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
);
42 static DEFINE_PER_CPU(struct pagevec
, lru_deactivate_pvecs
);
45 * This path almost never happens for VM activity - pages are normally
46 * freed via pagevecs. But it gets used by networking.
48 static void __page_cache_release(struct page
*page
)
52 struct zone
*zone
= page_zone(page
);
54 spin_lock_irqsave(&zone
->lru_lock
, flags
);
55 VM_BUG_ON(!PageLRU(page
));
57 del_page_from_lru(zone
, page
);
58 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
62 static void __put_single_page(struct page
*page
)
64 __page_cache_release(page
);
65 free_hot_cold_page(page
, 0);
68 static void __put_compound_page(struct page
*page
)
70 compound_page_dtor
*dtor
;
72 __page_cache_release(page
);
73 dtor
= get_compound_page_dtor(page
);
77 static void put_compound_page(struct page
*page
)
79 if (unlikely(PageTail(page
))) {
80 /* __split_huge_page_refcount can run under us */
81 struct page
*page_head
= page
->first_page
;
84 * If PageTail is still set after smp_rmb() we can be sure
85 * that the page->first_page we read wasn't a dangling pointer.
86 * See __split_huge_page_refcount() smp_wmb().
88 if (likely(PageTail(page
) && get_page_unless_zero(page_head
))) {
91 * Verify that our page_head wasn't converted
92 * to a a regular page before we got a
95 if (unlikely(!PageHead(page_head
))) {
96 /* PageHead is cleared after PageTail */
98 VM_BUG_ON(PageTail(page
));
102 * Only run compound_lock on a valid PageHead,
103 * after having it pinned with
104 * get_page_unless_zero() above.
107 /* page_head wasn't a dangling pointer */
108 flags
= compound_lock_irqsave(page_head
);
109 if (unlikely(!PageTail(page
))) {
110 /* __split_huge_page_refcount run before us */
111 compound_unlock_irqrestore(page_head
, flags
);
112 VM_BUG_ON(PageHead(page_head
));
114 if (put_page_testzero(page_head
))
115 __put_single_page(page_head
);
117 if (put_page_testzero(page
))
118 __put_single_page(page
);
121 VM_BUG_ON(page_head
!= page
->first_page
);
123 * We can release the refcount taken by
124 * get_page_unless_zero now that
125 * split_huge_page_refcount is blocked on the
128 if (put_page_testzero(page_head
))
130 /* __split_huge_page_refcount will wait now */
131 VM_BUG_ON(atomic_read(&page
->_count
) <= 0);
132 atomic_dec(&page
->_count
);
133 VM_BUG_ON(atomic_read(&page_head
->_count
) <= 0);
134 compound_unlock_irqrestore(page_head
, flags
);
135 if (put_page_testzero(page_head
)) {
136 if (PageHead(page_head
))
137 __put_compound_page(page_head
);
139 __put_single_page(page_head
);
142 /* page_head is a dangling pointer */
143 VM_BUG_ON(PageTail(page
));
146 } else if (put_page_testzero(page
)) {
148 __put_compound_page(page
);
150 __put_single_page(page
);
154 void put_page(struct page
*page
)
156 if (unlikely(PageCompound(page
)))
157 put_compound_page(page
);
158 else if (put_page_testzero(page
))
159 __put_single_page(page
);
161 EXPORT_SYMBOL(put_page
);
164 * put_pages_list() - release a list of pages
165 * @pages: list of pages threaded on page->lru
167 * Release a list of pages which are strung together on page.lru. Currently
168 * used by read_cache_pages() and related error recovery code.
170 void put_pages_list(struct list_head
*pages
)
172 while (!list_empty(pages
)) {
175 victim
= list_entry(pages
->prev
, struct page
, lru
);
176 list_del(&victim
->lru
);
177 page_cache_release(victim
);
180 EXPORT_SYMBOL(put_pages_list
);
183 * pagevec_move_tail() must be called with IRQ disabled.
184 * Otherwise this may cause nasty races.
186 static void pagevec_move_tail(struct pagevec
*pvec
)
190 struct zone
*zone
= NULL
;
192 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
193 struct page
*page
= pvec
->pages
[i
];
194 struct zone
*pagezone
= page_zone(page
);
196 if (pagezone
!= zone
) {
198 spin_unlock(&zone
->lru_lock
);
200 spin_lock(&zone
->lru_lock
);
202 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
203 enum lru_list lru
= page_lru_base_type(page
);
204 list_move_tail(&page
->lru
, &zone
->lru
[lru
].list
);
205 mem_cgroup_rotate_reclaimable_page(page
);
210 spin_unlock(&zone
->lru_lock
);
211 __count_vm_events(PGROTATED
, pgmoved
);
212 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
213 pagevec_reinit(pvec
);
217 * Writeback is about to end against a page which has been marked for immediate
218 * reclaim. If it still appears to be reclaimable, move it to the tail of the
221 void rotate_reclaimable_page(struct page
*page
)
223 if (!PageLocked(page
) && !PageDirty(page
) && !PageActive(page
) &&
224 !PageUnevictable(page
) && PageLRU(page
)) {
225 struct pagevec
*pvec
;
228 page_cache_get(page
);
229 local_irq_save(flags
);
230 pvec
= &__get_cpu_var(lru_rotate_pvecs
);
231 if (!pagevec_add(pvec
, page
))
232 pagevec_move_tail(pvec
);
233 local_irq_restore(flags
);
237 static void update_page_reclaim_stat(struct zone
*zone
, struct page
*page
,
238 int file
, int rotated
)
240 struct zone_reclaim_stat
*reclaim_stat
= &zone
->reclaim_stat
;
241 struct zone_reclaim_stat
*memcg_reclaim_stat
;
243 memcg_reclaim_stat
= mem_cgroup_get_reclaim_stat_from_page(page
);
245 reclaim_stat
->recent_scanned
[file
]++;
247 reclaim_stat
->recent_rotated
[file
]++;
249 if (!memcg_reclaim_stat
)
252 memcg_reclaim_stat
->recent_scanned
[file
]++;
254 memcg_reclaim_stat
->recent_rotated
[file
]++;
258 * FIXME: speed this up?
260 void activate_page(struct page
*page
)
262 struct zone
*zone
= page_zone(page
);
264 spin_lock_irq(&zone
->lru_lock
);
265 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
266 int file
= page_is_file_cache(page
);
267 int lru
= page_lru_base_type(page
);
268 del_page_from_lru_list(zone
, page
, lru
);
272 add_page_to_lru_list(zone
, page
, lru
);
273 __count_vm_event(PGACTIVATE
);
275 update_page_reclaim_stat(zone
, page
, file
, 1);
277 spin_unlock_irq(&zone
->lru_lock
);
281 * Mark a page as having seen activity.
283 * inactive,unreferenced -> inactive,referenced
284 * inactive,referenced -> active,unreferenced
285 * active,unreferenced -> active,referenced
287 void mark_page_accessed(struct page
*page
)
289 if (!PageActive(page
) && !PageUnevictable(page
) &&
290 PageReferenced(page
) && PageLRU(page
)) {
292 ClearPageReferenced(page
);
293 } else if (!PageReferenced(page
)) {
294 SetPageReferenced(page
);
298 EXPORT_SYMBOL(mark_page_accessed
);
300 void __lru_cache_add(struct page
*page
, enum lru_list lru
)
302 struct pagevec
*pvec
= &get_cpu_var(lru_add_pvecs
)[lru
];
304 page_cache_get(page
);
305 if (!pagevec_add(pvec
, page
))
306 ____pagevec_lru_add(pvec
, lru
);
307 put_cpu_var(lru_add_pvecs
);
309 EXPORT_SYMBOL(__lru_cache_add
);
312 * lru_cache_add_lru - add a page to a page list
313 * @page: the page to be added to the LRU.
314 * @lru: the LRU list to which the page is added.
316 void lru_cache_add_lru(struct page
*page
, enum lru_list lru
)
318 if (PageActive(page
)) {
319 VM_BUG_ON(PageUnevictable(page
));
320 ClearPageActive(page
);
321 } else if (PageUnevictable(page
)) {
322 VM_BUG_ON(PageActive(page
));
323 ClearPageUnevictable(page
);
326 VM_BUG_ON(PageLRU(page
) || PageActive(page
) || PageUnevictable(page
));
327 __lru_cache_add(page
, lru
);
331 * add_page_to_unevictable_list - add a page to the unevictable list
332 * @page: the page to be added to the unevictable list
334 * Add page directly to its zone's unevictable list. To avoid races with
335 * tasks that might be making the page evictable, through eg. munlock,
336 * munmap or exit, while it's not on the lru, we want to add the page
337 * while it's locked or otherwise "invisible" to other tasks. This is
338 * difficult to do when using the pagevec cache, so bypass that.
340 void add_page_to_unevictable_list(struct page
*page
)
342 struct zone
*zone
= page_zone(page
);
344 spin_lock_irq(&zone
->lru_lock
);
345 SetPageUnevictable(page
);
347 add_page_to_lru_list(zone
, page
, LRU_UNEVICTABLE
);
348 spin_unlock_irq(&zone
->lru_lock
);
352 * If the page can not be invalidated, it is moved to the
353 * inactive list to speed up its reclaim. It is moved to the
354 * head of the list, rather than the tail, to give the flusher
355 * threads some time to write it out, as this is much more
356 * effective than the single-page writeout from reclaim.
358 * If the page isn't page_mapped and dirty/writeback, the page
359 * could reclaim asap using PG_reclaim.
361 * 1. active, mapped page -> none
362 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
363 * 3. inactive, mapped page -> none
364 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
365 * 5. inactive, clean -> inactive, tail
368 * In 4, why it moves inactive's head, the VM expects the page would
369 * be write it out by flusher threads as this is much more effective
370 * than the single-page writeout from reclaim.
372 static void lru_deactivate(struct page
*page
, struct zone
*zone
)
380 /* Some processes are using the page */
381 if (page_mapped(page
))
384 active
= PageActive(page
);
386 file
= page_is_file_cache(page
);
387 lru
= page_lru_base_type(page
);
388 del_page_from_lru_list(zone
, page
, lru
+ active
);
389 ClearPageActive(page
);
390 ClearPageReferenced(page
);
391 add_page_to_lru_list(zone
, page
, lru
);
393 if (PageWriteback(page
) || PageDirty(page
)) {
395 * PG_reclaim could be raced with end_page_writeback
396 * It can make readahead confusing. But race window
397 * is _really_ small and it's non-critical problem.
399 SetPageReclaim(page
);
402 * The page's writeback ends up during pagevec
403 * We moves tha page into tail of inactive.
405 list_move_tail(&page
->lru
, &zone
->lru
[lru
].list
);
406 mem_cgroup_rotate_reclaimable_page(page
);
407 __count_vm_event(PGROTATED
);
411 __count_vm_event(PGDEACTIVATE
);
412 update_page_reclaim_stat(zone
, page
, file
, 0);
415 static void ____pagevec_lru_deactivate(struct pagevec
*pvec
)
418 struct zone
*zone
= NULL
;
420 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
421 struct page
*page
= pvec
->pages
[i
];
422 struct zone
*pagezone
= page_zone(page
);
424 if (pagezone
!= zone
) {
426 spin_unlock_irq(&zone
->lru_lock
);
428 spin_lock_irq(&zone
->lru_lock
);
430 lru_deactivate(page
, zone
);
433 spin_unlock_irq(&zone
->lru_lock
);
435 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
436 pagevec_reinit(pvec
);
441 * Drain pages out of the cpu's pagevecs.
442 * Either "cpu" is the current CPU, and preemption has already been
443 * disabled; or "cpu" is being hot-unplugged, and is already dead.
445 static void drain_cpu_pagevecs(int cpu
)
447 struct pagevec
*pvecs
= per_cpu(lru_add_pvecs
, cpu
);
448 struct pagevec
*pvec
;
452 pvec
= &pvecs
[lru
- LRU_BASE
];
453 if (pagevec_count(pvec
))
454 ____pagevec_lru_add(pvec
, lru
);
457 pvec
= &per_cpu(lru_rotate_pvecs
, cpu
);
458 if (pagevec_count(pvec
)) {
461 /* No harm done if a racing interrupt already did this */
462 local_irq_save(flags
);
463 pagevec_move_tail(pvec
);
464 local_irq_restore(flags
);
467 pvec
= &per_cpu(lru_deactivate_pvecs
, cpu
);
468 if (pagevec_count(pvec
))
469 ____pagevec_lru_deactivate(pvec
);
473 * deactivate_page - forcefully deactivate a page
474 * @page: page to deactivate
476 * This function hints the VM that @page is a good reclaim candidate,
477 * for example if its invalidation fails due to the page being dirty
478 * or under writeback.
480 void deactivate_page(struct page
*page
)
482 if (likely(get_page_unless_zero(page
))) {
483 struct pagevec
*pvec
= &get_cpu_var(lru_deactivate_pvecs
);
485 if (!pagevec_add(pvec
, page
))
486 ____pagevec_lru_deactivate(pvec
);
487 put_cpu_var(lru_deactivate_pvecs
);
491 void lru_add_drain(void)
493 drain_cpu_pagevecs(get_cpu());
497 static void lru_add_drain_per_cpu(struct work_struct
*dummy
)
503 * Returns 0 for success
505 int lru_add_drain_all(void)
507 return schedule_on_each_cpu(lru_add_drain_per_cpu
);
511 * Batched page_cache_release(). Decrement the reference count on all the
512 * passed pages. If it fell to zero then remove the page from the LRU and
515 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
516 * for the remainder of the operation.
518 * The locking in this function is against shrink_inactive_list(): we recheck
519 * the page count inside the lock to see whether shrink_inactive_list()
520 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
523 void release_pages(struct page
**pages
, int nr
, int cold
)
526 struct pagevec pages_to_free
;
527 struct zone
*zone
= NULL
;
528 unsigned long uninitialized_var(flags
);
530 pagevec_init(&pages_to_free
, cold
);
531 for (i
= 0; i
< nr
; i
++) {
532 struct page
*page
= pages
[i
];
534 if (unlikely(PageCompound(page
))) {
536 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
539 put_compound_page(page
);
543 if (!put_page_testzero(page
))
547 struct zone
*pagezone
= page_zone(page
);
549 if (pagezone
!= zone
) {
551 spin_unlock_irqrestore(&zone
->lru_lock
,
554 spin_lock_irqsave(&zone
->lru_lock
, flags
);
556 VM_BUG_ON(!PageLRU(page
));
557 __ClearPageLRU(page
);
558 del_page_from_lru(zone
, page
);
561 if (!pagevec_add(&pages_to_free
, page
)) {
563 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
566 __pagevec_free(&pages_to_free
);
567 pagevec_reinit(&pages_to_free
);
571 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
573 pagevec_free(&pages_to_free
);
575 EXPORT_SYMBOL(release_pages
);
578 * The pages which we're about to release may be in the deferred lru-addition
579 * queues. That would prevent them from really being freed right now. That's
580 * OK from a correctness point of view but is inefficient - those pages may be
581 * cache-warm and we want to give them back to the page allocator ASAP.
583 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
584 * and __pagevec_lru_add_active() call release_pages() directly to avoid
587 void __pagevec_release(struct pagevec
*pvec
)
590 release_pages(pvec
->pages
, pagevec_count(pvec
), pvec
->cold
);
591 pagevec_reinit(pvec
);
594 EXPORT_SYMBOL(__pagevec_release
);
596 /* used by __split_huge_page_refcount() */
597 void lru_add_page_tail(struct zone
* zone
,
598 struct page
*page
, struct page
*page_tail
)
603 struct list_head
*head
;
605 VM_BUG_ON(!PageHead(page
));
606 VM_BUG_ON(PageCompound(page_tail
));
607 VM_BUG_ON(PageLRU(page_tail
));
608 VM_BUG_ON(!spin_is_locked(&zone
->lru_lock
));
610 SetPageLRU(page_tail
);
612 if (page_evictable(page_tail
, NULL
)) {
613 if (PageActive(page
)) {
614 SetPageActive(page_tail
);
616 lru
= LRU_ACTIVE_ANON
;
619 lru
= LRU_INACTIVE_ANON
;
621 update_page_reclaim_stat(zone
, page_tail
, file
, active
);
622 if (likely(PageLRU(page
)))
623 head
= page
->lru
.prev
;
625 head
= &zone
->lru
[lru
].list
;
626 __add_page_to_lru_list(zone
, page_tail
, lru
, head
);
628 SetPageUnevictable(page_tail
);
629 add_page_to_lru_list(zone
, page_tail
, LRU_UNEVICTABLE
);
634 * Add the passed pages to the LRU, then drop the caller's refcount
635 * on them. Reinitialises the caller's pagevec.
637 void ____pagevec_lru_add(struct pagevec
*pvec
, enum lru_list lru
)
640 struct zone
*zone
= NULL
;
642 VM_BUG_ON(is_unevictable_lru(lru
));
644 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
645 struct page
*page
= pvec
->pages
[i
];
646 struct zone
*pagezone
= page_zone(page
);
650 if (pagezone
!= zone
) {
652 spin_unlock_irq(&zone
->lru_lock
);
654 spin_lock_irq(&zone
->lru_lock
);
656 VM_BUG_ON(PageActive(page
));
657 VM_BUG_ON(PageUnevictable(page
));
658 VM_BUG_ON(PageLRU(page
));
660 active
= is_active_lru(lru
);
661 file
= is_file_lru(lru
);
664 update_page_reclaim_stat(zone
, page
, file
, active
);
665 add_page_to_lru_list(zone
, page
, lru
);
668 spin_unlock_irq(&zone
->lru_lock
);
669 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
670 pagevec_reinit(pvec
);
673 EXPORT_SYMBOL(____pagevec_lru_add
);
676 * Try to drop buffers from the pages in a pagevec
678 void pagevec_strip(struct pagevec
*pvec
)
682 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
683 struct page
*page
= pvec
->pages
[i
];
685 if (page_has_private(page
) && trylock_page(page
)) {
686 if (page_has_private(page
))
687 try_to_release_page(page
, 0);
694 * pagevec_lookup - gang pagecache lookup
695 * @pvec: Where the resulting pages are placed
696 * @mapping: The address_space to search
697 * @start: The starting page index
698 * @nr_pages: The maximum number of pages
700 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
701 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
702 * reference against the pages in @pvec.
704 * The search returns a group of mapping-contiguous pages with ascending
705 * indexes. There may be holes in the indices due to not-present pages.
707 * pagevec_lookup() returns the number of pages which were found.
709 unsigned pagevec_lookup(struct pagevec
*pvec
, struct address_space
*mapping
,
710 pgoff_t start
, unsigned nr_pages
)
712 pvec
->nr
= find_get_pages(mapping
, start
, nr_pages
, pvec
->pages
);
713 return pagevec_count(pvec
);
716 EXPORT_SYMBOL(pagevec_lookup
);
718 unsigned pagevec_lookup_tag(struct pagevec
*pvec
, struct address_space
*mapping
,
719 pgoff_t
*index
, int tag
, unsigned nr_pages
)
721 pvec
->nr
= find_get_pages_tag(mapping
, index
, tag
,
722 nr_pages
, pvec
->pages
);
723 return pagevec_count(pvec
);
726 EXPORT_SYMBOL(pagevec_lookup_tag
);
729 * Perform any setup for the swap system
731 void __init
swap_setup(void)
733 unsigned long megs
= totalram_pages
>> (20 - PAGE_SHIFT
);
736 bdi_init(swapper_space
.backing_dev_info
);
739 /* Use a smaller cluster for small-memory machines */
745 * Right now other parts of the system means that we
746 * _really_ don't want to cluster much more