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>
34 /* How many pages do we try to swap or page in/out together? */
37 static DEFINE_PER_CPU(struct pagevec
[NR_LRU_LISTS
], lru_add_pvecs
);
38 static DEFINE_PER_CPU(struct pagevec
, lru_rotate_pvecs
);
41 * This path almost never happens for VM activity - pages are normally
42 * freed via pagevecs. But it gets used by networking.
44 static void __page_cache_release(struct page
*page
)
48 struct zone
*zone
= page_zone(page
);
50 spin_lock_irqsave(&zone
->lru_lock
, flags
);
51 VM_BUG_ON(!PageLRU(page
));
53 del_page_from_lru(zone
, page
);
54 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
59 static void put_compound_page(struct page
*page
)
61 page
= compound_head(page
);
62 if (put_page_testzero(page
)) {
63 compound_page_dtor
*dtor
;
65 dtor
= get_compound_page_dtor(page
);
70 void put_page(struct page
*page
)
72 if (unlikely(PageCompound(page
)))
73 put_compound_page(page
);
74 else if (put_page_testzero(page
))
75 __page_cache_release(page
);
77 EXPORT_SYMBOL(put_page
);
80 * put_pages_list() - release a list of pages
81 * @pages: list of pages threaded on page->lru
83 * Release a list of pages which are strung together on page.lru. Currently
84 * used by read_cache_pages() and related error recovery code.
86 void put_pages_list(struct list_head
*pages
)
88 while (!list_empty(pages
)) {
91 victim
= list_entry(pages
->prev
, struct page
, lru
);
92 list_del(&victim
->lru
);
93 page_cache_release(victim
);
96 EXPORT_SYMBOL(put_pages_list
);
99 * pagevec_move_tail() must be called with IRQ disabled.
100 * Otherwise this may cause nasty races.
102 static void pagevec_move_tail(struct pagevec
*pvec
)
106 struct zone
*zone
= NULL
;
108 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
109 struct page
*page
= pvec
->pages
[i
];
110 struct zone
*pagezone
= page_zone(page
);
112 if (pagezone
!= zone
) {
114 spin_unlock(&zone
->lru_lock
);
116 spin_lock(&zone
->lru_lock
);
118 if (PageLRU(page
) && !PageActive(page
)) {
119 list_move_tail(&page
->lru
, &zone
->lru
[LRU_INACTIVE
].list
);
124 spin_unlock(&zone
->lru_lock
);
125 __count_vm_events(PGROTATED
, pgmoved
);
126 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
127 pagevec_reinit(pvec
);
131 * Writeback is about to end against a page which has been marked for immediate
132 * reclaim. If it still appears to be reclaimable, move it to the tail of the
135 void rotate_reclaimable_page(struct page
*page
)
137 if (!PageLocked(page
) && !PageDirty(page
) && !PageActive(page
) &&
139 struct pagevec
*pvec
;
142 page_cache_get(page
);
143 local_irq_save(flags
);
144 pvec
= &__get_cpu_var(lru_rotate_pvecs
);
145 if (!pagevec_add(pvec
, page
))
146 pagevec_move_tail(pvec
);
147 local_irq_restore(flags
);
152 * FIXME: speed this up?
154 void activate_page(struct page
*page
)
156 struct zone
*zone
= page_zone(page
);
158 spin_lock_irq(&zone
->lru_lock
);
159 if (PageLRU(page
) && !PageActive(page
)) {
160 del_page_from_inactive_list(zone
, page
);
162 add_page_to_active_list(zone
, page
);
163 __count_vm_event(PGACTIVATE
);
164 mem_cgroup_move_lists(page
, true);
166 spin_unlock_irq(&zone
->lru_lock
);
170 * Mark a page as having seen activity.
172 * inactive,unreferenced -> inactive,referenced
173 * inactive,referenced -> active,unreferenced
174 * active,unreferenced -> active,referenced
176 void mark_page_accessed(struct page
*page
)
178 if (!PageActive(page
) && PageReferenced(page
) && PageLRU(page
)) {
180 ClearPageReferenced(page
);
181 } else if (!PageReferenced(page
)) {
182 SetPageReferenced(page
);
186 EXPORT_SYMBOL(mark_page_accessed
);
188 void __lru_cache_add(struct page
*page
, enum lru_list lru
)
190 struct pagevec
*pvec
= &get_cpu_var(lru_add_pvecs
)[lru
];
192 page_cache_get(page
);
193 if (!pagevec_add(pvec
, page
))
194 ____pagevec_lru_add(pvec
, lru
);
195 put_cpu_var(lru_add_pvecs
);
199 * lru_cache_add_lru - add a page to a page list
200 * @page: the page to be added to the LRU.
201 * @lru: the LRU list to which the page is added.
203 void lru_cache_add_lru(struct page
*page
, enum lru_list lru
)
205 if (PageActive(page
)) {
206 ClearPageActive(page
);
209 VM_BUG_ON(PageLRU(page
) || PageActive(page
));
210 __lru_cache_add(page
, lru
);
214 * Drain pages out of the cpu's pagevecs.
215 * Either "cpu" is the current CPU, and preemption has already been
216 * disabled; or "cpu" is being hot-unplugged, and is already dead.
218 static void drain_cpu_pagevecs(int cpu
)
220 struct pagevec
*pvecs
= per_cpu(lru_add_pvecs
, cpu
);
221 struct pagevec
*pvec
;
225 pvec
= &pvecs
[lru
- LRU_BASE
];
226 if (pagevec_count(pvec
))
227 ____pagevec_lru_add(pvec
, lru
);
230 pvec
= &per_cpu(lru_rotate_pvecs
, cpu
);
231 if (pagevec_count(pvec
)) {
234 /* No harm done if a racing interrupt already did this */
235 local_irq_save(flags
);
236 pagevec_move_tail(pvec
);
237 local_irq_restore(flags
);
241 void lru_add_drain(void)
243 drain_cpu_pagevecs(get_cpu());
248 static void lru_add_drain_per_cpu(struct work_struct
*dummy
)
254 * Returns 0 for success
256 int lru_add_drain_all(void)
258 return schedule_on_each_cpu(lru_add_drain_per_cpu
);
264 * Returns 0 for success
266 int lru_add_drain_all(void)
274 * Batched page_cache_release(). Decrement the reference count on all the
275 * passed pages. If it fell to zero then remove the page from the LRU and
278 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
279 * for the remainder of the operation.
281 * The locking in this function is against shrink_inactive_list(): we recheck
282 * the page count inside the lock to see whether shrink_inactive_list()
283 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
286 void release_pages(struct page
**pages
, int nr
, int cold
)
289 struct pagevec pages_to_free
;
290 struct zone
*zone
= NULL
;
291 unsigned long uninitialized_var(flags
);
293 pagevec_init(&pages_to_free
, cold
);
294 for (i
= 0; i
< nr
; i
++) {
295 struct page
*page
= pages
[i
];
297 if (unlikely(PageCompound(page
))) {
299 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
302 put_compound_page(page
);
306 if (!put_page_testzero(page
))
310 struct zone
*pagezone
= page_zone(page
);
311 if (pagezone
!= zone
) {
313 spin_unlock_irqrestore(&zone
->lru_lock
,
316 spin_lock_irqsave(&zone
->lru_lock
, flags
);
318 VM_BUG_ON(!PageLRU(page
));
319 __ClearPageLRU(page
);
320 del_page_from_lru(zone
, page
);
323 if (!pagevec_add(&pages_to_free
, page
)) {
325 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
328 __pagevec_free(&pages_to_free
);
329 pagevec_reinit(&pages_to_free
);
333 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
335 pagevec_free(&pages_to_free
);
339 * The pages which we're about to release may be in the deferred lru-addition
340 * queues. That would prevent them from really being freed right now. That's
341 * OK from a correctness point of view but is inefficient - those pages may be
342 * cache-warm and we want to give them back to the page allocator ASAP.
344 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
345 * and __pagevec_lru_add_active() call release_pages() directly to avoid
348 void __pagevec_release(struct pagevec
*pvec
)
351 release_pages(pvec
->pages
, pagevec_count(pvec
), pvec
->cold
);
352 pagevec_reinit(pvec
);
355 EXPORT_SYMBOL(__pagevec_release
);
358 * pagevec_release() for pages which are known to not be on the LRU
360 * This function reinitialises the caller's pagevec.
362 void __pagevec_release_nonlru(struct pagevec
*pvec
)
365 struct pagevec pages_to_free
;
367 pagevec_init(&pages_to_free
, pvec
->cold
);
368 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
369 struct page
*page
= pvec
->pages
[i
];
371 VM_BUG_ON(PageLRU(page
));
372 if (put_page_testzero(page
))
373 pagevec_add(&pages_to_free
, page
);
375 pagevec_free(&pages_to_free
);
376 pagevec_reinit(pvec
);
380 * Add the passed pages to the LRU, then drop the caller's refcount
381 * on them. Reinitialises the caller's pagevec.
383 void ____pagevec_lru_add(struct pagevec
*pvec
, enum lru_list lru
)
386 struct zone
*zone
= NULL
;
388 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
389 struct page
*page
= pvec
->pages
[i
];
390 struct zone
*pagezone
= page_zone(page
);
392 if (pagezone
!= zone
) {
394 spin_unlock_irq(&zone
->lru_lock
);
396 spin_lock_irq(&zone
->lru_lock
);
398 VM_BUG_ON(PageLRU(page
));
400 if (is_active_lru(lru
))
402 add_page_to_lru_list(zone
, page
, lru
);
405 spin_unlock_irq(&zone
->lru_lock
);
406 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
407 pagevec_reinit(pvec
);
410 EXPORT_SYMBOL(____pagevec_lru_add
);
413 * Try to drop buffers from the pages in a pagevec
415 void pagevec_strip(struct pagevec
*pvec
)
419 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
420 struct page
*page
= pvec
->pages
[i
];
422 if (PagePrivate(page
) && trylock_page(page
)) {
423 if (PagePrivate(page
))
424 try_to_release_page(page
, 0);
431 * pagevec_swap_free - try to free swap space from the pages in a pagevec
432 * @pvec: pagevec with swapcache pages to free the swap space of
434 * The caller needs to hold an extra reference to each page and
435 * not hold the page lock on the pages. This function uses a
436 * trylock on the page lock so it may not always free the swap
437 * space associated with a page.
439 void pagevec_swap_free(struct pagevec
*pvec
)
443 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
444 struct page
*page
= pvec
->pages
[i
];
446 if (PageSwapCache(page
) && trylock_page(page
)) {
447 if (PageSwapCache(page
))
448 remove_exclusive_swap_page_ref(page
);
455 * pagevec_lookup - gang pagecache lookup
456 * @pvec: Where the resulting pages are placed
457 * @mapping: The address_space to search
458 * @start: The starting page index
459 * @nr_pages: The maximum number of pages
461 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
462 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
463 * reference against the pages in @pvec.
465 * The search returns a group of mapping-contiguous pages with ascending
466 * indexes. There may be holes in the indices due to not-present pages.
468 * pagevec_lookup() returns the number of pages which were found.
470 unsigned pagevec_lookup(struct pagevec
*pvec
, struct address_space
*mapping
,
471 pgoff_t start
, unsigned nr_pages
)
473 pvec
->nr
= find_get_pages(mapping
, start
, nr_pages
, pvec
->pages
);
474 return pagevec_count(pvec
);
477 EXPORT_SYMBOL(pagevec_lookup
);
479 unsigned pagevec_lookup_tag(struct pagevec
*pvec
, struct address_space
*mapping
,
480 pgoff_t
*index
, int tag
, unsigned nr_pages
)
482 pvec
->nr
= find_get_pages_tag(mapping
, index
, tag
,
483 nr_pages
, pvec
->pages
);
484 return pagevec_count(pvec
);
487 EXPORT_SYMBOL(pagevec_lookup_tag
);
491 * We tolerate a little inaccuracy to avoid ping-ponging the counter between
494 #define ACCT_THRESHOLD max(16, NR_CPUS * 2)
496 static DEFINE_PER_CPU(long, committed_space
);
498 void vm_acct_memory(long pages
)
503 local
= &__get_cpu_var(committed_space
);
505 if (*local
> ACCT_THRESHOLD
|| *local
< -ACCT_THRESHOLD
) {
506 atomic_long_add(*local
, &vm_committed_space
);
512 #ifdef CONFIG_HOTPLUG_CPU
514 /* Drop the CPU's cached committed space back into the central pool. */
515 static int cpu_swap_callback(struct notifier_block
*nfb
,
516 unsigned long action
,
521 committed
= &per_cpu(committed_space
, (long)hcpu
);
522 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
523 atomic_long_add(*committed
, &vm_committed_space
);
525 drain_cpu_pagevecs((long)hcpu
);
529 #endif /* CONFIG_HOTPLUG_CPU */
530 #endif /* CONFIG_SMP */
533 * Perform any setup for the swap system
535 void __init
swap_setup(void)
537 unsigned long megs
= num_physpages
>> (20 - PAGE_SHIFT
);
540 bdi_init(swapper_space
.backing_dev_info
);
543 /* Use a smaller cluster for small-memory machines */
549 * Right now other parts of the system means that we
550 * _really_ don't want to cluster much more
552 #ifdef CONFIG_HOTPLUG_CPU
553 hotcpu_notifier(cpu_swap_callback
, 0);