xref: /linux/mm/swap.c (revision ca64d84e93762f4e587e040a44ad9f6089afc777)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/mm/swap.c
4  *
5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
6  */
7 
8 /*
9  * This file contains the default values for the operation of the
10  * Linux VM subsystem. Fine-tuning documentation can be found in
11  * Documentation/admin-guide/sysctl/vm.rst.
12  * Started 18.12.91
13  * Swap aging added 23.2.95, Stephen Tweedie.
14  * Buffermem limits added 12.3.98, Rik van Riel.
15  */
16 
17 #include <linux/mm.h>
18 #include <linux/sched.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/pagevec.h>
24 #include <linux/init.h>
25 #include <linux/export.h>
26 #include <linux/mm_inline.h>
27 #include <linux/percpu_counter.h>
28 #include <linux/memremap.h>
29 #include <linux/percpu.h>
30 #include <linux/cpu.h>
31 #include <linux/notifier.h>
32 #include <linux/backing-dev.h>
33 #include <linux/memcontrol.h>
34 #include <linux/gfp.h>
35 #include <linux/uio.h>
36 #include <linux/hugetlb.h>
37 #include <linux/page_idle.h>
38 
39 #include "internal.h"
40 
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/pagemap.h>
43 
44 /* How many pages do we try to swap or page in/out together? */
45 int page_cluster;
46 
47 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
48 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
49 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
50 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
51 static DEFINE_PER_CPU(struct pagevec, lru_lazyfree_pvecs);
52 #ifdef CONFIG_SMP
53 static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
54 #endif
55 
56 /*
57  * This path almost never happens for VM activity - pages are normally
58  * freed via pagevecs.  But it gets used by networking.
59  */
60 static void __page_cache_release(struct page *page)
61 {
62 	if (PageLRU(page)) {
63 		pg_data_t *pgdat = page_pgdat(page);
64 		struct lruvec *lruvec;
65 		unsigned long flags;
66 
67 		spin_lock_irqsave(&pgdat->lru_lock, flags);
68 		lruvec = mem_cgroup_page_lruvec(page, pgdat);
69 		VM_BUG_ON_PAGE(!PageLRU(page), page);
70 		__ClearPageLRU(page);
71 		del_page_from_lru_list(page, lruvec, page_off_lru(page));
72 		spin_unlock_irqrestore(&pgdat->lru_lock, flags);
73 	}
74 	__ClearPageWaiters(page);
75 }
76 
77 static void __put_single_page(struct page *page)
78 {
79 	__page_cache_release(page);
80 	mem_cgroup_uncharge(page);
81 	free_unref_page(page);
82 }
83 
84 static void __put_compound_page(struct page *page)
85 {
86 	compound_page_dtor *dtor;
87 
88 	/*
89 	 * __page_cache_release() is supposed to be called for thp, not for
90 	 * hugetlb. This is because hugetlb page does never have PageLRU set
91 	 * (it's never listed to any LRU lists) and no memcg routines should
92 	 * be called for hugetlb (it has a separate hugetlb_cgroup.)
93 	 */
94 	if (!PageHuge(page))
95 		__page_cache_release(page);
96 	dtor = get_compound_page_dtor(page);
97 	(*dtor)(page);
98 }
99 
100 void __put_page(struct page *page)
101 {
102 	if (is_zone_device_page(page)) {
103 		put_dev_pagemap(page->pgmap);
104 
105 		/*
106 		 * The page belongs to the device that created pgmap. Do
107 		 * not return it to page allocator.
108 		 */
109 		return;
110 	}
111 
112 	if (unlikely(PageCompound(page)))
113 		__put_compound_page(page);
114 	else
115 		__put_single_page(page);
116 }
117 EXPORT_SYMBOL(__put_page);
118 
119 /**
120  * put_pages_list() - release a list of pages
121  * @pages: list of pages threaded on page->lru
122  *
123  * Release a list of pages which are strung together on page.lru.  Currently
124  * used by read_cache_pages() and related error recovery code.
125  */
126 void put_pages_list(struct list_head *pages)
127 {
128 	while (!list_empty(pages)) {
129 		struct page *victim;
130 
131 		victim = lru_to_page(pages);
132 		list_del(&victim->lru);
133 		put_page(victim);
134 	}
135 }
136 EXPORT_SYMBOL(put_pages_list);
137 
138 /*
139  * get_kernel_pages() - pin kernel pages in memory
140  * @kiov:	An array of struct kvec structures
141  * @nr_segs:	number of segments to pin
142  * @write:	pinning for read/write, currently ignored
143  * @pages:	array that receives pointers to the pages pinned.
144  *		Should be at least nr_segs long.
145  *
146  * Returns number of pages pinned. This may be fewer than the number
147  * requested. If nr_pages is 0 or negative, returns 0. If no pages
148  * were pinned, returns -errno. Each page returned must be released
149  * with a put_page() call when it is finished with.
150  */
151 int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write,
152 		struct page **pages)
153 {
154 	int seg;
155 
156 	for (seg = 0; seg < nr_segs; seg++) {
157 		if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE))
158 			return seg;
159 
160 		pages[seg] = kmap_to_page(kiov[seg].iov_base);
161 		get_page(pages[seg]);
162 	}
163 
164 	return seg;
165 }
166 EXPORT_SYMBOL_GPL(get_kernel_pages);
167 
168 /*
169  * get_kernel_page() - pin a kernel page in memory
170  * @start:	starting kernel address
171  * @write:	pinning for read/write, currently ignored
172  * @pages:	array that receives pointer to the page pinned.
173  *		Must be at least nr_segs long.
174  *
175  * Returns 1 if page is pinned. If the page was not pinned, returns
176  * -errno. The page returned must be released with a put_page() call
177  * when it is finished with.
178  */
179 int get_kernel_page(unsigned long start, int write, struct page **pages)
180 {
181 	const struct kvec kiov = {
182 		.iov_base = (void *)start,
183 		.iov_len = PAGE_SIZE
184 	};
185 
186 	return get_kernel_pages(&kiov, 1, write, pages);
187 }
188 EXPORT_SYMBOL_GPL(get_kernel_page);
189 
190 static void pagevec_lru_move_fn(struct pagevec *pvec,
191 	void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg),
192 	void *arg)
193 {
194 	int i;
195 	struct pglist_data *pgdat = NULL;
196 	struct lruvec *lruvec;
197 	unsigned long flags = 0;
198 
199 	for (i = 0; i < pagevec_count(pvec); i++) {
200 		struct page *page = pvec->pages[i];
201 		struct pglist_data *pagepgdat = page_pgdat(page);
202 
203 		if (pagepgdat != pgdat) {
204 			if (pgdat)
205 				spin_unlock_irqrestore(&pgdat->lru_lock, flags);
206 			pgdat = pagepgdat;
207 			spin_lock_irqsave(&pgdat->lru_lock, flags);
208 		}
209 
210 		lruvec = mem_cgroup_page_lruvec(page, pgdat);
211 		(*move_fn)(page, lruvec, arg);
212 	}
213 	if (pgdat)
214 		spin_unlock_irqrestore(&pgdat->lru_lock, flags);
215 	release_pages(pvec->pages, pvec->nr);
216 	pagevec_reinit(pvec);
217 }
218 
219 static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec,
220 				 void *arg)
221 {
222 	int *pgmoved = arg;
223 
224 	if (PageLRU(page) && !PageUnevictable(page)) {
225 		del_page_from_lru_list(page, lruvec, page_lru(page));
226 		ClearPageActive(page);
227 		add_page_to_lru_list_tail(page, lruvec, page_lru(page));
228 		(*pgmoved)++;
229 	}
230 }
231 
232 /*
233  * pagevec_move_tail() must be called with IRQ disabled.
234  * Otherwise this may cause nasty races.
235  */
236 static void pagevec_move_tail(struct pagevec *pvec)
237 {
238 	int pgmoved = 0;
239 
240 	pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved);
241 	__count_vm_events(PGROTATED, pgmoved);
242 }
243 
244 /*
245  * Writeback is about to end against a page which has been marked for immediate
246  * reclaim.  If it still appears to be reclaimable, move it to the tail of the
247  * inactive list.
248  */
249 void rotate_reclaimable_page(struct page *page)
250 {
251 	if (!PageLocked(page) && !PageDirty(page) &&
252 	    !PageUnevictable(page) && PageLRU(page)) {
253 		struct pagevec *pvec;
254 		unsigned long flags;
255 
256 		get_page(page);
257 		local_irq_save(flags);
258 		pvec = this_cpu_ptr(&lru_rotate_pvecs);
259 		if (!pagevec_add(pvec, page) || PageCompound(page))
260 			pagevec_move_tail(pvec);
261 		local_irq_restore(flags);
262 	}
263 }
264 
265 static void update_page_reclaim_stat(struct lruvec *lruvec,
266 				     int file, int rotated)
267 {
268 	struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
269 
270 	reclaim_stat->recent_scanned[file]++;
271 	if (rotated)
272 		reclaim_stat->recent_rotated[file]++;
273 }
274 
275 static void __activate_page(struct page *page, struct lruvec *lruvec,
276 			    void *arg)
277 {
278 	if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
279 		int file = page_is_file_lru(page);
280 		int lru = page_lru_base_type(page);
281 
282 		del_page_from_lru_list(page, lruvec, lru);
283 		SetPageActive(page);
284 		lru += LRU_ACTIVE;
285 		add_page_to_lru_list(page, lruvec, lru);
286 		trace_mm_lru_activate(page);
287 
288 		__count_vm_event(PGACTIVATE);
289 		update_page_reclaim_stat(lruvec, file, 1);
290 	}
291 }
292 
293 #ifdef CONFIG_SMP
294 static void activate_page_drain(int cpu)
295 {
296 	struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu);
297 
298 	if (pagevec_count(pvec))
299 		pagevec_lru_move_fn(pvec, __activate_page, NULL);
300 }
301 
302 static bool need_activate_page_drain(int cpu)
303 {
304 	return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
305 }
306 
307 void activate_page(struct page *page)
308 {
309 	page = compound_head(page);
310 	if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
311 		struct pagevec *pvec = &get_cpu_var(activate_page_pvecs);
312 
313 		get_page(page);
314 		if (!pagevec_add(pvec, page) || PageCompound(page))
315 			pagevec_lru_move_fn(pvec, __activate_page, NULL);
316 		put_cpu_var(activate_page_pvecs);
317 	}
318 }
319 
320 #else
321 static inline void activate_page_drain(int cpu)
322 {
323 }
324 
325 void activate_page(struct page *page)
326 {
327 	pg_data_t *pgdat = page_pgdat(page);
328 
329 	page = compound_head(page);
330 	spin_lock_irq(&pgdat->lru_lock);
331 	__activate_page(page, mem_cgroup_page_lruvec(page, pgdat), NULL);
332 	spin_unlock_irq(&pgdat->lru_lock);
333 }
334 #endif
335 
336 static void __lru_cache_activate_page(struct page *page)
337 {
338 	struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
339 	int i;
340 
341 	/*
342 	 * Search backwards on the optimistic assumption that the page being
343 	 * activated has just been added to this pagevec. Note that only
344 	 * the local pagevec is examined as a !PageLRU page could be in the
345 	 * process of being released, reclaimed, migrated or on a remote
346 	 * pagevec that is currently being drained. Furthermore, marking
347 	 * a remote pagevec's page PageActive potentially hits a race where
348 	 * a page is marked PageActive just after it is added to the inactive
349 	 * list causing accounting errors and BUG_ON checks to trigger.
350 	 */
351 	for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
352 		struct page *pagevec_page = pvec->pages[i];
353 
354 		if (pagevec_page == page) {
355 			SetPageActive(page);
356 			break;
357 		}
358 	}
359 
360 	put_cpu_var(lru_add_pvec);
361 }
362 
363 /*
364  * Mark a page as having seen activity.
365  *
366  * inactive,unreferenced	->	inactive,referenced
367  * inactive,referenced		->	active,unreferenced
368  * active,unreferenced		->	active,referenced
369  *
370  * When a newly allocated page is not yet visible, so safe for non-atomic ops,
371  * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
372  */
373 void mark_page_accessed(struct page *page)
374 {
375 	page = compound_head(page);
376 
377 	if (!PageReferenced(page)) {
378 		SetPageReferenced(page);
379 	} else if (PageUnevictable(page)) {
380 		/*
381 		 * Unevictable pages are on the "LRU_UNEVICTABLE" list. But,
382 		 * this list is never rotated or maintained, so marking an
383 		 * evictable page accessed has no effect.
384 		 */
385 	} else if (!PageActive(page)) {
386 		/*
387 		 * If the page is on the LRU, queue it for activation via
388 		 * activate_page_pvecs. Otherwise, assume the page is on a
389 		 * pagevec, mark it active and it'll be moved to the active
390 		 * LRU on the next drain.
391 		 */
392 		if (PageLRU(page))
393 			activate_page(page);
394 		else
395 			__lru_cache_activate_page(page);
396 		ClearPageReferenced(page);
397 		if (page_is_file_lru(page))
398 			workingset_activation(page);
399 	}
400 	if (page_is_idle(page))
401 		clear_page_idle(page);
402 }
403 EXPORT_SYMBOL(mark_page_accessed);
404 
405 static void __lru_cache_add(struct page *page)
406 {
407 	struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
408 
409 	get_page(page);
410 	if (!pagevec_add(pvec, page) || PageCompound(page))
411 		__pagevec_lru_add(pvec);
412 	put_cpu_var(lru_add_pvec);
413 }
414 
415 /**
416  * lru_cache_add_anon - add a page to the page lists
417  * @page: the page to add
418  */
419 void lru_cache_add_anon(struct page *page)
420 {
421 	if (PageActive(page))
422 		ClearPageActive(page);
423 	__lru_cache_add(page);
424 }
425 
426 void lru_cache_add_file(struct page *page)
427 {
428 	if (PageActive(page))
429 		ClearPageActive(page);
430 	__lru_cache_add(page);
431 }
432 EXPORT_SYMBOL(lru_cache_add_file);
433 
434 /**
435  * lru_cache_add - add a page to a page list
436  * @page: the page to be added to the LRU.
437  *
438  * Queue the page for addition to the LRU via pagevec. The decision on whether
439  * to add the page to the [in]active [file|anon] list is deferred until the
440  * pagevec is drained. This gives a chance for the caller of lru_cache_add()
441  * have the page added to the active list using mark_page_accessed().
442  */
443 void lru_cache_add(struct page *page)
444 {
445 	VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
446 	VM_BUG_ON_PAGE(PageLRU(page), page);
447 	__lru_cache_add(page);
448 }
449 
450 /**
451  * lru_cache_add_active_or_unevictable
452  * @page:  the page to be added to LRU
453  * @vma:   vma in which page is mapped for determining reclaimability
454  *
455  * Place @page on the active or unevictable LRU list, depending on its
456  * evictability.  Note that if the page is not evictable, it goes
457  * directly back onto it's zone's unevictable list, it does NOT use a
458  * per cpu pagevec.
459  */
460 void lru_cache_add_active_or_unevictable(struct page *page,
461 					 struct vm_area_struct *vma)
462 {
463 	VM_BUG_ON_PAGE(PageLRU(page), page);
464 
465 	if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
466 		SetPageActive(page);
467 	else if (!TestSetPageMlocked(page)) {
468 		/*
469 		 * We use the irq-unsafe __mod_zone_page_stat because this
470 		 * counter is not modified from interrupt context, and the pte
471 		 * lock is held(spinlock), which implies preemption disabled.
472 		 */
473 		__mod_zone_page_state(page_zone(page), NR_MLOCK,
474 				    hpage_nr_pages(page));
475 		count_vm_event(UNEVICTABLE_PGMLOCKED);
476 	}
477 	lru_cache_add(page);
478 }
479 
480 /*
481  * If the page can not be invalidated, it is moved to the
482  * inactive list to speed up its reclaim.  It is moved to the
483  * head of the list, rather than the tail, to give the flusher
484  * threads some time to write it out, as this is much more
485  * effective than the single-page writeout from reclaim.
486  *
487  * If the page isn't page_mapped and dirty/writeback, the page
488  * could reclaim asap using PG_reclaim.
489  *
490  * 1. active, mapped page -> none
491  * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
492  * 3. inactive, mapped page -> none
493  * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
494  * 5. inactive, clean -> inactive, tail
495  * 6. Others -> none
496  *
497  * In 4, why it moves inactive's head, the VM expects the page would
498  * be write it out by flusher threads as this is much more effective
499  * than the single-page writeout from reclaim.
500  */
501 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
502 			      void *arg)
503 {
504 	int lru, file;
505 	bool active;
506 
507 	if (!PageLRU(page))
508 		return;
509 
510 	if (PageUnevictable(page))
511 		return;
512 
513 	/* Some processes are using the page */
514 	if (page_mapped(page))
515 		return;
516 
517 	active = PageActive(page);
518 	file = page_is_file_lru(page);
519 	lru = page_lru_base_type(page);
520 
521 	del_page_from_lru_list(page, lruvec, lru + active);
522 	ClearPageActive(page);
523 	ClearPageReferenced(page);
524 
525 	if (PageWriteback(page) || PageDirty(page)) {
526 		/*
527 		 * PG_reclaim could be raced with end_page_writeback
528 		 * It can make readahead confusing.  But race window
529 		 * is _really_ small and  it's non-critical problem.
530 		 */
531 		add_page_to_lru_list(page, lruvec, lru);
532 		SetPageReclaim(page);
533 	} else {
534 		/*
535 		 * The page's writeback ends up during pagevec
536 		 * We moves tha page into tail of inactive.
537 		 */
538 		add_page_to_lru_list_tail(page, lruvec, lru);
539 		__count_vm_event(PGROTATED);
540 	}
541 
542 	if (active)
543 		__count_vm_event(PGDEACTIVATE);
544 	update_page_reclaim_stat(lruvec, file, 0);
545 }
546 
547 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
548 			    void *arg)
549 {
550 	if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
551 		int file = page_is_file_lru(page);
552 		int lru = page_lru_base_type(page);
553 
554 		del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE);
555 		ClearPageActive(page);
556 		ClearPageReferenced(page);
557 		add_page_to_lru_list(page, lruvec, lru);
558 
559 		__count_vm_events(PGDEACTIVATE, hpage_nr_pages(page));
560 		update_page_reclaim_stat(lruvec, file, 0);
561 	}
562 }
563 
564 static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
565 			    void *arg)
566 {
567 	if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
568 	    !PageSwapCache(page) && !PageUnevictable(page)) {
569 		bool active = PageActive(page);
570 
571 		del_page_from_lru_list(page, lruvec,
572 				       LRU_INACTIVE_ANON + active);
573 		ClearPageActive(page);
574 		ClearPageReferenced(page);
575 		/*
576 		 * Lazyfree pages are clean anonymous pages.  They have
577 		 * PG_swapbacked flag cleared, to distinguish them from normal
578 		 * anonymous pages
579 		 */
580 		ClearPageSwapBacked(page);
581 		add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
582 
583 		__count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
584 		count_memcg_page_event(page, PGLAZYFREE);
585 		update_page_reclaim_stat(lruvec, 1, 0);
586 	}
587 }
588 
589 /*
590  * Drain pages out of the cpu's pagevecs.
591  * Either "cpu" is the current CPU, and preemption has already been
592  * disabled; or "cpu" is being hot-unplugged, and is already dead.
593  */
594 void lru_add_drain_cpu(int cpu)
595 {
596 	struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
597 
598 	if (pagevec_count(pvec))
599 		__pagevec_lru_add(pvec);
600 
601 	pvec = &per_cpu(lru_rotate_pvecs, cpu);
602 	if (pagevec_count(pvec)) {
603 		unsigned long flags;
604 
605 		/* No harm done if a racing interrupt already did this */
606 		local_irq_save(flags);
607 		pagevec_move_tail(pvec);
608 		local_irq_restore(flags);
609 	}
610 
611 	pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
612 	if (pagevec_count(pvec))
613 		pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
614 
615 	pvec = &per_cpu(lru_deactivate_pvecs, cpu);
616 	if (pagevec_count(pvec))
617 		pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
618 
619 	pvec = &per_cpu(lru_lazyfree_pvecs, cpu);
620 	if (pagevec_count(pvec))
621 		pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
622 
623 	activate_page_drain(cpu);
624 }
625 
626 /**
627  * deactivate_file_page - forcefully deactivate a file page
628  * @page: page to deactivate
629  *
630  * This function hints the VM that @page is a good reclaim candidate,
631  * for example if its invalidation fails due to the page being dirty
632  * or under writeback.
633  */
634 void deactivate_file_page(struct page *page)
635 {
636 	/*
637 	 * In a workload with many unevictable page such as mprotect,
638 	 * unevictable page deactivation for accelerating reclaim is pointless.
639 	 */
640 	if (PageUnevictable(page))
641 		return;
642 
643 	if (likely(get_page_unless_zero(page))) {
644 		struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
645 
646 		if (!pagevec_add(pvec, page) || PageCompound(page))
647 			pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
648 		put_cpu_var(lru_deactivate_file_pvecs);
649 	}
650 }
651 
652 /*
653  * deactivate_page - deactivate a page
654  * @page: page to deactivate
655  *
656  * deactivate_page() moves @page to the inactive list if @page was on the active
657  * list and was not an unevictable page.  This is done to accelerate the reclaim
658  * of @page.
659  */
660 void deactivate_page(struct page *page)
661 {
662 	if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
663 		struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
664 
665 		get_page(page);
666 		if (!pagevec_add(pvec, page) || PageCompound(page))
667 			pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL);
668 		put_cpu_var(lru_deactivate_pvecs);
669 	}
670 }
671 
672 /**
673  * mark_page_lazyfree - make an anon page lazyfree
674  * @page: page to deactivate
675  *
676  * mark_page_lazyfree() moves @page to the inactive file list.
677  * This is done to accelerate the reclaim of @page.
678  */
679 void mark_page_lazyfree(struct page *page)
680 {
681 	if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
682 	    !PageSwapCache(page) && !PageUnevictable(page)) {
683 		struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs);
684 
685 		get_page(page);
686 		if (!pagevec_add(pvec, page) || PageCompound(page))
687 			pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
688 		put_cpu_var(lru_lazyfree_pvecs);
689 	}
690 }
691 
692 void lru_add_drain(void)
693 {
694 	lru_add_drain_cpu(get_cpu());
695 	put_cpu();
696 }
697 
698 #ifdef CONFIG_SMP
699 
700 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
701 
702 static void lru_add_drain_per_cpu(struct work_struct *dummy)
703 {
704 	lru_add_drain();
705 }
706 
707 /*
708  * Doesn't need any cpu hotplug locking because we do rely on per-cpu
709  * kworkers being shut down before our page_alloc_cpu_dead callback is
710  * executed on the offlined cpu.
711  * Calling this function with cpu hotplug locks held can actually lead
712  * to obscure indirect dependencies via WQ context.
713  */
714 void lru_add_drain_all(void)
715 {
716 	static seqcount_t seqcount = SEQCNT_ZERO(seqcount);
717 	static DEFINE_MUTEX(lock);
718 	static struct cpumask has_work;
719 	int cpu, seq;
720 
721 	/*
722 	 * Make sure nobody triggers this path before mm_percpu_wq is fully
723 	 * initialized.
724 	 */
725 	if (WARN_ON(!mm_percpu_wq))
726 		return;
727 
728 	seq = raw_read_seqcount_latch(&seqcount);
729 
730 	mutex_lock(&lock);
731 
732 	/*
733 	 * Piggyback on drain started and finished while we waited for lock:
734 	 * all pages pended at the time of our enter were drained from vectors.
735 	 */
736 	if (__read_seqcount_retry(&seqcount, seq))
737 		goto done;
738 
739 	raw_write_seqcount_latch(&seqcount);
740 
741 	cpumask_clear(&has_work);
742 
743 	for_each_online_cpu(cpu) {
744 		struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
745 
746 		if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
747 		    pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
748 		    pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
749 		    pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
750 		    pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
751 		    need_activate_page_drain(cpu)) {
752 			INIT_WORK(work, lru_add_drain_per_cpu);
753 			queue_work_on(cpu, mm_percpu_wq, work);
754 			cpumask_set_cpu(cpu, &has_work);
755 		}
756 	}
757 
758 	for_each_cpu(cpu, &has_work)
759 		flush_work(&per_cpu(lru_add_drain_work, cpu));
760 
761 done:
762 	mutex_unlock(&lock);
763 }
764 #else
765 void lru_add_drain_all(void)
766 {
767 	lru_add_drain();
768 }
769 #endif
770 
771 /**
772  * release_pages - batched put_page()
773  * @pages: array of pages to release
774  * @nr: number of pages
775  *
776  * Decrement the reference count on all the pages in @pages.  If it
777  * fell to zero, remove the page from the LRU and free it.
778  */
779 void release_pages(struct page **pages, int nr)
780 {
781 	int i;
782 	LIST_HEAD(pages_to_free);
783 	struct pglist_data *locked_pgdat = NULL;
784 	struct lruvec *lruvec;
785 	unsigned long uninitialized_var(flags);
786 	unsigned int uninitialized_var(lock_batch);
787 
788 	for (i = 0; i < nr; i++) {
789 		struct page *page = pages[i];
790 
791 		/*
792 		 * Make sure the IRQ-safe lock-holding time does not get
793 		 * excessive with a continuous string of pages from the
794 		 * same pgdat. The lock is held only if pgdat != NULL.
795 		 */
796 		if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
797 			spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
798 			locked_pgdat = NULL;
799 		}
800 
801 		if (is_huge_zero_page(page))
802 			continue;
803 
804 		if (is_zone_device_page(page)) {
805 			if (locked_pgdat) {
806 				spin_unlock_irqrestore(&locked_pgdat->lru_lock,
807 						       flags);
808 				locked_pgdat = NULL;
809 			}
810 			/*
811 			 * ZONE_DEVICE pages that return 'false' from
812 			 * put_devmap_managed_page() do not require special
813 			 * processing, and instead, expect a call to
814 			 * put_page_testzero().
815 			 */
816 			if (page_is_devmap_managed(page)) {
817 				put_devmap_managed_page(page);
818 				continue;
819 			}
820 		}
821 
822 		page = compound_head(page);
823 		if (!put_page_testzero(page))
824 			continue;
825 
826 		if (PageCompound(page)) {
827 			if (locked_pgdat) {
828 				spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
829 				locked_pgdat = NULL;
830 			}
831 			__put_compound_page(page);
832 			continue;
833 		}
834 
835 		if (PageLRU(page)) {
836 			struct pglist_data *pgdat = page_pgdat(page);
837 
838 			if (pgdat != locked_pgdat) {
839 				if (locked_pgdat)
840 					spin_unlock_irqrestore(&locked_pgdat->lru_lock,
841 									flags);
842 				lock_batch = 0;
843 				locked_pgdat = pgdat;
844 				spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
845 			}
846 
847 			lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
848 			VM_BUG_ON_PAGE(!PageLRU(page), page);
849 			__ClearPageLRU(page);
850 			del_page_from_lru_list(page, lruvec, page_off_lru(page));
851 		}
852 
853 		/* Clear Active bit in case of parallel mark_page_accessed */
854 		__ClearPageActive(page);
855 		__ClearPageWaiters(page);
856 
857 		list_add(&page->lru, &pages_to_free);
858 	}
859 	if (locked_pgdat)
860 		spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
861 
862 	mem_cgroup_uncharge_list(&pages_to_free);
863 	free_unref_page_list(&pages_to_free);
864 }
865 EXPORT_SYMBOL(release_pages);
866 
867 /*
868  * The pages which we're about to release may be in the deferred lru-addition
869  * queues.  That would prevent them from really being freed right now.  That's
870  * OK from a correctness point of view but is inefficient - those pages may be
871  * cache-warm and we want to give them back to the page allocator ASAP.
872  *
873  * So __pagevec_release() will drain those queues here.  __pagevec_lru_add()
874  * and __pagevec_lru_add_active() call release_pages() directly to avoid
875  * mutual recursion.
876  */
877 void __pagevec_release(struct pagevec *pvec)
878 {
879 	if (!pvec->percpu_pvec_drained) {
880 		lru_add_drain();
881 		pvec->percpu_pvec_drained = true;
882 	}
883 	release_pages(pvec->pages, pagevec_count(pvec));
884 	pagevec_reinit(pvec);
885 }
886 EXPORT_SYMBOL(__pagevec_release);
887 
888 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
889 /* used by __split_huge_page_refcount() */
890 void lru_add_page_tail(struct page *page, struct page *page_tail,
891 		       struct lruvec *lruvec, struct list_head *list)
892 {
893 	const int file = 0;
894 
895 	VM_BUG_ON_PAGE(!PageHead(page), page);
896 	VM_BUG_ON_PAGE(PageCompound(page_tail), page);
897 	VM_BUG_ON_PAGE(PageLRU(page_tail), page);
898 	lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock);
899 
900 	if (!list)
901 		SetPageLRU(page_tail);
902 
903 	if (likely(PageLRU(page)))
904 		list_add_tail(&page_tail->lru, &page->lru);
905 	else if (list) {
906 		/* page reclaim is reclaiming a huge page */
907 		get_page(page_tail);
908 		list_add_tail(&page_tail->lru, list);
909 	} else {
910 		/*
911 		 * Head page has not yet been counted, as an hpage,
912 		 * so we must account for each subpage individually.
913 		 *
914 		 * Put page_tail on the list at the correct position
915 		 * so they all end up in order.
916 		 */
917 		add_page_to_lru_list_tail(page_tail, lruvec,
918 					  page_lru(page_tail));
919 	}
920 
921 	if (!PageUnevictable(page))
922 		update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
923 }
924 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
925 
926 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
927 				 void *arg)
928 {
929 	enum lru_list lru;
930 	int was_unevictable = TestClearPageUnevictable(page);
931 
932 	VM_BUG_ON_PAGE(PageLRU(page), page);
933 
934 	/*
935 	 * Page becomes evictable in two ways:
936 	 * 1) Within LRU lock [munlock_vma_page() and __munlock_pagevec()].
937 	 * 2) Before acquiring LRU lock to put the page to correct LRU and then
938 	 *   a) do PageLRU check with lock [check_move_unevictable_pages]
939 	 *   b) do PageLRU check before lock [clear_page_mlock]
940 	 *
941 	 * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need
942 	 * following strict ordering:
943 	 *
944 	 * #0: __pagevec_lru_add_fn		#1: clear_page_mlock
945 	 *
946 	 * SetPageLRU()				TestClearPageMlocked()
947 	 * smp_mb() // explicit ordering	// above provides strict
948 	 *					// ordering
949 	 * PageMlocked()			PageLRU()
950 	 *
951 	 *
952 	 * if '#1' does not observe setting of PG_lru by '#0' and fails
953 	 * isolation, the explicit barrier will make sure that page_evictable
954 	 * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
955 	 * can be reordered after PageMlocked check and can make '#1' to fail
956 	 * the isolation of the page whose Mlocked bit is cleared (#0 is also
957 	 * looking at the same page) and the evictable page will be stranded
958 	 * in an unevictable LRU.
959 	 */
960 	SetPageLRU(page);
961 	smp_mb__after_atomic();
962 
963 	if (page_evictable(page)) {
964 		lru = page_lru(page);
965 		update_page_reclaim_stat(lruvec, page_is_file_lru(page),
966 					 PageActive(page));
967 		if (was_unevictable)
968 			count_vm_event(UNEVICTABLE_PGRESCUED);
969 	} else {
970 		lru = LRU_UNEVICTABLE;
971 		ClearPageActive(page);
972 		SetPageUnevictable(page);
973 		if (!was_unevictable)
974 			count_vm_event(UNEVICTABLE_PGCULLED);
975 	}
976 
977 	add_page_to_lru_list(page, lruvec, lru);
978 	trace_mm_lru_insertion(page, lru);
979 }
980 
981 /*
982  * Add the passed pages to the LRU, then drop the caller's refcount
983  * on them.  Reinitialises the caller's pagevec.
984  */
985 void __pagevec_lru_add(struct pagevec *pvec)
986 {
987 	pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
988 }
989 
990 /**
991  * pagevec_lookup_entries - gang pagecache lookup
992  * @pvec:	Where the resulting entries are placed
993  * @mapping:	The address_space to search
994  * @start:	The starting entry index
995  * @nr_entries:	The maximum number of pages
996  * @indices:	The cache indices corresponding to the entries in @pvec
997  *
998  * pagevec_lookup_entries() will search for and return a group of up
999  * to @nr_pages pages and shadow entries in the mapping.  All
1000  * entries are placed in @pvec.  pagevec_lookup_entries() takes a
1001  * reference against actual pages in @pvec.
1002  *
1003  * The search returns a group of mapping-contiguous entries with
1004  * ascending indexes.  There may be holes in the indices due to
1005  * not-present entries.
1006  *
1007  * Only one subpage of a Transparent Huge Page is returned in one call:
1008  * allowing truncate_inode_pages_range() to evict the whole THP without
1009  * cycling through a pagevec of extra references.
1010  *
1011  * pagevec_lookup_entries() returns the number of entries which were
1012  * found.
1013  */
1014 unsigned pagevec_lookup_entries(struct pagevec *pvec,
1015 				struct address_space *mapping,
1016 				pgoff_t start, unsigned nr_entries,
1017 				pgoff_t *indices)
1018 {
1019 	pvec->nr = find_get_entries(mapping, start, nr_entries,
1020 				    pvec->pages, indices);
1021 	return pagevec_count(pvec);
1022 }
1023 
1024 /**
1025  * pagevec_remove_exceptionals - pagevec exceptionals pruning
1026  * @pvec:	The pagevec to prune
1027  *
1028  * pagevec_lookup_entries() fills both pages and exceptional radix
1029  * tree entries into the pagevec.  This function prunes all
1030  * exceptionals from @pvec without leaving holes, so that it can be
1031  * passed on to page-only pagevec operations.
1032  */
1033 void pagevec_remove_exceptionals(struct pagevec *pvec)
1034 {
1035 	int i, j;
1036 
1037 	for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
1038 		struct page *page = pvec->pages[i];
1039 		if (!xa_is_value(page))
1040 			pvec->pages[j++] = page;
1041 	}
1042 	pvec->nr = j;
1043 }
1044 
1045 /**
1046  * pagevec_lookup_range - gang pagecache lookup
1047  * @pvec:	Where the resulting pages are placed
1048  * @mapping:	The address_space to search
1049  * @start:	The starting page index
1050  * @end:	The final page index
1051  *
1052  * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE
1053  * pages in the mapping starting from index @start and upto index @end
1054  * (inclusive).  The pages are placed in @pvec.  pagevec_lookup() takes a
1055  * reference against the pages in @pvec.
1056  *
1057  * The search returns a group of mapping-contiguous pages with ascending
1058  * indexes.  There may be holes in the indices due to not-present pages. We
1059  * also update @start to index the next page for the traversal.
1060  *
1061  * pagevec_lookup_range() returns the number of pages which were found. If this
1062  * number is smaller than PAGEVEC_SIZE, the end of specified range has been
1063  * reached.
1064  */
1065 unsigned pagevec_lookup_range(struct pagevec *pvec,
1066 		struct address_space *mapping, pgoff_t *start, pgoff_t end)
1067 {
1068 	pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
1069 					pvec->pages);
1070 	return pagevec_count(pvec);
1071 }
1072 EXPORT_SYMBOL(pagevec_lookup_range);
1073 
1074 unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
1075 		struct address_space *mapping, pgoff_t *index, pgoff_t end,
1076 		xa_mark_t tag)
1077 {
1078 	pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1079 					PAGEVEC_SIZE, pvec->pages);
1080 	return pagevec_count(pvec);
1081 }
1082 EXPORT_SYMBOL(pagevec_lookup_range_tag);
1083 
1084 unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
1085 		struct address_space *mapping, pgoff_t *index, pgoff_t end,
1086 		xa_mark_t tag, unsigned max_pages)
1087 {
1088 	pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1089 		min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
1090 	return pagevec_count(pvec);
1091 }
1092 EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
1093 /*
1094  * Perform any setup for the swap system
1095  */
1096 void __init swap_setup(void)
1097 {
1098 	unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
1099 
1100 	/* Use a smaller cluster for small-memory machines */
1101 	if (megs < 16)
1102 		page_cluster = 2;
1103 	else
1104 		page_cluster = 3;
1105 	/*
1106 	 * Right now other parts of the system means that we
1107 	 * _really_ don't want to cluster much more
1108 	 */
1109 }
1110 
1111 #ifdef CONFIG_DEV_PAGEMAP_OPS
1112 void put_devmap_managed_page(struct page *page)
1113 {
1114 	int count;
1115 
1116 	if (WARN_ON_ONCE(!page_is_devmap_managed(page)))
1117 		return;
1118 
1119 	count = page_ref_dec_return(page);
1120 
1121 	/*
1122 	 * devmap page refcounts are 1-based, rather than 0-based: if
1123 	 * refcount is 1, then the page is free and the refcount is
1124 	 * stable because nobody holds a reference on the page.
1125 	 */
1126 	if (count == 1)
1127 		free_devmap_managed_page(page);
1128 	else if (!count)
1129 		__put_page(page);
1130 }
1131 EXPORT_SYMBOL(put_devmap_managed_page);
1132 #endif
1133