blob: a3a0a2f1f7c3dc48c43494b949af6aee66adcf8f [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Lee Schermerhorn64d65192008-10-18 20:26:52 -070036#include "internal.h"
37
Mel Gormanc6286c92013-07-03 15:02:26 -070038#define CREATE_TRACE_POINTS
39#include <trace/events/pagemap.h>
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/* How many pages do we try to swap or page in/out together? */
42int page_cluster;
43
Mel Gorman13f7f782013-07-03 15:02:28 -070044static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
Vegard Nossumf84f95042008-07-23 21:28:14 -070045static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
Minchan Kimcc5993b2015-04-15 16:13:26 -070046static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -070047
Adrian Bunkb2213852006-09-25 23:31:02 -070048/*
49 * This path almost never happens for VM activity - pages are normally
50 * freed via pagevecs. But it gets used by networking.
51 */
Harvey Harrison920c7a52008-02-04 22:29:26 -080052static void __page_cache_release(struct page *page)
Adrian Bunkb2213852006-09-25 23:31:02 -070053{
54 if (PageLRU(page)) {
Adrian Bunkb2213852006-09-25 23:31:02 -070055 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070056 struct lruvec *lruvec;
57 unsigned long flags;
Adrian Bunkb2213852006-09-25 23:31:02 -070058
59 spin_lock_irqsave(&zone->lru_lock, flags);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070060 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -080061 VM_BUG_ON_PAGE(!PageLRU(page), page);
Adrian Bunkb2213852006-09-25 23:31:02 -070062 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -070063 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Adrian Bunkb2213852006-09-25 23:31:02 -070064 spin_unlock_irqrestore(&zone->lru_lock, flags);
65 }
Johannes Weiner0a31bc92014-08-08 14:19:22 -070066 mem_cgroup_uncharge(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080067}
68
69static void __put_single_page(struct page *page)
70{
71 __page_cache_release(page);
Mel Gormanb745bc82014-06-04 16:10:22 -070072 free_hot_cold_page(page, false);
Adrian Bunkb2213852006-09-25 23:31:02 -070073}
74
Andrea Arcangeli91807062011-01-13 15:46:32 -080075static void __put_compound_page(struct page *page)
76{
77 compound_page_dtor *dtor;
78
Naoya Horiguchi822fc612015-04-15 16:14:35 -070079 /*
80 * __page_cache_release() is supposed to be called for thp, not for
81 * hugetlb. This is because hugetlb page does never have PageLRU set
82 * (it's never listed to any LRU lists) and no memcg routines should
83 * be called for hugetlb (it has a separate hugetlb_cgroup.)
84 */
85 if (!PageHuge(page))
86 __page_cache_release(page);
Andrea Arcangeli91807062011-01-13 15:46:32 -080087 dtor = get_compound_page_dtor(page);
88 (*dtor)(page);
89}
90
Jianyu Zhanc747ce72014-06-04 16:07:59 -070091/**
92 * Two special cases here: we could avoid taking compound_lock_irqsave
93 * and could skip the tail refcounting(in _mapcount).
94 *
95 * 1. Hugetlbfs page:
96 *
97 * PageHeadHuge will remain true until the compound page
98 * is released and enters the buddy allocator, and it could
99 * not be split by __split_huge_page_refcount().
100 *
101 * So if we see PageHeadHuge set, and we have the tail page pin,
102 * then we could safely put head page.
103 *
104 * 2. Slab THP page:
105 *
106 * PG_slab is cleared before the slab frees the head page, and
107 * tail pin cannot be the last reference left on the head page,
108 * because the slab code is free to reuse the compound page
109 * after a kfree/kmem_cache_free without having to check if
110 * there's any tail pin left. In turn all tail pinsmust be always
111 * released while the head is still pinned by the slab code
112 * and so we know PG_slab will be still set too.
113 *
114 * So if we see PageSlab set, and we have the tail page pin,
115 * then we could safely put head page.
116 */
117static __always_inline
118void put_unrefcounted_compound_page(struct page *page_head, struct page *page)
119{
120 /*
121 * If @page is a THP tail, we must read the tail page
122 * flags after the head page flags. The
123 * __split_huge_page_refcount side enforces write memory barriers
124 * between clearing PageTail and before the head page
125 * can be freed and reallocated.
126 */
127 smp_rmb();
128 if (likely(PageTail(page))) {
129 /*
130 * __split_huge_page_refcount cannot race
131 * here, see the comment above this function.
132 */
133 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
Jianyu Zhanc747ce72014-06-04 16:07:59 -0700134 if (put_page_testzero(page_head)) {
135 /*
136 * If this is the tail of a slab THP page,
137 * the tail pin must not be the last reference
138 * held on the page, because the PG_slab cannot
139 * be cleared before all tail pins (which skips
140 * the _mapcount tail refcounting) have been
141 * released.
142 *
143 * If this is the tail of a hugetlbfs page,
144 * the tail pin may be the last reference on
145 * the page instead, because PageHeadHuge will
146 * not go away until the compound page enters
147 * the buddy allocator.
148 */
149 VM_BUG_ON_PAGE(PageSlab(page_head), page_head);
150 __put_compound_page(page_head);
151 }
152 } else
153 /*
154 * __split_huge_page_refcount run before us,
155 * @page was a THP tail. The split @page_head
156 * has been freed and reallocated as slab or
157 * hugetlbfs page of smaller order (only
158 * possible if reallocated as slab on x86).
159 */
160 if (put_page_testzero(page))
161 __put_single_page(page);
162}
163
164static __always_inline
165void put_refcounted_compound_page(struct page *page_head, struct page *page)
166{
167 if (likely(page != page_head && get_page_unless_zero(page_head))) {
168 unsigned long flags;
169
170 /*
171 * @page_head wasn't a dangling pointer but it may not
172 * be a head page anymore by the time we obtain the
173 * lock. That is ok as long as it can't be freed from
174 * under us.
175 */
176 flags = compound_lock_irqsave(page_head);
177 if (unlikely(!PageTail(page))) {
178 /* __split_huge_page_refcount run before us */
179 compound_unlock_irqrestore(page_head, flags);
180 if (put_page_testzero(page_head)) {
181 /*
182 * The @page_head may have been freed
183 * and reallocated as a compound page
184 * of smaller order and then freed
185 * again. All we know is that it
186 * cannot have become: a THP page, a
187 * compound page of higher order, a
188 * tail page. That is because we
189 * still hold the refcount of the
190 * split THP tail and page_head was
191 * the THP head before the split.
192 */
193 if (PageHead(page_head))
194 __put_compound_page(page_head);
195 else
196 __put_single_page(page_head);
197 }
198out_put_single:
199 if (put_page_testzero(page))
200 __put_single_page(page);
201 return;
202 }
203 VM_BUG_ON_PAGE(page_head != page->first_page, page);
204 /*
205 * We can release the refcount taken by
206 * get_page_unless_zero() now that
207 * __split_huge_page_refcount() is blocked on the
208 * compound_lock.
209 */
210 if (put_page_testzero(page_head))
211 VM_BUG_ON_PAGE(1, page_head);
212 /* __split_huge_page_refcount will wait now */
213 VM_BUG_ON_PAGE(page_mapcount(page) <= 0, page);
214 atomic_dec(&page->_mapcount);
215 VM_BUG_ON_PAGE(atomic_read(&page_head->_count) <= 0, page_head);
216 VM_BUG_ON_PAGE(atomic_read(&page->_count) != 0, page);
217 compound_unlock_irqrestore(page_head, flags);
218
219 if (put_page_testzero(page_head)) {
220 if (PageHead(page_head))
221 __put_compound_page(page_head);
222 else
223 __put_single_page(page_head);
224 }
225 } else {
226 /* @page_head is a dangling pointer */
227 VM_BUG_ON_PAGE(PageTail(page), page);
228 goto out_put_single;
229 }
230}
231
Nick Piggin8519fb32006-02-07 12:58:52 -0800232static void put_compound_page(struct page *page)
233{
Andrew Morton26296ad2014-01-21 15:48:59 -0800234 struct page *page_head;
235
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700236 /*
237 * We see the PageCompound set and PageTail not set, so @page maybe:
238 * 1. hugetlbfs head page, or
239 * 2. THP head page.
240 */
Andrew Morton26296ad2014-01-21 15:48:59 -0800241 if (likely(!PageTail(page))) {
242 if (put_page_testzero(page)) {
243 /*
244 * By the time all refcounts have been released
245 * split_huge_page cannot run anymore from under us.
246 */
247 if (PageHead(page))
248 __put_compound_page(page);
249 else
250 __put_single_page(page);
251 }
252 return;
253 }
254
Andrew Morton26296ad2014-01-21 15:48:59 -0800255 /*
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700256 * We see the PageCompound set and PageTail set, so @page maybe:
257 * 1. a tail hugetlbfs page, or
258 * 2. a tail THP page, or
259 * 3. a split THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800260 *
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700261 * Case 3 is possible, as we may race with
262 * __split_huge_page_refcount tearing down a THP page.
Andrew Morton26296ad2014-01-21 15:48:59 -0800263 */
Jianyu Zhand2ee40e2014-06-04 16:08:02 -0700264 page_head = compound_head_by_tail(page);
Jianyu Zhan4bd3e8f2014-06-04 16:08:01 -0700265 if (!__compound_tail_refcounted(page_head))
266 put_unrefcounted_compound_page(page_head, page);
267 else
268 put_refcounted_compound_page(page_head, page);
Nick Piggin8519fb32006-02-07 12:58:52 -0800269}
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271void put_page(struct page *page)
272{
Nick Piggin8519fb32006-02-07 12:58:52 -0800273 if (unlikely(PageCompound(page)))
274 put_compound_page(page);
275 else if (put_page_testzero(page))
Andrea Arcangeli91807062011-01-13 15:46:32 -0800276 __put_single_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278EXPORT_SYMBOL(put_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700280/*
281 * This function is exported but must not be called by anything other
282 * than get_page(). It implements the slow path of get_page().
283 */
284bool __get_page_tail(struct page *page)
285{
286 /*
287 * This takes care of get_page() if run on a tail page
288 * returned by one of the get_user_pages/follow_page variants.
289 * get_user_pages/follow_page itself doesn't need the compound
290 * lock because it runs __get_page_tail_foll() under the
291 * proper PT lock that already serializes against
292 * split_huge_page().
293 */
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800294 unsigned long flags;
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800295 bool got;
David Rientjes668f9abb2014-03-03 15:38:18 -0800296 struct page *page_head = compound_head(page);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700297
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800298 /* Ref to put_compound_page() comment. */
Andrea Arcangeli3bfcd132014-01-21 15:48:56 -0800299 if (!__compound_tail_refcounted(page_head)) {
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800300 smp_rmb();
301 if (likely(PageTail(page))) {
302 /*
303 * This is a hugetlbfs page or a slab
304 * page. __split_huge_page_refcount
305 * cannot race here.
306 */
Sasha Levin309381fea2014-01-23 15:52:54 -0800307 VM_BUG_ON_PAGE(!PageHead(page_head), page_head);
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800308 __get_page_tail_foll(page, true);
309 return true;
310 } else {
311 /*
312 * __split_huge_page_refcount run
313 * before us, "page" was a THP
314 * tail. The split page_head has been
315 * freed and reallocated as slab or
316 * hugetlbfs page of smaller order
317 * (only possible if reallocated as
318 * slab on x86).
319 */
320 return false;
Pravin B Shelar5bf5f032012-05-29 15:06:49 -0700321 }
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800322 }
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800323
Andrea Arcangeliebf360f2014-01-21 15:48:51 -0800324 got = false;
325 if (likely(page != page_head && get_page_unless_zero(page_head))) {
Andrea Arcangeli27c73ae2013-11-21 14:32:02 -0800326 /*
327 * page_head wasn't a dangling pointer but it
328 * may not be a head page anymore by the time
329 * we obtain the lock. That is ok as long as it
330 * can't be freed from under us.
331 */
332 flags = compound_lock_irqsave(page_head);
333 /* here __split_huge_page_refcount won't run anymore */
334 if (likely(PageTail(page))) {
335 __get_page_tail_foll(page, false);
336 got = true;
337 }
338 compound_unlock_irqrestore(page_head, flags);
339 if (unlikely(!got))
340 put_page(page_head);
Andrea Arcangeli70b50f92011-11-02 13:36:59 -0700341 }
342 return got;
343}
344EXPORT_SYMBOL(__get_page_tail);
345
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700346/**
Randy Dunlap76824862008-03-19 17:00:40 -0700347 * put_pages_list() - release a list of pages
348 * @pages: list of pages threaded on page->lru
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700349 *
350 * Release a list of pages which are strung together on page.lru. Currently
351 * used by read_cache_pages() and related error recovery code.
Alexander Zarochentsev1d7ea732006-08-13 23:24:27 -0700352 */
353void put_pages_list(struct list_head *pages)
354{
355 while (!list_empty(pages)) {
356 struct page *victim;
357
358 victim = list_entry(pages->prev, struct page, lru);
359 list_del(&victim->lru);
360 page_cache_release(victim);
361 }
362}
363EXPORT_SYMBOL(put_pages_list);
364
Mel Gorman18022c52012-07-31 16:44:51 -0700365/*
366 * get_kernel_pages() - pin kernel pages in memory
367 * @kiov: An array of struct kvec structures
368 * @nr_segs: number of segments to pin
369 * @write: pinning for read/write, currently ignored
370 * @pages: array that receives pointers to the pages pinned.
371 * Should be at least nr_segs long.
372 *
373 * Returns number of pages pinned. This may be fewer than the number
374 * requested. If nr_pages is 0 or negative, returns 0. If no pages
375 * were pinned, returns -errno. Each page returned must be released
376 * with a put_page() call when it is finished with.
377 */
378int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
379 struct page **pages)
380{
381 int seg;
382
383 for (seg = 0; seg < nr_segs; seg++) {
384 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
385 return seg;
386
Mel Gorman5a178112012-07-31 16:45:02 -0700387 pages[seg] = kmap_to_page(kiov[seg].iov_base);
Mel Gorman18022c52012-07-31 16:44:51 -0700388 page_cache_get(pages[seg]);
389 }
390
391 return seg;
392}
393EXPORT_SYMBOL_GPL(get_kernel_pages);
394
395/*
396 * get_kernel_page() - pin a kernel page in memory
397 * @start: starting kernel address
398 * @write: pinning for read/write, currently ignored
399 * @pages: array that receives pointer to the page pinned.
400 * Must be at least nr_segs long.
401 *
402 * Returns 1 if page is pinned. If the page was not pinned, returns
403 * -errno. The page returned must be released with a put_page() call
404 * when it is finished with.
405 */
406int get_kernel_page(unsigned long start, int write, struct page **pages)
407{
408 const struct kvec kiov = {
409 .iov_base = (void *)start,
410 .iov_len = PAGE_SIZE
411 };
412
413 return get_kernel_pages(&kiov, 1, write, pages);
414}
415EXPORT_SYMBOL_GPL(get_kernel_page);
416
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700417static void pagevec_lru_move_fn(struct pagevec *pvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700418 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
419 void *arg)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700420{
421 int i;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700422 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700423 struct lruvec *lruvec;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700424 unsigned long flags = 0;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700425
426 for (i = 0; i < pagevec_count(pvec); i++) {
427 struct page *page = pvec->pages[i];
428 struct zone *pagezone = page_zone(page);
429
430 if (pagezone != zone) {
431 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700432 spin_unlock_irqrestore(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700433 zone = pagezone;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700434 spin_lock_irqsave(&zone->lru_lock, flags);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700435 }
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700436
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700437 lruvec = mem_cgroup_page_lruvec(page, zone);
438 (*move_fn)(page, lruvec, arg);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700439 }
440 if (zone)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700441 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds83896fb2011-01-17 14:42:34 -0800442 release_pages(pvec->pages, pvec->nr, pvec->cold);
443 pagevec_reinit(pvec);
Shaohua Lid8505de2011-01-13 15:47:33 -0800444}
445
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700446static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
447 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700448{
449 int *pgmoved = arg;
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700450
451 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
452 enum lru_list lru = page_lru_base_type(page);
Johannes Weiner925b7672012-01-12 17:18:15 -0800453 list_move_tail(&page->lru, &lruvec->lists[lru]);
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700454 (*pgmoved)++;
455 }
456}
457
458/*
459 * pagevec_move_tail() must be called with IRQ disabled.
460 * Otherwise this may cause nasty races.
461 */
462static void pagevec_move_tail(struct pagevec *pvec)
463{
464 int pgmoved = 0;
465
466 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
467 __count_vm_events(PGROTATED, pgmoved);
468}
469
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700470/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 * Writeback is about to end against a page which has been marked for immediate
472 * reclaim. If it still appears to be reclaimable, move it to the tail of the
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700473 * inactive list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 */
Shaohua Li3dd7ae82011-03-22 16:33:45 -0700475void rotate_reclaimable_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700477 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700478 !PageUnevictable(page) && PageLRU(page)) {
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700479 struct pagevec *pvec;
480 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700482 page_cache_get(page);
483 local_irq_save(flags);
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700484 pvec = this_cpu_ptr(&lru_rotate_pvecs);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700485 if (!pagevec_add(pvec, page))
486 pagevec_move_tail(pvec);
487 local_irq_restore(flags);
488 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
490
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700491static void update_page_reclaim_stat(struct lruvec *lruvec,
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800492 int file, int rotated)
493{
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700494 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800495
496 reclaim_stat->recent_scanned[file]++;
497 if (rotated)
498 reclaim_stat->recent_rotated[file]++;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800499}
500
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700501static void __activate_page(struct page *page, struct lruvec *lruvec,
502 void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503{
Linus Torvalds7a608572011-01-17 14:42:19 -0800504 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
505 int file = page_is_file_cache(page);
506 int lru = page_lru_base_type(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800507
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700508 del_page_from_lru_list(page, lruvec, lru);
Linus Torvalds7a608572011-01-17 14:42:19 -0800509 SetPageActive(page);
510 lru += LRU_ACTIVE;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700511 add_page_to_lru_list(page, lruvec, lru);
Mel Gorman24b7e582014-08-06 16:07:11 -0700512 trace_mm_lru_activate(page);
Linus Torvalds7a608572011-01-17 14:42:19 -0800513
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700514 __count_vm_event(PGACTIVATE);
515 update_page_reclaim_stat(lruvec, file, 1);
Linus Torvalds7a608572011-01-17 14:42:19 -0800516 }
Shaohua Lieb709b02011-05-24 17:12:55 -0700517}
518
519#ifdef CONFIG_SMP
520static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
521
522static void activate_page_drain(int cpu)
523{
524 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
525
526 if (pagevec_count(pvec))
527 pagevec_lru_move_fn(pvec, __activate_page, NULL);
528}
529
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700530static bool need_activate_page_drain(int cpu)
531{
532 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
533}
534
Shaohua Lieb709b02011-05-24 17:12:55 -0700535void activate_page(struct page *page)
536{
537 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
538 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
539
540 page_cache_get(page);
541 if (!pagevec_add(pvec, page))
542 pagevec_lru_move_fn(pvec, __activate_page, NULL);
543 put_cpu_var(activate_page_pvecs);
544 }
545}
546
547#else
548static inline void activate_page_drain(int cpu)
549{
550}
551
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700552static bool need_activate_page_drain(int cpu)
553{
554 return false;
555}
556
Shaohua Lieb709b02011-05-24 17:12:55 -0700557void activate_page(struct page *page)
558{
559 struct zone *zone = page_zone(page);
560
561 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700562 __activate_page(page, mem_cgroup_page_lruvec(page, zone), NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 spin_unlock_irq(&zone->lru_lock);
564}
Shaohua Lieb709b02011-05-24 17:12:55 -0700565#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Mel Gorman059285a2013-07-03 15:02:30 -0700567static void __lru_cache_activate_page(struct page *page)
568{
569 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
570 int i;
571
572 /*
573 * Search backwards on the optimistic assumption that the page being
574 * activated has just been added to this pagevec. Note that only
575 * the local pagevec is examined as a !PageLRU page could be in the
576 * process of being released, reclaimed, migrated or on a remote
577 * pagevec that is currently being drained. Furthermore, marking
578 * a remote pagevec's page PageActive potentially hits a race where
579 * a page is marked PageActive just after it is added to the inactive
580 * list causing accounting errors and BUG_ON checks to trigger.
581 */
582 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
583 struct page *pagevec_page = pvec->pages[i];
584
585 if (pagevec_page == page) {
586 SetPageActive(page);
587 break;
588 }
589 }
590
591 put_cpu_var(lru_add_pvec);
592}
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594/*
595 * Mark a page as having seen activity.
596 *
597 * inactive,unreferenced -> inactive,referenced
598 * inactive,referenced -> active,unreferenced
599 * active,unreferenced -> active,referenced
Hugh Dickinseb39d612014-08-06 16:06:43 -0700600 *
601 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
602 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800604void mark_page_accessed(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700606 if (!PageActive(page) && !PageUnevictable(page) &&
Mel Gorman059285a2013-07-03 15:02:30 -0700607 PageReferenced(page)) {
608
609 /*
610 * If the page is on the LRU, queue it for activation via
611 * activate_page_pvecs. Otherwise, assume the page is on a
612 * pagevec, mark it active and it'll be moved to the active
613 * LRU on the next drain.
614 */
615 if (PageLRU(page))
616 activate_page(page);
617 else
618 __lru_cache_activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 ClearPageReferenced(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700620 if (page_is_file_cache(page))
621 workingset_activation(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 } else if (!PageReferenced(page)) {
623 SetPageReferenced(page);
624 }
625}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626EXPORT_SYMBOL(mark_page_accessed);
627
Jianyu Zhan2329d372014-06-04 16:07:31 -0700628static void __lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Mel Gorman13f7f782013-07-03 15:02:28 -0700630 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 page_cache_get(page);
Robin Dongd741c9c2012-10-08 16:29:05 -0700633 if (!pagevec_space(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700634 __pagevec_lru_add(pvec);
Robin Dongd741c9c2012-10-08 16:29:05 -0700635 pagevec_add(pvec, page);
Mel Gorman13f7f782013-07-03 15:02:28 -0700636 put_cpu_var(lru_add_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
Jianyu Zhan2329d372014-06-04 16:07:31 -0700638
639/**
640 * lru_cache_add: add a page to the page lists
641 * @page: the page to add
642 */
643void lru_cache_add_anon(struct page *page)
644{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700645 if (PageActive(page))
646 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700647 __lru_cache_add(page);
648}
649
650void lru_cache_add_file(struct page *page)
651{
Mel Gorman6fb81a12014-06-04 16:10:28 -0700652 if (PageActive(page))
653 ClearPageActive(page);
Jianyu Zhan2329d372014-06-04 16:07:31 -0700654 __lru_cache_add(page);
655}
656EXPORT_SYMBOL(lru_cache_add_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700658/**
Mel Gormanc53954a2013-07-03 15:02:34 -0700659 * lru_cache_add - add a page to a page list
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700660 * @page: the page to be added to the LRU.
Jianyu Zhan2329d372014-06-04 16:07:31 -0700661 *
662 * Queue the page for addition to the LRU via pagevec. The decision on whether
663 * to add the page to the [in]active [file|anon] list is deferred until the
664 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
665 * have the page added to the active list using mark_page_accessed().
KOSAKI Motohirof04e9eb2008-10-18 20:26:19 -0700666 */
Mel Gormanc53954a2013-07-03 15:02:34 -0700667void lru_cache_add(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668{
Sasha Levin309381fea2014-01-23 15:52:54 -0800669 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
670 VM_BUG_ON_PAGE(PageLRU(page), page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700671 __lru_cache_add(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700674/**
675 * add_page_to_unevictable_list - add a page to the unevictable list
676 * @page: the page to be added to the unevictable list
677 *
678 * Add page directly to its zone's unevictable list. To avoid races with
679 * tasks that might be making the page evictable, through eg. munlock,
680 * munmap or exit, while it's not on the lru, we want to add the page
681 * while it's locked or otherwise "invisible" to other tasks. This is
682 * difficult to do when using the pagevec cache, so bypass that.
683 */
684void add_page_to_unevictable_list(struct page *page)
685{
686 struct zone *zone = page_zone(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700687 struct lruvec *lruvec;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700688
689 spin_lock_irq(&zone->lru_lock);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700690 lruvec = mem_cgroup_page_lruvec(page, zone);
Naoya Horiguchief2a2cb2013-07-31 13:53:37 -0700691 ClearPageActive(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700692 SetPageUnevictable(page);
693 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700694 add_page_to_lru_list(page, lruvec, LRU_UNEVICTABLE);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700695 spin_unlock_irq(&zone->lru_lock);
696}
697
Johannes Weiner00501b52014-08-08 14:19:20 -0700698/**
699 * lru_cache_add_active_or_unevictable
700 * @page: the page to be added to LRU
701 * @vma: vma in which page is mapped for determining reclaimability
702 *
703 * Place @page on the active or unevictable LRU list, depending on its
704 * evictability. Note that if the page is not evictable, it goes
705 * directly back onto it's zone's unevictable list, it does NOT use a
706 * per cpu pagevec.
707 */
708void lru_cache_add_active_or_unevictable(struct page *page,
709 struct vm_area_struct *vma)
710{
711 VM_BUG_ON_PAGE(PageLRU(page), page);
712
713 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) {
714 SetPageActive(page);
715 lru_cache_add(page);
716 return;
717 }
718
719 if (!TestSetPageMlocked(page)) {
720 /*
721 * We use the irq-unsafe __mod_zone_page_stat because this
722 * counter is not modified from interrupt context, and the pte
723 * lock is held(spinlock), which implies preemption disabled.
724 */
725 __mod_zone_page_state(page_zone(page), NR_MLOCK,
726 hpage_nr_pages(page));
727 count_vm_event(UNEVICTABLE_PGMLOCKED);
728 }
729 add_page_to_unevictable_list(page);
730}
731
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700732/*
Minchan Kim31560182011-03-22 16:32:52 -0700733 * If the page can not be invalidated, it is moved to the
734 * inactive list to speed up its reclaim. It is moved to the
735 * head of the list, rather than the tail, to give the flusher
736 * threads some time to write it out, as this is much more
737 * effective than the single-page writeout from reclaim.
Minchan Kim278df9f2011-03-22 16:32:54 -0700738 *
739 * If the page isn't page_mapped and dirty/writeback, the page
740 * could reclaim asap using PG_reclaim.
741 *
742 * 1. active, mapped page -> none
743 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
744 * 3. inactive, mapped page -> none
745 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
746 * 5. inactive, clean -> inactive, tail
747 * 6. Others -> none
748 *
749 * In 4, why it moves inactive's head, the VM expects the page would
750 * be write it out by flusher threads as this is much more effective
751 * than the single-page writeout from reclaim.
Minchan Kim31560182011-03-22 16:32:52 -0700752 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700753static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700754 void *arg)
Minchan Kim31560182011-03-22 16:32:52 -0700755{
756 int lru, file;
Minchan Kim278df9f2011-03-22 16:32:54 -0700757 bool active;
Minchan Kim31560182011-03-22 16:32:52 -0700758
Minchan Kim278df9f2011-03-22 16:32:54 -0700759 if (!PageLRU(page))
Minchan Kim31560182011-03-22 16:32:52 -0700760 return;
761
Minchan Kimbad49d92011-05-11 15:13:30 -0700762 if (PageUnevictable(page))
763 return;
764
Minchan Kim31560182011-03-22 16:32:52 -0700765 /* Some processes are using the page */
766 if (page_mapped(page))
767 return;
768
Minchan Kim278df9f2011-03-22 16:32:54 -0700769 active = PageActive(page);
Minchan Kim31560182011-03-22 16:32:52 -0700770 file = page_is_file_cache(page);
771 lru = page_lru_base_type(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700772
773 del_page_from_lru_list(page, lruvec, lru + active);
Minchan Kim31560182011-03-22 16:32:52 -0700774 ClearPageActive(page);
775 ClearPageReferenced(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700776 add_page_to_lru_list(page, lruvec, lru);
Minchan Kim31560182011-03-22 16:32:52 -0700777
Minchan Kim278df9f2011-03-22 16:32:54 -0700778 if (PageWriteback(page) || PageDirty(page)) {
779 /*
780 * PG_reclaim could be raced with end_page_writeback
781 * It can make readahead confusing. But race window
782 * is _really_ small and it's non-critical problem.
783 */
784 SetPageReclaim(page);
785 } else {
786 /*
787 * The page's writeback ends up during pagevec
788 * We moves tha page into tail of inactive.
789 */
Johannes Weiner925b7672012-01-12 17:18:15 -0800790 list_move_tail(&page->lru, &lruvec->lists[lru]);
Minchan Kim278df9f2011-03-22 16:32:54 -0700791 __count_vm_event(PGROTATED);
792 }
793
794 if (active)
795 __count_vm_event(PGDEACTIVATE);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700796 update_page_reclaim_stat(lruvec, file, 0);
Minchan Kim31560182011-03-22 16:32:52 -0700797}
798
Minchan Kim31560182011-03-22 16:32:52 -0700799/*
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700800 * Drain pages out of the cpu's pagevecs.
801 * Either "cpu" is the current CPU, and preemption has already been
802 * disabled; or "cpu" is being hot-unplugged, and is already dead.
803 */
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700804void lru_add_drain_cpu(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
Mel Gorman13f7f782013-07-03 15:02:28 -0700806 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
Mel Gorman13f7f782013-07-03 15:02:28 -0700808 if (pagevec_count(pvec))
Mel Gormana0b8cab32013-07-03 15:02:32 -0700809 __pagevec_lru_add(pvec);
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700810
811 pvec = &per_cpu(lru_rotate_pvecs, cpu);
812 if (pagevec_count(pvec)) {
813 unsigned long flags;
814
815 /* No harm done if a racing interrupt already did this */
816 local_irq_save(flags);
817 pagevec_move_tail(pvec);
818 local_irq_restore(flags);
819 }
Minchan Kim31560182011-03-22 16:32:52 -0700820
Minchan Kimcc5993b2015-04-15 16:13:26 -0700821 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700822 if (pagevec_count(pvec))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700823 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
Shaohua Lieb709b02011-05-24 17:12:55 -0700824
825 activate_page_drain(cpu);
Minchan Kim31560182011-03-22 16:32:52 -0700826}
827
828/**
Minchan Kimcc5993b2015-04-15 16:13:26 -0700829 * deactivate_file_page - forcefully deactivate a file page
Minchan Kim31560182011-03-22 16:32:52 -0700830 * @page: page to deactivate
831 *
832 * This function hints the VM that @page is a good reclaim candidate,
833 * for example if its invalidation fails due to the page being dirty
834 * or under writeback.
835 */
Minchan Kimcc5993b2015-04-15 16:13:26 -0700836void deactivate_file_page(struct page *page)
Minchan Kim31560182011-03-22 16:32:52 -0700837{
Minchan Kim821ed6b2011-05-24 17:12:31 -0700838 /*
Minchan Kimcc5993b2015-04-15 16:13:26 -0700839 * In a workload with many unevictable page such as mprotect,
840 * unevictable page deactivation for accelerating reclaim is pointless.
Minchan Kim821ed6b2011-05-24 17:12:31 -0700841 */
842 if (PageUnevictable(page))
843 return;
844
Minchan Kim31560182011-03-22 16:32:52 -0700845 if (likely(get_page_unless_zero(page))) {
Minchan Kimcc5993b2015-04-15 16:13:26 -0700846 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700847
848 if (!pagevec_add(pvec, page))
Minchan Kimcc5993b2015-04-15 16:13:26 -0700849 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
850 put_cpu_var(lru_deactivate_file_pvecs);
Minchan Kim31560182011-03-22 16:32:52 -0700851 }
Andrew Morton80bfed92006-01-06 00:11:14 -0800852}
853
854void lru_add_drain(void)
855{
Konstantin Khlebnikovf0cb3c72012-03-21 16:34:06 -0700856 lru_add_drain_cpu(get_cpu());
Andrew Morton80bfed92006-01-06 00:11:14 -0800857 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858}
859
David Howellsc4028952006-11-22 14:57:56 +0000860static void lru_add_drain_per_cpu(struct work_struct *dummy)
Nick Piggin053837f2006-01-18 17:42:27 -0800861{
862 lru_add_drain();
863}
864
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700865static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
866
867void lru_add_drain_all(void)
Nick Piggin053837f2006-01-18 17:42:27 -0800868{
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700869 static DEFINE_MUTEX(lock);
870 static struct cpumask has_work;
871 int cpu;
872
873 mutex_lock(&lock);
874 get_online_cpus();
875 cpumask_clear(&has_work);
876
877 for_each_online_cpu(cpu) {
878 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
879
880 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
881 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
Minchan Kimcc5993b2015-04-15 16:13:26 -0700882 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
Chris Metcalf5fbc4612013-09-12 15:13:55 -0700883 need_activate_page_drain(cpu)) {
884 INIT_WORK(work, lru_add_drain_per_cpu);
885 schedule_work_on(cpu, work);
886 cpumask_set_cpu(cpu, &has_work);
887 }
888 }
889
890 for_each_cpu(cpu, &has_work)
891 flush_work(&per_cpu(lru_add_drain_work, cpu));
892
893 put_online_cpus();
894 mutex_unlock(&lock);
Nick Piggin053837f2006-01-18 17:42:27 -0800895}
896
Michal Hockoaabfb572014-10-09 15:28:52 -0700897/**
898 * release_pages - batched page_cache_release()
899 * @pages: array of pages to release
900 * @nr: number of pages
901 * @cold: whether the pages are cache cold
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 *
Michal Hockoaabfb572014-10-09 15:28:52 -0700903 * Decrement the reference count on all the pages in @pages. If it
904 * fell to zero, remove the page from the LRU and free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 */
Mel Gormanb745bc82014-06-04 16:10:22 -0700906void release_pages(struct page **pages, int nr, bool cold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
908 int i;
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800909 LIST_HEAD(pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct zone *zone = NULL;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700911 struct lruvec *lruvec;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700912 unsigned long uninitialized_var(flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700913 unsigned int uninitialized_var(lock_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 for (i = 0; i < nr; i++) {
916 struct page *page = pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Nick Piggin8519fb32006-02-07 12:58:52 -0800918 if (unlikely(PageCompound(page))) {
919 if (zone) {
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700920 spin_unlock_irqrestore(&zone->lru_lock, flags);
Nick Piggin8519fb32006-02-07 12:58:52 -0800921 zone = NULL;
922 }
923 put_compound_page(page);
924 continue;
925 }
926
Michal Hockoaabfb572014-10-09 15:28:52 -0700927 /*
928 * Make sure the IRQ-safe lock-holding time does not get
929 * excessive with a continuous string of pages from the
930 * same zone. The lock is held only if zone != NULL.
931 */
932 if (zone && ++lock_batch == SWAP_CLUSTER_MAX) {
933 spin_unlock_irqrestore(&zone->lru_lock, flags);
934 zone = NULL;
935 }
936
Nick Pigginb5810032005-10-29 18:16:12 -0700937 if (!put_page_testzero(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 continue;
939
Nick Piggin46453a62006-03-22 00:07:58 -0800940 if (PageLRU(page)) {
941 struct zone *pagezone = page_zone(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700942
Nick Piggin46453a62006-03-22 00:07:58 -0800943 if (pagezone != zone) {
944 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700945 spin_unlock_irqrestore(&zone->lru_lock,
946 flags);
Michal Hockoaabfb572014-10-09 15:28:52 -0700947 lock_batch = 0;
Nick Piggin46453a62006-03-22 00:07:58 -0800948 zone = pagezone;
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700949 spin_lock_irqsave(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 }
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700951
952 lruvec = mem_cgroup_page_lruvec(page, zone);
Sasha Levin309381fea2014-01-23 15:52:54 -0800953 VM_BUG_ON_PAGE(!PageLRU(page), page);
Nick Piggin67453912006-03-22 00:08:00 -0800954 __ClearPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700955 del_page_from_lru_list(page, lruvec, page_off_lru(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 }
Nick Piggin46453a62006-03-22 00:07:58 -0800957
Mel Gormanc53954a2013-07-03 15:02:34 -0700958 /* Clear Active bit in case of parallel mark_page_accessed */
Mel Gormane3741b52014-06-04 16:10:26 -0700959 __ClearPageActive(page);
Mel Gormanc53954a2013-07-03 15:02:34 -0700960
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800961 list_add(&page->lru, &pages_to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 }
963 if (zone)
Hisashi Hifumi902aaed2007-10-16 01:24:52 -0700964 spin_unlock_irqrestore(&zone->lru_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Johannes Weiner747db952014-08-08 14:19:24 -0700966 mem_cgroup_uncharge_list(&pages_to_free);
Konstantin Khlebnikovcc598502012-01-10 15:07:04 -0800967 free_hot_cold_page_list(&pages_to_free, cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968}
Miklos Szeredi0be85572010-10-27 15:34:46 -0700969EXPORT_SYMBOL(release_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971/*
972 * The pages which we're about to release may be in the deferred lru-addition
973 * queues. That would prevent them from really being freed right now. That's
974 * OK from a correctness point of view but is inefficient - those pages may be
975 * cache-warm and we want to give them back to the page allocator ASAP.
976 *
977 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
978 * and __pagevec_lru_add_active() call release_pages() directly to avoid
979 * mutual recursion.
980 */
981void __pagevec_release(struct pagevec *pvec)
982{
983 lru_add_drain();
984 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
985 pagevec_reinit(pvec);
986}
Steve French7f285702005-11-01 10:22:55 -0800987EXPORT_SYMBOL(__pagevec_release);
988
Hugh Dickins12d27102012-01-12 17:19:52 -0800989#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800990/* used by __split_huge_page_refcount() */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700991void lru_add_page_tail(struct page *page, struct page *page_tail,
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700992 struct lruvec *lruvec, struct list_head *list)
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800993{
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800994 const int file = 0;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800995
Sasha Levin309381fea2014-01-23 15:52:54 -0800996 VM_BUG_ON_PAGE(!PageHead(page), page);
997 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
998 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700999 VM_BUG_ON(NR_CPUS != 1 &&
1000 !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001001
Shaohua Li5bc7b8a2013-04-29 15:08:36 -07001002 if (!list)
1003 SetPageLRU(page_tail);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001004
Hugh Dickins12d27102012-01-12 17:19:52 -08001005 if (likely(PageLRU(page)))
1006 list_add_tail(&page_tail->lru, &page->lru);
Shaohua Li5bc7b8a2013-04-29 15:08:36 -07001007 else if (list) {
1008 /* page reclaim is reclaiming a huge page */
1009 get_page(page_tail);
1010 list_add_tail(&page_tail->lru, list);
1011 } else {
Hugh Dickins12d27102012-01-12 17:19:52 -08001012 struct list_head *list_head;
1013 /*
1014 * Head page has not yet been counted, as an hpage,
1015 * so we must account for each subpage individually.
1016 *
1017 * Use the standard add function to put page_tail on the list,
1018 * but then correct its position so they all end up in order.
1019 */
Kirill A. Shutemove180cf82013-07-31 13:53:39 -07001020 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
Hugh Dickins12d27102012-01-12 17:19:52 -08001021 list_head = page_tail->lru.prev;
1022 list_move_tail(&page_tail->lru, list_head);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001023 }
Hugh Dickins75121022012-03-05 14:59:18 -08001024
1025 if (!PageUnevictable(page))
Kirill A. Shutemove180cf82013-07-31 13:53:39 -07001026 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001027}
Hugh Dickins12d27102012-01-12 17:19:52 -08001028#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001029
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001030static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
1031 void *arg)
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001032{
Mel Gorman13f7f782013-07-03 15:02:28 -07001033 int file = page_is_file_cache(page);
1034 int active = PageActive(page);
1035 enum lru_list lru = page_lru(page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001036
Sasha Levin309381fea2014-01-23 15:52:54 -08001037 VM_BUG_ON_PAGE(PageLRU(page), page);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001038
1039 SetPageLRU(page);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001040 add_page_to_lru_list(page, lruvec, lru);
1041 update_page_reclaim_stat(lruvec, file, active);
Mel Gorman24b7e582014-08-06 16:07:11 -07001042 trace_mm_lru_insertion(page, lru);
Shaohua Li3dd7ae82011-03-22 16:33:45 -07001043}
1044
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 * Add the passed pages to the LRU, then drop the caller's refcount
1047 * on them. Reinitialises the caller's pagevec.
1048 */
Mel Gormana0b8cab32013-07-03 15:02:32 -07001049void __pagevec_lru_add(struct pagevec *pvec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050{
Mel Gormana0b8cab32013-07-03 15:02:32 -07001051 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052}
Hugh Dickins5095ae832012-01-12 17:19:58 -08001053EXPORT_SYMBOL(__pagevec_lru_add);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001056 * pagevec_lookup_entries - gang pagecache lookup
1057 * @pvec: Where the resulting entries are placed
1058 * @mapping: The address_space to search
1059 * @start: The starting entry index
1060 * @nr_entries: The maximum number of entries
1061 * @indices: The cache indices corresponding to the entries in @pvec
1062 *
1063 * pagevec_lookup_entries() will search for and return a group of up
1064 * to @nr_entries pages and shadow entries in the mapping. All
1065 * entries are placed in @pvec. pagevec_lookup_entries() takes a
1066 * reference against actual pages in @pvec.
1067 *
1068 * The search returns a group of mapping-contiguous entries with
1069 * ascending indexes. There may be holes in the indices due to
1070 * not-present entries.
1071 *
1072 * pagevec_lookup_entries() returns the number of entries which were
1073 * found.
1074 */
1075unsigned pagevec_lookup_entries(struct pagevec *pvec,
1076 struct address_space *mapping,
1077 pgoff_t start, unsigned nr_pages,
1078 pgoff_t *indices)
1079{
1080 pvec->nr = find_get_entries(mapping, start, nr_pages,
1081 pvec->pages, indices);
1082 return pagevec_count(pvec);
1083}
1084
1085/**
1086 * pagevec_remove_exceptionals - pagevec exceptionals pruning
1087 * @pvec: The pagevec to prune
1088 *
1089 * pagevec_lookup_entries() fills both pages and exceptional radix
1090 * tree entries into the pagevec. This function prunes all
1091 * exceptionals from @pvec without leaving holes, so that it can be
1092 * passed on to page-only pagevec operations.
1093 */
1094void pagevec_remove_exceptionals(struct pagevec *pvec)
1095{
1096 int i, j;
1097
1098 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1099 struct page *page = pvec->pages[i];
1100 if (!radix_tree_exceptional_entry(page))
1101 pvec->pages[j++] = page;
1102 }
1103 pvec->nr = j;
1104}
1105
1106/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 * pagevec_lookup - gang pagecache lookup
1108 * @pvec: Where the resulting pages are placed
1109 * @mapping: The address_space to search
1110 * @start: The starting page index
1111 * @nr_pages: The maximum number of pages
1112 *
1113 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
1114 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
1115 * reference against the pages in @pvec.
1116 *
1117 * The search returns a group of mapping-contiguous pages with ascending
1118 * indexes. There may be holes in the indices due to not-present pages.
1119 *
1120 * pagevec_lookup() returns the number of pages which were found.
1121 */
1122unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
1123 pgoff_t start, unsigned nr_pages)
1124{
1125 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
1126 return pagevec_count(pvec);
1127}
Christoph Hellwig78539fd2006-01-11 20:47:41 +11001128EXPORT_SYMBOL(pagevec_lookup);
1129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
1131 pgoff_t *index, int tag, unsigned nr_pages)
1132{
1133 pvec->nr = find_get_pages_tag(mapping, index, tag,
1134 nr_pages, pvec->pages);
1135 return pagevec_count(pvec);
1136}
Steve French7f285702005-11-01 10:22:55 -08001137EXPORT_SYMBOL(pagevec_lookup_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139/*
1140 * Perform any setup for the swap system
1141 */
1142void __init swap_setup(void)
1143{
Jan Beulich44813742009-09-21 17:03:05 -07001144 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001145#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -08001146 int i;
1147
Kirill A. Shutemov27ba0642015-02-10 14:09:59 -08001148 for (i = 0; i < MAX_SWAPFILES; i++)
Shaohua Li33806f02013-02-22 16:34:37 -08001149 spin_lock_init(&swapper_spaces[i].tree_lock);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001150#endif
1151
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 /* Use a smaller cluster for small-memory machines */
1153 if (megs < 16)
1154 page_cluster = 2;
1155 else
1156 page_cluster = 3;
1157 /*
1158 * Right now other parts of the system means that we
1159 * _really_ don't want to cluster much more
1160 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161}