blob: ab498ea04ae3948000ae091dfaad258853ad37fc [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);
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);
56 del_page_from_lru(zone, page);
57 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 */
80 struct page *page_head = page->first_page;
81 smp_rmb();
82 /*
83 * If PageTail is still set after smp_rmb() we can be sure
84 * that the page->first_page we read wasn't a dangling pointer.
85 * See __split_huge_page_refcount() smp_wmb().
86 */
87 if (likely(PageTail(page) && get_page_unless_zero(page_head))) {
88 unsigned long flags;
89 /*
90 * Verify that our page_head wasn't converted
91 * to a a regular page before we got a
92 * reference on it.
93 */
94 if (unlikely(!PageHead(page_head))) {
95 /* PageHead is cleared after PageTail */
96 smp_rmb();
97 VM_BUG_ON(PageTail(page));
98 goto out_put_head;
99 }
100 /*
101 * Only run compound_lock on a valid PageHead,
102 * after having it pinned with
103 * get_page_unless_zero() above.
104 */
105 smp_mb();
106 /* page_head wasn't a dangling pointer */
107 flags = compound_lock_irqsave(page_head);
108 if (unlikely(!PageTail(page))) {
109 /* __split_huge_page_refcount run before us */
110 compound_unlock_irqrestore(page_head, flags);
111 VM_BUG_ON(PageHead(page_head));
112 out_put_head:
113 if (put_page_testzero(page_head))
114 __put_single_page(page_head);
115 out_put_single:
116 if (put_page_testzero(page))
117 __put_single_page(page);
118 return;
119 }
120 VM_BUG_ON(page_head != page->first_page);
121 /*
122 * We can release the refcount taken by
123 * get_page_unless_zero now that
124 * split_huge_page_refcount is blocked on the
125 * compound_lock.
126 */
127 if (put_page_testzero(page_head))
128 VM_BUG_ON(1);
129 /* __split_huge_page_refcount will wait now */
130 VM_BUG_ON(atomic_read(&page->_count) <= 0);
131 atomic_dec(&page->_count);
132 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
133 compound_unlock_irqrestore(page_head, flags);
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800134 if (put_page_testzero(page_head)) {
135 if (PageHead(page_head))
136 __put_compound_page(page_head);
137 else
138 __put_single_page(page_head);
139 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800140 } else {
141 /* page_head is a dangling pointer */
142 VM_BUG_ON(PageTail(page));
143 goto out_put_single;
144 }
145 } else if (put_page_testzero(page)) {
146 if (PageHead(page))
147 __put_compound_page(page);
148 else
149 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800150 }
151}
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153void put_page(struct page *page)
154{
Nick Piggin8519fb32006-02-07 12:58:52 -0800155 if (unlikely(PageCompound(page)))
156 put_compound_page(page);
157 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800158 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159}
160EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700162/**
Randy Dunlap76824862008-03-19 17:00:40 -0700163 * put_pages_list() - release a list of pages
164 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700165 *
166 * Release a list of pages which are strung together on page.lru. Currently
167 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700168 */
169void put_pages_list(struct list_head *pages)
170{
171 while (!list_empty(pages)) {
172 struct page *victim;
173
174 victim = list_entry(pages->prev, struct page, lru);
175 list_del(&victim->lru);
176 page_cache_release(victim);
177 }
178}
179EXPORT_SYMBOL(put_pages_list);
180
Shaohua Lid8505de2011-01-13 15:47:33 -0800181static void pagevec_lru_move_fn(struct pagevec *pvec,
182 void (*move_fn)(struct page *page, void *arg),
183 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700184{
185 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700186 struct zone *zone = NULL;
Shaohua Lid8505de2011-01-13 15:47:33 -0800187 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700188
189 for (i = 0; i < pagevec_count(pvec); i++) {
190 struct page *page = pvec->pages[i];
191 struct zone *pagezone = page_zone(page);
192
193 if (pagezone != zone) {
194 if (zone)
Shaohua Lid8505de2011-01-13 15:47:33 -0800195 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700196 zone = pagezone;
Shaohua Lid8505de2011-01-13 15:47:33 -0800197 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700198 }
Shaohua Lid8505de2011-01-13 15:47:33 -0800199
200 (*move_fn)(page, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700201 }
202 if (zone)
Shaohua Lid8505de2011-01-13 15:47:33 -0800203 spin_unlock_irqrestore(&zone->lru_lock, flags);
204 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700205 pagevec_reinit(pvec);
206}
207
Shaohua Lid8505de2011-01-13 15:47:33 -0800208static void pagevec_move_tail_fn(struct page *page, void *arg)
209{
210 int *pgmoved = arg;
211 struct zone *zone = page_zone(page);
212
213 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
214 int lru = page_lru_base_type(page);
215 list_move_tail(&page->lru, &zone->lru[lru].list);
216 (*pgmoved)++;
217 }
218}
219
220/*
221 * pagevec_move_tail() must be called with IRQ disabled.
222 * Otherwise this may cause nasty races.
223 */
224static void pagevec_move_tail(struct pagevec *pvec)
225{
226 int pgmoved = 0;
227
228 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
229 __count_vm_events(PGROTATED, pgmoved);
230}
231
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700232/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 * Writeback is about to end against a page which has been marked for immediate
234 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700235 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 */
Shaohua Lid8505de2011-01-13 15:47:33 -0800237void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700239 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700240 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700241 struct pagevec *pvec;
242 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700244 page_cache_get(page);
245 local_irq_save(flags);
246 pvec = &__get_cpu_var(lru_rotate_pvecs);
247 if (!pagevec_add(pvec, page))
248 pagevec_move_tail(pvec);
249 local_irq_restore(flags);
250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800253static void update_page_reclaim_stat(struct zone *zone, struct page *page,
254 int file, int rotated)
255{
256 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
257 struct zone_reclaim_stat *memcg_reclaim_stat;
258
259 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
260
261 reclaim_stat->recent_scanned[file]++;
262 if (rotated)
263 reclaim_stat->recent_rotated[file]++;
264
265 if (!memcg_reclaim_stat)
266 return;
267
268 memcg_reclaim_stat->recent_scanned[file]++;
269 if (rotated)
270 memcg_reclaim_stat->recent_rotated[file]++;
271}
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273/*
Linus Torvalds7a608572011-01-17 14:42:19 -0800274 * FIXME: speed this up?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800276void activate_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
278 struct zone *zone = page_zone(page);
279
280 spin_lock_irq(&zone->lru_lock);
Linus Torvalds7a608572011-01-17 14:42:19 -0800281 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
282 int file = page_is_file_cache(page);
283 int lru = page_lru_base_type(page);
284 del_page_from_lru_list(zone, page, lru);
285
286 SetPageActive(page);
287 lru += LRU_ACTIVE;
288 add_page_to_lru_list(zone, page, lru);
289 __count_vm_event(PGACTIVATE);
290
291 update_page_reclaim_stat(zone, page, file, 1);
292 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 spin_unlock_irq(&zone->lru_lock);
294}
295
296/*
297 * Mark a page as having seen activity.
298 *
299 * inactive,unreferenced -> inactive,referenced
300 * inactive,referenced -> active,unreferenced
301 * active,unreferenced -> active,referenced
302 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800303void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700305 if (!PageActive(page) && !PageUnevictable(page) &&
306 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 activate_page(page);
308 ClearPageReferenced(page);
309 } else if (!PageReferenced(page)) {
310 SetPageReferenced(page);
311 }
312}
313
314EXPORT_SYMBOL(mark_page_accessed);
315
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700316void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700318 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320 page_cache_get(page);
321 if (!pagevec_add(pvec, page))
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700322 ____pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 put_cpu_var(lru_add_pvecs);
324}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200325EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700327/**
328 * lru_cache_add_lru - add a page to a page list
329 * @page: the page to be added to the LRU.
330 * @lru: the LRU list to which the page is added.
331 */
332void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700334 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700335 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700336 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700337 } else if (PageUnevictable(page)) {
338 VM_BUG_ON(PageActive(page));
339 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700340 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700342 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700343 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700346/**
347 * add_page_to_unevictable_list - add a page to the unevictable list
348 * @page: the page to be added to the unevictable list
349 *
350 * Add page directly to its zone's unevictable list. To avoid races with
351 * tasks that might be making the page evictable, through eg. munlock,
352 * munmap or exit, while it's not on the lru, we want to add the page
353 * while it's locked or otherwise "invisible" to other tasks. This is
354 * difficult to do when using the pagevec cache, so bypass that.
355 */
356void add_page_to_unevictable_list(struct page *page)
357{
358 struct zone *zone = page_zone(page);
359
360 spin_lock_irq(&zone->lru_lock);
361 SetPageUnevictable(page);
362 SetPageLRU(page);
363 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
364 spin_unlock_irq(&zone->lru_lock);
365}
366
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700367/*
368 * Drain pages out of the cpu's pagevecs.
369 * Either "cpu" is the current CPU, and preemption has already been
370 * disabled; or "cpu" is being hot-unplugged, and is already dead.
371 */
372static void drain_cpu_pagevecs(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700374 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700375 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700376 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700378 for_each_lru(lru) {
379 pvec = &pvecs[lru - LRU_BASE];
380 if (pagevec_count(pvec))
381 ____pagevec_lru_add(pvec, lru);
382 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700383
384 pvec = &per_cpu(lru_rotate_pvecs, cpu);
385 if (pagevec_count(pvec)) {
386 unsigned long flags;
387
388 /* No harm done if a racing interrupt already did this */
389 local_irq_save(flags);
390 pagevec_move_tail(pvec);
391 local_irq_restore(flags);
392 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800393}
394
395void lru_add_drain(void)
396{
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700397 drain_cpu_pagevecs(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800398 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399}
400
David Howellsc4028952006-11-22 14:57:56 +0000401static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800402{
403 lru_add_drain();
404}
405
406/*
407 * Returns 0 for success
408 */
409int lru_add_drain_all(void)
410{
David Howellsc4028952006-11-22 14:57:56 +0000411 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800412}
413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 * Batched page_cache_release(). Decrement the reference count on all the
416 * passed pages. If it fell to zero then remove the page from the LRU and
417 * free it.
418 *
419 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
420 * for the remainder of the operation.
421 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700422 * The locking in this function is against shrink_inactive_list(): we recheck
423 * the page count inside the lock to see whether shrink_inactive_list()
424 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
425 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 */
427void release_pages(struct page **pages, int nr, int cold)
428{
429 int i;
430 struct pagevec pages_to_free;
431 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700432 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
434 pagevec_init(&pages_to_free, cold);
435 for (i = 0; i < nr; i++) {
436 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Nick Piggin8519fb32006-02-07 12:58:52 -0800438 if (unlikely(PageCompound(page))) {
439 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700440 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800441 zone = NULL;
442 }
443 put_compound_page(page);
444 continue;
445 }
446
Nick Pigginb5810032005-10-29 18:16:12 -0700447 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 continue;
449
Nick Piggin46453a62006-03-22 00:07:58 -0800450 if (PageLRU(page)) {
451 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700452
Nick Piggin46453a62006-03-22 00:07:58 -0800453 if (pagezone != zone) {
454 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700455 spin_unlock_irqrestore(&zone->lru_lock,
456 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800457 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700458 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
Nick Piggin725d7042006-09-25 23:30:55 -0700460 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800461 __ClearPageLRU(page);
Nick Piggin46453a62006-03-22 00:07:58 -0800462 del_page_from_lru(zone, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 }
Nick Piggin46453a62006-03-22 00:07:58 -0800464
465 if (!pagevec_add(&pages_to_free, page)) {
466 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700467 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800468 zone = NULL;
469 }
470 __pagevec_free(&pages_to_free);
471 pagevec_reinit(&pages_to_free);
472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700475 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
477 pagevec_free(&pages_to_free);
478}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700479EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481/*
482 * The pages which we're about to release may be in the deferred lru-addition
483 * queues. That would prevent them from really being freed right now. That's
484 * OK from a correctness point of view but is inefficient - those pages may be
485 * cache-warm and we want to give them back to the page allocator ASAP.
486 *
487 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
488 * and __pagevec_lru_add_active() call release_pages() directly to avoid
489 * mutual recursion.
490 */
491void __pagevec_release(struct pagevec *pvec)
492{
493 lru_add_drain();
494 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
495 pagevec_reinit(pvec);
496}
497
Steve French7f285702005-11-01 10:22:55 -0800498EXPORT_SYMBOL(__pagevec_release);
499
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800500/* used by __split_huge_page_refcount() */
501void lru_add_page_tail(struct zone* zone,
502 struct page *page, struct page *page_tail)
503{
504 int active;
505 enum lru_list lru;
506 const int file = 0;
507 struct list_head *head;
508
509 VM_BUG_ON(!PageHead(page));
510 VM_BUG_ON(PageCompound(page_tail));
511 VM_BUG_ON(PageLRU(page_tail));
512 VM_BUG_ON(!spin_is_locked(&zone->lru_lock));
513
514 SetPageLRU(page_tail);
515
516 if (page_evictable(page_tail, NULL)) {
517 if (PageActive(page)) {
518 SetPageActive(page_tail);
519 active = 1;
520 lru = LRU_ACTIVE_ANON;
521 } else {
522 active = 0;
523 lru = LRU_INACTIVE_ANON;
524 }
525 update_page_reclaim_stat(zone, page_tail, file, active);
526 if (likely(PageLRU(page)))
527 head = page->lru.prev;
528 else
529 head = &zone->lru[lru].list;
530 __add_page_to_lru_list(zone, page_tail, lru, head);
531 } else {
532 SetPageUnevictable(page_tail);
533 add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE);
534 }
535}
536
Shaohua Lid8505de2011-01-13 15:47:33 -0800537static void ____pagevec_lru_add_fn(struct page *page, void *arg)
538{
539 enum lru_list lru = (enum lru_list)arg;
540 struct zone *zone = page_zone(page);
541 int file = is_file_lru(lru);
542 int active = is_active_lru(lru);
543
544 VM_BUG_ON(PageActive(page));
545 VM_BUG_ON(PageUnevictable(page));
546 VM_BUG_ON(PageLRU(page));
547
548 SetPageLRU(page);
549 if (active)
550 SetPageActive(page);
551 update_page_reclaim_stat(zone, page, file, active);
552 add_page_to_lru_list(zone, page, lru);
553}
554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 * Add the passed pages to the LRU, then drop the caller's refcount
557 * on them. Reinitialises the caller's pagevec.
558 */
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700559void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700561 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Shaohua Lid8505de2011-01-13 15:47:33 -0800563 pagevec_lru_move_fn(pvec, ____pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564}
565
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700566EXPORT_SYMBOL(____pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568/*
569 * Try to drop buffers from the pages in a pagevec
570 */
571void pagevec_strip(struct pagevec *pvec)
572{
573 int i;
574
575 for (i = 0; i < pagevec_count(pvec); i++) {
576 struct page *page = pvec->pages[i];
577
David Howells266cf652009-04-03 16:42:36 +0100578 if (page_has_private(page) && trylock_page(page)) {
579 if (page_has_private(page))
Christoph Lameter5b40dc72006-03-16 23:04:07 -0800580 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 unlock_page(page);
582 }
583 }
584}
585
586/**
587 * pagevec_lookup - gang pagecache lookup
588 * @pvec: Where the resulting pages are placed
589 * @mapping: The address_space to search
590 * @start: The starting page index
591 * @nr_pages: The maximum number of pages
592 *
593 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
594 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
595 * reference against the pages in @pvec.
596 *
597 * The search returns a group of mapping-contiguous pages with ascending
598 * indexes. There may be holes in the indices due to not-present pages.
599 *
600 * pagevec_lookup() returns the number of pages which were found.
601 */
602unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
603 pgoff_t start, unsigned nr_pages)
604{
605 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
606 return pagevec_count(pvec);
607}
608
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100609EXPORT_SYMBOL(pagevec_lookup);
610
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
612 pgoff_t *index, int tag, unsigned nr_pages)
613{
614 pvec->nr = find_get_pages_tag(mapping, index, tag,
615 nr_pages, pvec->pages);
616 return pagevec_count(pvec);
617}
618
Steve French7f285702005-11-01 10:22:55 -0800619EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621/*
622 * Perform any setup for the swap system
623 */
624void __init swap_setup(void)
625{
Jan Beulich44813742009-09-21 17:03:05 -0700626 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700628#ifdef CONFIG_SWAP
629 bdi_init(swapper_space.backing_dev_info);
630#endif
631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 /* Use a smaller cluster for small-memory machines */
633 if (megs < 16)
634 page_cluster = 2;
635 else
636 page_cluster = 3;
637 /*
638 * Right now other parts of the system means that we
639 * _really_ don't want to cluster much more
640 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}