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/export.h>
25 #include <linux/mm_inline.h>
26 #include <linux/percpu_counter.h>
27 #include <linux/percpu.h>
28 #include <linux/cpu.h>
29 #include <linux/notifier.h>
30 #include <linux/backing-dev.h>
31 #include <linux/memcontrol.h>
32 #include <linux/gfp.h>
33 #include <linux/uio.h>
34 #include <linux/hugetlb.h>
38 /* How many pages do we try to swap or page in/out together? */
41 static DEFINE_PER_CPU(struct pagevec
[NR_LRU_LISTS
], lru_add_pvecs
);
42 static DEFINE_PER_CPU(struct pagevec
, lru_rotate_pvecs
);
43 static DEFINE_PER_CPU(struct pagevec
, lru_deactivate_pvecs
);
46 * This path almost never happens for VM activity - pages are normally
47 * freed via pagevecs. But it gets used by networking.
49 static void __page_cache_release(struct page
*page
)
52 struct zone
*zone
= page_zone(page
);
53 struct lruvec
*lruvec
;
56 spin_lock_irqsave(&zone
->lru_lock
, flags
);
57 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
58 VM_BUG_ON(!PageLRU(page
));
60 del_page_from_lru_list(page
, lruvec
, page_off_lru(page
));
61 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
65 static void __put_single_page(struct page
*page
)
67 __page_cache_release(page
);
68 free_hot_cold_page(page
, 0);
71 static void __put_compound_page(struct page
*page
)
73 compound_page_dtor
*dtor
;
75 __page_cache_release(page
);
76 dtor
= get_compound_page_dtor(page
);
80 static void put_compound_page(struct page
*page
)
82 if (unlikely(PageTail(page
))) {
83 /* __split_huge_page_refcount can run under us */
84 struct page
*page_head
= compound_head(page
);
86 if (likely(page
!= page_head
&&
87 get_page_unless_zero(page_head
))) {
91 * THP can not break up slab pages so avoid taking
92 * compound_lock(). Slab performs non-atomic bit ops
93 * on page->flags for better performance. In particular
94 * slab_unlock() in slub used to be a hot path. It is
95 * still hot on arches that do not support
96 * this_cpu_cmpxchg_double().
98 if (PageSlab(page_head
) || PageHeadHuge(page_head
)) {
99 if (likely(PageTail(page
))) {
101 * __split_huge_page_refcount
104 VM_BUG_ON(!PageHead(page_head
));
105 atomic_dec(&page
->_mapcount
);
106 if (put_page_testzero(page_head
))
108 if (put_page_testzero(page_head
))
109 __put_compound_page(page_head
);
113 * __split_huge_page_refcount
114 * run before us, "page" was a
115 * THP tail. The split
116 * page_head has been freed
117 * and reallocated as slab or
118 * hugetlbfs page of smaller
119 * order (only possible if
120 * reallocated as slab on
126 * page_head wasn't a dangling pointer but it
127 * may not be a head page anymore by the time
128 * we obtain the lock. That is ok as long as it
129 * can't be freed from under us.
131 flags
= compound_lock_irqsave(page_head
);
132 if (unlikely(!PageTail(page
))) {
133 /* __split_huge_page_refcount run before us */
134 compound_unlock_irqrestore(page_head
, flags
);
136 if (put_page_testzero(page_head
)) {
138 * The head page may have been
139 * freed and reallocated as a
140 * compound page of smaller
141 * order and then freed again.
142 * All we know is that it
143 * cannot have become: a THP
144 * page, a compound page of
145 * higher order, a tail page.
146 * That is because we still
147 * hold the refcount of the
149 * page_head was the THP head
152 if (PageHead(page_head
))
153 __put_compound_page(page_head
);
155 __put_single_page(page_head
);
158 if (put_page_testzero(page
))
159 __put_single_page(page
);
162 VM_BUG_ON(page_head
!= page
->first_page
);
164 * We can release the refcount taken by
165 * get_page_unless_zero() now that
166 * __split_huge_page_refcount() is blocked on
169 if (put_page_testzero(page_head
))
171 /* __split_huge_page_refcount will wait now */
172 VM_BUG_ON(page_mapcount(page
) <= 0);
173 atomic_dec(&page
->_mapcount
);
174 VM_BUG_ON(atomic_read(&page_head
->_count
) <= 0);
175 VM_BUG_ON(atomic_read(&page
->_count
) != 0);
176 compound_unlock_irqrestore(page_head
, flags
);
178 if (put_page_testzero(page_head
)) {
179 if (PageHead(page_head
))
180 __put_compound_page(page_head
);
182 __put_single_page(page_head
);
185 /* page_head is a dangling pointer */
186 VM_BUG_ON(PageTail(page
));
189 } else if (put_page_testzero(page
)) {
191 __put_compound_page(page
);
193 __put_single_page(page
);
197 void put_page(struct page
*page
)
199 if (unlikely(PageCompound(page
)))
200 put_compound_page(page
);
201 else if (put_page_testzero(page
))
202 __put_single_page(page
);
204 EXPORT_SYMBOL(put_page
);
207 * This function is exported but must not be called by anything other
208 * than get_page(). It implements the slow path of get_page().
210 bool __get_page_tail(struct page
*page
)
213 * This takes care of get_page() if run on a tail page
214 * returned by one of the get_user_pages/follow_page variants.
215 * get_user_pages/follow_page itself doesn't need the compound
216 * lock because it runs __get_page_tail_foll() under the
217 * proper PT lock that already serializes against
222 struct page
*page_head
= compound_head(page
);
224 if (likely(page
!= page_head
&& get_page_unless_zero(page_head
))) {
225 /* Ref to put_compound_page() comment. */
226 if (PageSlab(page_head
) || PageHeadHuge(page_head
)) {
227 if (likely(PageTail(page
))) {
229 * This is a hugetlbfs page or a slab
230 * page. __split_huge_page_refcount
233 VM_BUG_ON(!PageHead(page_head
));
234 __get_page_tail_foll(page
, false);
238 * __split_huge_page_refcount run
239 * before us, "page" was a THP
240 * tail. The split page_head has been
241 * freed and reallocated as slab or
242 * hugetlbfs page of smaller order
243 * (only possible if reallocated as
252 * page_head wasn't a dangling pointer but it
253 * may not be a head page anymore by the time
254 * we obtain the lock. That is ok as long as it
255 * can't be freed from under us.
257 flags
= compound_lock_irqsave(page_head
);
258 /* here __split_huge_page_refcount won't run anymore */
259 if (likely(PageTail(page
))) {
260 __get_page_tail_foll(page
, false);
263 compound_unlock_irqrestore(page_head
, flags
);
269 EXPORT_SYMBOL(__get_page_tail
);
272 * put_pages_list() - release a list of pages
273 * @pages: list of pages threaded on page->lru
275 * Release a list of pages which are strung together on page.lru. Currently
276 * used by read_cache_pages() and related error recovery code.
278 void put_pages_list(struct list_head
*pages
)
280 while (!list_empty(pages
)) {
283 victim
= list_entry(pages
->prev
, struct page
, lru
);
284 list_del(&victim
->lru
);
285 page_cache_release(victim
);
288 EXPORT_SYMBOL(put_pages_list
);
291 * get_kernel_pages() - pin kernel pages in memory
292 * @kiov: An array of struct kvec structures
293 * @nr_segs: number of segments to pin
294 * @write: pinning for read/write, currently ignored
295 * @pages: array that receives pointers to the pages pinned.
296 * Should be at least nr_segs long.
298 * Returns number of pages pinned. This may be fewer than the number
299 * requested. If nr_pages is 0 or negative, returns 0. If no pages
300 * were pinned, returns -errno. Each page returned must be released
301 * with a put_page() call when it is finished with.
303 int get_kernel_pages(const struct kvec
*kiov
, int nr_segs
, int write
,
308 for (seg
= 0; seg
< nr_segs
; seg
++) {
309 if (WARN_ON(kiov
[seg
].iov_len
!= PAGE_SIZE
))
312 pages
[seg
] = kmap_to_page(kiov
[seg
].iov_base
);
313 page_cache_get(pages
[seg
]);
318 EXPORT_SYMBOL_GPL(get_kernel_pages
);
321 * get_kernel_page() - pin a kernel page in memory
322 * @start: starting kernel address
323 * @write: pinning for read/write, currently ignored
324 * @pages: array that receives pointer to the page pinned.
325 * Must be at least nr_segs long.
327 * Returns 1 if page is pinned. If the page was not pinned, returns
328 * -errno. The page returned must be released with a put_page() call
329 * when it is finished with.
331 int get_kernel_page(unsigned long start
, int write
, struct page
**pages
)
333 const struct kvec kiov
= {
334 .iov_base
= (void *)start
,
338 return get_kernel_pages(&kiov
, 1, write
, pages
);
340 EXPORT_SYMBOL_GPL(get_kernel_page
);
342 static void pagevec_lru_move_fn(struct pagevec
*pvec
,
343 void (*move_fn
)(struct page
*page
, struct lruvec
*lruvec
, void *arg
),
347 struct zone
*zone
= NULL
;
348 struct lruvec
*lruvec
;
349 unsigned long flags
= 0;
351 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
352 struct page
*page
= pvec
->pages
[i
];
353 struct zone
*pagezone
= page_zone(page
);
355 if (pagezone
!= zone
) {
357 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
359 spin_lock_irqsave(&zone
->lru_lock
, flags
);
362 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
363 (*move_fn
)(page
, lruvec
, arg
);
366 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
367 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
368 pagevec_reinit(pvec
);
371 static void pagevec_move_tail_fn(struct page
*page
, struct lruvec
*lruvec
,
376 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
377 enum lru_list lru
= page_lru_base_type(page
);
378 list_move_tail(&page
->lru
, &lruvec
->lists
[lru
]);
384 * pagevec_move_tail() must be called with IRQ disabled.
385 * Otherwise this may cause nasty races.
387 static void pagevec_move_tail(struct pagevec
*pvec
)
391 pagevec_lru_move_fn(pvec
, pagevec_move_tail_fn
, &pgmoved
);
392 __count_vm_events(PGROTATED
, pgmoved
);
396 * Writeback is about to end against a page which has been marked for immediate
397 * reclaim. If it still appears to be reclaimable, move it to the tail of the
400 void rotate_reclaimable_page(struct page
*page
)
402 if (!PageLocked(page
) && !PageDirty(page
) && !PageActive(page
) &&
403 !PageUnevictable(page
) && PageLRU(page
)) {
404 struct pagevec
*pvec
;
407 page_cache_get(page
);
408 local_irq_save(flags
);
409 pvec
= &__get_cpu_var(lru_rotate_pvecs
);
410 if (!pagevec_add(pvec
, page
))
411 pagevec_move_tail(pvec
);
412 local_irq_restore(flags
);
416 static void update_page_reclaim_stat(struct lruvec
*lruvec
,
417 int file
, int rotated
)
419 struct zone_reclaim_stat
*reclaim_stat
= &lruvec
->reclaim_stat
;
421 reclaim_stat
->recent_scanned
[file
]++;
423 reclaim_stat
->recent_rotated
[file
]++;
426 static void __activate_page(struct page
*page
, struct lruvec
*lruvec
,
429 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
430 int file
= page_is_file_cache(page
);
431 int lru
= page_lru_base_type(page
);
433 del_page_from_lru_list(page
, lruvec
, lru
);
436 add_page_to_lru_list(page
, lruvec
, lru
);
438 __count_vm_event(PGACTIVATE
);
439 update_page_reclaim_stat(lruvec
, file
, 1);
444 static DEFINE_PER_CPU(struct pagevec
, activate_page_pvecs
);
446 static void activate_page_drain(int cpu
)
448 struct pagevec
*pvec
= &per_cpu(activate_page_pvecs
, cpu
);
450 if (pagevec_count(pvec
))
451 pagevec_lru_move_fn(pvec
, __activate_page
, NULL
);
454 void activate_page(struct page
*page
)
456 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
457 struct pagevec
*pvec
= &get_cpu_var(activate_page_pvecs
);
459 page_cache_get(page
);
460 if (!pagevec_add(pvec
, page
))
461 pagevec_lru_move_fn(pvec
, __activate_page
, NULL
);
462 put_cpu_var(activate_page_pvecs
);
467 static inline void activate_page_drain(int cpu
)
471 void activate_page(struct page
*page
)
473 struct zone
*zone
= page_zone(page
);
475 spin_lock_irq(&zone
->lru_lock
);
476 __activate_page(page
, mem_cgroup_page_lruvec(page
, zone
), NULL
);
477 spin_unlock_irq(&zone
->lru_lock
);
482 * Mark a page as having seen activity.
484 * inactive,unreferenced -> inactive,referenced
485 * inactive,referenced -> active,unreferenced
486 * active,unreferenced -> active,referenced
488 void mark_page_accessed(struct page
*page
)
490 if (!PageActive(page
) && !PageUnevictable(page
) &&
491 PageReferenced(page
) && PageLRU(page
)) {
493 ClearPageReferenced(page
);
494 } else if (!PageReferenced(page
)) {
495 SetPageReferenced(page
);
498 EXPORT_SYMBOL(mark_page_accessed
);
501 * Order of operations is important: flush the pagevec when it's already
502 * full, not when adding the last page, to make sure that last page is
503 * not added to the LRU directly when passed to this function. Because
504 * mark_page_accessed() (called after this when writing) only activates
505 * pages that are on the LRU, linear writes in subpage chunks would see
506 * every PAGEVEC_SIZE page activated, which is unexpected.
508 void __lru_cache_add(struct page
*page
, enum lru_list lru
)
510 struct pagevec
*pvec
= &get_cpu_var(lru_add_pvecs
)[lru
];
512 page_cache_get(page
);
513 if (!pagevec_space(pvec
))
514 __pagevec_lru_add(pvec
, lru
);
515 pagevec_add(pvec
, page
);
516 put_cpu_var(lru_add_pvecs
);
518 EXPORT_SYMBOL(__lru_cache_add
);
521 * lru_cache_add_lru - add a page to a page list
522 * @page: the page to be added to the LRU.
523 * @lru: the LRU list to which the page is added.
525 void lru_cache_add_lru(struct page
*page
, enum lru_list lru
)
527 if (PageActive(page
)) {
528 VM_BUG_ON(PageUnevictable(page
));
529 ClearPageActive(page
);
530 } else if (PageUnevictable(page
)) {
531 VM_BUG_ON(PageActive(page
));
532 ClearPageUnevictable(page
);
535 VM_BUG_ON(PageLRU(page
) || PageActive(page
) || PageUnevictable(page
));
536 __lru_cache_add(page
, lru
);
540 * add_page_to_unevictable_list - add a page to the unevictable list
541 * @page: the page to be added to the unevictable list
543 * Add page directly to its zone's unevictable list. To avoid races with
544 * tasks that might be making the page evictable, through eg. munlock,
545 * munmap or exit, while it's not on the lru, we want to add the page
546 * while it's locked or otherwise "invisible" to other tasks. This is
547 * difficult to do when using the pagevec cache, so bypass that.
549 void add_page_to_unevictable_list(struct page
*page
)
551 struct zone
*zone
= page_zone(page
);
552 struct lruvec
*lruvec
;
554 spin_lock_irq(&zone
->lru_lock
);
555 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
556 SetPageUnevictable(page
);
558 add_page_to_lru_list(page
, lruvec
, LRU_UNEVICTABLE
);
559 spin_unlock_irq(&zone
->lru_lock
);
563 * If the page can not be invalidated, it is moved to the
564 * inactive list to speed up its reclaim. It is moved to the
565 * head of the list, rather than the tail, to give the flusher
566 * threads some time to write it out, as this is much more
567 * effective than the single-page writeout from reclaim.
569 * If the page isn't page_mapped and dirty/writeback, the page
570 * could reclaim asap using PG_reclaim.
572 * 1. active, mapped page -> none
573 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
574 * 3. inactive, mapped page -> none
575 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
576 * 5. inactive, clean -> inactive, tail
579 * In 4, why it moves inactive's head, the VM expects the page would
580 * be write it out by flusher threads as this is much more effective
581 * than the single-page writeout from reclaim.
583 static void lru_deactivate_fn(struct page
*page
, struct lruvec
*lruvec
,
592 if (PageUnevictable(page
))
595 /* Some processes are using the page */
596 if (page_mapped(page
))
599 active
= PageActive(page
);
600 file
= page_is_file_cache(page
);
601 lru
= page_lru_base_type(page
);
603 del_page_from_lru_list(page
, lruvec
, lru
+ active
);
604 ClearPageActive(page
);
605 ClearPageReferenced(page
);
606 add_page_to_lru_list(page
, lruvec
, lru
);
608 if (PageWriteback(page
) || PageDirty(page
)) {
610 * PG_reclaim could be raced with end_page_writeback
611 * It can make readahead confusing. But race window
612 * is _really_ small and it's non-critical problem.
614 SetPageReclaim(page
);
617 * The page's writeback ends up during pagevec
618 * We moves tha page into tail of inactive.
620 list_move_tail(&page
->lru
, &lruvec
->lists
[lru
]);
621 __count_vm_event(PGROTATED
);
625 __count_vm_event(PGDEACTIVATE
);
626 update_page_reclaim_stat(lruvec
, file
, 0);
630 * Drain pages out of the cpu's pagevecs.
631 * Either "cpu" is the current CPU, and preemption has already been
632 * disabled; or "cpu" is being hot-unplugged, and is already dead.
634 void lru_add_drain_cpu(int cpu
)
636 struct pagevec
*pvecs
= per_cpu(lru_add_pvecs
, cpu
);
637 struct pagevec
*pvec
;
641 pvec
= &pvecs
[lru
- LRU_BASE
];
642 if (pagevec_count(pvec
))
643 __pagevec_lru_add(pvec
, lru
);
646 pvec
= &per_cpu(lru_rotate_pvecs
, cpu
);
647 if (pagevec_count(pvec
)) {
650 /* No harm done if a racing interrupt already did this */
651 local_irq_save(flags
);
652 pagevec_move_tail(pvec
);
653 local_irq_restore(flags
);
656 pvec
= &per_cpu(lru_deactivate_pvecs
, cpu
);
657 if (pagevec_count(pvec
))
658 pagevec_lru_move_fn(pvec
, lru_deactivate_fn
, NULL
);
660 activate_page_drain(cpu
);
664 * deactivate_page - forcefully deactivate a page
665 * @page: page to deactivate
667 * This function hints the VM that @page is a good reclaim candidate,
668 * for example if its invalidation fails due to the page being dirty
669 * or under writeback.
671 void deactivate_page(struct page
*page
)
674 * In a workload with many unevictable page such as mprotect, unevictable
675 * page deactivation for accelerating reclaim is pointless.
677 if (PageUnevictable(page
))
680 if (likely(get_page_unless_zero(page
))) {
681 struct pagevec
*pvec
= &get_cpu_var(lru_deactivate_pvecs
);
683 if (!pagevec_add(pvec
, page
))
684 pagevec_lru_move_fn(pvec
, lru_deactivate_fn
, NULL
);
685 put_cpu_var(lru_deactivate_pvecs
);
689 void lru_add_drain(void)
691 lru_add_drain_cpu(get_cpu());
695 static void lru_add_drain_per_cpu(struct work_struct
*dummy
)
701 * Returns 0 for success
703 int lru_add_drain_all(void)
705 return schedule_on_each_cpu(lru_add_drain_per_cpu
);
709 * Batched page_cache_release(). Decrement the reference count on all the
710 * passed pages. If it fell to zero then remove the page from the LRU and
713 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
714 * for the remainder of the operation.
716 * The locking in this function is against shrink_inactive_list(): we recheck
717 * the page count inside the lock to see whether shrink_inactive_list()
718 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
721 void release_pages(struct page
**pages
, int nr
, int cold
)
724 LIST_HEAD(pages_to_free
);
725 struct zone
*zone
= NULL
;
726 struct lruvec
*lruvec
;
727 unsigned long uninitialized_var(flags
);
729 for (i
= 0; i
< nr
; i
++) {
730 struct page
*page
= pages
[i
];
732 if (unlikely(PageCompound(page
))) {
734 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
737 put_compound_page(page
);
741 if (!put_page_testzero(page
))
745 struct zone
*pagezone
= page_zone(page
);
747 if (pagezone
!= zone
) {
749 spin_unlock_irqrestore(&zone
->lru_lock
,
752 spin_lock_irqsave(&zone
->lru_lock
, flags
);
755 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
756 VM_BUG_ON(!PageLRU(page
));
757 __ClearPageLRU(page
);
758 del_page_from_lru_list(page
, lruvec
, page_off_lru(page
));
761 list_add(&page
->lru
, &pages_to_free
);
764 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
766 free_hot_cold_page_list(&pages_to_free
, cold
);
768 EXPORT_SYMBOL(release_pages
);
771 * The pages which we're about to release may be in the deferred lru-addition
772 * queues. That would prevent them from really being freed right now. That's
773 * OK from a correctness point of view but is inefficient - those pages may be
774 * cache-warm and we want to give them back to the page allocator ASAP.
776 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
777 * and __pagevec_lru_add_active() call release_pages() directly to avoid
780 void __pagevec_release(struct pagevec
*pvec
)
783 release_pages(pvec
->pages
, pagevec_count(pvec
), pvec
->cold
);
784 pagevec_reinit(pvec
);
786 EXPORT_SYMBOL(__pagevec_release
);
788 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
789 /* used by __split_huge_page_refcount() */
790 void lru_add_page_tail(struct page
*page
, struct page
*page_tail
,
791 struct lruvec
*lruvec
, struct list_head
*list
)
793 int uninitialized_var(active
);
797 VM_BUG_ON(!PageHead(page
));
798 VM_BUG_ON(PageCompound(page_tail
));
799 VM_BUG_ON(PageLRU(page_tail
));
800 VM_BUG_ON(NR_CPUS
!= 1 &&
801 !spin_is_locked(&lruvec_zone(lruvec
)->lru_lock
));
804 SetPageLRU(page_tail
);
806 if (page_evictable(page_tail
)) {
807 if (PageActive(page
)) {
808 SetPageActive(page_tail
);
810 lru
= LRU_ACTIVE_ANON
;
813 lru
= LRU_INACTIVE_ANON
;
816 SetPageUnevictable(page_tail
);
817 lru
= LRU_UNEVICTABLE
;
820 if (likely(PageLRU(page
)))
821 list_add_tail(&page_tail
->lru
, &page
->lru
);
823 /* page reclaim is reclaiming a huge page */
825 list_add_tail(&page_tail
->lru
, list
);
827 struct list_head
*list_head
;
829 * Head page has not yet been counted, as an hpage,
830 * so we must account for each subpage individually.
832 * Use the standard add function to put page_tail on the list,
833 * but then correct its position so they all end up in order.
835 add_page_to_lru_list(page_tail
, lruvec
, lru
);
836 list_head
= page_tail
->lru
.prev
;
837 list_move_tail(&page_tail
->lru
, list_head
);
840 if (!PageUnevictable(page
))
841 update_page_reclaim_stat(lruvec
, file
, active
);
843 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
845 static void __pagevec_lru_add_fn(struct page
*page
, struct lruvec
*lruvec
,
848 enum lru_list lru
= (enum lru_list
)arg
;
849 int file
= is_file_lru(lru
);
850 int active
= is_active_lru(lru
);
852 VM_BUG_ON(PageActive(page
));
853 VM_BUG_ON(PageUnevictable(page
));
854 VM_BUG_ON(PageLRU(page
));
859 add_page_to_lru_list(page
, lruvec
, lru
);
860 update_page_reclaim_stat(lruvec
, file
, active
);
864 * Add the passed pages to the LRU, then drop the caller's refcount
865 * on them. Reinitialises the caller's pagevec.
867 void __pagevec_lru_add(struct pagevec
*pvec
, enum lru_list lru
)
869 VM_BUG_ON(is_unevictable_lru(lru
));
871 pagevec_lru_move_fn(pvec
, __pagevec_lru_add_fn
, (void *)lru
);
873 EXPORT_SYMBOL(__pagevec_lru_add
);
876 * pagevec_lookup - gang pagecache lookup
877 * @pvec: Where the resulting pages are placed
878 * @mapping: The address_space to search
879 * @start: The starting page index
880 * @nr_pages: The maximum number of pages
882 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
883 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
884 * reference against the pages in @pvec.
886 * The search returns a group of mapping-contiguous pages with ascending
887 * indexes. There may be holes in the indices due to not-present pages.
889 * pagevec_lookup() returns the number of pages which were found.
891 unsigned pagevec_lookup(struct pagevec
*pvec
, struct address_space
*mapping
,
892 pgoff_t start
, unsigned nr_pages
)
894 pvec
->nr
= find_get_pages(mapping
, start
, nr_pages
, pvec
->pages
);
895 return pagevec_count(pvec
);
897 EXPORT_SYMBOL(pagevec_lookup
);
899 unsigned pagevec_lookup_tag(struct pagevec
*pvec
, struct address_space
*mapping
,
900 pgoff_t
*index
, int tag
, unsigned nr_pages
)
902 pvec
->nr
= find_get_pages_tag(mapping
, index
, tag
,
903 nr_pages
, pvec
->pages
);
904 return pagevec_count(pvec
);
906 EXPORT_SYMBOL(pagevec_lookup_tag
);
909 * Perform any setup for the swap system
911 void __init
swap_setup(void)
913 unsigned long megs
= totalram_pages
>> (20 - PAGE_SHIFT
);
917 bdi_init(swapper_spaces
[0].backing_dev_info
);
918 for (i
= 0; i
< MAX_SWAPFILES
; i
++) {
919 spin_lock_init(&swapper_spaces
[i
].tree_lock
);
920 INIT_LIST_HEAD(&swapper_spaces
[i
].i_mmap_nonlinear
);
924 /* Use a smaller cluster for small-memory machines */
926 page_cluster
= 0; // disable swap read-ahead
934 * Right now other parts of the system means that we
935 * _really_ don't want to cluster much more