blob: c8d6df556ce64600467ed92f8dbbb1b1639e4e28 [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
Jianyu Zhanc747ce72014-06-04 16:07:59 -070082/**
83 * Two special cases here: we could avoid taking compound_lock_irqsave
84 * and could skip the tail refcounting(in _mapcount).
85 *
86 * 1. Hugetlbfs page:
87 *
88 * PageHeadHuge will remain true until the compound page
89 * is released and enters the buddy allocator, and it could
90 * not be split by __split_huge_page_refcount().
91 *
92 * So if we see PageHeadHuge set, and we have the tail page pin,
93 * then we could safely put head page.
94 *
95 * 2. Slab THP page:
96 *
97 * PG_slab is cleared before the slab frees the head page, and
98 * tail pin cannot be the last reference left on the head page,
99 * because the slab code is free to reuse the compound page
100 * after a kfree/kmem_cache_free without having to check if
101 * there's any tail pin left. In turn all tail pinsmust be always
102 * released while the head is still pinned by the slab code
103 * and so we know PG_slab will be still set too.
104 *
105 * So if we see PageSlab set, and we have the tail page pin,
106 * then we could safely put head page.
107 */
108static __always_inline
109void put_unrefcounted_compound_page(struct page *page_head, struct page *page)
110{
111 /*
112 * If @page is a THP tail, we must read the tail page
113 * flags after the head page flags. The
114 * __split_huge_page_refcount side enforces write memory barriers
115 * between clearing PageTail and before the head page
116 * can be freed and reallocated.
117 */
118 smp_rmb();
119 if (likely(PageTail(page))) {
120 /*
121 * __split_huge_page_refcount cannot race
122 * here, see the comment above this function.
123 */
124 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
125 VM_BUG_ON_PAGE(page_mapcount(page) != 0, page);
126 if (put_page_testzero(page_head)) {
127 /*
128 * If this is the tail of a slab THP page,
129 * the tail pin must not be the last reference
130 * held on the page, because the PG_slab cannot
131 * be cleared before all tail pins (which skips
132 * the _mapcount tail refcounting) have been
133 * released.
134 *
135 * If this is the tail of a hugetlbfs page,
136 * the tail pin may be the last reference on
137 * the page instead, because PageHeadHuge will
138 * not go away until the compound page enters
139 * the buddy allocator.
140 */
141 VM_BUG_ON_PAGE(PageSlab(page_head), page_head);
142 __put_compound_page(page_head);
143 }
144 } else
145 /*
146 * __split_huge_page_refcount run before us,
147 * @page was a THP tail. The split @page_head
148 * has been freed and reallocated as slab or
149 * hugetlbfs page of smaller order (only
150 * possible if reallocated as slab on x86).
151 */
152 if (put_page_testzero(page))
153 __put_single_page(page);
154}
155
156static __always_inline
157void put_refcounted_compound_page(struct page *page_head, struct page *page)
158{
159 if (likely(page != page_head && get_page_unless_zero(page_head))) {
160 unsigned long flags;
161
162 /*
163 * @page_head wasn't a dangling pointer but it may not
164 * be a head page anymore by the time we obtain the
165 * lock. That is ok as long as it can't be freed from
166 * under us.
167 */
168 flags = compound_lock_irqsave(page_head);
169 if (unlikely(!PageTail(page))) {
170 /* __split_huge_page_refcount run before us */
171 compound_unlock_irqrestore(page_head, flags);
172 if (put_page_testzero(page_head)) {
173 /*
174 * The @page_head may have been freed
175 * and reallocated as a compound page
176 * of smaller order and then freed
177 * again. All we know is that it
178 * cannot have become: a THP page, a
179 * compound page of higher order, a
180 * tail page. That is because we
181 * still hold the refcount of the
182 * split THP tail and page_head was
183 * the THP head before the split.
184 */
185 if (PageHead(page_head))
186 __put_compound_page(page_head);
187 else
188 __put_single_page(page_head);
189 }
190out_put_single:
191 if (put_page_testzero(page))
192 __put_single_page(page);
193 return;
194 }
195 VM_BUG_ON_PAGE(page_head != page->first_page, page);
196 /*
197 * We can release the refcount taken by
198 * get_page_unless_zero() now that
199 * __split_huge_page_refcount() is blocked on the
200 * compound_lock.
201 */
202 if (put_page_testzero(page_head))
203 VM_BUG_ON_PAGE(1, page_head);
204 /* __split_huge_page_refcount will wait now */
205 VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page);
206 atomic_dec(&page->_mapcount);
207 VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head);
208 VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page);
209 compound_unlock_irqrestore(page_head, flags);
210
211 if (put_page_testzero(page_head)) {
212 if (PageHead(page_head))
213 __put_compound_page(page_head);
214 else
215 __put_single_page(page_head);
216 }
217 } else {
218 /* @page_head is a dangling pointer */
219 VM_BUG_ON_PAGE(PageTail(page), page);
220 goto out_put_single;
221 }
222}
223
Nick Piggin8519fb32006-02-07 12:58:52 -0800224static void put_compound_page(struct page *page)
225{
Andrew Morton26296ad2014-01-21 15:48:59 -0800226 struct page *page_head;
227
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700228 /*
229 * We see the PageCompound set and PageTail not set, so @page maybe:
230 * 1. hugetlbfs head page, or
231 * 2. THP head page.
232 */
Andrew Morton26296ad2014-01-21 15:48:59 -0800233 if (likely(!PageTail(page))) {
234 if (put_page_testzero(page)) {
235 /*
236 * By the time all refcounts have been released
237 * split_huge_page cannot run anymore from under us.
238 */
239 if (PageHead(page))
240 __put_compound_page(page);
241 else
242 __put_single_page(page);
243 }
244 return;
245 }
246
Andrew Morton26296ad2014-01-21 15:48:59 -0800247 /*
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700248 * We see the PageCompound set and PageTail set, so @page maybe:
249 * 1. a tail hugetlbfs page, or
250 * 2. a tail THP page, or
251 * 3. a split THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800252 *
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700253 * Case 3 is possible, as we may race with
254 * __split_huge_page_refcount tearing down a THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800255 */
Jianyu Zhand2ee40e2014-06-04 16:08:02 -0700256 page_head = compound_head_by_tail(page);
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700257 if (!__compound_tail_refcounted(page_head))
258 put_unrefcounted_compound_page(page_head, page);
259 else
260 put_refcounted_compound_page(page_head, page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800261}
262
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263void put_page(struct page *page)
264{
Nick Piggin8519fb32006-02-07 12:58:52 -0800265 if (unlikely(PageCompound(page)))
266 put_compound_page(page);
267 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800268 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269}
270EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700272/*
273 * This function is exported but must not be called by anything other
274 * than get_page(). It implements the slow path of get_page().
275 */
276bool __get_page_tail(struct page *page)
277{
278 /*
279 * This takes care of get_page() if run on a tail page
280 * returned by one of the get_user_pages/follow_page variants.
281 * get_user_pages/follow_page itself doesn't need the compound
282 * lock because it runs __get_page_tail_foll() under the
283 * proper PT lock that already serializes against
284 * split_huge_page().
285 */
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800286 unsigned long flags;
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800287 bool got;
David Rientjes668f9abb2014-03-03 15:38:18 -0800288 struct page *page_head = compound_head(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700289
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800290 /* Ref to put_compound_page() comment. */
Andrea Arcangeli3bfcd132014-01-21 15:48:56 -0800291 if (!__compound_tail_refcounted(page_head)) {
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800292 smp_rmb();
293 if (likely(PageTail(page))) {
294 /*
295 * This is a hugetlbfs page or a slab
296 * page. __split_huge_page_refcount
297 * cannot race here.
298 */
Sasha Levin309381fea2014-01-23 15:52:54 -0800299 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800300 __get_page_tail_foll(page, true);
301 return true;
302 } else {
303 /*
304 * __split_huge_page_refcount run
305 * before us, "page" was a THP
306 * tail. The split page_head has been
307 * freed and reallocated as slab or
308 * hugetlbfs page of smaller order
309 * (only possible if reallocated as
310 * slab on x86).
311 */
312 return false;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700313 }
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800314 }
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800315
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800316 got = false;
317 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800318 /*
319 * page_head wasn't a dangling pointer but it
320 * may not be a head page anymore by the time
321 * we obtain the lock. That is ok as long as it
322 * can't be freed from under us.
323 */
324 flags = compound_lock_irqsave(page_head);
325 /* here __split_huge_page_refcount won't run anymore */
326 if (likely(PageTail(page))) {
327 __get_page_tail_foll(page, false);
328 got = true;
329 }
330 compound_unlock_irqrestore(page_head, flags);
331 if (unlikely(!got))
332 put_page(page_head);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700333 }
334 return got;
335}
336EXPORT_SYMBOL(__get_page_tail);
337
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700338/**
Randy Dunlap76824862008-03-19 17:00:40 -0700339 * put_pages_list() - release a list of pages
340 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700341 *
342 * Release a list of pages which are strung together on page.lru. Currently
343 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700344 */
345void put_pages_list(struct list_head *pages)
346{
347 while (!list_empty(pages)) {
348 struct page *victim;
349
350 victim = list_entry(pages->prev, struct page, lru);
351 list_del(&victim->lru);
352 page_cache_release(victim);
353 }
354}
355EXPORT_SYMBOL(put_pages_list);
356
Mel Gorman18022c52012-07-31 16:44:51 -0700357/*
358 * get_kernel_pages() - pin kernel pages in memory
359 * @kiov: An array of struct kvec structures
360 * @nr_segs: number of segments to pin
361 * @write: pinning for read/write, currently ignored
362 * @pages: array that receives pointers to the pages pinned.
363 * Should be at least nr_segs long.
364 *
365 * Returns number of pages pinned. This may be fewer than the number
366 * requested. If nr_pages is 0 or negative, returns 0. If no pages
367 * were pinned, returns -errno. Each page returned must be released
368 * with a put_page() call when it is finished with.
369 */
370int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
371 struct page **pages)
372{
373 int seg;
374
375 for (seg = 0; seg < nr_segs; seg++) {
376 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
377 return seg;
378
Mel Gorman5a178112012-07-31 16:45:02 -0700379 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700380 page_cache_get(pages[seg]);
381 }
382
383 return seg;
384}
385EXPORT_SYMBOL_GPL(get_kernel_pages);
386
387/*
388 * get_kernel_page() - pin a kernel page in memory
389 * @start: starting kernel address
390 * @write: pinning for read/write, currently ignored
391 * @pages: array that receives pointer to the page pinned.
392 * Must be at least nr_segs long.
393 *
394 * Returns 1 if page is pinned. If the page was not pinned, returns
395 * -errno. The page returned must be released with a put_page() call
396 * when it is finished with.
397 */
398int get_kernel_page(unsigned long start, int write, struct page **pages)
399{
400 const struct kvec kiov = {
401 .iov_base = (void *)start,
402 .iov_len = PAGE_SIZE
403 };
404
405 return get_kernel_pages(&kiov, 1, write, pages);
406}
407EXPORT_SYMBOL_GPL(get_kernel_page);
408
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700409static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700410 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
411 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700412{
413 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700414 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700415 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700416 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700417
418 for (i = 0; i < pagevec_count(pvec); i++) {
419 struct page *page = pvec->pages[i];
420 struct zone *pagezone = page_zone(page);
421
422 if (pagezone != zone) {
423 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700424 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700425 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700426 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700427 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700428
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700429 lruvec = mem_cgroup_page_lruvec(page, zone);
430 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700431 }
432 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700433 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800434 release_pages(pvec->pages, pvec->nr, pvec->cold);
435 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800436}
437
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700438static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
439 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700440{
441 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700442
443 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
444 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800445 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700446 (*pgmoved)++;
447 }
448}
449
450/*
451 * pagevec_move_tail() must be called with IRQ disabled.
452 * Otherwise this may cause nasty races.
453 */
454static void pagevec_move_tail(struct pagevec *pvec)
455{
456 int pgmoved = 0;
457
458 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
459 __count_vm_events(PGROTATED, pgmoved);
460}
461
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700462/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 * Writeback is about to end against a page which has been marked for immediate
464 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700465 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700467void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700469 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700470 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700471 struct pagevec *pvec;
472 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700474 page_cache_get(page);
475 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700476 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700477 if (!pagevec_add(pvec, page))
478 pagevec_move_tail(pvec);
479 local_irq_restore(flags);
480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481}
482
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700483static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800484 int file, int rotated)
485{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700486 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800487
488 reclaim_stat->recent_scanned[file]++;
489 if (rotated)
490 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800491}
492
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700493static void __activate_page(struct page *page, struct lruvec *lruvec,
494 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Linus Torvalds7a608572011-01-17 14:42:19 -0800496 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
497 int file = page_is_file_cache(page);
498 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800499
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700500 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800501 SetPageActive(page);
502 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700503 add_page_to_lru_list(page, lruvec, lru);
Mel Gormanc6286c92013-07-03 15:02:26 -0700504 trace_mm_lru_activate(page, page_to_pfn(page));
Linus Torvalds7a608572011-01-17 14:42:19 -0800505
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700506 __count_vm_event(PGACTIVATE);
507 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800508 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700509}
510
511#ifdef CONFIG_SMP
512static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
513
514static void activate_page_drain(int cpu)
515{
516 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
517
518 if (pagevec_count(pvec))
519 pagevec_lru_move_fn(pvec, __activate_page, NULL);
520}
521
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700522static bool need_activate_page_drain(int cpu)
523{
524 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
525}
526
Shaohua Lieb709b02011-05-24 17:12:55 -0700527void activate_page(struct page *page)
528{
529 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
530 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
531
532 page_cache_get(page);
533 if (!pagevec_add(pvec, page))
534 pagevec_lru_move_fn(pvec, __activate_page, NULL);
535 put_cpu_var(activate_page_pvecs);
536 }
537}
538
539#else
540static inline void activate_page_drain(int cpu)
541{
542}
543
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700544static bool need_activate_page_drain(int cpu)
545{
546 return false;
547}
548
Shaohua Lieb709b02011-05-24 17:12:55 -0700549void activate_page(struct page *page)
550{
551 struct zone *zone = page_zone(page);
552
553 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700554 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 spin_unlock_irq(&zone->lru_lock);
556}
Shaohua Lieb709b02011-05-24 17:12:55 -0700557#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Mel Gorman059285a2013-07-03 15:02:30 -0700559static void __lru_cache_activate_page(struct page *page)
560{
561 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
562 int i;
563
564 /*
565 * Search backwards on the optimistic assumption that the page being
566 * activated has just been added to this pagevec. Note that only
567 * the local pagevec is examined as a !PageLRU page could be in the
568 * process of being released, reclaimed, migrated or on a remote
569 * pagevec that is currently being drained. Furthermore, marking
570 * a remote pagevec's page PageActive potentially hits a race where
571 * a page is marked PageActive just after it is added to the inactive
572 * list causing accounting errors and BUG_ON checks to trigger.
573 */
574 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
575 struct page *pagevec_page = pvec->pages[i];
576
577 if (pagevec_page == page) {
578 SetPageActive(page);
579 break;
580 }
581 }
582
583 put_cpu_var(lru_add_pvec);
584}
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586/*
587 * Mark a page as having seen activity.
588 *
589 * inactive,unreferenced -> inactive,referenced
590 * inactive,referenced -> active,unreferenced
591 * active,unreferenced -> active,referenced
592 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800593void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700595 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700596 PageReferenced(page)) {
597
598 /*
599 * If the page is on the LRU, queue it for activation via
600 * activate_page_pvecs. Otherwise, assume the page is on a
601 * pagevec, mark it active and it'll be moved to the active
602 * LRU on the next drain.
603 */
604 if (PageLRU(page))
605 activate_page(page);
606 else
607 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700609 if (page_is_file_cache(page))
610 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 } else if (!PageReferenced(page)) {
612 SetPageReferenced(page);
613 }
614}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615EXPORT_SYMBOL(mark_page_accessed);
616
Jianyu Zhan2329d372014-06-04 16:07:31 -0700617static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
Mel Gorman13f7f782013-07-03 15:02:28 -0700619 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700622 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700623 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700624 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700625 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700627
628/**
629 * lru_cache_add: add a page to the page lists
630 * @page: the page to add
631 */
632void lru_cache_add_anon(struct page *page)
633{
634 ClearPageActive(page);
635 __lru_cache_add(page);
636}
637
638void lru_cache_add_file(struct page *page)
639{
640 ClearPageActive(page);
641 __lru_cache_add(page);
642}
643EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700645/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700646 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700647 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700648 *
649 * Queue the page for addition to the LRU via pagevec. The decision on whether
650 * to add the page to the [in]active [file|anon] list is deferred until the
651 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
652 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700653 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700654void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
Sasha Levin309381fea2014-01-23 15:52:54 -0800656 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
657 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700658 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
660
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700661/**
662 * add_page_to_unevictable_list - add a page to the unevictable list
663 * @page: the page to be added to the unevictable list
664 *
665 * Add page directly to its zone's unevictable list. To avoid races with
666 * tasks that might be making the page evictable, through eg. munlock,
667 * munmap or exit, while it's not on the lru, we want to add the page
668 * while it's locked or otherwise "invisible" to other tasks. This is
669 * difficult to do when using the pagevec cache, so bypass that.
670 */
671void add_page_to_unevictable_list(struct page *page)
672{
673 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700674 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700675
676 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700677 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700678 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700679 SetPageUnevictable(page);
680 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700681 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700682 spin_unlock_irq(&zone->lru_lock);
683}
684
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700685/*
Minchan Kim31560182011-03-22 16:32:52 -0700686 * If the page can not be invalidated, it is moved to the
687 * inactive list to speed up its reclaim. It is moved to the
688 * head of the list, rather than the tail, to give the flusher
689 * threads some time to write it out, as this is much more
690 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700691 *
692 * If the page isn't page_mapped and dirty/writeback, the page
693 * could reclaim asap using PG_reclaim.
694 *
695 * 1. active, mapped page -> none
696 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
697 * 3. inactive, mapped page -> none
698 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
699 * 5. inactive, clean -> inactive, tail
700 * 6. Others -> none
701 *
702 * In 4, why it moves inactive's head, the VM expects the page would
703 * be write it out by flusher threads as this is much more effective
704 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700705 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700706static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
707 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700708{
709 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700710 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700711
Minchan Kim278df9f2011-03-22 16:32:54 -0700712 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700713 return;
714
Minchan Kimbad49d92011-05-11 15:13:30 -0700715 if (PageUnevictable(page))
716 return;
717
Minchan Kim31560182011-03-22 16:32:52 -0700718 /* Some processes are using the page */
719 if (page_mapped(page))
720 return;
721
Minchan Kim278df9f2011-03-22 16:32:54 -0700722 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700723 file = page_is_file_cache(page);
724 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700725
726 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700727 ClearPageActive(page);
728 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700729 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700730
Minchan Kim278df9f2011-03-22 16:32:54 -0700731 if (PageWriteback(page) || PageDirty(page)) {
732 /*
733 * PG_reclaim could be raced with end_page_writeback
734 * It can make readahead confusing. But race window
735 * is _really_ small and it's non-critical problem.
736 */
737 SetPageReclaim(page);
738 } else {
739 /*
740 * The page's writeback ends up during pagevec
741 * We moves tha page into tail of inactive.
742 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800743 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700744 __count_vm_event(PGROTATED);
745 }
746
747 if (active)
748 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700749 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700750}
751
Minchan Kim31560182011-03-22 16:32:52 -0700752/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700753 * Drain pages out of the cpu's pagevecs.
754 * Either "cpu" is the current CPU, and preemption has already been
755 * disabled; or "cpu" is being hot-unplugged, and is already dead.
756 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700757void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
Mel Gorman13f7f782013-07-03 15:02:28 -0700759 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Mel Gorman13f7f782013-07-03 15:02:28 -0700761 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700762 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700763
764 pvec = &per_cpu(lru_rotate_pvecs, cpu);
765 if (pagevec_count(pvec)) {
766 unsigned long flags;
767
768 /* No harm done if a racing interrupt already did this */
769 local_irq_save(flags);
770 pagevec_move_tail(pvec);
771 local_irq_restore(flags);
772 }
Minchan Kim31560182011-03-22 16:32:52 -0700773
774 pvec = &per_cpu(lru_deactivate_pvecs, cpu);
775 if (pagevec_count(pvec))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700776 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700777
778 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700779}
780
781/**
782 * deactivate_page - forcefully deactivate a page
783 * @page: page to deactivate
784 *
785 * This function hints the VM that @page is a good reclaim candidate,
786 * for example if its invalidation fails due to the page being dirty
787 * or under writeback.
788 */
789void deactivate_page(struct page *page)
790{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700791 /*
792 * In a workload with many unevictable page such as mprotect, unevictable
793 * page deactivation for accelerating reclaim is pointless.
794 */
795 if (PageUnevictable(page))
796 return;
797
Minchan Kim31560182011-03-22 16:32:52 -0700798 if (likely(get_page_unless_zero(page))) {
799 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
800
801 if (!pagevec_add(pvec, page))
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700802 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
Minchan Kim31560182011-03-22 16:32:52 -0700803 put_cpu_var(lru_deactivate_pvecs);
804 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800805}
806
807void lru_add_drain(void)
808{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700809 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800810 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
David Howellsc4028952006-11-22 14:57:56 +0000813static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800814{
815 lru_add_drain();
816}
817
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700818static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
819
820void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800821{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700822 static DEFINE_MUTEX(lock);
823 static struct cpumask has_work;
824 int cpu;
825
826 mutex_lock(&lock);
827 get_online_cpus();
828 cpumask_clear(&has_work);
829
830 for_each_online_cpu(cpu) {
831 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
832
833 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
834 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
835 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
836 need_activate_page_drain(cpu)) {
837 INIT_WORK(work, lru_add_drain_per_cpu);
838 schedule_work_on(cpu, work);
839 cpumask_set_cpu(cpu, &has_work);
840 }
841 }
842
843 for_each_cpu(cpu, &has_work)
844 flush_work(&per_cpu(lru_add_drain_work, cpu));
845
846 put_online_cpus();
847 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800848}
849
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 * Batched page_cache_release(). Decrement the reference count on all the
852 * passed pages. If it fell to zero then remove the page from the LRU and
853 * free it.
854 *
855 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
856 * for the remainder of the operation.
857 *
Fernando Luis Vazquez Caoab33dc02008-07-29 22:33:40 -0700858 * The locking in this function is against shrink_inactive_list(): we recheck
859 * the page count inside the lock to see whether shrink_inactive_list()
860 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
861 * will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 */
863void release_pages(struct page **pages, int nr, int cold)
864{
865 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800866 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700868 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700869 unsigned long uninitialized_var(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 for (i = 0; i < nr; i++) {
872 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Nick Piggin8519fb32006-02-07 12:58:52 -0800874 if (unlikely(PageCompound(page))) {
875 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700876 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800877 zone = NULL;
878 }
879 put_compound_page(page);
880 continue;
881 }
882
Nick Pigginb5810032005-10-29 18:16:12 -0700883 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 continue;
885
Nick Piggin46453a62006-03-22 00:07:58 -0800886 if (PageLRU(page)) {
887 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700888
Nick Piggin46453a62006-03-22 00:07:58 -0800889 if (pagezone != zone) {
890 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700891 spin_unlock_irqrestore(&zone->lru_lock,
892 flags);
Nick Piggin46453a62006-03-22 00:07:58 -0800893 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700894 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700896
897 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -0800898 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800899 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700900 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 }
Nick Piggin46453a62006-03-22 00:07:58 -0800902
Mel Gormanc53954a2013-07-03 15:02:34 -0700903 /* Clear Active bit in case of parallel mark_page_accessed */
904 ClearPageActive(page);
905
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800906 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
908 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700909 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800911 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700913EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915/*
916 * The pages which we're about to release may be in the deferred lru-addition
917 * queues. That would prevent them from really being freed right now. That's
918 * OK from a correctness point of view but is inefficient - those pages may be
919 * cache-warm and we want to give them back to the page allocator ASAP.
920 *
921 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
922 * and __pagevec_lru_add_active() call release_pages() directly to avoid
923 * mutual recursion.
924 */
925void __pagevec_release(struct pagevec *pvec)
926{
927 lru_add_drain();
928 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
929 pagevec_reinit(pvec);
930}
Steve French7f285702005-11-01 10:22:55 -0800931EXPORT_SYMBOL(__pagevec_release);
932
Hugh Dickins12d27102012-01-12 17:19:52 -0800933#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800934/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700935void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700936 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800937{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800938 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800939
Sasha Levin309381fea2014-01-23 15:52:54 -0800940 VM_BUG_ON_PAGE(!PageHead(page), page);
941 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
942 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700943 VM_BUG_ON(NR_CPUS != 1 &&
944 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800945
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700946 if (!list)
947 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800948
Hugh Dickins12d27102012-01-12 17:19:52 -0800949 if (likely(PageLRU(page)))
950 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700951 else if (list) {
952 /* page reclaim is reclaiming a huge page */
953 get_page(page_tail);
954 list_add_tail(&page_tail->lru, list);
955 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -0800956 struct list_head *list_head;
957 /*
958 * Head page has not yet been counted, as an hpage,
959 * so we must account for each subpage individually.
960 *
961 * Use the standard add function to put page_tail on the list,
962 * but then correct its position so they all end up in order.
963 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700964 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -0800965 list_head = page_tail->lru.prev;
966 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800967 }
Hugh Dickins75121022012-03-05 14:59:18 -0800968
969 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -0700970 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800971}
Hugh Dickins12d27102012-01-12 17:19:52 -0800972#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800973
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700974static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
975 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700976{
Mel Gorman13f7f782013-07-03 15:02:28 -0700977 int file = page_is_file_cache(page);
978 int active = PageActive(page);
979 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700980
Sasha Levin309381fea2014-01-23 15:52:54 -0800981 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700982
983 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700984 add_page_to_lru_list(page, lruvec, lru);
985 update_page_reclaim_stat(lruvec, file, active);
Mel Gormanc6286c92013-07-03 15:02:26 -0700986 trace_mm_lru_insertion(page, page_to_pfn(page), lru, trace_pagemap_flags(page));
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700987}
988
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 * Add the passed pages to the LRU, then drop the caller's refcount
991 * on them. Reinitialises the caller's pagevec.
992 */
Mel Gormana0b8cab32013-07-03 15:02:32 -0700993void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
Mel Gormana0b8cab32013-07-03 15:02:32 -0700995 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
Hugh Dickins5095ae832012-01-12 17:19:58 -0800997EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001000 * pagevec_lookup_entries - gang pagecache lookup
1001 * @pvec: Where the resulting entries are placed
1002 * @mapping: The address_space to search
1003 * @start: The starting entry index
1004 * @nr_entries: The maximum number of entries
1005 * @indices: The cache indices corresponding to the entries in @pvec
1006 *
1007 * pagevec_lookup_entries() will search for and return a group of up
1008 * to @nr_entries pages and shadow entries in the mapping. All
1009 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1010 * reference against actual pages in @pvec.
1011 *
1012 * The search returns a group of mapping-contiguous entries with
1013 * ascending indexes. There may be holes in the indices due to
1014 * not-present entries.
1015 *
1016 * pagevec_lookup_entries() returns the number of entries which were
1017 * found.
1018 */
1019unsigned pagevec_lookup_entries(struct pagevec *pvec,
1020 struct address_space *mapping,
1021 pgoff_t start, unsigned nr_pages,
1022 pgoff_t *indices)
1023{
1024 pvec->nr = find_get_entries(mapping, start, nr_pages,
1025 pvec->pages, indices);
1026 return pagevec_count(pvec);
1027}
1028
1029/**
1030 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1031 * @pvec: The pagevec to prune
1032 *
1033 * pagevec_lookup_entries() fills both pages and exceptional radix
1034 * tree entries into the pagevec. This function prunes all
1035 * exceptionals from @pvec without leaving holes, so that it can be
1036 * passed on to page-only pagevec operations.
1037 */
1038void pagevec_remove_exceptionals(struct pagevec *pvec)
1039{
1040 int i, j;
1041
1042 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1043 struct page *page = pvec->pages[i];
1044 if (!radix_tree_exceptional_entry(page))
1045 pvec->pages[j++] = page;
1046 }
1047 pvec->nr = j;
1048}
1049
1050/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 * pagevec_lookup - gang pagecache lookup
1052 * @pvec: Where the resulting pages are placed
1053 * @mapping: The address_space to search
1054 * @start: The starting page index
1055 * @nr_pages: The maximum number of pages
1056 *
1057 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1058 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1059 * reference against the pages in @pvec.
1060 *
1061 * The search returns a group of mapping-contiguous pages with ascending
1062 * indexes. There may be holes in the indices due to not-present pages.
1063 *
1064 * pagevec_lookup() returns the number of pages which were found.
1065 */
1066unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1067 pgoff_t start, unsigned nr_pages)
1068{
1069 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1070 return pagevec_count(pvec);
1071}
Christoph Hellwig78539fd2006-01-11 20:47:41 +11001072EXPORT_SYMBOL(pagevec_lookup);
1073
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1075 pgoff_t *index, int tag, unsigned nr_pages)
1076{
1077 pvec->nr = find_get_pages_tag(mapping, index, tag,
1078 nr_pages, pvec->pages);
1079 return pagevec_count(pvec);
1080}
Steve French7f285702005-11-01 10:22:55 -08001081EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083/*
1084 * Perform any setup for the swap system
1085 */
1086void __init swap_setup(void)
1087{
Jan Beulich44813742009-09-21 17:03:05 -07001088 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001089#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -08001090 int i;
1091
Mikulas Patocka8077c0d2013-10-14 12:14:13 -04001092 if (bdi_init(swapper_spaces[0].backing_dev_info))
1093 panic("Failed to init swap bdi");
Shaohua Li33806f02013-02-22 16:34:37 -08001094 for (i = 0; i < MAX_SWAPFILES; i++) {
1095 spin_lock_init(&swapper_spaces[i].tree_lock);
1096 INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
1097 }
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001098#endif
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 /* Use a smaller cluster for small-memory machines */
1101 if (megs < 16)
1102 page_cluster = 2;
1103 else
1104 page_cluster = 3;
1105 /*
1106 * Right now other parts of the system means that we
1107 * _really_ don't want to cluster much more
1108 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109}