1/*
2 * linux/mm/swap.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * This file contains the default values for the operation of the
9 * 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>
24#include <linux/export.h>
25#include <linux/mm_inline.h>
26#include <linux/percpu_counter.h>
27#include <linux/memremap.h>
28#include <linux/percpu.h>
29#include <linux/cpu.h>
30#include <linux/notifier.h>
31#include <linux/backing-dev.h>
32#include <linux/memcontrol.h>
33#include <linux/gfp.h>
34#include <linux/uio.h>
35#include <linux/hugetlb.h>
36#include <linux/page_idle.h>
37
38#include "internal.h"
39
40#define CREATE_TRACE_POINTS
41#include <trace/events/pagemap.h>
42
43/* How many pages do we try to swap or page in/out together? */
44int page_cluster;
45
46static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
47static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
48static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
49static DEFINE_PER_CPU(struct pagevec, lru_lazyfree_pvecs);
50#ifdef CONFIG_SMP
51static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
52#endif
53
54/*
55 * This path almost never happens for VM activity - pages are normally
56 * freed via pagevecs. But it gets used by networking.
57 */
58static void __page_cache_release(struct page *page)
59{
60 if (PageLRU(page)) {
61 pg_data_t *pgdat = page_pgdat(page);
62 struct lruvec *lruvec;
63 unsigned long flags;
64
65 spin_lock_irqsave(&pgdat->lru_lock, flags);
66 lruvec = mem_cgroup_page_lruvec(page, pgdat);
67 VM_BUG_ON_PAGE(!PageLRU(page), page);
68 __ClearPageLRU(page);
69 del_page_from_lru_list(page, lruvec, page_off_lru(page));
70 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
71 }
72 __ClearPageWaiters(page);
73 mem_cgroup_uncharge(page);
74}
75
76static void __put_single_page(struct page *page)
77{
78 __page_cache_release(page);
79 free_unref_page(page);
80}
81
82static void __put_compound_page(struct page *page)
83{
84 compound_page_dtor *dtor;
85
86 /*
87 * __page_cache_release() is supposed to be called for thp, not for
88 * hugetlb. This is because hugetlb page does never have PageLRU set
89 * (it's never listed to any LRU lists) and no memcg routines should
90 * be called for hugetlb (it has a separate hugetlb_cgroup.)
91 */
92 if (!PageHuge(page))
93 __page_cache_release(page);
94 dtor = get_compound_page_dtor(page);
95 (*dtor)(page);
96}
97
98void __put_page(struct page *page)
99{
100 if (is_zone_device_page(page)) {
101 put_dev_pagemap(page->pgmap);
102
103 /*
104 * The page belongs to the device that created pgmap. Do
105 * not return it to page allocator.
106 */
107 return;
108 }
109
110 if (unlikely(PageCompound(page)))
111 __put_compound_page(page);
112 else
113 __put_single_page(page);
114}
115EXPORT_SYMBOL(__put_page);
116
117/**
118 * put_pages_list() - release a list of pages
119 * @pages: list of pages threaded on page->lru
120 *
121 * Release a list of pages which are strung together on page.lru. Currently
122 * used by read_cache_pages() and related error recovery code.
123 */
124void put_pages_list(struct list_head *pages)
125{
126 while (!list_empty(pages)) {
127 struct page *victim;
128
129 victim = lru_to_page(pages);
130 list_del(&victim->lru);
131 put_page(victim);
132 }
133}
134EXPORT_SYMBOL(put_pages_list);
135
136/*
137 * get_kernel_pages() - pin kernel pages in memory
138 * @kiov: An array of struct kvec structures
139 * @nr_segs: number of segments to pin
140 * @write: pinning for read/write, currently ignored
141 * @pages: array that receives pointers to the pages pinned.
142 * Should be at least nr_segs long.
143 *
144 * Returns number of pages pinned. This may be fewer than the number
145 * requested. If nr_pages is 0 or negative, returns 0. If no pages
146 * were pinned, returns -errno. Each page returned must be released
147 * with a put_page() call when it is finished with.
148 */
149int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
150 struct page **pages)
151{
152 int seg;
153
154 for (seg = 0; seg < nr_segs; seg++) {
155 if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
156 return seg;
157
158 pages[seg] = kmap_to_page(kiov[seg].iov_base);
159 get_page(pages[seg]);
160 }
161
162 return seg;
163}
164EXPORT_SYMBOL_GPL(get_kernel_pages);
165
166/*
167 * get_kernel_page() - pin a kernel page in memory
168 * @start: starting kernel address
169 * @write: pinning for read/write, currently ignored
170 * @pages: array that receives pointer to the page pinned.
171 * Must be at least nr_segs long.
172 *
173 * Returns 1 if page is pinned. If the page was not pinned, returns
174 * -errno. The page returned must be released with a put_page() call
175 * when it is finished with.
176 */
177int get_kernel_page(unsigned long start, int write, struct page **pages)
178{
179 const struct kvec kiov = {
180 .iov_base = (void *)start,
181 .iov_len = PAGE_SIZE
182 };
183
184 return get_kernel_pages(&kiov, 1, write, pages);
185}
186EXPORT_SYMBOL_GPL(get_kernel_page);
187
188static void pagevec_lru_move_fn(struct pagevec *pvec,
189 void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
190 void *arg)
191{
192 int i;
193 struct pglist_data *pgdat = NULL;
194 struct lruvec *lruvec;
195 unsigned long flags = 0;
196
197 for (i = 0; i < pagevec_count(pvec); i++) {
198 struct page *page = pvec->pages[i];
199 struct pglist_data *pagepgdat = page_pgdat(page);
200
201 if (pagepgdat != pgdat) {
202 if (pgdat)
203 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
204 pgdat = pagepgdat;
205 spin_lock_irqsave(&pgdat->lru_lock, flags);
206 }
207
208 lruvec = mem_cgroup_page_lruvec(page, pgdat);
209 (*move_fn)(page, lruvec, arg);
210 }
211 if (pgdat)
212 spin_unlock_irqrestore(&pgdat->lru_lock, flags);
213 release_pages(pvec->pages, pvec->nr);
214 pagevec_reinit(pvec);
215}
216
217static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
218 void *arg)
219{
220 int *pgmoved = arg;
221
222 if (PageLRU(page) && !PageUnevictable(page)) {
223 del_page_from_lru_list(page, lruvec, page_lru(page));
224 ClearPageActive(page);
225 add_page_to_lru_list_tail(page, lruvec, page_lru(page));
226 (*pgmoved)++;
227 }
228}
229
230/*
231 * pagevec_move_tail() must be called with IRQ disabled.
232 * Otherwise this may cause nasty races.
233 */
234static void pagevec_move_tail(struct pagevec *pvec)
235{
236 int pgmoved = 0;
237
238 pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
239 __count_vm_events(PGROTATED, pgmoved);
240}
241
242/*
243 * Writeback is about to end against a page which has been marked for immediate
244 * reclaim. If it still appears to be reclaimable, move it to the tail of the
245 * inactive list.
246 */
247void rotate_reclaimable_page(struct page *page)
248{
249 if (!PageLocked(page) && !PageDirty(page) &&
250 !PageUnevictable(page) && PageLRU(page)) {
251 struct pagevec *pvec;
252 unsigned long flags;
253
254 get_page(page);
255 local_irq_save(flags);
256 pvec = this_cpu_ptr(&lru_rotate_pvecs);
257 if (!pagevec_add(pvec, page) || PageCompound(page))
258 pagevec_move_tail(pvec);
259 local_irq_restore(flags);
260 }
261}
262
263static void update_page_reclaim_stat(struct lruvec *lruvec,
264 int file, int rotated)
265{
266 struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
267
268 reclaim_stat->recent_scanned[file]++;
269 if (rotated)
270 reclaim_stat->recent_rotated[file]++;
271}
272
273static void __activate_page(struct page *page, struct lruvec *lruvec,
274 void *arg)
275{
276 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
277 int file = page_is_file_cache(page);
278 int lru = page_lru_base_type(page);
279
280 del_page_from_lru_list(page, lruvec, lru);
281 SetPageActive(page);
282 lru += LRU_ACTIVE;
283 add_page_to_lru_list(page, lruvec, lru);
284 trace_mm_lru_activate(page);
285
286 __count_vm_event(PGACTIVATE);
287 update_page_reclaim_stat(lruvec, file, 1);
288 }
289}
290
291#ifdef CONFIG_SMP
292static void activate_page_drain(int cpu)
293{
294 struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
295
296 if (pagevec_count(pvec))
297 pagevec_lru_move_fn(pvec, __activate_page, NULL);
298}
299
300static bool need_activate_page_drain(int cpu)
301{
302 return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
303}
304
305void activate_page(struct page *page)
306{
307 page = compound_head(page);
308 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
309 struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
310
311 get_page(page);
312 if (!pagevec_add(pvec, page) || PageCompound(page))
313 pagevec_lru_move_fn(pvec, __activate_page, NULL);
314 put_cpu_var(activate_page_pvecs);
315 }
316}
317
318#else
319static inline void activate_page_drain(int cpu)
320{
321}
322
323void activate_page(struct page *page)
324{
325 pg_data_t *pgdat = page_pgdat(page);
326
327 page = compound_head(page);
328 spin_lock_irq(&pgdat->lru_lock);
329 __activate_page(page, mem_cgroup_page_lruvec(page, pgdat), NULL);
330 spin_unlock_irq(&pgdat->lru_lock);
331}
332#endif
333
334static void __lru_cache_activate_page(struct page *page)
335{
336 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
337 int i;
338
339 /*
340 * Search backwards on the optimistic assumption that the page being
341 * activated has just been added to this pagevec. Note that only
342 * the local pagevec is examined as a !PageLRU page could be in the
343 * process of being released, reclaimed, migrated or on a remote
344 * pagevec that is currently being drained. Furthermore, marking
345 * a remote pagevec's page PageActive potentially hits a race where
346 * a page is marked PageActive just after it is added to the inactive
347 * list causing accounting errors and BUG_ON checks to trigger.
348 */
349 for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
350 struct page *pagevec_page = pvec->pages[i];
351
352 if (pagevec_page == page) {
353 SetPageActive(page);
354 break;
355 }
356 }
357
358 put_cpu_var(lru_add_pvec);
359}
360
361/*
362 * Mark a page as having seen activity.
363 *
364 * inactive,unreferenced -> inactive,referenced
365 * inactive,referenced -> active,unreferenced
366 * active,unreferenced -> active,referenced
367 *
368 * When a newly allocated page is not yet visible, so safe for non-atomic ops,
369 * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
370 */
371void mark_page_accessed(struct page *page)
372{
373 page = compound_head(page);
374 if (!PageActive(page) && !PageUnevictable(page) &&
375 PageReferenced(page)) {
376
377 /*
378 * If the page is on the LRU, queue it for activation via
379 * activate_page_pvecs. Otherwise, assume the page is on a
380 * pagevec, mark it active and it'll be moved to the active
381 * LRU on the next drain.
382 */
383 if (PageLRU(page))
384 activate_page(page);
385 else
386 __lru_cache_activate_page(page);
387 ClearPageReferenced(page);
388 if (page_is_file_cache(page))
389 workingset_activation(page);
390 } else if (!PageReferenced(page)) {
391 SetPageReferenced(page);
392 }
393 if (page_is_idle(page))
394 clear_page_idle(page);
395}
396EXPORT_SYMBOL(mark_page_accessed);
397
398static void __lru_cache_add(struct page *page)
399{
400 struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
401
402 get_page(page);
403 if (!pagevec_add(pvec, page) || PageCompound(page))
404 __pagevec_lru_add(pvec);
405 put_cpu_var(lru_add_pvec);
406}
407
408/**
409 * lru_cache_add_anon - add a page to the page lists
410 * @page: the page to add
411 */
412void lru_cache_add_anon(struct page *page)
413{
414 if (PageActive(page))
415 ClearPageActive(page);
416 __lru_cache_add(page);
417}
418
419void lru_cache_add_file(struct page *page)
420{
421 if (PageActive(page))
422 ClearPageActive(page);
423 __lru_cache_add(page);
424}
425EXPORT_SYMBOL(lru_cache_add_file);
426
427/**
428 * lru_cache_add - add a page to a page list
429 * @page: the page to be added to the LRU.
430 *
431 * Queue the page for addition to the LRU via pagevec. The decision on whether
432 * to add the page to the [in]active [file|anon] list is deferred until the
433 * pagevec is drained. This gives a chance for the caller of lru_cache_add()
434 * have the page added to the active list using mark_page_accessed().
435 */
436void lru_cache_add(struct page *page)
437{
438 VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
439 VM_BUG_ON_PAGE(PageLRU(page), page);
440 __lru_cache_add(page);
441}
442
443/**
444 * lru_cache_add_active_or_unevictable
445 * @page: the page to be added to LRU
446 * @vma: vma in which page is mapped for determining reclaimability
447 *
448 * Place @page on the active or unevictable LRU list, depending on its
449 * evictability. Note that if the page is not evictable, it goes
450 * directly back onto it's zone's unevictable list, it does NOT use a
451 * per cpu pagevec.
452 */
453void lru_cache_add_active_or_unevictable(struct page *page,
454 struct vm_area_struct *vma)
455{
456 VM_BUG_ON_PAGE(PageLRU(page), page);
457
458 if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
459 SetPageActive(page);
460 else if (!TestSetPageMlocked(page)) {
461 /*
462 * We use the irq-unsafe __mod_zone_page_stat because this
463 * counter is not modified from interrupt context, and the pte
464 * lock is held(spinlock), which implies preemption disabled.
465 */
466 __mod_zone_page_state(page_zone(page), NR_MLOCK,
467 hpage_nr_pages(page));
468 count_vm_event(UNEVICTABLE_PGMLOCKED);
469 }
470 lru_cache_add(page);
471}
472
473/*
474 * If the page can not be invalidated, it is moved to the
475 * inactive list to speed up its reclaim. It is moved to the
476 * head of the list, rather than the tail, to give the flusher
477 * threads some time to write it out, as this is much more
478 * effective than the single-page writeout from reclaim.
479 *
480 * If the page isn't page_mapped and dirty/writeback, the page
481 * could reclaim asap using PG_reclaim.
482 *
483 * 1. active, mapped page -> none
484 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
485 * 3. inactive, mapped page -> none
486 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
487 * 5. inactive, clean -> inactive, tail
488 * 6. Others -> none
489 *
490 * In 4, why it moves inactive's head, the VM expects the page would
491 * be write it out by flusher threads as this is much more effective
492 * than the single-page writeout from reclaim.
493 */
494static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
495 void *arg)
496{
497 int lru, file;
498 bool active;
499
500 if (!PageLRU(page))
501 return;
502
503 if (PageUnevictable(page))
504 return;
505
506 /* Some processes are using the page */
507 if (page_mapped(page))
508 return;
509
510 active = PageActive(page);
511 file = page_is_file_cache(page);
512 lru = page_lru_base_type(page);
513
514 del_page_from_lru_list(page, lruvec, lru + active);
515 ClearPageActive(page);
516 ClearPageReferenced(page);
517 add_page_to_lru_list(page, lruvec, lru);
518
519 if (PageWriteback(page) || PageDirty(page)) {
520 /*
521 * PG_reclaim could be raced with end_page_writeback
522 * It can make readahead confusing. But race window
523 * is _really_ small and it's non-critical problem.
524 */
525 SetPageReclaim(page);
526 } else {
527 /*
528 * The page's writeback ends up during pagevec
529 * We moves tha page into tail of inactive.
530 */
531 list_move_tail(&page->lru, &lruvec->lists[lru]);
532 __count_vm_event(PGROTATED);
533 }
534
535 if (active)
536 __count_vm_event(PGDEACTIVATE);
537 update_page_reclaim_stat(lruvec, file, 0);
538}
539
540
541static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
542 void *arg)
543{
544 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
545 !PageSwapCache(page) && !PageUnevictable(page)) {
546 bool active = PageActive(page);
547
548 del_page_from_lru_list(page, lruvec,
549 LRU_INACTIVE_ANON + active);
550 ClearPageActive(page);
551 ClearPageReferenced(page);
552 /*
553 * lazyfree pages are clean anonymous pages. They have
554 * SwapBacked flag cleared to distinguish normal anonymous
555 * pages
556 */
557 ClearPageSwapBacked(page);
558 add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
559
560 __count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
561 count_memcg_page_event(page, PGLAZYFREE);
562 update_page_reclaim_stat(lruvec, 1, 0);
563 }
564}
565
566/*
567 * Drain pages out of the cpu's pagevecs.
568 * Either "cpu" is the current CPU, and preemption has already been
569 * disabled; or "cpu" is being hot-unplugged, and is already dead.
570 */
571void lru_add_drain_cpu(int cpu)
572{
573 struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
574
575 if (pagevec_count(pvec))
576 __pagevec_lru_add(pvec);
577
578 pvec = &per_cpu(lru_rotate_pvecs, cpu);
579 if (pagevec_count(pvec)) {
580 unsigned long flags;
581
582 /* No harm done if a racing interrupt already did this */
583 local_irq_save(flags);
584 pagevec_move_tail(pvec);
585 local_irq_restore(flags);
586 }
587
588 pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
589 if (pagevec_count(pvec))
590 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
591
592 pvec = &per_cpu(lru_lazyfree_pvecs, cpu);
593 if (pagevec_count(pvec))
594 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
595
596 activate_page_drain(cpu);
597}
598
599/**
600 * deactivate_file_page - forcefully deactivate a file page
601 * @page: page to deactivate
602 *
603 * This function hints the VM that @page is a good reclaim candidate,
604 * for example if its invalidation fails due to the page being dirty
605 * or under writeback.
606 */
607void deactivate_file_page(struct page *page)
608{
609 /*
610 * In a workload with many unevictable page such as mprotect,
611 * unevictable page deactivation for accelerating reclaim is pointless.
612 */
613 if (PageUnevictable(page))
614 return;
615
616 if (likely(get_page_unless_zero(page))) {
617 struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
618
619 if (!pagevec_add(pvec, page) || PageCompound(page))
620 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
621 put_cpu_var(lru_deactivate_file_pvecs);
622 }
623}
624
625/**
626 * mark_page_lazyfree - make an anon page lazyfree
627 * @page: page to deactivate
628 *
629 * mark_page_lazyfree() moves @page to the inactive file list.
630 * This is done to accelerate the reclaim of @page.
631 */
632void mark_page_lazyfree(struct page *page)
633{
634 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
635 !PageSwapCache(page) && !PageUnevictable(page)) {
636 struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs);
637
638 get_page(page);
639 if (!pagevec_add(pvec, page) || PageCompound(page))
640 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
641 put_cpu_var(lru_lazyfree_pvecs);
642 }
643}
644
645void lru_add_drain(void)
646{
647 lru_add_drain_cpu(get_cpu());
648 put_cpu();
649}
650
651#ifdef CONFIG_SMP
652
653static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
654
655static void lru_add_drain_per_cpu(struct work_struct *dummy)
656{
657 lru_add_drain();
658}
659
660/*
661 * Doesn't need any cpu hotplug locking because we do rely on per-cpu
662 * kworkers being shut down before our page_alloc_cpu_dead callback is
663 * executed on the offlined cpu.
664 * Calling this function with cpu hotplug locks held can actually lead
665 * to obscure indirect dependencies via WQ context.
666 */
667void lru_add_drain_all(void)
668{
669 static DEFINE_MUTEX(lock);
670 static struct cpumask has_work;
671 int cpu;
672
673 /*
674 * Make sure nobody triggers this path before mm_percpu_wq is fully
675 * initialized.
676 */
677 if (WARN_ON(!mm_percpu_wq))
678 return;
679
680 mutex_lock(&lock);
681 cpumask_clear(&has_work);
682
683 for_each_online_cpu(cpu) {
684 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
685
686 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
687 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
688 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
689 pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
690 need_activate_page_drain(cpu)) {
691 INIT_WORK(work, lru_add_drain_per_cpu);
692 queue_work_on(cpu, mm_percpu_wq, work);
693 cpumask_set_cpu(cpu, &has_work);
694 }
695 }
696
697 for_each_cpu(cpu, &has_work)
698 flush_work(&per_cpu(lru_add_drain_work, cpu));
699
700 mutex_unlock(&lock);
701}
702#else
703void lru_add_drain_all(void)
704{
705 lru_add_drain();
706}
707#endif
708
709/**
710 * release_pages - batched put_page()
711 * @pages: array of pages to release
712 * @nr: number of pages
713 *
714 * Decrement the reference count on all the pages in @pages. If it
715 * fell to zero, remove the page from the LRU and free it.
716 */
717void release_pages(struct page **pages, int nr)
718{
719 int i;
720 LIST_HEAD(pages_to_free);
721 struct pglist_data *locked_pgdat = NULL;
722 struct lruvec *lruvec;
723 unsigned long uninitialized_var(flags);
724 unsigned int uninitialized_var(lock_batch);
725
726 for (i = 0; i < nr; i++) {
727 struct page *page = pages[i];
728
729 /*
730 * Make sure the IRQ-safe lock-holding time does not get
731 * excessive with a continuous string of pages from the
732 * same pgdat. The lock is held only if pgdat != NULL.
733 */
734 if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
735 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
736 locked_pgdat = NULL;
737 }
738
739 if (is_huge_zero_page(page))
740 continue;
741
742 /* Device public page can not be huge page */
743 if (is_device_public_page(page)) {
744 if (locked_pgdat) {
745 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
746 flags);
747 locked_pgdat = NULL;
748 }
749 put_devmap_managed_page(page);
750 continue;
751 }
752
753 page = compound_head(page);
754 if (!put_page_testzero(page))
755 continue;
756
757 if (PageCompound(page)) {
758 if (locked_pgdat) {
759 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
760 locked_pgdat = NULL;
761 }
762 __put_compound_page(page);
763 continue;
764 }
765
766 if (PageLRU(page)) {
767 struct pglist_data *pgdat = page_pgdat(page);
768
769 if (pgdat != locked_pgdat) {
770 if (locked_pgdat)
771 spin_unlock_irqrestore(&locked_pgdat->lru_lock,
772 flags);
773 lock_batch = 0;
774 locked_pgdat = pgdat;
775 spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
776 }
777
778 lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
779 VM_BUG_ON_PAGE(!PageLRU(page), page);
780 __ClearPageLRU(page);
781 del_page_from_lru_list(page, lruvec, page_off_lru(page));
782 }
783
784 /* Clear Active bit in case of parallel mark_page_accessed */
785 __ClearPageActive(page);
786 __ClearPageWaiters(page);
787
788 list_add(&page->lru, &pages_to_free);
789 }
790 if (locked_pgdat)
791 spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
792
793 mem_cgroup_uncharge_list(&pages_to_free);
794 free_unref_page_list(&pages_to_free);
795}
796EXPORT_SYMBOL(release_pages);
797
798/*
799 * The pages which we're about to release may be in the deferred lru-addition
800 * queues. That would prevent them from really being freed right now. That's
801 * OK from a correctness point of view but is inefficient - those pages may be
802 * cache-warm and we want to give them back to the page allocator ASAP.
803 *
804 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
805 * and __pagevec_lru_add_active() call release_pages() directly to avoid
806 * mutual recursion.
807 */
808void __pagevec_release(struct pagevec *pvec)
809{
810 if (!pvec->percpu_pvec_drained) {
811 lru_add_drain();
812 pvec->percpu_pvec_drained = true;
813 }
814 release_pages(pvec->pages, pagevec_count(pvec));
815 pagevec_reinit(pvec);
816}
817EXPORT_SYMBOL(__pagevec_release);
818
819#ifdef CONFIG_TRANSPARENT_HUGEPAGE
820/* used by __split_huge_page_refcount() */
821void lru_add_page_tail(struct page *page, struct page *page_tail,
822 struct lruvec *lruvec, struct list_head *list)
823{
824 const int file = 0;
825
826 VM_BUG_ON_PAGE(!PageHead(page), page);
827 VM_BUG_ON_PAGE(PageCompound(page_tail), page);
828 VM_BUG_ON_PAGE(PageLRU(page_tail), page);
829 lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock);
830
831 if (!list)
832 SetPageLRU(page_tail);
833
834 if (likely(PageLRU(page)))
835 list_add_tail(&page_tail->lru, &page->lru);
836 else if (list) {
837 /* page reclaim is reclaiming a huge page */
838 get_page(page_tail);
839 list_add_tail(&page_tail->lru, list);
840 } else {
841 struct list_head *list_head;
842 /*
843 * Head page has not yet been counted, as an hpage,
844 * so we must account for each subpage individually.
845 *
846 * Use the standard add function to put page_tail on the list,
847 * but then correct its position so they all end up in order.
848 */
849 add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
850 list_head = page_tail->lru.prev;
851 list_move_tail(&page_tail->lru, list_head);
852 }
853
854 if (!PageUnevictable(page))
855 update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
856}
857#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
858
859static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
860 void *arg)
861{
862 enum lru_list lru;
863 int was_unevictable = TestClearPageUnevictable(page);
864
865 VM_BUG_ON_PAGE(PageLRU(page), page);
866
867 SetPageLRU(page);
868 /*
869 * Page becomes evictable in two ways:
870 * 1) Within LRU lock [munlock_vma_pages() and __munlock_pagevec()].
871 * 2) Before acquiring LRU lock to put the page to correct LRU and then
872 * a) do PageLRU check with lock [check_move_unevictable_pages]
873 * b) do PageLRU check before lock [clear_page_mlock]
874 *
875 * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need
876 * following strict ordering:
877 *
878 * #0: __pagevec_lru_add_fn #1: clear_page_mlock
879 *
880 * SetPageLRU() TestClearPageMlocked()
881 * smp_mb() // explicit ordering // above provides strict
882 * // ordering
883 * PageMlocked() PageLRU()
884 *
885 *
886 * if '#1' does not observe setting of PG_lru by '#0' and fails
887 * isolation, the explicit barrier will make sure that page_evictable
888 * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
889 * can be reordered after PageMlocked check and can make '#1' to fail
890 * the isolation of the page whose Mlocked bit is cleared (#0 is also
891 * looking at the same page) and the evictable page will be stranded
892 * in an unevictable LRU.
893 */
894 smp_mb();
895
896 if (page_evictable(page)) {
897 lru = page_lru(page);
898 update_page_reclaim_stat(lruvec, page_is_file_cache(page),
899 PageActive(page));
900 if (was_unevictable)
901 count_vm_event(UNEVICTABLE_PGRESCUED);
902 } else {
903 lru = LRU_UNEVICTABLE;
904 ClearPageActive(page);
905 SetPageUnevictable(page);
906 if (!was_unevictable)
907 count_vm_event(UNEVICTABLE_PGCULLED);
908 }
909
910 add_page_to_lru_list(page, lruvec, lru);
911 trace_mm_lru_insertion(page, lru);
912}
913
914/*
915 * Add the passed pages to the LRU, then drop the caller's refcount
916 * on them. Reinitialises the caller's pagevec.
917 */
918void __pagevec_lru_add(struct pagevec *pvec)
919{
920 pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
921}
922EXPORT_SYMBOL(__pagevec_lru_add);
923
924/**
925 * pagevec_lookup_entries - gang pagecache lookup
926 * @pvec: Where the resulting entries are placed
927 * @mapping: The address_space to search
928 * @start: The starting entry index
929 * @nr_entries: The maximum number of pages
930 * @indices: The cache indices corresponding to the entries in @pvec
931 *
932 * pagevec_lookup_entries() will search for and return a group of up
933 * to @nr_pages pages and shadow entries in the mapping. All
934 * entries are placed in @pvec. pagevec_lookup_entries() takes a
935 * reference against actual pages in @pvec.
936 *
937 * The search returns a group of mapping-contiguous entries with
938 * ascending indexes. There may be holes in the indices due to
939 * not-present entries.
940 *
941 * pagevec_lookup_entries() returns the number of entries which were
942 * found.
943 */
944unsigned pagevec_lookup_entries(struct pagevec *pvec,
945 struct address_space *mapping,
946 pgoff_t start, unsigned nr_entries,
947 pgoff_t *indices)
948{
949 pvec->nr = find_get_entries(mapping, start, nr_entries,
950 pvec->pages, indices);
951 return pagevec_count(pvec);
952}
953
954/**
955 * pagevec_remove_exceptionals - pagevec exceptionals pruning
956 * @pvec: The pagevec to prune
957 *
958 * pagevec_lookup_entries() fills both pages and exceptional radix
959 * tree entries into the pagevec. This function prunes all
960 * exceptionals from @pvec without leaving holes, so that it can be
961 * passed on to page-only pagevec operations.
962 */
963void pagevec_remove_exceptionals(struct pagevec *pvec)
964{
965 int i, j;
966
967 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
968 struct page *page = pvec->pages[i];
969 if (!xa_is_value(page))
970 pvec->pages[j++] = page;
971 }
972 pvec->nr = j;
973}
974
975/**
976 * pagevec_lookup_range - gang pagecache lookup
977 * @pvec: Where the resulting pages are placed
978 * @mapping: The address_space to search
979 * @start: The starting page index
980 * @end: The final page index
981 *
982 * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE
983 * pages in the mapping starting from index @start and upto index @end
984 * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a
985 * reference against the pages in @pvec.
986 *
987 * The search returns a group of mapping-contiguous pages with ascending
988 * indexes. There may be holes in the indices due to not-present pages. We
989 * also update @start to index the next page for the traversal.
990 *
991 * pagevec_lookup_range() returns the number of pages which were found. If this
992 * number is smaller than PAGEVEC_SIZE, the end of specified range has been
993 * reached.
994 */
995unsigned pagevec_lookup_range(struct pagevec *pvec,
996 struct address_space *mapping, pgoff_t *start, pgoff_t end)
997{
998 pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
999 pvec->pages);
1000 return pagevec_count(pvec);
1001}
1002EXPORT_SYMBOL(pagevec_lookup_range);
1003
1004unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
1005 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1006 xa_mark_t tag)
1007{
1008 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1009 PAGEVEC_SIZE, pvec->pages);
1010 return pagevec_count(pvec);
1011}
1012EXPORT_SYMBOL(pagevec_lookup_range_tag);
1013
1014unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
1015 struct address_space *mapping, pgoff_t *index, pgoff_t end,
1016 xa_mark_t tag, unsigned max_pages)
1017{
1018 pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1019 min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
1020 return pagevec_count(pvec);
1021}
1022EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
1023/*
1024 * Perform any setup for the swap system
1025 */
1026void __init swap_setup(void)
1027{
1028 unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
1029
1030 /* Use a smaller cluster for small-memory machines */
1031 if (megs < 16)
1032 page_cluster = 2;
1033 else
1034 page_cluster = 3;
1035 /*
1036 * Right now other parts of the system means that we
1037 * _really_ don't want to cluster much more
1038 */
1039}
1040