]>
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/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 #define CREATE_TRACE_POINTS
39 #include <trace/events/pagemap.h>
41 /* How many pages do we try to swap or page in/out together? */
44 static DEFINE_PER_CPU(struct pagevec
, lru_add_pvec
);
45 static DEFINE_PER_CPU(struct pagevec
, lru_rotate_pvecs
);
46 static DEFINE_PER_CPU(struct pagevec
, lru_deactivate_pvecs
);
49 * This path almost never happens for VM activity - pages are normally
50 * freed via pagevecs. But it gets used by networking.
52 static void __page_cache_release(struct page
*page
)
55 struct zone
*zone
= page_zone(page
);
56 struct lruvec
*lruvec
;
59 spin_lock_irqsave(&zone
->lru_lock
, flags
);
60 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
61 VM_BUG_ON(!PageLRU(page
));
63 del_page_from_lru_list(page
, lruvec
, page_off_lru(page
));
64 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
68 static void __put_single_page(struct page
*page
)
70 __page_cache_release(page
);
71 free_hot_cold_page(page
, 0);
74 static void __put_compound_page(struct page
*page
)
76 compound_page_dtor
*dtor
;
78 __page_cache_release(page
);
79 dtor
= get_compound_page_dtor(page
);
83 static void put_compound_page(struct page
*page
)
86 * hugetlbfs pages cannot be split from under us. If this is a
87 * hugetlbfs page, check refcount on head page and release the page if
88 * the refcount becomes zero.
91 page
= compound_head(page
);
92 if (put_page_testzero(page
))
93 __put_compound_page(page
);
98 if (unlikely(PageTail(page
))) {
99 /* __split_huge_page_refcount can run under us */
100 struct page
*page_head
= compound_trans_head(page
);
102 if (likely(page
!= page_head
&&
103 get_page_unless_zero(page_head
))) {
107 * THP can not break up slab pages so avoid taking
108 * compound_lock(). Slab performs non-atomic bit ops
109 * on page->flags for better performance. In particular
110 * slab_unlock() in slub used to be a hot path. It is
111 * still hot on arches that do not support
112 * this_cpu_cmpxchg_double().
114 if (PageSlab(page_head
)) {
115 if (PageTail(page
)) {
116 if (put_page_testzero(page_head
))
119 atomic_dec(&page
->_mapcount
);
125 * page_head wasn't a dangling pointer but it
126 * may not be a head page anymore by the time
127 * we obtain the lock. That is ok as long as it
128 * can't be freed from under us.
130 flags
= compound_lock_irqsave(page_head
);
131 if (unlikely(!PageTail(page
))) {
132 /* __split_huge_page_refcount run before us */
133 compound_unlock_irqrestore(page_head
, flags
);
135 if (put_page_testzero(page_head
))
136 __put_single_page(page_head
);
138 if (put_page_testzero(page
))
139 __put_single_page(page
);
142 VM_BUG_ON(page_head
!= page
->first_page
);
144 * We can release the refcount taken by
145 * get_page_unless_zero() now that
146 * __split_huge_page_refcount() is blocked on
149 if (put_page_testzero(page_head
))
151 /* __split_huge_page_refcount will wait now */
152 VM_BUG_ON(page_mapcount(page
) <= 0);
153 atomic_dec(&page
->_mapcount
);
154 VM_BUG_ON(atomic_read(&page_head
->_count
) <= 0);
155 VM_BUG_ON(atomic_read(&page
->_count
) != 0);
156 compound_unlock_irqrestore(page_head
, flags
);
159 if (put_page_testzero(page_head
)) {
160 if (PageHead(page_head
))
161 __put_compound_page(page_head
);
163 __put_single_page(page_head
);
166 /* page_head is a dangling pointer */
167 VM_BUG_ON(PageTail(page
));
170 } else if (put_page_testzero(page
)) {
172 __put_compound_page(page
);
174 __put_single_page(page
);
178 void put_page(struct page
*page
)
180 if (unlikely(PageCompound(page
)))
181 put_compound_page(page
);
182 else if (put_page_testzero(page
))
183 __put_single_page(page
);
185 EXPORT_SYMBOL(put_page
);
188 * This function is exported but must not be called by anything other
189 * than get_page(). It implements the slow path of get_page().
191 bool __get_page_tail(struct page
*page
)
194 * This takes care of get_page() if run on a tail page
195 * returned by one of the get_user_pages/follow_page variants.
196 * get_user_pages/follow_page itself doesn't need the compound
197 * lock because it runs __get_page_tail_foll() under the
198 * proper PT lock that already serializes against
202 struct page
*page_head
;
205 * If this is a hugetlbfs page it cannot be split under us. Simply
206 * increment refcount for the head page.
208 if (PageHuge(page
)) {
209 page_head
= compound_head(page
);
210 atomic_inc(&page_head
->_count
);
215 page_head
= compound_trans_head(page
);
216 if (likely(page
!= page_head
&&
217 get_page_unless_zero(page_head
))) {
219 /* Ref to put_compound_page() comment. */
220 if (PageSlab(page_head
)) {
221 if (likely(PageTail(page
))) {
222 __get_page_tail_foll(page
, false);
231 * page_head wasn't a dangling pointer but it
232 * may not be a head page anymore by the time
233 * we obtain the lock. That is ok as long as it
234 * can't be freed from under us.
236 flags
= compound_lock_irqsave(page_head
);
237 /* here __split_huge_page_refcount won't run anymore */
238 if (likely(PageTail(page
))) {
239 __get_page_tail_foll(page
, false);
242 compound_unlock_irqrestore(page_head
, flags
);
249 EXPORT_SYMBOL(__get_page_tail
);
252 * put_pages_list() - release a list of pages
253 * @pages: list of pages threaded on page->lru
255 * Release a list of pages which are strung together on page.lru. Currently
256 * used by read_cache_pages() and related error recovery code.
258 void put_pages_list(struct list_head
*pages
)
260 while (!list_empty(pages
)) {
263 victim
= list_entry(pages
->prev
, struct page
, lru
);
264 list_del(&victim
->lru
);
265 page_cache_release(victim
);
268 EXPORT_SYMBOL(put_pages_list
);
271 * get_kernel_pages() - pin kernel pages in memory
272 * @kiov: An array of struct kvec structures
273 * @nr_segs: number of segments to pin
274 * @write: pinning for read/write, currently ignored
275 * @pages: array that receives pointers to the pages pinned.
276 * Should be at least nr_segs long.
278 * Returns number of pages pinned. This may be fewer than the number
279 * requested. If nr_pages is 0 or negative, returns 0. If no pages
280 * were pinned, returns -errno. Each page returned must be released
281 * with a put_page() call when it is finished with.
283 int get_kernel_pages(const struct kvec
*kiov
, int nr_segs
, int write
,
288 for (seg
= 0; seg
< nr_segs
; seg
++) {
289 if (WARN_ON(kiov
[seg
].iov_len
!= PAGE_SIZE
))
292 pages
[seg
] = kmap_to_page(kiov
[seg
].iov_base
);
293 page_cache_get(pages
[seg
]);
298 EXPORT_SYMBOL_GPL(get_kernel_pages
);
301 * get_kernel_page() - pin a kernel page in memory
302 * @start: starting kernel address
303 * @write: pinning for read/write, currently ignored
304 * @pages: array that receives pointer to the page pinned.
305 * Must be at least nr_segs long.
307 * Returns 1 if page is pinned. If the page was not pinned, returns
308 * -errno. The page returned must be released with a put_page() call
309 * when it is finished with.
311 int get_kernel_page(unsigned long start
, int write
, struct page
**pages
)
313 const struct kvec kiov
= {
314 .iov_base
= (void *)start
,
318 return get_kernel_pages(&kiov
, 1, write
, pages
);
320 EXPORT_SYMBOL_GPL(get_kernel_page
);
322 static void pagevec_lru_move_fn(struct pagevec
*pvec
,
323 void (*move_fn
)(struct page
*page
, struct lruvec
*lruvec
, void *arg
),
327 struct zone
*zone
= NULL
;
328 struct lruvec
*lruvec
;
329 unsigned long flags
= 0;
331 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
332 struct page
*page
= pvec
->pages
[i
];
333 struct zone
*pagezone
= page_zone(page
);
335 if (pagezone
!= zone
) {
337 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
339 spin_lock_irqsave(&zone
->lru_lock
, flags
);
342 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
343 (*move_fn
)(page
, lruvec
, arg
);
346 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
347 release_pages(pvec
->pages
, pvec
->nr
, pvec
->cold
);
348 pagevec_reinit(pvec
);
351 static void pagevec_move_tail_fn(struct page
*page
, struct lruvec
*lruvec
,
356 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
357 enum lru_list lru
= page_lru_base_type(page
);
358 list_move_tail(&page
->lru
, &lruvec
->lists
[lru
]);
364 * pagevec_move_tail() must be called with IRQ disabled.
365 * Otherwise this may cause nasty races.
367 static void pagevec_move_tail(struct pagevec
*pvec
)
371 pagevec_lru_move_fn(pvec
, pagevec_move_tail_fn
, &pgmoved
);
372 __count_vm_events(PGROTATED
, pgmoved
);
376 * Writeback is about to end against a page which has been marked for immediate
377 * reclaim. If it still appears to be reclaimable, move it to the tail of the
380 void rotate_reclaimable_page(struct page
*page
)
382 if (!PageLocked(page
) && !PageDirty(page
) && !PageActive(page
) &&
383 !PageUnevictable(page
) && PageLRU(page
)) {
384 struct pagevec
*pvec
;
387 page_cache_get(page
);
388 local_irq_save(flags
);
389 pvec
= &__get_cpu_var(lru_rotate_pvecs
);
390 if (!pagevec_add(pvec
, page
))
391 pagevec_move_tail(pvec
);
392 local_irq_restore(flags
);
396 static void update_page_reclaim_stat(struct lruvec
*lruvec
,
397 int file
, int rotated
)
399 struct zone_reclaim_stat
*reclaim_stat
= &lruvec
->reclaim_stat
;
401 reclaim_stat
->recent_scanned
[file
]++;
403 reclaim_stat
->recent_rotated
[file
]++;
406 static void __activate_page(struct page
*page
, struct lruvec
*lruvec
,
409 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
410 int file
= page_is_file_cache(page
);
411 int lru
= page_lru_base_type(page
);
413 del_page_from_lru_list(page
, lruvec
, lru
);
416 add_page_to_lru_list(page
, lruvec
, lru
);
417 trace_mm_lru_activate(page
, page_to_pfn(page
));
419 __count_vm_event(PGACTIVATE
);
420 update_page_reclaim_stat(lruvec
, file
, 1);
425 static DEFINE_PER_CPU(struct pagevec
, activate_page_pvecs
);
427 static void activate_page_drain(int cpu
)
429 struct pagevec
*pvec
= &per_cpu(activate_page_pvecs
, cpu
);
431 if (pagevec_count(pvec
))
432 pagevec_lru_move_fn(pvec
, __activate_page
, NULL
);
435 void activate_page(struct page
*page
)
437 if (PageLRU(page
) && !PageActive(page
) && !PageUnevictable(page
)) {
438 struct pagevec
*pvec
= &get_cpu_var(activate_page_pvecs
);
440 page_cache_get(page
);
441 if (!pagevec_add(pvec
, page
))
442 pagevec_lru_move_fn(pvec
, __activate_page
, NULL
);
443 put_cpu_var(activate_page_pvecs
);
448 static inline void activate_page_drain(int cpu
)
452 void activate_page(struct page
*page
)
454 struct zone
*zone
= page_zone(page
);
456 spin_lock_irq(&zone
->lru_lock
);
457 __activate_page(page
, mem_cgroup_page_lruvec(page
, zone
), NULL
);
458 spin_unlock_irq(&zone
->lru_lock
);
462 static void __lru_cache_activate_page(struct page
*page
)
464 struct pagevec
*pvec
= &get_cpu_var(lru_add_pvec
);
468 * Search backwards on the optimistic assumption that the page being
469 * activated has just been added to this pagevec. Note that only
470 * the local pagevec is examined as a !PageLRU page could be in the
471 * process of being released, reclaimed, migrated or on a remote
472 * pagevec that is currently being drained. Furthermore, marking
473 * a remote pagevec's page PageActive potentially hits a race where
474 * a page is marked PageActive just after it is added to the inactive
475 * list causing accounting errors and BUG_ON checks to trigger.
477 for (i
= pagevec_count(pvec
) - 1; i
>= 0; i
--) {
478 struct page
*pagevec_page
= pvec
->pages
[i
];
480 if (pagevec_page
== page
) {
486 put_cpu_var(lru_add_pvec
);
490 * Mark a page as having seen activity.
492 * inactive,unreferenced -> inactive,referenced
493 * inactive,referenced -> active,unreferenced
494 * active,unreferenced -> active,referenced
496 void mark_page_accessed(struct page
*page
)
498 if (!PageActive(page
) && !PageUnevictable(page
) &&
499 PageReferenced(page
)) {
502 * If the page is on the LRU, queue it for activation via
503 * activate_page_pvecs. Otherwise, assume the page is on a
504 * pagevec, mark it active and it'll be moved to the active
505 * LRU on the next drain.
510 __lru_cache_activate_page(page
);
511 ClearPageReferenced(page
);
512 } else if (!PageReferenced(page
)) {
513 SetPageReferenced(page
);
516 EXPORT_SYMBOL(mark_page_accessed
);
519 * Queue the page for addition to the LRU via pagevec. The decision on whether
520 * to add the page to the [in]active [file|anon] list is deferred until the
521 * pagevec is drained. This gives a chance for the caller of __lru_cache_add()
522 * have the page added to the active list using mark_page_accessed().
524 void __lru_cache_add(struct page
*page
)
526 struct pagevec
*pvec
= &get_cpu_var(lru_add_pvec
);
528 page_cache_get(page
);
529 if (!pagevec_space(pvec
))
530 __pagevec_lru_add(pvec
);
531 pagevec_add(pvec
, page
);
532 put_cpu_var(lru_add_pvec
);
534 EXPORT_SYMBOL(__lru_cache_add
);
537 * lru_cache_add - add a page to a page list
538 * @page: the page to be added to the LRU.
540 void lru_cache_add(struct page
*page
)
542 VM_BUG_ON(PageActive(page
) && PageUnevictable(page
));
543 VM_BUG_ON(PageLRU(page
));
544 __lru_cache_add(page
);
548 * add_page_to_unevictable_list - add a page to the unevictable list
549 * @page: the page to be added to the unevictable list
551 * Add page directly to its zone's unevictable list. To avoid races with
552 * tasks that might be making the page evictable, through eg. munlock,
553 * munmap or exit, while it's not on the lru, we want to add the page
554 * while it's locked or otherwise "invisible" to other tasks. This is
555 * difficult to do when using the pagevec cache, so bypass that.
557 void add_page_to_unevictable_list(struct page
*page
)
559 struct zone
*zone
= page_zone(page
);
560 struct lruvec
*lruvec
;
562 spin_lock_irq(&zone
->lru_lock
);
563 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
564 ClearPageActive(page
);
565 SetPageUnevictable(page
);
567 add_page_to_lru_list(page
, lruvec
, LRU_UNEVICTABLE
);
568 spin_unlock_irq(&zone
->lru_lock
);
572 * If the page can not be invalidated, it is moved to the
573 * inactive list to speed up its reclaim. It is moved to the
574 * head of the list, rather than the tail, to give the flusher
575 * threads some time to write it out, as this is much more
576 * effective than the single-page writeout from reclaim.
578 * If the page isn't page_mapped and dirty/writeback, the page
579 * could reclaim asap using PG_reclaim.
581 * 1. active, mapped page -> none
582 * 2. active, dirty/writeback page -> inactive, head, PG_reclaim
583 * 3. inactive, mapped page -> none
584 * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim
585 * 5. inactive, clean -> inactive, tail
588 * In 4, why it moves inactive's head, the VM expects the page would
589 * be write it out by flusher threads as this is much more effective
590 * than the single-page writeout from reclaim.
592 static void lru_deactivate_fn(struct page
*page
, struct lruvec
*lruvec
,
601 if (PageUnevictable(page
))
604 /* Some processes are using the page */
605 if (page_mapped(page
))
608 active
= PageActive(page
);
609 file
= page_is_file_cache(page
);
610 lru
= page_lru_base_type(page
);
612 del_page_from_lru_list(page
, lruvec
, lru
+ active
);
613 ClearPageActive(page
);
614 ClearPageReferenced(page
);
615 add_page_to_lru_list(page
, lruvec
, lru
);
617 if (PageWriteback(page
) || PageDirty(page
)) {
619 * PG_reclaim could be raced with end_page_writeback
620 * It can make readahead confusing. But race window
621 * is _really_ small and it's non-critical problem.
623 SetPageReclaim(page
);
626 * The page's writeback ends up during pagevec
627 * We moves tha page into tail of inactive.
629 list_move_tail(&page
->lru
, &lruvec
->lists
[lru
]);
630 __count_vm_event(PGROTATED
);
634 __count_vm_event(PGDEACTIVATE
);
635 update_page_reclaim_stat(lruvec
, file
, 0);
639 * Drain pages out of the cpu's pagevecs.
640 * Either "cpu" is the current CPU, and preemption has already been
641 * disabled; or "cpu" is being hot-unplugged, and is already dead.
643 void lru_add_drain_cpu(int cpu
)
645 struct pagevec
*pvec
= &per_cpu(lru_add_pvec
, cpu
);
647 if (pagevec_count(pvec
))
648 __pagevec_lru_add(pvec
);
650 pvec
= &per_cpu(lru_rotate_pvecs
, cpu
);
651 if (pagevec_count(pvec
)) {
654 /* No harm done if a racing interrupt already did this */
655 local_irq_save(flags
);
656 pagevec_move_tail(pvec
);
657 local_irq_restore(flags
);
660 pvec
= &per_cpu(lru_deactivate_pvecs
, cpu
);
661 if (pagevec_count(pvec
))
662 pagevec_lru_move_fn(pvec
, lru_deactivate_fn
, NULL
);
664 activate_page_drain(cpu
);
668 * deactivate_page - forcefully deactivate a page
669 * @page: page to deactivate
671 * This function hints the VM that @page is a good reclaim candidate,
672 * for example if its invalidation fails due to the page being dirty
673 * or under writeback.
675 void deactivate_page(struct page
*page
)
678 * In a workload with many unevictable page such as mprotect, unevictable
679 * page deactivation for accelerating reclaim is pointless.
681 if (PageUnevictable(page
))
684 if (likely(get_page_unless_zero(page
))) {
685 struct pagevec
*pvec
= &get_cpu_var(lru_deactivate_pvecs
);
687 if (!pagevec_add(pvec
, page
))
688 pagevec_lru_move_fn(pvec
, lru_deactivate_fn
, NULL
);
689 put_cpu_var(lru_deactivate_pvecs
);
693 void lru_add_drain(void)
695 lru_add_drain_cpu(get_cpu());
699 static void lru_add_drain_per_cpu(struct work_struct
*dummy
)
705 * Returns 0 for success
707 int lru_add_drain_all(void)
709 return schedule_on_each_cpu(lru_add_drain_per_cpu
);
713 * Batched page_cache_release(). Decrement the reference count on all the
714 * passed pages. If it fell to zero then remove the page from the LRU and
717 * Avoid taking zone->lru_lock if possible, but if it is taken, retain it
718 * for the remainder of the operation.
720 * The locking in this function is against shrink_inactive_list(): we recheck
721 * the page count inside the lock to see whether shrink_inactive_list()
722 * grabbed the page via the LRU. If it did, give up: shrink_inactive_list()
725 void release_pages(struct page
**pages
, int nr
, int cold
)
728 LIST_HEAD(pages_to_free
);
729 struct zone
*zone
= NULL
;
730 struct lruvec
*lruvec
;
731 unsigned long uninitialized_var(flags
);
733 for (i
= 0; i
< nr
; i
++) {
734 struct page
*page
= pages
[i
];
736 if (unlikely(PageCompound(page
))) {
738 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
741 put_compound_page(page
);
745 if (!put_page_testzero(page
))
749 struct zone
*pagezone
= page_zone(page
);
751 if (pagezone
!= zone
) {
753 spin_unlock_irqrestore(&zone
->lru_lock
,
756 spin_lock_irqsave(&zone
->lru_lock
, flags
);
759 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
760 VM_BUG_ON(!PageLRU(page
));
761 __ClearPageLRU(page
);
762 del_page_from_lru_list(page
, lruvec
, page_off_lru(page
));
765 /* Clear Active bit in case of parallel mark_page_accessed */
766 ClearPageActive(page
);
768 list_add(&page
->lru
, &pages_to_free
);
771 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
773 free_hot_cold_page_list(&pages_to_free
, cold
);
775 EXPORT_SYMBOL(release_pages
);
778 * The pages which we're about to release may be in the deferred lru-addition
779 * queues. That would prevent them from really being freed right now. That's
780 * OK from a correctness point of view but is inefficient - those pages may be
781 * cache-warm and we want to give them back to the page allocator ASAP.
783 * So __pagevec_release() will drain those queues here. __pagevec_lru_add()
784 * and __pagevec_lru_add_active() call release_pages() directly to avoid
787 void __pagevec_release(struct pagevec
*pvec
)
790 release_pages(pvec
->pages
, pagevec_count(pvec
), pvec
->cold
);
791 pagevec_reinit(pvec
);
793 EXPORT_SYMBOL(__pagevec_release
);
795 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
796 /* used by __split_huge_page_refcount() */
797 void lru_add_page_tail(struct page
*page
, struct page
*page_tail
,
798 struct lruvec
*lruvec
, struct list_head
*list
)
802 VM_BUG_ON(!PageHead(page
));
803 VM_BUG_ON(PageCompound(page_tail
));
804 VM_BUG_ON(PageLRU(page_tail
));
805 VM_BUG_ON(NR_CPUS
!= 1 &&
806 !spin_is_locked(&lruvec_zone(lruvec
)->lru_lock
));
809 SetPageLRU(page_tail
);
811 if (likely(PageLRU(page
)))
812 list_add_tail(&page_tail
->lru
, &page
->lru
);
814 /* page reclaim is reclaiming a huge page */
816 list_add_tail(&page_tail
->lru
, list
);
818 struct list_head
*list_head
;
820 * Head page has not yet been counted, as an hpage,
821 * so we must account for each subpage individually.
823 * Use the standard add function to put page_tail on the list,
824 * but then correct its position so they all end up in order.
826 add_page_to_lru_list(page_tail
, lruvec
, page_lru(page_tail
));
827 list_head
= page_tail
->lru
.prev
;
828 list_move_tail(&page_tail
->lru
, list_head
);
831 if (!PageUnevictable(page
))
832 update_page_reclaim_stat(lruvec
, file
, PageActive(page_tail
));
834 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
836 static void __pagevec_lru_add_fn(struct page
*page
, struct lruvec
*lruvec
,
839 int file
= page_is_file_cache(page
);
840 int active
= PageActive(page
);
841 enum lru_list lru
= page_lru(page
);
843 VM_BUG_ON(PageLRU(page
));
846 add_page_to_lru_list(page
, lruvec
, lru
);
847 update_page_reclaim_stat(lruvec
, file
, active
);
848 trace_mm_lru_insertion(page
, page_to_pfn(page
), lru
, trace_pagemap_flags(page
));
852 * Add the passed pages to the LRU, then drop the caller's refcount
853 * on them. Reinitialises the caller's pagevec.
855 void __pagevec_lru_add(struct pagevec
*pvec
)
857 pagevec_lru_move_fn(pvec
, __pagevec_lru_add_fn
, NULL
);
859 EXPORT_SYMBOL(__pagevec_lru_add
);
862 * pagevec_lookup - gang pagecache lookup
863 * @pvec: Where the resulting pages are placed
864 * @mapping: The address_space to search
865 * @start: The starting page index
866 * @nr_pages: The maximum number of pages
868 * pagevec_lookup() will search for and return a group of up to @nr_pages pages
869 * in the mapping. The pages are placed in @pvec. pagevec_lookup() takes a
870 * reference against the pages in @pvec.
872 * The search returns a group of mapping-contiguous pages with ascending
873 * indexes. There may be holes in the indices due to not-present pages.
875 * pagevec_lookup() returns the number of pages which were found.
877 unsigned pagevec_lookup(struct pagevec
*pvec
, struct address_space
*mapping
,
878 pgoff_t start
, unsigned nr_pages
)
880 pvec
->nr
= find_get_pages(mapping
, start
, nr_pages
, pvec
->pages
);
881 return pagevec_count(pvec
);
883 EXPORT_SYMBOL(pagevec_lookup
);
885 unsigned pagevec_lookup_tag(struct pagevec
*pvec
, struct address_space
*mapping
,
886 pgoff_t
*index
, int tag
, unsigned nr_pages
)
888 pvec
->nr
= find_get_pages_tag(mapping
, index
, tag
,
889 nr_pages
, pvec
->pages
);
890 return pagevec_count(pvec
);
892 EXPORT_SYMBOL(pagevec_lookup_tag
);
895 * Perform any setup for the swap system
897 void __init
swap_setup(void)
899 unsigned long megs
= totalram_pages
>> (20 - PAGE_SHIFT
);
903 bdi_init(swapper_spaces
[0].backing_dev_info
);
904 for (i
= 0; i
< MAX_SWAPFILES
; i
++) {
905 spin_lock_init(&swapper_spaces
[i
].tree_lock
);
906 INIT_LIST_HEAD(&swapper_spaces
[i
].i_mmap_nonlinear
);
910 /* Use a smaller cluster for small-memory machines */
916 * Right now other parts of the system means that we
917 * _really_ don't want to cluster much more