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>
33 /* How many pages do we try to swap or page in/out together? */
36 static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
37 static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
38 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs) = { 0, };
41 * This path almost never happens for VM activity - pages are normally
42 * freed via pagevecs. But it gets used by networking.
44 static void fastcall __page_cache_release(struct page *page)
48 struct zone *zone = page_zone(page);
50 spin_lock_irqsave(&zone->lru_lock, flags);
51 VM_BUG_ON(!PageLRU(page));
53 del_page_from_lru(zone, page);
54 spin_unlock_irqrestore(&zone->lru_lock, flags);
59 static void put_compound_page(struct page *page)
61 page = compound_head(page);
62 if (put_page_testzero(page)) {
63 compound_page_dtor *dtor;
65 dtor = get_compound_page_dtor(page);
70 void put_page(struct page *page)
72 if (unlikely(PageCompound(page)))
73 put_compound_page(page);
74 else if (put_page_testzero(page))
75 __page_cache_release(page);
77 EXPORT_SYMBOL(put_page);
80 * put_pages_list(): release a list of pages
82 * Release a list of pages which are strung together on page.lru. Currently
83 * used by read_cache_pages() and related error recovery code.
85 * @pages: list of pages threaded on page->lru
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 fastcall 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);
179 spin_unlock_irq(&zone->lru_lock);
183 * Mark a page as having seen activity.
185 * inactive,unreferenced -> inactive,referenced
186 * inactive,referenced -> active,unreferenced
187 * active,unreferenced -> active,referenced
189 void fastcall mark_page_accessed(struct page *page)
191 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
193 ClearPageReferenced(page);
194 } else if (!PageReferenced(page)) {
195 SetPageReferenced(page);
199 EXPORT_SYMBOL(mark_page_accessed);
202 * lru_cache_add: add a page to the page lists
203 * @page: the page to add
205 void fastcall lru_cache_add(struct page *page)
207 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
209 page_cache_get(page);
210 if (!pagevec_add(pvec, page))
211 __pagevec_lru_add(pvec);
212 put_cpu_var(lru_add_pvecs);
215 void fastcall lru_cache_add_active(struct page *page)
217 struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
219 page_cache_get(page);
220 if (!pagevec_add(pvec, page))
221 __pagevec_lru_add_active(pvec);
222 put_cpu_var(lru_add_active_pvecs);
226 * Drain pages out of the cpu's pagevecs.
227 * Either "cpu" is the current CPU, and preemption has already been
228 * disabled; or "cpu" is being hot-unplugged, and is already dead.
230 static void drain_cpu_pagevecs(int cpu)
232 struct pagevec *pvec;
234 pvec = &per_cpu(lru_add_pvecs, cpu);
235 if (pagevec_count(pvec))
236 __pagevec_lru_add(pvec);
238 pvec = &per_cpu(lru_add_active_pvecs, cpu);
239 if (pagevec_count(pvec))
240 __pagevec_lru_add_active(pvec);
242 pvec = &per_cpu(lru_rotate_pvecs, cpu);
243 if (pagevec_count(pvec)) {
246 /* No harm done if a racing interrupt already did this */
247 local_irq_save(flags);
248 pagevec_move_tail(pvec);
249 local_irq_restore(flags);
253 void lru_add_drain(void)
255 drain_cpu_pagevecs(get_cpu());
260 static void lru_add_drain_per_cpu(struct work_struct *dummy)
266 * Returns 0 for success
268 int lru_add_drain_all(void)
270 return schedule_on_each_cpu(lru_add_drain_per_cpu);
276 * Returns 0 for success
278 int lru_add_drain_all(void)
286 * Batched page_cache_release(). Decrement the reference count on all the
287 * passed pages. If it fell to zero then remove the page from the LRU and
290 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
291 * for the remainder of the operation.
293 * The locking in this function is against shrink_cache(): we recheck the
294 * page count inside the lock to see whether shrink_cache grabbed the page
295 * via the LRU. If it did, give up: shrink_cache will free it.
297 void release_pages(struct page **pages, int nr, int cold)
300 struct pagevec pages_to_free;
301 struct zone *zone = NULL;
302 unsigned long uninitialized_var(flags);
304 pagevec_init(&pages_to_free, cold);
305 for (i = 0; i < nr; i++) {
306 struct page *page = pages[i];
308 if (unlikely(PageCompound(page))) {
310 spin_unlock_irqrestore(&zone->lru_lock, flags);
313 put_compound_page(page);
317 if (!put_page_testzero(page))
321 struct zone *pagezone = page_zone(page);
322 if (pagezone != zone) {
324 spin_unlock_irqrestore(&zone->lru_lock,
327 spin_lock_irqsave(&zone->lru_lock, flags);
329 VM_BUG_ON(!PageLRU(page));
330 __ClearPageLRU(page);
331 del_page_from_lru(zone, page);
334 if (!pagevec_add(&pages_to_free, page)) {
336 spin_unlock_irqrestore(&zone->lru_lock, flags);
339 __pagevec_free(&pages_to_free);
340 pagevec_reinit(&pages_to_free);
344 spin_unlock_irqrestore(&zone->lru_lock, flags);
346 pagevec_free(&pages_to_free);
350 * The pages which we're about to release may be in the deferred lru-addition
351 * queues. That would prevent them from really being freed right now. That's
352 * OK from a correctness point of view but is inefficient - those pages may be
353 * cache-warm and we want to give them back to the page allocator ASAP.
355 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
356 * and __pagevec_lru_add_active() call release_pages() directly to avoid
359 void __pagevec_release(struct pagevec *pvec)
362 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
363 pagevec_reinit(pvec);
366 EXPORT_SYMBOL(__pagevec_release);
369 * pagevec_release() for pages which are known to not be on the LRU
371 * This function reinitialises the caller's pagevec.
373 void __pagevec_release_nonlru(struct pagevec *pvec)
376 struct pagevec pages_to_free;
378 pagevec_init(&pages_to_free, pvec->cold);
379 for (i = 0; i < pagevec_count(pvec); i++) {
380 struct page *page = pvec->pages[i];
382 VM_BUG_ON(PageLRU(page));
383 if (put_page_testzero(page))
384 pagevec_add(&pages_to_free, page);
386 pagevec_free(&pages_to_free);
387 pagevec_reinit(pvec);
391 * Add the passed pages to the LRU, then drop the caller's refcount
392 * on them. Reinitialises the caller's pagevec.
394 void __pagevec_lru_add(struct pagevec *pvec)
397 struct zone *zone = NULL;
399 for (i = 0; i < pagevec_count(pvec); i++) {
400 struct page *page = pvec->pages[i];
401 struct zone *pagezone = page_zone(page);
403 if (pagezone != zone) {
405 spin_unlock_irq(&zone->lru_lock);
407 spin_lock_irq(&zone->lru_lock);
409 VM_BUG_ON(PageLRU(page));
411 add_page_to_inactive_list(zone, page);
414 spin_unlock_irq(&zone->lru_lock);
415 release_pages(pvec->pages, pvec->nr, pvec->cold);
416 pagevec_reinit(pvec);
419 EXPORT_SYMBOL(__pagevec_lru_add);
421 void __pagevec_lru_add_active(struct pagevec *pvec)
424 struct zone *zone = NULL;
426 for (i = 0; i < pagevec_count(pvec); i++) {
427 struct page *page = pvec->pages[i];
428 struct zone *pagezone = page_zone(page);
430 if (pagezone != zone) {
432 spin_unlock_irq(&zone->lru_lock);
434 spin_lock_irq(&zone->lru_lock);
436 VM_BUG_ON(PageLRU(page));
438 VM_BUG_ON(PageActive(page));
440 add_page_to_active_list(zone, page);
443 spin_unlock_irq(&zone->lru_lock);
444 release_pages(pvec->pages, pvec->nr, pvec->cold);
445 pagevec_reinit(pvec);
449 * Try to drop buffers from the pages in a pagevec
451 void pagevec_strip(struct pagevec *pvec)
455 for (i = 0; i < pagevec_count(pvec); i++) {
456 struct page *page = pvec->pages[i];
458 if (PagePrivate(page) && !TestSetPageLocked(page)) {
459 if (PagePrivate(page))
460 try_to_release_page(page, 0);
467 * pagevec_lookup - gang pagecache lookup
468 * @pvec: Where the resulting pages are placed
469 * @mapping: The address_space to search
470 * @start: The starting page index
471 * @nr_pages: The maximum number of pages
473 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
474 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
475 * reference against the pages in @pvec.
477 * The search returns a group of mapping-contiguous pages with ascending
478 * indexes. There may be holes in the indices due to not-present pages.
480 * pagevec_lookup() returns the number of pages which were found.
482 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
483 pgoff_t start, unsigned nr_pages)
485 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
486 return pagevec_count(pvec);
489 EXPORT_SYMBOL(pagevec_lookup);
491 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
492 pgoff_t *index, int tag, unsigned nr_pages)
494 pvec->nr = find_get_pages_tag(mapping, index, tag,
495 nr_pages, pvec->pages);
496 return pagevec_count(pvec);
499 EXPORT_SYMBOL(pagevec_lookup_tag);
503 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
506 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
508 static DEFINE_PER_CPU(long, committed_space) = 0;
510 void vm_acct_memory(long pages)
515 local = &__get_cpu_var(committed_space);
517 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
518 atomic_add(*local, &vm_committed_space);
524 #ifdef CONFIG_HOTPLUG_CPU
526 /* Drop the CPU's cached committed space back into the central pool. */
527 static int cpu_swap_callback(struct notifier_block *nfb,
528 unsigned long action,
533 committed = &per_cpu(committed_space, (long)hcpu);
534 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
535 atomic_add(*committed, &vm_committed_space);
537 drain_cpu_pagevecs((long)hcpu);
541 #endif /* CONFIG_HOTPLUG_CPU */
542 #endif /* CONFIG_SMP */
545 * Perform any setup for the swap system
547 void __init swap_setup(void)
549 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
552 bdi_init(swapper_space.backing_dev_info);
555 /* Use a smaller cluster for small-memory machines */
561 * Right now other parts of the system means that we
562 * _really_ don't want to cluster much more
564 #ifdef CONFIG_HOTPLUG_CPU
565 hotcpu_notifier(cpu_swap_callback, 0);