blob: 87627f181c3f333075cb15773bd3d45db3ec1ea9 [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>
24#include <linux/module.h>
25#include <linux/mm_inline.h>
26#include <linux/buffer_head.h> /* for try_to_release_page() */
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/percpu_counter.h>
28#include <linux/percpu.h>
29#include <linux/cpu.h>
30#include <linux/notifier.h>
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -070031#include <linux/backing-dev.h>
Balbir Singh66e17072008-02-07 00:13:56 -080032#include <linux/memcontrol.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090033#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Lee Schermerhorn64d65192008-10-18 20:26:52 -070035#include "internal.h"
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/* How many pages do we try to swap or page in/out together? */
38int page_cluster;
39
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -070040static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
Vegard Nossumf84f95042008-07-23 21:28:14 -070041static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -070042static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070043
Adrian Bunkb2213852006-09-25 23:31:02 -070044/*
45 * This path almost never happens for VM activity - pages are normally
46 * freed via pagevecs. But it gets used by networking.
47 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080048static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070049{
50 if (PageLRU(page)) {
51 unsigned long flags;
52 struct zone *zone = page_zone(page);
53
54 spin_lock_irqsave(&zone->lru_lock, flags);
55 VM_BUG_ON(!PageLRU(page));
56 __ClearPageLRU(page);
57 del_page_from_lru(zone, page);
58 spin_unlock_irqrestore(&zone->lru_lock, flags);
59 }
Andrea Arcangeli91807062011-01-13 15:46:32 -080060}
61
62static void __put_single_page(struct page *page)
63{
64 __page_cache_release(page);
Li Hongfc916682010-03-05 13:41:54 -080065 free_hot_cold_page(page, 0);
Adrian Bunkb2213852006-09-25 23:31:02 -070066}
67
Andrea Arcangeli91807062011-01-13 15:46:32 -080068static void __put_compound_page(struct page *page)
69{
70 compound_page_dtor *dtor;
71
72 __page_cache_release(page);
73 dtor = get_compound_page_dtor(page);
74 (*dtor)(page);
75}
76
Nick Piggin8519fb32006-02-07 12:58:52 -080077static void put_compound_page(struct page *page)
78{
Andrea Arcangeli91807062011-01-13 15:46:32 -080079 if (unlikely(PageTail(page))) {
80 /* __split_huge_page_refcount can run under us */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070081 struct page *page_head = compound_trans_head(page);
82
83 if (likely(page != page_head &&
84 get_page_unless_zero(page_head))) {
Andrea Arcangeli91807062011-01-13 15:46:32 -080085 unsigned long flags;
86 /*
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070087 * page_head wasn't a dangling pointer but it
88 * may not be a head page anymore by the time
89 * we obtain the lock. That is ok as long as it
90 * can't be freed from under us.
Andrea Arcangeli91807062011-01-13 15:46:32 -080091 */
Andrea Arcangeli91807062011-01-13 15:46:32 -080092 flags = compound_lock_irqsave(page_head);
93 if (unlikely(!PageTail(page))) {
94 /* __split_huge_page_refcount run before us */
95 compound_unlock_irqrestore(page_head, flags);
96 VM_BUG_ON(PageHead(page_head));
Andrea Arcangeli91807062011-01-13 15:46:32 -080097 if (put_page_testzero(page_head))
98 __put_single_page(page_head);
99 out_put_single:
100 if (put_page_testzero(page))
101 __put_single_page(page);
102 return;
103 }
104 VM_BUG_ON(page_head != page->first_page);
105 /*
106 * We can release the refcount taken by
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700107 * get_page_unless_zero() now that
108 * __split_huge_page_refcount() is blocked on
109 * the compound_lock.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800110 */
111 if (put_page_testzero(page_head))
112 VM_BUG_ON(1);
113 /* __split_huge_page_refcount will wait now */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700114 VM_BUG_ON(page_mapcount(page) <= 0);
115 atomic_dec(&page->_mapcount);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800116 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700117 VM_BUG_ON(atomic_read(&page->_count) != 0);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800118 compound_unlock_irqrestore(page_head, flags);
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800119 if (put_page_testzero(page_head)) {
120 if (PageHead(page_head))
121 __put_compound_page(page_head);
122 else
123 __put_single_page(page_head);
124 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800125 } else {
126 /* page_head is a dangling pointer */
127 VM_BUG_ON(PageTail(page));
128 goto out_put_single;
129 }
130 } else if (put_page_testzero(page)) {
131 if (PageHead(page))
132 __put_compound_page(page);
133 else
134 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800135 }
136}
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138void put_page(struct page *page)
139{
Nick Piggin8519fb32006-02-07 12:58:52 -0800140 if (unlikely(PageCompound(page)))
141 put_compound_page(page);
142 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800143 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700147/*
148 * This function is exported but must not be called by anything other
149 * than get_page(). It implements the slow path of get_page().
150 */
151bool __get_page_tail(struct page *page)
152{
153 /*
154 * This takes care of get_page() if run on a tail page
155 * returned by one of the get_user_pages/follow_page variants.
156 * get_user_pages/follow_page itself doesn't need the compound
157 * lock because it runs __get_page_tail_foll() under the
158 * proper PT lock that already serializes against
159 * split_huge_page().
160 */
161 unsigned long flags;
162 bool got = false;
163 struct page *page_head = compound_trans_head(page);
164
165 if (likely(page != page_head && get_page_unless_zero(page_head))) {
166 /*
167 * page_head wasn't a dangling pointer but it
168 * may not be a head page anymore by the time
169 * we obtain the lock. That is ok as long as it
170 * can't be freed from under us.
171 */
172 flags = compound_lock_irqsave(page_head);
173 /* here __split_huge_page_refcount won't run anymore */
174 if (likely(PageTail(page))) {
175 __get_page_tail_foll(page, false);
176 got = true;
177 }
178 compound_unlock_irqrestore(page_head, flags);
179 if (unlikely(!got))
180 put_page(page_head);
181 }
182 return got;
183}
184EXPORT_SYMBOL(__get_page_tail);
185
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700186/**
Randy Dunlap76824862008-03-19 17:00:40 -0700187 * put_pages_list() - release a list of pages
188 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700189 *
190 * Release a list of pages which are strung together on page.lru. Currently
191 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700192 */
193void put_pages_list(struct list_head *pages)
194{
195 while (!list_empty(pages)) {
196 struct page *victim;
197
198 victim = list_entry(pages->prev, struct page, lru);
199 list_del(&victim->lru);
200 page_cache_release(victim);
201 }
202}
203EXPORT_SYMBOL(put_pages_list);
204
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700205static void pagevec_lru_move_fn(struct pagevec *pvec,
206 void (*move_fn)(struct page *page, void *arg),
207 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700208{
209 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700210 struct zone *zone = NULL;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700211 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700212
213 for (i = 0; i < pagevec_count(pvec); i++) {
214 struct page *page = pvec->pages[i];
215 struct zone *pagezone = page_zone(page);
216
217 if (pagezone != zone) {
218 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700219 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700220 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700221 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700222 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700223
224 (*move_fn)(page, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700225 }
226 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700227 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800228 release_pages(pvec->pages, pvec->nr, pvec->cold);
229 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800230}
231
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700232static void pagevec_move_tail_fn(struct page *page, void *arg)
233{
234 int *pgmoved = arg;
235 struct zone *zone = page_zone(page);
236
237 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
238 enum lru_list lru = page_lru_base_type(page);
239 list_move_tail(&page->lru, &zone->lru[lru].list);
240 mem_cgroup_rotate_reclaimable_page(page);
241 (*pgmoved)++;
242 }
243}
244
245/*
246 * pagevec_move_tail() must be called with IRQ disabled.
247 * Otherwise this may cause nasty races.
248 */
249static void pagevec_move_tail(struct pagevec *pvec)
250{
251 int pgmoved = 0;
252
253 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
254 __count_vm_events(PGROTATED, pgmoved);
255}
256
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700257/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 * Writeback is about to end against a page which has been marked for immediate
259 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700260 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700262void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700264 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700265 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700266 struct pagevec *pvec;
267 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700269 page_cache_get(page);
270 local_irq_save(flags);
271 pvec = &__get_cpu_var(lru_rotate_pvecs);
272 if (!pagevec_add(pvec, page))
273 pagevec_move_tail(pvec);
274 local_irq_restore(flags);
275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800278static void update_page_reclaim_stat(struct zone *zone, struct page *page,
279 int file, int rotated)
280{
281 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
282 struct zone_reclaim_stat *memcg_reclaim_stat;
283
284 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
285
286 reclaim_stat->recent_scanned[file]++;
287 if (rotated)
288 reclaim_stat->recent_rotated[file]++;
289
290 if (!memcg_reclaim_stat)
291 return;
292
293 memcg_reclaim_stat->recent_scanned[file]++;
294 if (rotated)
295 memcg_reclaim_stat->recent_rotated[file]++;
296}
297
Shaohua Lieb709b02011-05-24 17:12:55 -0700298static void __activate_page(struct page *page, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 struct zone *zone = page_zone(page);
301
Linus Torvalds7a608572011-01-17 14:42:19 -0800302 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
303 int file = page_is_file_cache(page);
304 int lru = page_lru_base_type(page);
305 del_page_from_lru_list(zone, page, lru);
306
307 SetPageActive(page);
308 lru += LRU_ACTIVE;
309 add_page_to_lru_list(zone, page, lru);
310 __count_vm_event(PGACTIVATE);
311
312 update_page_reclaim_stat(zone, page, file, 1);
313 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700314}
315
316#ifdef CONFIG_SMP
317static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
318
319static void activate_page_drain(int cpu)
320{
321 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
322
323 if (pagevec_count(pvec))
324 pagevec_lru_move_fn(pvec, __activate_page, NULL);
325}
326
327void activate_page(struct page *page)
328{
329 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
330 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
331
332 page_cache_get(page);
333 if (!pagevec_add(pvec, page))
334 pagevec_lru_move_fn(pvec, __activate_page, NULL);
335 put_cpu_var(activate_page_pvecs);
336 }
337}
338
339#else
340static inline void activate_page_drain(int cpu)
341{
342}
343
344void activate_page(struct page *page)
345{
346 struct zone *zone = page_zone(page);
347
348 spin_lock_irq(&zone->lru_lock);
349 __activate_page(page, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 spin_unlock_irq(&zone->lru_lock);
351}
Shaohua Lieb709b02011-05-24 17:12:55 -0700352#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354/*
355 * Mark a page as having seen activity.
356 *
357 * inactive,unreferenced -> inactive,referenced
358 * inactive,referenced -> active,unreferenced
359 * active,unreferenced -> active,referenced
360 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800361void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700363 if (!PageActive(page) && !PageUnevictable(page) &&
364 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 activate_page(page);
366 ClearPageReferenced(page);
367 } else if (!PageReferenced(page)) {
368 SetPageReferenced(page);
369 }
370}
371
372EXPORT_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))
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700380 ____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 {
479 /*
480 * The page's writeback ends up during pagevec
481 * We moves tha page into tail of inactive.
482 */
483 list_move_tail(&page->lru, &zone->lru[lru].list);
484 mem_cgroup_rotate_reclaimable_page(page);
485 __count_vm_event(PGROTATED);
486 }
487
488 if (active)
489 __count_vm_event(PGDEACTIVATE);
Minchan Kim31560182011-03-22 16:32:52 -0700490 update_page_reclaim_stat(zone, page, file, 0);
491}
492
Minchan Kim31560182011-03-22 16:32:52 -0700493/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700494 * Drain pages out of the cpu's pagevecs.
495 * Either "cpu" is the current CPU, and preemption has already been
496 * disabled; or "cpu" is being hot-unplugged, and is already dead.
497 */
498static void drain_cpu_pagevecs(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700500 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700501 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700502 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700504 for_each_lru(lru) {
505 pvec = &pvecs[lru - LRU_BASE];
506 if (pagevec_count(pvec))
507 ____pagevec_lru_add(pvec, lru);
508 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700509
510 pvec = &per_cpu(lru_rotate_pvecs, cpu);
511 if (pagevec_count(pvec)) {
512 unsigned long flags;
513
514 /* No harm done if a racing interrupt already did this */
515 local_irq_save(flags);
516 pagevec_move_tail(pvec);
517 local_irq_restore(flags);
518 }
Minchan Kim31560182011-03-22 16:32:52 -0700519
520 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
521 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700522 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700523
524 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700525}
526
527/**
528 * deactivate_page - forcefully deactivate a page
529 * @page: page to deactivate
530 *
531 * This function hints the VM that @page is a good reclaim candidate,
532 * for example if its invalidation fails due to the page being dirty
533 * or under writeback.
534 */
535void deactivate_page(struct page *page)
536{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700537 /*
538 * In a workload with many unevictable page such as mprotect, unevictable
539 * page deactivation for accelerating reclaim is pointless.
540 */
541 if (PageUnevictable(page))
542 return;
543
Minchan Kim31560182011-03-22 16:32:52 -0700544 if (likely(get_page_unless_zero(page))) {
545 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
546
547 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700548 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700549 put_cpu_var(lru_deactivate_pvecs);
550 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800551}
552
553void lru_add_drain(void)
554{
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700555 drain_cpu_pagevecs(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800556 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
David Howellsc4028952006-11-22 14:57:56 +0000559static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800560{
561 lru_add_drain();
562}
563
564/*
565 * Returns 0 for success
566 */
567int lru_add_drain_all(void)
568{
David Howellsc4028952006-11-22 14:57:56 +0000569 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 * Batched page_cache_release(). Decrement the reference count on all the
574 * passed pages. If it fell to zero then remove the page from the LRU and
575 * free it.
576 *
577 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
578 * for the remainder of the operation.
579 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700580 * The locking in this function is against shrink_inactive_list(): we recheck
581 * the page count inside the lock to see whether shrink_inactive_list()
582 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
583 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 */
585void release_pages(struct page **pages, int nr, int cold)
586{
587 int i;
588 struct pagevec pages_to_free;
589 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700590 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 pagevec_init(&pages_to_free, cold);
593 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);
Nick Piggin46453a62006-03-22 00:07:58 -0800620 del_page_from_lru(zone, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 }
Nick Piggin46453a62006-03-22 00:07:58 -0800622
623 if (!pagevec_add(&pages_to_free, page)) {
624 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700625 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800626 zone = NULL;
627 }
628 __pagevec_free(&pages_to_free);
629 pagevec_reinit(&pages_to_free);
630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
632 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700633 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635 pagevec_free(&pages_to_free);
636}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700637EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639/*
640 * The pages which we're about to release may be in the deferred lru-addition
641 * queues. That would prevent them from really being freed right now. That's
642 * OK from a correctness point of view but is inefficient - those pages may be
643 * cache-warm and we want to give them back to the page allocator ASAP.
644 *
645 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
646 * and __pagevec_lru_add_active() call release_pages() directly to avoid
647 * mutual recursion.
648 */
649void __pagevec_release(struct pagevec *pvec)
650{
651 lru_add_drain();
652 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
653 pagevec_reinit(pvec);
654}
655
Steve French7f285702005-11-01 10:22:55 -0800656EXPORT_SYMBOL(__pagevec_release);
657
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800658/* used by __split_huge_page_refcount() */
659void lru_add_page_tail(struct zone* zone,
660 struct page *page, struct page *page_tail)
661{
662 int active;
663 enum lru_list lru;
664 const int file = 0;
665 struct list_head *head;
666
667 VM_BUG_ON(!PageHead(page));
668 VM_BUG_ON(PageCompound(page_tail));
669 VM_BUG_ON(PageLRU(page_tail));
670 VM_BUG_ON(!spin_is_locked(&zone->lru_lock));
671
672 SetPageLRU(page_tail);
673
674 if (page_evictable(page_tail, NULL)) {
675 if (PageActive(page)) {
676 SetPageActive(page_tail);
677 active = 1;
678 lru = LRU_ACTIVE_ANON;
679 } else {
680 active = 0;
681 lru = LRU_INACTIVE_ANON;
682 }
683 update_page_reclaim_stat(zone, page_tail, file, active);
684 if (likely(PageLRU(page)))
685 head = page->lru.prev;
686 else
687 head = &zone->lru[lru].list;
688 __add_page_to_lru_list(zone, page_tail, lru, head);
689 } else {
690 SetPageUnevictable(page_tail);
691 add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE);
692 }
693}
694
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700695static void ____pagevec_lru_add_fn(struct page *page, void *arg)
696{
697 enum lru_list lru = (enum lru_list)arg;
698 struct zone *zone = page_zone(page);
699 int file = is_file_lru(lru);
700 int active = is_active_lru(lru);
701
702 VM_BUG_ON(PageActive(page));
703 VM_BUG_ON(PageUnevictable(page));
704 VM_BUG_ON(PageLRU(page));
705
706 SetPageLRU(page);
707 if (active)
708 SetPageActive(page);
709 update_page_reclaim_stat(zone, page, file, active);
710 add_page_to_lru_list(zone, page, lru);
711}
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 * Add the passed pages to the LRU, then drop the caller's refcount
715 * on them. Reinitialises the caller's pagevec.
716 */
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700717void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700719 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700721 pagevec_lru_move_fn(pvec, ____pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700724EXPORT_SYMBOL(____pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726/*
727 * Try to drop buffers from the pages in a pagevec
728 */
729void pagevec_strip(struct pagevec *pvec)
730{
731 int i;
732
733 for (i = 0; i < pagevec_count(pvec); i++) {
734 struct page *page = pvec->pages[i];
735
David Howells266cf652009-04-03 16:42:36 +0100736 if (page_has_private(page) && trylock_page(page)) {
737 if (page_has_private(page))
Christoph Lameter5b40dc72006-03-16 23:04:07 -0800738 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 unlock_page(page);
740 }
741 }
742}
743
744/**
745 * pagevec_lookup - gang pagecache lookup
746 * @pvec: Where the resulting pages are placed
747 * @mapping: The address_space to search
748 * @start: The starting page index
749 * @nr_pages: The maximum number of pages
750 *
751 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
752 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
753 * reference against the pages in @pvec.
754 *
755 * The search returns a group of mapping-contiguous pages with ascending
756 * indexes. There may be holes in the indices due to not-present pages.
757 *
758 * pagevec_lookup() returns the number of pages which were found.
759 */
760unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
761 pgoff_t start, unsigned nr_pages)
762{
763 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
764 return pagevec_count(pvec);
765}
766
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100767EXPORT_SYMBOL(pagevec_lookup);
768
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
770 pgoff_t *index, int tag, unsigned nr_pages)
771{
772 pvec->nr = find_get_pages_tag(mapping, index, tag,
773 nr_pages, pvec->pages);
774 return pagevec_count(pvec);
775}
776
Steve French7f285702005-11-01 10:22:55 -0800777EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779/*
780 * Perform any setup for the swap system
781 */
782void __init swap_setup(void)
783{
Jan Beulich44813742009-09-21 17:03:05 -0700784 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700786#ifdef CONFIG_SWAP
787 bdi_init(swapper_space.backing_dev_info);
788#endif
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /* Use a smaller cluster for small-memory machines */
791 if (megs < 16)
792 page_cluster = 2;
793 else
794 page_cluster = 3;
795 /*
796 * Right now other parts of the system means that we
797 * _really_ don't want to cluster much more
798 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799}