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>
40 #include <asm/tlbflush.h>
41 #include <asm/div64.h>
43 #include <linux/swapops.h>
48 /* Incremented by the number of inactive pages that were scanned */
49 unsigned long nr_scanned
;
51 /* This context's GFP mask */
56 /* Can pages be swapped as part of reclaim? */
59 /* This context's SWAP_CLUSTER_MAX. If freeing memory for
60 * suspend, we effectively ignore SWAP_CLUSTER_MAX.
61 * In this context, it doesn't matter that we scan the
62 * whole list at once. */
67 int all_unreclaimable
;
71 * The list of shrinker callbacks used by to apply pressure to
76 struct list_head list
;
77 int seeks
; /* seeks to recreate an obj */
78 long nr
; /* objs pending delete */
81 #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
83 #ifdef ARCH_HAS_PREFETCH
84 #define prefetch_prev_lru_page(_page, _base, _field) \
86 if ((_page)->lru.prev != _base) { \
89 prev = lru_to_page(&(_page->lru)); \
90 prefetch(&prev->_field); \
94 #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
97 #ifdef ARCH_HAS_PREFETCHW
98 #define prefetchw_prev_lru_page(_page, _base, _field) \
100 if ((_page)->lru.prev != _base) { \
103 prev = lru_to_page(&(_page->lru)); \
104 prefetchw(&prev->_field); \
108 #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
112 * From 0 .. 100. Higher means more swappy.
114 int vm_swappiness
= 60;
115 long vm_total_pages
; /* The total number of pages which the VM controls */
117 static LIST_HEAD(shrinker_list
);
118 static DECLARE_RWSEM(shrinker_rwsem
);
121 * Add a shrinker callback to be called from the vm
123 struct shrinker
*set_shrinker(int seeks
, shrinker_t theshrinker
)
125 struct shrinker
*shrinker
;
127 shrinker
= kmalloc(sizeof(*shrinker
), GFP_KERNEL
);
129 shrinker
->shrinker
= theshrinker
;
130 shrinker
->seeks
= seeks
;
132 down_write(&shrinker_rwsem
);
133 list_add_tail(&shrinker
->list
, &shrinker_list
);
134 up_write(&shrinker_rwsem
);
138 EXPORT_SYMBOL(set_shrinker
);
143 void remove_shrinker(struct shrinker
*shrinker
)
145 down_write(&shrinker_rwsem
);
146 list_del(&shrinker
->list
);
147 up_write(&shrinker_rwsem
);
150 EXPORT_SYMBOL(remove_shrinker
);
152 #define SHRINK_BATCH 128
154 * Call the shrink functions to age shrinkable caches
156 * Here we assume it costs one seek to replace a lru page and that it also
157 * takes a seek to recreate a cache object. With this in mind we age equal
158 * percentages of the lru and ageable caches. This should balance the seeks
159 * generated by these structures.
161 * If the vm encounted mapped pages on the LRU it increase the pressure on
162 * slab to avoid swapping.
164 * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
166 * `lru_pages' represents the number of on-LRU pages in all the zones which
167 * are eligible for the caller's allocation attempt. It is used for balancing
168 * slab reclaim versus page reclaim.
170 * Returns the number of slab objects which we shrunk.
172 unsigned long shrink_slab(unsigned long scanned
, gfp_t gfp_mask
,
173 unsigned long lru_pages
)
175 struct shrinker
*shrinker
;
176 unsigned long ret
= 0;
179 scanned
= SWAP_CLUSTER_MAX
;
181 if (!down_read_trylock(&shrinker_rwsem
))
182 return 1; /* Assume we'll be able to shrink next time */
184 list_for_each_entry(shrinker
, &shrinker_list
, list
) {
185 unsigned long long delta
;
186 unsigned long total_scan
;
187 unsigned long max_pass
= (*shrinker
->shrinker
)(0, gfp_mask
);
189 delta
= (4 * scanned
) / shrinker
->seeks
;
191 do_div(delta
, lru_pages
+ 1);
192 shrinker
->nr
+= delta
;
193 if (shrinker
->nr
< 0) {
194 printk(KERN_ERR
"%s: nr=%ld\n",
195 __FUNCTION__
, shrinker
->nr
);
196 shrinker
->nr
= max_pass
;
200 * Avoid risking looping forever due to too large nr value:
201 * never try to free more than twice the estimate number of
204 if (shrinker
->nr
> max_pass
* 2)
205 shrinker
->nr
= max_pass
* 2;
207 total_scan
= shrinker
->nr
;
210 while (total_scan
>= SHRINK_BATCH
) {
211 long this_scan
= SHRINK_BATCH
;
215 nr_before
= (*shrinker
->shrinker
)(0, gfp_mask
);
216 shrink_ret
= (*shrinker
->shrinker
)(this_scan
, gfp_mask
);
217 if (shrink_ret
== -1)
219 if (shrink_ret
< nr_before
)
220 ret
+= nr_before
- shrink_ret
;
221 count_vm_events(SLABS_SCANNED
, this_scan
);
222 total_scan
-= this_scan
;
227 shrinker
->nr
+= total_scan
;
229 up_read(&shrinker_rwsem
);
233 /* Called without lock on whether page is mapped, so answer is unstable */
234 static inline int page_mapping_inuse(struct page
*page
)
236 struct address_space
*mapping
;
238 /* Page is in somebody's page tables. */
239 if (page_mapped(page
))
242 /* Be more reluctant to reclaim swapcache than pagecache */
243 if (PageSwapCache(page
))
246 mapping
= page_mapping(page
);
250 /* File is mmap'd by somebody? */
251 return mapping_mapped(mapping
);
254 static inline int is_page_cache_freeable(struct page
*page
)
256 return page_count(page
) - !!PagePrivate(page
) == 2;
259 static int may_write_to_queue(struct backing_dev_info
*bdi
)
261 if (current
->flags
& PF_SWAPWRITE
)
263 if (!bdi_write_congested(bdi
))
265 if (bdi
== current
->backing_dev_info
)
271 * We detected a synchronous write error writing a page out. Probably
272 * -ENOSPC. We need to propagate that into the address_space for a subsequent
273 * fsync(), msync() or close().
275 * The tricky part is that after writepage we cannot touch the mapping: nothing
276 * prevents it from being freed up. But we have a ref on the page and once
277 * that page is locked, the mapping is pinned.
279 * We're allowed to run sleeping lock_page() here because we know the caller has
282 static void handle_write_error(struct address_space
*mapping
,
283 struct page
*page
, int error
)
286 if (page_mapping(page
) == mapping
) {
287 if (error
== -ENOSPC
)
288 set_bit(AS_ENOSPC
, &mapping
->flags
);
290 set_bit(AS_EIO
, &mapping
->flags
);
295 /* possible outcome of pageout() */
297 /* failed to write page out, page is locked */
299 /* move page to the active list, page is locked */
301 /* page has been sent to the disk successfully, page is unlocked */
303 /* page is clean and locked */
308 * pageout is called by shrink_page_list() for each dirty page.
309 * Calls ->writepage().
311 static pageout_t
pageout(struct page
*page
, struct address_space
*mapping
)
314 * If the page is dirty, only perform writeback if that write
315 * will be non-blocking. To prevent this allocation from being
316 * stalled by pagecache activity. But note that there may be
317 * stalls if we need to run get_block(). We could test
318 * PagePrivate for that.
320 * If this process is currently in generic_file_write() against
321 * this page's queue, we can perform writeback even if that
324 * If the page is swapcache, write it back even if that would
325 * block, for some throttling. This happens by accident, because
326 * swap_backing_dev_info is bust: it doesn't reflect the
327 * congestion state of the swapdevs. Easy to fix, if needed.
328 * See swapfile.c:page_queue_congested().
330 if (!is_page_cache_freeable(page
))
334 * Some data journaling orphaned pages can have
335 * page->mapping == NULL while being dirty with clean buffers.
337 if (PagePrivate(page
)) {
338 if (try_to_free_buffers(page
)) {
339 ClearPageDirty(page
);
340 printk("%s: orphaned page\n", __FUNCTION__
);
346 if (mapping
->a_ops
->writepage
== NULL
)
347 return PAGE_ACTIVATE
;
348 if (!may_write_to_queue(mapping
->backing_dev_info
))
351 if (clear_page_dirty_for_io(page
)) {
353 struct writeback_control wbc
= {
354 .sync_mode
= WB_SYNC_NONE
,
355 .nr_to_write
= SWAP_CLUSTER_MAX
,
357 .range_end
= LLONG_MAX
,
362 SetPageReclaim(page
);
363 res
= mapping
->a_ops
->writepage(page
, &wbc
);
365 handle_write_error(mapping
, page
, res
);
366 if (res
== AOP_WRITEPAGE_ACTIVATE
) {
367 ClearPageReclaim(page
);
368 return PAGE_ACTIVATE
;
370 if (!PageWriteback(page
)) {
371 /* synchronous write or broken a_ops? */
372 ClearPageReclaim(page
);
374 inc_zone_page_state(page
, NR_VMSCAN_WRITE
);
381 int remove_mapping(struct address_space
*mapping
, struct page
*page
)
383 BUG_ON(!PageLocked(page
));
384 BUG_ON(mapping
!= page_mapping(page
));
386 write_lock_irq(&mapping
->tree_lock
);
389 * The non-racy check for busy page. It is critical to check
390 * PageDirty _after_ making sure that the page is freeable and
391 * not in use by anybody. (pagecache + us == 2)
393 if (unlikely(page_count(page
) != 2))
396 if (unlikely(PageDirty(page
)))
399 if (PageSwapCache(page
)) {
400 swp_entry_t swap
= { .val
= page_private(page
) };
401 __delete_from_swap_cache(page
);
402 write_unlock_irq(&mapping
->tree_lock
);
404 __put_page(page
); /* The pagecache ref */
408 __remove_from_page_cache(page
);
409 write_unlock_irq(&mapping
->tree_lock
);
414 write_unlock_irq(&mapping
->tree_lock
);
419 * shrink_page_list() returns the number of reclaimed pages
421 static unsigned long shrink_page_list(struct list_head
*page_list
,
422 struct scan_control
*sc
)
424 LIST_HEAD(ret_pages
);
425 struct pagevec freed_pvec
;
427 unsigned long nr_reclaimed
= 0;
431 pagevec_init(&freed_pvec
, 1);
432 while (!list_empty(page_list
)) {
433 struct address_space
*mapping
;
440 page
= lru_to_page(page_list
);
441 list_del(&page
->lru
);
443 if (TestSetPageLocked(page
))
446 VM_BUG_ON(PageActive(page
));
450 if (!sc
->may_swap
&& page_mapped(page
))
453 /* Double the slab pressure for mapped and swapcache pages */
454 if (page_mapped(page
) || PageSwapCache(page
))
457 if (PageWriteback(page
))
460 referenced
= page_referenced(page
, 1);
461 /* In active use or really unfreeable? Activate it. */
462 if (referenced
&& page_mapping_inuse(page
))
463 goto activate_locked
;
467 * Anonymous process memory has backing store?
468 * Try to allocate it some swap space here.
470 if (PageAnon(page
) && !PageSwapCache(page
))
471 if (!add_to_swap(page
, GFP_ATOMIC
))
472 goto activate_locked
;
473 #endif /* CONFIG_SWAP */
475 mapping
= page_mapping(page
);
476 may_enter_fs
= (sc
->gfp_mask
& __GFP_FS
) ||
477 (PageSwapCache(page
) && (sc
->gfp_mask
& __GFP_IO
));
480 * The page is mapped into the page tables of one or more
481 * processes. Try to unmap it here.
483 if (page_mapped(page
) && mapping
) {
484 switch (try_to_unmap(page
, 0)) {
486 goto activate_locked
;
490 ; /* try to free the page below */
494 if (PageDirty(page
)) {
499 if (!sc
->may_writepage
)
502 /* Page is dirty, try to write it out here */
503 switch(pageout(page
, mapping
)) {
507 goto activate_locked
;
509 if (PageWriteback(page
) || PageDirty(page
))
512 * A synchronous write - probably a ramdisk. Go
513 * ahead and try to reclaim the page.
515 if (TestSetPageLocked(page
))
517 if (PageDirty(page
) || PageWriteback(page
))
519 mapping
= page_mapping(page
);
521 ; /* try to free the page below */
526 * If the page has buffers, try to free the buffer mappings
527 * associated with this page. If we succeed we try to free
530 * We do this even if the page is PageDirty().
531 * try_to_release_page() does not perform I/O, but it is
532 * possible for a page to have PageDirty set, but it is actually
533 * clean (all its buffers are clean). This happens if the
534 * buffers were written out directly, with submit_bh(). ext3
535 * will do this, as well as the blockdev mapping.
536 * try_to_release_page() will discover that cleanness and will
537 * drop the buffers and mark the page clean - it can be freed.
539 * Rarely, pages can have buffers and no ->mapping. These are
540 * the pages which were not successfully invalidated in
541 * truncate_complete_page(). We try to drop those buffers here
542 * and if that worked, and the page is no longer mapped into
543 * process address space (page_count == 1) it can be freed.
544 * Otherwise, leave the page on the LRU so it is swappable.
546 if (PagePrivate(page
)) {
547 if (!try_to_release_page(page
, sc
->gfp_mask
))
548 goto activate_locked
;
549 if (!mapping
&& page_count(page
) == 1)
553 if (!mapping
|| !remove_mapping(mapping
, page
))
559 if (!pagevec_add(&freed_pvec
, page
))
560 __pagevec_release_nonlru(&freed_pvec
);
569 list_add(&page
->lru
, &ret_pages
);
570 VM_BUG_ON(PageLRU(page
));
572 list_splice(&ret_pages
, page_list
);
573 if (pagevec_count(&freed_pvec
))
574 __pagevec_release_nonlru(&freed_pvec
);
575 count_vm_events(PGACTIVATE
, pgactivate
);
580 * zone->lru_lock is heavily contended. Some of the functions that
581 * shrink the lists perform better by taking out a batch of pages
582 * and working on them outside the LRU lock.
584 * For pagecache intensive workloads, this function is the hottest
585 * spot in the kernel (apart from copy_*_user functions).
587 * Appropriate locks must be held before calling this function.
589 * @nr_to_scan: The number of pages to look through on the list.
590 * @src: The LRU list to pull pages off.
591 * @dst: The temp list to put pages on to.
592 * @scanned: The number of pages that were scanned.
594 * returns how many pages were moved onto *@dst.
596 static unsigned long isolate_lru_pages(unsigned long nr_to_scan
,
597 struct list_head
*src
, struct list_head
*dst
,
598 unsigned long *scanned
)
600 unsigned long nr_taken
= 0;
604 for (scan
= 0; scan
< nr_to_scan
&& !list_empty(src
); scan
++) {
605 struct list_head
*target
;
606 page
= lru_to_page(src
);
607 prefetchw_prev_lru_page(page
, src
, flags
);
609 VM_BUG_ON(!PageLRU(page
));
611 list_del(&page
->lru
);
613 if (likely(get_page_unless_zero(page
))) {
615 * Be careful not to clear PageLRU until after we're
616 * sure the page is not being freed elsewhere -- the
617 * page release code relies on it.
622 } /* else it is being freed elsewhere */
624 list_add(&page
->lru
, target
);
632 * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
635 static unsigned long shrink_inactive_list(unsigned long max_scan
,
636 struct zone
*zone
, struct scan_control
*sc
)
638 LIST_HEAD(page_list
);
640 unsigned long nr_scanned
= 0;
641 unsigned long nr_reclaimed
= 0;
643 pagevec_init(&pvec
, 1);
646 spin_lock_irq(&zone
->lru_lock
);
649 unsigned long nr_taken
;
650 unsigned long nr_scan
;
651 unsigned long nr_freed
;
653 nr_taken
= isolate_lru_pages(sc
->swap_cluster_max
,
654 &zone
->inactive_list
,
655 &page_list
, &nr_scan
);
656 zone
->nr_inactive
-= nr_taken
;
657 zone
->pages_scanned
+= nr_scan
;
658 spin_unlock_irq(&zone
->lru_lock
);
660 nr_scanned
+= nr_scan
;
661 nr_freed
= shrink_page_list(&page_list
, sc
);
662 nr_reclaimed
+= nr_freed
;
664 if (current_is_kswapd()) {
665 __count_zone_vm_events(PGSCAN_KSWAPD
, zone
, nr_scan
);
666 __count_vm_events(KSWAPD_STEAL
, nr_freed
);
668 __count_zone_vm_events(PGSCAN_DIRECT
, zone
, nr_scan
);
669 __count_vm_events(PGACTIVATE
, nr_freed
);
674 spin_lock(&zone
->lru_lock
);
676 * Put back any unfreeable pages.
678 while (!list_empty(&page_list
)) {
679 page
= lru_to_page(&page_list
);
680 VM_BUG_ON(PageLRU(page
));
682 list_del(&page
->lru
);
683 if (PageActive(page
))
684 add_page_to_active_list(zone
, page
);
686 add_page_to_inactive_list(zone
, page
);
687 if (!pagevec_add(&pvec
, page
)) {
688 spin_unlock_irq(&zone
->lru_lock
);
689 __pagevec_release(&pvec
);
690 spin_lock_irq(&zone
->lru_lock
);
693 } while (nr_scanned
< max_scan
);
694 spin_unlock(&zone
->lru_lock
);
697 pagevec_release(&pvec
);
701 static inline int zone_is_near_oom(struct zone
*zone
)
703 return zone
->pages_scanned
>= (zone
->nr_active
+ zone
->nr_inactive
)*3;
707 * This moves pages from the active list to the inactive list.
709 * We move them the other way if the page is referenced by one or more
710 * processes, from rmap.
712 * If the pages are mostly unmapped, the processing is fast and it is
713 * appropriate to hold zone->lru_lock across the whole operation. But if
714 * the pages are mapped, the processing is slow (page_referenced()) so we
715 * should drop zone->lru_lock around each page. It's impossible to balance
716 * this, so instead we remove the pages from the LRU while processing them.
717 * It is safe to rely on PG_active against the non-LRU pages in here because
718 * nobody will play with that bit on a non-LRU page.
720 * The downside is that we have to touch page->_count against each page.
721 * But we had to alter page->flags anyway.
723 static void shrink_active_list(unsigned long nr_pages
, struct zone
*zone
,
724 struct scan_control
*sc
)
726 unsigned long pgmoved
;
727 int pgdeactivate
= 0;
728 unsigned long pgscanned
;
729 LIST_HEAD(l_hold
); /* The pages which were snipped off */
730 LIST_HEAD(l_inactive
); /* Pages to go onto the inactive_list */
731 LIST_HEAD(l_active
); /* Pages to go onto the active_list */
734 int reclaim_mapped
= 0;
741 if (zone_is_near_oom(zone
))
742 goto force_reclaim_mapped
;
745 * `distress' is a measure of how much trouble we're having
746 * reclaiming pages. 0 -> no problems. 100 -> great trouble.
748 distress
= 100 >> zone
->prev_priority
;
751 * The point of this algorithm is to decide when to start
752 * reclaiming mapped memory instead of just pagecache. Work out
756 mapped_ratio
= ((global_page_state(NR_FILE_MAPPED
) +
757 global_page_state(NR_ANON_PAGES
)) * 100) /
761 * Now decide how much we really want to unmap some pages. The
762 * mapped ratio is downgraded - just because there's a lot of
763 * mapped memory doesn't necessarily mean that page reclaim
766 * The distress ratio is important - we don't want to start
769 * A 100% value of vm_swappiness overrides this algorithm
772 swap_tendency
= mapped_ratio
/ 2 + distress
+ sc
->swappiness
;
775 * Now use this metric to decide whether to start moving mapped
776 * memory onto the inactive list.
778 if (swap_tendency
>= 100)
779 force_reclaim_mapped
:
784 spin_lock_irq(&zone
->lru_lock
);
785 pgmoved
= isolate_lru_pages(nr_pages
, &zone
->active_list
,
786 &l_hold
, &pgscanned
);
787 zone
->pages_scanned
+= pgscanned
;
788 zone
->nr_active
-= pgmoved
;
789 spin_unlock_irq(&zone
->lru_lock
);
791 while (!list_empty(&l_hold
)) {
793 page
= lru_to_page(&l_hold
);
794 list_del(&page
->lru
);
795 if (page_mapped(page
)) {
796 if (!reclaim_mapped
||
797 (total_swap_pages
== 0 && PageAnon(page
)) ||
798 page_referenced(page
, 0)) {
799 list_add(&page
->lru
, &l_active
);
803 list_add(&page
->lru
, &l_inactive
);
806 pagevec_init(&pvec
, 1);
808 spin_lock_irq(&zone
->lru_lock
);
809 while (!list_empty(&l_inactive
)) {
810 page
= lru_to_page(&l_inactive
);
811 prefetchw_prev_lru_page(page
, &l_inactive
, flags
);
812 VM_BUG_ON(PageLRU(page
));
814 VM_BUG_ON(!PageActive(page
));
815 ClearPageActive(page
);
817 list_move(&page
->lru
, &zone
->inactive_list
);
819 if (!pagevec_add(&pvec
, page
)) {
820 zone
->nr_inactive
+= pgmoved
;
821 spin_unlock_irq(&zone
->lru_lock
);
822 pgdeactivate
+= pgmoved
;
824 if (buffer_heads_over_limit
)
825 pagevec_strip(&pvec
);
826 __pagevec_release(&pvec
);
827 spin_lock_irq(&zone
->lru_lock
);
830 zone
->nr_inactive
+= pgmoved
;
831 pgdeactivate
+= pgmoved
;
832 if (buffer_heads_over_limit
) {
833 spin_unlock_irq(&zone
->lru_lock
);
834 pagevec_strip(&pvec
);
835 spin_lock_irq(&zone
->lru_lock
);
839 while (!list_empty(&l_active
)) {
840 page
= lru_to_page(&l_active
);
841 prefetchw_prev_lru_page(page
, &l_active
, flags
);
842 VM_BUG_ON(PageLRU(page
));
844 VM_BUG_ON(!PageActive(page
));
845 list_move(&page
->lru
, &zone
->active_list
);
847 if (!pagevec_add(&pvec
, page
)) {
848 zone
->nr_active
+= pgmoved
;
850 spin_unlock_irq(&zone
->lru_lock
);
851 __pagevec_release(&pvec
);
852 spin_lock_irq(&zone
->lru_lock
);
855 zone
->nr_active
+= pgmoved
;
857 __count_zone_vm_events(PGREFILL
, zone
, pgscanned
);
858 __count_vm_events(PGDEACTIVATE
, pgdeactivate
);
859 spin_unlock_irq(&zone
->lru_lock
);
861 pagevec_release(&pvec
);
865 * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
867 static unsigned long shrink_zone(int priority
, struct zone
*zone
,
868 struct scan_control
*sc
)
870 unsigned long nr_active
;
871 unsigned long nr_inactive
;
872 unsigned long nr_to_scan
;
873 unsigned long nr_reclaimed
= 0;
875 atomic_inc(&zone
->reclaim_in_progress
);
878 * Add one to `nr_to_scan' just to make sure that the kernel will
879 * slowly sift through the active list.
881 zone
->nr_scan_active
+= (zone
->nr_active
>> priority
) + 1;
882 nr_active
= zone
->nr_scan_active
;
883 if (nr_active
>= sc
->swap_cluster_max
)
884 zone
->nr_scan_active
= 0;
888 zone
->nr_scan_inactive
+= (zone
->nr_inactive
>> priority
) + 1;
889 nr_inactive
= zone
->nr_scan_inactive
;
890 if (nr_inactive
>= sc
->swap_cluster_max
)
891 zone
->nr_scan_inactive
= 0;
895 while (nr_active
|| nr_inactive
) {
897 nr_to_scan
= min(nr_active
,
898 (unsigned long)sc
->swap_cluster_max
);
899 nr_active
-= nr_to_scan
;
900 shrink_active_list(nr_to_scan
, zone
, sc
);
904 nr_to_scan
= min(nr_inactive
,
905 (unsigned long)sc
->swap_cluster_max
);
906 nr_inactive
-= nr_to_scan
;
907 nr_reclaimed
+= shrink_inactive_list(nr_to_scan
, zone
,
912 throttle_vm_writeout();
914 atomic_dec(&zone
->reclaim_in_progress
);
919 * This is the direct reclaim path, for page-allocating processes. We only
920 * try to reclaim pages from zones which will satisfy the caller's allocation
923 * We reclaim from a zone even if that zone is over pages_high. Because:
924 * a) The caller may be trying to free *extra* pages to satisfy a higher-order
926 * b) The zones may be over pages_high but they must go *over* pages_high to
927 * satisfy the `incremental min' zone defense algorithm.
929 * Returns the number of reclaimed pages.
931 * If a zone is deemed to be full of pinned pages then just give it a light
932 * scan then give up on it.
934 static unsigned long shrink_zones(int priority
, struct zone
**zones
,
935 struct scan_control
*sc
)
937 unsigned long nr_reclaimed
= 0;
940 sc
->all_unreclaimable
= 1;
941 for (i
= 0; zones
[i
] != NULL
; i
++) {
942 struct zone
*zone
= zones
[i
];
944 if (!populated_zone(zone
))
947 if (!cpuset_zone_allowed(zone
, __GFP_HARDWALL
))
950 zone
->temp_priority
= priority
;
951 if (zone
->prev_priority
> priority
)
952 zone
->prev_priority
= priority
;
954 if (zone
->all_unreclaimable
&& priority
!= DEF_PRIORITY
)
955 continue; /* Let kswapd poll it */
957 sc
->all_unreclaimable
= 0;
959 nr_reclaimed
+= shrink_zone(priority
, zone
, sc
);
965 * This is the main entry point to direct page reclaim.
967 * If a full scan of the inactive list fails to free enough memory then we
968 * are "out of memory" and something needs to be killed.
970 * If the caller is !__GFP_FS then the probability of a failure is reasonably
971 * high - the zone may be full of dirty or under-writeback pages, which this
972 * caller can't do much about. We kick pdflush and take explicit naps in the
973 * hope that some of these pages can be written. But if the allocating task
974 * holds filesystem locks which prevent writeout this might not work, and the
975 * allocation attempt will fail.
977 unsigned long try_to_free_pages(struct zone
**zones
, gfp_t gfp_mask
)
981 unsigned long total_scanned
= 0;
982 unsigned long nr_reclaimed
= 0;
983 struct reclaim_state
*reclaim_state
= current
->reclaim_state
;
984 unsigned long lru_pages
= 0;
986 struct scan_control sc
= {
987 .gfp_mask
= gfp_mask
,
988 .may_writepage
= !laptop_mode
,
989 .swap_cluster_max
= SWAP_CLUSTER_MAX
,
991 .swappiness
= vm_swappiness
,
994 count_vm_event(ALLOCSTALL
);
996 for (i
= 0; zones
[i
] != NULL
; i
++) {
997 struct zone
*zone
= zones
[i
];
999 if (!cpuset_zone_allowed(zone
, __GFP_HARDWALL
))
1002 zone
->temp_priority
= DEF_PRIORITY
;
1003 lru_pages
+= zone
->nr_active
+ zone
->nr_inactive
;
1006 for (priority
= DEF_PRIORITY
; priority
>= 0; priority
--) {
1009 disable_swap_token();
1010 nr_reclaimed
+= shrink_zones(priority
, zones
, &sc
);
1011 shrink_slab(sc
.nr_scanned
, gfp_mask
, lru_pages
);
1012 if (reclaim_state
) {
1013 nr_reclaimed
+= reclaim_state
->reclaimed_slab
;
1014 reclaim_state
->reclaimed_slab
= 0;
1016 total_scanned
+= sc
.nr_scanned
;
1017 if (nr_reclaimed
>= sc
.swap_cluster_max
) {
1023 * Try to write back as many pages as we just scanned. This
1024 * tends to cause slow streaming writers to write data to the
1025 * disk smoothly, at the dirtying rate, which is nice. But
1026 * that's undesirable in laptop mode, where we *want* lumpy
1027 * writeout. So in laptop mode, write out the whole world.
1029 if (total_scanned
> sc
.swap_cluster_max
+
1030 sc
.swap_cluster_max
/ 2) {
1031 wakeup_pdflush(laptop_mode
? 0 : total_scanned
);
1032 sc
.may_writepage
= 1;
1035 /* Take a nap, wait for some writeback to complete */
1036 if (sc
.nr_scanned
&& priority
< DEF_PRIORITY
- 2)
1037 blk_congestion_wait(WRITE
, HZ
/10);
1039 /* top priority shrink_caches still had more to do? don't OOM, then */
1040 if (!sc
.all_unreclaimable
)
1043 for (i
= 0; zones
[i
] != 0; i
++) {
1044 struct zone
*zone
= zones
[i
];
1046 if (!cpuset_zone_allowed(zone
, __GFP_HARDWALL
))
1049 zone
->prev_priority
= zone
->temp_priority
;
1055 * For kswapd, balance_pgdat() will work across all this node's zones until
1056 * they are all at pages_high.
1058 * Returns the number of pages which were actually freed.
1060 * There is special handling here for zones which are full of pinned pages.
1061 * This can happen if the pages are all mlocked, or if they are all used by
1062 * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
1063 * What we do is to detect the case where all pages in the zone have been
1064 * scanned twice and there has been zero successful reclaim. Mark the zone as
1065 * dead and from now on, only perform a short scan. Basically we're polling
1066 * the zone for when the problem goes away.
1068 * kswapd scans the zones in the highmem->normal->dma direction. It skips
1069 * zones which have free_pages > pages_high, but once a zone is found to have
1070 * free_pages <= pages_high, we scan that zone and the lower zones regardless
1071 * of the number of free pages in the lower zones. This interoperates with
1072 * the page allocator fallback scheme to ensure that aging of pages is balanced
1075 static unsigned long balance_pgdat(pg_data_t
*pgdat
, int order
)
1080 unsigned long total_scanned
;
1081 unsigned long nr_reclaimed
;
1082 struct reclaim_state
*reclaim_state
= current
->reclaim_state
;
1083 struct scan_control sc
= {
1084 .gfp_mask
= GFP_KERNEL
,
1086 .swap_cluster_max
= SWAP_CLUSTER_MAX
,
1087 .swappiness
= vm_swappiness
,
1093 sc
.may_writepage
= !laptop_mode
;
1094 count_vm_event(PAGEOUTRUN
);
1096 for (i
= 0; i
< pgdat
->nr_zones
; i
++) {
1097 struct zone
*zone
= pgdat
->node_zones
+ i
;
1099 zone
->temp_priority
= DEF_PRIORITY
;
1102 for (priority
= DEF_PRIORITY
; priority
>= 0; priority
--) {
1103 int end_zone
= 0; /* Inclusive. 0 = ZONE_DMA */
1104 unsigned long lru_pages
= 0;
1106 /* The swap token gets in the way of swapout... */
1108 disable_swap_token();
1113 * Scan in the highmem->dma direction for the highest
1114 * zone which needs scanning
1116 for (i
= pgdat
->nr_zones
- 1; i
>= 0; i
--) {
1117 struct zone
*zone
= pgdat
->node_zones
+ i
;
1119 if (!populated_zone(zone
))
1122 if (zone
->all_unreclaimable
&& priority
!= DEF_PRIORITY
)
1125 if (!zone_watermark_ok(zone
, order
, zone
->pages_high
,
1133 for (i
= 0; i
<= end_zone
; i
++) {
1134 struct zone
*zone
= pgdat
->node_zones
+ i
;
1136 lru_pages
+= zone
->nr_active
+ zone
->nr_inactive
;
1140 * Now scan the zone in the dma->highmem direction, stopping
1141 * at the last zone which needs scanning.
1143 * We do this because the page allocator works in the opposite
1144 * direction. This prevents the page allocator from allocating
1145 * pages behind kswapd's direction of progress, which would
1146 * cause too much scanning of the lower zones.
1148 for (i
= 0; i
<= end_zone
; i
++) {
1149 struct zone
*zone
= pgdat
->node_zones
+ i
;
1152 if (!populated_zone(zone
))
1155 if (zone
->all_unreclaimable
&& priority
!= DEF_PRIORITY
)
1158 if (!zone_watermark_ok(zone
, order
, zone
->pages_high
,
1161 zone
->temp_priority
= priority
;
1162 if (zone
->prev_priority
> priority
)
1163 zone
->prev_priority
= priority
;
1165 nr_reclaimed
+= shrink_zone(priority
, zone
, &sc
);
1166 reclaim_state
->reclaimed_slab
= 0;
1167 nr_slab
= shrink_slab(sc
.nr_scanned
, GFP_KERNEL
,
1169 nr_reclaimed
+= reclaim_state
->reclaimed_slab
;
1170 total_scanned
+= sc
.nr_scanned
;
1171 if (zone
->all_unreclaimable
)
1173 if (nr_slab
== 0 && zone
->pages_scanned
>=
1174 (zone
->nr_active
+ zone
->nr_inactive
) * 6)
1175 zone
->all_unreclaimable
= 1;
1177 * If we've done a decent amount of scanning and
1178 * the reclaim ratio is low, start doing writepage
1179 * even in laptop mode
1181 if (total_scanned
> SWAP_CLUSTER_MAX
* 2 &&
1182 total_scanned
> nr_reclaimed
+ nr_reclaimed
/ 2)
1183 sc
.may_writepage
= 1;
1186 break; /* kswapd: all done */
1188 * OK, kswapd is getting into trouble. Take a nap, then take
1189 * another pass across the zones.
1191 if (total_scanned
&& priority
< DEF_PRIORITY
- 2)
1192 blk_congestion_wait(WRITE
, HZ
/10);
1195 * We do this so kswapd doesn't build up large priorities for
1196 * example when it is freeing in parallel with allocators. It
1197 * matches the direct reclaim path behaviour in terms of impact
1198 * on zone->*_priority.
1200 if (nr_reclaimed
>= SWAP_CLUSTER_MAX
)
1204 for (i
= 0; i
< pgdat
->nr_zones
; i
++) {
1205 struct zone
*zone
= pgdat
->node_zones
+ i
;
1207 zone
->prev_priority
= zone
->temp_priority
;
1209 if (!all_zones_ok
) {
1214 return nr_reclaimed
;
1218 * The background pageout daemon, started as a kernel thread
1219 * from the init process.
1221 * This basically trickles out pages so that we have _some_
1222 * free memory available even if there is no other activity
1223 * that frees anything up. This is needed for things like routing
1224 * etc, where we otherwise might have all activity going on in
1225 * asynchronous contexts that cannot page things out.
1227 * If there are applications that are active memory-allocators
1228 * (most normal use), this basically shouldn't matter.
1230 static int kswapd(void *p
)
1232 unsigned long order
;
1233 pg_data_t
*pgdat
= (pg_data_t
*)p
;
1234 struct task_struct
*tsk
= current
;
1236 struct reclaim_state reclaim_state
= {
1237 .reclaimed_slab
= 0,
1241 cpumask
= node_to_cpumask(pgdat
->node_id
);
1242 if (!cpus_empty(cpumask
))
1243 set_cpus_allowed(tsk
, cpumask
);
1244 current
->reclaim_state
= &reclaim_state
;
1247 * Tell the memory management that we're a "memory allocator",
1248 * and that if we need more memory we should get access to it
1249 * regardless (see "__alloc_pages()"). "kswapd" should
1250 * never get caught in the normal page freeing logic.
1252 * (Kswapd normally doesn't need memory anyway, but sometimes
1253 * you need a small amount of memory in order to be able to
1254 * page out something else, and this flag essentially protects
1255 * us from recursively trying to free more memory as we're
1256 * trying to free the first piece of memory in the first place).
1258 tsk
->flags
|= PF_MEMALLOC
| PF_SWAPWRITE
| PF_KSWAPD
;
1262 unsigned long new_order
;
1266 prepare_to_wait(&pgdat
->kswapd_wait
, &wait
, TASK_INTERRUPTIBLE
);
1267 new_order
= pgdat
->kswapd_max_order
;
1268 pgdat
->kswapd_max_order
= 0;
1269 if (order
< new_order
) {
1271 * Don't sleep if someone wants a larger 'order'
1277 order
= pgdat
->kswapd_max_order
;
1279 finish_wait(&pgdat
->kswapd_wait
, &wait
);
1281 balance_pgdat(pgdat
, order
);
1287 * A zone is low on free memory, so wake its kswapd task to service it.
1289 void wakeup_kswapd(struct zone
*zone
, int order
)
1293 if (!populated_zone(zone
))
1296 pgdat
= zone
->zone_pgdat
;
1297 if (zone_watermark_ok(zone
, order
, zone
->pages_low
, 0, 0))
1299 if (pgdat
->kswapd_max_order
< order
)
1300 pgdat
->kswapd_max_order
= order
;
1301 if (!cpuset_zone_allowed(zone
, __GFP_HARDWALL
))
1303 if (!waitqueue_active(&pgdat
->kswapd_wait
))
1305 wake_up_interruptible(&pgdat
->kswapd_wait
);
1310 * Helper function for shrink_all_memory(). Tries to reclaim 'nr_pages' pages
1311 * from LRU lists system-wide, for given pass and priority, and returns the
1312 * number of reclaimed pages
1314 * For pass > 3 we also try to shrink the LRU lists that contain a few pages
1316 static unsigned long shrink_all_zones(unsigned long nr_pages
, int pass
,
1317 int prio
, struct scan_control
*sc
)
1320 unsigned long nr_to_scan
, ret
= 0;
1322 for_each_zone(zone
) {
1324 if (!populated_zone(zone
))
1327 if (zone
->all_unreclaimable
&& prio
!= DEF_PRIORITY
)
1330 /* For pass = 0 we don't shrink the active list */
1332 zone
->nr_scan_active
+= (zone
->nr_active
>> prio
) + 1;
1333 if (zone
->nr_scan_active
>= nr_pages
|| pass
> 3) {
1334 zone
->nr_scan_active
= 0;
1335 nr_to_scan
= min(nr_pages
, zone
->nr_active
);
1336 shrink_active_list(nr_to_scan
, zone
, sc
);
1340 zone
->nr_scan_inactive
+= (zone
->nr_inactive
>> prio
) + 1;
1341 if (zone
->nr_scan_inactive
>= nr_pages
|| pass
> 3) {
1342 zone
->nr_scan_inactive
= 0;
1343 nr_to_scan
= min(nr_pages
, zone
->nr_inactive
);
1344 ret
+= shrink_inactive_list(nr_to_scan
, zone
, sc
);
1345 if (ret
>= nr_pages
)
1354 * Try to free `nr_pages' of memory, system-wide, and return the number of
1357 * Rather than trying to age LRUs the aim is to preserve the overall
1358 * LRU order by reclaiming preferentially
1359 * inactive > active > active referenced > active mapped
1361 unsigned long shrink_all_memory(unsigned long nr_pages
)
1363 unsigned long lru_pages
, nr_slab
;
1364 unsigned long ret
= 0;
1366 struct reclaim_state reclaim_state
;
1368 struct scan_control sc
= {
1369 .gfp_mask
= GFP_KERNEL
,
1371 .swap_cluster_max
= nr_pages
,
1373 .swappiness
= vm_swappiness
,
1376 current
->reclaim_state
= &reclaim_state
;
1380 lru_pages
+= zone
->nr_active
+ zone
->nr_inactive
;
1382 nr_slab
= global_page_state(NR_SLAB_RECLAIMABLE
);
1383 /* If slab caches are huge, it's better to hit them first */
1384 while (nr_slab
>= lru_pages
) {
1385 reclaim_state
.reclaimed_slab
= 0;
1386 shrink_slab(nr_pages
, sc
.gfp_mask
, lru_pages
);
1387 if (!reclaim_state
.reclaimed_slab
)
1390 ret
+= reclaim_state
.reclaimed_slab
;
1391 if (ret
>= nr_pages
)
1394 nr_slab
-= reclaim_state
.reclaimed_slab
;
1398 * We try to shrink LRUs in 5 passes:
1399 * 0 = Reclaim from inactive_list only
1400 * 1 = Reclaim from active list but don't reclaim mapped
1401 * 2 = 2nd pass of type 1
1402 * 3 = Reclaim mapped (normal reclaim)
1403 * 4 = 2nd pass of type 3
1405 for (pass
= 0; pass
< 5; pass
++) {
1408 /* Needed for shrinking slab caches later on */
1410 for_each_zone(zone
) {
1411 lru_pages
+= zone
->nr_active
;
1412 lru_pages
+= zone
->nr_inactive
;
1415 /* Force reclaiming mapped pages in the passes #3 and #4 */
1418 sc
.swappiness
= 100;
1421 for (prio
= DEF_PRIORITY
; prio
>= 0; prio
--) {
1422 unsigned long nr_to_scan
= nr_pages
- ret
;
1425 ret
+= shrink_all_zones(nr_to_scan
, prio
, pass
, &sc
);
1426 if (ret
>= nr_pages
)
1429 reclaim_state
.reclaimed_slab
= 0;
1430 shrink_slab(sc
.nr_scanned
, sc
.gfp_mask
, lru_pages
);
1431 ret
+= reclaim_state
.reclaimed_slab
;
1432 if (ret
>= nr_pages
)
1435 if (sc
.nr_scanned
&& prio
< DEF_PRIORITY
- 2)
1436 blk_congestion_wait(WRITE
, HZ
/ 10);
1443 * If ret = 0, we could not shrink LRUs, but there may be something
1448 reclaim_state
.reclaimed_slab
= 0;
1449 shrink_slab(nr_pages
, sc
.gfp_mask
, lru_pages
);
1450 ret
+= reclaim_state
.reclaimed_slab
;
1451 } while (ret
< nr_pages
&& reclaim_state
.reclaimed_slab
> 0);
1454 current
->reclaim_state
= NULL
;
1460 #ifdef CONFIG_HOTPLUG_CPU
1461 /* It's optimal to keep kswapds on the same CPUs as their memory, but
1462 not required for correctness. So if the last cpu in a node goes
1463 away, we get changed to run anywhere: as the first one comes back,
1464 restore their cpu bindings. */
1465 static int __devinit
cpu_callback(struct notifier_block
*nfb
,
1466 unsigned long action
, void *hcpu
)
1471 if (action
== CPU_ONLINE
) {
1472 for_each_online_pgdat(pgdat
) {
1473 mask
= node_to_cpumask(pgdat
->node_id
);
1474 if (any_online_cpu(mask
) != NR_CPUS
)
1475 /* One of our CPUs online: restore mask */
1476 set_cpus_allowed(pgdat
->kswapd
, mask
);
1481 #endif /* CONFIG_HOTPLUG_CPU */
1484 * This kswapd start function will be called by init and node-hot-add.
1485 * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
1487 int kswapd_run(int nid
)
1489 pg_data_t
*pgdat
= NODE_DATA(nid
);
1495 pgdat
->kswapd
= kthread_run(kswapd
, pgdat
, "kswapd%d", nid
);
1496 if (IS_ERR(pgdat
->kswapd
)) {
1497 /* failure at boot is fatal */
1498 BUG_ON(system_state
== SYSTEM_BOOTING
);
1499 printk("Failed to start kswapd on node %d\n",nid
);
1505 static int __init
kswapd_init(void)
1510 for_each_online_node(nid
)
1512 hotcpu_notifier(cpu_callback
, 0);
1516 module_init(kswapd_init
)
1522 * If non-zero call zone_reclaim when the number of free pages falls below
1525 int zone_reclaim_mode __read_mostly
;
1527 #define RECLAIM_OFF 0
1528 #define RECLAIM_ZONE (1<<0) /* Run shrink_cache on the zone */
1529 #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
1530 #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
1533 * Priority for ZONE_RECLAIM. This determines the fraction of pages
1534 * of a node considered for each zone_reclaim. 4 scans 1/16th of
1537 #define ZONE_RECLAIM_PRIORITY 4
1540 * Percentage of pages in a zone that must be unmapped for zone_reclaim to
1543 int sysctl_min_unmapped_ratio
= 1;
1546 * If the number of slab pages in a zone grows beyond this percentage then
1547 * slab reclaim needs to occur.
1549 int sysctl_min_slab_ratio
= 5;
1552 * Try to free up some pages from this zone through reclaim.
1554 static int __zone_reclaim(struct zone
*zone
, gfp_t gfp_mask
, unsigned int order
)
1556 /* Minimum pages needed in order to stay on node */
1557 const unsigned long nr_pages
= 1 << order
;
1558 struct task_struct
*p
= current
;
1559 struct reclaim_state reclaim_state
;
1561 unsigned long nr_reclaimed
= 0;
1562 struct scan_control sc
= {
1563 .may_writepage
= !!(zone_reclaim_mode
& RECLAIM_WRITE
),
1564 .may_swap
= !!(zone_reclaim_mode
& RECLAIM_SWAP
),
1565 .swap_cluster_max
= max_t(unsigned long, nr_pages
,
1567 .gfp_mask
= gfp_mask
,
1568 .swappiness
= vm_swappiness
,
1570 unsigned long slab_reclaimable
;
1572 disable_swap_token();
1575 * We need to be able to allocate from the reserves for RECLAIM_SWAP
1576 * and we also need to be able to write out pages for RECLAIM_WRITE
1579 p
->flags
|= PF_MEMALLOC
| PF_SWAPWRITE
;
1580 reclaim_state
.reclaimed_slab
= 0;
1581 p
->reclaim_state
= &reclaim_state
;
1583 if (zone_page_state(zone
, NR_FILE_PAGES
) -
1584 zone_page_state(zone
, NR_FILE_MAPPED
) >
1585 zone
->min_unmapped_pages
) {
1587 * Free memory by calling shrink zone with increasing
1588 * priorities until we have enough memory freed.
1590 priority
= ZONE_RECLAIM_PRIORITY
;
1592 nr_reclaimed
+= shrink_zone(priority
, zone
, &sc
);
1594 } while (priority
>= 0 && nr_reclaimed
< nr_pages
);
1597 slab_reclaimable
= zone_page_state(zone
, NR_SLAB_RECLAIMABLE
);
1598 if (slab_reclaimable
> zone
->min_slab_pages
) {
1600 * shrink_slab() does not currently allow us to determine how
1601 * many pages were freed in this zone. So we take the current
1602 * number of slab pages and shake the slab until it is reduced
1603 * by the same nr_pages that we used for reclaiming unmapped
1606 * Note that shrink_slab will free memory on all zones and may
1609 while (shrink_slab(sc
.nr_scanned
, gfp_mask
, order
) &&
1610 zone_page_state(zone
, NR_SLAB_RECLAIMABLE
) >
1611 slab_reclaimable
- nr_pages
)
1615 * Update nr_reclaimed by the number of slab pages we
1616 * reclaimed from this zone.
1618 nr_reclaimed
+= slab_reclaimable
-
1619 zone_page_state(zone
, NR_SLAB_RECLAIMABLE
);
1622 p
->reclaim_state
= NULL
;
1623 current
->flags
&= ~(PF_MEMALLOC
| PF_SWAPWRITE
);
1624 return nr_reclaimed
>= nr_pages
;
1627 int zone_reclaim(struct zone
*zone
, gfp_t gfp_mask
, unsigned int order
)
1633 * Zone reclaim reclaims unmapped file backed pages and
1634 * slab pages if we are over the defined limits.
1636 * A small portion of unmapped file backed pages is needed for
1637 * file I/O otherwise pages read by file I/O will be immediately
1638 * thrown out if the zone is overallocated. So we do not reclaim
1639 * if less than a specified percentage of the zone is used by
1640 * unmapped file backed pages.
1642 if (zone_page_state(zone
, NR_FILE_PAGES
) -
1643 zone_page_state(zone
, NR_FILE_MAPPED
) <= zone
->min_unmapped_pages
1644 && zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)
1645 <= zone
->min_slab_pages
)
1649 * Avoid concurrent zone reclaims, do not reclaim in a zone that does
1650 * not have reclaimable pages and if we should not delay the allocation
1653 if (!(gfp_mask
& __GFP_WAIT
) ||
1654 zone
->all_unreclaimable
||
1655 atomic_read(&zone
->reclaim_in_progress
) > 0 ||
1656 (current
->flags
& PF_MEMALLOC
))
1660 * Only run zone reclaim on the local zone or on zones that do not
1661 * have associated processors. This will favor the local processor
1662 * over remote processors and spread off node memory allocations
1663 * as wide as possible.
1665 node_id
= zone_to_nid(zone
);
1666 mask
= node_to_cpumask(node_id
);
1667 if (!cpus_empty(mask
) && node_id
!= numa_node_id())
1669 return __zone_reclaim(zone
, gfp_mask
, order
);