blob: b135ec90cdeb92673e2e97389f7cf9afc3ae80c9 [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>
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);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070041
Adrian Bunkb2213852006-09-25 23:31:02 -070042/*
43 * This path almost never happens for VM activity - pages are normally
44 * freed via pagevecs. But it gets used by networking.
45 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080046static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070047{
48 if (PageLRU(page)) {
49 unsigned long flags;
50 struct zone *zone = page_zone(page);
51
52 spin_lock_irqsave(&zone->lru_lock, flags);
53 VM_BUG_ON(!PageLRU(page));
54 __ClearPageLRU(page);
55 del_page_from_lru(zone, page);
56 spin_unlock_irqrestore(&zone->lru_lock, flags);
57 }
58 free_hot_page(page);
59}
60
Nick Piggin8519fb32006-02-07 12:58:52 -080061static void put_compound_page(struct page *page)
62{
Christoph Lameterd85f3382007-05-06 14:49:39 -070063 page = compound_head(page);
Nick Piggin8519fb32006-02-07 12:58:52 -080064 if (put_page_testzero(page)) {
Andy Whitcroft33f2ef82006-12-06 20:33:32 -080065 compound_page_dtor *dtor;
Nick Piggin8519fb32006-02-07 12:58:52 -080066
Andy Whitcroft33f2ef82006-12-06 20:33:32 -080067 dtor = get_compound_page_dtor(page);
Nick Piggin8519fb32006-02-07 12:58:52 -080068 (*dtor)(page);
69 }
70}
71
Linus Torvalds1da177e2005-04-16 15:20:36 -070072void put_page(struct page *page)
73{
Nick Piggin8519fb32006-02-07 12:58:52 -080074 if (unlikely(PageCompound(page)))
75 put_compound_page(page);
76 else if (put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 __page_cache_release(page);
78}
79EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -070081/**
Randy Dunlap76824862008-03-19 17:00:40 -070082 * put_pages_list() - release a list of pages
83 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -070084 *
85 * Release a list of pages which are strung together on page.lru. Currently
86 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -070087 */
88void put_pages_list(struct list_head *pages)
89{
90 while (!list_empty(pages)) {
91 struct page *victim;
92
93 victim = list_entry(pages->prev, struct page, lru);
94 list_del(&victim->lru);
95 page_cache_release(victim);
96 }
97}
98EXPORT_SYMBOL(put_pages_list);
99
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700101 * pagevec_move_tail() must be called with IRQ disabled.
102 * Otherwise this may cause nasty races.
103 */
104static void pagevec_move_tail(struct pagevec *pvec)
105{
106 int i;
107 int pgmoved = 0;
108 struct zone *zone = NULL;
109
110 for (i = 0; i < pagevec_count(pvec); i++) {
111 struct page *page = pvec->pages[i];
112 struct zone *pagezone = page_zone(page);
113
114 if (pagezone != zone) {
115 if (zone)
116 spin_unlock(&zone->lru_lock);
117 zone = pagezone;
118 spin_lock(&zone->lru_lock);
119 }
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700120 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700121 int lru = page_is_file_cache(page);
122 list_move_tail(&page->lru, &zone->lru[lru].list);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700123 pgmoved++;
124 }
125 }
126 if (zone)
127 spin_unlock(&zone->lru_lock);
128 __count_vm_events(PGROTATED, pgmoved);
129 release_pages(pvec->pages, pvec->nr, pvec->cold);
130 pagevec_reinit(pvec);
131}
132
133/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * Writeback is about to end against a page which has been marked for immediate
135 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700136 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 */
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700138void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700140 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700141 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700142 struct pagevec *pvec;
143 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700145 page_cache_get(page);
146 local_irq_save(flags);
147 pvec = &__get_cpu_var(lru_rotate_pvecs);
148 if (!pagevec_add(pvec, page))
149 pagevec_move_tail(pvec);
150 local_irq_restore(flags);
151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152}
153
154/*
155 * FIXME: speed this up?
156 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800157void activate_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
159 struct zone *zone = page_zone(page);
160
161 spin_lock_irq(&zone->lru_lock);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700162 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700163 int file = page_is_file_cache(page);
164 int lru = LRU_BASE + file;
165 del_page_from_lru_list(zone, page, lru);
166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 SetPageActive(page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700168 lru += LRU_ACTIVE;
169 add_page_to_lru_list(zone, page, lru);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700170 __count_vm_event(PGACTIVATE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700171 mem_cgroup_move_lists(page, lru);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700172
173 zone->recent_rotated[!!file]++;
174 zone->recent_scanned[!!file]++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 }
176 spin_unlock_irq(&zone->lru_lock);
177}
178
179/*
180 * Mark a page as having seen activity.
181 *
182 * inactive,unreferenced -> inactive,referenced
183 * inactive,referenced -> active,unreferenced
184 * active,unreferenced -> active,referenced
185 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800186void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700188 if (!PageActive(page) && !PageUnevictable(page) &&
189 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 activate_page(page);
191 ClearPageReferenced(page);
192 } else if (!PageReferenced(page)) {
193 SetPageReferenced(page);
194 }
195}
196
197EXPORT_SYMBOL(mark_page_accessed);
198
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700199void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700201 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203 page_cache_get(page);
204 if (!pagevec_add(pvec, page))
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700205 ____pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 put_cpu_var(lru_add_pvecs);
207}
208
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700209/**
210 * lru_cache_add_lru - add a page to a page list
211 * @page: the page to be added to the LRU.
212 * @lru: the LRU list to which the page is added.
213 */
214void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700216 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700217 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700218 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700219 } else if (PageUnevictable(page)) {
220 VM_BUG_ON(PageActive(page));
221 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700222 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700224 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700225 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700228/**
229 * add_page_to_unevictable_list - add a page to the unevictable list
230 * @page: the page to be added to the unevictable list
231 *
232 * Add page directly to its zone's unevictable list. To avoid races with
233 * tasks that might be making the page evictable, through eg. munlock,
234 * munmap or exit, while it's not on the lru, we want to add the page
235 * while it's locked or otherwise "invisible" to other tasks. This is
236 * difficult to do when using the pagevec cache, so bypass that.
237 */
238void add_page_to_unevictable_list(struct page *page)
239{
240 struct zone *zone = page_zone(page);
241
242 spin_lock_irq(&zone->lru_lock);
243 SetPageUnevictable(page);
244 SetPageLRU(page);
245 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
246 spin_unlock_irq(&zone->lru_lock);
247}
248
Lee Schermerhorn64d65192008-10-18 20:26:52 -0700249/**
250 * lru_cache_add_active_or_unevictable
251 * @page: the page to be added to LRU
252 * @vma: vma in which page is mapped for determining reclaimability
253 *
254 * place @page on active or unevictable LRU list, depending on
255 * page_evictable(). Note that if the page is not evictable,
256 * it goes directly back onto it's zone's unevictable list. It does
257 * NOT use a per cpu pagevec.
258 */
259void lru_cache_add_active_or_unevictable(struct page *page,
260 struct vm_area_struct *vma)
261{
262 if (page_evictable(page, vma))
263 lru_cache_add_lru(page, LRU_ACTIVE + page_is_file_cache(page));
264 else
265 add_page_to_unevictable_list(page);
266}
267
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700268/*
269 * Drain pages out of the cpu's pagevecs.
270 * Either "cpu" is the current CPU, and preemption has already been
271 * disabled; or "cpu" is being hot-unplugged, and is already dead.
272 */
273static void drain_cpu_pagevecs(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700275 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700276 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700277 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700279 for_each_lru(lru) {
280 pvec = &pvecs[lru - LRU_BASE];
281 if (pagevec_count(pvec))
282 ____pagevec_lru_add(pvec, lru);
283 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700284
285 pvec = &per_cpu(lru_rotate_pvecs, cpu);
286 if (pagevec_count(pvec)) {
287 unsigned long flags;
288
289 /* No harm done if a racing interrupt already did this */
290 local_irq_save(flags);
291 pagevec_move_tail(pvec);
292 local_irq_restore(flags);
293 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800294}
295
296void lru_add_drain(void)
297{
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700298 drain_cpu_pagevecs(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800299 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
David Howellsc4028952006-11-22 14:57:56 +0000302static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800303{
304 lru_add_drain();
305}
306
307/*
308 * Returns 0 for success
309 */
310int lru_add_drain_all(void)
311{
David Howellsc4028952006-11-22 14:57:56 +0000312 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800313}
314
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * Batched page_cache_release(). Decrement the reference count on all the
317 * passed pages. If it fell to zero then remove the page from the LRU and
318 * free it.
319 *
320 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
321 * for the remainder of the operation.
322 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700323 * The locking in this function is against shrink_inactive_list(): we recheck
324 * the page count inside the lock to see whether shrink_inactive_list()
325 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
326 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 */
328void release_pages(struct page **pages, int nr, int cold)
329{
330 int i;
331 struct pagevec pages_to_free;
332 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700333 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 pagevec_init(&pages_to_free, cold);
336 for (i = 0; i < nr; i++) {
337 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Nick Piggin8519fb32006-02-07 12:58:52 -0800339 if (unlikely(PageCompound(page))) {
340 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700341 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800342 zone = NULL;
343 }
344 put_compound_page(page);
345 continue;
346 }
347
Nick Pigginb5810032005-10-29 18:16:12 -0700348 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 continue;
350
Nick Piggin46453a62006-03-22 00:07:58 -0800351 if (PageLRU(page)) {
352 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700353
Nick Piggin46453a62006-03-22 00:07:58 -0800354 if (pagezone != zone) {
355 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700356 spin_unlock_irqrestore(&zone->lru_lock,
357 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800358 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700359 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 }
Nick Piggin725d7042006-09-25 23:30:55 -0700361 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800362 __ClearPageLRU(page);
Nick Piggin46453a62006-03-22 00:07:58 -0800363 del_page_from_lru(zone, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 }
Nick Piggin46453a62006-03-22 00:07:58 -0800365
366 if (!pagevec_add(&pages_to_free, page)) {
367 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700368 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800369 zone = NULL;
370 }
371 __pagevec_free(&pages_to_free);
372 pagevec_reinit(&pages_to_free);
373 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 }
375 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700376 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
378 pagevec_free(&pages_to_free);
379}
380
381/*
382 * The pages which we're about to release may be in the deferred lru-addition
383 * queues. That would prevent them from really being freed right now. That's
384 * OK from a correctness point of view but is inefficient - those pages may be
385 * cache-warm and we want to give them back to the page allocator ASAP.
386 *
387 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
388 * and __pagevec_lru_add_active() call release_pages() directly to avoid
389 * mutual recursion.
390 */
391void __pagevec_release(struct pagevec *pvec)
392{
393 lru_add_drain();
394 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
395 pagevec_reinit(pvec);
396}
397
Steve French7f285702005-11-01 10:22:55 -0800398EXPORT_SYMBOL(__pagevec_release);
399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400/*
401 * pagevec_release() for pages which are known to not be on the LRU
402 *
403 * This function reinitialises the caller's pagevec.
404 */
405void __pagevec_release_nonlru(struct pagevec *pvec)
406{
407 int i;
408 struct pagevec pages_to_free;
409
410 pagevec_init(&pages_to_free, pvec->cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 for (i = 0; i < pagevec_count(pvec); i++) {
412 struct page *page = pvec->pages[i];
413
Nick Piggin725d7042006-09-25 23:30:55 -0700414 VM_BUG_ON(PageLRU(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 if (put_page_testzero(page))
416 pagevec_add(&pages_to_free, page);
417 }
418 pagevec_free(&pages_to_free);
419 pagevec_reinit(pvec);
420}
421
422/*
423 * Add the passed pages to the LRU, then drop the caller's refcount
424 * on them. Reinitialises the caller's pagevec.
425 */
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700426void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
428 int i;
429 struct zone *zone = NULL;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700430 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
432 for (i = 0; i < pagevec_count(pvec); i++) {
433 struct page *page = pvec->pages[i];
434 struct zone *pagezone = page_zone(page);
Rik van Riel9ff473b2008-12-02 10:31:52 -0800435 int file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437 if (pagezone != zone) {
438 if (zone)
439 spin_unlock_irq(&zone->lru_lock);
440 zone = pagezone;
441 spin_lock_irq(&zone->lru_lock);
442 }
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700443 VM_BUG_ON(PageActive(page));
444 VM_BUG_ON(PageUnevictable(page));
Nick Piggin725d7042006-09-25 23:30:55 -0700445 VM_BUG_ON(PageLRU(page));
Nick Piggin8d438f92006-03-22 00:07:59 -0800446 SetPageLRU(page);
Rik van Riel9ff473b2008-12-02 10:31:52 -0800447 file = is_file_lru(lru);
448 zone->recent_scanned[file]++;
449 if (is_active_lru(lru)) {
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700450 SetPageActive(page);
Rik van Riel9ff473b2008-12-02 10:31:52 -0800451 zone->recent_rotated[file]++;
452 }
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700453 add_page_to_lru_list(zone, page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
455 if (zone)
456 spin_unlock_irq(&zone->lru_lock);
457 release_pages(pvec->pages, pvec->nr, pvec->cold);
458 pagevec_reinit(pvec);
459}
460
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700461EXPORT_SYMBOL(____pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463/*
464 * Try to drop buffers from the pages in a pagevec
465 */
466void pagevec_strip(struct pagevec *pvec)
467{
468 int i;
469
470 for (i = 0; i < pagevec_count(pvec); i++) {
471 struct page *page = pvec->pages[i];
472
Nick Piggin529ae9a2008-08-02 12:01:03 +0200473 if (PagePrivate(page) && trylock_page(page)) {
Christoph Lameter5b40dc72006-03-16 23:04:07 -0800474 if (PagePrivate(page))
475 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 unlock_page(page);
477 }
478 }
479}
480
481/**
Rik van Riel68a223942008-10-18 20:26:23 -0700482 * pagevec_swap_free - try to free swap space from the pages in a pagevec
483 * @pvec: pagevec with swapcache pages to free the swap space of
484 *
485 * The caller needs to hold an extra reference to each page and
486 * not hold the page lock on the pages. This function uses a
487 * trylock on the page lock so it may not always free the swap
488 * space associated with a page.
489 */
490void pagevec_swap_free(struct pagevec *pvec)
491{
492 int i;
493
494 for (i = 0; i < pagevec_count(pvec); i++) {
495 struct page *page = pvec->pages[i];
496
497 if (PageSwapCache(page) && trylock_page(page)) {
498 if (PageSwapCache(page))
499 remove_exclusive_swap_page_ref(page);
500 unlock_page(page);
501 }
502 }
503}
504
505/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 * pagevec_lookup - gang pagecache lookup
507 * @pvec: Where the resulting pages are placed
508 * @mapping: The address_space to search
509 * @start: The starting page index
510 * @nr_pages: The maximum number of pages
511 *
512 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
513 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
514 * reference against the pages in @pvec.
515 *
516 * The search returns a group of mapping-contiguous pages with ascending
517 * indexes. There may be holes in the indices due to not-present pages.
518 *
519 * pagevec_lookup() returns the number of pages which were found.
520 */
521unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
522 pgoff_t start, unsigned nr_pages)
523{
524 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
525 return pagevec_count(pvec);
526}
527
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100528EXPORT_SYMBOL(pagevec_lookup);
529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
531 pgoff_t *index, int tag, unsigned nr_pages)
532{
533 pvec->nr = find_get_pages_tag(mapping, index, tag,
534 nr_pages, pvec->pages);
535 return pagevec_count(pvec);
536}
537
Steve French7f285702005-11-01 10:22:55 -0800538EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540#ifdef CONFIG_SMP
541/*
542 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
543 * CPUs
544 */
545#define ACCT_THRESHOLD max(16, NR_CPUS * 2)
546
Vegard Nossumf84f95042008-07-23 21:28:14 -0700547static DEFINE_PER_CPU(long, committed_space);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
549void vm_acct_memory(long pages)
550{
551 long *local;
552
553 preempt_disable();
554 local = &__get_cpu_var(committed_space);
555 *local += pages;
556 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
Alan Cox80119ef2008-05-23 13:04:31 -0700557 atomic_long_add(*local, &vm_committed_space);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 *local = 0;
559 }
560 preempt_enable();
561}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
563#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
565/* Drop the CPU's cached committed space back into the central pool. */
566static int cpu_swap_callback(struct notifier_block *nfb,
567 unsigned long action,
568 void *hcpu)
569{
570 long *committed;
571
572 committed = &per_cpu(committed_space, (long)hcpu);
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -0700573 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Alan Cox80119ef2008-05-23 13:04:31 -0700574 atomic_long_add(*committed, &vm_committed_space);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 *committed = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700576 drain_cpu_pagevecs((long)hcpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 }
578 return NOTIFY_OK;
579}
580#endif /* CONFIG_HOTPLUG_CPU */
581#endif /* CONFIG_SMP */
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583/*
584 * Perform any setup for the swap system
585 */
586void __init swap_setup(void)
587{
588 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
589
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700590#ifdef CONFIG_SWAP
591 bdi_init(swapper_space.backing_dev_info);
592#endif
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 /* Use a smaller cluster for small-memory machines */
595 if (megs < 16)
596 page_cluster = 2;
597 else
598 page_cluster = 3;
599 /*
600 * Right now other parts of the system means that we
601 * _really_ don't want to cluster much more
602 */
Ingo Molnar02316062006-12-06 20:38:17 -0800603#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 hotcpu_notifier(cpu_swap_callback, 0);
Ingo Molnar02316062006-12-06 20:38:17 -0800605#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606}