4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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.
12 * Swap aging added 23.2.95, Stephen Tweedie.
13 * Buffermem limits added 12.3.98, Rik van Riel.
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/percpu_counter.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>
34 /* How many pages do we try to swap or page in/out together? */
37 static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
38 static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
39 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs) = { 0, };
42 * This path almost never happens for VM activity - pages are normally
43 * freed via pagevecs. But it gets used by networking.
45 static void __page_cache_release(struct page *page)
49 struct zone *zone = page_zone(page);
51 spin_lock_irqsave(&zone->lru_lock, flags);
52 VM_BUG_ON(!PageLRU(page));
54 del_page_from_lru(zone, page);
55 spin_unlock_irqrestore(&zone->lru_lock, flags);
60 static void put_compound_page(struct page *page)
62 page = compound_head(page);
63 if (put_page_testzero(page)) {
64 compound_page_dtor *dtor;
66 dtor = get_compound_page_dtor(page);
71 void put_page(struct page *page)
73 if (unlikely(PageCompound(page)))
74 put_compound_page(page);
75 else if (put_page_testzero(page))
76 __page_cache_release(page);
78 EXPORT_SYMBOL(put_page);
81 * put_pages_list() - release a list of pages
82 * @pages: list of pages threaded on page->lru
84 * Release a list of pages which are strung together on page.lru. Currently
85 * used by read_cache_pages() and related error recovery code.
87 void put_pages_list(struct list_head *pages)
89 while (!list_empty(pages)) {
92 victim = list_entry(pages->prev, struct page, lru);
93 list_del(&victim->lru);
94 page_cache_release(victim);
97 EXPORT_SYMBOL(put_pages_list);
100 * pagevec_move_tail() must be called with IRQ disabled.
101 * Otherwise this may cause nasty races.
103 static void pagevec_move_tail(struct pagevec *pvec)
107 struct zone *zone = NULL;
109 for (i = 0; i < pagevec_count(pvec); i++) {
110 struct page *page = pvec->pages[i];
111 struct zone *pagezone = page_zone(page);
113 if (pagezone != zone) {
115 spin_unlock(&zone->lru_lock);
117 spin_lock(&zone->lru_lock);
119 if (PageLRU(page) && !PageActive(page)) {
120 list_move_tail(&page->lru, &zone->inactive_list);
125 spin_unlock(&zone->lru_lock);
126 __count_vm_events(PGROTATED, pgmoved);
127 release_pages(pvec->pages, pvec->nr, pvec->cold);
128 pagevec_reinit(pvec);
132 * Writeback is about to end against a page which has been marked for immediate
133 * reclaim. If it still appears to be reclaimable, move it to the tail of the
136 * Returns zero if it cleared PG_writeback.
138 int rotate_reclaimable_page(struct page *page)
140 struct pagevec *pvec;
143 if (PageLocked(page))
147 if (PageActive(page))
152 page_cache_get(page);
153 local_irq_save(flags);
154 pvec = &__get_cpu_var(lru_rotate_pvecs);
155 if (!pagevec_add(pvec, page))
156 pagevec_move_tail(pvec);
157 local_irq_restore(flags);
159 if (!test_clear_page_writeback(page))
166 * FIXME: speed this up?
168 void activate_page(struct page *page)
170 struct zone *zone = page_zone(page);
172 spin_lock_irq(&zone->lru_lock);
173 if (PageLRU(page) && !PageActive(page)) {
174 del_page_from_inactive_list(zone, page);
176 add_page_to_active_list(zone, page);
177 __count_vm_event(PGACTIVATE);
178 mem_cgroup_move_lists(page, true);
180 spin_unlock_irq(&zone->lru_lock);
184 * Mark a page as having seen activity.
186 * inactive,unreferenced -> inactive,referenced
187 * inactive,referenced -> active,unreferenced
188 * active,unreferenced -> active,referenced
190 void mark_page_accessed(struct page *page)
192 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
194 ClearPageReferenced(page);
195 } else if (!PageReferenced(page)) {
196 SetPageReferenced(page);
200 EXPORT_SYMBOL(mark_page_accessed);
203 * lru_cache_add: add a page to the page lists
204 * @page: the page to add
206 void lru_cache_add(struct page *page)
208 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
210 page_cache_get(page);
211 if (!pagevec_add(pvec, page))
212 __pagevec_lru_add(pvec);
213 put_cpu_var(lru_add_pvecs);
216 void lru_cache_add_active(struct page *page)
218 struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
220 page_cache_get(page);
221 if (!pagevec_add(pvec, page))
222 __pagevec_lru_add_active(pvec);
223 put_cpu_var(lru_add_active_pvecs);
227 * Drain pages out of the cpu's pagevecs.
228 * Either "cpu" is the current CPU, and preemption has already been
229 * disabled; or "cpu" is being hot-unplugged, and is already dead.
231 static void drain_cpu_pagevecs(int cpu)
233 struct pagevec *pvec;
235 pvec = &per_cpu(lru_add_pvecs, cpu);
236 if (pagevec_count(pvec))
237 __pagevec_lru_add(pvec);
239 pvec = &per_cpu(lru_add_active_pvecs, cpu);
240 if (pagevec_count(pvec))
241 __pagevec_lru_add_active(pvec);
243 pvec = &per_cpu(lru_rotate_pvecs, cpu);
244 if (pagevec_count(pvec)) {
247 /* No harm done if a racing interrupt already did this */
248 local_irq_save(flags);
249 pagevec_move_tail(pvec);
250 local_irq_restore(flags);
254 void lru_add_drain(void)
256 drain_cpu_pagevecs(get_cpu());
261 static void lru_add_drain_per_cpu(struct work_struct *dummy)
267 * Returns 0 for success
269 int lru_add_drain_all(void)
271 return schedule_on_each_cpu(lru_add_drain_per_cpu);
277 * Returns 0 for success
279 int lru_add_drain_all(void)
287 * Batched page_cache_release(). Decrement the reference count on all the
288 * passed pages. If it fell to zero then remove the page from the LRU and
291 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
292 * for the remainder of the operation.
294 * The locking in this function is against shrink_cache(): we recheck the
295 * page count inside the lock to see whether shrink_cache grabbed the page
296 * via the LRU. If it did, give up: shrink_cache will free it.
298 void release_pages(struct page **pages, int nr, int cold)
301 struct pagevec pages_to_free;
302 struct zone *zone = NULL;
303 unsigned long uninitialized_var(flags);
305 pagevec_init(&pages_to_free, cold);
306 for (i = 0; i < nr; i++) {
307 struct page *page = pages[i];
309 if (unlikely(PageCompound(page))) {
311 spin_unlock_irqrestore(&zone->lru_lock, flags);
314 put_compound_page(page);
318 if (!put_page_testzero(page))
322 struct zone *pagezone = page_zone(page);
323 if (pagezone != zone) {
325 spin_unlock_irqrestore(&zone->lru_lock,
328 spin_lock_irqsave(&zone->lru_lock, flags);
330 VM_BUG_ON(!PageLRU(page));
331 __ClearPageLRU(page);
332 del_page_from_lru(zone, page);
335 if (!pagevec_add(&pages_to_free, page)) {
337 spin_unlock_irqrestore(&zone->lru_lock, flags);
340 __pagevec_free(&pages_to_free);
341 pagevec_reinit(&pages_to_free);
345 spin_unlock_irqrestore(&zone->lru_lock, flags);
347 pagevec_free(&pages_to_free);
351 * The pages which we're about to release may be in the deferred lru-addition
352 * queues. That would prevent them from really being freed right now. That's
353 * OK from a correctness point of view but is inefficient - those pages may be
354 * cache-warm and we want to give them back to the page allocator ASAP.
356 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
357 * and __pagevec_lru_add_active() call release_pages() directly to avoid
360 void __pagevec_release(struct pagevec *pvec)
363 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
364 pagevec_reinit(pvec);
367 EXPORT_SYMBOL(__pagevec_release);
370 * pagevec_release() for pages which are known to not be on the LRU
372 * This function reinitialises the caller's pagevec.
374 void __pagevec_release_nonlru(struct pagevec *pvec)
377 struct pagevec pages_to_free;
379 pagevec_init(&pages_to_free, pvec->cold);
380 for (i = 0; i < pagevec_count(pvec); i++) {
381 struct page *page = pvec->pages[i];
383 VM_BUG_ON(PageLRU(page));
384 if (put_page_testzero(page))
385 pagevec_add(&pages_to_free, page);
387 pagevec_free(&pages_to_free);
388 pagevec_reinit(pvec);
392 * Add the passed pages to the LRU, then drop the caller's refcount
393 * on them. Reinitialises the caller's pagevec.
395 void __pagevec_lru_add(struct pagevec *pvec)
398 struct zone *zone = NULL;
400 for (i = 0; i < pagevec_count(pvec); i++) {
401 struct page *page = pvec->pages[i];
402 struct zone *pagezone = page_zone(page);
404 if (pagezone != zone) {
406 spin_unlock_irq(&zone->lru_lock);
408 spin_lock_irq(&zone->lru_lock);
410 VM_BUG_ON(PageLRU(page));
412 add_page_to_inactive_list(zone, page);
415 spin_unlock_irq(&zone->lru_lock);
416 release_pages(pvec->pages, pvec->nr, pvec->cold);
417 pagevec_reinit(pvec);
420 EXPORT_SYMBOL(__pagevec_lru_add);
422 void __pagevec_lru_add_active(struct pagevec *pvec)
425 struct zone *zone = NULL;
427 for (i = 0; i < pagevec_count(pvec); i++) {
428 struct page *page = pvec->pages[i];
429 struct zone *pagezone = page_zone(page);
431 if (pagezone != zone) {
433 spin_unlock_irq(&zone->lru_lock);
435 spin_lock_irq(&zone->lru_lock);
437 VM_BUG_ON(PageLRU(page));
439 VM_BUG_ON(PageActive(page));
441 add_page_to_active_list(zone, page);
444 spin_unlock_irq(&zone->lru_lock);
445 release_pages(pvec->pages, pvec->nr, pvec->cold);
446 pagevec_reinit(pvec);
450 * Try to drop buffers from the pages in a pagevec
452 void pagevec_strip(struct pagevec *pvec)
456 for (i = 0; i < pagevec_count(pvec); i++) {
457 struct page *page = pvec->pages[i];
459 if (PagePrivate(page) && !TestSetPageLocked(page)) {
460 if (PagePrivate(page))
461 try_to_release_page(page, 0);
468 * pagevec_lookup - gang pagecache lookup
469 * @pvec: Where the resulting pages are placed
470 * @mapping: The address_space to search
471 * @start: The starting page index
472 * @nr_pages: The maximum number of pages
474 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
475 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
476 * reference against the pages in @pvec.
478 * The search returns a group of mapping-contiguous pages with ascending
479 * indexes. There may be holes in the indices due to not-present pages.
481 * pagevec_lookup() returns the number of pages which were found.
483 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
484 pgoff_t start, unsigned nr_pages)
486 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
487 return pagevec_count(pvec);
490 EXPORT_SYMBOL(pagevec_lookup);
492 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
493 pgoff_t *index, int tag, unsigned nr_pages)
495 pvec->nr = find_get_pages_tag(mapping, index, tag,
496 nr_pages, pvec->pages);
497 return pagevec_count(pvec);
500 EXPORT_SYMBOL(pagevec_lookup_tag);
504 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
507 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
509 static DEFINE_PER_CPU(long, committed_space) = 0;
511 void vm_acct_memory(long pages)
516 local = &__get_cpu_var(committed_space);
518 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
519 atomic_add(*local, &vm_committed_space);
525 #ifdef CONFIG_HOTPLUG_CPU
527 /* Drop the CPU's cached committed space back into the central pool. */
528 static int cpu_swap_callback(struct notifier_block *nfb,
529 unsigned long action,
534 committed = &per_cpu(committed_space, (long)hcpu);
535 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
536 atomic_add(*committed, &vm_committed_space);
538 drain_cpu_pagevecs((long)hcpu);
542 #endif /* CONFIG_HOTPLUG_CPU */
543 #endif /* CONFIG_SMP */
546 * Perform any setup for the swap system
548 void __init swap_setup(void)
550 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
553 bdi_init(swapper_space.backing_dev_info);
556 /* Use a smaller cluster for small-memory machines */
562 * Right now other parts of the system means that we
563 * _really_ don't want to cluster much more
565 #ifdef CONFIG_HOTPLUG_CPU
566 hotcpu_notifier(cpu_swap_callback, 0);