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);
59 free_hot_cold_page(page, 0);
62 static void put_compound_page(struct page *page)
64 page = compound_head(page);
65 if (put_page_testzero(page)) {
66 compound_page_dtor *dtor;
68 dtor = get_compound_page_dtor(page);
73 void put_page(struct page *page)
75 if (unlikely(PageCompound(page)))
76 put_compound_page(page);
77 else if (put_page_testzero(page))
78 __page_cache_release(page);
80 EXPORT_SYMBOL(put_page);
83 * put_pages_list() - release a list of pages
84 * @pages: list of pages threaded on page->lru
86 * Release a list of pages which are strung together on page.lru. Currently
87 * used by read_cache_pages() and related error recovery code.
89 void put_pages_list(struct list_head *pages)
91 while (!list_empty(pages)) {
94 victim = list_entry(pages->prev, struct page, lru);
95 list_del(&victim->lru);
96 page_cache_release(victim);
99 EXPORT_SYMBOL(put_pages_list);
102 * pagevec_move_tail() must be called with IRQ disabled.
103 * Otherwise this may cause nasty races.
105 static void pagevec_move_tail(struct pagevec *pvec)
109 struct zone *zone = NULL;
111 for (i = 0; i < pagevec_count(pvec); i++) {
112 struct page *page = pvec->pages[i];
113 struct zone *pagezone = page_zone(page);
115 if (pagezone != zone) {
117 spin_unlock(&zone->lru_lock);
119 spin_lock(&zone->lru_lock);
121 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
122 int lru = page_lru_base_type(page);
123 list_move_tail(&page->lru, &zone->lru[lru].list);
128 spin_unlock(&zone->lru_lock);
129 __count_vm_events(PGROTATED, pgmoved);
130 release_pages(pvec->pages, pvec->nr, pvec->cold);
131 pagevec_reinit(pvec);
135 * Writeback is about to end against a page which has been marked for immediate
136 * reclaim. If it still appears to be reclaimable, move it to the tail of the
139 void rotate_reclaimable_page(struct page *page)
141 if (!PageLocked(page) && !PageDirty(page) && !PageActive(page) &&
142 !PageUnevictable(page) && PageLRU(page)) {
143 struct pagevec *pvec;
146 page_cache_get(page);
147 local_irq_save(flags);
148 pvec = &__get_cpu_var(lru_rotate_pvecs);
149 if (!pagevec_add(pvec, page))
150 pagevec_move_tail(pvec);
151 local_irq_restore(flags);
155 static void update_page_reclaim_stat(struct zone *zone, struct page *page,
156 int file, int rotated)
158 struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
159 struct zone_reclaim_stat *memcg_reclaim_stat;
161 memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
163 reclaim_stat->recent_scanned[file]++;
165 reclaim_stat->recent_rotated[file]++;
167 if (!memcg_reclaim_stat)
170 memcg_reclaim_stat->recent_scanned[file]++;
172 memcg_reclaim_stat->recent_rotated[file]++;
176 * FIXME: speed this up?
178 void activate_page(struct page *page)
180 struct zone *zone = page_zone(page);
182 spin_lock_irq(&zone->lru_lock);
183 if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
184 int file = page_is_file_cache(page);
185 int lru = page_lru_base_type(page);
186 del_page_from_lru_list(zone, page, lru);
190 add_page_to_lru_list(zone, page, lru);
191 __count_vm_event(PGACTIVATE);
193 update_page_reclaim_stat(zone, page, file, 1);
195 spin_unlock_irq(&zone->lru_lock);
199 * Mark a page as having seen activity.
201 * inactive,unreferenced -> inactive,referenced
202 * inactive,referenced -> active,unreferenced
203 * active,unreferenced -> active,referenced
205 void mark_page_accessed(struct page *page)
207 if (!PageActive(page) && !PageUnevictable(page) &&
208 PageReferenced(page) && PageLRU(page)) {
210 ClearPageReferenced(page);
211 } else if (!PageReferenced(page)) {
212 SetPageReferenced(page);
216 EXPORT_SYMBOL(mark_page_accessed);
218 void __lru_cache_add(struct page *page, enum lru_list lru)
220 struct pagevec *pvec = &get_cpu_var(lru_add_pvecs)[lru];
222 page_cache_get(page);
223 if (!pagevec_add(pvec, page))
224 ____pagevec_lru_add(pvec, lru);
225 put_cpu_var(lru_add_pvecs);
227 EXPORT_SYMBOL(__lru_cache_add);
230 * lru_cache_add_lru - add a page to a page list
231 * @page: the page to be added to the LRU.
232 * @lru: the LRU list to which the page is added.
234 void lru_cache_add_lru(struct page *page, enum lru_list lru)
236 if (PageActive(page)) {
237 VM_BUG_ON(PageUnevictable(page));
238 ClearPageActive(page);
239 } else if (PageUnevictable(page)) {
240 VM_BUG_ON(PageActive(page));
241 ClearPageUnevictable(page);
244 VM_BUG_ON(PageLRU(page) || PageActive(page) || PageUnevictable(page));
245 __lru_cache_add(page, lru);
249 * add_page_to_unevictable_list - add a page to the unevictable list
250 * @page: the page to be added to the unevictable list
252 * Add page directly to its zone's unevictable list. To avoid races with
253 * tasks that might be making the page evictable, through eg. munlock,
254 * munmap or exit, while it's not on the lru, we want to add the page
255 * while it's locked or otherwise "invisible" to other tasks. This is
256 * difficult to do when using the pagevec cache, so bypass that.
258 void add_page_to_unevictable_list(struct page *page)
260 struct zone *zone = page_zone(page);
262 spin_lock_irq(&zone->lru_lock);
263 SetPageUnevictable(page);
265 add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
266 spin_unlock_irq(&zone->lru_lock);
270 * Drain pages out of the cpu's pagevecs.
271 * Either "cpu" is the current CPU, and preemption has already been
272 * disabled; or "cpu" is being hot-unplugged, and is already dead.
274 static void drain_cpu_pagevecs(int cpu)
276 struct pagevec *pvecs = per_cpu(lru_add_pvecs, cpu);
277 struct pagevec *pvec;
281 pvec = &pvecs[lru - LRU_BASE];
282 if (pagevec_count(pvec))
283 ____pagevec_lru_add(pvec, lru);
286 pvec = &per_cpu(lru_rotate_pvecs, cpu);
287 if (pagevec_count(pvec)) {
290 /* No harm done if a racing interrupt already did this */
291 local_irq_save(flags);
292 pagevec_move_tail(pvec);
293 local_irq_restore(flags);
297 void lru_add_drain(void)
299 drain_cpu_pagevecs(get_cpu());
303 static void lru_add_drain_per_cpu(struct work_struct *dummy)
309 * Returns 0 for success
311 int lru_add_drain_all(void)
313 return schedule_on_each_cpu(lru_add_drain_per_cpu);
317 * Batched page_cache_release(). Decrement the reference count on all the
318 * passed pages. If it fell to zero then remove the page from the LRU and
321 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
322 * for the remainder of the operation.
324 * The locking in this function is against shrink_inactive_list(): we recheck
325 * the page count inside the lock to see whether shrink_inactive_list()
326 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
329 void release_pages(struct page **pages, int nr, int cold)
332 struct pagevec pages_to_free;
333 struct zone *zone = NULL;
334 unsigned long uninitialized_var(flags);
336 pagevec_init(&pages_to_free, cold);
337 for (i = 0; i < nr; i++) {
338 struct page *page = pages[i];
340 if (unlikely(PageCompound(page))) {
342 spin_unlock_irqrestore(&zone->lru_lock, flags);
345 put_compound_page(page);
349 if (!put_page_testzero(page))
353 struct zone *pagezone = page_zone(page);
355 if (pagezone != zone) {
357 spin_unlock_irqrestore(&zone->lru_lock,
360 spin_lock_irqsave(&zone->lru_lock, flags);
362 VM_BUG_ON(!PageLRU(page));
363 __ClearPageLRU(page);
364 del_page_from_lru(zone, page);
367 if (!pagevec_add(&pages_to_free, page)) {
369 spin_unlock_irqrestore(&zone->lru_lock, flags);
372 __pagevec_free(&pages_to_free);
373 pagevec_reinit(&pages_to_free);
377 spin_unlock_irqrestore(&zone->lru_lock, flags);
379 pagevec_free(&pages_to_free);
383 * The pages which we're about to release may be in the deferred lru-addition
384 * queues. That would prevent them from really being freed right now. That's
385 * OK from a correctness point of view but is inefficient - those pages may be
386 * cache-warm and we want to give them back to the page allocator ASAP.
388 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
389 * and __pagevec_lru_add_active() call release_pages() directly to avoid
392 void __pagevec_release(struct pagevec *pvec)
395 release_pages(pvec->pages, pagevec_count(pvec), pvec->cold);
396 pagevec_reinit(pvec);
399 EXPORT_SYMBOL(__pagevec_release);
402 * Add the passed pages to the LRU, then drop the caller's refcount
403 * on them. Reinitialises the caller's pagevec.
405 void ____pagevec_lru_add(struct pagevec *pvec, enum lru_list lru)
408 struct zone *zone = NULL;
410 VM_BUG_ON(is_unevictable_lru(lru));
412 for (i = 0; i < pagevec_count(pvec); i++) {
413 struct page *page = pvec->pages[i];
414 struct zone *pagezone = page_zone(page);
418 if (pagezone != zone) {
420 spin_unlock_irq(&zone->lru_lock);
422 spin_lock_irq(&zone->lru_lock);
424 VM_BUG_ON(PageActive(page));
425 VM_BUG_ON(PageUnevictable(page));
426 VM_BUG_ON(PageLRU(page));
428 active = is_active_lru(lru);
429 file = is_file_lru(lru);
432 update_page_reclaim_stat(zone, page, file, active);
433 add_page_to_lru_list(zone, page, lru);
436 spin_unlock_irq(&zone->lru_lock);
437 release_pages(pvec->pages, pvec->nr, pvec->cold);
438 pagevec_reinit(pvec);
441 EXPORT_SYMBOL(____pagevec_lru_add);
444 * Try to drop buffers from the pages in a pagevec
446 void pagevec_strip(struct pagevec *pvec)
450 for (i = 0; i < pagevec_count(pvec); i++) {
451 struct page *page = pvec->pages[i];
453 if (page_has_private(page) && trylock_page(page)) {
454 if (page_has_private(page))
455 try_to_release_page(page, 0);
462 * pagevec_lookup - gang pagecache lookup
463 * @pvec: Where the resulting pages are placed
464 * @mapping: The address_space to search
465 * @start: The starting page index
466 * @nr_pages: The maximum number of pages
468 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
469 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
470 * reference against the pages in @pvec.
472 * The search returns a group of mapping-contiguous pages with ascending
473 * indexes. There may be holes in the indices due to not-present pages.
475 * pagevec_lookup() returns the number of pages which were found.
477 unsigned pagevec_lookup(struct pagevec *pvec, struct address_space *mapping,
478 pgoff_t start, unsigned nr_pages)
480 pvec->nr = find_get_pages(mapping, start, nr_pages, pvec->pages);
481 return pagevec_count(pvec);
484 EXPORT_SYMBOL(pagevec_lookup);
486 unsigned pagevec_lookup_tag(struct pagevec *pvec, struct address_space *mapping,
487 pgoff_t *index, int tag, unsigned nr_pages)
489 pvec->nr = find_get_pages_tag(mapping, index, tag,
490 nr_pages, pvec->pages);
491 return pagevec_count(pvec);
494 EXPORT_SYMBOL(pagevec_lookup_tag);
497 * Perform any setup for the swap system
499 void __init swap_setup(void)
501 unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
504 bdi_init(swapper_space.backing_dev_info);
507 /* Use a smaller cluster for small-memory machines */
513 * Right now other parts of the system means that we
514 * _really_ don't want to cluster much more