4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie.
7 * kswapd added: 7.1.96 sct
8 * Removed kswapd_ctl limits, and swap out as many pages as needed
9 * to bring the system back to freepages.high: 2.4.97, Rik van Riel.
10 * Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com).
11 * Multiqueue VM started 5.8.00, Rik van Riel.
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/kernel_stat.h>
18 #include <linux/swap.h>
19 #include <linux/pagemap.h>
20 #include <linux/init.h>
21 #include <linux/highmem.h>
22 #include <linux/vmstat.h>
23 #include <linux/file.h>
24 #include <linux/writeback.h>
25 #include <linux/blkdev.h>
26 #include <linux/buffer_head.h> /* for try_to_release_page(),
27 buffer_heads_over_limit */
28 #include <linux/mm_inline.h>
29 #include <linux/pagevec.h>
30 #include <linux/backing-dev.h>
31 #include <linux/rmap.h>
32 #include <linux/topology.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/notifier.h>
36 #include <linux/rwsem.h>
37 #include <linux/delay.h>
38 #include <linux/kthread.h>
39 #include <linux/freezer.h>
40 #include <linux/memcontrol.h>
41 #include <linux/delayacct.h>
42 #include <linux/sysctl.h>
44 #include <asm/tlbflush.h>
45 #include <asm/div64.h>
47 #include <linux/swapops.h>
52 /* Incremented by the number of inactive pages that were scanned */
53 unsigned long nr_scanned;
55 /* Number of pages freed so far during a call to shrink_zones() */
56 unsigned long nr_reclaimed;
58 /* This context's GFP mask */
63 /* Can pages be swapped as part of reclaim? */
66 /* This context's SWAP_CLUSTER_MAX. If freeing memory for
67 * suspend, we effectively ignore SWAP_CLUSTER_MAX.
68 * In this context, it doesn't matter that we scan the
69 * whole list at once. */
74 int all_unreclaimable;
78 /* Which cgroup do we reclaim from */
79 struct mem_cgroup *mem_cgroup;
81 /* Pluggable isolate pages callback */
82 unsigned long (*isolate_pages)(unsigned long nr, struct list_head *dst,
83 unsigned long *scanned, int order, int mode,
84 struct zone *z, struct mem_cgroup *mem_cont,
85 int active, int file);
88 #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
90 #ifdef ARCH_HAS_PREFETCH
91 #define prefetch_prev_lru_page(_page, _base, _field) \
93 if ((_page)->lru.prev != _base) { \
96 prev = lru_to_page(&(_page->lru)); \
97 prefetch(&prev->_field); \
101 #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
104 #ifdef ARCH_HAS_PREFETCHW
105 #define prefetchw_prev_lru_page(_page, _base, _field) \
107 if ((_page)->lru.prev != _base) { \
110 prev = lru_to_page(&(_page->lru)); \
111 prefetchw(&prev->_field); \
115 #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
119 * From 0 .. 100. Higher means more swappy.
121 int vm_swappiness = 60;
122 long vm_total_pages; /* The total number of pages which the VM controls */
124 static LIST_HEAD(shrinker_list);
125 static DECLARE_RWSEM(shrinker_rwsem);
127 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
128 #define scan_global_lru(sc) (!(sc)->mem_cgroup)
130 #define scan_global_lru(sc) (1)
134 * Add a shrinker callback to be called from the vm
136 void register_shrinker(struct shrinker *shrinker)
139 down_write(&shrinker_rwsem);
140 list_add_tail(&shrinker->list, &shrinker_list);
141 up_write(&shrinker_rwsem);
143 EXPORT_SYMBOL(register_shrinker);
148 void unregister_shrinker(struct shrinker *shrinker)
150 down_write(&shrinker_rwsem);
151 list_del(&shrinker->list);
152 up_write(&shrinker_rwsem);
154 EXPORT_SYMBOL(unregister_shrinker);
156 #define SHRINK_BATCH 128
158 * Call the shrink functions to age shrinkable caches
160 * Here we assume it costs one seek to replace a lru page and that it also
161 * takes a seek to recreate a cache object. With this in mind we age equal
162 * percentages of the lru and ageable caches. This should balance the seeks
163 * generated by these structures.
165 * If the vm encountered mapped pages on the LRU it increase the pressure on
166 * slab to avoid swapping.
168 * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
170 * `lru_pages' represents the number of on-LRU pages in all the zones which
171 * are eligible for the caller's allocation attempt. It is used for balancing
172 * slab reclaim versus page reclaim.
174 * Returns the number of slab objects which we shrunk.
176 unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask,
177 unsigned long lru_pages)
179 struct shrinker *shrinker;
180 unsigned long ret = 0;
183 scanned = SWAP_CLUSTER_MAX;
185 if (!down_read_trylock(&shrinker_rwsem))
186 return 1; /* Assume we'll be able to shrink next time */
188 list_for_each_entry(shrinker, &shrinker_list, list) {
189 unsigned long long delta;
190 unsigned long total_scan;
191 unsigned long max_pass = (*shrinker->shrink)(0, gfp_mask);
193 delta = (4 * scanned) / shrinker->seeks;
195 do_div(delta, lru_pages + 1);
196 shrinker->nr += delta;
197 if (shrinker->nr < 0) {
198 printk(KERN_ERR "%s: nr=%ld\n",
199 __func__, shrinker->nr);
200 shrinker->nr = max_pass;
204 * Avoid risking looping forever due to too large nr value:
205 * never try to free more than twice the estimate number of
208 if (shrinker->nr > max_pass * 2)
209 shrinker->nr = max_pass * 2;
211 total_scan = shrinker->nr;
214 while (total_scan >= SHRINK_BATCH) {
215 long this_scan = SHRINK_BATCH;
219 nr_before = (*shrinker->shrink)(0, gfp_mask);
220 shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask);
221 if (shrink_ret == -1)
223 if (shrink_ret < nr_before)
224 ret += nr_before - shrink_ret;
225 count_vm_events(SLABS_SCANNED, this_scan);
226 total_scan -= this_scan;
231 shrinker->nr += total_scan;
233 up_read(&shrinker_rwsem);
237 /* Called without lock on whether page is mapped, so answer is unstable */
238 static inline int page_mapping_inuse(struct page *page)
240 struct address_space *mapping;
242 /* Page is in somebody's page tables. */
243 if (page_mapped(page))
246 /* Be more reluctant to reclaim swapcache than pagecache */
247 if (PageSwapCache(page))
250 mapping = page_mapping(page);
254 /* File is mmap'd by somebody? */
255 return mapping_mapped(mapping);
258 static inline int is_page_cache_freeable(struct page *page)
260 return page_count(page) - !!PagePrivate(page) == 2;
263 static int may_write_to_queue(struct backing_dev_info *bdi)
265 if (current->flags & PF_SWAPWRITE)
267 if (!bdi_write_congested(bdi))
269 if (bdi == current->backing_dev_info)
275 * We detected a synchronous write error writing a page out. Probably
276 * -ENOSPC. We need to propagate that into the address_space for a subsequent
277 * fsync(), msync() or close().
279 * The tricky part is that after writepage we cannot touch the mapping: nothing
280 * prevents it from being freed up. But we have a ref on the page and once
281 * that page is locked, the mapping is pinned.
283 * We're allowed to run sleeping lock_page() here because we know the caller has
286 static void handle_write_error(struct address_space *mapping,
287 struct page *page, int error)
290 if (page_mapping(page) == mapping)
291 mapping_set_error(mapping, error);
295 /* Request for sync pageout. */
301 /* possible outcome of pageout() */
303 /* failed to write page out, page is locked */
305 /* move page to the active list, page is locked */
307 /* page has been sent to the disk successfully, page is unlocked */
309 /* page is clean and locked */
314 * pageout is called by shrink_page_list() for each dirty page.
315 * Calls ->writepage().
317 static pageout_t pageout(struct page *page, struct address_space *mapping,
318 enum pageout_io sync_writeback)
321 * If the page is dirty, only perform writeback if that write
322 * will be non-blocking. To prevent this allocation from being
323 * stalled by pagecache activity. But note that there may be
324 * stalls if we need to run get_block(). We could test
325 * PagePrivate for that.
327 * If this process is currently in generic_file_write() against
328 * this page's queue, we can perform writeback even if that
331 * If the page is swapcache, write it back even if that would
332 * block, for some throttling. This happens by accident, because
333 * swap_backing_dev_info is bust: it doesn't reflect the
334 * congestion state of the swapdevs. Easy to fix, if needed.
335 * See swapfile.c:page_queue_congested().
337 if (!is_page_cache_freeable(page))
341 * Some data journaling orphaned pages can have
342 * page->mapping == NULL while being dirty with clean buffers.
344 if (PagePrivate(page)) {
345 if (try_to_free_buffers(page)) {
346 ClearPageDirty(page);
347 printk("%s: orphaned page\n", __func__);
353 if (mapping->a_ops->writepage == NULL)
354 return PAGE_ACTIVATE;
355 if (!may_write_to_queue(mapping->backing_dev_info))
358 if (clear_page_dirty_for_io(page)) {
360 struct writeback_control wbc = {
361 .sync_mode = WB_SYNC_NONE,
362 .nr_to_write = SWAP_CLUSTER_MAX,
364 .range_end = LLONG_MAX,
369 SetPageReclaim(page);
370 res = mapping->a_ops->writepage(page, &wbc);
372 handle_write_error(mapping, page, res);
373 if (res == AOP_WRITEPAGE_ACTIVATE) {
374 ClearPageReclaim(page);
375 return PAGE_ACTIVATE;
379 * Wait on writeback if requested to. This happens when
380 * direct reclaiming a large contiguous area and the
381 * first attempt to free a range of pages fails.
383 if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC)
384 wait_on_page_writeback(page);
386 if (!PageWriteback(page)) {
387 /* synchronous write or broken a_ops? */
388 ClearPageReclaim(page);
390 inc_zone_page_state(page, NR_VMSCAN_WRITE);
398 * Same as remove_mapping, but if the page is removed from the mapping, it
399 * gets returned with a refcount of 0.
401 static int __remove_mapping(struct address_space *mapping, struct page *page)
403 BUG_ON(!PageLocked(page));
404 BUG_ON(mapping != page_mapping(page));
406 spin_lock_irq(&mapping->tree_lock);
408 * The non racy check for a busy page.
410 * Must be careful with the order of the tests. When someone has
411 * a ref to the page, it may be possible that they dirty it then
412 * drop the reference. So if PageDirty is tested before page_count
413 * here, then the following race may occur:
415 * get_user_pages(&page);
416 * [user mapping goes away]
418 * !PageDirty(page) [good]
419 * SetPageDirty(page);
421 * !page_count(page) [good, discard it]
423 * [oops, our write_to data is lost]
425 * Reversing the order of the tests ensures such a situation cannot
426 * escape unnoticed. The smp_rmb is needed to ensure the page->flags
427 * load is not satisfied before that of page->_count.
429 * Note that if SetPageDirty is always performed via set_page_dirty,
430 * and thus under tree_lock, then this ordering is not required.
432 if (!page_freeze_refs(page, 2))
434 /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
435 if (unlikely(PageDirty(page))) {
436 page_unfreeze_refs(page, 2);
440 if (PageSwapCache(page)) {
441 swp_entry_t swap = { .val = page_private(page) };
442 __delete_from_swap_cache(page);
443 spin_unlock_irq(&mapping->tree_lock);
446 __remove_from_page_cache(page);
447 spin_unlock_irq(&mapping->tree_lock);
453 spin_unlock_irq(&mapping->tree_lock);
458 * Attempt to detach a locked page from its ->mapping. If it is dirty or if
459 * someone else has a ref on the page, abort and return 0. If it was
460 * successfully detached, return 1. Assumes the caller has a single ref on
463 int remove_mapping(struct address_space *mapping, struct page *page)
465 if (__remove_mapping(mapping, page)) {
467 * Unfreezing the refcount with 1 rather than 2 effectively
468 * drops the pagecache ref for us without requiring another
471 page_unfreeze_refs(page, 1);
478 * putback_lru_page - put previously isolated page onto appropriate LRU list
479 * @page: page to be put back to appropriate lru list
481 * Add previously isolated @page to appropriate LRU list.
482 * Page may still be unevictable for other reasons.
484 * lru_lock must not be held, interrupts must be enabled.
486 #ifdef CONFIG_UNEVICTABLE_LRU
487 void putback_lru_page(struct page *page)
490 int active = !!TestClearPageActive(page);
491 int was_unevictable = PageUnevictable(page);
493 VM_BUG_ON(PageLRU(page));
496 ClearPageUnevictable(page);
498 if (page_evictable(page, NULL)) {
500 * For evictable pages, we can use the cache.
501 * In event of a race, worst case is we end up with an
502 * unevictable page on [in]active list.
503 * We know how to handle that.
505 lru = active + page_is_file_cache(page);
506 lru_cache_add_lru(page, lru);
509 * Put unevictable pages directly on zone's unevictable
512 lru = LRU_UNEVICTABLE;
513 add_page_to_unevictable_list(page);
515 mem_cgroup_move_lists(page, lru);
518 * page's status can change while we move it among lru. If an evictable
519 * page is on unevictable list, it never be freed. To avoid that,
520 * check after we added it to the list, again.
522 if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) {
523 if (!isolate_lru_page(page)) {
527 /* This means someone else dropped this page from LRU
528 * So, it will be freed or putback to LRU again. There is
529 * nothing to do here.
533 if (was_unevictable && lru != LRU_UNEVICTABLE)
534 count_vm_event(UNEVICTABLE_PGRESCUED);
535 else if (!was_unevictable && lru == LRU_UNEVICTABLE)
536 count_vm_event(UNEVICTABLE_PGCULLED);
538 put_page(page); /* drop ref from isolate */
541 #else /* CONFIG_UNEVICTABLE_LRU */
543 void putback_lru_page(struct page *page)
546 VM_BUG_ON(PageLRU(page));
548 lru = !!TestClearPageActive(page) + page_is_file_cache(page);
549 lru_cache_add_lru(page, lru);
550 mem_cgroup_move_lists(page, lru);
553 #endif /* CONFIG_UNEVICTABLE_LRU */
557 * shrink_page_list() returns the number of reclaimed pages
559 static unsigned long shrink_page_list(struct list_head *page_list,
560 struct scan_control *sc,
561 enum pageout_io sync_writeback)
563 LIST_HEAD(ret_pages);
564 struct pagevec freed_pvec;
566 unsigned long nr_reclaimed = 0;
570 pagevec_init(&freed_pvec, 1);
571 while (!list_empty(page_list)) {
572 struct address_space *mapping;
579 page = lru_to_page(page_list);
580 list_del(&page->lru);
582 if (!trylock_page(page))
585 VM_BUG_ON(PageActive(page));
589 if (unlikely(!page_evictable(page, NULL)))
592 if (!sc->may_swap && page_mapped(page))
595 /* Double the slab pressure for mapped and swapcache pages */
596 if (page_mapped(page) || PageSwapCache(page))
599 may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
600 (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
602 if (PageWriteback(page)) {
604 * Synchronous reclaim is performed in two passes,
605 * first an asynchronous pass over the list to
606 * start parallel writeback, and a second synchronous
607 * pass to wait for the IO to complete. Wait here
608 * for any page for which writeback has already
611 if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs)
612 wait_on_page_writeback(page);
617 referenced = page_referenced(page, 1, sc->mem_cgroup);
618 /* In active use or really unfreeable? Activate it. */
619 if (sc->order <= PAGE_ALLOC_COSTLY_ORDER &&
620 referenced && page_mapping_inuse(page))
621 goto activate_locked;
624 * Anonymous process memory has backing store?
625 * Try to allocate it some swap space here.
627 if (PageAnon(page) && !PageSwapCache(page)) {
628 if (!(sc->gfp_mask & __GFP_IO))
630 if (!add_to_swap(page))
631 goto activate_locked;
635 mapping = page_mapping(page);
638 * The page is mapped into the page tables of one or more
639 * processes. Try to unmap it here.
641 if (page_mapped(page) && mapping) {
642 switch (try_to_unmap(page, 0)) {
644 goto activate_locked;
650 ; /* try to free the page below */
654 if (PageDirty(page)) {
655 if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && referenced)
659 if (!sc->may_writepage)
662 /* Page is dirty, try to write it out here */
663 switch (pageout(page, mapping, sync_writeback)) {
667 goto activate_locked;
669 if (PageWriteback(page) || PageDirty(page))
672 * A synchronous write - probably a ramdisk. Go
673 * ahead and try to reclaim the page.
675 if (!trylock_page(page))
677 if (PageDirty(page) || PageWriteback(page))
679 mapping = page_mapping(page);
681 ; /* try to free the page below */
686 * If the page has buffers, try to free the buffer mappings
687 * associated with this page. If we succeed we try to free
690 * We do this even if the page is PageDirty().
691 * try_to_release_page() does not perform I/O, but it is
692 * possible for a page to have PageDirty set, but it is actually
693 * clean (all its buffers are clean). This happens if the
694 * buffers were written out directly, with submit_bh(). ext3
695 * will do this, as well as the blockdev mapping.
696 * try_to_release_page() will discover that cleanness and will
697 * drop the buffers and mark the page clean - it can be freed.
699 * Rarely, pages can have buffers and no ->mapping. These are
700 * the pages which were not successfully invalidated in
701 * truncate_complete_page(). We try to drop those buffers here
702 * and if that worked, and the page is no longer mapped into
703 * process address space (page_count == 1) it can be freed.
704 * Otherwise, leave the page on the LRU so it is swappable.
706 if (PagePrivate(page)) {
707 if (!try_to_release_page(page, sc->gfp_mask))
708 goto activate_locked;
709 if (!mapping && page_count(page) == 1) {
711 if (put_page_testzero(page))
715 * rare race with speculative reference.
716 * the speculative reference will free
717 * this page shortly, so we may
718 * increment nr_reclaimed here (and
719 * leave it off the LRU).
727 if (!mapping || !__remove_mapping(mapping, page))
731 * At this point, we have no other references and there is
732 * no way to pick any more up (removed from LRU, removed
733 * from pagecache). Can use non-atomic bitops now (and
734 * we obviously don't have to worry about waking up a process
735 * waiting on the page lock, because there are no references.
737 __clear_page_locked(page);
740 if (!pagevec_add(&freed_pvec, page)) {
741 __pagevec_free(&freed_pvec);
742 pagevec_reinit(&freed_pvec);
747 if (PageSwapCache(page))
748 try_to_free_swap(page);
750 putback_lru_page(page);
754 /* Not a candidate for swapping, so reclaim swap space. */
755 if (PageSwapCache(page) && vm_swap_full())
756 try_to_free_swap(page);
757 VM_BUG_ON(PageActive(page));
763 list_add(&page->lru, &ret_pages);
764 VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
766 list_splice(&ret_pages, page_list);
767 if (pagevec_count(&freed_pvec))
768 __pagevec_free(&freed_pvec);
769 count_vm_events(PGACTIVATE, pgactivate);
773 /* LRU Isolation modes. */
774 #define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */
775 #define ISOLATE_ACTIVE 1 /* Isolate active pages. */
776 #define ISOLATE_BOTH 2 /* Isolate both active and inactive pages. */
779 * Attempt to remove the specified page from its LRU. Only take this page
780 * if it is of the appropriate PageActive status. Pages which are being
781 * freed elsewhere are also ignored.
783 * page: page to consider
784 * mode: one of the LRU isolation modes defined above
786 * returns 0 on success, -ve errno on failure.
788 int __isolate_lru_page(struct page *page, int mode, int file)
792 /* Only take pages on the LRU. */
797 * When checking the active state, we need to be sure we are
798 * dealing with comparible boolean values. Take the logical not
801 if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
804 if (mode != ISOLATE_BOTH && (!page_is_file_cache(page) != !file))
808 * When this function is being called for lumpy reclaim, we
809 * initially look into all LRU pages, active, inactive and
810 * unevictable; only give shrink_page_list evictable pages.
812 if (PageUnevictable(page))
816 if (likely(get_page_unless_zero(page))) {
818 * Be careful not to clear PageLRU until after we're
819 * sure the page is not being freed elsewhere -- the
820 * page release code relies on it.
830 * zone->lru_lock is heavily contended. Some of the functions that
831 * shrink the lists perform better by taking out a batch of pages
832 * and working on them outside the LRU lock.
834 * For pagecache intensive workloads, this function is the hottest
835 * spot in the kernel (apart from copy_*_user functions).
837 * Appropriate locks must be held before calling this function.
839 * @nr_to_scan: The number of pages to look through on the list.
840 * @src: The LRU list to pull pages off.
841 * @dst: The temp list to put pages on to.
842 * @scanned: The number of pages that were scanned.
843 * @order: The caller's attempted allocation order
844 * @mode: One of the LRU isolation modes
845 * @file: True [1] if isolating file [!anon] pages
847 * returns how many pages were moved onto *@dst.
849 static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
850 struct list_head *src, struct list_head *dst,
851 unsigned long *scanned, int order, int mode, int file)
853 unsigned long nr_taken = 0;
856 for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
859 unsigned long end_pfn;
860 unsigned long page_pfn;
863 page = lru_to_page(src);
864 prefetchw_prev_lru_page(page, src, flags);
866 VM_BUG_ON(!PageLRU(page));
868 switch (__isolate_lru_page(page, mode, file)) {
870 list_move(&page->lru, dst);
875 /* else it is being freed elsewhere */
876 list_move(&page->lru, src);
887 * Attempt to take all pages in the order aligned region
888 * surrounding the tag page. Only take those pages of
889 * the same active state as that tag page. We may safely
890 * round the target page pfn down to the requested order
891 * as the mem_map is guarenteed valid out to MAX_ORDER,
892 * where that page is in a different zone we will detect
893 * it from its zone id and abort this block scan.
895 zone_id = page_zone_id(page);
896 page_pfn = page_to_pfn(page);
897 pfn = page_pfn & ~((1 << order) - 1);
898 end_pfn = pfn + (1 << order);
899 for (; pfn < end_pfn; pfn++) {
900 struct page *cursor_page;
902 /* The target page is in the block, ignore it. */
903 if (unlikely(pfn == page_pfn))
906 /* Avoid holes within the zone. */
907 if (unlikely(!pfn_valid_within(pfn)))
910 cursor_page = pfn_to_page(pfn);
912 /* Check that we have not crossed a zone boundary. */
913 if (unlikely(page_zone_id(cursor_page) != zone_id))
915 switch (__isolate_lru_page(cursor_page, mode, file)) {
917 list_move(&cursor_page->lru, dst);
923 /* else it is being freed elsewhere */
924 list_move(&cursor_page->lru, src);
926 break; /* ! on LRU or wrong list */
935 static unsigned long isolate_pages_global(unsigned long nr,
936 struct list_head *dst,
937 unsigned long *scanned, int order,
938 int mode, struct zone *z,
939 struct mem_cgroup *mem_cont,
940 int active, int file)
947 return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order,
952 * clear_active_flags() is a helper for shrink_active_list(), clearing
953 * any active bits from the pages in the list.
955 static unsigned long clear_active_flags(struct list_head *page_list,
962 list_for_each_entry(page, page_list, lru) {
963 lru = page_is_file_cache(page);
964 if (PageActive(page)) {
966 ClearPageActive(page);
976 * isolate_lru_page - tries to isolate a page from its LRU list
977 * @page: page to isolate from its LRU list
979 * Isolates a @page from an LRU list, clears PageLRU and adjusts the
980 * vmstat statistic corresponding to whatever LRU list the page was on.
982 * Returns 0 if the page was removed from an LRU list.
983 * Returns -EBUSY if the page was not on an LRU list.
985 * The returned page will have PageLRU() cleared. If it was found on
986 * the active list, it will have PageActive set. If it was found on
987 * the unevictable list, it will have the PageUnevictable bit set. That flag
988 * may need to be cleared by the caller before letting the page go.
990 * The vmstat statistic corresponding to the list on which the page was
991 * found will be decremented.
994 * (1) Must be called with an elevated refcount on the page. This is a
995 * fundamentnal difference from isolate_lru_pages (which is called
996 * without a stable reference).
997 * (2) the lru_lock must not be held.
998 * (3) interrupts must be enabled.
1000 int isolate_lru_page(struct page *page)
1004 if (PageLRU(page)) {
1005 struct zone *zone = page_zone(page);
1007 spin_lock_irq(&zone->lru_lock);
1008 if (PageLRU(page) && get_page_unless_zero(page)) {
1009 int lru = page_lru(page);
1013 del_page_from_lru_list(zone, page, lru);
1015 spin_unlock_irq(&zone->lru_lock);
1021 * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
1022 * of reclaimed pages
1024 static unsigned long shrink_inactive_list(unsigned long max_scan,
1025 struct zone *zone, struct scan_control *sc,
1026 int priority, int file)
1028 LIST_HEAD(page_list);
1029 struct pagevec pvec;
1030 unsigned long nr_scanned = 0;
1031 unsigned long nr_reclaimed = 0;
1033 pagevec_init(&pvec, 1);
1036 spin_lock_irq(&zone->lru_lock);
1039 unsigned long nr_taken;
1040 unsigned long nr_scan;
1041 unsigned long nr_freed;
1042 unsigned long nr_active;
1043 unsigned int count[NR_LRU_LISTS] = { 0, };
1044 int mode = ISOLATE_INACTIVE;
1047 * If we need a large contiguous chunk of memory, or have
1048 * trouble getting a small set of contiguous pages, we
1049 * will reclaim both active and inactive pages.
1051 * We use the same threshold as pageout congestion_wait below.
1053 if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
1054 mode = ISOLATE_BOTH;
1055 else if (sc->order && priority < DEF_PRIORITY - 2)
1056 mode = ISOLATE_BOTH;
1058 nr_taken = sc->isolate_pages(sc->swap_cluster_max,
1059 &page_list, &nr_scan, sc->order, mode,
1060 zone, sc->mem_cgroup, 0, file);
1061 nr_active = clear_active_flags(&page_list, count);
1062 __count_vm_events(PGDEACTIVATE, nr_active);
1064 __mod_zone_page_state(zone, NR_ACTIVE_FILE,
1065 -count[LRU_ACTIVE_FILE]);
1066 __mod_zone_page_state(zone, NR_INACTIVE_FILE,
1067 -count[LRU_INACTIVE_FILE]);
1068 __mod_zone_page_state(zone, NR_ACTIVE_ANON,
1069 -count[LRU_ACTIVE_ANON]);
1070 __mod_zone_page_state(zone, NR_INACTIVE_ANON,
1071 -count[LRU_INACTIVE_ANON]);
1073 if (scan_global_lru(sc)) {
1074 zone->pages_scanned += nr_scan;
1075 zone->recent_scanned[0] += count[LRU_INACTIVE_ANON];
1076 zone->recent_scanned[0] += count[LRU_ACTIVE_ANON];
1077 zone->recent_scanned[1] += count[LRU_INACTIVE_FILE];
1078 zone->recent_scanned[1] += count[LRU_ACTIVE_FILE];
1080 spin_unlock_irq(&zone->lru_lock);
1082 nr_scanned += nr_scan;
1083 nr_freed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC);
1086 * If we are direct reclaiming for contiguous pages and we do
1087 * not reclaim everything in the list, try again and wait
1088 * for IO to complete. This will stall high-order allocations
1089 * but that should be acceptable to the caller
1091 if (nr_freed < nr_taken && !current_is_kswapd() &&
1092 sc->order > PAGE_ALLOC_COSTLY_ORDER) {
1093 congestion_wait(WRITE, HZ/10);
1096 * The attempt at page out may have made some
1097 * of the pages active, mark them inactive again.
1099 nr_active = clear_active_flags(&page_list, count);
1100 count_vm_events(PGDEACTIVATE, nr_active);
1102 nr_freed += shrink_page_list(&page_list, sc,
1106 nr_reclaimed += nr_freed;
1107 local_irq_disable();
1108 if (current_is_kswapd()) {
1109 __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scan);
1110 __count_vm_events(KSWAPD_STEAL, nr_freed);
1111 } else if (scan_global_lru(sc))
1112 __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan);
1114 __count_zone_vm_events(PGSTEAL, zone, nr_freed);
1119 spin_lock(&zone->lru_lock);
1121 * Put back any unfreeable pages.
1123 while (!list_empty(&page_list)) {
1125 page = lru_to_page(&page_list);
1126 VM_BUG_ON(PageLRU(page));
1127 list_del(&page->lru);
1128 if (unlikely(!page_evictable(page, NULL))) {
1129 spin_unlock_irq(&zone->lru_lock);
1130 putback_lru_page(page);
1131 spin_lock_irq(&zone->lru_lock);
1135 lru = page_lru(page);
1136 add_page_to_lru_list(zone, page, lru);
1137 mem_cgroup_move_lists(page, lru);
1138 if (PageActive(page) && scan_global_lru(sc)) {
1139 int file = !!page_is_file_cache(page);
1140 zone->recent_rotated[file]++;
1142 if (!pagevec_add(&pvec, page)) {
1143 spin_unlock_irq(&zone->lru_lock);
1144 __pagevec_release(&pvec);
1145 spin_lock_irq(&zone->lru_lock);
1148 } while (nr_scanned < max_scan);
1149 spin_unlock(&zone->lru_lock);
1152 pagevec_release(&pvec);
1153 return nr_reclaimed;
1157 * We are about to scan this zone at a certain priority level. If that priority
1158 * level is smaller (ie: more urgent) than the previous priority, then note
1159 * that priority level within the zone. This is done so that when the next
1160 * process comes in to scan this zone, it will immediately start out at this
1161 * priority level rather than having to build up its own scanning priority.
1162 * Here, this priority affects only the reclaim-mapped threshold.
1164 static inline void note_zone_scanning_priority(struct zone *zone, int priority)
1166 if (priority < zone->prev_priority)
1167 zone->prev_priority = priority;
1170 static inline int zone_is_near_oom(struct zone *zone)
1172 return zone->pages_scanned >= (zone_lru_pages(zone) * 3);
1176 * This moves pages from the active list to the inactive list.
1178 * We move them the other way if the page is referenced by one or more
1179 * processes, from rmap.
1181 * If the pages are mostly unmapped, the processing is fast and it is
1182 * appropriate to hold zone->lru_lock across the whole operation. But if
1183 * the pages are mapped, the processing is slow (page_referenced()) so we
1184 * should drop zone->lru_lock around each page. It's impossible to balance
1185 * this, so instead we remove the pages from the LRU while processing them.
1186 * It is safe to rely on PG_active against the non-LRU pages in here because
1187 * nobody will play with that bit on a non-LRU page.
1189 * The downside is that we have to touch page->_count against each page.
1190 * But we had to alter page->flags anyway.
1194 static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
1195 struct scan_control *sc, int priority, int file)
1197 unsigned long pgmoved;
1198 int pgdeactivate = 0;
1199 unsigned long pgscanned;
1200 LIST_HEAD(l_hold); /* The pages which were snipped off */
1201 LIST_HEAD(l_inactive);
1203 struct pagevec pvec;
1207 spin_lock_irq(&zone->lru_lock);
1208 pgmoved = sc->isolate_pages(nr_pages, &l_hold, &pgscanned, sc->order,
1209 ISOLATE_ACTIVE, zone,
1210 sc->mem_cgroup, 1, file);
1212 * zone->pages_scanned is used for detect zone's oom
1213 * mem_cgroup remembers nr_scan by itself.
1215 if (scan_global_lru(sc)) {
1216 zone->pages_scanned += pgscanned;
1217 zone->recent_scanned[!!file] += pgmoved;
1221 __mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved);
1223 __mod_zone_page_state(zone, NR_ACTIVE_ANON, -pgmoved);
1224 spin_unlock_irq(&zone->lru_lock);
1227 while (!list_empty(&l_hold)) {
1229 page = lru_to_page(&l_hold);
1230 list_del(&page->lru);
1232 if (unlikely(!page_evictable(page, NULL))) {
1233 putback_lru_page(page);
1237 /* page_referenced clears PageReferenced */
1238 if (page_mapping_inuse(page) &&
1239 page_referenced(page, 0, sc->mem_cgroup))
1242 list_add(&page->lru, &l_inactive);
1245 spin_lock_irq(&zone->lru_lock);
1247 * Count referenced pages from currently used mappings as
1248 * rotated, even though they are moved to the inactive list.
1249 * This helps balance scan pressure between file and anonymous
1250 * pages in get_scan_ratio.
1252 if (scan_global_lru(sc))
1253 zone->recent_rotated[!!file] += pgmoved;
1256 * Move the pages to the [file or anon] inactive list.
1258 pagevec_init(&pvec, 1);
1261 lru = LRU_BASE + file * LRU_FILE;
1262 while (!list_empty(&l_inactive)) {
1263 page = lru_to_page(&l_inactive);
1264 prefetchw_prev_lru_page(page, &l_inactive, flags);
1265 VM_BUG_ON(PageLRU(page));
1267 VM_BUG_ON(!PageActive(page));
1268 ClearPageActive(page);
1270 list_move(&page->lru, &zone->lru[lru].list);
1271 mem_cgroup_move_lists(page, lru);
1273 if (!pagevec_add(&pvec, page)) {
1274 __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
1275 spin_unlock_irq(&zone->lru_lock);
1276 pgdeactivate += pgmoved;
1278 if (buffer_heads_over_limit)
1279 pagevec_strip(&pvec);
1280 __pagevec_release(&pvec);
1281 spin_lock_irq(&zone->lru_lock);
1284 __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
1285 pgdeactivate += pgmoved;
1286 if (buffer_heads_over_limit) {
1287 spin_unlock_irq(&zone->lru_lock);
1288 pagevec_strip(&pvec);
1289 spin_lock_irq(&zone->lru_lock);
1291 __count_zone_vm_events(PGREFILL, zone, pgscanned);
1292 __count_vm_events(PGDEACTIVATE, pgdeactivate);
1293 spin_unlock_irq(&zone->lru_lock);
1295 pagevec_swap_free(&pvec);
1297 pagevec_release(&pvec);
1300 static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
1301 struct zone *zone, struct scan_control *sc, int priority)
1303 int file = is_file_lru(lru);
1305 if (lru == LRU_ACTIVE_FILE) {
1306 shrink_active_list(nr_to_scan, zone, sc, priority, file);
1310 if (lru == LRU_ACTIVE_ANON &&
1311 (!scan_global_lru(sc) || inactive_anon_is_low(zone))) {
1312 shrink_active_list(nr_to_scan, zone, sc, priority, file);
1315 return shrink_inactive_list(nr_to_scan, zone, sc, priority, file);
1319 * Determine how aggressively the anon and file LRU lists should be
1320 * scanned. The relative value of each set of LRU lists is determined
1321 * by looking at the fraction of the pages scanned we did rotate back
1322 * onto the active list instead of evict.
1324 * percent[0] specifies how much pressure to put on ram/swap backed
1325 * memory, while percent[1] determines pressure on the file LRUs.
1327 static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
1328 unsigned long *percent)
1330 unsigned long anon, file, free;
1331 unsigned long anon_prio, file_prio;
1332 unsigned long ap, fp;
1334 /* If we have no swap space, do not bother scanning anon pages. */
1335 if (nr_swap_pages <= 0) {
1341 anon = zone_page_state(zone, NR_ACTIVE_ANON) +
1342 zone_page_state(zone, NR_INACTIVE_ANON);
1343 file = zone_page_state(zone, NR_ACTIVE_FILE) +
1344 zone_page_state(zone, NR_INACTIVE_FILE);
1345 free = zone_page_state(zone, NR_FREE_PAGES);
1347 /* If we have very few page cache pages, force-scan anon pages. */
1348 if (unlikely(file + free <= zone->pages_high)) {
1355 * OK, so we have swap space and a fair amount of page cache
1356 * pages. We use the recently rotated / recently scanned
1357 * ratios to determine how valuable each cache is.
1359 * Because workloads change over time (and to avoid overflow)
1360 * we keep these statistics as a floating average, which ends
1361 * up weighing recent references more than old ones.
1363 * anon in [0], file in [1]
1365 if (unlikely(zone->recent_scanned[0] > anon / 4)) {
1366 spin_lock_irq(&zone->lru_lock);
1367 zone->recent_scanned[0] /= 2;
1368 zone->recent_rotated[0] /= 2;
1369 spin_unlock_irq(&zone->lru_lock);
1372 if (unlikely(zone->recent_scanned[1] > file / 4)) {
1373 spin_lock_irq(&zone->lru_lock);
1374 zone->recent_scanned[1] /= 2;
1375 zone->recent_rotated[1] /= 2;
1376 spin_unlock_irq(&zone->lru_lock);
1380 * With swappiness at 100, anonymous and file have the same priority.
1381 * This scanning priority is essentially the inverse of IO cost.
1383 anon_prio = sc->swappiness;
1384 file_prio = 200 - sc->swappiness;
1387 * The amount of pressure on anon vs file pages is inversely
1388 * proportional to the fraction of recently scanned pages on
1389 * each list that were recently referenced and in active use.
1391 ap = (anon_prio + 1) * (zone->recent_scanned[0] + 1);
1392 ap /= zone->recent_rotated[0] + 1;
1394 fp = (file_prio + 1) * (zone->recent_scanned[1] + 1);
1395 fp /= zone->recent_rotated[1] + 1;
1397 /* Normalize to percentages */
1398 percent[0] = 100 * ap / (ap + fp + 1);
1399 percent[1] = 100 - percent[0];
1404 * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
1406 static void shrink_zone(int priority, struct zone *zone,
1407 struct scan_control *sc)
1409 unsigned long nr[NR_LRU_LISTS];
1410 unsigned long nr_to_scan;
1411 unsigned long percent[2]; /* anon @ 0; file @ 1 */
1413 unsigned long nr_reclaimed = sc->nr_reclaimed;
1414 unsigned long swap_cluster_max = sc->swap_cluster_max;
1416 get_scan_ratio(zone, sc, percent);
1418 for_each_evictable_lru(l) {
1419 if (scan_global_lru(sc)) {
1420 int file = is_file_lru(l);
1423 scan = zone_page_state(zone, NR_LRU_BASE + l);
1426 scan = (scan * percent[file]) / 100;
1428 zone->lru[l].nr_scan += scan;
1429 nr[l] = zone->lru[l].nr_scan;
1430 if (nr[l] >= swap_cluster_max)
1431 zone->lru[l].nr_scan = 0;
1436 * This reclaim occurs not because zone memory shortage
1437 * but because memory controller hits its limit.
1438 * Don't modify zone reclaim related data.
1440 nr[l] = mem_cgroup_calc_reclaim(sc->mem_cgroup, zone,
1445 while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
1446 nr[LRU_INACTIVE_FILE]) {
1447 for_each_evictable_lru(l) {
1449 nr_to_scan = min(nr[l], swap_cluster_max);
1450 nr[l] -= nr_to_scan;
1452 nr_reclaimed += shrink_list(l, nr_to_scan,
1453 zone, sc, priority);
1457 * On large memory systems, scan >> priority can become
1458 * really large. This is fine for the starting priority;
1459 * we want to put equal scanning pressure on each zone.
1460 * However, if the VM has a harder time of freeing pages,
1461 * with multiple processes reclaiming pages, the total
1462 * freeing target can get unreasonably large.
1464 if (nr_reclaimed > swap_cluster_max &&
1465 priority < DEF_PRIORITY && !current_is_kswapd())
1469 sc->nr_reclaimed = nr_reclaimed;
1472 * Even if we did not try to evict anon pages at all, we want to
1473 * rebalance the anon lru active/inactive ratio.
1475 if (!scan_global_lru(sc) || inactive_anon_is_low(zone))
1476 shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1477 else if (!scan_global_lru(sc))
1478 shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1480 throttle_vm_writeout(sc->gfp_mask);
1484 * This is the direct reclaim path, for page-allocating processes. We only
1485 * try to reclaim pages from zones which will satisfy the caller's allocation
1488 * We reclaim from a zone even if that zone is over pages_high. Because:
1489 * a) The caller may be trying to free *extra* pages to satisfy a higher-order
1491 * b) The zones may be over pages_high but they must go *over* pages_high to
1492 * satisfy the `incremental min' zone defense algorithm.
1494 * If a zone is deemed to be full of pinned pages then just give it a light
1495 * scan then give up on it.
1497 static void shrink_zones(int priority, struct zonelist *zonelist,
1498 struct scan_control *sc)
1500 enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
1504 sc->all_unreclaimable = 1;
1505 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1506 if (!populated_zone(zone))
1509 * Take care memory controller reclaiming has small influence
1512 if (scan_global_lru(sc)) {
1513 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1515 note_zone_scanning_priority(zone, priority);
1517 if (zone_is_all_unreclaimable(zone) &&
1518 priority != DEF_PRIORITY)
1519 continue; /* Let kswapd poll it */
1520 sc->all_unreclaimable = 0;
1523 * Ignore cpuset limitation here. We just want to reduce
1524 * # of used pages by us regardless of memory shortage.
1526 sc->all_unreclaimable = 0;
1527 mem_cgroup_note_reclaim_priority(sc->mem_cgroup,
1531 shrink_zone(priority, zone, sc);
1536 * This is the main entry point to direct page reclaim.
1538 * If a full scan of the inactive list fails to free enough memory then we
1539 * are "out of memory" and something needs to be killed.
1541 * If the caller is !__GFP_FS then the probability of a failure is reasonably
1542 * high - the zone may be full of dirty or under-writeback pages, which this
1543 * caller can't do much about. We kick pdflush and take explicit naps in the
1544 * hope that some of these pages can be written. But if the allocating task
1545 * holds filesystem locks which prevent writeout this might not work, and the
1546 * allocation attempt will fail.
1548 * returns: 0, if no pages reclaimed
1549 * else, the number of pages reclaimed
1551 static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
1552 struct scan_control *sc)
1555 unsigned long ret = 0;
1556 unsigned long total_scanned = 0;
1557 struct reclaim_state *reclaim_state = current->reclaim_state;
1558 unsigned long lru_pages = 0;
1561 enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
1563 delayacct_freepages_start();
1565 if (scan_global_lru(sc))
1566 count_vm_event(ALLOCSTALL);
1568 * mem_cgroup will not do shrink_slab.
1570 if (scan_global_lru(sc)) {
1571 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1573 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1576 lru_pages += zone_lru_pages(zone);
1580 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
1583 disable_swap_token();
1584 shrink_zones(priority, zonelist, sc);
1586 * Don't shrink slabs when reclaiming memory from
1587 * over limit cgroups
1589 if (scan_global_lru(sc)) {
1590 shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages);
1591 if (reclaim_state) {
1592 sc->nr_reclaimed += reclaim_state->reclaimed_slab;
1593 reclaim_state->reclaimed_slab = 0;
1596 total_scanned += sc->nr_scanned;
1597 if (sc->nr_reclaimed >= sc->swap_cluster_max) {
1598 ret = sc->nr_reclaimed;
1603 * Try to write back as many pages as we just scanned. This
1604 * tends to cause slow streaming writers to write data to the
1605 * disk smoothly, at the dirtying rate, which is nice. But
1606 * that's undesirable in laptop mode, where we *want* lumpy
1607 * writeout. So in laptop mode, write out the whole world.
1609 if (total_scanned > sc->swap_cluster_max +
1610 sc->swap_cluster_max / 2) {
1611 wakeup_pdflush(laptop_mode ? 0 : total_scanned);
1612 sc->may_writepage = 1;
1615 /* Take a nap, wait for some writeback to complete */
1616 if (sc->nr_scanned && priority < DEF_PRIORITY - 2)
1617 congestion_wait(WRITE, HZ/10);
1619 /* top priority shrink_zones still had more to do? don't OOM, then */
1620 if (!sc->all_unreclaimable && scan_global_lru(sc))
1621 ret = sc->nr_reclaimed;
1624 * Now that we've scanned all the zones at this priority level, note
1625 * that level within the zone so that the next thread which performs
1626 * scanning of this zone will immediately start out at this priority
1627 * level. This affects only the decision whether or not to bring
1628 * mapped pages onto the inactive list.
1633 if (scan_global_lru(sc)) {
1634 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1636 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1639 zone->prev_priority = priority;
1642 mem_cgroup_record_reclaim_priority(sc->mem_cgroup, priority);
1644 delayacct_freepages_end();
1649 unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
1652 struct scan_control sc = {
1653 .gfp_mask = gfp_mask,
1654 .may_writepage = !laptop_mode,
1655 .swap_cluster_max = SWAP_CLUSTER_MAX,
1657 .swappiness = vm_swappiness,
1660 .isolate_pages = isolate_pages_global,
1663 return do_try_to_free_pages(zonelist, &sc);
1666 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1668 unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
1671 struct scan_control sc = {
1672 .may_writepage = !laptop_mode,
1674 .swap_cluster_max = SWAP_CLUSTER_MAX,
1675 .swappiness = vm_swappiness,
1677 .mem_cgroup = mem_cont,
1678 .isolate_pages = mem_cgroup_isolate_pages,
1680 struct zonelist *zonelist;
1682 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
1683 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
1684 zonelist = NODE_DATA(numa_node_id())->node_zonelists;
1685 return do_try_to_free_pages(zonelist, &sc);
1690 * For kswapd, balance_pgdat() will work across all this node's zones until
1691 * they are all at pages_high.
1693 * Returns the number of pages which were actually freed.
1695 * There is special handling here for zones which are full of pinned pages.
1696 * This can happen if the pages are all mlocked, or if they are all used by
1697 * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
1698 * What we do is to detect the case where all pages in the zone have been
1699 * scanned twice and there has been zero successful reclaim. Mark the zone as
1700 * dead and from now on, only perform a short scan. Basically we're polling
1701 * the zone for when the problem goes away.
1703 * kswapd scans the zones in the highmem->normal->dma direction. It skips
1704 * zones which have free_pages > pages_high, but once a zone is found to have
1705 * free_pages <= pages_high, we scan that zone and the lower zones regardless
1706 * of the number of free pages in the lower zones. This interoperates with
1707 * the page allocator fallback scheme to ensure that aging of pages is balanced
1710 static unsigned long balance_pgdat(pg_data_t *pgdat, int order)
1715 unsigned long total_scanned;
1716 struct reclaim_state *reclaim_state = current->reclaim_state;
1717 struct scan_control sc = {
1718 .gfp_mask = GFP_KERNEL,
1720 .swap_cluster_max = SWAP_CLUSTER_MAX,
1721 .swappiness = vm_swappiness,
1724 .isolate_pages = isolate_pages_global,
1727 * temp_priority is used to remember the scanning priority at which
1728 * this zone was successfully refilled to free_pages == pages_high.
1730 int temp_priority[MAX_NR_ZONES];
1734 sc.nr_reclaimed = 0;
1735 sc.may_writepage = !laptop_mode;
1736 count_vm_event(PAGEOUTRUN);
1738 for (i = 0; i < pgdat->nr_zones; i++)
1739 temp_priority[i] = DEF_PRIORITY;
1741 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
1742 int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */
1743 unsigned long lru_pages = 0;
1745 /* The swap token gets in the way of swapout... */
1747 disable_swap_token();
1752 * Scan in the highmem->dma direction for the highest
1753 * zone which needs scanning
1755 for (i = pgdat->nr_zones - 1; i >= 0; i--) {
1756 struct zone *zone = pgdat->node_zones + i;
1758 if (!populated_zone(zone))
1761 if (zone_is_all_unreclaimable(zone) &&
1762 priority != DEF_PRIORITY)
1766 * Do some background aging of the anon list, to give
1767 * pages a chance to be referenced before reclaiming.
1769 if (inactive_anon_is_low(zone))
1770 shrink_active_list(SWAP_CLUSTER_MAX, zone,
1773 if (!zone_watermark_ok(zone, order, zone->pages_high,
1782 for (i = 0; i <= end_zone; i++) {
1783 struct zone *zone = pgdat->node_zones + i;
1785 lru_pages += zone_lru_pages(zone);
1789 * Now scan the zone in the dma->highmem direction, stopping
1790 * at the last zone which needs scanning.
1792 * We do this because the page allocator works in the opposite
1793 * direction. This prevents the page allocator from allocating
1794 * pages behind kswapd's direction of progress, which would
1795 * cause too much scanning of the lower zones.
1797 for (i = 0; i <= end_zone; i++) {
1798 struct zone *zone = pgdat->node_zones + i;
1801 if (!populated_zone(zone))
1804 if (zone_is_all_unreclaimable(zone) &&
1805 priority != DEF_PRIORITY)
1808 if (!zone_watermark_ok(zone, order, zone->pages_high,
1811 temp_priority[i] = priority;
1813 note_zone_scanning_priority(zone, priority);
1815 * We put equal pressure on every zone, unless one
1816 * zone has way too many pages free already.
1818 if (!zone_watermark_ok(zone, order, 8*zone->pages_high,
1820 shrink_zone(priority, zone, &sc);
1821 reclaim_state->reclaimed_slab = 0;
1822 nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL,
1824 sc.nr_reclaimed += reclaim_state->reclaimed_slab;
1825 total_scanned += sc.nr_scanned;
1826 if (zone_is_all_unreclaimable(zone))
1828 if (nr_slab == 0 && zone->pages_scanned >=
1829 (zone_lru_pages(zone) * 6))
1831 ZONE_ALL_UNRECLAIMABLE);
1833 * If we've done a decent amount of scanning and
1834 * the reclaim ratio is low, start doing writepage
1835 * even in laptop mode
1837 if (total_scanned > SWAP_CLUSTER_MAX * 2 &&
1838 total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2)
1839 sc.may_writepage = 1;
1842 break; /* kswapd: all done */
1844 * OK, kswapd is getting into trouble. Take a nap, then take
1845 * another pass across the zones.
1847 if (total_scanned && priority < DEF_PRIORITY - 2)
1848 congestion_wait(WRITE, HZ/10);
1851 * We do this so kswapd doesn't build up large priorities for
1852 * example when it is freeing in parallel with allocators. It
1853 * matches the direct reclaim path behaviour in terms of impact
1854 * on zone->*_priority.
1856 if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX)
1861 * Note within each zone the priority level at which this zone was
1862 * brought into a happy state. So that the next thread which scans this
1863 * zone will start out at that priority level.
1865 for (i = 0; i < pgdat->nr_zones; i++) {
1866 struct zone *zone = pgdat->node_zones + i;
1868 zone->prev_priority = temp_priority[i];
1870 if (!all_zones_ok) {
1878 return sc.nr_reclaimed;
1882 * The background pageout daemon, started as a kernel thread
1883 * from the init process.
1885 * This basically trickles out pages so that we have _some_
1886 * free memory available even if there is no other activity
1887 * that frees anything up. This is needed for things like routing
1888 * etc, where we otherwise might have all activity going on in
1889 * asynchronous contexts that cannot page things out.
1891 * If there are applications that are active memory-allocators
1892 * (most normal use), this basically shouldn't matter.
1894 static int kswapd(void *p)
1896 unsigned long order;
1897 pg_data_t *pgdat = (pg_data_t*)p;
1898 struct task_struct *tsk = current;
1900 struct reclaim_state reclaim_state = {
1901 .reclaimed_slab = 0,
1903 node_to_cpumask_ptr(cpumask, pgdat->node_id);
1905 if (!cpumask_empty(cpumask))
1906 set_cpus_allowed_ptr(tsk, cpumask);
1907 current->reclaim_state = &reclaim_state;
1910 * Tell the memory management that we're a "memory allocator",
1911 * and that if we need more memory we should get access to it
1912 * regardless (see "__alloc_pages()"). "kswapd" should
1913 * never get caught in the normal page freeing logic.
1915 * (Kswapd normally doesn't need memory anyway, but sometimes
1916 * you need a small amount of memory in order to be able to
1917 * page out something else, and this flag essentially protects
1918 * us from recursively trying to free more memory as we're
1919 * trying to free the first piece of memory in the first place).
1921 tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
1926 unsigned long new_order;
1928 prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
1929 new_order = pgdat->kswapd_max_order;
1930 pgdat->kswapd_max_order = 0;
1931 if (order < new_order) {
1933 * Don't sleep if someone wants a larger 'order'
1938 if (!freezing(current))
1941 order = pgdat->kswapd_max_order;
1943 finish_wait(&pgdat->kswapd_wait, &wait);
1945 if (!try_to_freeze()) {
1946 /* We can speed up thawing tasks if we don't call
1947 * balance_pgdat after returning from the refrigerator
1949 balance_pgdat(pgdat, order);
1956 * A zone is low on free memory, so wake its kswapd task to service it.
1958 void wakeup_kswapd(struct zone *zone, int order)
1962 if (!populated_zone(zone))
1965 pgdat = zone->zone_pgdat;
1966 if (zone_watermark_ok(zone, order, zone->pages_low, 0, 0))
1968 if (pgdat->kswapd_max_order < order)
1969 pgdat->kswapd_max_order = order;
1970 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1972 if (!waitqueue_active(&pgdat->kswapd_wait))
1974 wake_up_interruptible(&pgdat->kswapd_wait);
1977 unsigned long global_lru_pages(void)
1979 return global_page_state(NR_ACTIVE_ANON)
1980 + global_page_state(NR_ACTIVE_FILE)
1981 + global_page_state(NR_INACTIVE_ANON)
1982 + global_page_state(NR_INACTIVE_FILE);
1987 * Helper function for shrink_all_memory(). Tries to reclaim 'nr_pages' pages
1988 * from LRU lists system-wide, for given pass and priority, and returns the
1989 * number of reclaimed pages
1991 * For pass > 3 we also try to shrink the LRU lists that contain a few pages
1993 static unsigned long shrink_all_zones(unsigned long nr_pages, int prio,
1994 int pass, struct scan_control *sc)
1997 unsigned long nr_to_scan, ret = 0;
2000 for_each_zone(zone) {
2002 if (!populated_zone(zone))
2005 if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY)
2008 for_each_evictable_lru(l) {
2009 /* For pass = 0, we don't shrink the active list */
2011 (l == LRU_ACTIVE || l == LRU_ACTIVE_FILE))
2014 zone->lru[l].nr_scan +=
2015 (zone_page_state(zone, NR_LRU_BASE + l)
2017 if (zone->lru[l].nr_scan >= nr_pages || pass > 3) {
2018 zone->lru[l].nr_scan = 0;
2019 nr_to_scan = min(nr_pages,
2020 zone_page_state(zone,
2022 ret += shrink_list(l, nr_to_scan, zone,
2024 if (ret >= nr_pages)
2034 * Try to free `nr_pages' of memory, system-wide, and return the number of
2037 * Rather than trying to age LRUs the aim is to preserve the overall
2038 * LRU order by reclaiming preferentially
2039 * inactive > active > active referenced > active mapped
2041 unsigned long shrink_all_memory(unsigned long nr_pages)
2043 unsigned long lru_pages, nr_slab;
2044 unsigned long ret = 0;
2046 struct reclaim_state reclaim_state;
2047 struct scan_control sc = {
2048 .gfp_mask = GFP_KERNEL,
2050 .swap_cluster_max = nr_pages,
2052 .swappiness = vm_swappiness,
2053 .isolate_pages = isolate_pages_global,
2056 current->reclaim_state = &reclaim_state;
2058 lru_pages = global_lru_pages();
2059 nr_slab = global_page_state(NR_SLAB_RECLAIMABLE);
2060 /* If slab caches are huge, it's better to hit them first */
2061 while (nr_slab >= lru_pages) {
2062 reclaim_state.reclaimed_slab = 0;
2063 shrink_slab(nr_pages, sc.gfp_mask, lru_pages);
2064 if (!reclaim_state.reclaimed_slab)
2067 ret += reclaim_state.reclaimed_slab;
2068 if (ret >= nr_pages)
2071 nr_slab -= reclaim_state.reclaimed_slab;
2075 * We try to shrink LRUs in 5 passes:
2076 * 0 = Reclaim from inactive_list only
2077 * 1 = Reclaim from active list but don't reclaim mapped
2078 * 2 = 2nd pass of type 1
2079 * 3 = Reclaim mapped (normal reclaim)
2080 * 4 = 2nd pass of type 3
2082 for (pass = 0; pass < 5; pass++) {
2085 /* Force reclaiming mapped pages in the passes #3 and #4 */
2088 sc.swappiness = 100;
2091 for (prio = DEF_PRIORITY; prio >= 0; prio--) {
2092 unsigned long nr_to_scan = nr_pages - ret;
2095 ret += shrink_all_zones(nr_to_scan, prio, pass, &sc);
2096 if (ret >= nr_pages)
2099 reclaim_state.reclaimed_slab = 0;
2100 shrink_slab(sc.nr_scanned, sc.gfp_mask,
2101 global_lru_pages());
2102 ret += reclaim_state.reclaimed_slab;
2103 if (ret >= nr_pages)
2106 if (sc.nr_scanned && prio < DEF_PRIORITY - 2)
2107 congestion_wait(WRITE, HZ / 10);
2112 * If ret = 0, we could not shrink LRUs, but there may be something
2117 reclaim_state.reclaimed_slab = 0;
2118 shrink_slab(nr_pages, sc.gfp_mask, global_lru_pages());
2119 ret += reclaim_state.reclaimed_slab;
2120 } while (ret < nr_pages && reclaim_state.reclaimed_slab > 0);
2124 current->reclaim_state = NULL;
2130 /* It's optimal to keep kswapds on the same CPUs as their memory, but
2131 not required for correctness. So if the last cpu in a node goes
2132 away, we get changed to run anywhere: as the first one comes back,
2133 restore their cpu bindings. */
2134 static int __devinit cpu_callback(struct notifier_block *nfb,
2135 unsigned long action, void *hcpu)
2139 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
2140 for_each_node_state(nid, N_HIGH_MEMORY) {
2141 pg_data_t *pgdat = NODE_DATA(nid);
2142 node_to_cpumask_ptr(mask, pgdat->node_id);
2144 if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
2145 /* One of our CPUs online: restore mask */
2146 set_cpus_allowed_ptr(pgdat->kswapd, mask);
2153 * This kswapd start function will be called by init and node-hot-add.
2154 * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
2156 int kswapd_run(int nid)
2158 pg_data_t *pgdat = NODE_DATA(nid);
2164 pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
2165 if (IS_ERR(pgdat->kswapd)) {
2166 /* failure at boot is fatal */
2167 BUG_ON(system_state == SYSTEM_BOOTING);
2168 printk("Failed to start kswapd on node %d\n",nid);
2174 static int __init kswapd_init(void)
2179 for_each_node_state(nid, N_HIGH_MEMORY)
2181 hotcpu_notifier(cpu_callback, 0);
2185 module_init(kswapd_init)
2191 * If non-zero call zone_reclaim when the number of free pages falls below
2194 int zone_reclaim_mode __read_mostly;
2196 #define RECLAIM_OFF 0
2197 #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
2198 #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
2199 #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
2202 * Priority for ZONE_RECLAIM. This determines the fraction of pages
2203 * of a node considered for each zone_reclaim. 4 scans 1/16th of
2206 #define ZONE_RECLAIM_PRIORITY 4
2209 * Percentage of pages in a zone that must be unmapped for zone_reclaim to
2212 int sysctl_min_unmapped_ratio = 1;
2215 * If the number of slab pages in a zone grows beyond this percentage then
2216 * slab reclaim needs to occur.
2218 int sysctl_min_slab_ratio = 5;
2221 * Try to free up some pages from this zone through reclaim.
2223 static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
2225 /* Minimum pages needed in order to stay on node */
2226 const unsigned long nr_pages = 1 << order;
2227 struct task_struct *p = current;
2228 struct reclaim_state reclaim_state;
2230 struct scan_control sc = {
2231 .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
2232 .may_swap = !!(zone_reclaim_mode & RECLAIM_SWAP),
2233 .swap_cluster_max = max_t(unsigned long, nr_pages,
2235 .gfp_mask = gfp_mask,
2236 .swappiness = vm_swappiness,
2237 .isolate_pages = isolate_pages_global,
2239 unsigned long slab_reclaimable;
2241 disable_swap_token();
2244 * We need to be able to allocate from the reserves for RECLAIM_SWAP
2245 * and we also need to be able to write out pages for RECLAIM_WRITE
2248 p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
2249 reclaim_state.reclaimed_slab = 0;
2250 p->reclaim_state = &reclaim_state;
2252 if (zone_page_state(zone, NR_FILE_PAGES) -
2253 zone_page_state(zone, NR_FILE_MAPPED) >
2254 zone->min_unmapped_pages) {
2256 * Free memory by calling shrink zone with increasing
2257 * priorities until we have enough memory freed.
2259 priority = ZONE_RECLAIM_PRIORITY;
2261 note_zone_scanning_priority(zone, priority);
2262 shrink_zone(priority, zone, &sc);
2264 } while (priority >= 0 && sc.nr_reclaimed < nr_pages);
2267 slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2268 if (slab_reclaimable > zone->min_slab_pages) {
2270 * shrink_slab() does not currently allow us to determine how
2271 * many pages were freed in this zone. So we take the current
2272 * number of slab pages and shake the slab until it is reduced
2273 * by the same nr_pages that we used for reclaiming unmapped
2276 * Note that shrink_slab will free memory on all zones and may
2279 while (shrink_slab(sc.nr_scanned, gfp_mask, order) &&
2280 zone_page_state(zone, NR_SLAB_RECLAIMABLE) >
2281 slab_reclaimable - nr_pages)
2285 * Update nr_reclaimed by the number of slab pages we
2286 * reclaimed from this zone.
2288 sc.nr_reclaimed += slab_reclaimable -
2289 zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2292 p->reclaim_state = NULL;
2293 current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
2294 return sc.nr_reclaimed >= nr_pages;
2297 int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
2303 * Zone reclaim reclaims unmapped file backed pages and
2304 * slab pages if we are over the defined limits.
2306 * A small portion of unmapped file backed pages is needed for
2307 * file I/O otherwise pages read by file I/O will be immediately
2308 * thrown out if the zone is overallocated. So we do not reclaim
2309 * if less than a specified percentage of the zone is used by
2310 * unmapped file backed pages.
2312 if (zone_page_state(zone, NR_FILE_PAGES) -
2313 zone_page_state(zone, NR_FILE_MAPPED) <= zone->min_unmapped_pages
2314 && zone_page_state(zone, NR_SLAB_RECLAIMABLE)
2315 <= zone->min_slab_pages)
2318 if (zone_is_all_unreclaimable(zone))
2322 * Do not scan if the allocation should not be delayed.
2324 if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
2328 * Only run zone reclaim on the local zone or on zones that do not
2329 * have associated processors. This will favor the local processor
2330 * over remote processors and spread off node memory allocations
2331 * as wide as possible.
2333 node_id = zone_to_nid(zone);
2334 if (node_state(node_id, N_CPU) && node_id != numa_node_id())
2337 if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
2339 ret = __zone_reclaim(zone, gfp_mask, order);
2340 zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
2346 #ifdef CONFIG_UNEVICTABLE_LRU
2348 * page_evictable - test whether a page is evictable
2349 * @page: the page to test
2350 * @vma: the VMA in which the page is or will be mapped, may be NULL
2352 * Test whether page is evictable--i.e., should be placed on active/inactive
2353 * lists vs unevictable list. The vma argument is !NULL when called from the
2354 * fault path to determine how to instantate a new page.
2356 * Reasons page might not be evictable:
2357 * (1) page's mapping marked unevictable
2358 * (2) page is part of an mlocked VMA
2361 int page_evictable(struct page *page, struct vm_area_struct *vma)
2364 if (mapping_unevictable(page_mapping(page)))
2367 if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page)))
2374 * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list
2375 * @page: page to check evictability and move to appropriate lru list
2376 * @zone: zone page is in
2378 * Checks a page for evictability and moves the page to the appropriate
2381 * Restrictions: zone->lru_lock must be held, page must be on LRU and must
2382 * have PageUnevictable set.
2384 static void check_move_unevictable_page(struct page *page, struct zone *zone)
2386 VM_BUG_ON(PageActive(page));
2389 ClearPageUnevictable(page);
2390 if (page_evictable(page, NULL)) {
2391 enum lru_list l = LRU_INACTIVE_ANON + page_is_file_cache(page);
2393 __dec_zone_state(zone, NR_UNEVICTABLE);
2394 list_move(&page->lru, &zone->lru[l].list);
2395 __inc_zone_state(zone, NR_INACTIVE_ANON + l);
2396 __count_vm_event(UNEVICTABLE_PGRESCUED);
2399 * rotate unevictable list
2401 SetPageUnevictable(page);
2402 list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list);
2403 if (page_evictable(page, NULL))
2409 * scan_mapping_unevictable_pages - scan an address space for evictable pages
2410 * @mapping: struct address_space to scan for evictable pages
2412 * Scan all pages in mapping. Check unevictable pages for
2413 * evictability and move them to the appropriate zone lru list.
2415 void scan_mapping_unevictable_pages(struct address_space *mapping)
2418 pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >>
2421 struct pagevec pvec;
2423 if (mapping->nrpages == 0)
2426 pagevec_init(&pvec, 0);
2427 while (next < end &&
2428 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
2434 for (i = 0; i < pagevec_count(&pvec); i++) {
2435 struct page *page = pvec.pages[i];
2436 pgoff_t page_index = page->index;
2437 struct zone *pagezone = page_zone(page);
2440 if (page_index > next)
2444 if (pagezone != zone) {
2446 spin_unlock_irq(&zone->lru_lock);
2448 spin_lock_irq(&zone->lru_lock);
2451 if (PageLRU(page) && PageUnevictable(page))
2452 check_move_unevictable_page(page, zone);
2455 spin_unlock_irq(&zone->lru_lock);
2456 pagevec_release(&pvec);
2458 count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned);
2464 * scan_zone_unevictable_pages - check unevictable list for evictable pages
2465 * @zone - zone of which to scan the unevictable list
2467 * Scan @zone's unevictable LRU lists to check for pages that have become
2468 * evictable. Move those that have to @zone's inactive list where they
2469 * become candidates for reclaim, unless shrink_inactive_zone() decides
2470 * to reactivate them. Pages that are still unevictable are rotated
2471 * back onto @zone's unevictable list.
2473 #define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */
2474 static void scan_zone_unevictable_pages(struct zone *zone)
2476 struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list;
2478 unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE);
2480 while (nr_to_scan > 0) {
2481 unsigned long batch_size = min(nr_to_scan,
2482 SCAN_UNEVICTABLE_BATCH_SIZE);
2484 spin_lock_irq(&zone->lru_lock);
2485 for (scan = 0; scan < batch_size; scan++) {
2486 struct page *page = lru_to_page(l_unevictable);
2488 if (!trylock_page(page))
2491 prefetchw_prev_lru_page(page, l_unevictable, flags);
2493 if (likely(PageLRU(page) && PageUnevictable(page)))
2494 check_move_unevictable_page(page, zone);
2498 spin_unlock_irq(&zone->lru_lock);
2500 nr_to_scan -= batch_size;
2506 * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages
2508 * A really big hammer: scan all zones' unevictable LRU lists to check for
2509 * pages that have become evictable. Move those back to the zones'
2510 * inactive list where they become candidates for reclaim.
2511 * This occurs when, e.g., we have unswappable pages on the unevictable lists,
2512 * and we add swap to the system. As such, it runs in the context of a task
2513 * that has possibly/probably made some previously unevictable pages
2516 static void scan_all_zones_unevictable_pages(void)
2520 for_each_zone(zone) {
2521 scan_zone_unevictable_pages(zone);
2526 * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of
2527 * all nodes' unevictable lists for evictable pages
2529 unsigned long scan_unevictable_pages;
2531 int scan_unevictable_handler(struct ctl_table *table, int write,
2532 struct file *file, void __user *buffer,
2533 size_t *length, loff_t *ppos)
2535 proc_doulongvec_minmax(table, write, file, buffer, length, ppos);
2537 if (write && *(unsigned long *)table->data)
2538 scan_all_zones_unevictable_pages();
2540 scan_unevictable_pages = 0;
2545 * per node 'scan_unevictable_pages' attribute. On demand re-scan of
2546 * a specified node's per zone unevictable lists for evictable pages.
2549 static ssize_t read_scan_unevictable_node(struct sys_device *dev,
2550 struct sysdev_attribute *attr,
2553 return sprintf(buf, "0\n"); /* always zero; should fit... */
2556 static ssize_t write_scan_unevictable_node(struct sys_device *dev,
2557 struct sysdev_attribute *attr,
2558 const char *buf, size_t count)
2560 struct zone *node_zones = NODE_DATA(dev->id)->node_zones;
2563 unsigned long req = strict_strtoul(buf, 10, &res);
2566 return 1; /* zero is no-op */
2568 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2569 if (!populated_zone(zone))
2571 scan_zone_unevictable_pages(zone);
2577 static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
2578 read_scan_unevictable_node,
2579 write_scan_unevictable_node);
2581 int scan_unevictable_register_node(struct node *node)
2583 return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages);
2586 void scan_unevictable_unregister_node(struct node *node)
2588 sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages);