blob: 4e35f3ff0427107023b95ca2222c9fd663d3eba7 [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>
Khalid Aziz09642082013-09-11 14:22:20 -070034#include <linux/hugetlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Lee Schermerhorn64d65192008-10-18 20:26:52 -070036#include "internal.h"
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038/* How many pages do we try to swap or page in/out together? */
39int page_cluster;
40
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -070041static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
Vegard Nossumf84f95042008-07-23 21:28:14 -070042static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -070043static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070044
Adrian Bunkb2213852006-09-25 23:31:02 -070045/*
46 * This path almost never happens for VM activity - pages are normally
47 * freed via pagevecs. But it gets used by networking.
48 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080049static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070050{
51 if (PageLRU(page)) {
Adrian Bunkb2213852006-09-25 23:31:02 -070052 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070053 struct lruvec *lruvec;
54 unsigned long flags;
Adrian Bunkb2213852006-09-25 23:31:02 -070055
56 spin_lock_irqsave(&zone->lru_lock, flags);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070057 lruvec = mem_cgroup_page_lruvec(page, zone);
Adrian Bunkb2213852006-09-25 23:31:02 -070058 VM_BUG_ON(!PageLRU(page));
59 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070060 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070061 spin_unlock_irqrestore(&zone->lru_lock, flags);
62 }
Andrea Arcangeli91807062011-01-13 15:46:32 -080063}
64
65static void __put_single_page(struct page *page)
66{
67 __page_cache_release(page);
Li Hongfc916682010-03-05 13:41:54 -080068 free_hot_cold_page(page, 0);
Adrian Bunkb2213852006-09-25 23:31:02 -070069}
70
Andrea Arcangeli91807062011-01-13 15:46:32 -080071static void __put_compound_page(struct page *page)
72{
73 compound_page_dtor *dtor;
74
75 __page_cache_release(page);
76 dtor = get_compound_page_dtor(page);
77 (*dtor)(page);
78}
79
Nick Piggin8519fb32006-02-07 12:58:52 -080080static void put_compound_page(struct page *page)
81{
Andrea Arcangeli91807062011-01-13 15:46:32 -080082 if (unlikely(PageTail(page))) {
83 /* __split_huge_page_refcount can run under us */
David Rientjesdef52ac2014-03-03 15:38:18 -080084 struct page *page_head = compound_head(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -070085
86 if (likely(page != page_head &&
87 get_page_unless_zero(page_head))) {
Andrea Arcangeli91807062011-01-13 15:46:32 -080088 unsigned long flags;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -070089
90 /*
91 * THP can not break up slab pages so avoid taking
92 * compound_lock(). Slab performs non-atomic bit ops
93 * on page->flags for better performance. In particular
94 * slab_unlock() in slub used to be a hot path. It is
95 * still hot on arches that do not support
96 * this_cpu_cmpxchg_double().
97 */
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -080098 if (PageSlab(page_head) || PageHeadHuge(page_head)) {
99 if (likely(PageTail(page))) {
100 /*
101 * __split_huge_page_refcount
102 * cannot race here.
103 */
104 VM_BUG_ON(!PageHead(page_head));
105 atomic_dec(&page->_mapcount);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700106 if (put_page_testzero(page_head))
107 VM_BUG_ON(1);
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800108 if (put_page_testzero(page_head))
109 __put_compound_page(page_head);
110 return;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700111 } else
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800112 /*
113 * __split_huge_page_refcount
114 * run before us, "page" was a
115 * THP tail. The split
116 * page_head has been freed
117 * and reallocated as slab or
118 * hugetlbfs page of smaller
119 * order (only possible if
120 * reallocated as slab on
121 * x86).
122 */
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700123 goto skip_lock;
124 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800125 /*
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700126 * page_head wasn't a dangling pointer but it
127 * may not be a head page anymore by the time
128 * we obtain the lock. That is ok as long as it
129 * can't be freed from under us.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800130 */
Andrea Arcangeli91807062011-01-13 15:46:32 -0800131 flags = compound_lock_irqsave(page_head);
132 if (unlikely(!PageTail(page))) {
133 /* __split_huge_page_refcount run before us */
134 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700135skip_lock:
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800136 if (put_page_testzero(page_head)) {
137 /*
138 * The head page may have been
139 * freed and reallocated as a
140 * compound page of smaller
141 * order and then freed again.
142 * All we know is that it
143 * cannot have become: a THP
144 * page, a compound page of
145 * higher order, a tail page.
146 * That is because we still
147 * hold the refcount of the
148 * split THP tail and
149 * page_head was the THP head
150 * before the split.
151 */
152 if (PageHead(page_head))
153 __put_compound_page(page_head);
154 else
155 __put_single_page(page_head);
156 }
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700157out_put_single:
Andrea Arcangeli91807062011-01-13 15:46:32 -0800158 if (put_page_testzero(page))
159 __put_single_page(page);
160 return;
161 }
162 VM_BUG_ON(page_head != page->first_page);
163 /*
164 * We can release the refcount taken by
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700165 * get_page_unless_zero() now that
166 * __split_huge_page_refcount() is blocked on
167 * the compound_lock.
Andrea Arcangeli91807062011-01-13 15:46:32 -0800168 */
169 if (put_page_testzero(page_head))
170 VM_BUG_ON(1);
171 /* __split_huge_page_refcount will wait now */
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700172 VM_BUG_ON(page_mapcount(page) <= 0);
173 atomic_dec(&page->_mapcount);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800174 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700175 VM_BUG_ON(atomic_read(&page->_count) != 0);
Andrea Arcangeli91807062011-01-13 15:46:32 -0800176 compound_unlock_irqrestore(page_head, flags);
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700177
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800178 if (put_page_testzero(page_head)) {
179 if (PageHead(page_head))
180 __put_compound_page(page_head);
181 else
182 __put_single_page(page_head);
183 }
Andrea Arcangeli91807062011-01-13 15:46:32 -0800184 } else {
185 /* page_head is a dangling pointer */
186 VM_BUG_ON(PageTail(page));
187 goto out_put_single;
188 }
189 } else if (put_page_testzero(page)) {
190 if (PageHead(page))
191 __put_compound_page(page);
192 else
193 __put_single_page(page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800194 }
195}
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197void put_page(struct page *page)
198{
Nick Piggin8519fb32006-02-07 12:58:52 -0800199 if (unlikely(PageCompound(page)))
200 put_compound_page(page);
201 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800202 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700206/*
207 * This function is exported but must not be called by anything other
208 * than get_page(). It implements the slow path of get_page().
209 */
210bool __get_page_tail(struct page *page)
211{
212 /*
213 * This takes care of get_page() if run on a tail page
214 * returned by one of the get_user_pages/follow_page variants.
215 * get_user_pages/follow_page itself doesn't need the compound
216 * lock because it runs __get_page_tail_foll() under the
217 * proper PT lock that already serializes against
218 * split_huge_page().
219 */
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800220 unsigned long flags;
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700221 bool got = false;
David Rientjesdef52ac2014-03-03 15:38:18 -0800222 struct page *page_head = compound_head(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700223
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800224 if (likely(page != page_head && get_page_unless_zero(page_head))) {
225 /* Ref to put_compound_page() comment. */
226 if (PageSlab(page_head) || PageHeadHuge(page_head)) {
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700227 if (likely(PageTail(page))) {
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800228 /*
229 * This is a hugetlbfs page or a slab
230 * page. __split_huge_page_refcount
231 * cannot race here.
232 */
233 VM_BUG_ON(!PageHead(page_head));
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700234 __get_page_tail_foll(page, false);
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800235 return true;
236 } else {
237 /*
238 * __split_huge_page_refcount run
239 * before us, "page" was a THP
240 * tail. The split page_head has been
241 * freed and reallocated as slab or
242 * hugetlbfs page of smaller order
243 * (only possible if reallocated as
244 * slab on x86).
245 */
Khalid Aziz09642082013-09-11 14:22:20 -0700246 put_page(page_head);
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800247 return false;
248 }
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700249 }
Andrea Arcangeli17b6ada2013-11-21 14:32:02 -0800250
251 /*
252 * page_head wasn't a dangling pointer but it
253 * may not be a head page anymore by the time
254 * we obtain the lock. That is ok as long as it
255 * can't be freed from under us.
256 */
257 flags = compound_lock_irqsave(page_head);
258 /* here __split_huge_page_refcount won't run anymore */
259 if (likely(PageTail(page))) {
260 __get_page_tail_foll(page, false);
261 got = true;
262 }
263 compound_unlock_irqrestore(page_head, flags);
264 if (unlikely(!got))
265 put_page(page_head);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700266 }
267 return got;
268}
269EXPORT_SYMBOL(__get_page_tail);
270
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700271/**
Randy Dunlap76824862008-03-19 17:00:40 -0700272 * put_pages_list() - release a list of pages
273 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700274 *
275 * Release a list of pages which are strung together on page.lru. Currently
276 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700277 */
278void put_pages_list(struct list_head *pages)
279{
280 while (!list_empty(pages)) {
281 struct page *victim;
282
283 victim = list_entry(pages->prev, struct page, lru);
284 list_del(&victim->lru);
285 page_cache_release(victim);
286 }
287}
288EXPORT_SYMBOL(put_pages_list);
289
Mel Gorman18022c52012-07-31 16:44:51 -0700290/*
291 * get_kernel_pages() - pin kernel pages in memory
292 * @kiov: An array of struct kvec structures
293 * @nr_segs: number of segments to pin
294 * @write: pinning for read/write, currently ignored
295 * @pages: array that receives pointers to the pages pinned.
296 * Should be at least nr_segs long.
297 *
298 * Returns number of pages pinned. This may be fewer than the number
299 * requested. If nr_pages is 0 or negative, returns 0. If no pages
300 * were pinned, returns -errno. Each page returned must be released
301 * with a put_page() call when it is finished with.
302 */
303int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
304 struct page **pages)
305{
306 int seg;
307
308 for (seg = 0; seg < nr_segs; seg++) {
309 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
310 return seg;
311
Mel Gorman5a178112012-07-31 16:45:02 -0700312 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700313 page_cache_get(pages[seg]);
314 }
315
316 return seg;
317}
318EXPORT_SYMBOL_GPL(get_kernel_pages);
319
320/*
321 * get_kernel_page() - pin a kernel page in memory
322 * @start: starting kernel address
323 * @write: pinning for read/write, currently ignored
324 * @pages: array that receives pointer to the page pinned.
325 * Must be at least nr_segs long.
326 *
327 * Returns 1 if page is pinned. If the page was not pinned, returns
328 * -errno. The page returned must be released with a put_page() call
329 * when it is finished with.
330 */
331int get_kernel_page(unsigned long start, int write, struct page **pages)
332{
333 const struct kvec kiov = {
334 .iov_base = (void *)start,
335 .iov_len = PAGE_SIZE
336 };
337
338 return get_kernel_pages(&kiov, 1, write, pages);
339}
340EXPORT_SYMBOL_GPL(get_kernel_page);
341
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700342static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700343 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
344 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700345{
346 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700347 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700348 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700349 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700350
351 for (i = 0; i < pagevec_count(pvec); i++) {
352 struct page *page = pvec->pages[i];
353 struct zone *pagezone = page_zone(page);
354
355 if (pagezone != zone) {
356 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700357 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700358 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700359 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700360 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700361
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700362 lruvec = mem_cgroup_page_lruvec(page, zone);
363 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700364 }
365 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700366 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800367 release_pages(pvec->pages, pvec->nr, pvec->cold);
368 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800369}
370
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700371static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
372 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700373{
374 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700375
376 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
377 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800378 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700379 (*pgmoved)++;
380 }
381}
382
383/*
384 * pagevec_move_tail() must be called with IRQ disabled.
385 * Otherwise this may cause nasty races.
386 */
387static void pagevec_move_tail(struct pagevec *pvec)
388{
389 int pgmoved = 0;
390
391 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
392 __count_vm_events(PGROTATED, pgmoved);
393}
394
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700395/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 * Writeback is about to end against a page which has been marked for immediate
397 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700398 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700400void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700402 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700403 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700404 struct pagevec *pvec;
405 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700407 page_cache_get(page);
408 local_irq_save(flags);
409 pvec = &__get_cpu_var(lru_rotate_pvecs);
410 if (!pagevec_add(pvec, page))
411 pagevec_move_tail(pvec);
412 local_irq_restore(flags);
413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700416static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800417 int file, int rotated)
418{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700419 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800420
421 reclaim_stat->recent_scanned[file]++;
422 if (rotated)
423 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800424}
425
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700426static void __activate_page(struct page *page, struct lruvec *lruvec,
427 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
Linus Torvalds7a608572011-01-17 14:42:19 -0800429 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
430 int file = page_is_file_cache(page);
431 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800432
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700433 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800434 SetPageActive(page);
435 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700436 add_page_to_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800437
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700438 __count_vm_event(PGACTIVATE);
439 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800440 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700441}
442
443#ifdef CONFIG_SMP
444static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
445
446static void activate_page_drain(int cpu)
447{
448 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
449
450 if (pagevec_count(pvec))
451 pagevec_lru_move_fn(pvec, __activate_page, NULL);
452}
453
454void activate_page(struct page *page)
455{
456 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
457 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
458
459 page_cache_get(page);
460 if (!pagevec_add(pvec, page))
461 pagevec_lru_move_fn(pvec, __activate_page, NULL);
462 put_cpu_var(activate_page_pvecs);
463 }
464}
465
466#else
467static inline void activate_page_drain(int cpu)
468{
469}
470
471void activate_page(struct page *page)
472{
473 struct zone *zone = page_zone(page);
474
475 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700476 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 spin_unlock_irq(&zone->lru_lock);
478}
Shaohua Lieb709b02011-05-24 17:12:55 -0700479#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481/*
482 * Mark a page as having seen activity.
483 *
484 * inactive,unreferenced -> inactive,referenced
485 * inactive,referenced -> active,unreferenced
486 * active,unreferenced -> active,referenced
487 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800488void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700490 if (!PageActive(page) && !PageUnevictable(page) &&
491 PageReferenced(page) && PageLRU(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 activate_page(page);
493 ClearPageReferenced(page);
494 } else if (!PageReferenced(page)) {
495 SetPageReferenced(page);
496 }
497}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498EXPORT_SYMBOL(mark_page_accessed);
499
Robin Dongd741c9c2012-10-08 16:29:05 -0700500/*
501 * Order of operations is important: flush the pagevec when it's already
502 * full, not when adding the last page, to make sure that last page is
503 * not added to the LRU directly when passed to this function. Because
504 * mark_page_accessed() (called after this when writing) only activates
505 * pages that are on the LRU, linear writes in subpage chunks would see
506 * every PAGEVEC_SIZE page activated, which is unexpected.
507 */
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700508void __lru_cache_add(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700510 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
512 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700513 if (!pagevec_space(pvec))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800514 __pagevec_lru_add(pvec, lru);
Robin Dongd741c9c2012-10-08 16:29:05 -0700515 pagevec_add(pvec, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 put_cpu_var(lru_add_pvecs);
517}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200518EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700520/**
521 * lru_cache_add_lru - add a page to a page list
522 * @page: the page to be added to the LRU.
523 * @lru: the LRU list to which the page is added.
524 */
525void lru_cache_add_lru(struct page *page, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700527 if (PageActive(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700528 VM_BUG_ON(PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700529 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700530 } else if (PageUnevictable(page)) {
531 VM_BUG_ON(PageActive(page));
532 ClearPageUnevictable(page);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700535 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700536 __lru_cache_add(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700539/**
540 * add_page_to_unevictable_list - add a page to the unevictable list
541 * @page: the page to be added to the unevictable list
542 *
543 * Add page directly to its zone's unevictable list. To avoid races with
544 * tasks that might be making the page evictable, through eg. munlock,
545 * munmap or exit, while it's not on the lru, we want to add the page
546 * while it's locked or otherwise "invisible" to other tasks. This is
547 * difficult to do when using the pagevec cache, so bypass that.
548 */
549void add_page_to_unevictable_list(struct page *page)
550{
551 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700552 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700553
554 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700555 lruvec = mem_cgroup_page_lruvec(page, zone);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700556 SetPageUnevictable(page);
557 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700558 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700559 spin_unlock_irq(&zone->lru_lock);
560}
561
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700562/*
Minchan Kim31560182011-03-22 16:32:52 -0700563 * If the page can not be invalidated, it is moved to the
564 * inactive list to speed up its reclaim. It is moved to the
565 * head of the list, rather than the tail, to give the flusher
566 * threads some time to write it out, as this is much more
567 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700568 *
569 * If the page isn't page_mapped and dirty/writeback, the page
570 * could reclaim asap using PG_reclaim.
571 *
572 * 1. active, mapped page -> none
573 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
574 * 3. inactive, mapped page -> none
575 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
576 * 5. inactive, clean -> inactive, tail
577 * 6. Others -> none
578 *
579 * In 4, why it moves inactive's head, the VM expects the page would
580 * be write it out by flusher threads as this is much more effective
581 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700582 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700583static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
584 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700585{
586 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700587 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700588
Minchan Kim278df9f2011-03-22 16:32:54 -0700589 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700590 return;
591
Minchan Kimbad49d92011-05-11 15:13:30 -0700592 if (PageUnevictable(page))
593 return;
594
Minchan Kim31560182011-03-22 16:32:52 -0700595 /* Some processes are using the page */
596 if (page_mapped(page))
597 return;
598
Minchan Kim278df9f2011-03-22 16:32:54 -0700599 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700600 file = page_is_file_cache(page);
601 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700602
603 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700604 ClearPageActive(page);
605 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700606 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700607
Minchan Kim278df9f2011-03-22 16:32:54 -0700608 if (PageWriteback(page) || PageDirty(page)) {
609 /*
610 * PG_reclaim could be raced with end_page_writeback
611 * It can make readahead confusing. But race window
612 * is _really_ small and it's non-critical problem.
613 */
614 SetPageReclaim(page);
615 } else {
616 /*
617 * The page's writeback ends up during pagevec
618 * We moves tha page into tail of inactive.
619 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800620 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700621 __count_vm_event(PGROTATED);
622 }
623
624 if (active)
625 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700626 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700627}
628
Minchan Kim31560182011-03-22 16:32:52 -0700629/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700630 * Drain pages out of the cpu's pagevecs.
631 * Either "cpu" is the current CPU, and preemption has already been
632 * disabled; or "cpu" is being hot-unplugged, and is already dead.
633 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700634void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700636 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700637 struct pagevec *pvec;
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700638 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700640 for_each_lru(lru) {
641 pvec = &pvecs[lru - LRU_BASE];
642 if (pagevec_count(pvec))
Hugh Dickins5095ae832012-01-12 17:19:58 -0800643 __pagevec_lru_add(pvec, lru);
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700644 }
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700645
646 pvec = &per_cpu(lru_rotate_pvecs, cpu);
647 if (pagevec_count(pvec)) {
648 unsigned long flags;
649
650 /* No harm done if a racing interrupt already did this */
651 local_irq_save(flags);
652 pagevec_move_tail(pvec);
653 local_irq_restore(flags);
654 }
Minchan Kim31560182011-03-22 16:32:52 -0700655
656 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
657 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700658 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700659
660 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700661}
662
663/**
664 * deactivate_page - forcefully deactivate a page
665 * @page: page to deactivate
666 *
667 * This function hints the VM that @page is a good reclaim candidate,
668 * for example if its invalidation fails due to the page being dirty
669 * or under writeback.
670 */
671void deactivate_page(struct page *page)
672{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700673 /*
674 * In a workload with many unevictable page such as mprotect, unevictable
675 * page deactivation for accelerating reclaim is pointless.
676 */
677 if (PageUnevictable(page))
678 return;
679
Minchan Kim31560182011-03-22 16:32:52 -0700680 if (likely(get_page_unless_zero(page))) {
681 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
682
683 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700684 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700685 put_cpu_var(lru_deactivate_pvecs);
686 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800687}
688
689void lru_add_drain(void)
690{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700691 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800692 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
David Howellsc4028952006-11-22 14:57:56 +0000695static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800696{
697 lru_add_drain();
698}
699
700/*
701 * Returns 0 for success
702 */
703int lru_add_drain_all(void)
704{
David Howellsc4028952006-11-22 14:57:56 +0000705 return schedule_on_each_cpu(lru_add_drain_per_cpu);
Nick Piggin053837f2006-01-18 17:42:27 -0800706}
707
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 * Batched page_cache_release(). Decrement the reference count on all the
710 * passed pages. If it fell to zero then remove the page from the LRU and
711 * free it.
712 *
713 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
714 * for the remainder of the operation.
715 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700716 * The locking in this function is against shrink_inactive_list(): we recheck
717 * the page count inside the lock to see whether shrink_inactive_list()
718 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
719 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 */
721void release_pages(struct page **pages, int nr, int cold)
722{
723 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800724 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700726 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700727 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 for (i = 0; i < nr; i++) {
730 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Nick Piggin8519fb32006-02-07 12:58:52 -0800732 if (unlikely(PageCompound(page))) {
733 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700734 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800735 zone = NULL;
736 }
737 put_compound_page(page);
738 continue;
739 }
740
Nick Pigginb5810032005-10-29 18:16:12 -0700741 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 continue;
743
Nick Piggin46453a62006-03-22 00:07:58 -0800744 if (PageLRU(page)) {
745 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700746
Nick Piggin46453a62006-03-22 00:07:58 -0800747 if (pagezone != zone) {
748 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700749 spin_unlock_irqrestore(&zone->lru_lock,
750 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800751 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700752 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700754
755 lruvec = mem_cgroup_page_lruvec(page, zone);
Nick Piggin725d7042006-09-25 23:30:55 -0700756 VM_BUG_ON(!PageLRU(page));
Nick Piggin67453912006-03-22 00:08:00 -0800757 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700758 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
Nick Piggin46453a62006-03-22 00:07:58 -0800760
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800761 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700764 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800766 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700768EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
770/*
771 * The pages which we're about to release may be in the deferred lru-addition
772 * queues. That would prevent them from really being freed right now. That's
773 * OK from a correctness point of view but is inefficient - those pages may be
774 * cache-warm and we want to give them back to the page allocator ASAP.
775 *
776 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
777 * and __pagevec_lru_add_active() call release_pages() directly to avoid
778 * mutual recursion.
779 */
780void __pagevec_release(struct pagevec *pvec)
781{
782 lru_add_drain();
783 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
784 pagevec_reinit(pvec);
785}
Steve French7f285702005-11-01 10:22:55 -0800786EXPORT_SYMBOL(__pagevec_release);
787
Hugh Dickins12d27102012-01-12 17:19:52 -0800788#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800789/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700790void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700791 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800792{
Hugh Dickins75121022012-03-05 14:59:18 -0800793 int uninitialized_var(active);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800794 enum lru_list lru;
795 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800796
797 VM_BUG_ON(!PageHead(page));
798 VM_BUG_ON(PageCompound(page_tail));
799 VM_BUG_ON(PageLRU(page_tail));
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700800 VM_BUG_ON(NR_CPUS != 1 &&
801 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800802
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700803 if (!list)
804 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800805
Hugh Dickins39b5f292012-10-08 16:33:18 -0700806 if (page_evictable(page_tail)) {
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800807 if (PageActive(page)) {
808 SetPageActive(page_tail);
809 active = 1;
810 lru = LRU_ACTIVE_ANON;
811 } else {
812 active = 0;
813 lru = LRU_INACTIVE_ANON;
814 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800815 } else {
816 SetPageUnevictable(page_tail);
Hugh Dickins12d27102012-01-12 17:19:52 -0800817 lru = LRU_UNEVICTABLE;
818 }
819
820 if (likely(PageLRU(page)))
821 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700822 else if (list) {
823 /* page reclaim is reclaiming a huge page */
824 get_page(page_tail);
825 list_add_tail(&page_tail->lru, list);
826 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800827 struct list_head *list_head;
828 /*
829 * Head page has not yet been counted, as an hpage,
830 * so we must account for each subpage individually.
831 *
832 * Use the standard add function to put page_tail on the list,
833 * but then correct its position so they all end up in order.
834 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700835 add_page_to_lru_list(page_tail, lruvec, lru);
Hugh Dickins12d27102012-01-12 17:19:52 -0800836 list_head = page_tail->lru.prev;
837 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800838 }
Hugh Dickins75121022012-03-05 14:59:18 -0800839
840 if (!PageUnevictable(page))
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700841 update_page_reclaim_stat(lruvec, file, active);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800842}
Hugh Dickins12d27102012-01-12 17:19:52 -0800843#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800844
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700845static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
846 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700847{
848 enum lru_list lru = (enum lru_list)arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700849 int file = is_file_lru(lru);
850 int active = is_active_lru(lru);
851
852 VM_BUG_ON(PageActive(page));
853 VM_BUG_ON(PageUnevictable(page));
854 VM_BUG_ON(PageLRU(page));
855
856 SetPageLRU(page);
857 if (active)
858 SetPageActive(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700859 add_page_to_lru_list(page, lruvec, lru);
860 update_page_reclaim_stat(lruvec, file, active);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700861}
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 * Add the passed pages to the LRU, then drop the caller's refcount
865 * on them. Reinitialises the caller's pagevec.
866 */
Hugh Dickins5095ae832012-01-12 17:19:58 -0800867void __pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700869 VM_BUG_ON(is_unevictable_lru(lru));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
Hugh Dickins5095ae832012-01-12 17:19:58 -0800871 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, (void *)lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800873EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875/**
876 * pagevec_lookup - gang pagecache lookup
877 * @pvec: Where the resulting pages are placed
878 * @mapping: The address_space to search
879 * @start: The starting page index
880 * @nr_pages: The maximum number of pages
881 *
882 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
883 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
884 * reference against the pages in @pvec.
885 *
886 * The search returns a group of mapping-contiguous pages with ascending
887 * indexes. There may be holes in the indices due to not-present pages.
888 *
889 * pagevec_lookup() returns the number of pages which were found.
890 */
891unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
892 pgoff_t start, unsigned nr_pages)
893{
894 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
895 return pagevec_count(pvec);
896}
Christoph Hellwig78539fd2006-01-11 20:47:41 +1100897EXPORT_SYMBOL(pagevec_lookup);
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
900 pgoff_t *index, int tag, unsigned nr_pages)
901{
902 pvec->nr = find_get_pages_tag(mapping, index, tag,
903 nr_pages, pvec->pages);
904 return pagevec_count(pvec);
905}
Steve French7f285702005-11-01 10:22:55 -0800906EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908/*
909 * Perform any setup for the swap system
910 */
911void __init swap_setup(void)
912{
Jan Beulich44813742009-09-21 17:03:05 -0700913 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700914#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -0800915 int i;
916
917 bdi_init(swapper_spaces[0].backing_dev_info);
918 for (i = 0; i < MAX_SWAPFILES; i++) {
919 spin_lock_init(&swapper_spaces[i].tree_lock);
920 INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
921 }
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700922#endif
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 /* Use a smaller cluster for small-memory machines */
925 if (megs < 16)
926 page_cluster = 2;
927 else
928 page_cluster = 3;
929 /*
930 * Right now other parts of the system means that we
931 * _really_ don't want to cluster much more
932 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933}