blob: 39395fb549c013e20b34d5fe0be8afa6eee5afbf [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>
Naoya Horiguchi822fc612015-04-15 16:14:35 -070034#include <linux/hugetlb.h>
Vladimir Davydov33c3fc72015-09-09 15:35:45 -070035#include <linux/page_idle.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Lee Schermerhorn64d65192008-10-18 20:26:52 -070037#include "internal.h"
38
Mel Gormanc6286c92013-07-03 15:02:26 -070039#define CREATE_TRACE_POINTS
40#include <trace/events/pagemap.h>
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/* How many pages do we try to swap or page in/out together? */
43int page_cluster;
44
Mel Gorman13f7f782013-07-03 15:02:28 -070045static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
Vegard Nossumf84f95042008-07-23 21:28:14 -070046static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
Minchan Kimcc5993b2015-04-15 16:13:26 -070047static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070048
Adrian Bunkb2213852006-09-25 23:31:02 -070049/*
50 * This path almost never happens for VM activity - pages are normally
51 * freed via pagevecs. But it gets used by networking.
52 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080053static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070054{
55 if (PageLRU(page)) {
Adrian Bunkb2213852006-09-25 23:31:02 -070056 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070057 struct lruvec *lruvec;
58 unsigned long flags;
Adrian Bunkb2213852006-09-25 23:31:02 -070059
60 spin_lock_irqsave(&zone->lru_lock, flags);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070061 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -080062 VM_BUG_ON_PAGE(!PageLRU(page), page);
Adrian Bunkb2213852006-09-25 23:31:02 -070063 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070064 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070065 spin_unlock_irqrestore(&zone->lru_lock, flags);
66 }
Johannes Weiner0a31bc92014-08-08 14:19:22 -070067 mem_cgroup_uncharge(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080068}
69
70static void __put_single_page(struct page *page)
71{
72 __page_cache_release(page);
Mel Gormanb745bc82014-06-04 16:10:22 -070073 free_hot_cold_page(page, false);
Adrian Bunkb2213852006-09-25 23:31:02 -070074}
75
Andrea Arcangeli91807062011-01-13 15:46:32 -080076static void __put_compound_page(struct page *page)
77{
78 compound_page_dtor *dtor;
79
Naoya Horiguchi822fc612015-04-15 16:14:35 -070080 /*
81 * __page_cache_release() is supposed to be called for thp, not for
82 * hugetlb. This is because hugetlb page does never have PageLRU set
83 * (it's never listed to any LRU lists) and no memcg routines should
84 * be called for hugetlb (it has a separate hugetlb_cgroup.)
85 */
86 if (!PageHuge(page))
87 __page_cache_release(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080088 dtor = get_compound_page_dtor(page);
89 (*dtor)(page);
90}
91
Jianyu Zhanc747ce72014-06-04 16:07:59 -070092/**
93 * Two special cases here: we could avoid taking compound_lock_irqsave
94 * and could skip the tail refcounting(in _mapcount).
95 *
96 * 1. Hugetlbfs page:
97 *
98 * PageHeadHuge will remain true until the compound page
99 * is released and enters the buddy allocator, and it could
100 * not be split by __split_huge_page_refcount().
101 *
102 * So if we see PageHeadHuge set, and we have the tail page pin,
103 * then we could safely put head page.
104 *
105 * 2. Slab THP page:
106 *
107 * PG_slab is cleared before the slab frees the head page, and
108 * tail pin cannot be the last reference left on the head page,
109 * because the slab code is free to reuse the compound page
110 * after a kfree/kmem_cache_free without having to check if
111 * there's any tail pin left. In turn all tail pinsmust be always
112 * released while the head is still pinned by the slab code
113 * and so we know PG_slab will be still set too.
114 *
115 * So if we see PageSlab set, and we have the tail page pin,
116 * then we could safely put head page.
117 */
118static __always_inline
119void put_unrefcounted_compound_page(struct page *page_head, struct page *page)
120{
121 /*
122 * If @page is a THP tail, we must read the tail page
123 * flags after the head page flags. The
124 * __split_huge_page_refcount side enforces write memory barriers
125 * between clearing PageTail and before the head page
126 * can be freed and reallocated.
127 */
128 smp_rmb();
129 if (likely(PageTail(page))) {
130 /*
131 * __split_huge_page_refcount cannot race
132 * here, see the comment above this function.
133 */
134 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
Jianyu Zhanc747ce72014-06-04 16:07:59 -0700135 if (put_page_testzero(page_head)) {
136 /*
137 * If this is the tail of a slab THP page,
138 * the tail pin must not be the last reference
139 * held on the page, because the PG_slab cannot
140 * be cleared before all tail pins (which skips
141 * the _mapcount tail refcounting) have been
142 * released.
143 *
144 * If this is the tail of a hugetlbfs page,
145 * the tail pin may be the last reference on
146 * the page instead, because PageHeadHuge will
147 * not go away until the compound page enters
148 * the buddy allocator.
149 */
150 VM_BUG_ON_PAGE(PageSlab(page_head), page_head);
151 __put_compound_page(page_head);
152 }
153 } else
154 /*
155 * __split_huge_page_refcount run before us,
156 * @page was a THP tail. The split @page_head
157 * has been freed and reallocated as slab or
158 * hugetlbfs page of smaller order (only
159 * possible if reallocated as slab on x86).
160 */
161 if (put_page_testzero(page))
162 __put_single_page(page);
163}
164
165static __always_inline
166void put_refcounted_compound_page(struct page *page_head, struct page *page)
167{
168 if (likely(page != page_head && get_page_unless_zero(page_head))) {
169 unsigned long flags;
170
171 /*
172 * @page_head wasn't a dangling pointer but it may not
173 * be a head page anymore by the time we obtain the
174 * lock. That is ok as long as it can't be freed from
175 * under us.
176 */
177 flags = compound_lock_irqsave(page_head);
178 if (unlikely(!PageTail(page))) {
179 /* __split_huge_page_refcount run before us */
180 compound_unlock_irqrestore(page_head, flags);
181 if (put_page_testzero(page_head)) {
182 /*
183 * The @page_head may have been freed
184 * and reallocated as a compound page
185 * of smaller order and then freed
186 * again. All we know is that it
187 * cannot have become: a THP page, a
188 * compound page of higher order, a
189 * tail page. That is because we
190 * still hold the refcount of the
191 * split THP tail and page_head was
192 * the THP head before the split.
193 */
194 if (PageHead(page_head))
195 __put_compound_page(page_head);
196 else
197 __put_single_page(page_head);
198 }
199out_put_single:
200 if (put_page_testzero(page))
201 __put_single_page(page);
202 return;
203 }
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800204 VM_BUG_ON_PAGE(page_head != compound_head(page), page);
Jianyu Zhanc747ce72014-06-04 16:07:59 -0700205 /*
206 * We can release the refcount taken by
207 * get_page_unless_zero() now that
208 * __split_huge_page_refcount() is blocked on the
209 * compound_lock.
210 */
211 if (put_page_testzero(page_head))
212 VM_BUG_ON_PAGE(1, page_head);
213 /* __split_huge_page_refcount will wait now */
214 VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page);
215 atomic_dec(&page->_mapcount);
216 VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head);
217 VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page);
218 compound_unlock_irqrestore(page_head, flags);
219
220 if (put_page_testzero(page_head)) {
221 if (PageHead(page_head))
222 __put_compound_page(page_head);
223 else
224 __put_single_page(page_head);
225 }
226 } else {
227 /* @page_head is a dangling pointer */
228 VM_BUG_ON_PAGE(PageTail(page), page);
229 goto out_put_single;
230 }
231}
232
Nick Piggin8519fb32006-02-07 12:58:52 -0800233static void put_compound_page(struct page *page)
234{
Andrew Morton26296ad2014-01-21 15:48:59 -0800235 struct page *page_head;
236
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700237 /*
238 * We see the PageCompound set and PageTail not set, so @page maybe:
239 * 1. hugetlbfs head page, or
240 * 2. THP head page.
241 */
Andrew Morton26296ad2014-01-21 15:48:59 -0800242 if (likely(!PageTail(page))) {
243 if (put_page_testzero(page)) {
244 /*
245 * By the time all refcounts have been released
246 * split_huge_page cannot run anymore from under us.
247 */
248 if (PageHead(page))
249 __put_compound_page(page);
250 else
251 __put_single_page(page);
252 }
253 return;
254 }
255
Andrew Morton26296ad2014-01-21 15:48:59 -0800256 /*
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700257 * We see the PageCompound set and PageTail set, so @page maybe:
258 * 1. a tail hugetlbfs page, or
259 * 2. a tail THP page, or
260 * 3. a split THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800261 *
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700262 * Case 3 is possible, as we may race with
263 * __split_huge_page_refcount tearing down a THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800264 */
Kirill A. Shutemov1d798ca2015-11-06 16:29:54 -0800265 page_head = compound_head(page);
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700266 if (!__compound_tail_refcounted(page_head))
267 put_unrefcounted_compound_page(page_head, page);
268 else
269 put_refcounted_compound_page(page_head, page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800270}
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272void put_page(struct page *page)
273{
Nick Piggin8519fb32006-02-07 12:58:52 -0800274 if (unlikely(PageCompound(page)))
275 put_compound_page(page);
276 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800277 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
279EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700281/*
282 * This function is exported but must not be called by anything other
283 * than get_page(). It implements the slow path of get_page().
284 */
285bool __get_page_tail(struct page *page)
286{
287 /*
288 * This takes care of get_page() if run on a tail page
289 * returned by one of the get_user_pages/follow_page variants.
290 * get_user_pages/follow_page itself doesn't need the compound
291 * lock because it runs __get_page_tail_foll() under the
292 * proper PT lock that already serializes against
293 * split_huge_page().
294 */
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800295 unsigned long flags;
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800296 bool got;
David Rientjes668f9abb2014-03-03 15:38:18 -0800297 struct page *page_head = compound_head(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700298
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800299 /* Ref to put_compound_page() comment. */
Andrea Arcangeli3bfcd132014-01-21 15:48:56 -0800300 if (!__compound_tail_refcounted(page_head)) {
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800301 smp_rmb();
302 if (likely(PageTail(page))) {
303 /*
304 * This is a hugetlbfs page or a slab
305 * page. __split_huge_page_refcount
306 * cannot race here.
307 */
Sasha Levin309381fea2014-01-23 15:52:54 -0800308 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800309 __get_page_tail_foll(page, true);
310 return true;
311 } else {
312 /*
313 * __split_huge_page_refcount run
314 * before us, "page" was a THP
315 * tail. The split page_head has been
316 * freed and reallocated as slab or
317 * hugetlbfs page of smaller order
318 * (only possible if reallocated as
319 * slab on x86).
320 */
321 return false;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700322 }
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800323 }
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800324
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800325 got = false;
326 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800327 /*
328 * page_head wasn't a dangling pointer but it
329 * may not be a head page anymore by the time
330 * we obtain the lock. That is ok as long as it
331 * can't be freed from under us.
332 */
333 flags = compound_lock_irqsave(page_head);
334 /* here __split_huge_page_refcount won't run anymore */
335 if (likely(PageTail(page))) {
336 __get_page_tail_foll(page, false);
337 got = true;
338 }
339 compound_unlock_irqrestore(page_head, flags);
340 if (unlikely(!got))
341 put_page(page_head);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700342 }
343 return got;
344}
345EXPORT_SYMBOL(__get_page_tail);
346
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700347/**
Randy Dunlap76824862008-03-19 17:00:40 -0700348 * put_pages_list() - release a list of pages
349 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700350 *
351 * Release a list of pages which are strung together on page.lru. Currently
352 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700353 */
354void put_pages_list(struct list_head *pages)
355{
356 while (!list_empty(pages)) {
357 struct page *victim;
358
359 victim = list_entry(pages->prev, struct page, lru);
360 list_del(&victim->lru);
361 page_cache_release(victim);
362 }
363}
364EXPORT_SYMBOL(put_pages_list);
365
Mel Gorman18022c52012-07-31 16:44:51 -0700366/*
367 * get_kernel_pages() - pin kernel pages in memory
368 * @kiov: An array of struct kvec structures
369 * @nr_segs: number of segments to pin
370 * @write: pinning for read/write, currently ignored
371 * @pages: array that receives pointers to the pages pinned.
372 * Should be at least nr_segs long.
373 *
374 * Returns number of pages pinned. This may be fewer than the number
375 * requested. If nr_pages is 0 or negative, returns 0. If no pages
376 * were pinned, returns -errno. Each page returned must be released
377 * with a put_page() call when it is finished with.
378 */
379int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
380 struct page **pages)
381{
382 int seg;
383
384 for (seg = 0; seg < nr_segs; seg++) {
385 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
386 return seg;
387
Mel Gorman5a178112012-07-31 16:45:02 -0700388 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700389 page_cache_get(pages[seg]);
390 }
391
392 return seg;
393}
394EXPORT_SYMBOL_GPL(get_kernel_pages);
395
396/*
397 * get_kernel_page() - pin a kernel page in memory
398 * @start: starting kernel address
399 * @write: pinning for read/write, currently ignored
400 * @pages: array that receives pointer to the page pinned.
401 * Must be at least nr_segs long.
402 *
403 * Returns 1 if page is pinned. If the page was not pinned, returns
404 * -errno. The page returned must be released with a put_page() call
405 * when it is finished with.
406 */
407int get_kernel_page(unsigned long start, int write, struct page **pages)
408{
409 const struct kvec kiov = {
410 .iov_base = (void *)start,
411 .iov_len = PAGE_SIZE
412 };
413
414 return get_kernel_pages(&kiov, 1, write, pages);
415}
416EXPORT_SYMBOL_GPL(get_kernel_page);
417
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700418static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700419 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
420 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700421{
422 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700423 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700424 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700425 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700426
427 for (i = 0; i < pagevec_count(pvec); i++) {
428 struct page *page = pvec->pages[i];
429 struct zone *pagezone = page_zone(page);
430
431 if (pagezone != zone) {
432 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700433 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700434 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700435 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700436 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700437
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700438 lruvec = mem_cgroup_page_lruvec(page, zone);
439 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700440 }
441 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700442 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800443 release_pages(pvec->pages, pvec->nr, pvec->cold);
444 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800445}
446
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700447static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
448 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700449{
450 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700451
452 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
453 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800454 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700455 (*pgmoved)++;
456 }
457}
458
459/*
460 * pagevec_move_tail() must be called with IRQ disabled.
461 * Otherwise this may cause nasty races.
462 */
463static void pagevec_move_tail(struct pagevec *pvec)
464{
465 int pgmoved = 0;
466
467 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
468 __count_vm_events(PGROTATED, pgmoved);
469}
470
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700471/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 * Writeback is about to end against a page which has been marked for immediate
473 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700474 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700476void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700478 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700479 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700480 struct pagevec *pvec;
481 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700483 page_cache_get(page);
484 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700485 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700486 if (!pagevec_add(pvec, page))
487 pagevec_move_tail(pvec);
488 local_irq_restore(flags);
489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700492static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800493 int file, int rotated)
494{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700495 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800496
497 reclaim_stat->recent_scanned[file]++;
498 if (rotated)
499 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800500}
501
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700502static void __activate_page(struct page *page, struct lruvec *lruvec,
503 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Linus Torvalds7a608572011-01-17 14:42:19 -0800505 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
506 int file = page_is_file_cache(page);
507 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800508
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700509 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800510 SetPageActive(page);
511 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700512 add_page_to_lru_list(page, lruvec, lru);
Mel Gorman24b7e582014-08-06 16:07:11 -0700513 trace_mm_lru_activate(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800514
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700515 __count_vm_event(PGACTIVATE);
516 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800517 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700518}
519
520#ifdef CONFIG_SMP
521static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
522
523static void activate_page_drain(int cpu)
524{
525 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
526
527 if (pagevec_count(pvec))
528 pagevec_lru_move_fn(pvec, __activate_page, NULL);
529}
530
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700531static bool need_activate_page_drain(int cpu)
532{
533 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
534}
535
Shaohua Lieb709b02011-05-24 17:12:55 -0700536void activate_page(struct page *page)
537{
538 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
539 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
540
541 page_cache_get(page);
542 if (!pagevec_add(pvec, page))
543 pagevec_lru_move_fn(pvec, __activate_page, NULL);
544 put_cpu_var(activate_page_pvecs);
545 }
546}
547
548#else
549static inline void activate_page_drain(int cpu)
550{
551}
552
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700553static bool need_activate_page_drain(int cpu)
554{
555 return false;
556}
557
Shaohua Lieb709b02011-05-24 17:12:55 -0700558void activate_page(struct page *page)
559{
560 struct zone *zone = page_zone(page);
561
562 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700563 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 spin_unlock_irq(&zone->lru_lock);
565}
Shaohua Lieb709b02011-05-24 17:12:55 -0700566#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Mel Gorman059285a2013-07-03 15:02:30 -0700568static void __lru_cache_activate_page(struct page *page)
569{
570 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
571 int i;
572
573 /*
574 * Search backwards on the optimistic assumption that the page being
575 * activated has just been added to this pagevec. Note that only
576 * the local pagevec is examined as a !PageLRU page could be in the
577 * process of being released, reclaimed, migrated or on a remote
578 * pagevec that is currently being drained. Furthermore, marking
579 * a remote pagevec's page PageActive potentially hits a race where
580 * a page is marked PageActive just after it is added to the inactive
581 * list causing accounting errors and BUG_ON checks to trigger.
582 */
583 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
584 struct page *pagevec_page = pvec->pages[i];
585
586 if (pagevec_page == page) {
587 SetPageActive(page);
588 break;
589 }
590 }
591
592 put_cpu_var(lru_add_pvec);
593}
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595/*
596 * Mark a page as having seen activity.
597 *
598 * inactive,unreferenced -> inactive,referenced
599 * inactive,referenced -> active,unreferenced
600 * active,unreferenced -> active,referenced
Hugh Dickinseb39d612014-08-06 16:06:43 -0700601 *
602 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
603 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800605void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700607 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700608 PageReferenced(page)) {
609
610 /*
611 * If the page is on the LRU, queue it for activation via
612 * activate_page_pvecs. Otherwise, assume the page is on a
613 * pagevec, mark it active and it'll be moved to the active
614 * LRU on the next drain.
615 */
616 if (PageLRU(page))
617 activate_page(page);
618 else
619 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700621 if (page_is_file_cache(page))
622 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 } else if (!PageReferenced(page)) {
624 SetPageReferenced(page);
625 }
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700626 if (page_is_idle(page))
627 clear_page_idle(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629EXPORT_SYMBOL(mark_page_accessed);
630
Jianyu Zhan2329d372014-06-04 16:07:31 -0700631static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Mel Gorman13f7f782013-07-03 15:02:28 -0700633 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700636 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700637 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700638 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700639 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700641
642/**
643 * lru_cache_add: add a page to the page lists
644 * @page: the page to add
645 */
646void lru_cache_add_anon(struct page *page)
647{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700648 if (PageActive(page))
649 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700650 __lru_cache_add(page);
651}
652
653void lru_cache_add_file(struct page *page)
654{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700655 if (PageActive(page))
656 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700657 __lru_cache_add(page);
658}
659EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700661/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700662 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700663 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700664 *
665 * Queue the page for addition to the LRU via pagevec. The decision on whether
666 * to add the page to the [in]active [file|anon] list is deferred until the
667 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
668 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700669 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700670void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671{
Sasha Levin309381fea2014-01-23 15:52:54 -0800672 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
673 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700674 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700677/**
678 * add_page_to_unevictable_list - add a page to the unevictable list
679 * @page: the page to be added to the unevictable list
680 *
681 * Add page directly to its zone's unevictable list. To avoid races with
682 * tasks that might be making the page evictable, through eg. munlock,
683 * munmap or exit, while it's not on the lru, we want to add the page
684 * while it's locked or otherwise "invisible" to other tasks. This is
685 * difficult to do when using the pagevec cache, so bypass that.
686 */
687void add_page_to_unevictable_list(struct page *page)
688{
689 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700690 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700691
692 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700693 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700694 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700695 SetPageUnevictable(page);
696 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700697 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700698 spin_unlock_irq(&zone->lru_lock);
699}
700
Johannes Weiner00501b52014-08-08 14:19:20 -0700701/**
702 * lru_cache_add_active_or_unevictable
703 * @page: the page to be added to LRU
704 * @vma: vma in which page is mapped for determining reclaimability
705 *
706 * Place @page on the active or unevictable LRU list, depending on its
707 * evictability. Note that if the page is not evictable, it goes
708 * directly back onto it's zone's unevictable list, it does NOT use a
709 * per cpu pagevec.
710 */
711void lru_cache_add_active_or_unevictable(struct page *page,
712 struct vm_area_struct *vma)
713{
714 VM_BUG_ON_PAGE(PageLRU(page), page);
715
716 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
717 SetPageActive(page);
718 lru_cache_add(page);
719 return;
720 }
721
722 if (!TestSetPageMlocked(page)) {
723 /*
724 * We use the irq-unsafe __mod_zone_page_stat because this
725 * counter is not modified from interrupt context, and the pte
726 * lock is held(spinlock), which implies preemption disabled.
727 */
728 __mod_zone_page_state(page_zone(page), NR_MLOCK,
729 hpage_nr_pages(page));
730 count_vm_event(UNEVICTABLE_PGMLOCKED);
731 }
732 add_page_to_unevictable_list(page);
733}
734
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700735/*
Minchan Kim31560182011-03-22 16:32:52 -0700736 * If the page can not be invalidated, it is moved to the
737 * inactive list to speed up its reclaim. It is moved to the
738 * head of the list, rather than the tail, to give the flusher
739 * threads some time to write it out, as this is much more
740 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700741 *
742 * If the page isn't page_mapped and dirty/writeback, the page
743 * could reclaim asap using PG_reclaim.
744 *
745 * 1. active, mapped page -> none
746 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
747 * 3. inactive, mapped page -> none
748 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
749 * 5. inactive, clean -> inactive, tail
750 * 6. Others -> none
751 *
752 * In 4, why it moves inactive's head, the VM expects the page would
753 * be write it out by flusher threads as this is much more effective
754 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700755 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700756static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700757 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700758{
759 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700760 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700761
Minchan Kim278df9f2011-03-22 16:32:54 -0700762 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700763 return;
764
Minchan Kimbad49d92011-05-11 15:13:30 -0700765 if (PageUnevictable(page))
766 return;
767
Minchan Kim31560182011-03-22 16:32:52 -0700768 /* Some processes are using the page */
769 if (page_mapped(page))
770 return;
771
Minchan Kim278df9f2011-03-22 16:32:54 -0700772 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700773 file = page_is_file_cache(page);
774 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700775
776 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700777 ClearPageActive(page);
778 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700779 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700780
Minchan Kim278df9f2011-03-22 16:32:54 -0700781 if (PageWriteback(page) || PageDirty(page)) {
782 /*
783 * PG_reclaim could be raced with end_page_writeback
784 * It can make readahead confusing. But race window
785 * is _really_ small and it's non-critical problem.
786 */
787 SetPageReclaim(page);
788 } else {
789 /*
790 * The page's writeback ends up during pagevec
791 * We moves tha page into tail of inactive.
792 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800793 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700794 __count_vm_event(PGROTATED);
795 }
796
797 if (active)
798 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700799 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700800}
801
Minchan Kim31560182011-03-22 16:32:52 -0700802/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700803 * Drain pages out of the cpu's pagevecs.
804 * Either "cpu" is the current CPU, and preemption has already been
805 * disabled; or "cpu" is being hot-unplugged, and is already dead.
806 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700807void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
Mel Gorman13f7f782013-07-03 15:02:28 -0700809 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
Mel Gorman13f7f782013-07-03 15:02:28 -0700811 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700812 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700813
814 pvec = &per_cpu(lru_rotate_pvecs, cpu);
815 if (pagevec_count(pvec)) {
816 unsigned long flags;
817
818 /* No harm done if a racing interrupt already did this */
819 local_irq_save(flags);
820 pagevec_move_tail(pvec);
821 local_irq_restore(flags);
822 }
Minchan Kim31560182011-03-22 16:32:52 -0700823
Minchan Kimcc5993b2015-04-15 16:13:26 -0700824 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700825 if (pagevec_count(pvec))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700826 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700827
828 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700829}
830
831/**
Minchan Kimcc5993b2015-04-15 16:13:26 -0700832 * deactivate_file_page - forcefully deactivate a file page
Minchan Kim31560182011-03-22 16:32:52 -0700833 * @page: page to deactivate
834 *
835 * This function hints the VM that @page is a good reclaim candidate,
836 * for example if its invalidation fails due to the page being dirty
837 * or under writeback.
838 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700839void deactivate_file_page(struct page *page)
Minchan Kim31560182011-03-22 16:32:52 -0700840{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700841 /*
Minchan Kimcc5993b2015-04-15 16:13:26 -0700842 * In a workload with many unevictable page such as mprotect,
843 * unevictable page deactivation for accelerating reclaim is pointless.
Minchan Kim821ed6b2011-05-24 17:12:31 -0700844 */
845 if (PageUnevictable(page))
846 return;
847
Minchan Kim31560182011-03-22 16:32:52 -0700848 if (likely(get_page_unless_zero(page))) {
Minchan Kimcc5993b2015-04-15 16:13:26 -0700849 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700850
851 if (!pagevec_add(pvec, page))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700852 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
853 put_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700854 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800855}
856
857void lru_add_drain(void)
858{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700859 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800860 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861}
862
David Howellsc4028952006-11-22 14:57:56 +0000863static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800864{
865 lru_add_drain();
866}
867
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700868static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
869
870void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800871{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700872 static DEFINE_MUTEX(lock);
873 static struct cpumask has_work;
874 int cpu;
875
876 mutex_lock(&lock);
877 get_online_cpus();
878 cpumask_clear(&has_work);
879
880 for_each_online_cpu(cpu) {
881 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
882
883 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
884 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
Minchan Kimcc5993b2015-04-15 16:13:26 -0700885 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700886 need_activate_page_drain(cpu)) {
887 INIT_WORK(work, lru_add_drain_per_cpu);
888 schedule_work_on(cpu, work);
889 cpumask_set_cpu(cpu, &has_work);
890 }
891 }
892
893 for_each_cpu(cpu, &has_work)
894 flush_work(&per_cpu(lru_add_drain_work, cpu));
895
896 put_online_cpus();
897 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800898}
899
Michal Hockoaabfb572014-10-09 15:28:52 -0700900/**
901 * release_pages - batched page_cache_release()
902 * @pages: array of pages to release
903 * @nr: number of pages
904 * @cold: whether the pages are cache cold
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 *
Michal Hockoaabfb572014-10-09 15:28:52 -0700906 * Decrement the reference count on all the pages in @pages. If it
907 * fell to zero, remove the page from the LRU and free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 */
Mel Gormanb745bc82014-06-04 16:10:22 -0700909void release_pages(struct page **pages, int nr, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
911 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800912 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700914 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700915 unsigned long uninitialized_var(flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700916 unsigned int uninitialized_var(lock_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 for (i = 0; i < nr; i++) {
919 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
Nick Piggin8519fb32006-02-07 12:58:52 -0800921 if (unlikely(PageCompound(page))) {
922 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700923 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800924 zone = NULL;
925 }
926 put_compound_page(page);
927 continue;
928 }
929
Michal Hockoaabfb572014-10-09 15:28:52 -0700930 /*
931 * Make sure the IRQ-safe lock-holding time does not get
932 * excessive with a continuous string of pages from the
933 * same zone. The lock is held only if zone != NULL.
934 */
935 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
936 spin_unlock_irqrestore(&zone->lru_lock, flags);
937 zone = NULL;
938 }
939
Nick Pigginb5810032005-10-29 18:16:12 -0700940 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 continue;
942
Nick Piggin46453a62006-03-22 00:07:58 -0800943 if (PageLRU(page)) {
944 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700945
Nick Piggin46453a62006-03-22 00:07:58 -0800946 if (pagezone != zone) {
947 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700948 spin_unlock_irqrestore(&zone->lru_lock,
949 flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700950 lock_batch = 0;
Nick Piggin46453a62006-03-22 00:07:58 -0800951 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700952 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700954
955 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -0800956 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800957 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700958 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
Nick Piggin46453a62006-03-22 00:07:58 -0800960
Mel Gormanc53954a2013-07-03 15:02:34 -0700961 /* Clear Active bit in case of parallel mark_page_accessed */
Mel Gormane3741b52014-06-04 16:10:26 -0700962 __ClearPageActive(page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700963
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800964 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 }
966 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700967 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Johannes Weiner747db952014-08-08 14:19:24 -0700969 mem_cgroup_uncharge_list(&pages_to_free);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800970 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700972EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
974/*
975 * The pages which we're about to release may be in the deferred lru-addition
976 * queues. That would prevent them from really being freed right now. That's
977 * OK from a correctness point of view but is inefficient - those pages may be
978 * cache-warm and we want to give them back to the page allocator ASAP.
979 *
980 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
981 * and __pagevec_lru_add_active() call release_pages() directly to avoid
982 * mutual recursion.
983 */
984void __pagevec_release(struct pagevec *pvec)
985{
986 lru_add_drain();
987 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
988 pagevec_reinit(pvec);
989}
Steve French7f285702005-11-01 10:22:55 -0800990EXPORT_SYMBOL(__pagevec_release);
991
Hugh Dickins12d27102012-01-12 17:19:52 -0800992#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800993/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700994void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700995 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800996{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800997 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800998
Sasha Levin309381fea2014-01-23 15:52:54 -0800999 VM_BUG_ON_PAGE(!PageHead(page), page);
1000 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
1001 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001002 VM_BUG_ON(NR_CPUS != 1 &&
1003 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001004
Shaohua Li5bc7b8a2013-04-29 15:08:36 -07001005 if (!list)
1006 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001007
Hugh Dickins12d27102012-01-12 17:19:52 -08001008 if (likely(PageLRU(page)))
1009 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -07001010 else if (list) {
1011 /* page reclaim is reclaiming a huge page */
1012 get_page(page_tail);
1013 list_add_tail(&page_tail->lru, list);
1014 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -08001015 struct list_head *list_head;
1016 /*
1017 * Head page has not yet been counted, as an hpage,
1018 * so we must account for each subpage individually.
1019 *
1020 * Use the standard add function to put page_tail on the list,
1021 * but then correct its position so they all end up in order.
1022 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -07001023 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -08001024 list_head = page_tail->lru.prev;
1025 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001026 }
Hugh Dickins75121022012-03-05 14:59:18 -08001027
1028 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -07001029 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001030}
Hugh Dickins12d27102012-01-12 17:19:52 -08001031#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001032
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001033static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
1034 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001035{
Mel Gorman13f7f782013-07-03 15:02:28 -07001036 int file = page_is_file_cache(page);
1037 int active = PageActive(page);
1038 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001039
Sasha Levin309381fea2014-01-23 15:52:54 -08001040 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001041
1042 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001043 add_page_to_lru_list(page, lruvec, lru);
1044 update_page_reclaim_stat(lruvec, file, active);
Mel Gorman24b7e582014-08-06 16:07:11 -07001045 trace_mm_lru_insertion(page, lru);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001046}
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 * Add the passed pages to the LRU, then drop the caller's refcount
1050 * on them. Reinitialises the caller's pagevec.
1051 */
Mel Gormana0b8cab32013-07-03 15:02:32 -07001052void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
Mel Gormana0b8cab32013-07-03 15:02:32 -07001054 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055}
Hugh Dickins5095ae832012-01-12 17:19:58 -08001056EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001059 * pagevec_lookup_entries - gang pagecache lookup
1060 * @pvec: Where the resulting entries are placed
1061 * @mapping: The address_space to search
1062 * @start: The starting entry index
1063 * @nr_entries: The maximum number of entries
1064 * @indices: The cache indices corresponding to the entries in @pvec
1065 *
1066 * pagevec_lookup_entries() will search for and return a group of up
1067 * to @nr_entries pages and shadow entries in the mapping. All
1068 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1069 * reference against actual pages in @pvec.
1070 *
1071 * The search returns a group of mapping-contiguous entries with
1072 * ascending indexes. There may be holes in the indices due to
1073 * not-present entries.
1074 *
1075 * pagevec_lookup_entries() returns the number of entries which were
1076 * found.
1077 */
1078unsigned pagevec_lookup_entries(struct pagevec *pvec,
1079 struct address_space *mapping,
1080 pgoff_t start, unsigned nr_pages,
1081 pgoff_t *indices)
1082{
1083 pvec->nr = find_get_entries(mapping, start, nr_pages,
1084 pvec->pages, indices);
1085 return pagevec_count(pvec);
1086}
1087
1088/**
1089 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1090 * @pvec: The pagevec to prune
1091 *
1092 * pagevec_lookup_entries() fills both pages and exceptional radix
1093 * tree entries into the pagevec. This function prunes all
1094 * exceptionals from @pvec without leaving holes, so that it can be
1095 * passed on to page-only pagevec operations.
1096 */
1097void pagevec_remove_exceptionals(struct pagevec *pvec)
1098{
1099 int i, j;
1100
1101 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1102 struct page *page = pvec->pages[i];
1103 if (!radix_tree_exceptional_entry(page))
1104 pvec->pages[j++] = page;
1105 }
1106 pvec->nr = j;
1107}
1108
1109/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 * pagevec_lookup - gang pagecache lookup
1111 * @pvec: Where the resulting pages are placed
1112 * @mapping: The address_space to search
1113 * @start: The starting page index
1114 * @nr_pages: The maximum number of pages
1115 *
1116 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1117 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1118 * reference against the pages in @pvec.
1119 *
1120 * The search returns a group of mapping-contiguous pages with ascending
1121 * indexes. There may be holes in the indices due to not-present pages.
1122 *
1123 * pagevec_lookup() returns the number of pages which were found.
1124 */
1125unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1126 pgoff_t start, unsigned nr_pages)
1127{
1128 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1129 return pagevec_count(pvec);
1130}
Christoph Hellwig78539fd2006-01-11 20:47:41 +11001131EXPORT_SYMBOL(pagevec_lookup);
1132
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1134 pgoff_t *index, int tag, unsigned nr_pages)
1135{
1136 pvec->nr = find_get_pages_tag(mapping, index, tag,
1137 nr_pages, pvec->pages);
1138 return pagevec_count(pvec);
1139}
Steve French7f285702005-11-01 10:22:55 -08001140EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142/*
1143 * Perform any setup for the swap system
1144 */
1145void __init swap_setup(void)
1146{
Jan Beulich44813742009-09-21 17:03:05 -07001147 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001148#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -08001149 int i;
1150
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08001151 for (i = 0; i < MAX_SWAPFILES; i++)
Shaohua Li33806f02013-02-22 16:34:37 -08001152 spin_lock_init(&swapper_spaces[i].tree_lock);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001153#endif
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 /* Use a smaller cluster for small-memory machines */
1156 if (megs < 16)
1157 page_cluster = 2;
1158 else
1159 page_cluster = 3;
1160 /*
1161 * Right now other parts of the system means that we
1162 * _really_ don't want to cluster much more
1163 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164}