4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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.
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/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>
34 /* How many pages do we try to swap or page in/out together? */
38 * This path almost never happens for VM activity - pages are normally
39 * freed via pagevecs. But it gets used by networking.
41 static void fastcall __page_cache_release(struct page *page)
45 struct zone *zone = page_zone(page);
47 spin_lock_irqsave(&zone->lru_lock, flags);
48 VM_BUG_ON(!PageLRU(page));
50 del_page_from_lru(zone, page);
51 spin_unlock_irqrestore(&zone->lru_lock, flags);
56 static void put_compound_page(struct page *page)
58 page = (struct page *)page_private(page);
59 if (put_page_testzero(page)) {
60 compound_page_dtor *dtor;
62 dtor = get_compound_page_dtor(page);
67 void put_page(struct page *page)
69 if (unlikely(PageCompound(page)))
70 put_compound_page(page);
71 else if (put_page_testzero(page))
72 __page_cache_release(page);
74 EXPORT_SYMBOL(put_page);
77 * put_pages_list(): release a list of pages
79 * Release a list of pages which are strung together on page.lru. Currently
80 * used by read_cache_pages() and related error recovery code.
82 * @pages: list of pages threaded on page->lru
84 void put_pages_list(struct list_head *pages)
86 while (!list_empty(pages)) {
89 victim = list_entry(pages->prev, struct page, lru);
90 list_del(&victim->lru);
91 page_cache_release(victim);
94 EXPORT_SYMBOL(put_pages_list);
97 * Writeback is about to end against a page which has been marked for immediate
98 * reclaim. If it still appears to be reclaimable, move it to the tail of the
99 * inactive list. The page still has PageWriteback set, which will pin it.
101 * We don't expect many pages to come through here, so don't bother batching
104 * To avoid placing the page at the tail of the LRU while PG_writeback is still
105 * set, this function will clear PG_writeback before performing the page
106 * motion. Do that inside the lru lock because once PG_writeback is cleared
107 * we may not touch the page.
109 * Returns zero if it cleared PG_writeback.
111 int rotate_reclaimable_page(struct page *page)
116 if (PageLocked(page))
120 if (PageActive(page))
125 zone = page_zone(page);
126 spin_lock_irqsave(&zone->lru_lock, flags);
127 if (PageLRU(page) && !PageActive(page)) {
128 list_move_tail(&page->lru, &zone->inactive_list);
129 __count_vm_event(PGROTATED);
131 if (!test_clear_page_writeback(page))
133 spin_unlock_irqrestore(&zone->lru_lock, flags);
138 * FIXME: speed this up?
140 void fastcall activate_page(struct page *page)
142 struct zone *zone = page_zone(page);
144 spin_lock_irq(&zone->lru_lock);
145 if (PageLRU(page) && !PageActive(page)) {
146 del_page_from_inactive_list(zone, page);
148 add_page_to_active_list(zone, page);
149 __count_vm_event(PGACTIVATE);
151 spin_unlock_irq(&zone->lru_lock);
155 * Mark a page as having seen activity.
157 * inactive,unreferenced -> inactive,referenced
158 * inactive,referenced -> active,unreferenced
159 * active,unreferenced -> active,referenced
161 void fastcall mark_page_accessed(struct page *page)
163 if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
165 ClearPageReferenced(page);
166 } else if (!PageReferenced(page)) {
167 SetPageReferenced(page);
171 EXPORT_SYMBOL(mark_page_accessed);
174 * lru_cache_add: add a page to the page lists
175 * @page: the page to add
177 static DEFINE_PER_CPU(struct pagevec, lru_add_pvecs) = { 0, };
178 static DEFINE_PER_CPU(struct pagevec, lru_add_active_pvecs) = { 0, };
180 void fastcall lru_cache_add(struct page *page)
182 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs);
184 page_cache_get(page);
185 if (!pagevec_add(pvec, page))
186 __pagevec_lru_add(pvec);
187 put_cpu_var(lru_add_pvecs);
190 void fastcall lru_cache_add_active(struct page *page)
192 struct pagevec *pvec = &get_cpu_var(lru_add_active_pvecs);
194 page_cache_get(page);
195 if (!pagevec_add(pvec, page))
196 __pagevec_lru_add_active(pvec);
197 put_cpu_var(lru_add_active_pvecs);
200 static void __lru_add_drain(int cpu)
202 struct pagevec *pvec = &per_cpu(lru_add_pvecs, cpu);
204 /* CPU is dead, so no locking needed. */
205 if (pagevec_count(pvec))
206 __pagevec_lru_add(pvec);
207 pvec = &per_cpu(lru_add_active_pvecs, cpu);
208 if (pagevec_count(pvec))
209 __pagevec_lru_add_active(pvec);
212 void lru_add_drain(void)
214 __lru_add_drain(get_cpu());
219 static void lru_add_drain_per_cpu(struct work_struct *dummy)
225 * Returns 0 for success
227 int lru_add_drain_all(void)
229 return schedule_on_each_cpu(lru_add_drain_per_cpu);
235 * Returns 0 for success
237 int lru_add_drain_all(void)
245 * Batched page_cache_release(). Decrement the reference count on all the
246 * passed pages. If it fell to zero then remove the page from the LRU and
249 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
250 * for the remainder of the operation.
252 * The locking in this function is against shrink_cache(): we recheck the
253 * page count inside the lock to see whether shrink_cache grabbed the page
254 * via the LRU. If it did, give up: shrink_cache will free it.
256 void release_pages(struct page **pages, int nr, int cold)
259 struct pagevec pages_to_free;
260 struct zone *zone = NULL;
262 pagevec_init(&pages_to_free, cold);
263 for (i = 0; i < nr; i++) {
264 struct page *page = pages[i];
266 if (unlikely(PageCompound(page))) {
268 spin_unlock_irq(&zone->lru_lock);
271 put_compound_page(page);
275 if (!put_page_testzero(page))
279 struct zone *pagezone = page_zone(page);
280 if (pagezone != zone) {
282 spin_unlock_irq(&zone->lru_lock);
284 spin_lock_irq(&zone->lru_lock);
286 VM_BUG_ON(!PageLRU(page));
287 __ClearPageLRU(page);
288 del_page_from_lru(zone, page);
291 if (!pagevec_add(&pages_to_free, page)) {
293 spin_unlock_irq(&zone->lru_lock);
296 __pagevec_free(&pages_to_free);
297 pagevec_reinit(&pages_to_free);
301 spin_unlock_irq(&zone->lru_lock);
303 pagevec_free(&pages_to_free);
307 * The pages which we're about to release may be in the deferred lru-addition
308 * queues. That would prevent them from really being freed right now. That's
309 * OK from a correctness point of view but is inefficient - those pages may be
310 * cache-warm and we want to give them back to the page allocator ASAP.
312 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
313 * and __pagevec_lru_add_active() call release_pages() directly to avoid
316 void __pagevec_release(struct pagevec *pvec)
319 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
320 pagevec_reinit(pvec);
323 EXPORT_SYMBOL(__pagevec_release);
326 * pagevec_release() for pages which are known to not be on the LRU
328 * This function reinitialises the caller's pagevec.
330 void __pagevec_release_nonlru(struct pagevec *pvec)
333 struct pagevec pages_to_free;
335 pagevec_init(&pages_to_free, pvec->cold);
336 for (i = 0; i < pagevec_count(pvec); i++) {
337 struct page *page = pvec->pages[i];
339 VM_BUG_ON(PageLRU(page));
340 if (put_page_testzero(page))
341 pagevec_add(&pages_to_free, page);
343 pagevec_free(&pages_to_free);
344 pagevec_reinit(pvec);
348 * Add the passed pages to the LRU, then drop the caller's refcount
349 * on them. Reinitialises the caller's pagevec.
351 void __pagevec_lru_add(struct pagevec *pvec)
354 struct zone *zone = NULL;
356 for (i = 0; i < pagevec_count(pvec); i++) {
357 struct page *page = pvec->pages[i];
358 struct zone *pagezone = page_zone(page);
360 if (pagezone != zone) {
362 spin_unlock_irq(&zone->lru_lock);
364 spin_lock_irq(&zone->lru_lock);
366 VM_BUG_ON(PageLRU(page));
368 add_page_to_inactive_list(zone, page);
371 spin_unlock_irq(&zone->lru_lock);
372 release_pages(pvec->pages, pvec->nr, pvec->cold);
373 pagevec_reinit(pvec);
376 EXPORT_SYMBOL(__pagevec_lru_add);
378 void __pagevec_lru_add_active(struct pagevec *pvec)
381 struct zone *zone = NULL;
383 for (i = 0; i < pagevec_count(pvec); i++) {
384 struct page *page = pvec->pages[i];
385 struct zone *pagezone = page_zone(page);
387 if (pagezone != zone) {
389 spin_unlock_irq(&zone->lru_lock);
391 spin_lock_irq(&zone->lru_lock);
393 VM_BUG_ON(PageLRU(page));
395 VM_BUG_ON(PageActive(page));
397 add_page_to_active_list(zone, page);
400 spin_unlock_irq(&zone->lru_lock);
401 release_pages(pvec->pages, pvec->nr, pvec->cold);
402 pagevec_reinit(pvec);
406 * Try to drop buffers from the pages in a pagevec
408 void pagevec_strip(struct pagevec *pvec)
412 for (i = 0; i < pagevec_count(pvec); i++) {
413 struct page *page = pvec->pages[i];
415 if (PagePrivate(page) && !TestSetPageLocked(page)) {
416 if (PagePrivate(page))
417 try_to_release_page(page, 0);
424 * pagevec_lookup - gang pagecache lookup
425 * @pvec: Where the resulting pages are placed
426 * @mapping: The address_space to search
427 * @start: The starting page index
428 * @nr_pages: The maximum number of pages
430 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
431 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
432 * reference against the pages in @pvec.
434 * The search returns a group of mapping-contiguous pages with ascending
435 * indexes. There may be holes in the indices due to not-present pages.
437 * pagevec_lookup() returns the number of pages which were found.
439 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
440 pgoff_t start, unsigned nr_pages)
442 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
443 return pagevec_count(pvec);
446 EXPORT_SYMBOL(pagevec_lookup);
448 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
449 pgoff_t *index, int tag, unsigned nr_pages)
451 pvec->nr = find_get_pages_tag(mapping, index, tag,
452 nr_pages, pvec->pages);
453 return pagevec_count(pvec);
456 EXPORT_SYMBOL(pagevec_lookup_tag);
460 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
463 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
465 static DEFINE_PER_CPU(long, committed_space) = 0;
467 void vm_acct_memory(long pages)
472 local = &__get_cpu_var(committed_space);
474 if (*local > ACCT_THRESHOLD || *local < -ACCT_THRESHOLD) {
475 atomic_add(*local, &vm_committed_space);
481 #ifdef CONFIG_HOTPLUG_CPU
483 /* Drop the CPU's cached committed space back into the central pool. */
484 static int cpu_swap_callback(struct notifier_block *nfb,
485 unsigned long action,
490 committed = &per_cpu(committed_space, (long)hcpu);
491 if (action == CPU_DEAD) {
492 atomic_add(*committed, &vm_committed_space);
494 __lru_add_drain((long)hcpu);
498 #endif /* CONFIG_HOTPLUG_CPU */
499 #endif /* CONFIG_SMP */
502 * Perform any setup for the swap system
504 void __init swap_setup(void)
506 unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);
508 /* Use a smaller cluster for small-memory machines */
514 * Right now other parts of the system means that we
515 * _really_ don't want to cluster much more
517 #ifdef CONFIG_HOTPLUG_CPU
518 hotcpu_notifier(cpu_swap_callback, 0);