blob: 70445ac3bad46d111452867d359e0bb772dcefb5 [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>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040024#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#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
Shaohua Lieb709b02011-05-24 17:12:55 -0700275static void __activate_page(struct page *page, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 struct zone *zone = page_zone(page);
278
Linus Torvalds7a608572011-01-17 14:42:19 -0800279 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
280 int file = page_is_file_cache(page);
281 int lru = page_lru_base_type(page);
282 del_page_from_lru_list(zone, page, lru);
283
284 SetPageActive(page);
285 lru += LRU_ACTIVE;
286 add_page_to_lru_list(zone, page, lru);
287 __count_vm_event(PGACTIVATE);
288
289 update_page_reclaim_stat(zone, page, file, 1);
290 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700291}
292
293#ifdef CONFIG_SMP
294static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
295
296static void activate_page_drain(int cpu)
297{
298 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
299
300 if (pagevec_count(pvec))
301 pagevec_lru_move_fn(pvec, __activate_page, NULL);
302}
303
304void activate_page(struct page *page)
305{
306 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
307 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
308
309 page_cache_get(page);
310 if (!pagevec_add(pvec, page))
311 pagevec_lru_move_fn(pvec, __activate_page, NULL);
312 put_cpu_var(activate_page_pvecs);
313 }
314}
315
316#else
317static inline void activate_page_drain(int cpu)
318{
319}
320
321void activate_page(struct page *page)
322{
323 struct zone *zone = page_zone(page);
324
325 spin_lock_irq(&zone->lru_lock);
326 __activate_page(page, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 spin_unlock_irq(&zone->lru_lock);
328}
Shaohua Lieb709b02011-05-24 17:12:55 -0700329#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331/*
332 * Mark a page as having seen activity.
333 *
334 * inactive,unreferenced -> inactive,referenced
335 * inactive,referenced -> active,unreferenced
336 * active,unreferenced -> active,referenced
337 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800338void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700340 if (!PageActive(page) && !PageUnevictable(page) &&
341 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 activate_page(page);
343 ClearPageReferenced(page);
344 } else if (!PageReferenced(page)) {
345 SetPageReferenced(page);
346 }
347}
348
349EXPORT_SYMBOL(mark_page_accessed);
350
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700351void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700353 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 page_cache_get(page);
356 if (!pagevec_add(pvec, page))
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700357 ____pagevec_lru_add(pvec, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 put_cpu_var(lru_add_pvecs);
359}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200360EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700362/**
363 * lru_cache_add_lru - add a page to a page list
364 * @page: the page to be added to the LRU.
365 * @lru: the LRU list to which the page is added.
366 */
367void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700369 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700370 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700371 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700372 } else if (PageUnevictable(page)) {
373 VM_BUG_ON(PageActive(page));
374 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700377 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700378 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700381/**
382 * add_page_to_unevictable_list - add a page to the unevictable list
383 * @page: the page to be added to the unevictable list
384 *
385 * Add page directly to its zone's unevictable list. To avoid races with
386 * tasks that might be making the page evictable, through eg. munlock,
387 * munmap or exit, while it's not on the lru, we want to add the page
388 * while it's locked or otherwise "invisible" to other tasks. This is
389 * difficult to do when using the pagevec cache, so bypass that.
390 */
391void add_page_to_unevictable_list(struct page *page)
392{
393 struct zone *zone = page_zone(page);
394
395 spin_lock_irq(&zone->lru_lock);
396 SetPageUnevictable(page);
397 SetPageLRU(page);
398 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
399 spin_unlock_irq(&zone->lru_lock);
400}
401
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700402/*
Minchan Kim31560182011-03-22 16:32:52 -0700403 * If the page can not be invalidated, it is moved to the
404 * inactive list to speed up its reclaim. It is moved to the
405 * head of the list, rather than the tail, to give the flusher
406 * threads some time to write it out, as this is much more
407 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700408 *
409 * If the page isn't page_mapped and dirty/writeback, the page
410 * could reclaim asap using PG_reclaim.
411 *
412 * 1. active, mapped page -> none
413 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
414 * 3. inactive, mapped page -> none
415 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
416 * 5. inactive, clean -> inactive, tail
417 * 6. Others -> none
418 *
419 * In 4, why it moves inactive's head, the VM expects the page would
420 * be write it out by flusher threads as this is much more effective
421 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700422 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700423static void lru_deactivate_fn(struct page *page, void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700424{
425 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700426 bool active;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700427 struct zone *zone = page_zone(page);
Minchan Kim31560182011-03-22 16:32:52 -0700428
Minchan Kim278df9f2011-03-22 16:32:54 -0700429 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700430 return;
431
Minchan Kimbad49d92011-05-11 15:13:30 -0700432 if (PageUnevictable(page))
433 return;
434
Minchan Kim31560182011-03-22 16:32:52 -0700435 /* Some processes are using the page */
436 if (page_mapped(page))
437 return;
438
Minchan Kim278df9f2011-03-22 16:32:54 -0700439 active = PageActive(page);
440
Minchan Kim31560182011-03-22 16:32:52 -0700441 file = page_is_file_cache(page);
442 lru = page_lru_base_type(page);
Minchan Kim278df9f2011-03-22 16:32:54 -0700443 del_page_from_lru_list(zone, page, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700444 ClearPageActive(page);
445 ClearPageReferenced(page);
446 add_page_to_lru_list(zone, page, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700447
Minchan Kim278df9f2011-03-22 16:32:54 -0700448 if (PageWriteback(page) || PageDirty(page)) {
449 /*
450 * PG_reclaim could be raced with end_page_writeback
451 * It can make readahead confusing. But race window
452 * is _really_ small and it's non-critical problem.
453 */
454 SetPageReclaim(page);
455 } else {
456 /*
457 * The page's writeback ends up during pagevec
458 * We moves tha page into tail of inactive.
459 */
460 list_move_tail(&page->lru, &zone->lru[lru].list);
461 mem_cgroup_rotate_reclaimable_page(page);
462 __count_vm_event(PGROTATED);
463 }
464
465 if (active)
466 __count_vm_event(PGDEACTIVATE);
Minchan Kim31560182011-03-22 16:32:52 -0700467 update_page_reclaim_stat(zone, page, file, 0);
468}
469
Minchan Kim31560182011-03-22 16:32:52 -0700470/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700471 * Drain pages out of the cpu's pagevecs.
472 * Either "cpu" is the current CPU, and preemption has already been
473 * disabled; or "cpu" is being hot-unplugged, and is already dead.
474 */
475static void drain_cpu_pagevecs(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700477 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700478 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700479 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700481 for_each_lru(lru) {
482 pvec = &pvecs[lru - LRU_BASE];
483 if (pagevec_count(pvec))
484 ____pagevec_lru_add(pvec, lru);
485 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700486
487 pvec = &per_cpu(lru_rotate_pvecs, cpu);
488 if (pagevec_count(pvec)) {
489 unsigned long flags;
490
491 /* No harm done if a racing interrupt already did this */
492 local_irq_save(flags);
493 pagevec_move_tail(pvec);
494 local_irq_restore(flags);
495 }
Minchan Kim31560182011-03-22 16:32:52 -0700496
497 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
498 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700499 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700500
501 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700502}
503
504/**
505 * deactivate_page - forcefully deactivate a page
506 * @page: page to deactivate
507 *
508 * This function hints the VM that @page is a good reclaim candidate,
509 * for example if its invalidation fails due to the page being dirty
510 * or under writeback.
511 */
512void deactivate_page(struct page *page)
513{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700514 /*
515 * In a workload with many unevictable page such as mprotect, unevictable
516 * page deactivation for accelerating reclaim is pointless.
517 */
518 if (PageUnevictable(page))
519 return;
520
Minchan Kim31560182011-03-22 16:32:52 -0700521 if (likely(get_page_unless_zero(page))) {
522 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
523
524 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700525 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700526 put_cpu_var(lru_deactivate_pvecs);
527 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800528}
529
530void lru_add_drain(void)
531{
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700532 drain_cpu_pagevecs(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800533 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534}
535
David Howellsc4028952006-11-22 14:57:56 +0000536static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800537{
538 lru_add_drain();
539}
540
541/*
542 * Returns 0 for success
543 */
544int lru_add_drain_all(void)
545{
David Howellsc4028952006-11-22 14:57:56 +0000546 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800547}
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 * Batched page_cache_release(). Decrement the reference count on all the
551 * passed pages. If it fell to zero then remove the page from the LRU and
552 * free it.
553 *
554 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
555 * for the remainder of the operation.
556 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700557 * The locking in this function is against shrink_inactive_list(): we recheck
558 * the page count inside the lock to see whether shrink_inactive_list()
559 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
560 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 */
562void release_pages(struct page **pages, int nr, int cold)
563{
564 int i;
565 struct pagevec pages_to_free;
566 struct zone *zone = NULL;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700567 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
569 pagevec_init(&pages_to_free, cold);
570 for (i = 0; i < nr; i++) {
571 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Nick Piggin8519fb32006-02-07 12:58:52 -0800573 if (unlikely(PageCompound(page))) {
574 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700575 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800576 zone = NULL;
577 }
578 put_compound_page(page);
579 continue;
580 }
581
Nick Pigginb5810032005-10-29 18:16:12 -0700582 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 continue;
584
Nick Piggin46453a62006-03-22 00:07:58 -0800585 if (PageLRU(page)) {
586 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700587
Nick Piggin46453a62006-03-22 00:07:58 -0800588 if (pagezone != zone) {
589 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700590 spin_unlock_irqrestore(&zone->lru_lock,
591 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800592 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700593 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Nick Piggin725d7042006-09-25 23:30:55 -0700595 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800596 __ClearPageLRU(page);
Nick Piggin46453a62006-03-22 00:07:58 -0800597 del_page_from_lru(zone, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 }
Nick Piggin46453a62006-03-22 00:07:58 -0800599
600 if (!pagevec_add(&pages_to_free, page)) {
601 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700602 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800603 zone = NULL;
604 }
605 __pagevec_free(&pages_to_free);
606 pagevec_reinit(&pages_to_free);
607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 }
609 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700610 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 pagevec_free(&pages_to_free);
613}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700614EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
616/*
617 * The pages which we're about to release may be in the deferred lru-addition
618 * queues. That would prevent them from really being freed right now. That's
619 * OK from a correctness point of view but is inefficient - those pages may be
620 * cache-warm and we want to give them back to the page allocator ASAP.
621 *
622 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
623 * and __pagevec_lru_add_active() call release_pages() directly to avoid
624 * mutual recursion.
625 */
626void __pagevec_release(struct pagevec *pvec)
627{
628 lru_add_drain();
629 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
630 pagevec_reinit(pvec);
631}
632
Steve French7f285702005-11-01 10:22:55 -0800633EXPORT_SYMBOL(__pagevec_release);
634
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800635/* used by __split_huge_page_refcount() */
636void lru_add_page_tail(struct zone* zone,
637 struct page *page, struct page *page_tail)
638{
639 int active;
640 enum lru_list lru;
641 const int file = 0;
642 struct list_head *head;
643
644 VM_BUG_ON(!PageHead(page));
645 VM_BUG_ON(PageCompound(page_tail));
646 VM_BUG_ON(PageLRU(page_tail));
647 VM_BUG_ON(!spin_is_locked(&zone->lru_lock));
648
649 SetPageLRU(page_tail);
650
651 if (page_evictable(page_tail, NULL)) {
652 if (PageActive(page)) {
653 SetPageActive(page_tail);
654 active = 1;
655 lru = LRU_ACTIVE_ANON;
656 } else {
657 active = 0;
658 lru = LRU_INACTIVE_ANON;
659 }
660 update_page_reclaim_stat(zone, page_tail, file, active);
661 if (likely(PageLRU(page)))
662 head = page->lru.prev;
663 else
664 head = &zone->lru[lru].list;
665 __add_page_to_lru_list(zone, page_tail, lru, head);
666 } else {
667 SetPageUnevictable(page_tail);
668 add_page_to_lru_list(zone, page_tail, LRU_UNEVICTABLE);
669 }
670}
671
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700672static void ____pagevec_lru_add_fn(struct page *page, void *arg)
673{
674 enum lru_list lru = (enum lru_list)arg;
675 struct zone *zone = page_zone(page);
676 int file = is_file_lru(lru);
677 int active = is_active_lru(lru);
678
679 VM_BUG_ON(PageActive(page));
680 VM_BUG_ON(PageUnevictable(page));
681 VM_BUG_ON(PageLRU(page));
682
683 SetPageLRU(page);
684 if (active)
685 SetPageActive(page);
686 update_page_reclaim_stat(zone, page, file, active);
687 add_page_to_lru_list(zone, page, lru);
688}
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 * Add the passed pages to the LRU, then drop the caller's refcount
692 * on them. Reinitialises the caller's pagevec.
693 */
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700694void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700696 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700698 pagevec_lru_move_fn(pvec, ____pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699}
700
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700701EXPORT_SYMBOL(____pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
703/*
704 * Try to drop buffers from the pages in a pagevec
705 */
706void pagevec_strip(struct pagevec *pvec)
707{
708 int i;
709
710 for (i = 0; i < pagevec_count(pvec); i++) {
711 struct page *page = pvec->pages[i];
712
David Howells266cf652009-04-03 16:42:36 +0100713 if (page_has_private(page) && trylock_page(page)) {
714 if (page_has_private(page))
Christoph Lameter5b40dc72006-03-16 23:04:07 -0800715 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 unlock_page(page);
717 }
718 }
719}
720
721/**
722 * pagevec_lookup - gang pagecache lookup
723 * @pvec: Where the resulting pages are placed
724 * @mapping: The address_space to search
725 * @start: The starting page index
726 * @nr_pages: The maximum number of pages
727 *
728 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
729 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
730 * reference against the pages in @pvec.
731 *
732 * The search returns a group of mapping-contiguous pages with ascending
733 * indexes. There may be holes in the indices due to not-present pages.
734 *
735 * pagevec_lookup() returns the number of pages which were found.
736 */
737unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
738 pgoff_t start, unsigned nr_pages)
739{
740 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
741 return pagevec_count(pvec);
742}
743
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100744EXPORT_SYMBOL(pagevec_lookup);
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
747 pgoff_t *index, int tag, unsigned nr_pages)
748{
749 pvec->nr = find_get_pages_tag(mapping, index, tag,
750 nr_pages, pvec->pages);
751 return pagevec_count(pvec);
752}
753
Steve French7f285702005-11-01 10:22:55 -0800754EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756/*
757 * Perform any setup for the swap system
758 */
759void __init swap_setup(void)
760{
Jan Beulich44813742009-09-21 17:03:05 -0700761 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700763#ifdef CONFIG_SWAP
764 bdi_init(swapper_space.backing_dev_info);
765#endif
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /* Use a smaller cluster for small-memory machines */
768 if (megs < 16)
769 page_cluster = 2;
770 else
771 page_cluster = 3;
772 /*
773 * Right now other parts of the system means that we
774 * _really_ don't want to cluster much more
775 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}