blob: a448db377cb046d0871506f156b430dceed818fe [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 */
81 struct page *page_head = page->first_page;
82 smp_rmb();
83 /*
84 * If PageTail is still set after smp_rmb() we can be sure
85 * that the page->first_page we read wasn't a dangling pointer.
86 * See __split_huge_page_refcount() smp_wmb().
87 */
88 if (likely(PageTail(page) && get_page_unless_zero(page_head))) {
89 unsigned long flags;
90 /*
91 * Verify that our page_head wasn't converted
92 * to a a regular page before we got a
93 * reference on it.
94 */
95 if (unlikely(!PageHead(page_head))) {
96 /* PageHead is cleared after PageTail */
97 smp_rmb();
98 VM_BUG_ON(PageTail(page));
99 goto out_put_head;
100 }
101 /*
102 * Only run compound_lock on a valid PageHead,
103 * after having it pinned with
104 * get_page_unless_zero() above.
105 */
106 smp_mb();
107 /* page_head wasn't a dangling pointer */
108 flags = compound_lock_irqsave(page_head);
109 if (unlikely(!PageTail(page))) {
110 /* __split_huge_page_refcount run before us */
111 compound_unlock_irqrestore(page_head, flags);
112 VM_BUG_ON(PageHead(page_head));
113 out_put_head:
114 if (put_page_testzero(page_head))
115 __put_single_page(page_head);
116 out_put_single:
117 if (put_page_testzero(page))
118 __put_single_page(page);
119 return;
120 }
121 VM_BUG_ON(page_head != page->first_page);
122 /*
123 * We can release the refcount taken by
124 * get_page_unless_zero now that
125 * split_huge_page_refcount is blocked on the
126 * compound_lock.
127 */
128 if (put_page_testzero(page_head))
129 VM_BUG_ON(1);
130 /* __split_huge_page_refcount will wait now */
131 VM_BUG_ON(atomic_read(&page->_count) <= 0);
132 atomic_dec(&page->_count);
133 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
134 compound_unlock_irqrestore(page_head, flags);
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800135 if (put_page_testzero(page_head)) {
136 if (PageHead(page_head))
137 __put_compound_page(page_head);
138 else
139 __put_single_page(page_head);
140 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800141 } else {
142 /* page_head is a dangling pointer */
143 VM_BUG_ON(PageTail(page));
144 goto out_put_single;
145 }
146 } else if (put_page_testzero(page)) {
147 if (PageHead(page))
148 __put_compound_page(page);
149 else
150 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800151 }
152}
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154void put_page(struct page *page)
155{
Nick Piggin8519fb32006-02-07 12:58:52 -0800156 if (unlikely(PageCompound(page)))
157 put_compound_page(page);
158 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800159 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700163/**
Randy Dunlap76824862008-03-19 17:00:40 -0700164 * put_pages_list() - release a list of pages
165 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700166 *
167 * Release a list of pages which are strung together on page.lru. Currently
168 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700169 */
170void put_pages_list(struct list_head *pages)
171{
172 while (!list_empty(pages)) {
173 struct page *victim;
174
175 victim = list_entry(pages->prev, struct page, lru);
176 list_del(&victim->lru);
177 page_cache_release(victim);
178 }
179}
180EXPORT_SYMBOL(put_pages_list);
181
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700182static void pagevec_lru_move_fn(struct pagevec *pvec,
183 void (*move_fn)(struct page *page, void *arg),
184 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700185{
186 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700187 struct zone *zone = NULL;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700188 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700189
190 for (i = 0; i < pagevec_count(pvec); i++) {
191 struct page *page = pvec->pages[i];
192 struct zone *pagezone = page_zone(page);
193
194 if (pagezone != zone) {
195 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700196 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700197 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700198 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700199 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700200
201 (*move_fn)(page, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700202 }
203 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700204 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800205 release_pages(pvec->pages, pvec->nr, pvec->cold);
206 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800207}
208
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700209static void pagevec_move_tail_fn(struct page *page, void *arg)
210{
211 int *pgmoved = arg;
212 struct zone *zone = page_zone(page);
213
214 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
215 enum lru_list lru = page_lru_base_type(page);
216 list_move_tail(&page->lru, &zone->lru[lru].list);
217 mem_cgroup_rotate_reclaimable_page(page);
218 (*pgmoved)++;
219 }
220}
221
222/*
223 * pagevec_move_tail() must be called with IRQ disabled.
224 * Otherwise this may cause nasty races.
225 */
226static void pagevec_move_tail(struct pagevec *pvec)
227{
228 int pgmoved = 0;
229
230 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
231 __count_vm_events(PGROTATED, pgmoved);
232}
233
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700234/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 * Writeback is about to end against a page which has been marked for immediate
236 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700237 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700239void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700241 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700242 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700243 struct pagevec *pvec;
244 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700246 page_cache_get(page);
247 local_irq_save(flags);
248 pvec = &__get_cpu_var(lru_rotate_pvecs);
249 if (!pagevec_add(pvec, page))
250 pagevec_move_tail(pvec);
251 local_irq_restore(flags);
252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800255static void update_page_reclaim_stat(struct zone *zone, struct page *page,
256 int file, int rotated)
257{
258 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
259 struct zone_reclaim_stat *memcg_reclaim_stat;
260
261 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
262
263 reclaim_stat->recent_scanned[file]++;
264 if (rotated)
265 reclaim_stat->recent_rotated[file]++;
266
267 if (!memcg_reclaim_stat)
268 return;
269
270 memcg_reclaim_stat->recent_scanned[file]++;
271 if (rotated)
272 memcg_reclaim_stat->recent_rotated[file]++;
273}
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275/*
Linus Torvalds7a608572011-01-17 14:42:19 -0800276 * FIXME: speed this up?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800278void activate_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
280 struct zone *zone = page_zone(page);
281
282 spin_lock_irq(&zone->lru_lock);
Linus Torvalds7a608572011-01-17 14:42:19 -0800283 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
284 int file = page_is_file_cache(page);
285 int lru = page_lru_base_type(page);
286 del_page_from_lru_list(zone, page, lru);
287
288 SetPageActive(page);
289 lru += LRU_ACTIVE;
290 add_page_to_lru_list(zone, page, lru);
291 __count_vm_event(PGACTIVATE);
292
293 update_page_reclaim_stat(zone, page, file, 1);
294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 spin_unlock_irq(&zone->lru_lock);
296}
297
298/*
299 * Mark a page as having seen activity.
300 *
301 * inactive,unreferenced -> inactive,referenced
302 * inactive,referenced -> active,unreferenced
303 * active,unreferenced -> active,referenced
304 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800305void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700307 if (!PageActive(page) && !PageUnevictable(page) &&
308 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 activate_page(page);
310 ClearPageReferenced(page);
311 } else if (!PageReferenced(page)) {
312 SetPageReferenced(page);
313 }
314}
315
316EXPORT_SYMBOL(mark_page_accessed);
317
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700318void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700320 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322 page_cache_get(page);
323 if (!pagevec_add(pvec, page))
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700324 ____pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 put_cpu_var(lru_add_pvecs);
326}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200327EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700329/**
330 * lru_cache_add_lru - add a page to a page list
331 * @page: the page to be added to the LRU.
332 * @lru: the LRU list to which the page is added.
333 */
334void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700336 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700337 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700338 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700339 } else if (PageUnevictable(page)) {
340 VM_BUG_ON(PageActive(page));
341 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700342 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700344 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700345 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346}
347
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700348/**
349 * add_page_to_unevictable_list - add a page to the unevictable list
350 * @page: the page to be added to the unevictable list
351 *
352 * Add page directly to its zone's unevictable list. To avoid races with
353 * tasks that might be making the page evictable, through eg. munlock,
354 * munmap or exit, while it's not on the lru, we want to add the page
355 * while it's locked or otherwise "invisible" to other tasks. This is
356 * difficult to do when using the pagevec cache, so bypass that.
357 */
358void add_page_to_unevictable_list(struct page *page)
359{
360 struct zone *zone = page_zone(page);
361
362 spin_lock_irq(&zone->lru_lock);
363 SetPageUnevictable(page);
364 SetPageLRU(page);
365 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
366 spin_unlock_irq(&zone->lru_lock);
367}
368
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700369/*
Minchan Kim31560182011-03-22 16:32:52 -0700370 * If the page can not be invalidated, it is moved to the
371 * inactive list to speed up its reclaim. It is moved to the
372 * head of the list, rather than the tail, to give the flusher
373 * threads some time to write it out, as this is much more
374 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700375 *
376 * If the page isn't page_mapped and dirty/writeback, the page
377 * could reclaim asap using PG_reclaim.
378 *
379 * 1. active, mapped page -> none
380 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
381 * 3. inactive, mapped page -> none
382 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
383 * 5. inactive, clean -> inactive, tail
384 * 6. Others -> none
385 *
386 * In 4, why it moves inactive's head, the VM expects the page would
387 * be write it out by flusher threads as this is much more effective
388 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700389 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700390static void lru_deactivate_fn(struct page *page, void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700391{
392 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700393 bool active;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700394 struct zone *zone = page_zone(page);
Minchan Kim31560182011-03-22 16:32:52 -0700395
Minchan Kim278df9f2011-03-22 16:32:54 -0700396 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700397 return;
398
399 /* Some processes are using the page */
400 if (page_mapped(page))
401 return;
402
Minchan Kim278df9f2011-03-22 16:32:54 -0700403 active = PageActive(page);
404
Minchan Kim31560182011-03-22 16:32:52 -0700405 file = page_is_file_cache(page);
406 lru = page_lru_base_type(page);
Minchan Kim278df9f2011-03-22 16:32:54 -0700407 del_page_from_lru_list(zone, page, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700408 ClearPageActive(page);
409 ClearPageReferenced(page);
410 add_page_to_lru_list(zone, page, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700411
Minchan Kim278df9f2011-03-22 16:32:54 -0700412 if (PageWriteback(page) || PageDirty(page)) {
413 /*
414 * PG_reclaim could be raced with end_page_writeback
415 * It can make readahead confusing. But race window
416 * is _really_ small and it's non-critical problem.
417 */
418 SetPageReclaim(page);
419 } else {
420 /*
421 * The page's writeback ends up during pagevec
422 * We moves tha page into tail of inactive.
423 */
424 list_move_tail(&page->lru, &zone->lru[lru].list);
425 mem_cgroup_rotate_reclaimable_page(page);
426 __count_vm_event(PGROTATED);
427 }
428
429 if (active)
430 __count_vm_event(PGDEACTIVATE);
Minchan Kim31560182011-03-22 16:32:52 -0700431 update_page_reclaim_stat(zone, page, file, 0);
432}
433
Minchan Kim31560182011-03-22 16:32:52 -0700434/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700435 * Drain pages out of the cpu's pagevecs.
436 * Either "cpu" is the current CPU, and preemption has already been
437 * disabled; or "cpu" is being hot-unplugged, and is already dead.
438 */
439static void drain_cpu_pagevecs(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700441 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700442 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700443 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700445 for_each_lru(lru) {
446 pvec = &pvecs[lru - LRU_BASE];
447 if (pagevec_count(pvec))
448 ____pagevec_lru_add(pvec, lru);
449 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700450
451 pvec = &per_cpu(lru_rotate_pvecs, cpu);
452 if (pagevec_count(pvec)) {
453 unsigned long flags;
454
455 /* No harm done if a racing interrupt already did this */
456 local_irq_save(flags);
457 pagevec_move_tail(pvec);
458 local_irq_restore(flags);
459 }
Minchan Kim31560182011-03-22 16:32:52 -0700460
461 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
462 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700463 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700464}
465
466/**
467 * deactivate_page - forcefully deactivate a page
468 * @page: page to deactivate
469 *
470 * This function hints the VM that @page is a good reclaim candidate,
471 * for example if its invalidation fails due to the page being dirty
472 * or under writeback.
473 */
474void deactivate_page(struct page *page)
475{
476 if (likely(get_page_unless_zero(page))) {
477 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
478
479 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700480 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700481 put_cpu_var(lru_deactivate_pvecs);
482 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800483}
484
485void lru_add_drain(void)
486{
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700487 drain_cpu_pagevecs(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800488 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
490
David Howellsc4028952006-11-22 14:57:56 +0000491static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800492{
493 lru_add_drain();
494}
495
496/*
497 * Returns 0 for success
498 */
499int lru_add_drain_all(void)
500{
David Howellsc4028952006-11-22 14:57:56 +0000501 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800502}
503
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 * Batched page_cache_release(). Decrement the reference count on all the
506 * passed pages. If it fell to zero then remove the page from the LRU and
507 * free it.
508 *
509 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
510 * for the remainder of the operation.
511 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700512 * The locking in this function is against shrink_inactive_list(): we recheck
513 * the page count inside the lock to see whether shrink_inactive_list()
514 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
515 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 */
517void release_pages(struct page **pages, int nr, int cold)
518{
519 int i;
520 struct pagevec pages_to_free;
521 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700522 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
524 pagevec_init(&pages_to_free, cold);
525 for (i = 0; i < nr; i++) {
526 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Nick Piggin8519fb32006-02-07 12:58:52 -0800528 if (unlikely(PageCompound(page))) {
529 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700530 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800531 zone = NULL;
532 }
533 put_compound_page(page);
534 continue;
535 }
536
Nick Pigginb5810032005-10-29 18:16:12 -0700537 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 continue;
539
Nick Piggin46453a62006-03-22 00:07:58 -0800540 if (PageLRU(page)) {
541 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700542
Nick Piggin46453a62006-03-22 00:07:58 -0800543 if (pagezone != zone) {
544 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700545 spin_unlock_irqrestore(&zone->lru_lock,
546 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800547 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700548 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
Nick Piggin725d7042006-09-25 23:30:55 -0700550 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800551 __ClearPageLRU(page);
Nick Piggin46453a62006-03-22 00:07:58 -0800552 del_page_from_lru(zone, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
Nick Piggin46453a62006-03-22 00:07:58 -0800554
555 if (!pagevec_add(&pages_to_free, page)) {
556 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700557 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800558 zone = NULL;
559 }
560 __pagevec_free(&pages_to_free);
561 pagevec_reinit(&pages_to_free);
562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 }
564 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700565 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567 pagevec_free(&pages_to_free);
568}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700569EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
571/*
572 * The pages which we're about to release may be in the deferred lru-addition
573 * queues. That would prevent them from really being freed right now. That's
574 * OK from a correctness point of view but is inefficient - those pages may be
575 * cache-warm and we want to give them back to the page allocator ASAP.
576 *
577 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
578 * and __pagevec_lru_add_active() call release_pages() directly to avoid
579 * mutual recursion.
580 */
581void __pagevec_release(struct pagevec *pvec)
582{
583 lru_add_drain();
584 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
585 pagevec_reinit(pvec);
586}
587
Steve French7f285702005-11-01 10:22:55 -0800588EXPORT_SYMBOL(__pagevec_release);
589
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800590/* used by __split_huge_page_refcount() */
591void lru_add_page_tail(struct zone* zone,
592 struct page *page, struct page *page_tail)
593{
594 int active;
595 enum lru_list lru;
596 const int file = 0;
597 struct list_head *head;
598
599 VM_BUG_ON(!PageHead(page));
600 VM_BUG_ON(PageCompound(page_tail));
601 VM_BUG_ON(PageLRU(page_tail));
602 VM_BUG_ON(!spin_is_locked(&zone->lru_lock));
603
604 SetPageLRU(page_tail);
605
606 if (page_evictable(page_tail, NULL)) {
607 if (PageActive(page)) {
608 SetPageActive(page_tail);
609 active = 1;
610 lru = LRU_ACTIVE_ANON;
611 } else {
612 active = 0;
613 lru = LRU_INACTIVE_ANON;
614 }
615 update_page_reclaim_stat(zone, page_tail, file, active);
616 if (likely(PageLRU(page)))
617 head = page->lru.prev;
618 else
619 head = &zone->lru[lru].list;
620 __add_page_to_lru_list(zone, page_tail, lru, head);
621 } else {
622 SetPageUnevictable(page_tail);
623 add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE);
624 }
625}
626
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700627static void ____pagevec_lru_add_fn(struct page *page, void *arg)
628{
629 enum lru_list lru = (enum lru_list)arg;
630 struct zone *zone = page_zone(page);
631 int file = is_file_lru(lru);
632 int active = is_active_lru(lru);
633
634 VM_BUG_ON(PageActive(page));
635 VM_BUG_ON(PageUnevictable(page));
636 VM_BUG_ON(PageLRU(page));
637
638 SetPageLRU(page);
639 if (active)
640 SetPageActive(page);
641 update_page_reclaim_stat(zone, page, file, active);
642 add_page_to_lru_list(zone, page, lru);
643}
644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 * Add the passed pages to the LRU, then drop the caller's refcount
647 * on them. Reinitialises the caller's pagevec.
648 */
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700649void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700651 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700653 pagevec_lru_move_fn(pvec, ____pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654}
655
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700656EXPORT_SYMBOL(____pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658/*
659 * Try to drop buffers from the pages in a pagevec
660 */
661void pagevec_strip(struct pagevec *pvec)
662{
663 int i;
664
665 for (i = 0; i < pagevec_count(pvec); i++) {
666 struct page *page = pvec->pages[i];
667
David Howells266cf652009-04-03 16:42:36 +0100668 if (page_has_private(page) && trylock_page(page)) {
669 if (page_has_private(page))
Christoph Lameter5b40dc72006-03-16 23:04:07 -0800670 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 unlock_page(page);
672 }
673 }
674}
675
676/**
677 * pagevec_lookup - gang pagecache lookup
678 * @pvec: Where the resulting pages are placed
679 * @mapping: The address_space to search
680 * @start: The starting page index
681 * @nr_pages: The maximum number of pages
682 *
683 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
684 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
685 * reference against the pages in @pvec.
686 *
687 * The search returns a group of mapping-contiguous pages with ascending
688 * indexes. There may be holes in the indices due to not-present pages.
689 *
690 * pagevec_lookup() returns the number of pages which were found.
691 */
692unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
693 pgoff_t start, unsigned nr_pages)
694{
695 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
696 return pagevec_count(pvec);
697}
698
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100699EXPORT_SYMBOL(pagevec_lookup);
700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
702 pgoff_t *index, int tag, unsigned nr_pages)
703{
704 pvec->nr = find_get_pages_tag(mapping, index, tag,
705 nr_pages, pvec->pages);
706 return pagevec_count(pvec);
707}
708
Steve French7f285702005-11-01 10:22:55 -0800709EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711/*
712 * Perform any setup for the swap system
713 */
714void __init swap_setup(void)
715{
Jan Beulich44813742009-09-21 17:03:05 -0700716 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700718#ifdef CONFIG_SWAP
719 bdi_init(swapper_space.backing_dev_info);
720#endif
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 /* Use a smaller cluster for small-memory machines */
723 if (megs < 16)
724 page_cluster = 2;
725 else
726 page_cluster = 3;
727 /*
728 * Right now other parts of the system means that we
729 * _really_ don't want to cluster much more
730 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}