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>
33 #include <linux/gfp.h>
37 /* How many pages do we try to swap or page in/out together? */
40 static DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
41 static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
44 * This path almost never happens for VM activity - pages are normally
45 * freed via pagevecs. But it gets used by networking.
47 static void __page_cache_release(struct page *page)
51 struct zone *zone = page_zone(page);
53 spin_lock_irqsave(&zone->lru_lock, flags);
54 VM_BUG_ON(!PageLRU(page));
56 del_page_from_lru(zone, page);
57 spin_unlock_irqrestore(&zone->lru_lock, flags);
61 static void __put_single_page(struct page *page)
63 __page_cache_release(page);
64 free_hot_cold_page(page, 0);
67 static void __put_compound_page(struct page *page)
69 compound_page_dtor *dtor;
71 __page_cache_release(page);
72 dtor = get_compound_page_dtor(page);
76 static void put_compound_page(struct page *page)
78 if (unlikely(PageTail(page))) {
79 /* __split_huge_page_refcount can run under us */
80 struct page *page_head = page->first_page;
83 * If PageTail is still set after smp_rmb() we can be sure
84 * that the page->first_page we read wasn't a dangling pointer.
85 * See __split_huge_page_refcount() smp_wmb().
87 if (likely(PageTail(page) && get_page_unless_zero(page_head))) {
90 * Verify that our page_head wasn't converted
91 * to a a regular page before we got a
94 if (unlikely(!PageHead(page_head))) {
95 /* PageHead is cleared after PageTail */
97 VM_BUG_ON(PageTail(page));
101 * Only run compound_lock on a valid PageHead,
102 * after having it pinned with
103 * get_page_unless_zero() above.
106 /* page_head wasn't a dangling pointer */
107 flags = compound_lock_irqsave(page_head);
108 if (unlikely(!PageTail(page))) {
109 /* __split_huge_page_refcount run before us */
110 compound_unlock_irqrestore(page_head, flags);
111 VM_BUG_ON(PageHead(page_head));
113 if (put_page_testzero(page_head))
114 __put_single_page(page_head);
116 if (put_page_testzero(page))
117 __put_single_page(page);
120 VM_BUG_ON(page_head != page->first_page);
122 * We can release the refcount taken by
123 * get_page_unless_zero now that
124 * split_huge_page_refcount is blocked on the
127 if (put_page_testzero(page_head))
129 /* __split_huge_page_refcount will wait now */
130 VM_BUG_ON(atomic_read(&page->_count) <= 0);
131 atomic_dec(&page->_count);
132 VM_BUG_ON(atomic_read(&page_head->_count) <= 0);
133 compound_unlock_irqrestore(page_head, flags);
134 if (put_page_testzero(page_head))
135 __put_compound_page(page_head);
137 /* page_head is a dangling pointer */
138 VM_BUG_ON(PageTail(page));
141 } else if (put_page_testzero(page)) {
143 __put_compound_page(page);
145 __put_single_page(page);
149 void put_page(struct page *page)
151 if (unlikely(PageCompound(page)))
152 put_compound_page(page);
153 else if (put_page_testzero(page))
154 __put_single_page(page);
156 EXPORT_SYMBOL(put_page);
159 * put_pages_list() - release a list of pages
160 * @pages: list of pages threaded on page->lru
162 * Release a list of pages which are strung together on page.lru. Currently
163 * used by read_cache_pages() and related error recovery code.
165 void put_pages_list(struct list_head *pages)
167 while (!list_empty(pages)) {
170 victim = list_entry(pages->prev, struct page, lru);
171 list_del(&victim->lru);
172 page_cache_release(victim);
175 EXPORT_SYMBOL(put_pages_list);
178 * pagevec_move_tail() must be called with IRQ disabled.
179 * Otherwise this may cause nasty races.
181 static void pagevec_move_tail(struct pagevec *pvec)
185 struct zone *zone = NULL;
187 for (i = 0; i < pagevec_count(pvec); i++) {
188 struct page *page = pvec->pages[i];
189 struct zone *pagezone = page_zone(page);
191 if (pagezone != zone) {
193 spin_unlock(&zone->lru_lock);
195 spin_lock(&zone->lru_lock);
197 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
198 int lru = page_lru_base_type(page);
199 list_move_tail(&page->lru, &zone->lru[lru].list);
204 spin_unlock(&zone->lru_lock);
205 __count_vm_events(PGROTATED, pgmoved);
206 release_pages(pvec->pages, pvec->nr, pvec->cold);
207 pagevec_reinit(pvec);
211 * Writeback is about to end against a page which has been marked for immediate
212 * reclaim. If it still appears to be reclaimable, move it to the tail of the
215 void rotate_reclaimable_page(struct page *page)
217 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
218 !PageUnevictable(page) && PageLRU(page)) {
219 struct pagevec *pvec;
222 page_cache_get(page);
223 local_irq_save(flags);
224 pvec = &__get_cpu_var(lru_rotate_pvecs);
225 if (!pagevec_add(pvec, page))
226 pagevec_move_tail(pvec);
227 local_irq_restore(flags);
231 static void update_page_reclaim_stat(struct zone *zone, struct page *page,
232 int file, int rotated)
234 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
235 struct zone_reclaim_stat *memcg_reclaim_stat;
237 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
239 reclaim_stat->recent_scanned[file]++;
241 reclaim_stat->recent_rotated[file]++;
243 if (!memcg_reclaim_stat)
246 memcg_reclaim_stat->recent_scanned[file]++;
248 memcg_reclaim_stat->recent_rotated[file]++;
252 * FIXME: speed this up?
254 void activate_page(struct page *page)
256 struct zone *zone = page_zone(page);
258 spin_lock_irq(&zone->lru_lock);
259 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
260 int file = page_is_file_cache(page);
261 int lru = page_lru_base_type(page);
262 del_page_from_lru_list(zone, page, lru);
266 add_page_to_lru_list(zone, page, lru);
267 __count_vm_event(PGACTIVATE);
269 update_page_reclaim_stat(zone, page, file, 1);
271 spin_unlock_irq(&zone->lru_lock);
275 * Mark a page as having seen activity.
277 * inactive,unreferenced -> inactive,referenced
278 * inactive,referenced -> active,unreferenced
279 * active,unreferenced -> active,referenced
281 void mark_page_accessed(struct page *page)
283 if (!PageActive(page) && !PageUnevictable(page) &&
284 PageReferenced(page) && PageLRU(page)) {
286 ClearPageReferenced(page);
287 } else if (!PageReferenced(page)) {
288 SetPageReferenced(page);
292 EXPORT_SYMBOL(mark_page_accessed);
294 void __lru_cache_add(struct page *page, enum lru_list lru)
296 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
298 page_cache_get(page);
299 if (!pagevec_add(pvec, page))
300 ____pagevec_lru_add(pvec, lru);
301 put_cpu_var(lru_add_pvecs);
303 EXPORT_SYMBOL(__lru_cache_add);
306 * lru_cache_add_lru - add a page to a page list
307 * @page: the page to be added to the LRU.
308 * @lru: the LRU list to which the page is added.
310 void lru_cache_add_lru(struct page *page, enum lru_list lru)
312 if (PageActive(page)) {
313 VM_BUG_ON(PageUnevictable(page));
314 ClearPageActive(page);
315 } else if (PageUnevictable(page)) {
316 VM_BUG_ON(PageActive(page));
317 ClearPageUnevictable(page);
320 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
321 __lru_cache_add(page, lru);
325 * add_page_to_unevictable_list - add a page to the unevictable list
326 * @page: the page to be added to the unevictable list
328 * Add page directly to its zone's unevictable list. To avoid races with
329 * tasks that might be making the page evictable, through eg. munlock,
330 * munmap or exit, while it's not on the lru, we want to add the page
331 * while it's locked or otherwise "invisible" to other tasks. This is
332 * difficult to do when using the pagevec cache, so bypass that.
334 void add_page_to_unevictable_list(struct page *page)
336 struct zone *zone = page_zone(page);
338 spin_lock_irq(&zone->lru_lock);
339 SetPageUnevictable(page);
341 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
342 spin_unlock_irq(&zone->lru_lock);
346 * Drain pages out of the cpu's pagevecs.
347 * Either "cpu" is the current CPU, and preemption has already been
348 * disabled; or "cpu" is being hot-unplugged, and is already dead.
350 static void drain_cpu_pagevecs(int cpu)
352 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
353 struct pagevec *pvec;
357 pvec = &pvecs[lru - LRU_BASE];
358 if (pagevec_count(pvec))
359 ____pagevec_lru_add(pvec, lru);
362 pvec = &per_cpu(lru_rotate_pvecs, cpu);
363 if (pagevec_count(pvec)) {
366 /* No harm done if a racing interrupt already did this */
367 local_irq_save(flags);
368 pagevec_move_tail(pvec);
369 local_irq_restore(flags);
373 void lru_add_drain(void)
375 drain_cpu_pagevecs(get_cpu());
379 static void lru_add_drain_per_cpu(struct work_struct *dummy)
385 * Returns 0 for success
387 int lru_add_drain_all(void)
389 return schedule_on_each_cpu(lru_add_drain_per_cpu);
393 * Batched page_cache_release(). Decrement the reference count on all the
394 * passed pages. If it fell to zero then remove the page from the LRU and
397 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
398 * for the remainder of the operation.
400 * The locking in this function is against shrink_inactive_list(): we recheck
401 * the page count inside the lock to see whether shrink_inactive_list()
402 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
405 void release_pages(struct page **pages, int nr, int cold)
408 struct pagevec pages_to_free;
409 struct zone *zone = NULL;
410 unsigned long uninitialized_var(flags);
412 pagevec_init(&pages_to_free, cold);
413 for (i = 0; i < nr; i++) {
414 struct page *page = pages[i];
416 if (unlikely(PageCompound(page))) {
418 spin_unlock_irqrestore(&zone->lru_lock, flags);
421 put_compound_page(page);
425 if (!put_page_testzero(page))
429 struct zone *pagezone = page_zone(page);
431 if (pagezone != zone) {
433 spin_unlock_irqrestore(&zone->lru_lock,
436 spin_lock_irqsave(&zone->lru_lock, flags);
438 VM_BUG_ON(!PageLRU(page));
439 __ClearPageLRU(page);
440 del_page_from_lru(zone, page);
443 if (!pagevec_add(&pages_to_free, page)) {
445 spin_unlock_irqrestore(&zone->lru_lock, flags);
448 __pagevec_free(&pages_to_free);
449 pagevec_reinit(&pages_to_free);
453 spin_unlock_irqrestore(&zone->lru_lock, flags);
455 pagevec_free(&pages_to_free);
457 EXPORT_SYMBOL(release_pages);
460 * The pages which we're about to release may be in the deferred lru-addition
461 * queues. That would prevent them from really being freed right now. That's
462 * OK from a correctness point of view but is inefficient - those pages may be
463 * cache-warm and we want to give them back to the page allocator ASAP.
465 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
466 * and __pagevec_lru_add_active() call release_pages() directly to avoid
469 void __pagevec_release(struct pagevec *pvec)
472 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
473 pagevec_reinit(pvec);
476 EXPORT_SYMBOL(__pagevec_release);
479 * Add the passed pages to the LRU, then drop the caller's refcount
480 * on them. Reinitialises the caller's pagevec.
482 void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
485 struct zone *zone = NULL;
487 VM_BUG_ON(is_unevictable_lru(lru));
489 for (i = 0; i < pagevec_count(pvec); i++) {
490 struct page *page = pvec->pages[i];
491 struct zone *pagezone = page_zone(page);
495 if (pagezone != zone) {
497 spin_unlock_irq(&zone->lru_lock);
499 spin_lock_irq(&zone->lru_lock);
501 VM_BUG_ON(PageActive(page));
502 VM_BUG_ON(PageUnevictable(page));
503 VM_BUG_ON(PageLRU(page));
505 active = is_active_lru(lru);
506 file = is_file_lru(lru);
509 update_page_reclaim_stat(zone, page, file, active);
510 add_page_to_lru_list(zone, page, lru);
513 spin_unlock_irq(&zone->lru_lock);
514 release_pages(pvec->pages, pvec->nr, pvec->cold);
515 pagevec_reinit(pvec);
518 EXPORT_SYMBOL(____pagevec_lru_add);
521 * Try to drop buffers from the pages in a pagevec
523 void pagevec_strip(struct pagevec *pvec)
527 for (i = 0; i < pagevec_count(pvec); i++) {
528 struct page *page = pvec->pages[i];
530 if (page_has_private(page) && trylock_page(page)) {
531 if (page_has_private(page))
532 try_to_release_page(page, 0);
539 * pagevec_lookup - gang pagecache lookup
540 * @pvec: Where the resulting pages are placed
541 * @mapping: The address_space to search
542 * @start: The starting page index
543 * @nr_pages: The maximum number of pages
545 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
546 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
547 * reference against the pages in @pvec.
549 * The search returns a group of mapping-contiguous pages with ascending
550 * indexes. There may be holes in the indices due to not-present pages.
552 * pagevec_lookup() returns the number of pages which were found.
554 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
555 pgoff_t start, unsigned nr_pages)
557 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
558 return pagevec_count(pvec);
561 EXPORT_SYMBOL(pagevec_lookup);
563 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
564 pgoff_t *index, int tag, unsigned nr_pages)
566 pvec->nr = find_get_pages_tag(mapping, index, tag,
567 nr_pages, pvec->pages);
568 return pagevec_count(pvec);
571 EXPORT_SYMBOL(pagevec_lookup_tag);
574 * Perform any setup for the swap system
576 void __init swap_setup(void)
578 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
581 bdi_init(swapper_space.backing_dev_info);
584 /* Use a smaller cluster for small-memory machines */
590 * Right now other parts of the system means that we
591 * _really_ don't want to cluster much more