xref: /openbmc/linux/mm/swap.c (revision 15d90a6a)
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? */
44 int page_cluster;
45 
46 static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
47 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
48 static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
49 static DEFINE_PER_CPU(struct pagevec, lru_lazyfree_pvecs);
50 #ifdef CONFIG_SMP
51 static 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  */
58 static void __page_cache_release(struct page *page)
59 {
60 	if (PageLRU(page)) {
61 		struct zone *zone = page_zone(page);
62 		struct lruvec *lruvec;
63 		unsigned long flags;
64 
65 		spin_lock_irqsave(zone_lru_lock(zone), flags);
66 		lruvec = mem_cgroup_page_lruvec(page, zone->zone_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(zone_lru_lock(zone), flags);
71 	}
72 	__ClearPageWaiters(page);
73 	mem_cgroup_uncharge(page);
74 }
75 
76 static void __put_single_page(struct page *page)
77 {
78 	__page_cache_release(page);
79 	free_unref_page(page);
80 }
81 
82 static 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 
98 void __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 }
115 EXPORT_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  */
124 void 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 }
134 EXPORT_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  */
149 int 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 }
164 EXPORT_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  */
177 int 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 }
186 EXPORT_SYMBOL_GPL(get_kernel_page);
187 
188 static 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 
217 static 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  */
234 static 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  */
247 void 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 
263 static 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 
273 static 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
292 static 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 
300 static bool need_activate_page_drain(int cpu)
301 {
302 	return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0;
303 }
304 
305 void 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
319 static inline void activate_page_drain(int cpu)
320 {
321 }
322 
323 static bool need_activate_page_drain(int cpu)
324 {
325 	return false;
326 }
327 
328 void activate_page(struct page *page)
329 {
330 	struct zone *zone = page_zone(page);
331 
332 	page = compound_head(page);
333 	spin_lock_irq(zone_lru_lock(zone));
334 	__activate_page(page, mem_cgroup_page_lruvec(page, zone->zone_pgdat), NULL);
335 	spin_unlock_irq(zone_lru_lock(zone));
336 }
337 #endif
338 
339 static void __lru_cache_activate_page(struct page *page)
340 {
341 	struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
342 	int i;
343 
344 	/*
345 	 * Search backwards on the optimistic assumption that the page being
346 	 * activated has just been added to this pagevec. Note that only
347 	 * the local pagevec is examined as a !PageLRU page could be in the
348 	 * process of being released, reclaimed, migrated or on a remote
349 	 * pagevec that is currently being drained. Furthermore, marking
350 	 * a remote pagevec's page PageActive potentially hits a race where
351 	 * a page is marked PageActive just after it is added to the inactive
352 	 * list causing accounting errors and BUG_ON checks to trigger.
353 	 */
354 	for (i = pagevec_count(pvec) - 1; i >= 0; i--) {
355 		struct page *pagevec_page = pvec->pages[i];
356 
357 		if (pagevec_page == page) {
358 			SetPageActive(page);
359 			break;
360 		}
361 	}
362 
363 	put_cpu_var(lru_add_pvec);
364 }
365 
366 /*
367  * Mark a page as having seen activity.
368  *
369  * inactive,unreferenced	->	inactive,referenced
370  * inactive,referenced		->	active,unreferenced
371  * active,unreferenced		->	active,referenced
372  *
373  * When a newly allocated page is not yet visible, so safe for non-atomic ops,
374  * __SetPageReferenced(page) may be substituted for mark_page_accessed(page).
375  */
376 void mark_page_accessed(struct page *page)
377 {
378 	page = compound_head(page);
379 	if (!PageActive(page) && !PageUnevictable(page) &&
380 			PageReferenced(page)) {
381 
382 		/*
383 		 * If the page is on the LRU, queue it for activation via
384 		 * activate_page_pvecs. Otherwise, assume the page is on a
385 		 * pagevec, mark it active and it'll be moved to the active
386 		 * LRU on the next drain.
387 		 */
388 		if (PageLRU(page))
389 			activate_page(page);
390 		else
391 			__lru_cache_activate_page(page);
392 		ClearPageReferenced(page);
393 		if (page_is_file_cache(page))
394 			workingset_activation(page);
395 	} else if (!PageReferenced(page)) {
396 		SetPageReferenced(page);
397 	}
398 	if (page_is_idle(page))
399 		clear_page_idle(page);
400 }
401 EXPORT_SYMBOL(mark_page_accessed);
402 
403 static void __lru_cache_add(struct page *page)
404 {
405 	struct pagevec *pvec = &get_cpu_var(lru_add_pvec);
406 
407 	get_page(page);
408 	if (!pagevec_add(pvec, page) || PageCompound(page))
409 		__pagevec_lru_add(pvec);
410 	put_cpu_var(lru_add_pvec);
411 }
412 
413 /**
414  * lru_cache_add_anon - add a page to the page lists
415  * @page: the page to add
416  */
417 void lru_cache_add_anon(struct page *page)
418 {
419 	if (PageActive(page))
420 		ClearPageActive(page);
421 	__lru_cache_add(page);
422 }
423 
424 void lru_cache_add_file(struct page *page)
425 {
426 	if (PageActive(page))
427 		ClearPageActive(page);
428 	__lru_cache_add(page);
429 }
430 EXPORT_SYMBOL(lru_cache_add_file);
431 
432 /**
433  * lru_cache_add - add a page to a page list
434  * @page: the page to be added to the LRU.
435  *
436  * Queue the page for addition to the LRU via pagevec. The decision on whether
437  * to add the page to the [in]active [file|anon] list is deferred until the
438  * pagevec is drained. This gives a chance for the caller of lru_cache_add()
439  * have the page added to the active list using mark_page_accessed().
440  */
441 void lru_cache_add(struct page *page)
442 {
443 	VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page);
444 	VM_BUG_ON_PAGE(PageLRU(page), page);
445 	__lru_cache_add(page);
446 }
447 
448 /**
449  * lru_cache_add_active_or_unevictable
450  * @page:  the page to be added to LRU
451  * @vma:   vma in which page is mapped for determining reclaimability
452  *
453  * Place @page on the active or unevictable LRU list, depending on its
454  * evictability.  Note that if the page is not evictable, it goes
455  * directly back onto it's zone's unevictable list, it does NOT use a
456  * per cpu pagevec.
457  */
458 void lru_cache_add_active_or_unevictable(struct page *page,
459 					 struct vm_area_struct *vma)
460 {
461 	VM_BUG_ON_PAGE(PageLRU(page), page);
462 
463 	if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
464 		SetPageActive(page);
465 	else if (!TestSetPageMlocked(page)) {
466 		/*
467 		 * We use the irq-unsafe __mod_zone_page_stat because this
468 		 * counter is not modified from interrupt context, and the pte
469 		 * lock is held(spinlock), which implies preemption disabled.
470 		 */
471 		__mod_zone_page_state(page_zone(page), NR_MLOCK,
472 				    hpage_nr_pages(page));
473 		count_vm_event(UNEVICTABLE_PGMLOCKED);
474 	}
475 	lru_cache_add(page);
476 }
477 
478 /*
479  * If the page can not be invalidated, it is moved to the
480  * inactive list to speed up its reclaim.  It is moved to the
481  * head of the list, rather than the tail, to give the flusher
482  * threads some time to write it out, as this is much more
483  * effective than the single-page writeout from reclaim.
484  *
485  * If the page isn't page_mapped and dirty/writeback, the page
486  * could reclaim asap using PG_reclaim.
487  *
488  * 1. active, mapped page -> none
489  * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
490  * 3. inactive, mapped page -> none
491  * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
492  * 5. inactive, clean -> inactive, tail
493  * 6. Others -> none
494  *
495  * In 4, why it moves inactive's head, the VM expects the page would
496  * be write it out by flusher threads as this is much more effective
497  * than the single-page writeout from reclaim.
498  */
499 static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
500 			      void *arg)
501 {
502 	int lru, file;
503 	bool active;
504 
505 	if (!PageLRU(page))
506 		return;
507 
508 	if (PageUnevictable(page))
509 		return;
510 
511 	/* Some processes are using the page */
512 	if (page_mapped(page))
513 		return;
514 
515 	active = PageActive(page);
516 	file = page_is_file_cache(page);
517 	lru = page_lru_base_type(page);
518 
519 	del_page_from_lru_list(page, lruvec, lru + active);
520 	ClearPageActive(page);
521 	ClearPageReferenced(page);
522 	add_page_to_lru_list(page, lruvec, lru);
523 
524 	if (PageWriteback(page) || PageDirty(page)) {
525 		/*
526 		 * PG_reclaim could be raced with end_page_writeback
527 		 * It can make readahead confusing.  But race window
528 		 * is _really_ small and  it's non-critical problem.
529 		 */
530 		SetPageReclaim(page);
531 	} else {
532 		/*
533 		 * The page's writeback ends up during pagevec
534 		 * We moves tha page into tail of inactive.
535 		 */
536 		list_move_tail(&page->lru, &lruvec->lists[lru]);
537 		__count_vm_event(PGROTATED);
538 	}
539 
540 	if (active)
541 		__count_vm_event(PGDEACTIVATE);
542 	update_page_reclaim_stat(lruvec, file, 0);
543 }
544 
545 
546 static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
547 			    void *arg)
548 {
549 	if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
550 	    !PageSwapCache(page) && !PageUnevictable(page)) {
551 		bool active = PageActive(page);
552 
553 		del_page_from_lru_list(page, lruvec,
554 				       LRU_INACTIVE_ANON + active);
555 		ClearPageActive(page);
556 		ClearPageReferenced(page);
557 		/*
558 		 * lazyfree pages are clean anonymous pages. They have
559 		 * SwapBacked flag cleared to distinguish normal anonymous
560 		 * pages
561 		 */
562 		ClearPageSwapBacked(page);
563 		add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
564 
565 		__count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
566 		count_memcg_page_event(page, PGLAZYFREE);
567 		update_page_reclaim_stat(lruvec, 1, 0);
568 	}
569 }
570 
571 /*
572  * Drain pages out of the cpu's pagevecs.
573  * Either "cpu" is the current CPU, and preemption has already been
574  * disabled; or "cpu" is being hot-unplugged, and is already dead.
575  */
576 void lru_add_drain_cpu(int cpu)
577 {
578 	struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu);
579 
580 	if (pagevec_count(pvec))
581 		__pagevec_lru_add(pvec);
582 
583 	pvec = &per_cpu(lru_rotate_pvecs, cpu);
584 	if (pagevec_count(pvec)) {
585 		unsigned long flags;
586 
587 		/* No harm done if a racing interrupt already did this */
588 		local_irq_save(flags);
589 		pagevec_move_tail(pvec);
590 		local_irq_restore(flags);
591 	}
592 
593 	pvec = &per_cpu(lru_deactivate_file_pvecs, cpu);
594 	if (pagevec_count(pvec))
595 		pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
596 
597 	pvec = &per_cpu(lru_lazyfree_pvecs, cpu);
598 	if (pagevec_count(pvec))
599 		pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
600 
601 	activate_page_drain(cpu);
602 }
603 
604 /**
605  * deactivate_file_page - forcefully deactivate a file page
606  * @page: page to deactivate
607  *
608  * This function hints the VM that @page is a good reclaim candidate,
609  * for example if its invalidation fails due to the page being dirty
610  * or under writeback.
611  */
612 void deactivate_file_page(struct page *page)
613 {
614 	/*
615 	 * In a workload with many unevictable page such as mprotect,
616 	 * unevictable page deactivation for accelerating reclaim is pointless.
617 	 */
618 	if (PageUnevictable(page))
619 		return;
620 
621 	if (likely(get_page_unless_zero(page))) {
622 		struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs);
623 
624 		if (!pagevec_add(pvec, page) || PageCompound(page))
625 			pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
626 		put_cpu_var(lru_deactivate_file_pvecs);
627 	}
628 }
629 
630 /**
631  * mark_page_lazyfree - make an anon page lazyfree
632  * @page: page to deactivate
633  *
634  * mark_page_lazyfree() moves @page to the inactive file list.
635  * This is done to accelerate the reclaim of @page.
636  */
637 void mark_page_lazyfree(struct page *page)
638 {
639 	if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
640 	    !PageSwapCache(page) && !PageUnevictable(page)) {
641 		struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs);
642 
643 		get_page(page);
644 		if (!pagevec_add(pvec, page) || PageCompound(page))
645 			pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
646 		put_cpu_var(lru_lazyfree_pvecs);
647 	}
648 }
649 
650 void lru_add_drain(void)
651 {
652 	lru_add_drain_cpu(get_cpu());
653 	put_cpu();
654 }
655 
656 static void lru_add_drain_per_cpu(struct work_struct *dummy)
657 {
658 	lru_add_drain();
659 }
660 
661 static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work);
662 
663 /*
664  * Doesn't need any cpu hotplug locking because we do rely on per-cpu
665  * kworkers being shut down before our page_alloc_cpu_dead callback is
666  * executed on the offlined cpu.
667  * Calling this function with cpu hotplug locks held can actually lead
668  * to obscure indirect dependencies via WQ context.
669  */
670 void lru_add_drain_all(void)
671 {
672 	static DEFINE_MUTEX(lock);
673 	static struct cpumask has_work;
674 	int cpu;
675 
676 	/*
677 	 * Make sure nobody triggers this path before mm_percpu_wq is fully
678 	 * initialized.
679 	 */
680 	if (WARN_ON(!mm_percpu_wq))
681 		return;
682 
683 	mutex_lock(&lock);
684 	cpumask_clear(&has_work);
685 
686 	for_each_online_cpu(cpu) {
687 		struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
688 
689 		if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
690 		    pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
691 		    pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
692 		    pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
693 		    need_activate_page_drain(cpu)) {
694 			INIT_WORK(work, lru_add_drain_per_cpu);
695 			queue_work_on(cpu, mm_percpu_wq, work);
696 			cpumask_set_cpu(cpu, &has_work);
697 		}
698 	}
699 
700 	for_each_cpu(cpu, &has_work)
701 		flush_work(&per_cpu(lru_add_drain_work, cpu));
702 
703 	mutex_unlock(&lock);
704 }
705 
706 /**
707  * release_pages - batched put_page()
708  * @pages: array of pages to release
709  * @nr: number of pages
710  *
711  * Decrement the reference count on all the pages in @pages.  If it
712  * fell to zero, remove the page from the LRU and free it.
713  */
714 void release_pages(struct page **pages, int nr)
715 {
716 	int i;
717 	LIST_HEAD(pages_to_free);
718 	struct pglist_data *locked_pgdat = NULL;
719 	struct lruvec *lruvec;
720 	unsigned long uninitialized_var(flags);
721 	unsigned int uninitialized_var(lock_batch);
722 
723 	for (i = 0; i < nr; i++) {
724 		struct page *page = pages[i];
725 
726 		/*
727 		 * Make sure the IRQ-safe lock-holding time does not get
728 		 * excessive with a continuous string of pages from the
729 		 * same pgdat. The lock is held only if pgdat != NULL.
730 		 */
731 		if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) {
732 			spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
733 			locked_pgdat = NULL;
734 		}
735 
736 		if (is_huge_zero_page(page))
737 			continue;
738 
739 		/* Device public page can not be huge page */
740 		if (is_device_public_page(page)) {
741 			if (locked_pgdat) {
742 				spin_unlock_irqrestore(&locked_pgdat->lru_lock,
743 						       flags);
744 				locked_pgdat = NULL;
745 			}
746 			put_devmap_managed_page(page);
747 			continue;
748 		}
749 
750 		page = compound_head(page);
751 		if (!put_page_testzero(page))
752 			continue;
753 
754 		if (PageCompound(page)) {
755 			if (locked_pgdat) {
756 				spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
757 				locked_pgdat = NULL;
758 			}
759 			__put_compound_page(page);
760 			continue;
761 		}
762 
763 		if (PageLRU(page)) {
764 			struct pglist_data *pgdat = page_pgdat(page);
765 
766 			if (pgdat != locked_pgdat) {
767 				if (locked_pgdat)
768 					spin_unlock_irqrestore(&locked_pgdat->lru_lock,
769 									flags);
770 				lock_batch = 0;
771 				locked_pgdat = pgdat;
772 				spin_lock_irqsave(&locked_pgdat->lru_lock, flags);
773 			}
774 
775 			lruvec = mem_cgroup_page_lruvec(page, locked_pgdat);
776 			VM_BUG_ON_PAGE(!PageLRU(page), page);
777 			__ClearPageLRU(page);
778 			del_page_from_lru_list(page, lruvec, page_off_lru(page));
779 		}
780 
781 		/* Clear Active bit in case of parallel mark_page_accessed */
782 		__ClearPageActive(page);
783 		__ClearPageWaiters(page);
784 
785 		list_add(&page->lru, &pages_to_free);
786 	}
787 	if (locked_pgdat)
788 		spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags);
789 
790 	mem_cgroup_uncharge_list(&pages_to_free);
791 	free_unref_page_list(&pages_to_free);
792 }
793 EXPORT_SYMBOL(release_pages);
794 
795 /*
796  * The pages which we're about to release may be in the deferred lru-addition
797  * queues.  That would prevent them from really being freed right now.  That's
798  * OK from a correctness point of view but is inefficient - those pages may be
799  * cache-warm and we want to give them back to the page allocator ASAP.
800  *
801  * So __pagevec_release() will drain those queues here.  __pagevec_lru_add()
802  * and __pagevec_lru_add_active() call release_pages() directly to avoid
803  * mutual recursion.
804  */
805 void __pagevec_release(struct pagevec *pvec)
806 {
807 	if (!pvec->percpu_pvec_drained) {
808 		lru_add_drain();
809 		pvec->percpu_pvec_drained = true;
810 	}
811 	release_pages(pvec->pages, pagevec_count(pvec));
812 	pagevec_reinit(pvec);
813 }
814 EXPORT_SYMBOL(__pagevec_release);
815 
816 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
817 /* used by __split_huge_page_refcount() */
818 void lru_add_page_tail(struct page *page, struct page *page_tail,
819 		       struct lruvec *lruvec, struct list_head *list)
820 {
821 	const int file = 0;
822 
823 	VM_BUG_ON_PAGE(!PageHead(page), page);
824 	VM_BUG_ON_PAGE(PageCompound(page_tail), page);
825 	VM_BUG_ON_PAGE(PageLRU(page_tail), page);
826 	lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock);
827 
828 	if (!list)
829 		SetPageLRU(page_tail);
830 
831 	if (likely(PageLRU(page)))
832 		list_add_tail(&page_tail->lru, &page->lru);
833 	else if (list) {
834 		/* page reclaim is reclaiming a huge page */
835 		get_page(page_tail);
836 		list_add_tail(&page_tail->lru, list);
837 	} else {
838 		struct list_head *list_head;
839 		/*
840 		 * Head page has not yet been counted, as an hpage,
841 		 * so we must account for each subpage individually.
842 		 *
843 		 * Use the standard add function to put page_tail on the list,
844 		 * but then correct its position so they all end up in order.
845 		 */
846 		add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail));
847 		list_head = page_tail->lru.prev;
848 		list_move_tail(&page_tail->lru, list_head);
849 	}
850 
851 	if (!PageUnevictable(page))
852 		update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
853 }
854 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
855 
856 static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
857 				 void *arg)
858 {
859 	enum lru_list lru;
860 	int was_unevictable = TestClearPageUnevictable(page);
861 
862 	VM_BUG_ON_PAGE(PageLRU(page), page);
863 
864 	SetPageLRU(page);
865 	/*
866 	 * Page becomes evictable in two ways:
867 	 * 1) Within LRU lock [munlock_vma_pages() and __munlock_pagevec()].
868 	 * 2) Before acquiring LRU lock to put the page to correct LRU and then
869 	 *   a) do PageLRU check with lock [check_move_unevictable_pages]
870 	 *   b) do PageLRU check before lock [clear_page_mlock]
871 	 *
872 	 * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need
873 	 * following strict ordering:
874 	 *
875 	 * #0: __pagevec_lru_add_fn		#1: clear_page_mlock
876 	 *
877 	 * SetPageLRU()				TestClearPageMlocked()
878 	 * smp_mb() // explicit ordering	// above provides strict
879 	 *					// ordering
880 	 * PageMlocked()			PageLRU()
881 	 *
882 	 *
883 	 * if '#1' does not observe setting of PG_lru by '#0' and fails
884 	 * isolation, the explicit barrier will make sure that page_evictable
885 	 * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
886 	 * can be reordered after PageMlocked check and can make '#1' to fail
887 	 * the isolation of the page whose Mlocked bit is cleared (#0 is also
888 	 * looking at the same page) and the evictable page will be stranded
889 	 * in an unevictable LRU.
890 	 */
891 	smp_mb();
892 
893 	if (page_evictable(page)) {
894 		lru = page_lru(page);
895 		update_page_reclaim_stat(lruvec, page_is_file_cache(page),
896 					 PageActive(page));
897 		if (was_unevictable)
898 			count_vm_event(UNEVICTABLE_PGRESCUED);
899 	} else {
900 		lru = LRU_UNEVICTABLE;
901 		ClearPageActive(page);
902 		SetPageUnevictable(page);
903 		if (!was_unevictable)
904 			count_vm_event(UNEVICTABLE_PGCULLED);
905 	}
906 
907 	add_page_to_lru_list(page, lruvec, lru);
908 	trace_mm_lru_insertion(page, lru);
909 }
910 
911 /*
912  * Add the passed pages to the LRU, then drop the caller's refcount
913  * on them.  Reinitialises the caller's pagevec.
914  */
915 void __pagevec_lru_add(struct pagevec *pvec)
916 {
917 	pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL);
918 }
919 EXPORT_SYMBOL(__pagevec_lru_add);
920 
921 /**
922  * pagevec_lookup_entries - gang pagecache lookup
923  * @pvec:	Where the resulting entries are placed
924  * @mapping:	The address_space to search
925  * @start:	The starting entry index
926  * @nr_entries:	The maximum number of pages
927  * @indices:	The cache indices corresponding to the entries in @pvec
928  *
929  * pagevec_lookup_entries() will search for and return a group of up
930  * to @nr_pages pages and shadow entries in the mapping.  All
931  * entries are placed in @pvec.  pagevec_lookup_entries() takes a
932  * reference against actual pages in @pvec.
933  *
934  * The search returns a group of mapping-contiguous entries with
935  * ascending indexes.  There may be holes in the indices due to
936  * not-present entries.
937  *
938  * pagevec_lookup_entries() returns the number of entries which were
939  * found.
940  */
941 unsigned pagevec_lookup_entries(struct pagevec *pvec,
942 				struct address_space *mapping,
943 				pgoff_t start, unsigned nr_entries,
944 				pgoff_t *indices)
945 {
946 	pvec->nr = find_get_entries(mapping, start, nr_entries,
947 				    pvec->pages, indices);
948 	return pagevec_count(pvec);
949 }
950 
951 /**
952  * pagevec_remove_exceptionals - pagevec exceptionals pruning
953  * @pvec:	The pagevec to prune
954  *
955  * pagevec_lookup_entries() fills both pages and exceptional radix
956  * tree entries into the pagevec.  This function prunes all
957  * exceptionals from @pvec without leaving holes, so that it can be
958  * passed on to page-only pagevec operations.
959  */
960 void pagevec_remove_exceptionals(struct pagevec *pvec)
961 {
962 	int i, j;
963 
964 	for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
965 		struct page *page = pvec->pages[i];
966 		if (!xa_is_value(page))
967 			pvec->pages[j++] = page;
968 	}
969 	pvec->nr = j;
970 }
971 
972 /**
973  * pagevec_lookup_range - gang pagecache lookup
974  * @pvec:	Where the resulting pages are placed
975  * @mapping:	The address_space to search
976  * @start:	The starting page index
977  * @end:	The final page index
978  *
979  * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE
980  * pages in the mapping starting from index @start and upto index @end
981  * (inclusive).  The pages are placed in @pvec.  pagevec_lookup() takes a
982  * reference against the pages in @pvec.
983  *
984  * The search returns a group of mapping-contiguous pages with ascending
985  * indexes.  There may be holes in the indices due to not-present pages. We
986  * also update @start to index the next page for the traversal.
987  *
988  * pagevec_lookup_range() returns the number of pages which were found. If this
989  * number is smaller than PAGEVEC_SIZE, the end of specified range has been
990  * reached.
991  */
992 unsigned pagevec_lookup_range(struct pagevec *pvec,
993 		struct address_space *mapping, pgoff_t *start, pgoff_t end)
994 {
995 	pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE,
996 					pvec->pages);
997 	return pagevec_count(pvec);
998 }
999 EXPORT_SYMBOL(pagevec_lookup_range);
1000 
1001 unsigned pagevec_lookup_range_tag(struct pagevec *pvec,
1002 		struct address_space *mapping, pgoff_t *index, pgoff_t end,
1003 		xa_mark_t tag)
1004 {
1005 	pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1006 					PAGEVEC_SIZE, pvec->pages);
1007 	return pagevec_count(pvec);
1008 }
1009 EXPORT_SYMBOL(pagevec_lookup_range_tag);
1010 
1011 unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec,
1012 		struct address_space *mapping, pgoff_t *index, pgoff_t end,
1013 		xa_mark_t tag, unsigned max_pages)
1014 {
1015 	pvec->nr = find_get_pages_range_tag(mapping, index, end, tag,
1016 		min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages);
1017 	return pagevec_count(pvec);
1018 }
1019 EXPORT_SYMBOL(pagevec_lookup_range_nr_tag);
1020 /*
1021  * Perform any setup for the swap system
1022  */
1023 void __init swap_setup(void)
1024 {
1025 	unsigned long megs = totalram_pages() >> (20 - PAGE_SHIFT);
1026 
1027 	/* Use a smaller cluster for small-memory machines */
1028 	if (megs < 16)
1029 		page_cluster = 2;
1030 	else
1031 		page_cluster = 3;
1032 	/*
1033 	 * Right now other parts of the system means that we
1034 	 * _really_ don't want to cluster much more
1035 	 */
1036 }
1037