blob: c02f93611a849e6c79bdbae2e13ee6e4cff6eb37 [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
Linus Torvalds83896fb2011-01-17 14:42:34 -0800181/*
182 * pagevec_move_tail() must be called with IRQ disabled.
183 * Otherwise this may cause nasty races.
184 */
185static void pagevec_move_tail(struct pagevec *pvec)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700186{
187 int i;
Linus Torvalds83896fb2011-01-17 14:42:34 -0800188 int pgmoved = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700189 struct zone *zone = NULL;
190
191 for (i = 0; i < pagevec_count(pvec); i++) {
192 struct page *page = pvec->pages[i];
193 struct zone *pagezone = page_zone(page);
194
195 if (pagezone != zone) {
196 if (zone)
Linus Torvalds83896fb2011-01-17 14:42:34 -0800197 spin_unlock(&zone->lru_lock);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700198 zone = pagezone;
Linus Torvalds83896fb2011-01-17 14:42:34 -0800199 spin_lock(&zone->lru_lock);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700200 }
Linus Torvalds83896fb2011-01-17 14:42:34 -0800201 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
202 int lru = page_lru_base_type(page);
203 list_move_tail(&page->lru, &zone->lru[lru].list);
204 pgmoved++;
205 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700206 }
207 if (zone)
Linus Torvalds83896fb2011-01-17 14:42:34 -0800208 spin_unlock(&zone->lru_lock);
Shaohua Lid8505de2011-01-13 15:47:33 -0800209 __count_vm_events(PGROTATED, pgmoved);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800210 release_pages(pvec->pages, pvec->nr, pvec->cold);
211 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800212}
213
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700214/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 * Writeback is about to end against a page which has been marked for immediate
216 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700217 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 */
Linus Torvalds83896fb2011-01-17 14:42:34 -0800219void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700221 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700222 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700223 struct pagevec *pvec;
224 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700226 page_cache_get(page);
227 local_irq_save(flags);
228 pvec = &__get_cpu_var(lru_rotate_pvecs);
229 if (!pagevec_add(pvec, page))
230 pagevec_move_tail(pvec);
231 local_irq_restore(flags);
232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800235static void update_page_reclaim_stat(struct zone *zone, struct page *page,
236 int file, int rotated)
237{
238 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
239 struct zone_reclaim_stat *memcg_reclaim_stat;
240
241 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
242
243 reclaim_stat->recent_scanned[file]++;
244 if (rotated)
245 reclaim_stat->recent_rotated[file]++;
246
247 if (!memcg_reclaim_stat)
248 return;
249
250 memcg_reclaim_stat->recent_scanned[file]++;
251 if (rotated)
252 memcg_reclaim_stat->recent_rotated[file]++;
253}
254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255/*
Linus Torvalds7a608572011-01-17 14:42:19 -0800256 * FIXME: speed this up?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800258void activate_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
260 struct zone *zone = page_zone(page);
261
262 spin_lock_irq(&zone->lru_lock);
Linus Torvalds7a608572011-01-17 14:42:19 -0800263 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
264 int file = page_is_file_cache(page);
265 int lru = page_lru_base_type(page);
266 del_page_from_lru_list(zone, page, lru);
267
268 SetPageActive(page);
269 lru += LRU_ACTIVE;
270 add_page_to_lru_list(zone, page, lru);
271 __count_vm_event(PGACTIVATE);
272
273 update_page_reclaim_stat(zone, page, file, 1);
274 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 spin_unlock_irq(&zone->lru_lock);
276}
277
278/*
279 * Mark a page as having seen activity.
280 *
281 * inactive,unreferenced -> inactive,referenced
282 * inactive,referenced -> active,unreferenced
283 * active,unreferenced -> active,referenced
284 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800285void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700287 if (!PageActive(page) && !PageUnevictable(page) &&
288 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 activate_page(page);
290 ClearPageReferenced(page);
291 } else if (!PageReferenced(page)) {
292 SetPageReferenced(page);
293 }
294}
295
296EXPORT_SYMBOL(mark_page_accessed);
297
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700298void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700300 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
302 page_cache_get(page);
303 if (!pagevec_add(pvec, page))
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700304 ____pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 put_cpu_var(lru_add_pvecs);
306}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200307EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700309/**
310 * lru_cache_add_lru - add a page to a page list
311 * @page: the page to be added to the LRU.
312 * @lru: the LRU list to which the page is added.
313 */
314void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700316 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700317 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700318 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700319 } else if (PageUnevictable(page)) {
320 VM_BUG_ON(PageActive(page));
321 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700322 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700324 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700325 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700328/**
329 * add_page_to_unevictable_list - add a page to the unevictable list
330 * @page: the page to be added to the unevictable list
331 *
332 * Add page directly to its zone's unevictable list. To avoid races with
333 * tasks that might be making the page evictable, through eg. munlock,
334 * munmap or exit, while it's not on the lru, we want to add the page
335 * while it's locked or otherwise "invisible" to other tasks. This is
336 * difficult to do when using the pagevec cache, so bypass that.
337 */
338void add_page_to_unevictable_list(struct page *page)
339{
340 struct zone *zone = page_zone(page);
341
342 spin_lock_irq(&zone->lru_lock);
343 SetPageUnevictable(page);
344 SetPageLRU(page);
345 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
346 spin_unlock_irq(&zone->lru_lock);
347}
348
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700349/*
350 * Drain pages out of the cpu's pagevecs.
351 * Either "cpu" is the current CPU, and preemption has already been
352 * disabled; or "cpu" is being hot-unplugged, and is already dead.
353 */
354static void drain_cpu_pagevecs(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700356 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700357 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700358 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700360 for_each_lru(lru) {
361 pvec = &pvecs[lru - LRU_BASE];
362 if (pagevec_count(pvec))
363 ____pagevec_lru_add(pvec, lru);
364 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700365
366 pvec = &per_cpu(lru_rotate_pvecs, cpu);
367 if (pagevec_count(pvec)) {
368 unsigned long flags;
369
370 /* No harm done if a racing interrupt already did this */
371 local_irq_save(flags);
372 pagevec_move_tail(pvec);
373 local_irq_restore(flags);
374 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800375}
376
377void lru_add_drain(void)
378{
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700379 drain_cpu_pagevecs(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800380 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381}
382
David Howellsc4028952006-11-22 14:57:56 +0000383static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800384{
385 lru_add_drain();
386}
387
388/*
389 * Returns 0 for success
390 */
391int lru_add_drain_all(void)
392{
David Howellsc4028952006-11-22 14:57:56 +0000393 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800394}
395
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 * Batched page_cache_release(). Decrement the reference count on all the
398 * passed pages. If it fell to zero then remove the page from the LRU and
399 * free it.
400 *
401 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
402 * for the remainder of the operation.
403 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700404 * The locking in this function is against shrink_inactive_list(): we recheck
405 * the page count inside the lock to see whether shrink_inactive_list()
406 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
407 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 */
409void release_pages(struct page **pages, int nr, int cold)
410{
411 int i;
412 struct pagevec pages_to_free;
413 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700414 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 pagevec_init(&pages_to_free, cold);
417 for (i = 0; i < nr; i++) {
418 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Nick Piggin8519fb32006-02-07 12:58:52 -0800420 if (unlikely(PageCompound(page))) {
421 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700422 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800423 zone = NULL;
424 }
425 put_compound_page(page);
426 continue;
427 }
428
Nick Pigginb5810032005-10-29 18:16:12 -0700429 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 continue;
431
Nick Piggin46453a62006-03-22 00:07:58 -0800432 if (PageLRU(page)) {
433 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700434
Nick Piggin46453a62006-03-22 00:07:58 -0800435 if (pagezone != zone) {
436 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700437 spin_unlock_irqrestore(&zone->lru_lock,
438 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800439 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700440 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
Nick Piggin725d7042006-09-25 23:30:55 -0700442 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800443 __ClearPageLRU(page);
Nick Piggin46453a62006-03-22 00:07:58 -0800444 del_page_from_lru(zone, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 }
Nick Piggin46453a62006-03-22 00:07:58 -0800446
447 if (!pagevec_add(&pages_to_free, page)) {
448 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700449 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800450 zone = NULL;
451 }
452 __pagevec_free(&pages_to_free);
453 pagevec_reinit(&pages_to_free);
454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
456 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700457 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 pagevec_free(&pages_to_free);
460}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700461EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463/*
464 * The pages which we're about to release may be in the deferred lru-addition
465 * queues. That would prevent them from really being freed right now. That's
466 * OK from a correctness point of view but is inefficient - those pages may be
467 * cache-warm and we want to give them back to the page allocator ASAP.
468 *
469 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
470 * and __pagevec_lru_add_active() call release_pages() directly to avoid
471 * mutual recursion.
472 */
473void __pagevec_release(struct pagevec *pvec)
474{
475 lru_add_drain();
476 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
477 pagevec_reinit(pvec);
478}
479
Steve French7f285702005-11-01 10:22:55 -0800480EXPORT_SYMBOL(__pagevec_release);
481
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800482/* used by __split_huge_page_refcount() */
483void lru_add_page_tail(struct zone* zone,
484 struct page *page, struct page *page_tail)
485{
486 int active;
487 enum lru_list lru;
488 const int file = 0;
489 struct list_head *head;
490
491 VM_BUG_ON(!PageHead(page));
492 VM_BUG_ON(PageCompound(page_tail));
493 VM_BUG_ON(PageLRU(page_tail));
494 VM_BUG_ON(!spin_is_locked(&zone->lru_lock));
495
496 SetPageLRU(page_tail);
497
498 if (page_evictable(page_tail, NULL)) {
499 if (PageActive(page)) {
500 SetPageActive(page_tail);
501 active = 1;
502 lru = LRU_ACTIVE_ANON;
503 } else {
504 active = 0;
505 lru = LRU_INACTIVE_ANON;
506 }
507 update_page_reclaim_stat(zone, page_tail, file, active);
508 if (likely(PageLRU(page)))
509 head = page->lru.prev;
510 else
511 head = &zone->lru[lru].list;
512 __add_page_to_lru_list(zone, page_tail, lru, head);
513 } else {
514 SetPageUnevictable(page_tail);
515 add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE);
516 }
517}
518
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 * Add the passed pages to the LRU, then drop the caller's refcount
521 * on them. Reinitialises the caller's pagevec.
522 */
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700523void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
Linus Torvalds83896fb2011-01-17 14:42:34 -0800525 int i;
526 struct zone *zone = NULL;
527
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700528 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Linus Torvalds83896fb2011-01-17 14:42:34 -0800530 for (i = 0; i < pagevec_count(pvec); i++) {
531 struct page *page = pvec->pages[i];
532 struct zone *pagezone = page_zone(page);
533 int file;
534 int active;
535
536 if (pagezone != zone) {
537 if (zone)
538 spin_unlock_irq(&zone->lru_lock);
539 zone = pagezone;
540 spin_lock_irq(&zone->lru_lock);
541 }
542 VM_BUG_ON(PageActive(page));
543 VM_BUG_ON(PageUnevictable(page));
544 VM_BUG_ON(PageLRU(page));
545 SetPageLRU(page);
546 active = is_active_lru(lru);
547 file = is_file_lru(lru);
548 if (active)
549 SetPageActive(page);
550 update_page_reclaim_stat(zone, page, file, active);
551 add_page_to_lru_list(zone, page, lru);
552 }
553 if (zone)
554 spin_unlock_irq(&zone->lru_lock);
555 release_pages(pvec->pages, pvec->nr, pvec->cold);
556 pagevec_reinit(pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700559EXPORT_SYMBOL(____pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561/*
562 * Try to drop buffers from the pages in a pagevec
563 */
564void pagevec_strip(struct pagevec *pvec)
565{
566 int i;
567
568 for (i = 0; i < pagevec_count(pvec); i++) {
569 struct page *page = pvec->pages[i];
570
David Howells266cf652009-04-03 16:42:36 +0100571 if (page_has_private(page) && trylock_page(page)) {
572 if (page_has_private(page))
Christoph Lameter5b40dc72006-03-16 23:04:07 -0800573 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 unlock_page(page);
575 }
576 }
577}
578
579/**
580 * pagevec_lookup - gang pagecache lookup
581 * @pvec: Where the resulting pages are placed
582 * @mapping: The address_space to search
583 * @start: The starting page index
584 * @nr_pages: The maximum number of pages
585 *
586 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
587 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
588 * reference against the pages in @pvec.
589 *
590 * The search returns a group of mapping-contiguous pages with ascending
591 * indexes. There may be holes in the indices due to not-present pages.
592 *
593 * pagevec_lookup() returns the number of pages which were found.
594 */
595unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
596 pgoff_t start, unsigned nr_pages)
597{
598 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
599 return pagevec_count(pvec);
600}
601
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100602EXPORT_SYMBOL(pagevec_lookup);
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
605 pgoff_t *index, int tag, unsigned nr_pages)
606{
607 pvec->nr = find_get_pages_tag(mapping, index, tag,
608 nr_pages, pvec->pages);
609 return pagevec_count(pvec);
610}
611
Steve French7f285702005-11-01 10:22:55 -0800612EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614/*
615 * Perform any setup for the swap system
616 */
617void __init swap_setup(void)
618{
Jan Beulich44813742009-09-21 17:03:05 -0700619 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700621#ifdef CONFIG_SWAP
622 bdi_init(swapper_space.backing_dev_info);
623#endif
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 /* Use a smaller cluster for small-memory machines */
626 if (megs < 16)
627 page_cluster = 2;
628 else
629 page_cluster = 3;
630 /*
631 * Right now other parts of the system means that we
632 * _really_ don't want to cluster much more
633 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}