when using the pagevec cache, so bypass that.
413 */
414void add_page_to_unevictable_list(struct page *page)
415{
416 struct zone *zone = page_zone(page);
417
418 spin_lock_irq(&zone->lru_lock);
419 SetPageUnevictable(page);
420 SetPageLRU(page);
421 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
422 spin_unlock_irq(&zone->lru_lock);
423}
424
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700425/*
Minchan Kim31560182011-03-22 16:32:52 -0700426 * If the page can not be invalidated, it is moved to the
427 * inactive list to speed up its reclaim. It is moved to the
428 * head of the list, rather than the tail, to give the flusher
429 * threads some time to write it out, as this is much more
430 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700431 *
432 * If the page isn't page_mapped and dirty/writeback, the page
433 * could reclaim asap using PG_reclaim.
434 *
435 * 1. active, mapped page -> none
436 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
437 * 3. inactive, mapped page -> none
438 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
439 * 5. inactive, clean -> inactive, tail
440 * 6. Others -> none
441 *
442 * In 4, why it moves inactive's head, the VM expects the page would
443 * be write it out by flusher threads as this is much more effective
444 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700445 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700446static void lru_deactivate_fn(struct page *page, void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700447{
448 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700449 bool active;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700450 struct zone *zone = page_zone(page);
Minchan Kim31560182011-03-22 16:32:52 -0700451
Minchan Kim278df9f2011-03-22 16:32:54 -0700452 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700453 return;
454
Minchan Kimbad49d92011-05-11 15:13:30 -0700455 if (PageUnevictable(page))
456 return;
457
Minchan Kim31560182011-03-22 16:32:52 -0700458 /* Some processes are using the page */
459 if (page_mapped(page))
460 return;
461
Minchan Kim278df9f2011-03-22 16:32:54 -0700462 active = PageActive(page);
463
Minchan Kim31560182011-03-22 16:32:52 -0700464 file = page_is_file_cache(page);
465 lru = page_lru_base_type(page);
Minchan Kim278df9f2011-03-22 16:32:54 -0700466 del_page_from_lru_list(zone, page, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700467 ClearPageActive(page);
468 ClearPageReferenced(page);
469 add_page_to_lru_list(zone, page, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700470
Minchan Kim278df9f2011-03-22 16:32:54 -0700471 if (PageWriteback(page) || PageDirty(page)) {
472 /*
473 * PG_reclaim could be raced with end_page_writeback
474 * It can make readahead confusing. But race window
475 * is _really_ small and it's non-critical problem.
476 */
477 SetPageReclaim(page);
478 } else {
Johannes Weiner925b7672012-01-12 17:18:15 -0800479 struct lruvec *lruvec;
Minchan Kim278df9f2011-03-22 16:32:54 -0700480 /*
481 * The page's writeback ends up during pagevec
482 * We moves tha page into tail of inactive.
483 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800484 lruvec = mem_cgroup_lru_move_lists(zone, page, lru, lru);
485 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700486 __count_vm_event(PGROTATED);
487 }
488
489 if (active)
490 __count_vm_event(PGDEACTIVATE);
Minchan Kim31560182011-03-22 16:32:52 -0700491 update_page_reclaim_stat(zone, page, file, 0);
492}
493
Minchan Kim31560182011-03-22 16:32:52 -0700494/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700495 * Drain pages out of the cpu's pagevecs.
496 * Either "cpu" is the current CPU, and preemption has already been
497 * disabled; or "cpu" is being hot-unplugged, and is already dead.
498 */
499static void drain_cpu_pagevecs(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700501 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700502 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700503 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700505 for_each_lru(lru) {
506 pvec = &pvecs[lru - LRU_BASE];
507 if (pagevec_count(pvec))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800508 __pagevec_lru_add(pvec, lru);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700509 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700510
511 pvec = &per_cpu(lru_rotate_pvecs, cpu);
512 if (pagevec_count(pvec)) {
513 unsigned long flags;
514
515 /* No harm done if a racing interrupt already did this */
516 local_irq_save(flags);
517 pagevec_move_tail(pvec);
518 local_irq_restore(flags);
519 }
Minchan Kim31560182011-03-22 16:32:52 -0700520
521 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
522 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700523 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700524
525 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700526}
527
528/**
529 * deactivate_page - forcefully deactivate a page
530 * @page: page to deactivate
531 *
532 * This function hints the VM that @page is a good reclaim candidate,
533 * for example if its invalidation fails due to the page being dirty
534 * or under writeback.
535 */
536void deactivate_page(struct page *page)
537{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700538 /*
539 * In a workload with many unevictable page such as mprotect, unevictable
540 * page deactivation for accelerating reclaim is pointless.
541 */
542 if (PageUnevictable(page))
543 return;
544
Minchan Kim31560182011-03-22 16:32:52 -0700545 if (likely(get_page_unless_zero(page))) {
546 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
547
548 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700549 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700550 put_cpu_var(lru_deactivate_pvecs);
551 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800552}
553
554void lru_add_drain(void)
555{
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700556 drain_cpu_pagevecs(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800557 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558}
559
David Howellsc4028952006-11-22 14:57:56 +0000560static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800561{
562 lru_add_drain();
563}
564
565/*
566 * Returns 0 for success
567 */
568int lru_add_drain_all(void)
569{
David Howellsc4028952006-11-22 14:57:56 +0000570 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800571}
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 * Batched page_cache_release(). Decrement the reference count on all the
575 * passed pages. If it fell to zero then remove the page from the LRU and
576 * free it.
577 *
578 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
579 * for the remainder of the operation.
580 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700581 * The locking in this function is against shrink_inactive_list(): we recheck
582 * the page count inside the lock to see whether shrink_inactive_list()
583 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
584 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 */
586void release_pages(struct page **pages, int nr, int cold)
587{
588 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800589 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700591 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 for (i = 0; i < nr; i++) {
594 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Nick Piggin8519fb32006-02-07 12:58:52 -0800596 if (unlikely(PageCompound(page))) {
597 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700598 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800599 zone = NULL;
600 }
601 put_compound_page(page);
602 continue;
603 }
604
Nick Pigginb5810032005-10-29 18:16:12 -0700605 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 continue;
607
Nick Piggin46453a62006-03-22 00:07:58 -0800608 if (PageLRU(page)) {
609 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700610
Nick Piggin46453a62006-03-22 00:07:58 -0800611 if (pagezone != zone) {
612 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700613 spin_unlock_irqrestore(&zone->lru_lock,
614 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800615 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700616 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
Nick Piggin725d7042006-09-25 23:30:55 -0700618 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800619 __ClearPageLRU(page);
Hugh Dickins1c1c53d2012-01-12 17:20:04 -0800620 del_page_from_lru_list(zone, page, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 }
Nick Piggin46453a62006-03-22 00:07:58 -0800622
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800623 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
625 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700626 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800628 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700630EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632/*
633 * The pages which we're about to release may be in the deferred lru-addition
634 * queues. That would prevent them from really being freed right now. That's
635 * OK from a correctness point of view but is inefficient - those pages may be
636 * cache-warm and we want to give them back to the page allocator ASAP.
637 *
638 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
639 * and __pagevec_lru_add_active() call release_pages() directly to avoid
640 * mutual recursion.
641 */
642void __pagevec_release(struct pagevec *pvec)
643{
644 lru_add_drain();
645 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
646 pagevec_reinit(pvec);
647}
Steve French7f285702005-11-01 10:22:55 -0800648EXPORT_SYMBOL(__pagevec_release);
649
Hugh Dickins12d27102012-01-12 17:19:52 -0800650#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800651/* used by __split_huge_page_refcount() */
652void lru_add_page_tail(struct zone* zone,
653 struct page *page, struct page *page_tail)
654{
655 int active;
656 enum lru_list lru;
657 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800658
659 VM_BUG_ON(!PageHead(page));
660 VM_BUG_ON(PageCompound(page_tail));
661 VM_BUG_ON(PageLRU(page_tail));
Hugh Dickinsb9980cd2012-02-08 17:13:40 -0800662 VM_BUG_ON(NR_CPUS != 1 && !spin_is_locked(&zone->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800663
664 SetPageLRU(page_tail);
665
666 if (page_evictable(page_tail, NULL)) {
667 if (PageActive(page)) {
668 SetPageActive(page_tail);
669 active = 1;
670 lru = LRU_ACTIVE_ANON;
671 } else {
672 active = 0;
673 lru = LRU_INACTIVE_ANON;
674 }
675 update_page_reclaim_stat(zone, page_tail, file, active);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800676 } else {
677 SetPageUnevictable(page_tail);
Hugh Dickins12d27102012-01-12 17:19:52 -0800678 lru = LRU_UNEVICTABLE;
679 }
680
681 if (likely(PageLRU(page)))
682 list_add_tail(&page_tail->lru, &page->lru);
683 else {
684 struct list_head *list_head;
685 /*
686 * Head page has not yet been counted, as an hpage,
687 * so we must account for each subpage individually.
688 *
689 * Use the standard add function to put page_tail on the list,
690 * but then correct its position so they all end up in order.
691 */
692 add_page_to_lru_list(zone, page_tail, lru);
693 list_head = page_tail->lru.prev;
694 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800695 }
696}
Hugh Dickins12d27102012-01-12 17:19:52 -0800697#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800698
Hugh Dickins5095ae832012-01-12 17:19:58 -0800699static void __pagevec_lru_add_fn(struct page *page, void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700700{
701 enum lru_list lru = (enum lru_list)arg;
702 struct zone *zone = page_zone(page);
703 int file = is_file_lru(lru);
704 int active = is_active_lru(lru);
705
706 VM_BUG_ON(PageActive(page));
707 VM_BUG_ON(PageUnevictable(page));
708 VM_BUG_ON(PageLRU(page));
709
710 SetPageLRU(page);
711 if (active)
712 SetPageActive(page);
713 update_page_reclaim_stat(zone, page, file, active);
714 add_page_to_lru_list(zone, page, lru);
715}
716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 * Add the passed pages to the LRU, then drop the caller's refcount
719 * on them. Reinitialises the caller's pagevec.
720 */
Hugh Dickins5095ae832012-01-12 17:19:58 -0800721void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700723 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
Hugh Dickins5095ae832012-01-12 17:19:58 -0800725 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800727EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729/**
730 * pagevec_lookup - gang pagecache lookup
731 * @pvec: Where the resulting pages are placed
732 * @mapping: The address_space to search
733 * @start: The starting page index
734 * @nr_pages: The maximum number of pages
735 *
736 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
737 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
738 * reference against the pages in @pvec.
739 *
740 * The search returns a group of mapping-contiguous pages with ascending
741 * indexes. There may be holes in the indices due to not-present pages.
742 *
743 * pagevec_lookup() returns the number of pages which were found.
744 */
745unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
746 pgoff_t start, unsigned nr_pages)
747{
748 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
749 return pagevec_count(pvec);
750}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100751EXPORT_SYMBOL(pagevec_lookup);
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
754 pgoff_t *index, int tag, unsigned nr_pages)
755{
756 pvec->nr = find_get_pages_tag(mapping, index, tag,
757 nr_pages, pvec->pages);
758 return pagevec_count(pvec);
759}
Steve French7f285702005-11-01 10:22:55 -0800760EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762/*
763 * Perform any setup for the swap system
764 */
765void __init swap_setup(void)
766{
Jan Beulich44813742009-09-21 17:03:05 -0700767 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700769#ifdef CONFIG_SWAP
770 bdi_init(swapper_space.backing_dev_info);
771#endif
772
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 /* Use a smaller cluster for small-memory machines */
774 if (megs < 16)
775 page_cluster = 2;
776 else
777 page_cluster = 3;
778 /*
779 * Right now other parts of the system means that we
780 * _really_ don't want to cluster much more
781 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}