blob: 9ce43ba4498bee042934f484d40f2b7f2fcf48ce [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 }
Andrea Arcangeli91807062011-01-13 15:46:32 -080065}
66
67static void __put_single_page(struct page *page)
68{
69 __page_cache_release(page);
Li Hongfc916682010-03-05 13:41:54 -080070 free_hot_cold_page(page, 0);
Adrian Bunkb2213852006-09-25 23:31:02 -070071}
72
Andrea Arcangeli91807062011-01-13 15:46:32 -080073static void __put_compound_page(struct page *page)
74{
75 compound_page_dtor *dtor;
76
77 __page_cache_release(page);
78 dtor = get_compound_page_dtor(page);
79 (*dtor)(page);
80}
81
Nick Piggin8519fb32006-02-07 12:58:52 -080082static void put_compound_page(struct page *page)
83{
Andrew Morton26296ad2014-01-21 15:48:59 -080084 struct page *page_head;
85
86 if (likely(!PageTail(page))) {
87 if (put_page_testzero(page)) {
88 /*
89 * By the time all refcounts have been released
90 * split_huge_page cannot run anymore from under us.
91 */
92 if (PageHead(page))
93 __put_compound_page(page);
94 else
95 __put_single_page(page);
96 }
97 return;
98 }
99
100 /* __split_huge_page_refcount can run under us */
David Rientjes668f9abb2014-03-03 15:38:18 -0800101 page_head = compound_head(page);
Andrew Morton26296ad2014-01-21 15:48:59 -0800102
103 /*
104 * THP can not break up slab pages so avoid taking
105 * compound_lock() and skip the tail page refcounting (in
106 * _mapcount) too. Slab performs non-atomic bit ops on
107 * page->flags for better performance. In particular
108 * slab_unlock() in slub used to be a hot path. It is still
109 * hot on arches that do not support
110 * this_cpu_cmpxchg_double().
111 *
112 * If "page" is part of a slab or hugetlbfs page it cannot be
113 * splitted and the head page cannot change from under us. And
114 * if "page" is part of a THP page under splitting, if the
115 * head page pointed by the THP tail isn't a THP head anymore,
116 * we'll find PageTail clear after smp_rmb() and we'll treat
117 * it as a single page.
118 */
119 if (!__compound_tail_refcounted(page_head)) {
120 /*
121 * If "page" is a THP tail, we must read the tail page
122 * flags after the head page flags. The
123 * split_huge_page side enforces write memory barriers
124 * between clearing PageTail and before the head page
125 * can be freed and reallocated.
126 */
127 smp_rmb();
128 if (likely(PageTail(page))) {
129 /*
130 * __split_huge_page_refcount cannot race
131 * here.
132 */
Sasha Levin309381fea2014-01-23 15:52:54 -0800133 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
134 VM_BUG_ON_PAGE(page_mapcount(page) != 0, page);
Andrew Morton26296ad2014-01-21 15:48:59 -0800135 if (put_page_testzero(page_head)) {
136 /*
137 * If this is the tail of a slab
138 * compound page, the tail pin must
139 * not be the last reference held on
140 * the page, because the PG_slab
141 * cannot be cleared before all tail
142 * pins (which skips the _mapcount
143 * tail refcounting) have been
144 * released. For hugetlbfs the tail
145 * pin may be the last reference on
146 * the page instead, because
147 * PageHeadHuge will not go away until
148 * the compound page enters the buddy
149 * allocator.
150 */
Sasha Levin309381fea2014-01-23 15:52:54 -0800151 VM_BUG_ON_PAGE(PageSlab(page_head), page_head);
Andrew Morton26296ad2014-01-21 15:48:59 -0800152 __put_compound_page(page_head);
153 }
154 return;
155 } else
156 /*
157 * __split_huge_page_refcount run before us,
158 * "page" was a THP tail. The split page_head
159 * has been freed and reallocated as slab or
160 * hugetlbfs page of smaller order (only
161 * possible if reallocated as slab on x86).
162 */
163 goto out_put_single;
164 }
165
166 if (likely(page != page_head && get_page_unless_zero(page_head))) {
167 unsigned long flags;
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700168
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800169 /*
Andrew Morton26296ad2014-01-21 15:48:59 -0800170 * page_head wasn't a dangling pointer but it may not
171 * be a head page anymore by the time we obtain the
172 * lock. That is ok as long as it can't be freed from
173 * under us.
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800174 */
Andrew Morton26296ad2014-01-21 15:48:59 -0800175 flags = compound_lock_irqsave(page_head);
176 if (unlikely(!PageTail(page))) {
177 /* __split_huge_page_refcount run before us */
Andrea Arcangeli91807062011-01-13 15:46:32 -0800178 compound_unlock_irqrestore(page_head, flags);
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800179 if (put_page_testzero(page_head)) {
Andrew Morton26296ad2014-01-21 15:48:59 -0800180 /*
181 * The head page may have been freed
182 * and reallocated as a compound page
183 * of smaller order and then freed
184 * again. All we know is that it
185 * cannot have become: a THP page, a
186 * compound page of higher order, a
187 * tail page. That is because we
188 * still hold the refcount of the
189 * split THP tail and page_head was
190 * the THP head before the split.
191 */
Andrea Arcangelia95a82e2011-01-13 15:46:33 -0800192 if (PageHead(page_head))
193 __put_compound_page(page_head);
194 else
195 __put_single_page(page_head);
196 }
Andrew Morton26296ad2014-01-21 15:48:59 -0800197out_put_single:
198 if (put_page_testzero(page))
199 __put_single_page(page);
200 return;
Andrea Arcangeli91807062011-01-13 15:46:32 -0800201 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800202 VM_BUG_ON_PAGE(page_head != page->first_page, page);
Andrew Morton26296ad2014-01-21 15:48:59 -0800203 /*
204 * We can release the refcount taken by
205 * get_page_unless_zero() now that
206 * __split_huge_page_refcount() is blocked on the
207 * compound_lock.
208 */
209 if (put_page_testzero(page_head))
Sasha Levin309381fea2014-01-23 15:52:54 -0800210 VM_BUG_ON_PAGE(1, page_head);
Andrew Morton26296ad2014-01-21 15:48:59 -0800211 /* __split_huge_page_refcount will wait now */
Sasha Levin309381fea2014-01-23 15:52:54 -0800212 VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page);
Andrew Morton26296ad2014-01-21 15:48:59 -0800213 atomic_dec(&page->_mapcount);
Sasha Levin309381fea2014-01-23 15:52:54 -0800214 VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head);
215 VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page);
Andrew Morton26296ad2014-01-21 15:48:59 -0800216 compound_unlock_irqrestore(page_head, flags);
217
218 if (put_page_testzero(page_head)) {
219 if (PageHead(page_head))
220 __put_compound_page(page_head);
221 else
222 __put_single_page(page_head);
223 }
224 } else {
225 /* page_head is a dangling pointer */
Sasha Levin309381fea2014-01-23 15:52:54 -0800226 VM_BUG_ON_PAGE(PageTail(page), page);
Andrew Morton26296ad2014-01-21 15:48:59 -0800227 goto out_put_single;
Nick Piggin8519fb32006-02-07 12:58:52 -0800228 }
229}
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231void put_page(struct page *page)
232{
Nick Piggin8519fb32006-02-07 12:58:52 -0800233 if (unlikely(PageCompound(page)))
234 put_compound_page(page);
235 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800236 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700240/*
241 * This function is exported but must not be called by anything other
242 * than get_page(). It implements the slow path of get_page().
243 */
244bool __get_page_tail(struct page *page)
245{
246 /*
247 * This takes care of get_page() if run on a tail page
248 * returned by one of the get_user_pages/follow_page variants.
249 * get_user_pages/follow_page itself doesn't need the compound
250 * lock because it runs __get_page_tail_foll() under the
251 * proper PT lock that already serializes against
252 * split_huge_page().
253 */
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800254 unsigned long flags;
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800255 bool got;
David Rientjes668f9abb2014-03-03 15:38:18 -0800256 struct page *page_head = compound_head(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700257
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800258 /* Ref to put_compound_page() comment. */
Andrea Arcangeli3bfcd132014-01-21 15:48:56 -0800259 if (!__compound_tail_refcounted(page_head)) {
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800260 smp_rmb();
261 if (likely(PageTail(page))) {
262 /*
263 * This is a hugetlbfs page or a slab
264 * page. __split_huge_page_refcount
265 * cannot race here.
266 */
Sasha Levin309381fea2014-01-23 15:52:54 -0800267 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800268 __get_page_tail_foll(page, true);
269 return true;
270 } else {
271 /*
272 * __split_huge_page_refcount run
273 * before us, "page" was a THP
274 * tail. The split page_head has been
275 * freed and reallocated as slab or
276 * hugetlbfs page of smaller order
277 * (only possible if reallocated as
278 * slab on x86).
279 */
280 return false;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700281 }
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800282 }
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800283
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800284 got = false;
285 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800286 /*
287 * page_head wasn't a dangling pointer but it
288 * may not be a head page anymore by the time
289 * we obtain the lock. That is ok as long as it
290 * can't be freed from under us.
291 */
292 flags = compound_lock_irqsave(page_head);
293 /* here __split_huge_page_refcount won't run anymore */
294 if (likely(PageTail(page))) {
295 __get_page_tail_foll(page, false);
296 got = true;
297 }
298 compound_unlock_irqrestore(page_head, flags);
299 if (unlikely(!got))
300 put_page(page_head);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700301 }
302 return got;
303}
304EXPORT_SYMBOL(__get_page_tail);
305
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700306/**
Randy Dunlap76824862008-03-19 17:00:40 -0700307 * put_pages_list() - release a list of pages
308 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700309 *
310 * Release a list of pages which are strung together on page.lru. Currently
311 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700312 */
313void put_pages_list(struct list_head *pages)
314{
315 while (!list_empty(pages)) {
316 struct page *victim;
317
318 victim = list_entry(pages->prev, struct page, lru);
319 list_del(&victim->lru);
320 page_cache_release(victim);
321 }
322}
323EXPORT_SYMBOL(put_pages_list);
324
Mel Gorman18022c52012-07-31 16:44:51 -0700325/*
326 * get_kernel_pages() - pin kernel pages in memory
327 * @kiov: An array of struct kvec structures
328 * @nr_segs: number of segments to pin
329 * @write: pinning for read/write, currently ignored
330 * @pages: array that receives pointers to the pages pinned.
331 * Should be at least nr_segs long.
332 *
333 * Returns number of pages pinned. This may be fewer than the number
334 * requested. If nr_pages is 0 or negative, returns 0. If no pages
335 * were pinned, returns -errno. Each page returned must be released
336 * with a put_page() call when it is finished with.
337 */
338int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
339 struct page **pages)
340{
341 int seg;
342
343 for (seg = 0; seg < nr_segs; seg++) {
344 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
345 return seg;
346
Mel Gorman5a178112012-07-31 16:45:02 -0700347 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700348 page_cache_get(pages[seg]);
349 }
350
351 return seg;
352}
353EXPORT_SYMBOL_GPL(get_kernel_pages);
354
355/*
356 * get_kernel_page() - pin a kernel page in memory
357 * @start: starting kernel address
358 * @write: pinning for read/write, currently ignored
359 * @pages: array that receives pointer to the page pinned.
360 * Must be at least nr_segs long.
361 *
362 * Returns 1 if page is pinned. If the page was not pinned, returns
363 * -errno. The page returned must be released with a put_page() call
364 * when it is finished with.
365 */
366int get_kernel_page(unsigned long start, int write, struct page **pages)
367{
368 const struct kvec kiov = {
369 .iov_base = (void *)start,
370 .iov_len = PAGE_SIZE
371 };
372
373 return get_kernel_pages(&kiov, 1, write, pages);
374}
375EXPORT_SYMBOL_GPL(get_kernel_page);
376
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700377static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700378 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
379 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700380{
381 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700382 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700383 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700384 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700385
386 for (i = 0; i < pagevec_count(pvec); i++) {
387 struct page *page = pvec->pages[i];
388 struct zone *pagezone = page_zone(page);
389
390 if (pagezone != zone) {
391 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700392 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700393 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700394 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700395 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700396
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700397 lruvec = mem_cgroup_page_lruvec(page, zone);
398 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700399 }
400 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700401 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800402 release_pages(pvec->pages, pvec->nr, pvec->cold);
403 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800404}
405
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700406static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
407 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700408{
409 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700410
411 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
412 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800413 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700414 (*pgmoved)++;
415 }
416}
417
418/*
419 * pagevec_move_tail() must be called with IRQ disabled.
420 * Otherwise this may cause nasty races.
421 */
422static void pagevec_move_tail(struct pagevec *pvec)
423{
424 int pgmoved = 0;
425
426 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
427 __count_vm_events(PGROTATED, pgmoved);
428}
429
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700430/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 * Writeback is about to end against a page which has been marked for immediate
432 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700433 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700435void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700437 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700438 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700439 struct pagevec *pvec;
440 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700442 page_cache_get(page);
443 local_irq_save(flags);
444 pvec = &__get_cpu_var(lru_rotate_pvecs);
445 if (!pagevec_add(pvec, page))
446 pagevec_move_tail(pvec);
447 local_irq_restore(flags);
448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449}
450
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700451static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800452 int file, int rotated)
453{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700454 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800455
456 reclaim_stat->recent_scanned[file]++;
457 if (rotated)
458 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800459}
460
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700461static void __activate_page(struct page *page, struct lruvec *lruvec,
462 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
Linus Torvalds7a608572011-01-17 14:42:19 -0800464 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
465 int file = page_is_file_cache(page);
466 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800467
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700468 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800469 SetPageActive(page);
470 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700471 add_page_to_lru_list(page, lruvec, lru);
Mel Gormanc6286c92013-07-03 15:02:26 -0700472 trace_mm_lru_activate(page, page_to_pfn(page));
Linus Torvalds7a608572011-01-17 14:42:19 -0800473
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700474 __count_vm_event(PGACTIVATE);
475 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800476 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700477}
478
479#ifdef CONFIG_SMP
480static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
481
482static void activate_page_drain(int cpu)
483{
484 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
485
486 if (pagevec_count(pvec))
487 pagevec_lru_move_fn(pvec, __activate_page, NULL);
488}
489
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700490static bool need_activate_page_drain(int cpu)
491{
492 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
493}
494
Shaohua Lieb709b02011-05-24 17:12:55 -0700495void activate_page(struct page *page)
496{
497 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
498 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
499
500 page_cache_get(page);
501 if (!pagevec_add(pvec, page))
502 pagevec_lru_move_fn(pvec, __activate_page, NULL);
503 put_cpu_var(activate_page_pvecs);
504 }
505}
506
507#else
508static inline void activate_page_drain(int cpu)
509{
510}
511
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700512static bool need_activate_page_drain(int cpu)
513{
514 return false;
515}
516
Shaohua Lieb709b02011-05-24 17:12:55 -0700517void activate_page(struct page *page)
518{
519 struct zone *zone = page_zone(page);
520
521 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700522 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 spin_unlock_irq(&zone->lru_lock);
524}
Shaohua Lieb709b02011-05-24 17:12:55 -0700525#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Mel Gorman059285a2013-07-03 15:02:30 -0700527static void __lru_cache_activate_page(struct page *page)
528{
529 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
530 int i;
531
532 /*
533 * Search backwards on the optimistic assumption that the page being
534 * activated has just been added to this pagevec. Note that only
535 * the local pagevec is examined as a !PageLRU page could be in the
536 * process of being released, reclaimed, migrated or on a remote
537 * pagevec that is currently being drained. Furthermore, marking
538 * a remote pagevec's page PageActive potentially hits a race where
539 * a page is marked PageActive just after it is added to the inactive
540 * list causing accounting errors and BUG_ON checks to trigger.
541 */
542 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
543 struct page *pagevec_page = pvec->pages[i];
544
545 if (pagevec_page == page) {
546 SetPageActive(page);
547 break;
548 }
549 }
550
551 put_cpu_var(lru_add_pvec);
552}
553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554/*
555 * Mark a page as having seen activity.
556 *
557 * inactive,unreferenced -> inactive,referenced
558 * inactive,referenced -> active,unreferenced
559 * active,unreferenced -> active,referenced
560 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800561void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700563 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700564 PageReferenced(page)) {
565
566 /*
567 * If the page is on the LRU, queue it for activation via
568 * activate_page_pvecs. Otherwise, assume the page is on a
569 * pagevec, mark it active and it'll be moved to the active
570 * LRU on the next drain.
571 */
572 if (PageLRU(page))
573 activate_page(page);
574 else
575 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700577 if (page_is_file_cache(page))
578 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 } else if (!PageReferenced(page)) {
580 SetPageReferenced(page);
581 }
582}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583EXPORT_SYMBOL(mark_page_accessed);
584
Robin Dongd741c9c2012-10-08 16:29:05 -0700585/*
Mel Gorman13f7f782013-07-03 15:02:28 -0700586 * Queue the page for addition to the LRU via pagevec. The decision on whether
587 * to add the page to the [in]active [file|anon] list is deferred until the
588 * pagevec is drained. This gives a chance for the caller of __lru_cache_add()
589 * have the page added to the active list using mark_page_accessed().
Robin Dongd741c9c2012-10-08 16:29:05 -0700590 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700591void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
Mel Gorman13f7f782013-07-03 15:02:28 -0700593 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700596 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700597 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700598 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700599 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
Miklos Szeredi47846b02010-05-25 15:06:06 +0200601EXPORT_SYMBOL(__lru_cache_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700603/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700604 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700605 * @page: the page to be added to the LRU.
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700606 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700607void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Sasha Levin309381fea2014-01-23 15:52:54 -0800609 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
610 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700611 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700614/**
615 * add_page_to_unevictable_list - add a page to the unevictable list
616 * @page: the page to be added to the unevictable list
617 *
618 * Add page directly to its zone's unevictable list. To avoid races with
619 * tasks that might be making the page evictable, through eg. munlock,
620 * munmap or exit, while it's not on the lru, we want to add the page
621 * while it's locked or otherwise "invisible" to other tasks. This is
622 * difficult to do when using the pagevec cache, so bypass that.
623 */
624void add_page_to_unevictable_list(struct page *page)
625{
626 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700627 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700628
629 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700630 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700631 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700632 SetPageUnevictable(page);
633 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700634 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700635 spin_unlock_irq(&zone->lru_lock);
636}
637
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700638/*
Minchan Kim31560182011-03-22 16:32:52 -0700639 * If the page can not be invalidated, it is moved to the
640 * inactive list to speed up its reclaim. It is moved to the
641 * head of the list, rather than the tail, to give the flusher
642 * threads some time to write it out, as this is much more
643 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700644 *
645 * If the page isn't page_mapped and dirty/writeback, the page
646 * could reclaim asap using PG_reclaim.
647 *
648 * 1. active, mapped page -> none
649 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
650 * 3. inactive, mapped page -> none
651 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
652 * 5. inactive, clean -> inactive, tail
653 * 6. Others -> none
654 *
655 * In 4, why it moves inactive's head, the VM expects the page would
656 * be write it out by flusher threads as this is much more effective
657 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700658 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700659static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
660 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700661{
662 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700663 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700664
Minchan Kim278df9f2011-03-22 16:32:54 -0700665 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700666 return;
667
Minchan Kimbad49d92011-05-11 15:13:30 -0700668 if (PageUnevictable(page))
669 return;
670
Minchan Kim31560182011-03-22 16:32:52 -0700671 /* Some processes are using the page */
672 if (page_mapped(page))
673 return;
674
Minchan Kim278df9f2011-03-22 16:32:54 -0700675 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700676 file = page_is_file_cache(page);
677 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700678
679 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700680 ClearPageActive(page);
681 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700682 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700683
Minchan Kim278df9f2011-03-22 16:32:54 -0700684 if (PageWriteback(page) || PageDirty(page)) {
685 /*
686 * PG_reclaim could be raced with end_page_writeback
687 * It can make readahead confusing. But race window
688 * is _really_ small and it's non-critical problem.
689 */
690 SetPageReclaim(page);
691 } else {
692 /*
693 * The page's writeback ends up during pagevec
694 * We moves tha page into tail of inactive.
695 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800696 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700697 __count_vm_event(PGROTATED);
698 }
699
700 if (active)
701 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700702 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700703}
704
Minchan Kim31560182011-03-22 16:32:52 -0700705/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700706 * Drain pages out of the cpu's pagevecs.
707 * Either "cpu" is the current CPU, and preemption has already been
708 * disabled; or "cpu" is being hot-unplugged, and is already dead.
709 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700710void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
Mel Gorman13f7f782013-07-03 15:02:28 -0700712 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Mel Gorman13f7f782013-07-03 15:02:28 -0700714 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700715 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700716
717 pvec = &per_cpu(lru_rotate_pvecs, cpu);
718 if (pagevec_count(pvec)) {
719 unsigned long flags;
720
721 /* No harm done if a racing interrupt already did this */
722 local_irq_save(flags);
723 pagevec_move_tail(pvec);
724 local_irq_restore(flags);
725 }
Minchan Kim31560182011-03-22 16:32:52 -0700726
727 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
728 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700729 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700730
731 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700732}
733
734/**
735 * deactivate_page - forcefully deactivate a page
736 * @page: page to deactivate
737 *
738 * This function hints the VM that @page is a good reclaim candidate,
739 * for example if its invalidation fails due to the page being dirty
740 * or under writeback.
741 */
742void deactivate_page(struct page *page)
743{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700744 /*
745 * In a workload with many unevictable page such as mprotect, unevictable
746 * page deactivation for accelerating reclaim is pointless.
747 */
748 if (PageUnevictable(page))
749 return;
750
Minchan Kim31560182011-03-22 16:32:52 -0700751 if (likely(get_page_unless_zero(page))) {
752 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
753
754 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700755 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700756 put_cpu_var(lru_deactivate_pvecs);
757 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800758}
759
760void lru_add_drain(void)
761{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700762 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800763 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764}
765
David Howellsc4028952006-11-22 14:57:56 +0000766static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800767{
768 lru_add_drain();
769}
770
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700771static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
772
773void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800774{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700775 static DEFINE_MUTEX(lock);
776 static struct cpumask has_work;
777 int cpu;
778
779 mutex_lock(&lock);
780 get_online_cpus();
781 cpumask_clear(&has_work);
782
783 for_each_online_cpu(cpu) {
784 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
785
786 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
787 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
788 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
789 need_activate_page_drain(cpu)) {
790 INIT_WORK(work, lru_add_drain_per_cpu);
791 schedule_work_on(cpu, work);
792 cpumask_set_cpu(cpu, &has_work);
793 }
794 }
795
796 for_each_cpu(cpu, &has_work)
797 flush_work(&per_cpu(lru_add_drain_work, cpu));
798
799 put_online_cpus();
800 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800801}
802
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 * Batched page_cache_release(). Decrement the reference count on all the
805 * passed pages. If it fell to zero then remove the page from the LRU and
806 * free it.
807 *
808 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
809 * for the remainder of the operation.
810 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700811 * The locking in this function is against shrink_inactive_list(): we recheck
812 * the page count inside the lock to see whether shrink_inactive_list()
813 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
814 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 */
816void release_pages(struct page **pages, int nr, int cold)
817{
818 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800819 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700821 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700822 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 for (i = 0; i < nr; i++) {
825 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Nick Piggin8519fb32006-02-07 12:58:52 -0800827 if (unlikely(PageCompound(page))) {
828 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700829 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800830 zone = NULL;
831 }
832 put_compound_page(page);
833 continue;
834 }
835
Nick Pigginb5810032005-10-29 18:16:12 -0700836 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 continue;
838
Nick Piggin46453a62006-03-22 00:07:58 -0800839 if (PageLRU(page)) {
840 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700841
Nick Piggin46453a62006-03-22 00:07:58 -0800842 if (pagezone != zone) {
843 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700844 spin_unlock_irqrestore(&zone->lru_lock,
845 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800846 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700847 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700849
850 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -0800851 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800852 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700853 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 }
Nick Piggin46453a62006-03-22 00:07:58 -0800855
Mel Gormanc53954a2013-07-03 15:02:34 -0700856 /* Clear Active bit in case of parallel mark_page_accessed */
857 ClearPageActive(page);
858
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800859 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
861 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700862 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800864 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700866EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868/*
869 * The pages which we're about to release may be in the deferred lru-addition
870 * queues. That would prevent them from really being freed right now. That's
871 * OK from a correctness point of view but is inefficient - those pages may be
872 * cache-warm and we want to give them back to the page allocator ASAP.
873 *
874 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
875 * and __pagevec_lru_add_active() call release_pages() directly to avoid
876 * mutual recursion.
877 */
878void __pagevec_release(struct pagevec *pvec)
879{
880 lru_add_drain();
881 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
882 pagevec_reinit(pvec);
883}
Steve French7f285702005-11-01 10:22:55 -0800884EXPORT_SYMBOL(__pagevec_release);
885
Hugh Dickins12d27102012-01-12 17:19:52 -0800886#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800887/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700888void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700889 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800890{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800891 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800892
Sasha Levin309381fea2014-01-23 15:52:54 -0800893 VM_BUG_ON_PAGE(!PageHead(page), page);
894 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
895 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700896 VM_BUG_ON(NR_CPUS != 1 &&
897 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800898
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700899 if (!list)
900 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800901
Hugh Dickins12d27102012-01-12 17:19:52 -0800902 if (likely(PageLRU(page)))
903 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700904 else if (list) {
905 /* page reclaim is reclaiming a huge page */
906 get_page(page_tail);
907 list_add_tail(&page_tail->lru, list);
908 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800909 struct list_head *list_head;
910 /*
911 * Head page has not yet been counted, as an hpage,
912 * so we must account for each subpage individually.
913 *
914 * Use the standard add function to put page_tail on the list,
915 * but then correct its position so they all end up in order.
916 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700917 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -0800918 list_head = page_tail->lru.prev;
919 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800920 }
Hugh Dickins75121022012-03-05 14:59:18 -0800921
922 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700923 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800924}
Hugh Dickins12d27102012-01-12 17:19:52 -0800925#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800926
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700927static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
928 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700929{
Mel Gorman13f7f782013-07-03 15:02:28 -0700930 int file = page_is_file_cache(page);
931 int active = PageActive(page);
932 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700933
Sasha Levin309381fea2014-01-23 15:52:54 -0800934 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700935
936 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700937 add_page_to_lru_list(page, lruvec, lru);
938 update_page_reclaim_stat(lruvec, file, active);
Mel Gormanc6286c92013-07-03 15:02:26 -0700939 trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page));
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700940}
941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 * Add the passed pages to the LRU, then drop the caller's refcount
944 * on them. Reinitialises the caller's pagevec.
945 */
Mel Gormana0b8cab32013-07-03 15:02:32 -0700946void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
Mel Gormana0b8cab32013-07-03 15:02:32 -0700948 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800950EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952/**
Johannes Weiner0cd61442014-04-03 14:47:46 -0700953 * pagevec_lookup_entries - gang pagecache lookup
954 * @pvec: Where the resulting entries are placed
955 * @mapping: The address_space to search
956 * @start: The starting entry index
957 * @nr_entries: The maximum number of entries
958 * @indices: The cache indices corresponding to the entries in @pvec
959 *
960 * pagevec_lookup_entries() will search for and return a group of up
961 * to @nr_entries pages and shadow entries in the mapping. All
962 * entries are placed in @pvec. pagevec_lookup_entries() takes a
963 * reference against actual pages in @pvec.
964 *
965 * The search returns a group of mapping-contiguous entries with
966 * ascending indexes. There may be holes in the indices due to
967 * not-present entries.
968 *
969 * pagevec_lookup_entries() returns the number of entries which were
970 * found.
971 */
972unsigned pagevec_lookup_entries(struct pagevec *pvec,
973 struct address_space *mapping,
974 pgoff_t start, unsigned nr_pages,
975 pgoff_t *indices)
976{
977 pvec->nr = find_get_entries(mapping, start, nr_pages,
978 pvec->pages, indices);
979 return pagevec_count(pvec);
980}
981
982/**
983 * pagevec_remove_exceptionals - pagevec exceptionals pruning
984 * @pvec: The pagevec to prune
985 *
986 * pagevec_lookup_entries() fills both pages and exceptional radix
987 * tree entries into the pagevec. This function prunes all
988 * exceptionals from @pvec without leaving holes, so that it can be
989 * passed on to page-only pagevec operations.
990 */
991void pagevec_remove_exceptionals(struct pagevec *pvec)
992{
993 int i, j;
994
995 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
996 struct page *page = pvec->pages[i];
997 if (!radix_tree_exceptional_entry(page))
998 pvec->pages[j++] = page;
999 }
1000 pvec->nr = j;
1001}
1002
1003/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 * pagevec_lookup - gang pagecache lookup
1005 * @pvec: Where the resulting pages are placed
1006 * @mapping: The address_space to search
1007 * @start: The starting page index
1008 * @nr_pages: The maximum number of pages
1009 *
1010 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1011 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1012 * reference against the pages in @pvec.
1013 *
1014 * The search returns a group of mapping-contiguous pages with ascending
1015 * indexes. There may be holes in the indices due to not-present pages.
1016 *
1017 * pagevec_lookup() returns the number of pages which were found.
1018 */
1019unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1020 pgoff_t start, unsigned nr_pages)
1021{
1022 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1023 return pagevec_count(pvec);
1024}
Christoph Hellwig78539fd2006-01-11 20:47:41 +11001025EXPORT_SYMBOL(pagevec_lookup);
1026
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1028 pgoff_t *index, int tag, unsigned nr_pages)
1029{
1030 pvec->nr = find_get_pages_tag(mapping, index, tag,
1031 nr_pages, pvec->pages);
1032 return pagevec_count(pvec);
1033}
Steve French7f285702005-11-01 10:22:55 -08001034EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036/*
1037 * Perform any setup for the swap system
1038 */
1039void __init swap_setup(void)
1040{
Jan Beulich44813742009-09-21 17:03:05 -07001041 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001042#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -08001043 int i;
1044
Mikulas Patocka8077c0d2013-10-14 12:14:13 -04001045 if (bdi_init(swapper_spaces[0].backing_dev_info))
1046 panic("Failed to init swap bdi");
Shaohua Li33806f02013-02-22 16:34:37 -08001047 for (i = 0; i < MAX_SWAPFILES; i++) {
1048 spin_lock_init(&swapper_spaces[i].tree_lock);
1049 INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
1050 }
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001051#endif
1052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 /* Use a smaller cluster for small-memory machines */
1054 if (megs < 16)
1055 page_cluster = 2;
1056 else
1057 page_cluster = 3;
1058 /*
1059 * Right now other parts of the system means that we
1060 * _really_ don't want to cluster much more
1061 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062}