blob: 8a12b33936b45c1c49a312ea04e08338472b98fe [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);
Sasha Levin309381fea2014-01-23 15:52:54 -080060 VM_BUG_ON_PAGE(!PageLRU(page), page);
Adrian Bunkb2213852006-09-25 23:31:02 -070061 __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 }
Johannes Weiner0a31bc92014-08-08 14:19:22 -070065 mem_cgroup_uncharge(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080066}
67
68static void __put_single_page(struct page *page)
69{
70 __page_cache_release(page);
Mel Gormanb745bc82014-06-04 16:10:22 -070071 free_hot_cold_page(page, false);
Adrian Bunkb2213852006-09-25 23:31:02 -070072}
73
Andrea Arcangeli91807062011-01-13 15:46:32 -080074static void __put_compound_page(struct page *page)
75{
76 compound_page_dtor *dtor;
77
78 __page_cache_release(page);
79 dtor = get_compound_page_dtor(page);
80 (*dtor)(page);
81}
82
Jianyu Zhanc747ce72014-06-04 16:07:59 -070083/**
84 * Two special cases here: we could avoid taking compound_lock_irqsave
85 * and could skip the tail refcounting(in _mapcount).
86 *
87 * 1. Hugetlbfs page:
88 *
89 * PageHeadHuge will remain true until the compound page
90 * is released and enters the buddy allocator, and it could
91 * not be split by __split_huge_page_refcount().
92 *
93 * So if we see PageHeadHuge set, and we have the tail page pin,
94 * then we could safely put head page.
95 *
96 * 2. Slab THP page:
97 *
98 * PG_slab is cleared before the slab frees the head page, and
99 * tail pin cannot be the last reference left on the head page,
100 * because the slab code is free to reuse the compound page
101 * after a kfree/kmem_cache_free without having to check if
102 * there's any tail pin left. In turn all tail pinsmust be always
103 * released while the head is still pinned by the slab code
104 * and so we know PG_slab will be still set too.
105 *
106 * So if we see PageSlab set, and we have the tail page pin,
107 * then we could safely put head page.
108 */
109static __always_inline
110void put_unrefcounted_compound_page(struct page *page_head, struct page *page)
111{
112 /*
113 * If @page is a THP tail, we must read the tail page
114 * flags after the head page flags. The
115 * __split_huge_page_refcount side enforces write memory barriers
116 * between clearing PageTail and before the head page
117 * can be freed and reallocated.
118 */
119 smp_rmb();
120 if (likely(PageTail(page))) {
121 /*
122 * __split_huge_page_refcount cannot race
123 * here, see the comment above this function.
124 */
125 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
126 VM_BUG_ON_PAGE(page_mapcount(page) != 0, page);
127 if (put_page_testzero(page_head)) {
128 /*
129 * If this is the tail of a slab THP page,
130 * the tail pin must not be the last reference
131 * held on the page, because the PG_slab cannot
132 * be cleared before all tail pins (which skips
133 * the _mapcount tail refcounting) have been
134 * released.
135 *
136 * If this is the tail of a hugetlbfs page,
137 * the tail pin may be the last reference on
138 * the page instead, because PageHeadHuge will
139 * not go away until the compound page enters
140 * the buddy allocator.
141 */
142 VM_BUG_ON_PAGE(PageSlab(page_head), page_head);
143 __put_compound_page(page_head);
144 }
145 } else
146 /*
147 * __split_huge_page_refcount run before us,
148 * @page was a THP tail. The split @page_head
149 * has been freed and reallocated as slab or
150 * hugetlbfs page of smaller order (only
151 * possible if reallocated as slab on x86).
152 */
153 if (put_page_testzero(page))
154 __put_single_page(page);
155}
156
157static __always_inline
158void put_refcounted_compound_page(struct page *page_head, struct page *page)
159{
160 if (likely(page != page_head && get_page_unless_zero(page_head))) {
161 unsigned long flags;
162
163 /*
164 * @page_head wasn't a dangling pointer but it may not
165 * be a head page anymore by the time we obtain the
166 * lock. That is ok as long as it can't be freed from
167 * under us.
168 */
169 flags = compound_lock_irqsave(page_head);
170 if (unlikely(!PageTail(page))) {
171 /* __split_huge_page_refcount run before us */
172 compound_unlock_irqrestore(page_head, flags);
173 if (put_page_testzero(page_head)) {
174 /*
175 * The @page_head may have been freed
176 * and reallocated as a compound page
177 * of smaller order and then freed
178 * again. All we know is that it
179 * cannot have become: a THP page, a
180 * compound page of higher order, a
181 * tail page. That is because we
182 * still hold the refcount of the
183 * split THP tail and page_head was
184 * the THP head before the split.
185 */
186 if (PageHead(page_head))
187 __put_compound_page(page_head);
188 else
189 __put_single_page(page_head);
190 }
191out_put_single:
192 if (put_page_testzero(page))
193 __put_single_page(page);
194 return;
195 }
196 VM_BUG_ON_PAGE(page_head != page->first_page, page);
197 /*
198 * We can release the refcount taken by
199 * get_page_unless_zero() now that
200 * __split_huge_page_refcount() is blocked on the
201 * compound_lock.
202 */
203 if (put_page_testzero(page_head))
204 VM_BUG_ON_PAGE(1, page_head);
205 /* __split_huge_page_refcount will wait now */
206 VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page);
207 atomic_dec(&page->_mapcount);
208 VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head);
209 VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page);
210 compound_unlock_irqrestore(page_head, flags);
211
212 if (put_page_testzero(page_head)) {
213 if (PageHead(page_head))
214 __put_compound_page(page_head);
215 else
216 __put_single_page(page_head);
217 }
218 } else {
219 /* @page_head is a dangling pointer */
220 VM_BUG_ON_PAGE(PageTail(page), page);
221 goto out_put_single;
222 }
223}
224
Nick Piggin8519fb32006-02-07 12:58:52 -0800225static void put_compound_page(struct page *page)
226{
Andrew Morton26296ad2014-01-21 15:48:59 -0800227 struct page *page_head;
228
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700229 /*
230 * We see the PageCompound set and PageTail not set, so @page maybe:
231 * 1. hugetlbfs head page, or
232 * 2. THP head page.
233 */
Andrew Morton26296ad2014-01-21 15:48:59 -0800234 if (likely(!PageTail(page))) {
235 if (put_page_testzero(page)) {
236 /*
237 * By the time all refcounts have been released
238 * split_huge_page cannot run anymore from under us.
239 */
240 if (PageHead(page))
241 __put_compound_page(page);
242 else
243 __put_single_page(page);
244 }
245 return;
246 }
247
Andrew Morton26296ad2014-01-21 15:48:59 -0800248 /*
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700249 * We see the PageCompound set and PageTail set, so @page maybe:
250 * 1. a tail hugetlbfs page, or
251 * 2. a tail THP page, or
252 * 3. a split THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800253 *
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700254 * Case 3 is possible, as we may race with
255 * __split_huge_page_refcount tearing down a THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800256 */
Jianyu Zhand2ee40e2014-06-04 16:08:02 -0700257 page_head = compound_head_by_tail(page);
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700258 if (!__compound_tail_refcounted(page_head))
259 put_unrefcounted_compound_page(page_head, page);
260 else
261 put_refcounted_compound_page(page_head, page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800262}
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264void put_page(struct page *page)
265{
Nick Piggin8519fb32006-02-07 12:58:52 -0800266 if (unlikely(PageCompound(page)))
267 put_compound_page(page);
268 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800269 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270}
271EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700273/*
274 * This function is exported but must not be called by anything other
275 * than get_page(). It implements the slow path of get_page().
276 */
277bool __get_page_tail(struct page *page)
278{
279 /*
280 * This takes care of get_page() if run on a tail page
281 * returned by one of the get_user_pages/follow_page variants.
282 * get_user_pages/follow_page itself doesn't need the compound
283 * lock because it runs __get_page_tail_foll() under the
284 * proper PT lock that already serializes against
285 * split_huge_page().
286 */
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800287 unsigned long flags;
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800288 bool got;
David Rientjes668f9abb2014-03-03 15:38:18 -0800289 struct page *page_head = compound_head(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700290
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800291 /* Ref to put_compound_page() comment. */
Andrea Arcangeli3bfcd132014-01-21 15:48:56 -0800292 if (!__compound_tail_refcounted(page_head)) {
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800293 smp_rmb();
294 if (likely(PageTail(page))) {
295 /*
296 * This is a hugetlbfs page or a slab
297 * page. __split_huge_page_refcount
298 * cannot race here.
299 */
Sasha Levin309381fea2014-01-23 15:52:54 -0800300 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800301 __get_page_tail_foll(page, true);
302 return true;
303 } else {
304 /*
305 * __split_huge_page_refcount run
306 * before us, "page" was a THP
307 * tail. The split page_head has been
308 * freed and reallocated as slab or
309 * hugetlbfs page of smaller order
310 * (only possible if reallocated as
311 * slab on x86).
312 */
313 return false;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700314 }
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800315 }
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800316
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800317 got = false;
318 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800319 /*
320 * page_head wasn't a dangling pointer but it
321 * may not be a head page anymore by the time
322 * we obtain the lock. That is ok as long as it
323 * can't be freed from under us.
324 */
325 flags = compound_lock_irqsave(page_head);
326 /* here __split_huge_page_refcount won't run anymore */
327 if (likely(PageTail(page))) {
328 __get_page_tail_foll(page, false);
329 got = true;
330 }
331 compound_unlock_irqrestore(page_head, flags);
332 if (unlikely(!got))
333 put_page(page_head);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700334 }
335 return got;
336}
337EXPORT_SYMBOL(__get_page_tail);
338
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700339/**
Randy Dunlap76824862008-03-19 17:00:40 -0700340 * put_pages_list() - release a list of pages
341 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700342 *
343 * Release a list of pages which are strung together on page.lru. Currently
344 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700345 */
346void put_pages_list(struct list_head *pages)
347{
348 while (!list_empty(pages)) {
349 struct page *victim;
350
351 victim = list_entry(pages->prev, struct page, lru);
352 list_del(&victim->lru);
353 page_cache_release(victim);
354 }
355}
356EXPORT_SYMBOL(put_pages_list);
357
Mel Gorman18022c52012-07-31 16:44:51 -0700358/*
359 * get_kernel_pages() - pin kernel pages in memory
360 * @kiov: An array of struct kvec structures
361 * @nr_segs: number of segments to pin
362 * @write: pinning for read/write, currently ignored
363 * @pages: array that receives pointers to the pages pinned.
364 * Should be at least nr_segs long.
365 *
366 * Returns number of pages pinned. This may be fewer than the number
367 * requested. If nr_pages is 0 or negative, returns 0. If no pages
368 * were pinned, returns -errno. Each page returned must be released
369 * with a put_page() call when it is finished with.
370 */
371int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
372 struct page **pages)
373{
374 int seg;
375
376 for (seg = 0; seg < nr_segs; seg++) {
377 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
378 return seg;
379
Mel Gorman5a178112012-07-31 16:45:02 -0700380 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700381 page_cache_get(pages[seg]);
382 }
383
384 return seg;
385}
386EXPORT_SYMBOL_GPL(get_kernel_pages);
387
388/*
389 * get_kernel_page() - pin a kernel page in memory
390 * @start: starting kernel address
391 * @write: pinning for read/write, currently ignored
392 * @pages: array that receives pointer to the page pinned.
393 * Must be at least nr_segs long.
394 *
395 * Returns 1 if page is pinned. If the page was not pinned, returns
396 * -errno. The page returned must be released with a put_page() call
397 * when it is finished with.
398 */
399int get_kernel_page(unsigned long start, int write, struct page **pages)
400{
401 const struct kvec kiov = {
402 .iov_base = (void *)start,
403 .iov_len = PAGE_SIZE
404 };
405
406 return get_kernel_pages(&kiov, 1, write, pages);
407}
408EXPORT_SYMBOL_GPL(get_kernel_page);
409
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700410static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700411 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
412 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700413{
414 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700415 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700416 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700417 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700418
419 for (i = 0; i < pagevec_count(pvec); i++) {
420 struct page *page = pvec->pages[i];
421 struct zone *pagezone = page_zone(page);
422
423 if (pagezone != zone) {
424 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700425 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700426 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700427 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700428 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700429
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700430 lruvec = mem_cgroup_page_lruvec(page, zone);
431 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700432 }
433 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700434 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800435 release_pages(pvec->pages, pvec->nr, pvec->cold);
436 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800437}
438
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700439static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
440 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700441{
442 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700443
444 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
445 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800446 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700447 (*pgmoved)++;
448 }
449}
450
451/*
452 * pagevec_move_tail() must be called with IRQ disabled.
453 * Otherwise this may cause nasty races.
454 */
455static void pagevec_move_tail(struct pagevec *pvec)
456{
457 int pgmoved = 0;
458
459 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
460 __count_vm_events(PGROTATED, pgmoved);
461}
462
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700463/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 * Writeback is about to end against a page which has been marked for immediate
465 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700466 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700468void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700470 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700471 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700472 struct pagevec *pvec;
473 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700475 page_cache_get(page);
476 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700477 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700478 if (!pagevec_add(pvec, page))
479 pagevec_move_tail(pvec);
480 local_irq_restore(flags);
481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700484static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800485 int file, int rotated)
486{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700487 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800488
489 reclaim_stat->recent_scanned[file]++;
490 if (rotated)
491 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800492}
493
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700494static void __activate_page(struct page *page, struct lruvec *lruvec,
495 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Linus Torvalds7a608572011-01-17 14:42:19 -0800497 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
498 int file = page_is_file_cache(page);
499 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800500
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700501 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800502 SetPageActive(page);
503 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700504 add_page_to_lru_list(page, lruvec, lru);
Mel Gorman24b7e582014-08-06 16:07:11 -0700505 trace_mm_lru_activate(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800506
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700507 __count_vm_event(PGACTIVATE);
508 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800509 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700510}
511
512#ifdef CONFIG_SMP
513static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
514
515static void activate_page_drain(int cpu)
516{
517 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
518
519 if (pagevec_count(pvec))
520 pagevec_lru_move_fn(pvec, __activate_page, NULL);
521}
522
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700523static bool need_activate_page_drain(int cpu)
524{
525 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
526}
527
Shaohua Lieb709b02011-05-24 17:12:55 -0700528void activate_page(struct page *page)
529{
530 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
531 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
532
533 page_cache_get(page);
534 if (!pagevec_add(pvec, page))
535 pagevec_lru_move_fn(pvec, __activate_page, NULL);
536 put_cpu_var(activate_page_pvecs);
537 }
538}
539
540#else
541static inline void activate_page_drain(int cpu)
542{
543}
544
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700545static bool need_activate_page_drain(int cpu)
546{
547 return false;
548}
549
Shaohua Lieb709b02011-05-24 17:12:55 -0700550void activate_page(struct page *page)
551{
552 struct zone *zone = page_zone(page);
553
554 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700555 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 spin_unlock_irq(&zone->lru_lock);
557}
Shaohua Lieb709b02011-05-24 17:12:55 -0700558#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Mel Gorman059285a2013-07-03 15:02:30 -0700560static void __lru_cache_activate_page(struct page *page)
561{
562 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
563 int i;
564
565 /*
566 * Search backwards on the optimistic assumption that the page being
567 * activated has just been added to this pagevec. Note that only
568 * the local pagevec is examined as a !PageLRU page could be in the
569 * process of being released, reclaimed, migrated or on a remote
570 * pagevec that is currently being drained. Furthermore, marking
571 * a remote pagevec's page PageActive potentially hits a race where
572 * a page is marked PageActive just after it is added to the inactive
573 * list causing accounting errors and BUG_ON checks to trigger.
574 */
575 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
576 struct page *pagevec_page = pvec->pages[i];
577
578 if (pagevec_page == page) {
579 SetPageActive(page);
580 break;
581 }
582 }
583
584 put_cpu_var(lru_add_pvec);
585}
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587/*
588 * Mark a page as having seen activity.
589 *
590 * inactive,unreferenced -> inactive,referenced
591 * inactive,referenced -> active,unreferenced
592 * active,unreferenced -> active,referenced
Hugh Dickinseb39d612014-08-06 16:06:43 -0700593 *
594 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
595 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800597void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700599 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700600 PageReferenced(page)) {
601
602 /*
603 * If the page is on the LRU, queue it for activation via
604 * activate_page_pvecs. Otherwise, assume the page is on a
605 * pagevec, mark it active and it'll be moved to the active
606 * LRU on the next drain.
607 */
608 if (PageLRU(page))
609 activate_page(page);
610 else
611 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700613 if (page_is_file_cache(page))
614 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 } else if (!PageReferenced(page)) {
616 SetPageReferenced(page);
617 }
618}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619EXPORT_SYMBOL(mark_page_accessed);
620
Jianyu Zhan2329d372014-06-04 16:07:31 -0700621static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Mel Gorman13f7f782013-07-03 15:02:28 -0700623 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700626 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700627 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700628 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700629 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700631
632/**
633 * lru_cache_add: add a page to the page lists
634 * @page: the page to add
635 */
636void lru_cache_add_anon(struct page *page)
637{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700638 if (PageActive(page))
639 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700640 __lru_cache_add(page);
641}
642
643void lru_cache_add_file(struct page *page)
644{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700645 if (PageActive(page))
646 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700647 __lru_cache_add(page);
648}
649EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700651/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700652 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700653 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700654 *
655 * Queue the page for addition to the LRU via pagevec. The decision on whether
656 * to add the page to the [in]active [file|anon] list is deferred until the
657 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
658 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700659 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700660void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
Sasha Levin309381fea2014-01-23 15:52:54 -0800662 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
663 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700664 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700667/**
668 * add_page_to_unevictable_list - add a page to the unevictable list
669 * @page: the page to be added to the unevictable list
670 *
671 * Add page directly to its zone's unevictable list. To avoid races with
672 * tasks that might be making the page evictable, through eg. munlock,
673 * munmap or exit, while it's not on the lru, we want to add the page
674 * while it's locked or otherwise "invisible" to other tasks. This is
675 * difficult to do when using the pagevec cache, so bypass that.
676 */
677void add_page_to_unevictable_list(struct page *page)
678{
679 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700680 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700681
682 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700683 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700684 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700685 SetPageUnevictable(page);
686 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700687 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700688 spin_unlock_irq(&zone->lru_lock);
689}
690
Johannes Weiner00501b52014-08-08 14:19:20 -0700691/**
692 * lru_cache_add_active_or_unevictable
693 * @page: the page to be added to LRU
694 * @vma: vma in which page is mapped for determining reclaimability
695 *
696 * Place @page on the active or unevictable LRU list, depending on its
697 * evictability. Note that if the page is not evictable, it goes
698 * directly back onto it's zone's unevictable list, it does NOT use a
699 * per cpu pagevec.
700 */
701void lru_cache_add_active_or_unevictable(struct page *page,
702 struct vm_area_struct *vma)
703{
704 VM_BUG_ON_PAGE(PageLRU(page), page);
705
706 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
707 SetPageActive(page);
708 lru_cache_add(page);
709 return;
710 }
711
712 if (!TestSetPageMlocked(page)) {
713 /*
714 * We use the irq-unsafe __mod_zone_page_stat because this
715 * counter is not modified from interrupt context, and the pte
716 * lock is held(spinlock), which implies preemption disabled.
717 */
718 __mod_zone_page_state(page_zone(page), NR_MLOCK,
719 hpage_nr_pages(page));
720 count_vm_event(UNEVICTABLE_PGMLOCKED);
721 }
722 add_page_to_unevictable_list(page);
723}
724
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700725/*
Minchan Kim31560182011-03-22 16:32:52 -0700726 * If the page can not be invalidated, it is moved to the
727 * inactive list to speed up its reclaim. It is moved to the
728 * head of the list, rather than the tail, to give the flusher
729 * threads some time to write it out, as this is much more
730 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700731 *
732 * If the page isn't page_mapped and dirty/writeback, the page
733 * could reclaim asap using PG_reclaim.
734 *
735 * 1. active, mapped page -> none
736 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
737 * 3. inactive, mapped page -> none
738 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
739 * 5. inactive, clean -> inactive, tail
740 * 6. Others -> none
741 *
742 * In 4, why it moves inactive's head, the VM expects the page would
743 * be write it out by flusher threads as this is much more effective
744 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700745 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700746static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
747 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700748{
749 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700750 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700751
Minchan Kim278df9f2011-03-22 16:32:54 -0700752 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700753 return;
754
Minchan Kimbad49d92011-05-11 15:13:30 -0700755 if (PageUnevictable(page))
756 return;
757
Minchan Kim31560182011-03-22 16:32:52 -0700758 /* Some processes are using the page */
759 if (page_mapped(page))
760 return;
761
Minchan Kim278df9f2011-03-22 16:32:54 -0700762 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700763 file = page_is_file_cache(page);
764 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700765
766 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700767 ClearPageActive(page);
768 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700769 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700770
Minchan Kim278df9f2011-03-22 16:32:54 -0700771 if (PageWriteback(page) || PageDirty(page)) {
772 /*
773 * PG_reclaim could be raced with end_page_writeback
774 * It can make readahead confusing. But race window
775 * is _really_ small and it's non-critical problem.
776 */
777 SetPageReclaim(page);
778 } else {
779 /*
780 * The page's writeback ends up during pagevec
781 * We moves tha page into tail of inactive.
782 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800783 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700784 __count_vm_event(PGROTATED);
785 }
786
787 if (active)
788 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700789 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700790}
791
Minchan Kim31560182011-03-22 16:32:52 -0700792/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700793 * Drain pages out of the cpu's pagevecs.
794 * Either "cpu" is the current CPU, and preemption has already been
795 * disabled; or "cpu" is being hot-unplugged, and is already dead.
796 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700797void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
Mel Gorman13f7f782013-07-03 15:02:28 -0700799 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Mel Gorman13f7f782013-07-03 15:02:28 -0700801 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700802 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700803
804 pvec = &per_cpu(lru_rotate_pvecs, cpu);
805 if (pagevec_count(pvec)) {
806 unsigned long flags;
807
808 /* No harm done if a racing interrupt already did this */
809 local_irq_save(flags);
810 pagevec_move_tail(pvec);
811 local_irq_restore(flags);
812 }
Minchan Kim31560182011-03-22 16:32:52 -0700813
814 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
815 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700816 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700817
818 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700819}
820
821/**
822 * deactivate_page - forcefully deactivate a page
823 * @page: page to deactivate
824 *
825 * This function hints the VM that @page is a good reclaim candidate,
826 * for example if its invalidation fails due to the page being dirty
827 * or under writeback.
828 */
829void deactivate_page(struct page *page)
830{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700831 /*
832 * In a workload with many unevictable page such as mprotect, unevictable
833 * page deactivation for accelerating reclaim is pointless.
834 */
835 if (PageUnevictable(page))
836 return;
837
Minchan Kim31560182011-03-22 16:32:52 -0700838 if (likely(get_page_unless_zero(page))) {
839 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
840
841 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700842 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700843 put_cpu_var(lru_deactivate_pvecs);
844 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800845}
846
847void lru_add_drain(void)
848{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700849 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800850 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
852
David Howellsc4028952006-11-22 14:57:56 +0000853static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800854{
855 lru_add_drain();
856}
857
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700858static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
859
860void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800861{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700862 static DEFINE_MUTEX(lock);
863 static struct cpumask has_work;
864 int cpu;
865
866 mutex_lock(&lock);
867 get_online_cpus();
868 cpumask_clear(&has_work);
869
870 for_each_online_cpu(cpu) {
871 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
872
873 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
874 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
875 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
876 need_activate_page_drain(cpu)) {
877 INIT_WORK(work, lru_add_drain_per_cpu);
878 schedule_work_on(cpu, work);
879 cpumask_set_cpu(cpu, &has_work);
880 }
881 }
882
883 for_each_cpu(cpu, &has_work)
884 flush_work(&per_cpu(lru_add_drain_work, cpu));
885
886 put_online_cpus();
887 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800888}
889
Michal Hockoaabfb572014-10-09 15:28:52 -0700890/**
891 * release_pages - batched page_cache_release()
892 * @pages: array of pages to release
893 * @nr: number of pages
894 * @cold: whether the pages are cache cold
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 *
Michal Hockoaabfb572014-10-09 15:28:52 -0700896 * Decrement the reference count on all the pages in @pages. If it
897 * fell to zero, remove the page from the LRU and free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 */
Mel Gormanb745bc82014-06-04 16:10:22 -0700899void release_pages(struct page **pages, int nr, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
901 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800902 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700904 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700905 unsigned long uninitialized_var(flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700906 unsigned int uninitialized_var(lock_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 for (i = 0; i < nr; i++) {
909 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Nick Piggin8519fb32006-02-07 12:58:52 -0800911 if (unlikely(PageCompound(page))) {
912 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700913 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800914 zone = NULL;
915 }
916 put_compound_page(page);
917 continue;
918 }
919
Michal Hockoaabfb572014-10-09 15:28:52 -0700920 /*
921 * Make sure the IRQ-safe lock-holding time does not get
922 * excessive with a continuous string of pages from the
923 * same zone. The lock is held only if zone != NULL.
924 */
925 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
926 spin_unlock_irqrestore(&zone->lru_lock, flags);
927 zone = NULL;
928 }
929
Nick Pigginb5810032005-10-29 18:16:12 -0700930 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 continue;
932
Nick Piggin46453a62006-03-22 00:07:58 -0800933 if (PageLRU(page)) {
934 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700935
Nick Piggin46453a62006-03-22 00:07:58 -0800936 if (pagezone != zone) {
937 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700938 spin_unlock_irqrestore(&zone->lru_lock,
939 flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700940 lock_batch = 0;
Nick Piggin46453a62006-03-22 00:07:58 -0800941 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700942 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700944
945 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -0800946 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800947 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700948 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 }
Nick Piggin46453a62006-03-22 00:07:58 -0800950
Mel Gormanc53954a2013-07-03 15:02:34 -0700951 /* Clear Active bit in case of parallel mark_page_accessed */
Mel Gormane3741b52014-06-04 16:10:26 -0700952 __ClearPageActive(page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700953
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800954 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
956 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700957 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Johannes Weiner747db952014-08-08 14:19:24 -0700959 mem_cgroup_uncharge_list(&pages_to_free);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800960 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700962EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
964/*
965 * The pages which we're about to release may be in the deferred lru-addition
966 * queues. That would prevent them from really being freed right now. That's
967 * OK from a correctness point of view but is inefficient - those pages may be
968 * cache-warm and we want to give them back to the page allocator ASAP.
969 *
970 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
971 * and __pagevec_lru_add_active() call release_pages() directly to avoid
972 * mutual recursion.
973 */
974void __pagevec_release(struct pagevec *pvec)
975{
976 lru_add_drain();
977 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
978 pagevec_reinit(pvec);
979}
Steve French7f285702005-11-01 10:22:55 -0800980EXPORT_SYMBOL(__pagevec_release);
981
Hugh Dickins12d27102012-01-12 17:19:52 -0800982#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800983/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700984void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700985 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800986{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800987 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800988
Sasha Levin309381fea2014-01-23 15:52:54 -0800989 VM_BUG_ON_PAGE(!PageHead(page), page);
990 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
991 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700992 VM_BUG_ON(NR_CPUS != 1 &&
993 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800994
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700995 if (!list)
996 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800997
Hugh Dickins12d27102012-01-12 17:19:52 -0800998 if (likely(PageLRU(page)))
999 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -07001000 else if (list) {
1001 /* page reclaim is reclaiming a huge page */
1002 get_page(page_tail);
1003 list_add_tail(&page_tail->lru, list);
1004 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -08001005 struct list_head *list_head;
1006 /*
1007 * Head page has not yet been counted, as an hpage,
1008 * so we must account for each subpage individually.
1009 *
1010 * Use the standard add function to put page_tail on the list,
1011 * but then correct its position so they all end up in order.
1012 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -07001013 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -08001014 list_head = page_tail->lru.prev;
1015 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001016 }
Hugh Dickins75121022012-03-05 14:59:18 -08001017
1018 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -07001019 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001020}
Hugh Dickins12d27102012-01-12 17:19:52 -08001021#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001022
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001023static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
1024 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001025{
Mel Gorman13f7f782013-07-03 15:02:28 -07001026 int file = page_is_file_cache(page);
1027 int active = PageActive(page);
1028 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001029
Sasha Levin309381fea2014-01-23 15:52:54 -08001030 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001031
1032 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001033 add_page_to_lru_list(page, lruvec, lru);
1034 update_page_reclaim_stat(lruvec, file, active);
Mel Gorman24b7e582014-08-06 16:07:11 -07001035 trace_mm_lru_insertion(page, lru);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001036}
1037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 * Add the passed pages to the LRU, then drop the caller's refcount
1040 * on them. Reinitialises the caller's pagevec.
1041 */
Mel Gormana0b8cab32013-07-03 15:02:32 -07001042void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043{
Mel Gormana0b8cab32013-07-03 15:02:32 -07001044 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045}
Hugh Dickins5095ae832012-01-12 17:19:58 -08001046EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001049 * pagevec_lookup_entries - gang pagecache lookup
1050 * @pvec: Where the resulting entries are placed
1051 * @mapping: The address_space to search
1052 * @start: The starting entry index
1053 * @nr_entries: The maximum number of entries
1054 * @indices: The cache indices corresponding to the entries in @pvec
1055 *
1056 * pagevec_lookup_entries() will search for and return a group of up
1057 * to @nr_entries pages and shadow entries in the mapping. All
1058 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1059 * reference against actual pages in @pvec.
1060 *
1061 * The search returns a group of mapping-contiguous entries with
1062 * ascending indexes. There may be holes in the indices due to
1063 * not-present entries.
1064 *
1065 * pagevec_lookup_entries() returns the number of entries which were
1066 * found.
1067 */
1068unsigned pagevec_lookup_entries(struct pagevec *pvec,
1069 struct address_space *mapping,
1070 pgoff_t start, unsigned nr_pages,
1071 pgoff_t *indices)
1072{
1073 pvec->nr = find_get_entries(mapping, start, nr_pages,
1074 pvec->pages, indices);
1075 return pagevec_count(pvec);
1076}
1077
1078/**
1079 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1080 * @pvec: The pagevec to prune
1081 *
1082 * pagevec_lookup_entries() fills both pages and exceptional radix
1083 * tree entries into the pagevec. This function prunes all
1084 * exceptionals from @pvec without leaving holes, so that it can be
1085 * passed on to page-only pagevec operations.
1086 */
1087void pagevec_remove_exceptionals(struct pagevec *pvec)
1088{
1089 int i, j;
1090
1091 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1092 struct page *page = pvec->pages[i];
1093 if (!radix_tree_exceptional_entry(page))
1094 pvec->pages[j++] = page;
1095 }
1096 pvec->nr = j;
1097}
1098
1099/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 * pagevec_lookup - gang pagecache lookup
1101 * @pvec: Where the resulting pages are placed
1102 * @mapping: The address_space to search
1103 * @start: The starting page index
1104 * @nr_pages: The maximum number of pages
1105 *
1106 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1107 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1108 * reference against the pages in @pvec.
1109 *
1110 * The search returns a group of mapping-contiguous pages with ascending
1111 * indexes. There may be holes in the indices due to not-present pages.
1112 *
1113 * pagevec_lookup() returns the number of pages which were found.
1114 */
1115unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1116 pgoff_t start, unsigned nr_pages)
1117{
1118 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1119 return pagevec_count(pvec);
1120}
Christoph Hellwig78539fd2006-01-11 20:47:41 +11001121EXPORT_SYMBOL(pagevec_lookup);
1122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1124 pgoff_t *index, int tag, unsigned nr_pages)
1125{
1126 pvec->nr = find_get_pages_tag(mapping, index, tag,
1127 nr_pages, pvec->pages);
1128 return pagevec_count(pvec);
1129}
Steve French7f285702005-11-01 10:22:55 -08001130EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132/*
1133 * Perform any setup for the swap system
1134 */
1135void __init swap_setup(void)
1136{
Jan Beulich44813742009-09-21 17:03:05 -07001137 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001138#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -08001139 int i;
1140
Mikulas Patocka8077c0d2013-10-14 12:14:13 -04001141 if (bdi_init(swapper_spaces[0].backing_dev_info))
1142 panic("Failed to init swap bdi");
Shaohua Li33806f02013-02-22 16:34:37 -08001143 for (i = 0; i < MAX_SWAPFILES; i++) {
1144 spin_lock_init(&swapper_spaces[i].tree_lock);
1145 INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
1146 }
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001147#endif
1148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 /* Use a smaller cluster for small-memory machines */
1150 if (megs < 16)
1151 page_cluster = 2;
1152 else
1153 page_cluster = 3;
1154 /*
1155 * Right now other parts of the system means that we
1156 * _really_ don't want to cluster much more
1157 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158}