]>
git.proxmox.com Git - mirror_ubuntu-zesty-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
);
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 if (PageHead(page_head
))
136 __put_compound_page(page_head
);
138 __put_single_page(page_head
);
141 /* page_head is a dangling pointer */
142 VM_BUG_ON(PageTail(page
));
145 } else if (put_page_testzero(page
)) {
147 __put_compound_page(page
);
149 __put_single_page(page
);
153 void put_page(struct page
*page
)
155 if (unlikely(PageCompound(page
)))
156 put_compound_page(page
);
157 else if (put_page_testzero(page
))
158 __put_single_page(page
);
160 EXPORT_SYMBOL(put_page
);
163 * put_pages_list() - release a list of pages
164 * @pages: list of pages threaded on page->lru
166 * Release a list of pages which are strung together on page.lru. Currently
167 * used by read_cache_pages() and related error recovery code.
169 void put_pages_list(struct list_head
*pages
)
171 while (!list_empty(pages
)) {
174 victim
= list_entry(pages
->prev
, struct page
, lru
);
175 list_del(&victim
->lru
);
176 page_cache_release(victim
);
179 EXPORT_SYMBOL(put_pages_list
);
181 static void pagevec_lru_move_fn(struct pagevec
*pvec
,
182 void (*move_fn
)(struct page
*page
, void *arg
),
186 struct zone
*zone
= NULL
;
187 unsigned long flags
= 0;
189 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
190 struct page
*page
= pvec
->pages
[i
];
191 struct zone
*pagezone
= page_zone(page
);
193 if (pagezone
!= zone
) {
195 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
197 spin_lock_irqsave(&zone
->lru_lock
, flags
);
200 (*move_fn
)(page
, arg
);
203 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
204 release_pages(pvec
->pages
, pagevec_count(pvec
), pvec
->cold
);
205 pagevec_reinit(pvec
);
208 static void pagevec_move_tail_fn(struct page
*page
, void *arg
)
211 struct zone
*zone
= page_zone(page
);
213 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
214 int lru
= page_lru_base_type(page
);
215 list_move_tail(&page
->lru
, &zone
->lru
[lru
].list
);
221 * pagevec_move_tail() must be called with IRQ disabled.
222 * Otherwise this may cause nasty races.
224 static void pagevec_move_tail(struct pagevec
*pvec
)
228 pagevec_lru_move_fn(pvec
, pagevec_move_tail_fn
, &pgmoved
);
229 __count_vm_events(PGROTATED
, pgmoved
);
233 * Writeback is about to end against a page which has been marked for immediate
234 * reclaim. If it still appears to be reclaimable, move it to the tail of the
237 void rotate_reclaimable_page(struct page
*page
)
239 if (!PageLocked(page
) && !PageDirty(page
) && !PageActive(page
) &&
240 !PageUnevictable(page
) && PageLRU(page
)) {
241 struct pagevec
*pvec
;
244 page_cache_get(page
);
245 local_irq_save(flags
);
246 pvec
= &__get_cpu_var(lru_rotate_pvecs
);
247 if (!pagevec_add(pvec
, page
))
248 pagevec_move_tail(pvec
);
249 local_irq_restore(flags
);
253 static void update_page_reclaim_stat(struct zone
*zone
, struct page
*page
,
254 int file
, int rotated
)
256 struct zone_reclaim_stat
*reclaim_stat
= &zone
->reclaim_stat
;
257 struct zone_reclaim_stat
*memcg_reclaim_stat
;
259 memcg_reclaim_stat
= mem_cgroup_get_reclaim_stat_from_page(page
);
261 reclaim_stat
->recent_scanned
[file
]++;
263 reclaim_stat
->recent_rotated
[file
]++;
265 if (!memcg_reclaim_stat
)
268 memcg_reclaim_stat
->recent_scanned
[file
]++;
270 memcg_reclaim_stat
->recent_rotated
[file
]++;
274 * FIXME: speed this up?
276 void activate_page(struct page
*page
)
278 struct zone
*zone
= page_zone(page
);
280 spin_lock_irq(&zone
->lru_lock
);
281 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
282 int file
= page_is_file_cache(page
);
283 int lru
= page_lru_base_type(page
);
284 del_page_from_lru_list(zone
, page
, lru
);
288 add_page_to_lru_list(zone
, page
, lru
);
289 __count_vm_event(PGACTIVATE
);
291 update_page_reclaim_stat(zone
, page
, file
, 1);
293 spin_unlock_irq(&zone
->lru_lock
);
297 * Mark a page as having seen activity.
299 * inactive,unreferenced -> inactive,referenced
300 * inactive,referenced -> active,unreferenced
301 * active,unreferenced -> active,referenced
303 void mark_page_accessed(struct page
*page
)
305 if (!PageActive(page
) && !PageUnevictable(page
) &&
306 PageReferenced(page
) && PageLRU(page
)) {
308 ClearPageReferenced(page
);
309 } else if (!PageReferenced(page
)) {
310 SetPageReferenced(page
);
314 EXPORT_SYMBOL(mark_page_accessed
);
316 void __lru_cache_add(struct page
*page
, enum lru_list lru
)
318 struct pagevec
*pvec
= &get_cpu_var(lru_add_pvecs
)[lru
];
320 page_cache_get(page
);
321 if (!pagevec_add(pvec
, page
))
322 ____pagevec_lru_add(pvec
, lru
);
323 put_cpu_var(lru_add_pvecs
);
325 EXPORT_SYMBOL(__lru_cache_add
);
328 * lru_cache_add_lru - add a page to a page list
329 * @page: the page to be added to the LRU.
330 * @lru: the LRU list to which the page is added.
332 void lru_cache_add_lru(struct page
*page
, enum lru_list lru
)
334 if (PageActive(page
)) {
335 VM_BUG_ON(PageUnevictable(page
));
336 ClearPageActive(page
);
337 } else if (PageUnevictable(page
)) {
338 VM_BUG_ON(PageActive(page
));
339 ClearPageUnevictable(page
);
342 VM_BUG_ON(PageLRU(page
) || PageActive(page
) || PageUnevictable(page
));
343 __lru_cache_add(page
, lru
);
347 * add_page_to_unevictable_list - add a page to the unevictable list
348 * @page: the page to be added to the unevictable list
350 * Add page directly to its zone's unevictable list. To avoid races with
351 * tasks that might be making the page evictable, through eg. munlock,
352 * munmap or exit, while it's not on the lru, we want to add the page
353 * while it's locked or otherwise "invisible" to other tasks. This is
354 * difficult to do when using the pagevec cache, so bypass that.
356 void add_page_to_unevictable_list(struct page
*page
)
358 struct zone
*zone
= page_zone(page
);
360 spin_lock_irq(&zone
->lru_lock
);
361 SetPageUnevictable(page
);
363 add_page_to_lru_list(zone
, page
, LRU_UNEVICTABLE
);
364 spin_unlock_irq(&zone
->lru_lock
);
368 * Drain pages out of the cpu's pagevecs.
369 * Either "cpu" is the current CPU, and preemption has already been
370 * disabled; or "cpu" is being hot-unplugged, and is already dead.
372 static void drain_cpu_pagevecs(int cpu
)
374 struct pagevec
*pvecs
= per_cpu(lru_add_pvecs
, cpu
);
375 struct pagevec
*pvec
;
379 pvec
= &pvecs
[lru
- LRU_BASE
];
380 if (pagevec_count(pvec
))
381 ____pagevec_lru_add(pvec
, lru
);
384 pvec
= &per_cpu(lru_rotate_pvecs
, cpu
);
385 if (pagevec_count(pvec
)) {
388 /* No harm done if a racing interrupt already did this */
389 local_irq_save(flags
);
390 pagevec_move_tail(pvec
);
391 local_irq_restore(flags
);
395 void lru_add_drain(void)
397 drain_cpu_pagevecs(get_cpu());
401 static void lru_add_drain_per_cpu(struct work_struct
*dummy
)
407 * Returns 0 for success
409 int lru_add_drain_all(void)
411 return schedule_on_each_cpu(lru_add_drain_per_cpu
);
415 * Batched page_cache_release(). Decrement the reference count on all the
416 * passed pages. If it fell to zero then remove the page from the LRU and
419 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
420 * for the remainder of the operation.
422 * The locking in this function is against shrink_inactive_list(): we recheck
423 * the page count inside the lock to see whether shrink_inactive_list()
424 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
427 void release_pages(struct page
**pages
, int nr
, int cold
)
430 struct pagevec pages_to_free
;
431 struct zone
*zone
= NULL
;
432 unsigned long uninitialized_var(flags
);
434 pagevec_init(&pages_to_free
, cold
);
435 for (i
= 0; i
< nr
; i
++) {
436 struct page
*page
= pages
[i
];
438 if (unlikely(PageCompound(page
))) {
440 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
443 put_compound_page(page
);
447 if (!put_page_testzero(page
))
451 struct zone
*pagezone
= page_zone(page
);
453 if (pagezone
!= zone
) {
455 spin_unlock_irqrestore(&zone
->lru_lock
,
458 spin_lock_irqsave(&zone
->lru_lock
, flags
);
460 VM_BUG_ON(!PageLRU(page
));
461 __ClearPageLRU(page
);
462 del_page_from_lru(zone
, page
);
465 if (!pagevec_add(&pages_to_free
, page
)) {
467 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
470 __pagevec_free(&pages_to_free
);
471 pagevec_reinit(&pages_to_free
);
475 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
477 pagevec_free(&pages_to_free
);
479 EXPORT_SYMBOL(release_pages
);
482 * The pages which we're about to release may be in the deferred lru-addition
483 * queues. That would prevent them from really being freed right now. That's
484 * OK from a correctness point of view but is inefficient - those pages may be
485 * cache-warm and we want to give them back to the page allocator ASAP.
487 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
488 * and __pagevec_lru_add_active() call release_pages() directly to avoid
491 void __pagevec_release(struct pagevec
*pvec
)
494 release_pages(pvec
->pages
, pagevec_count(pvec
), pvec
->cold
);
495 pagevec_reinit(pvec
);
498 EXPORT_SYMBOL(__pagevec_release
);
500 /* used by __split_huge_page_refcount() */
501 void lru_add_page_tail(struct zone
* zone
,
502 struct page
*page
, struct page
*page_tail
)
507 struct list_head
*head
;
509 VM_BUG_ON(!PageHead(page
));
510 VM_BUG_ON(PageCompound(page_tail
));
511 VM_BUG_ON(PageLRU(page_tail
));
512 VM_BUG_ON(!spin_is_locked(&zone
->lru_lock
));
514 SetPageLRU(page_tail
);
516 if (page_evictable(page_tail
, NULL
)) {
517 if (PageActive(page
)) {
518 SetPageActive(page_tail
);
520 lru
= LRU_ACTIVE_ANON
;
523 lru
= LRU_INACTIVE_ANON
;
525 update_page_reclaim_stat(zone
, page_tail
, file
, active
);
526 if (likely(PageLRU(page
)))
527 head
= page
->lru
.prev
;
529 head
= &zone
->lru
[lru
].list
;
530 __add_page_to_lru_list(zone
, page_tail
, lru
, head
);
532 SetPageUnevictable(page_tail
);
533 add_page_to_lru_list(zone
, page_tail
, LRU_UNEVICTABLE
);
537 static void ____pagevec_lru_add_fn(struct page
*page
, void *arg
)
539 enum lru_list lru
= (enum lru_list
)arg
;
540 struct zone
*zone
= page_zone(page
);
541 int file
= is_file_lru(lru
);
542 int active
= is_active_lru(lru
);
544 VM_BUG_ON(PageActive(page
));
545 VM_BUG_ON(PageUnevictable(page
));
546 VM_BUG_ON(PageLRU(page
));
551 update_page_reclaim_stat(zone
, page
, file
, active
);
552 add_page_to_lru_list(zone
, page
, lru
);
556 * Add the passed pages to the LRU, then drop the caller's refcount
557 * on them. Reinitialises the caller's pagevec.
559 void ____pagevec_lru_add(struct pagevec
*pvec
, enum lru_list lru
)
561 VM_BUG_ON(is_unevictable_lru(lru
));
563 pagevec_lru_move_fn(pvec
, ____pagevec_lru_add_fn
, (void *)lru
);
566 EXPORT_SYMBOL(____pagevec_lru_add
);
569 * Try to drop buffers from the pages in a pagevec
571 void pagevec_strip(struct pagevec
*pvec
)
575 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
576 struct page
*page
= pvec
->pages
[i
];
578 if (page_has_private(page
) && trylock_page(page
)) {
579 if (page_has_private(page
))
580 try_to_release_page(page
, 0);
587 * pagevec_lookup - gang pagecache lookup
588 * @pvec: Where the resulting pages are placed
589 * @mapping: The address_space to search
590 * @start: The starting page index
591 * @nr_pages: The maximum number of pages
593 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
594 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
595 * reference against the pages in @pvec.
597 * The search returns a group of mapping-contiguous pages with ascending
598 * indexes. There may be holes in the indices due to not-present pages.
600 * pagevec_lookup() returns the number of pages which were found.
602 unsigned pagevec_lookup(struct pagevec
*pvec
, struct address_space
*mapping
,
603 pgoff_t start
, unsigned nr_pages
)
605 pvec
->nr
= find_get_pages(mapping
, start
, nr_pages
, pvec
->pages
);
606 return pagevec_count(pvec
);
609 EXPORT_SYMBOL(pagevec_lookup
);
611 unsigned pagevec_lookup_tag(struct pagevec
*pvec
, struct address_space
*mapping
,
612 pgoff_t
*index
, int tag
, unsigned nr_pages
)
614 pvec
->nr
= find_get_pages_tag(mapping
, index
, tag
,
615 nr_pages
, pvec
->pages
);
616 return pagevec_count(pvec
);
619 EXPORT_SYMBOL(pagevec_lookup_tag
);
622 * Perform any setup for the swap system
624 void __init
swap_setup(void)
626 unsigned long megs
= totalram_pages
>> (20 - PAGE_SHIFT
);
629 bdi_init(swapper_space
.backing_dev_info
);
632 /* Use a smaller cluster for small-memory machines */
638 * Right now other parts of the system means that we
639 * _really_ don't want to cluster much more