swap.c revision f8891e5e1f93a128c3900f82035e8541357896a7 
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 opereation 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/module.h>
25#include <linux/mm_inline.h>
26#include <linux/buffer_head.h>	/* for try_to_release_page() */
27#include <linux/module.h>
28#include <linux/percpu_counter.h>
29#include <linux/percpu.h>
30#include <linux/cpu.h>
31#include <linux/notifier.h>
32#include <linux/init.h>
33
34/* How many pages do we try to swap or page in/out together? */
35int page_cluster;
36
37static void put_compound_page(struct page *page)
38{
39	page = (struct page *)page_private(page);
40	if (put_page_testzero(page)) {
41		void (*dtor)(struct page *page);
42
43		dtor = (void (*)(struct page *))page[1].lru.next;
44		(*dtor)(page);
45	}
46}
47
48void put_page(struct page *page)
49{
50	if (unlikely(PageCompound(page)))
51		put_compound_page(page);
52	else if (put_page_testzero(page))
53		__page_cache_release(page);
54}
55EXPORT_SYMBOL(put_page);
56
57/*
58 * Writeback is about to end against a page which has been marked for immediate
59 * reclaim.  If it still appears to be reclaimable, move it to the tail of the
60 * inactive list.  The page still has PageWriteback set, which will pin it.
61 *
62 * We don't expect many pages to come through here, so don't bother batching
63 * things up.
64 *
65 * To avoid placing the page at the tail of the LRU while PG_writeback is still
66 * set, this function will clear PG_writeback before performing the page
67 * motion.  Do that inside the lru lock because once PG_writeback is cleared
68 * we may not touch the page.
69 *
70 * Returns zero if it cleared PG_writeback.
71 */
72int rotate_reclaimable_page(struct page *page)
73{
74	struct zone *zone;
75	unsigned long flags;
76
77	if (PageLocked(page))
78		return 1;
79	if (PageDirty(page))
80		return 1;
81	if (PageActive(page))
82		return 1;
83	if (!PageLRU(page))
84		return 1;
85
86	zone = page_zone(page);
87	spin_lock_irqsave(&zone->lru_lock, flags);
88	if (PageLRU(page) && !PageActive(page)) {
89		list_move_tail(&page->lru, &zone->inactive_list);
90		__count_vm_event(PGROTATED);
91	}
92	if (!test_clear_page_writeback(page))
93		BUG();
94	spin_unlock_irqrestore(&zone->lru_lock, flags);
95	return 0;
96}
97
98/*
99 * FIXME: speed this up?
100 */
101void fastcall activate_page(struct page *page)
102{
103	struct zone *zone = page_zone(page);
104
105	spin_lock_irq(&zone->lru_lock);
106	if (PageLRU(page) && !PageActive(page)) {
107		del_page_from_inactive_list(zone, page);
108		SetPageActive(page);
109		add_page_to_active_list(zone, page);
110		__count_vm_event(PGACTIVATE);
111	}
112	spin_unlock_irq(&zone->lru_lock);
113}
114
115/*
116 * Mark a page as having seen activity.
117 *
118 * inactive,unreferenced	->	inactive,referenced
119 * inactive,referenced		->	active,unreferenced
120 * active,unreferenced		->	active,referenced
121 */
122void fastcall mark_page_accessed(struct page *page)
123{
124	if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
125		activate_page(page);
126		ClearPageReferenced(page);
127	} else if (!PageReferenced(page)) {
128		SetPageReferenced(page);
129	}
130}
131
132EXPORT_SYMBOL(mark_page_accessed);
133
134/**
135 * lru_cache_add: add a page to the page lists
136 * @page: the page to add
137 */
138static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
139static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
140
141void fastcall lru_cache_add(struct page *page)
142{
143	struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
144
145	page_cache_get(page);
146	if (!pagevec_add(pvec, page))
147		__pagevec_lru_add(pvec);
148	put_cpu_var(lru_add_pvecs);
149}
150
151void fastcall lru_cache_add_active(struct page *page)
152{
153	struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
154
155	page_cache_get(page);
156	if (!pagevec_add(pvec, page))
157		__pagevec_lru_add_active(pvec);
158	put_cpu_var(lru_add_active_pvecs);
159}
160
161static void __lru_add_drain(int cpu)
162{
163	struct pagevec *pvec = &per_cpu(lru_add_pvecs, cpu);
164
165	/* CPU is dead, so no locking needed. */
166	if (pagevec_count(pvec))
167		__pagevec_lru_add(pvec);
168	pvec = &per_cpu(lru_add_active_pvecs, cpu);
169	if (pagevec_count(pvec))
170		__pagevec_lru_add_active(pvec);
171}
172
173void lru_add_drain(void)
174{
175	__lru_add_drain(get_cpu());
176	put_cpu();
177}
178
179#ifdef CONFIG_NUMA
180static void lru_add_drain_per_cpu(void *dummy)
181{
182	lru_add_drain();
183}
184
185/*
186 * Returns 0 for success
187 */
188int lru_add_drain_all(void)
189{
190	return schedule_on_each_cpu(lru_add_drain_per_cpu, NULL);
191}
192
193#else
194
195/*
196 * Returns 0 for success
197 */
198int lru_add_drain_all(void)
199{
200	lru_add_drain();
201	return 0;
202}
203#endif
204
205/*
206 * This path almost never happens for VM activity - pages are normally
207 * freed via pagevecs.  But it gets used by networking.
208 */
209void fastcall __page_cache_release(struct page *page)
210{
211	if (PageLRU(page)) {
212		unsigned long flags;
213		struct zone *zone = page_zone(page);
214
215		spin_lock_irqsave(&zone->lru_lock, flags);
216		BUG_ON(!PageLRU(page));
217		__ClearPageLRU(page);
218		del_page_from_lru(zone, page);
219		spin_unlock_irqrestore(&zone->lru_lock, flags);
220	}
221	free_hot_page(page);
222}
223EXPORT_SYMBOL(__page_cache_release);
224
225/*
226 * Batched page_cache_release().  Decrement the reference count on all the
227 * passed pages.  If it fell to zero then remove the page from the LRU and
228 * free it.
229 *
230 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
231 * for the remainder of the operation.
232 *
233 * The locking in this function is against shrink_cache(): we recheck the
234 * page count inside the lock to see whether shrink_cache grabbed the page
235 * via the LRU.  If it did, give up: shrink_cache will free it.
236 */
237void release_pages(struct page **pages, int nr, int cold)
238{
239	int i;
240	struct pagevec pages_to_free;
241	struct zone *zone = NULL;
242
243	pagevec_init(&pages_to_free, cold);
244	for (i = 0; i < nr; i++) {
245		struct page *page = pages[i];
246
247		if (unlikely(PageCompound(page))) {
248			if (zone) {
249				spin_unlock_irq(&zone->lru_lock);
250				zone = NULL;
251			}
252			put_compound_page(page);
253			continue;
254		}
255
256		if (!put_page_testzero(page))
257			continue;
258
259		if (PageLRU(page)) {
260			struct zone *pagezone = page_zone(page);
261			if (pagezone != zone) {
262				if (zone)
263					spin_unlock_irq(&zone->lru_lock);
264				zone = pagezone;
265				spin_lock_irq(&zone->lru_lock);
266			}
267			BUG_ON(!PageLRU(page));
268			__ClearPageLRU(page);
269			del_page_from_lru(zone, page);
270		}
271
272		if (!pagevec_add(&pages_to_free, page)) {
273			if (zone) {
274				spin_unlock_irq(&zone->lru_lock);
275				zone = NULL;
276			}
277			__pagevec_free(&pages_to_free);
278			pagevec_reinit(&pages_to_free);
279  		}
280	}
281	if (zone)
282		spin_unlock_irq(&zone->lru_lock);
283
284	pagevec_free(&pages_to_free);
285}
286
287/*
288 * The pages which we're about to release may be in the deferred lru-addition
289 * queues.  That would prevent them from really being freed right now.  That's
290 * OK from a correctness point of view but is inefficient - those pages may be
291 * cache-warm and we want to give them back to the page allocator ASAP.
292 *
293 * So __pagevec_release() will drain those queues here.  __pagevec_lru_add()
294 * and __pagevec_lru_add_active() call release_pages() directly to avoid
295 * mutual recursion.
296 */
297void __pagevec_release(struct pagevec *pvec)
298{
299	lru_add_drain();
300	release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
301	pagevec_reinit(pvec);
302}
303
304EXPORT_SYMBOL(__pagevec_release);
305
306/*
307 * pagevec_release() for pages which are known to not be on the LRU
308 *
309 * This function reinitialises the caller's pagevec.
310 */
311void __pagevec_release_nonlru(struct pagevec *pvec)
312{
313	int i;
314	struct pagevec pages_to_free;
315
316	pagevec_init(&pages_to_free, pvec->cold);
317	for (i = 0; i < pagevec_count(pvec); i++) {
318		struct page *page = pvec->pages[i];
319
320		BUG_ON(PageLRU(page));
321		if (put_page_testzero(page))
322			pagevec_add(&pages_to_free, page);
323	}
324	pagevec_free(&pages_to_free);
325	pagevec_reinit(pvec);
326}
327
328/*
329 * Add the passed pages to the LRU, then drop the caller's refcount
330 * on them.  Reinitialises the caller's pagevec.
331 */
332void __pagevec_lru_add(struct pagevec *pvec)
333{
334	int i;
335	struct zone *zone = NULL;
336
337	for (i = 0; i < pagevec_count(pvec); i++) {
338		struct page *page = pvec->pages[i];
339		struct zone *pagezone = page_zone(page);
340
341		if (pagezone != zone) {
342			if (zone)
343				spin_unlock_irq(&zone->lru_lock);
344			zone = pagezone;
345			spin_lock_irq(&zone->lru_lock);
346		}
347		BUG_ON(PageLRU(page));
348		SetPageLRU(page);
349		add_page_to_inactive_list(zone, page);
350	}
351	if (zone)
352		spin_unlock_irq(&zone->lru_lock);
353	release_pages(pvec->pages, pvec->nr, pvec->cold);
354	pagevec_reinit(pvec);
355}
356
357EXPORT_SYMBOL(__pagevec_lru_add);
358
359void __pagevec_lru_add_active(struct pagevec *pvec)
360{
361	int i;
362	struct zone *zone = NULL;
363
364	for (i = 0; i < pagevec_count(pvec); i++) {
365		struct page *page = pvec->pages[i];
366		struct zone *pagezone = page_zone(page);
367
368		if (pagezone != zone) {
369			if (zone)
370				spin_unlock_irq(&zone->lru_lock);
371			zone = pagezone;
372			spin_lock_irq(&zone->lru_lock);
373		}
374		BUG_ON(PageLRU(page));
375		SetPageLRU(page);
376		BUG_ON(PageActive(page));
377		SetPageActive(page);
378		add_page_to_active_list(zone, page);
379	}
380	if (zone)
381		spin_unlock_irq(&zone->lru_lock);
382	release_pages(pvec->pages, pvec->nr, pvec->cold);
383	pagevec_reinit(pvec);
384}
385
386/*
387 * Try to drop buffers from the pages in a pagevec
388 */
389void pagevec_strip(struct pagevec *pvec)
390{
391	int i;
392
393	for (i = 0; i < pagevec_count(pvec); i++) {
394		struct page *page = pvec->pages[i];
395
396		if (PagePrivate(page) && !TestSetPageLocked(page)) {
397			if (PagePrivate(page))
398				try_to_release_page(page, 0);
399			unlock_page(page);
400		}
401	}
402}
403
404/**
405 * pagevec_lookup - gang pagecache lookup
406 * @pvec:	Where the resulting pages are placed
407 * @mapping:	The address_space to search
408 * @start:	The starting page index
409 * @nr_pages:	The maximum number of pages
410 *
411 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
412 * in the mapping.  The pages are placed in @pvec.  pagevec_lookup() takes a
413 * reference against the pages in @pvec.
414 *
415 * The search returns a group of mapping-contiguous pages with ascending
416 * indexes.  There may be holes in the indices due to not-present pages.
417 *
418 * pagevec_lookup() returns the number of pages which were found.
419 */
420unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
421		pgoff_t start, unsigned nr_pages)
422{
423	pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
424	return pagevec_count(pvec);
425}
426
427EXPORT_SYMBOL(pagevec_lookup);
428
429unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
430		pgoff_t *index, int tag, unsigned nr_pages)
431{
432	pvec->nr = find_get_pages_tag(mapping, index, tag,
433					nr_pages, pvec->pages);
434	return pagevec_count(pvec);
435}
436
437EXPORT_SYMBOL(pagevec_lookup_tag);
438
439#ifdef CONFIG_SMP
440/*
441 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
442 * CPUs
443 */
444#define ACCT_THRESHOLD	max(16, NR_CPUS * 2)
445
446static DEFINE_PER_CPU(long, committed_space) = 0;
447
448void vm_acct_memory(long pages)
449{
450	long *local;
451
452	preempt_disable();
453	local = &__get_cpu_var(committed_space);
454	*local += pages;
455	if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
456		atomic_add(*local, &vm_committed_space);
457		*local = 0;
458	}
459	preempt_enable();
460}
461
462#ifdef CONFIG_HOTPLUG_CPU
463
464/* Drop the CPU's cached committed space back into the central pool. */
465static int cpu_swap_callback(struct notifier_block *nfb,
466			     unsigned long action,
467			     void *hcpu)
468{
469	long *committed;
470
471	committed = &per_cpu(committed_space, (long)hcpu);
472	if (action == CPU_DEAD) {
473		atomic_add(*committed, &vm_committed_space);
474		*committed = 0;
475		__lru_add_drain((long)hcpu);
476	}
477	return NOTIFY_OK;
478}
479#endif /* CONFIG_HOTPLUG_CPU */
480#endif /* CONFIG_SMP */
481
482/*
483 * Perform any setup for the swap system
484 */
485void __init swap_setup(void)
486{
487	unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
488
489	/* Use a smaller cluster for small-memory machines */
490	if (megs < 16)
491		page_cluster = 2;
492	else
493		page_cluster = 3;
494	/*
495	 * Right now other parts of the system means that we
496	 * _really_ don't want to cluster much more
497	 */
498	hotcpu_notifier(cpu_swap_callback, 0);
499}
500