]>
git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - mm/swap.c
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
);
182 static void pagevec_lru_move_fn(struct pagevec
*pvec
,
183 void (*move_fn
)(struct page
*page
, void *arg
),
187 struct zone
*zone
= NULL
;
188 unsigned long flags
= 0;
190 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
191 struct page
*page
= pvec
->pages
[i
];
192 struct zone
*pagezone
= page_zone(page
);
194 if (pagezone
!= zone
) {
196 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
198 spin_lock_irqsave(&zone
->lru_lock
, flags
);
201 (*move_fn
)(page
, arg
);
204 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
205 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
206 pagevec_reinit(pvec
);
209 static void pagevec_move_tail_fn(struct page
*page
, void *arg
)
212 struct zone
*zone
= page_zone(page
);
214 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
215 enum lru_list lru
= page_lru_base_type(page
);
216 list_move_tail(&page
->lru
, &zone
->lru
[lru
].list
);
217 mem_cgroup_rotate_reclaimable_page(page
);
223 * pagevec_move_tail() must be called with IRQ disabled.
224 * Otherwise this may cause nasty races.
226 static void pagevec_move_tail(struct pagevec
*pvec
)
230 pagevec_lru_move_fn(pvec
, pagevec_move_tail_fn
, &pgmoved
);
231 __count_vm_events(PGROTATED
, pgmoved
);
235 * Writeback is about to end against a page which has been marked for immediate
236 * reclaim. If it still appears to be reclaimable, move it to the tail of the
239 void rotate_reclaimable_page(struct page
*page
)
241 if (!PageLocked(page
) && !PageDirty(page
) && !PageActive(page
) &&
242 !PageUnevictable(page
) && PageLRU(page
)) {
243 struct pagevec
*pvec
;
246 page_cache_get(page
);
247 local_irq_save(flags
);
248 pvec
= &__get_cpu_var(lru_rotate_pvecs
);
249 if (!pagevec_add(pvec
, page
))
250 pagevec_move_tail(pvec
);
251 local_irq_restore(flags
);
255 static void update_page_reclaim_stat(struct zone
*zone
, struct page
*page
,
256 int file
, int rotated
)
258 struct zone_reclaim_stat
*reclaim_stat
= &zone
->reclaim_stat
;
259 struct zone_reclaim_stat
*memcg_reclaim_stat
;
261 memcg_reclaim_stat
= mem_cgroup_get_reclaim_stat_from_page(page
);
263 reclaim_stat
->recent_scanned
[file
]++;
265 reclaim_stat
->recent_rotated
[file
]++;
267 if (!memcg_reclaim_stat
)
270 memcg_reclaim_stat
->recent_scanned
[file
]++;
272 memcg_reclaim_stat
->recent_rotated
[file
]++;
275 static void __activate_page(struct page
*page
, void *arg
)
277 struct zone
*zone
= page_zone(page
);
279 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
280 int file
= page_is_file_cache(page
);
281 int lru
= page_lru_base_type(page
);
282 del_page_from_lru_list(zone
, page
, lru
);
286 add_page_to_lru_list(zone
, page
, lru
);
287 __count_vm_event(PGACTIVATE
);
289 update_page_reclaim_stat(zone
, page
, file
, 1);
294 static DEFINE_PER_CPU(struct pagevec
, activate_page_pvecs
);
296 static void activate_page_drain(int cpu
)
298 struct pagevec
*pvec
= &per_cpu(activate_page_pvecs
, cpu
);
300 if (pagevec_count(pvec
))
301 pagevec_lru_move_fn(pvec
, __activate_page
, NULL
);
304 void activate_page(struct page
*page
)
306 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
307 struct pagevec
*pvec
= &get_cpu_var(activate_page_pvecs
);
309 page_cache_get(page
);
310 if (!pagevec_add(pvec
, page
))
311 pagevec_lru_move_fn(pvec
, __activate_page
, NULL
);
312 put_cpu_var(activate_page_pvecs
);
317 static inline void activate_page_drain(int cpu
)
321 void activate_page(struct page
*page
)
323 struct zone
*zone
= page_zone(page
);
325 spin_lock_irq(&zone
->lru_lock
);
326 __activate_page(page
, NULL
);
327 spin_unlock_irq(&zone
->lru_lock
);
332 * Mark a page as having seen activity.
334 * inactive,unreferenced -> inactive,referenced
335 * inactive,referenced -> active,unreferenced
336 * active,unreferenced -> active,referenced
338 void mark_page_accessed(struct page
*page
)
340 if (!PageActive(page
) && !PageUnevictable(page
) &&
341 PageReferenced(page
) && PageLRU(page
)) {
343 ClearPageReferenced(page
);
344 } else if (!PageReferenced(page
)) {
345 SetPageReferenced(page
);
349 EXPORT_SYMBOL(mark_page_accessed
);
351 void __lru_cache_add(struct page
*page
, enum lru_list lru
)
353 struct pagevec
*pvec
= &get_cpu_var(lru_add_pvecs
)[lru
];
355 page_cache_get(page
);
356 if (!pagevec_add(pvec
, page
))
357 ____pagevec_lru_add(pvec
, lru
);
358 put_cpu_var(lru_add_pvecs
);
360 EXPORT_SYMBOL(__lru_cache_add
);
363 * lru_cache_add_lru - add a page to a page list
364 * @page: the page to be added to the LRU.
365 * @lru: the LRU list to which the page is added.
367 void lru_cache_add_lru(struct page
*page
, enum lru_list lru
)
369 if (PageActive(page
)) {
370 VM_BUG_ON(PageUnevictable(page
));
371 ClearPageActive(page
);
372 } else if (PageUnevictable(page
)) {
373 VM_BUG_ON(PageActive(page
));
374 ClearPageUnevictable(page
);
377 VM_BUG_ON(PageLRU(page
) || PageActive(page
) || PageUnevictable(page
));
378 __lru_cache_add(page
, lru
);
382 * add_page_to_unevictable_list - add a page to the unevictable list
383 * @page: the page to be added to the unevictable list
385 * Add page directly to its zone's unevictable list. To avoid races with
386 * tasks that might be making the page evictable, through eg. munlock,
387 * munmap or exit, while it's not on the lru, we want to add the page
388 * while it's locked or otherwise "invisible" to other tasks. This is
389 * difficult to do when using the pagevec cache, so bypass that.
391 void add_page_to_unevictable_list(struct page
*page
)
393 struct zone
*zone
= page_zone(page
);
395 spin_lock_irq(&zone
->lru_lock
);
396 SetPageUnevictable(page
);
398 add_page_to_lru_list(zone
, page
, LRU_UNEVICTABLE
);
399 spin_unlock_irq(&zone
->lru_lock
);
403 * If the page can not be invalidated, it is moved to the
404 * inactive list to speed up its reclaim. It is moved to the
405 * head of the list, rather than the tail, to give the flusher
406 * threads some time to write it out, as this is much more
407 * effective than the single-page writeout from reclaim.
409 * If the page isn't page_mapped and dirty/writeback, the page
410 * could reclaim asap using PG_reclaim.
412 * 1. active, mapped page -> none
413 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
414 * 3. inactive, mapped page -> none
415 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
416 * 5. inactive, clean -> inactive, tail
419 * In 4, why it moves inactive's head, the VM expects the page would
420 * be write it out by flusher threads as this is much more effective
421 * than the single-page writeout from reclaim.
423 static void lru_deactivate_fn(struct page
*page
, void *arg
)
427 struct zone
*zone
= page_zone(page
);
432 if (PageUnevictable(page
))
435 /* Some processes are using the page */
436 if (page_mapped(page
))
439 active
= PageActive(page
);
441 file
= page_is_file_cache(page
);
442 lru
= page_lru_base_type(page
);
443 del_page_from_lru_list(zone
, page
, lru
+ active
);
444 ClearPageActive(page
);
445 ClearPageReferenced(page
);
446 add_page_to_lru_list(zone
, page
, lru
);
448 if (PageWriteback(page
) || PageDirty(page
)) {
450 * PG_reclaim could be raced with end_page_writeback
451 * It can make readahead confusing. But race window
452 * is _really_ small and it's non-critical problem.
454 SetPageReclaim(page
);
457 * The page's writeback ends up during pagevec
458 * We moves tha page into tail of inactive.
460 list_move_tail(&page
->lru
, &zone
->lru
[lru
].list
);
461 mem_cgroup_rotate_reclaimable_page(page
);
462 __count_vm_event(PGROTATED
);
466 __count_vm_event(PGDEACTIVATE
);
467 update_page_reclaim_stat(zone
, page
, file
, 0);
471 * Drain pages out of the cpu's pagevecs.
472 * Either "cpu" is the current CPU, and preemption has already been
473 * disabled; or "cpu" is being hot-unplugged, and is already dead.
475 static void drain_cpu_pagevecs(int cpu
)
477 struct pagevec
*pvecs
= per_cpu(lru_add_pvecs
, cpu
);
478 struct pagevec
*pvec
;
482 pvec
= &pvecs
[lru
- LRU_BASE
];
483 if (pagevec_count(pvec
))
484 ____pagevec_lru_add(pvec
, lru
);
487 pvec
= &per_cpu(lru_rotate_pvecs
, cpu
);
488 if (pagevec_count(pvec
)) {
491 /* No harm done if a racing interrupt already did this */
492 local_irq_save(flags
);
493 pagevec_move_tail(pvec
);
494 local_irq_restore(flags
);
497 pvec
= &per_cpu(lru_deactivate_pvecs
, cpu
);
498 if (pagevec_count(pvec
))
499 pagevec_lru_move_fn(pvec
, lru_deactivate_fn
, NULL
);
501 activate_page_drain(cpu
);
505 * deactivate_page - forcefully deactivate a page
506 * @page: page to deactivate
508 * This function hints the VM that @page is a good reclaim candidate,
509 * for example if its invalidation fails due to the page being dirty
510 * or under writeback.
512 void deactivate_page(struct page
*page
)
515 * In a workload with many unevictable page such as mprotect, unevictable
516 * page deactivation for accelerating reclaim is pointless.
518 if (PageUnevictable(page
))
521 if (likely(get_page_unless_zero(page
))) {
522 struct pagevec
*pvec
= &get_cpu_var(lru_deactivate_pvecs
);
524 if (!pagevec_add(pvec
, page
))
525 pagevec_lru_move_fn(pvec
, lru_deactivate_fn
, NULL
);
526 put_cpu_var(lru_deactivate_pvecs
);
530 void lru_add_drain(void)
532 drain_cpu_pagevecs(get_cpu());
536 static void lru_add_drain_per_cpu(struct work_struct
*dummy
)
542 * Returns 0 for success
544 int lru_add_drain_all(void)
546 return schedule_on_each_cpu(lru_add_drain_per_cpu
);
550 * Batched page_cache_release(). Decrement the reference count on all the
551 * passed pages. If it fell to zero then remove the page from the LRU and
554 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
555 * for the remainder of the operation.
557 * The locking in this function is against shrink_inactive_list(): we recheck
558 * the page count inside the lock to see whether shrink_inactive_list()
559 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
562 void release_pages(struct page
**pages
, int nr
, int cold
)
565 struct pagevec pages_to_free
;
566 struct zone
*zone
= NULL
;
567 unsigned long uninitialized_var(flags
);
569 pagevec_init(&pages_to_free
, cold
);
570 for (i
= 0; i
< nr
; i
++) {
571 struct page
*page
= pages
[i
];
573 if (unlikely(PageCompound(page
))) {
575 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
578 put_compound_page(page
);
582 if (!put_page_testzero(page
))
586 struct zone
*pagezone
= page_zone(page
);
588 if (pagezone
!= zone
) {
590 spin_unlock_irqrestore(&zone
->lru_lock
,
593 spin_lock_irqsave(&zone
->lru_lock
, flags
);
595 VM_BUG_ON(!PageLRU(page
));
596 __ClearPageLRU(page
);
597 del_page_from_lru(zone
, page
);
600 if (!pagevec_add(&pages_to_free
, page
)) {
602 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
605 __pagevec_free(&pages_to_free
);
606 pagevec_reinit(&pages_to_free
);
610 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
612 pagevec_free(&pages_to_free
);
614 EXPORT_SYMBOL(release_pages
);
617 * The pages which we're about to release may be in the deferred lru-addition
618 * queues. That would prevent them from really being freed right now. That's
619 * OK from a correctness point of view but is inefficient - those pages may be
620 * cache-warm and we want to give them back to the page allocator ASAP.
622 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
623 * and __pagevec_lru_add_active() call release_pages() directly to avoid
626 void __pagevec_release(struct pagevec
*pvec
)
629 release_pages(pvec
->pages
, pagevec_count(pvec
), pvec
->cold
);
630 pagevec_reinit(pvec
);
633 EXPORT_SYMBOL(__pagevec_release
);
635 /* used by __split_huge_page_refcount() */
636 void lru_add_page_tail(struct zone
* zone
,
637 struct page
*page
, struct page
*page_tail
)
642 struct list_head
*head
;
644 VM_BUG_ON(!PageHead(page
));
645 VM_BUG_ON(PageCompound(page_tail
));
646 VM_BUG_ON(PageLRU(page_tail
));
647 VM_BUG_ON(!spin_is_locked(&zone
->lru_lock
));
649 SetPageLRU(page_tail
);
651 if (page_evictable(page_tail
, NULL
)) {
652 if (PageActive(page
)) {
653 SetPageActive(page_tail
);
655 lru
= LRU_ACTIVE_ANON
;
658 lru
= LRU_INACTIVE_ANON
;
660 update_page_reclaim_stat(zone
, page_tail
, file
, active
);
661 if (likely(PageLRU(page
)))
662 head
= page
->lru
.prev
;
664 head
= &zone
->lru
[lru
].list
;
665 __add_page_to_lru_list(zone
, page_tail
, lru
, head
);
667 SetPageUnevictable(page_tail
);
668 add_page_to_lru_list(zone
, page_tail
, LRU_UNEVICTABLE
);
672 static void ____pagevec_lru_add_fn(struct page
*page
, void *arg
)
674 enum lru_list lru
= (enum lru_list
)arg
;
675 struct zone
*zone
= page_zone(page
);
676 int file
= is_file_lru(lru
);
677 int active
= is_active_lru(lru
);
679 VM_BUG_ON(PageActive(page
));
680 VM_BUG_ON(PageUnevictable(page
));
681 VM_BUG_ON(PageLRU(page
));
686 update_page_reclaim_stat(zone
, page
, file
, active
);
687 add_page_to_lru_list(zone
, page
, lru
);
691 * Add the passed pages to the LRU, then drop the caller's refcount
692 * on them. Reinitialises the caller's pagevec.
694 void ____pagevec_lru_add(struct pagevec
*pvec
, enum lru_list lru
)
696 VM_BUG_ON(is_unevictable_lru(lru
));
698 pagevec_lru_move_fn(pvec
, ____pagevec_lru_add_fn
, (void *)lru
);
701 EXPORT_SYMBOL(____pagevec_lru_add
);
704 * Try to drop buffers from the pages in a pagevec
706 void pagevec_strip(struct pagevec
*pvec
)
710 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
711 struct page
*page
= pvec
->pages
[i
];
713 if (page_has_private(page
) && trylock_page(page
)) {
714 if (page_has_private(page
))
715 try_to_release_page(page
, 0);
722 * pagevec_lookup - gang pagecache lookup
723 * @pvec: Where the resulting pages are placed
724 * @mapping: The address_space to search
725 * @start: The starting page index
726 * @nr_pages: The maximum number of pages
728 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
729 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
730 * reference against the pages in @pvec.
732 * The search returns a group of mapping-contiguous pages with ascending
733 * indexes. There may be holes in the indices due to not-present pages.
735 * pagevec_lookup() returns the number of pages which were found.
737 unsigned pagevec_lookup(struct pagevec
*pvec
, struct address_space
*mapping
,
738 pgoff_t start
, unsigned nr_pages
)
740 pvec
->nr
= find_get_pages(mapping
, start
, nr_pages
, pvec
->pages
);
741 return pagevec_count(pvec
);
744 EXPORT_SYMBOL(pagevec_lookup
);
746 unsigned pagevec_lookup_tag(struct pagevec
*pvec
, struct address_space
*mapping
,
747 pgoff_t
*index
, int tag
, unsigned nr_pages
)
749 pvec
->nr
= find_get_pages_tag(mapping
, index
, tag
,
750 nr_pages
, pvec
->pages
);
751 return pagevec_count(pvec
);
754 EXPORT_SYMBOL(pagevec_lookup_tag
);
757 * Perform any setup for the swap system
759 void __init
swap_setup(void)
761 unsigned long megs
= totalram_pages
>> (20 - PAGE_SHIFT
);
764 bdi_init(swapper_space
.backing_dev_info
);
767 /* Use a smaller cluster for small-memory machines */
773 * Right now other parts of the system means that we
774 * _really_ don't want to cluster much more