blob: 62b78a6e224f2f10682e9fa11f28b8d01da1c324 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/percpu_counter.h>
27#include <linux/percpu.h>
28#include <linux/cpu.h>
29#include <linux/notifier.h>
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -070030#include <linux/backing-dev.h>
Balbir Singh66e17072008-02-07 00:13:56 -080031#include <linux/memcontrol.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/gfp.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070033#include <linux/uio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Lee Schermerhorn64d65192008-10-18 20:26:52 -070035#include "internal.h"
36
Mel Gormanc6286c92013-07-03 15:02:26 -070037#define CREATE_TRACE_POINTS
38#include <trace/events/pagemap.h>
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/* How many pages do we try to swap or page in/out together? */
41int page_cluster;
42
Mel Gorman13f7f782013-07-03 15:02:28 -070043static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
Vegard Nossumf84f95042008-07-23 21:28:14 -070044static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -070045static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070046
Adrian Bunkb2213852006-09-25 23:31:02 -070047/*
48 * This path almost never happens for VM activity - pages are normally
49 * freed via pagevecs. But it gets used by networking.
50 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080051static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070052{
53 if (PageLRU(page)) {
Adrian Bunkb2213852006-09-25 23:31:02 -070054 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070055 struct lruvec *lruvec;
56 unsigned long flags;
Adrian Bunkb2213852006-09-25 23:31:02 -070057
58 spin_lock_irqsave(&zone->lru_lock, flags);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070059 lruvec = mem_cgroup_page_lruvec(page, zone);
Adrian Bunkb2213852006-09-25 23:31:02 -070060 VM_BUG_ON(!PageLRU(page));
61 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070062 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070063 spin_unlock_irqrestore(&zone->lru_lock, flags);
64 }
Andrea Arcangeli91807062011-01-13 15:46:32 -080065}
66
67static void __put_single_page(struct page *page)
68{
69 __page_cache_release(page);
Li Hongfc916682010-03-05 13:41:54 -080070 free_hot_cold_page(page, 0);
Adrian Bunkb2213852006-09-25 23:31:02 -070071}
72
Andrea Arcangeli91807062011-01-13 15:46:32 -080073static void __put_compound_page(struct page *page)
74{
75 compound_page_dtor *dtor;
76
77 __page_cache_release(page);
78 dtor = get_compound_page_dtor(page);
79 (*dtor)(page);
80}
81
Nick Piggin8519fb32006-02-07 12:58:52 -080082static void put_compound_page(struct page *page)
83{
Andrea Arcangeli91807062011-01-13 15:46:32 -080084 if (unlikely(PageTail(page))) {
85 /* __split_huge_page_refcount can run under us */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070086 struct page *page_head = compound_trans_head(page);
87
88 if (likely(page != page_head &&
89 get_page_unless_zero(page_head))) {
Andrea Arcangeli91807062011-01-13 15:46:32 -080090 unsigned long flags;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -070091
92 /*
93 * THP can not break up slab pages so avoid taking
94 * compound_lock(). Slab performs non-atomic bit ops
95 * on page->flags for better performance. In particular
96 * slab_unlock() in slub used to be a hot path. It is
97 * still hot on arches that do not support
98 * this_cpu_cmpxchg_double().
99 */
100 if (PageSlab(page_head)) {
101 if (PageTail(page)) {
102 if (put_page_testzero(page_head))
103 VM_BUG_ON(1);
104
105 atomic_dec(&page->_mapcount);
106 goto skip_lock_tail;
107 } else
108 goto skip_lock;
109 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800110 /*
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700111 * page_head wasn't a dangling pointer but it
112 * may not be a head page anymore by the time
113 * we obtain the lock. That is ok as long as it
114 * can't be freed from under us.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800115 */
Andrea Arcangeli91807062011-01-13 15:46:32 -0800116 flags = compound_lock_irqsave(page_head);
117 if (unlikely(!PageTail(page))) {
118 /* __split_huge_page_refcount run before us */
119 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700120skip_lock:
Andrea Arcangeli91807062011-01-13 15:46:32 -0800121 if (put_page_testzero(page_head))
122 __put_single_page(page_head);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700123out_put_single:
Andrea Arcangeli91807062011-01-13 15:46:32 -0800124 if (put_page_testzero(page))
125 __put_single_page(page);
126 return;
127 }
128 VM_BUG_ON(page_head != page->first_page);
129 /*
130 * We can release the refcount taken by
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700131 * get_page_unless_zero() now that
132 * __split_huge_page_refcount() is blocked on
133 * the compound_lock.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800134 */
135 if (put_page_testzero(page_head))
136 VM_BUG_ON(1);
137 /* __split_huge_page_refcount will wait now */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700138 VM_BUG_ON(page_mapcount(page) <= 0);
139 atomic_dec(&page->_mapcount);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800140 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700141 VM_BUG_ON(atomic_read(&page->_count) != 0);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800142 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700143
144skip_lock_tail:
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800145 if (put_page_testzero(page_head)) {
146 if (PageHead(page_head))
147 __put_compound_page(page_head);
148 else
149 __put_single_page(page_head);
150 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800151 } else {
152 /* page_head is a dangling pointer */
153 VM_BUG_ON(PageTail(page));
154 goto out_put_single;
155 }
156 } else if (put_page_testzero(page)) {
157 if (PageHead(page))
158 __put_compound_page(page);
159 else
160 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800161 }
162}
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164void put_page(struct page *page)
165{
Nick Piggin8519fb32006-02-07 12:58:52 -0800166 if (unlikely(PageCompound(page)))
167 put_compound_page(page);
168 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800169 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
171EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700173/*
174 * This function is exported but must not be called by anything other
175 * than get_page(). It implements the slow path of get_page().
176 */
177bool __get_page_tail(struct page *page)
178{
179 /*
180 * This takes care of get_page() if run on a tail page
181 * returned by one of the get_user_pages/follow_page variants.
182 * get_user_pages/follow_page itself doesn't need the compound
183 * lock because it runs __get_page_tail_foll() under the
184 * proper PT lock that already serializes against
185 * split_huge_page().
186 */
187 unsigned long flags;
188 bool got = false;
189 struct page *page_head = compound_trans_head(page);
190
191 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700192
193 /* Ref to put_compound_page() comment. */
194 if (PageSlab(page_head)) {
195 if (likely(PageTail(page))) {
196 __get_page_tail_foll(page, false);
197 return true;
198 } else {
199 put_page(page_head);
200 return false;
201 }
202 }
203
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700204 /*
205 * page_head wasn't a dangling pointer but it
206 * may not be a head page anymore by the time
207 * we obtain the lock. That is ok as long as it
208 * can't be freed from under us.
209 */
210 flags = compound_lock_irqsave(page_head);
211 /* here __split_huge_page_refcount won't run anymore */
212 if (likely(PageTail(page))) {
213 __get_page_tail_foll(page, false);
214 got = true;
215 }
216 compound_unlock_irqrestore(page_head, flags);
217 if (unlikely(!got))
218 put_page(page_head);
219 }
220 return got;
221}
222EXPORT_SYMBOL(__get_page_tail);
223
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700224/**
Randy Dunlap76824862008-03-19 17:00:40 -0700225 * put_pages_list() - release a list of pages
226 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700227 *
228 * Release a list of pages which are strung together on page.lru. Currently
229 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700230 */
231void put_pages_list(struct list_head *pages)
232{
233 while (!list_empty(pages)) {
234 struct page *victim;
235
236 victim = list_entry(pages->prev, struct page, lru);
237 list_del(&victim->lru);
238 page_cache_release(victim);
239 }
240}
241EXPORT_SYMBOL(put_pages_list);
242
Mel Gorman18022c52012-07-31 16:44:51 -0700243/*
244 * get_kernel_pages() - pin kernel pages in memory
245 * @kiov: An array of struct kvec structures
246 * @nr_segs: number of segments to pin
247 * @write: pinning for read/write, currently ignored
248 * @pages: array that receives pointers to the pages pinned.
249 * Should be at least nr_segs long.
250 *
251 * Returns number of pages pinned. This may be fewer than the number
252 * requested. If nr_pages is 0 or negative, returns 0. If no pages
253 * were pinned, returns -errno. Each page returned must be released
254 * with a put_page() call when it is finished with.
255 */
256int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
257 struct page **pages)
258{
259 int seg;
260
261 for (seg = 0; seg < nr_segs; seg++) {
262 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
263 return seg;
264
Mel Gorman5a178112012-07-31 16:45:02 -0700265 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700266 page_cache_get(pages[seg]);
267 }
268
269 return seg;
270}
271EXPORT_SYMBOL_GPL(get_kernel_pages);
272
273/*
274 * get_kernel_page() - pin a kernel page in memory
275 * @start: starting kernel address
276 * @write: pinning for read/write, currently ignored
277 * @pages: array that receives pointer to the page pinned.
278 * Must be at least nr_segs long.
279 *
280 * Returns 1 if page is pinned. If the page was not pinned, returns
281 * -errno. The page returned must be released with a put_page() call
282 * when it is finished with.
283 */
284int get_kernel_page(unsigned long start, int write, struct page **pages)
285{
286 const struct kvec kiov = {
287 .iov_base = (void *)start,
288 .iov_len = PAGE_SIZE
289 };
290
291 return get_kernel_pages(&kiov, 1, write, pages);
292}
293EXPORT_SYMBOL_GPL(get_kernel_page);
294
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700295static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700296 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
297 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700298{
299 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700300 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700301 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700302 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700303
304 for (i = 0; i < pagevec_count(pvec); i++) {
305 struct page *page = pvec->pages[i];
306 struct zone *pagezone = page_zone(page);
307
308 if (pagezone != zone) {
309 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700310 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700311 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700312 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700313 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700314
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700315 lruvec = mem_cgroup_page_lruvec(page, zone);
316 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700317 }
318 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700319 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800320 release_pages(pvec->pages, pvec->nr, pvec->cold);
321 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800322}
323
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700324static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
325 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700326{
327 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700328
329 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
330 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800331 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700332 (*pgmoved)++;
333 }
334}
335
336/*
337 * pagevec_move_tail() must be called with IRQ disabled.
338 * Otherwise this may cause nasty races.
339 */
340static void pagevec_move_tail(struct pagevec *pvec)
341{
342 int pgmoved = 0;
343
344 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
345 __count_vm_events(PGROTATED, pgmoved);
346}
347
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700348/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 * Writeback is about to end against a page which has been marked for immediate
350 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700351 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700353void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700355 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700356 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700357 struct pagevec *pvec;
358 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700360 page_cache_get(page);
361 local_irq_save(flags);
362 pvec = &__get_cpu_var(lru_rotate_pvecs);
363 if (!pagevec_add(pvec, page))
364 pagevec_move_tail(pvec);
365 local_irq_restore(flags);
366 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700369static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800370 int file, int rotated)
371{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700372 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800373
374 reclaim_stat->recent_scanned[file]++;
375 if (rotated)
376 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800377}
378
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700379static void __activate_page(struct page *page, struct lruvec *lruvec,
380 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Linus Torvalds7a608572011-01-17 14:42:19 -0800382 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
383 int file = page_is_file_cache(page);
384 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800385
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700386 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800387 SetPageActive(page);
388 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700389 add_page_to_lru_list(page, lruvec, lru);
Mel Gormanc6286c92013-07-03 15:02:26 -0700390 trace_mm_lru_activate(page, page_to_pfn(page));
Linus Torvalds7a608572011-01-17 14:42:19 -0800391
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700392 __count_vm_event(PGACTIVATE);
393 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800394 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700395}
396
397#ifdef CONFIG_SMP
398static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
399
400static void activate_page_drain(int cpu)
401{
402 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
403
404 if (pagevec_count(pvec))
405 pagevec_lru_move_fn(pvec, __activate_page, NULL);
406}
407
408void activate_page(struct page *page)
409{
410 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
411 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
412
413 page_cache_get(page);
414 if (!pagevec_add(pvec, page))
415 pagevec_lru_move_fn(pvec, __activate_page, NULL);
416 put_cpu_var(activate_page_pvecs);
417 }
418}
419
420#else
421static inline void activate_page_drain(int cpu)
422{
423}
424
425void activate_page(struct page *page)
426{
427 struct zone *zone = page_zone(page);
428
429 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700430 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 spin_unlock_irq(&zone->lru_lock);
432}
Shaohua Lieb709b02011-05-24 17:12:55 -0700433#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Mel Gorman059285a2013-07-03 15:02:30 -0700435static void __lru_cache_activate_page(struct page *page)
436{
437 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
438 int i;
439
440 /*
441 * Search backwards on the optimistic assumption that the page being
442 * activated has just been added to this pagevec. Note that only
443 * the local pagevec is examined as a !PageLRU page could be in the
444 * process of being released, reclaimed, migrated or on a remote
445 * pagevec that is currently being drained. Furthermore, marking
446 * a remote pagevec's page PageActive potentially hits a race where
447 * a page is marked PageActive just after it is added to the inactive
448 * list causing accounting errors and BUG_ON checks to trigger.
449 */
450 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
451 struct page *pagevec_page = pvec->pages[i];
452
453 if (pagevec_page == page) {
454 SetPageActive(page);
455 break;
456 }
457 }
458
459 put_cpu_var(lru_add_pvec);
460}
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462/*
463 * Mark a page as having seen activity.
464 *
465 * inactive,unreferenced -> inactive,referenced
466 * inactive,referenced -> active,unreferenced
467 * active,unreferenced -> active,referenced
468 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800469void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700471 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700472 PageReferenced(page)) {
473
474 /*
475 * If the page is on the LRU, queue it for activation via
476 * activate_page_pvecs. Otherwise, assume the page is on a
477 * pagevec, mark it active and it'll be moved to the active
478 * LRU on the next drain.
479 */
480 if (PageLRU(page))
481 activate_page(page);
482 else
483 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 ClearPageReferenced(page);
485 } else if (!PageReferenced(page)) {
486 SetPageReferenced(page);
487 }
488}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489EXPORT_SYMBOL(mark_page_accessed);
490
Robin Dongd741c9c2012-10-08 16:29:05 -0700491/*
Mel Gorman13f7f782013-07-03 15:02:28 -0700492 * Queue the page for addition to the LRU via pagevec. The decision on whether
493 * to add the page to the [in]active [file|anon] list is deferred until the
494 * pagevec is drained. This gives a chance for the caller of __lru_cache_add()
495 * have the page added to the active list using mark_page_accessed().
Robin Dongd741c9c2012-10-08 16:29:05 -0700496 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700497void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
Mel Gorman13f7f782013-07-03 15:02:28 -0700499 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700502 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700503 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700504 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700505 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200507EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700509/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700510 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700511 * @page: the page to be added to the LRU.
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700512 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700513void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700515 VM_BUG_ON(PageActive(page) && PageUnevictable(page));
Mel Gorman13f7f782013-07-03 15:02:28 -0700516 VM_BUG_ON(PageLRU(page));
Mel Gormanc53954a2013-07-03 15:02:34 -0700517 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518}
519
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700520/**
521 * add_page_to_unevictable_list - add a page to the unevictable list
522 * @page: the page to be added to the unevictable list
523 *
524 * Add page directly to its zone's unevictable list. To avoid races with
525 * tasks that might be making the page evictable, through eg. munlock,
526 * munmap or exit, while it's not on the lru, we want to add the page
527 * while it's locked or otherwise "invisible" to other tasks. This is
528 * difficult to do when using the pagevec cache, so bypass that.
529 */
530void add_page_to_unevictable_list(struct page *page)
531{
532 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700533 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700534
535 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700536 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700537 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700538 SetPageUnevictable(page);
539 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700540 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700541 spin_unlock_irq(&zone->lru_lock);
542}
543
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700544/*
Minchan Kim31560182011-03-22 16:32:52 -0700545 * If the page can not be invalidated, it is moved to the
546 * inactive list to speed up its reclaim. It is moved to the
547 * head of the list, rather than the tail, to give the flusher
548 * threads some time to write it out, as this is much more
549 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700550 *
551 * If the page isn't page_mapped and dirty/writeback, the page
552 * could reclaim asap using PG_reclaim.
553 *
554 * 1. active, mapped page -> none
555 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
556 * 3. inactive, mapped page -> none
557 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
558 * 5. inactive, clean -> inactive, tail
559 * 6. Others -> none
560 *
561 * In 4, why it moves inactive's head, the VM expects the page would
562 * be write it out by flusher threads as this is much more effective
563 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700564 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700565static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
566 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700567{
568 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700569 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700570
Minchan Kim278df9f2011-03-22 16:32:54 -0700571 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700572 return;
573
Minchan Kimbad49d92011-05-11 15:13:30 -0700574 if (PageUnevictable(page))
575 return;
576
Minchan Kim31560182011-03-22 16:32:52 -0700577 /* Some processes are using the page */
578 if (page_mapped(page))
579 return;
580
Minchan Kim278df9f2011-03-22 16:32:54 -0700581 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700582 file = page_is_file_cache(page);
583 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700584
585 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700586 ClearPageActive(page);
587 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700588 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700589
Minchan Kim278df9f2011-03-22 16:32:54 -0700590 if (PageWriteback(page) || PageDirty(page)) {
591 /*
592 * PG_reclaim could be raced with end_page_writeback
593 * It can make readahead confusing. But race window
594 * is _really_ small and it's non-critical problem.
595 */
596 SetPageReclaim(page);
597 } else {
598 /*
599 * The page's writeback ends up during pagevec
600 * We moves tha page into tail of inactive.
601 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800602 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700603 __count_vm_event(PGROTATED);
604 }
605
606 if (active)
607 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700608 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700609}
610
Minchan Kim31560182011-03-22 16:32:52 -0700611/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700612 * Drain pages out of the cpu's pagevecs.
613 * Either "cpu" is the current CPU, and preemption has already been
614 * disabled; or "cpu" is being hot-unplugged, and is already dead.
615 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700616void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Mel Gorman13f7f782013-07-03 15:02:28 -0700618 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Mel Gorman13f7f782013-07-03 15:02:28 -0700620 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700621 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700622
623 pvec = &per_cpu(lru_rotate_pvecs, cpu);
624 if (pagevec_count(pvec)) {
625 unsigned long flags;
626
627 /* No harm done if a racing interrupt already did this */
628 local_irq_save(flags);
629 pagevec_move_tail(pvec);
630 local_irq_restore(flags);
631 }
Minchan Kim31560182011-03-22 16:32:52 -0700632
633 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
634 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700635 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700636
637 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700638}
639
640/**
641 * deactivate_page - forcefully deactivate a page
642 * @page: page to deactivate
643 *
644 * This function hints the VM that @page is a good reclaim candidate,
645 * for example if its invalidation fails due to the page being dirty
646 * or under writeback.
647 */
648void deactivate_page(struct page *page)
649{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700650 /*
651 * In a workload with many unevictable page such as mprotect, unevictable
652 * page deactivation for accelerating reclaim is pointless.
653 */
654 if (PageUnevictable(page))
655 return;
656
Minchan Kim31560182011-03-22 16:32:52 -0700657 if (likely(get_page_unless_zero(page))) {
658 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
659
660 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700661 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700662 put_cpu_var(lru_deactivate_pvecs);
663 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800664}
665
666void lru_add_drain(void)
667{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700668 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800669 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
671
David Howellsc4028952006-11-22 14:57:56 +0000672static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800673{
674 lru_add_drain();
675}
676
677/*
678 * Returns 0 for success
679 */
680int lru_add_drain_all(void)
681{
David Howellsc4028952006-11-22 14:57:56 +0000682 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800683}
684
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 * Batched page_cache_release(). Decrement the reference count on all the
687 * passed pages. If it fell to zero then remove the page from the LRU and
688 * free it.
689 *
690 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
691 * for the remainder of the operation.
692 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700693 * The locking in this function is against shrink_inactive_list(): we recheck
694 * the page count inside the lock to see whether shrink_inactive_list()
695 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
696 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 */
698void release_pages(struct page **pages, int nr, int cold)
699{
700 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800701 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700703 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700704 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 for (i = 0; i < nr; i++) {
707 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Nick Piggin8519fb32006-02-07 12:58:52 -0800709 if (unlikely(PageCompound(page))) {
710 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700711 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800712 zone = NULL;
713 }
714 put_compound_page(page);
715 continue;
716 }
717
Nick Pigginb5810032005-10-29 18:16:12 -0700718 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 continue;
720
Nick Piggin46453a62006-03-22 00:07:58 -0800721 if (PageLRU(page)) {
722 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700723
Nick Piggin46453a62006-03-22 00:07:58 -0800724 if (pagezone != zone) {
725 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700726 spin_unlock_irqrestore(&zone->lru_lock,
727 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800728 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700729 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700731
732 lruvec = mem_cgroup_page_lruvec(page, zone);
Nick Piggin725d7042006-09-25 23:30:55 -0700733 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800734 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700735 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
Nick Piggin46453a62006-03-22 00:07:58 -0800737
Mel Gormanc53954a2013-07-03 15:02:34 -0700738 /* Clear Active bit in case of parallel mark_page_accessed */
739 ClearPageActive(page);
740
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800741 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
743 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700744 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800746 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700748EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750/*
751 * The pages which we're about to release may be in the deferred lru-addition
752 * queues. That would prevent them from really being freed right now. That's
753 * OK from a correctness point of view but is inefficient - those pages may be
754 * cache-warm and we want to give them back to the page allocator ASAP.
755 *
756 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
757 * and __pagevec_lru_add_active() call release_pages() directly to avoid
758 * mutual recursion.
759 */
760void __pagevec_release(struct pagevec *pvec)
761{
762 lru_add_drain();
763 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
764 pagevec_reinit(pvec);
765}
Steve French7f285702005-11-01 10:22:55 -0800766EXPORT_SYMBOL(__pagevec_release);
767
Hugh Dickins12d27102012-01-12 17:19:52 -0800768#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800769/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700770void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700771 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800772{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800773 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800774
775 VM_BUG_ON(!PageHead(page));
776 VM_BUG_ON(PageCompound(page_tail));
777 VM_BUG_ON(PageLRU(page_tail));
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700778 VM_BUG_ON(NR_CPUS != 1 &&
779 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800780
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700781 if (!list)
782 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800783
Hugh Dickins12d27102012-01-12 17:19:52 -0800784 if (likely(PageLRU(page)))
785 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700786 else if (list) {
787 /* page reclaim is reclaiming a huge page */
788 get_page(page_tail);
789 list_add_tail(&page_tail->lru, list);
790 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800791 struct list_head *list_head;
792 /*
793 * Head page has not yet been counted, as an hpage,
794 * so we must account for each subpage individually.
795 *
796 * Use the standard add function to put page_tail on the list,
797 * but then correct its position so they all end up in order.
798 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700799 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -0800800 list_head = page_tail->lru.prev;
801 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800802 }
Hugh Dickins75121022012-03-05 14:59:18 -0800803
804 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700805 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800806}
Hugh Dickins12d27102012-01-12 17:19:52 -0800807#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800808
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700809static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
810 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700811{
Mel Gorman13f7f782013-07-03 15:02:28 -0700812 int file = page_is_file_cache(page);
813 int active = PageActive(page);
814 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700815
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700816 VM_BUG_ON(PageLRU(page));
817
818 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700819 add_page_to_lru_list(page, lruvec, lru);
820 update_page_reclaim_stat(lruvec, file, active);
Mel Gormanc6286c92013-07-03 15:02:26 -0700821 trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page));
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700822}
823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 * Add the passed pages to the LRU, then drop the caller's refcount
826 * on them. Reinitialises the caller's pagevec.
827 */
Mel Gormana0b8cab32013-07-03 15:02:32 -0700828void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
Mel Gormana0b8cab32013-07-03 15:02:32 -0700830 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800832EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834/**
835 * pagevec_lookup - gang pagecache lookup
836 * @pvec: Where the resulting pages are placed
837 * @mapping: The address_space to search
838 * @start: The starting page index
839 * @nr_pages: The maximum number of pages
840 *
841 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
842 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
843 * reference against the pages in @pvec.
844 *
845 * The search returns a group of mapping-contiguous pages with ascending
846 * indexes. There may be holes in the indices due to not-present pages.
847 *
848 * pagevec_lookup() returns the number of pages which were found.
849 */
850unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
851 pgoff_t start, unsigned nr_pages)
852{
853 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
854 return pagevec_count(pvec);
855}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100856EXPORT_SYMBOL(pagevec_lookup);
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
859 pgoff_t *index, int tag, unsigned nr_pages)
860{
861 pvec->nr = find_get_pages_tag(mapping, index, tag,
862 nr_pages, pvec->pages);
863 return pagevec_count(pvec);
864}
Steve French7f285702005-11-01 10:22:55 -0800865EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867/*
868 * Perform any setup for the swap system
869 */
870void __init swap_setup(void)
871{
Jan Beulich44813742009-09-21 17:03:05 -0700872 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700873#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -0800874 int i;
875
876 bdi_init(swapper_spaces[0].backing_dev_info);
877 for (i = 0; i < MAX_SWAPFILES; i++) {
878 spin_lock_init(&swapper_spaces[i].tree_lock);
879 INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
880 }
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700881#endif
882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 /* Use a smaller cluster for small-memory machines */
884 if (megs < 16)
885 page_cluster = 2;
886 else
887 page_cluster = 3;
888 /*
889 * Right now other parts of the system means that we
890 * _really_ don't want to cluster much more
891 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}