blob: d254306562cda93af097d60ed024a71479c9da4d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/vmscan.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 *
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.
12 */
13
14#include <linux/mm.h>
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>
Andrew Mortone129b5c2006-09-27 01:50:00 -070022#include <linux/vmstat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#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>
Rafael J. Wysocki248a0302006-03-22 00:09:04 -080037#include <linux/delay.h>
Yasunori Goto3218ae12006-06-27 02:53:33 -070038#include <linux/kthread.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080039#include <linux/freezer.h>
Balbir Singh66e17072008-02-07 00:13:56 -080040#include <linux/memcontrol.h>
Keika Kobayashi873b4772008-07-25 01:48:52 -070041#include <linux/delayacct.h>
Lee Schermerhornaf936a12008-10-18 20:26:53 -070042#include <linux/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44#include <asm/tlbflush.h>
45#include <asm/div64.h>
46
47#include <linux/swapops.h>
48
Nick Piggin0f8053a2006-03-22 00:08:33 -080049#include "internal.h"
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051struct scan_control {
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 /* Incremented by the number of inactive pages that were scanned */
53 unsigned long nr_scanned;
54
Rik van Riela79311c2009-01-06 14:40:01 -080055 /* Number of pages freed so far during a call to shrink_zones() */
56 unsigned long nr_reclaimed;
57
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 /* This context's GFP mask */
Al Viro6daa0e22005-10-21 03:18:50 -040059 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61 int may_writepage;
62
Johannes Weinera6dc60f82009-03-31 15:19:30 -070063 /* Can mapped pages be reclaimed? */
64 int may_unmap;
Christoph Lameterf1fd1062006-01-18 17:42:30 -080065
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -070066 /* Can pages be swapped as part of reclaim? */
67 int may_swap;
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 /* This context's SWAP_CLUSTER_MAX. If freeing memory for
70 * suspend, we effectively ignore SWAP_CLUSTER_MAX.
71 * In this context, it doesn't matter that we scan the
72 * whole list at once. */
73 int swap_cluster_max;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -070074
75 int swappiness;
Nick Piggin408d8542006-09-25 23:31:27 -070076
77 int all_unreclaimable;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -070078
79 int order;
Balbir Singh66e17072008-02-07 00:13:56 -080080
81 /* Which cgroup do we reclaim from */
82 struct mem_cgroup *mem_cgroup;
83
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -070084 /*
85 * Nodemask of nodes allowed by the caller. If NULL, all nodes
86 * are scanned.
87 */
88 nodemask_t *nodemask;
89
Balbir Singh66e17072008-02-07 00:13:56 -080090 /* Pluggable isolate pages callback */
91 unsigned long (*isolate_pages)(unsigned long nr, struct list_head *dst,
92 unsigned long *scanned, int order, int mode,
93 struct zone *z, struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -070094 int active, int file);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095};
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
98
99#ifdef ARCH_HAS_PREFETCH
100#define prefetch_prev_lru_page(_page, _base, _field) \
101 do { \
102 if ((_page)->lru.prev != _base) { \
103 struct page *prev; \
104 \
105 prev = lru_to_page(&(_page->lru)); \
106 prefetch(&prev->_field); \
107 } \
108 } while (0)
109#else
110#define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
111#endif
112
113#ifdef ARCH_HAS_PREFETCHW
114#define prefetchw_prev_lru_page(_page, _base, _field) \
115 do { \
116 if ((_page)->lru.prev != _base) { \
117 struct page *prev; \
118 \
119 prev = lru_to_page(&(_page->lru)); \
120 prefetchw(&prev->_field); \
121 } \
122 } while (0)
123#else
124#define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
125#endif
126
127/*
128 * From 0 .. 100. Higher means more swappy.
129 */
130int vm_swappiness = 60;
Andrew Mortonbd1e22b2006-06-23 02:03:47 -0700131long vm_total_pages; /* The total number of pages which the VM controls */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133static LIST_HEAD(shrinker_list);
134static DECLARE_RWSEM(shrinker_rwsem);
135
Balbir Singh00f0b822008-03-04 14:28:39 -0800136#ifdef CONFIG_CGROUP_MEM_RES_CTLR
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800137#define scanning_global_lru(sc) (!(sc)->mem_cgroup)
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -0800138#else
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800139#define scanning_global_lru(sc) (1)
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -0800140#endif
141
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800142static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone,
143 struct scan_control *sc)
144{
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800145 if (!scanning_global_lru(sc))
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800146 return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone);
147
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800148 return &zone->reclaim_stat;
149}
150
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -0800151static unsigned long zone_nr_pages(struct zone *zone, struct scan_control *sc,
152 enum lru_list lru)
153{
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800154 if (!scanning_global_lru(sc))
KOSAKI Motohiroa3d8e052009-01-07 18:08:19 -0800155 return mem_cgroup_zone_nr_pages(sc->mem_cgroup, zone, lru);
156
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -0800157 return zone_page_state(zone, NR_LRU_BASE + lru);
158}
159
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161/*
162 * Add a shrinker callback to be called from the vm
163 */
Rusty Russell8e1f9362007-07-17 04:03:17 -0700164void register_shrinker(struct shrinker *shrinker)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165{
Rusty Russell8e1f9362007-07-17 04:03:17 -0700166 shrinker->nr = 0;
167 down_write(&shrinker_rwsem);
168 list_add_tail(&shrinker->list, &shrinker_list);
169 up_write(&shrinker_rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
Rusty Russell8e1f9362007-07-17 04:03:17 -0700171EXPORT_SYMBOL(register_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
173/*
174 * Remove one
175 */
Rusty Russell8e1f9362007-07-17 04:03:17 -0700176void unregister_shrinker(struct shrinker *shrinker)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
178 down_write(&shrinker_rwsem);
179 list_del(&shrinker->list);
180 up_write(&shrinker_rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181}
Rusty Russell8e1f9362007-07-17 04:03:17 -0700182EXPORT_SYMBOL(unregister_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184#define SHRINK_BATCH 128
185/*
186 * Call the shrink functions to age shrinkable caches
187 *
188 * Here we assume it costs one seek to replace a lru page and that it also
189 * takes a seek to recreate a cache object. With this in mind we age equal
190 * percentages of the lru and ageable caches. This should balance the seeks
191 * generated by these structures.
192 *
Simon Arlott183ff222007-10-20 01:27:18 +0200193 * If the vm encountered mapped pages on the LRU it increase the pressure on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 * slab to avoid swapping.
195 *
196 * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
197 *
198 * `lru_pages' represents the number of on-LRU pages in all the zones which
199 * are eligible for the caller's allocation attempt. It is used for balancing
200 * slab reclaim versus page reclaim.
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700201 *
202 * Returns the number of slab objects which we shrunk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 */
Andrew Morton69e05942006-03-22 00:08:19 -0800204unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask,
205 unsigned long lru_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
207 struct shrinker *shrinker;
Andrew Morton69e05942006-03-22 00:08:19 -0800208 unsigned long ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210 if (scanned == 0)
211 scanned = SWAP_CLUSTER_MAX;
212
213 if (!down_read_trylock(&shrinker_rwsem))
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700214 return 1; /* Assume we'll be able to shrink next time */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
216 list_for_each_entry(shrinker, &shrinker_list, list) {
217 unsigned long long delta;
218 unsigned long total_scan;
Rusty Russell8e1f9362007-07-17 04:03:17 -0700219 unsigned long max_pass = (*shrinker->shrink)(0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
221 delta = (4 * scanned) / shrinker->seeks;
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800222 delta *= max_pass;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 do_div(delta, lru_pages + 1);
224 shrinker->nr += delta;
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800225 if (shrinker->nr < 0) {
David Rientjes88c3bd72009-03-31 15:23:29 -0700226 printk(KERN_ERR "shrink_slab: %pF negative objects to "
227 "delete nr=%ld\n",
228 shrinker->shrink, shrinker->nr);
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800229 shrinker->nr = max_pass;
230 }
231
232 /*
233 * Avoid risking looping forever due to too large nr value:
234 * never try to free more than twice the estimate number of
235 * freeable entries.
236 */
237 if (shrinker->nr > max_pass * 2)
238 shrinker->nr = max_pass * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240 total_scan = shrinker->nr;
241 shrinker->nr = 0;
242
243 while (total_scan >= SHRINK_BATCH) {
244 long this_scan = SHRINK_BATCH;
245 int shrink_ret;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700246 int nr_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Rusty Russell8e1f9362007-07-17 04:03:17 -0700248 nr_before = (*shrinker->shrink)(0, gfp_mask);
249 shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 if (shrink_ret == -1)
251 break;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700252 if (shrink_ret < nr_before)
253 ret += nr_before - shrink_ret;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700254 count_vm_events(SLABS_SCANNED, this_scan);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 total_scan -= this_scan;
256
257 cond_resched();
258 }
259
260 shrinker->nr += total_scan;
261 }
262 up_read(&shrinker_rwsem);
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700263 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
266/* Called without lock on whether page is mapped, so answer is unstable */
267static inline int page_mapping_inuse(struct page *page)
268{
269 struct address_space *mapping;
270
271 /* Page is in somebody's page tables. */
272 if (page_mapped(page))
273 return 1;
274
275 /* Be more reluctant to reclaim swapcache than pagecache */
276 if (PageSwapCache(page))
277 return 1;
278
279 mapping = page_mapping(page);
280 if (!mapping)
281 return 0;
282
283 /* File is mmap'd by somebody? */
284 return mapping_mapped(mapping);
285}
286
287static inline int is_page_cache_freeable(struct page *page)
288{
David Howells266cf652009-04-03 16:42:36 +0100289 return page_count(page) - !!page_has_private(page) == 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291
292static int may_write_to_queue(struct backing_dev_info *bdi)
293{
Christoph Lameter930d9152006-01-08 01:00:47 -0800294 if (current->flags & PF_SWAPWRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return 1;
296 if (!bdi_write_congested(bdi))
297 return 1;
298 if (bdi == current->backing_dev_info)
299 return 1;
300 return 0;
301}
302
303/*
304 * We detected a synchronous write error writing a page out. Probably
305 * -ENOSPC. We need to propagate that into the address_space for a subsequent
306 * fsync(), msync() or close().
307 *
308 * The tricky part is that after writepage we cannot touch the mapping: nothing
309 * prevents it from being freed up. But we have a ref on the page and once
310 * that page is locked, the mapping is pinned.
311 *
312 * We're allowed to run sleeping lock_page() here because we know the caller has
313 * __GFP_FS.
314 */
315static void handle_write_error(struct address_space *mapping,
316 struct page *page, int error)
317{
318 lock_page(page);
Guillaume Chazarain3e9f45b2007-05-08 00:23:25 -0700319 if (page_mapping(page) == mapping)
320 mapping_set_error(mapping, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 unlock_page(page);
322}
323
Andy Whitcroftc661b072007-08-22 14:01:26 -0700324/* Request for sync pageout. */
325enum pageout_io {
326 PAGEOUT_IO_ASYNC,
327 PAGEOUT_IO_SYNC,
328};
329
Christoph Lameter04e62a22006-06-23 02:03:38 -0700330/* possible outcome of pageout() */
331typedef enum {
332 /* failed to write page out, page is locked */
333 PAGE_KEEP,
334 /* move page to the active list, page is locked */
335 PAGE_ACTIVATE,
336 /* page has been sent to the disk successfully, page is unlocked */
337 PAGE_SUCCESS,
338 /* page is clean and locked */
339 PAGE_CLEAN,
340} pageout_t;
341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342/*
Andrew Morton1742f192006-03-22 00:08:21 -0800343 * pageout is called by shrink_page_list() for each dirty page.
344 * Calls ->writepage().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 */
Andy Whitcroftc661b072007-08-22 14:01:26 -0700346static pageout_t pageout(struct page *page, struct address_space *mapping,
347 enum pageout_io sync_writeback)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
349 /*
350 * If the page is dirty, only perform writeback if that write
351 * will be non-blocking. To prevent this allocation from being
352 * stalled by pagecache activity. But note that there may be
353 * stalls if we need to run get_block(). We could test
354 * PagePrivate for that.
355 *
356 * If this process is currently in generic_file_write() against
357 * this page's queue, we can perform writeback even if that
358 * will block.
359 *
360 * If the page is swapcache, write it back even if that would
361 * block, for some throttling. This happens by accident, because
362 * swap_backing_dev_info is bust: it doesn't reflect the
363 * congestion state of the swapdevs. Easy to fix, if needed.
364 * See swapfile.c:page_queue_congested().
365 */
366 if (!is_page_cache_freeable(page))
367 return PAGE_KEEP;
368 if (!mapping) {
369 /*
370 * Some data journaling orphaned pages can have
371 * page->mapping == NULL while being dirty with clean buffers.
372 */
David Howells266cf652009-04-03 16:42:36 +0100373 if (page_has_private(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 if (try_to_free_buffers(page)) {
375 ClearPageDirty(page);
Harvey Harrisond40cee22008-04-30 00:55:07 -0700376 printk("%s: orphaned page\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 return PAGE_CLEAN;
378 }
379 }
380 return PAGE_KEEP;
381 }
382 if (mapping->a_ops->writepage == NULL)
383 return PAGE_ACTIVATE;
384 if (!may_write_to_queue(mapping->backing_dev_info))
385 return PAGE_KEEP;
386
387 if (clear_page_dirty_for_io(page)) {
388 int res;
389 struct writeback_control wbc = {
390 .sync_mode = WB_SYNC_NONE,
391 .nr_to_write = SWAP_CLUSTER_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700392 .range_start = 0,
393 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 .nonblocking = 1,
395 .for_reclaim = 1,
396 };
397
398 SetPageReclaim(page);
399 res = mapping->a_ops->writepage(page, &wbc);
400 if (res < 0)
401 handle_write_error(mapping, page, res);
Zach Brown994fc28c2005-12-15 14:28:17 -0800402 if (res == AOP_WRITEPAGE_ACTIVATE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 ClearPageReclaim(page);
404 return PAGE_ACTIVATE;
405 }
Andy Whitcroftc661b072007-08-22 14:01:26 -0700406
407 /*
408 * Wait on writeback if requested to. This happens when
409 * direct reclaiming a large contiguous area and the
410 * first attempt to free a range of pages fails.
411 */
412 if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC)
413 wait_on_page_writeback(page);
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 if (!PageWriteback(page)) {
416 /* synchronous write or broken a_ops? */
417 ClearPageReclaim(page);
418 }
Andrew Mortone129b5c2006-09-27 01:50:00 -0700419 inc_zone_page_state(page, NR_VMSCAN_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 return PAGE_SUCCESS;
421 }
422
423 return PAGE_CLEAN;
424}
425
Andrew Mortona649fd92006-10-17 00:09:36 -0700426/*
Nick Piggine2867812008-07-25 19:45:30 -0700427 * Same as remove_mapping, but if the page is removed from the mapping, it
428 * gets returned with a refcount of 0.
Andrew Mortona649fd92006-10-17 00:09:36 -0700429 */
Nick Piggine2867812008-07-25 19:45:30 -0700430static int __remove_mapping(struct address_space *mapping, struct page *page)
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800431{
Nick Piggin28e4d962006-09-25 23:31:23 -0700432 BUG_ON(!PageLocked(page));
433 BUG_ON(mapping != page_mapping(page));
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800434
Nick Piggin19fd6232008-07-25 19:45:32 -0700435 spin_lock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800436 /*
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700437 * The non racy check for a busy page.
438 *
439 * Must be careful with the order of the tests. When someone has
440 * a ref to the page, it may be possible that they dirty it then
441 * drop the reference. So if PageDirty is tested before page_count
442 * here, then the following race may occur:
443 *
444 * get_user_pages(&page);
445 * [user mapping goes away]
446 * write_to(page);
447 * !PageDirty(page) [good]
448 * SetPageDirty(page);
449 * put_page(page);
450 * !page_count(page) [good, discard it]
451 *
452 * [oops, our write_to data is lost]
453 *
454 * Reversing the order of the tests ensures such a situation cannot
455 * escape unnoticed. The smp_rmb is needed to ensure the page->flags
456 * load is not satisfied before that of page->_count.
457 *
458 * Note that if SetPageDirty is always performed via set_page_dirty,
459 * and thus under tree_lock, then this ordering is not required.
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800460 */
Nick Piggine2867812008-07-25 19:45:30 -0700461 if (!page_freeze_refs(page, 2))
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800462 goto cannot_free;
Nick Piggine2867812008-07-25 19:45:30 -0700463 /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
464 if (unlikely(PageDirty(page))) {
465 page_unfreeze_refs(page, 2);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800466 goto cannot_free;
Nick Piggine2867812008-07-25 19:45:30 -0700467 }
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800468
469 if (PageSwapCache(page)) {
470 swp_entry_t swap = { .val = page_private(page) };
471 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700472 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700473 mem_cgroup_uncharge_swapcache(page, swap);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800474 swap_free(swap);
Nick Piggine2867812008-07-25 19:45:30 -0700475 } else {
476 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700477 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700478 mem_cgroup_uncharge_cache_page(page);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800479 }
480
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800481 return 1;
482
483cannot_free:
Nick Piggin19fd6232008-07-25 19:45:32 -0700484 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800485 return 0;
486}
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488/*
Nick Piggine2867812008-07-25 19:45:30 -0700489 * Attempt to detach a locked page from its ->mapping. If it is dirty or if
490 * someone else has a ref on the page, abort and return 0. If it was
491 * successfully detached, return 1. Assumes the caller has a single ref on
492 * this page.
493 */
494int remove_mapping(struct address_space *mapping, struct page *page)
495{
496 if (__remove_mapping(mapping, page)) {
497 /*
498 * Unfreezing the refcount with 1 rather than 2 effectively
499 * drops the pagecache ref for us without requiring another
500 * atomic operation.
501 */
502 page_unfreeze_refs(page, 1);
503 return 1;
504 }
505 return 0;
506}
507
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700508/**
509 * putback_lru_page - put previously isolated page onto appropriate LRU list
510 * @page: page to be put back to appropriate lru list
511 *
512 * Add previously isolated @page to appropriate LRU list.
513 * Page may still be unevictable for other reasons.
514 *
515 * lru_lock must not be held, interrupts must be enabled.
516 */
517#ifdef CONFIG_UNEVICTABLE_LRU
518void putback_lru_page(struct page *page)
519{
520 int lru;
521 int active = !!TestClearPageActive(page);
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700522 int was_unevictable = PageUnevictable(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700523
524 VM_BUG_ON(PageLRU(page));
525
526redo:
527 ClearPageUnevictable(page);
528
529 if (page_evictable(page, NULL)) {
530 /*
531 * For evictable pages, we can use the cache.
532 * In event of a race, worst case is we end up with an
533 * unevictable page on [in]active list.
534 * We know how to handle that.
535 */
536 lru = active + page_is_file_cache(page);
537 lru_cache_add_lru(page, lru);
538 } else {
539 /*
540 * Put unevictable pages directly on zone's unevictable
541 * list.
542 */
543 lru = LRU_UNEVICTABLE;
544 add_page_to_unevictable_list(page);
545 }
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700546
547 /*
548 * page's status can change while we move it among lru. If an evictable
549 * page is on unevictable list, it never be freed. To avoid that,
550 * check after we added it to the list, again.
551 */
552 if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) {
553 if (!isolate_lru_page(page)) {
554 put_page(page);
555 goto redo;
556 }
557 /* This means someone else dropped this page from LRU
558 * So, it will be freed or putback to LRU again. There is
559 * nothing to do here.
560 */
561 }
562
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700563 if (was_unevictable && lru != LRU_UNEVICTABLE)
564 count_vm_event(UNEVICTABLE_PGRESCUED);
565 else if (!was_unevictable && lru == LRU_UNEVICTABLE)
566 count_vm_event(UNEVICTABLE_PGCULLED);
567
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700568 put_page(page); /* drop ref from isolate */
569}
570
571#else /* CONFIG_UNEVICTABLE_LRU */
572
573void putback_lru_page(struct page *page)
574{
575 int lru;
576 VM_BUG_ON(PageLRU(page));
577
578 lru = !!TestClearPageActive(page) + page_is_file_cache(page);
579 lru_cache_add_lru(page, lru);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700580 put_page(page);
581}
582#endif /* CONFIG_UNEVICTABLE_LRU */
583
584
Nick Piggine2867812008-07-25 19:45:30 -0700585/*
Andrew Morton1742f192006-03-22 00:08:21 -0800586 * shrink_page_list() returns the number of reclaimed pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
Andrew Morton1742f192006-03-22 00:08:21 -0800588static unsigned long shrink_page_list(struct list_head *page_list,
Andy Whitcroftc661b072007-08-22 14:01:26 -0700589 struct scan_control *sc,
590 enum pageout_io sync_writeback)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
592 LIST_HEAD(ret_pages);
593 struct pagevec freed_pvec;
594 int pgactivate = 0;
Andrew Morton05ff5132006-03-22 00:08:20 -0800595 unsigned long nr_reclaimed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 cond_resched();
598
599 pagevec_init(&freed_pvec, 1);
600 while (!list_empty(page_list)) {
601 struct address_space *mapping;
602 struct page *page;
603 int may_enter_fs;
604 int referenced;
605
606 cond_resched();
607
608 page = lru_to_page(page_list);
609 list_del(&page->lru);
610
Nick Piggin529ae9a2008-08-02 12:01:03 +0200611 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 goto keep;
613
Nick Piggin725d7042006-09-25 23:30:55 -0700614 VM_BUG_ON(PageActive(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
616 sc->nr_scanned++;
Christoph Lameter80e43422006-02-11 17:55:53 -0800617
Nick Pigginb291f002008-10-18 20:26:44 -0700618 if (unlikely(!page_evictable(page, NULL)))
619 goto cull_mlocked;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700620
Johannes Weinera6dc60f82009-03-31 15:19:30 -0700621 if (!sc->may_unmap && page_mapped(page))
Christoph Lameter80e43422006-02-11 17:55:53 -0800622 goto keep_locked;
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 /* Double the slab pressure for mapped and swapcache pages */
625 if (page_mapped(page) || PageSwapCache(page))
626 sc->nr_scanned++;
627
Andy Whitcroftc661b072007-08-22 14:01:26 -0700628 may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
629 (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
630
631 if (PageWriteback(page)) {
632 /*
633 * Synchronous reclaim is performed in two passes,
634 * first an asynchronous pass over the list to
635 * start parallel writeback, and a second synchronous
636 * pass to wait for the IO to complete. Wait here
637 * for any page for which writeback has already
638 * started.
639 */
640 if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs)
641 wait_on_page_writeback(page);
Andrew Morton4dd4b922008-03-24 12:29:52 -0700642 else
Andy Whitcroftc661b072007-08-22 14:01:26 -0700643 goto keep_locked;
644 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Balbir Singhbed71612008-02-07 00:14:01 -0800646 referenced = page_referenced(page, 1, sc->mem_cgroup);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 /* In active use or really unfreeable? Activate it. */
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700648 if (sc->order <= PAGE_ALLOC_COSTLY_ORDER &&
649 referenced && page_mapping_inuse(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 goto activate_locked;
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 /*
653 * Anonymous process memory has backing store?
654 * Try to allocate it some swap space here.
655 */
Nick Pigginb291f002008-10-18 20:26:44 -0700656 if (PageAnon(page) && !PageSwapCache(page)) {
Hugh Dickins63eb6b92008-11-19 15:36:37 -0800657 if (!(sc->gfp_mask & __GFP_IO))
658 goto keep_locked;
Hugh Dickinsac47b002009-01-06 14:39:39 -0800659 if (!add_to_swap(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 goto activate_locked;
Hugh Dickins63eb6b92008-11-19 15:36:37 -0800661 may_enter_fs = 1;
Nick Pigginb291f002008-10-18 20:26:44 -0700662 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 mapping = page_mapping(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
666 /*
667 * The page is mapped into the page tables of one or more
668 * processes. Try to unmap it here.
669 */
670 if (page_mapped(page) && mapping) {
Christoph Lametera48d07a2006-02-01 03:05:38 -0800671 switch (try_to_unmap(page, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 case SWAP_FAIL:
673 goto activate_locked;
674 case SWAP_AGAIN:
675 goto keep_locked;
Nick Pigginb291f002008-10-18 20:26:44 -0700676 case SWAP_MLOCK:
677 goto cull_mlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 case SWAP_SUCCESS:
679 ; /* try to free the page below */
680 }
681 }
682
683 if (PageDirty(page)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700684 if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && referenced)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 goto keep_locked;
Andrew Morton4dd4b922008-03-24 12:29:52 -0700686 if (!may_enter_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 goto keep_locked;
Christoph Lameter52a83632006-02-01 03:05:28 -0800688 if (!sc->may_writepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 goto keep_locked;
690
691 /* Page is dirty, try to write it out here */
Andy Whitcroftc661b072007-08-22 14:01:26 -0700692 switch (pageout(page, mapping, sync_writeback)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 case PAGE_KEEP:
694 goto keep_locked;
695 case PAGE_ACTIVATE:
696 goto activate_locked;
697 case PAGE_SUCCESS:
Andrew Morton4dd4b922008-03-24 12:29:52 -0700698 if (PageWriteback(page) || PageDirty(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 goto keep;
700 /*
701 * A synchronous write - probably a ramdisk. Go
702 * ahead and try to reclaim the page.
703 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200704 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 goto keep;
706 if (PageDirty(page) || PageWriteback(page))
707 goto keep_locked;
708 mapping = page_mapping(page);
709 case PAGE_CLEAN:
710 ; /* try to free the page below */
711 }
712 }
713
714 /*
715 * If the page has buffers, try to free the buffer mappings
716 * associated with this page. If we succeed we try to free
717 * the page as well.
718 *
719 * We do this even if the page is PageDirty().
720 * try_to_release_page() does not perform I/O, but it is
721 * possible for a page to have PageDirty set, but it is actually
722 * clean (all its buffers are clean). This happens if the
723 * buffers were written out directly, with submit_bh(). ext3
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700724 * will do this, as well as the blockdev mapping.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 * try_to_release_page() will discover that cleanness and will
726 * drop the buffers and mark the page clean - it can be freed.
727 *
728 * Rarely, pages can have buffers and no ->mapping. These are
729 * the pages which were not successfully invalidated in
730 * truncate_complete_page(). We try to drop those buffers here
731 * and if that worked, and the page is no longer mapped into
732 * process address space (page_count == 1) it can be freed.
733 * Otherwise, leave the page on the LRU so it is swappable.
734 */
David Howells266cf652009-04-03 16:42:36 +0100735 if (page_has_private(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 if (!try_to_release_page(page, sc->gfp_mask))
737 goto activate_locked;
Nick Piggine2867812008-07-25 19:45:30 -0700738 if (!mapping && page_count(page) == 1) {
739 unlock_page(page);
740 if (put_page_testzero(page))
741 goto free_it;
742 else {
743 /*
744 * rare race with speculative reference.
745 * the speculative reference will free
746 * this page shortly, so we may
747 * increment nr_reclaimed here (and
748 * leave it off the LRU).
749 */
750 nr_reclaimed++;
751 continue;
752 }
753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
755
Nick Piggine2867812008-07-25 19:45:30 -0700756 if (!mapping || !__remove_mapping(mapping, page))
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800757 goto keep_locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Nick Piggina978d6f2008-10-18 20:26:58 -0700759 /*
760 * At this point, we have no other references and there is
761 * no way to pick any more up (removed from LRU, removed
762 * from pagecache). Can use non-atomic bitops now (and
763 * we obviously don't have to worry about waking up a process
764 * waiting on the page lock, because there are no references.
765 */
766 __clear_page_locked(page);
Nick Piggine2867812008-07-25 19:45:30 -0700767free_it:
Andrew Morton05ff5132006-03-22 00:08:20 -0800768 nr_reclaimed++;
Nick Piggine2867812008-07-25 19:45:30 -0700769 if (!pagevec_add(&freed_pvec, page)) {
770 __pagevec_free(&freed_pvec);
771 pagevec_reinit(&freed_pvec);
772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 continue;
774
Nick Pigginb291f002008-10-18 20:26:44 -0700775cull_mlocked:
Hugh Dickins63d6c5a2009-01-06 14:39:38 -0800776 if (PageSwapCache(page))
777 try_to_free_swap(page);
Nick Pigginb291f002008-10-18 20:26:44 -0700778 unlock_page(page);
779 putback_lru_page(page);
780 continue;
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782activate_locked:
Rik van Riel68a223942008-10-18 20:26:23 -0700783 /* Not a candidate for swapping, so reclaim swap space. */
784 if (PageSwapCache(page) && vm_swap_full())
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800785 try_to_free_swap(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700786 VM_BUG_ON(PageActive(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 SetPageActive(page);
788 pgactivate++;
789keep_locked:
790 unlock_page(page);
791keep:
792 list_add(&page->lru, &ret_pages);
Nick Pigginb291f002008-10-18 20:26:44 -0700793 VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 }
795 list_splice(&ret_pages, page_list);
796 if (pagevec_count(&freed_pvec))
Nick Piggine2867812008-07-25 19:45:30 -0700797 __pagevec_free(&freed_pvec);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700798 count_vm_events(PGACTIVATE, pgactivate);
Andrew Morton05ff5132006-03-22 00:08:20 -0800799 return nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700802/* LRU Isolation modes. */
803#define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */
804#define ISOLATE_ACTIVE 1 /* Isolate active pages. */
805#define ISOLATE_BOTH 2 /* Isolate both active and inactive pages. */
806
807/*
808 * Attempt to remove the specified page from its LRU. Only take this page
809 * if it is of the appropriate PageActive status. Pages which are being
810 * freed elsewhere are also ignored.
811 *
812 * page: page to consider
813 * mode: one of the LRU isolation modes defined above
814 *
815 * returns 0 on success, -ve errno on failure.
816 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700817int __isolate_lru_page(struct page *page, int mode, int file)
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700818{
819 int ret = -EINVAL;
820
821 /* Only take pages on the LRU. */
822 if (!PageLRU(page))
823 return ret;
824
825 /*
826 * When checking the active state, we need to be sure we are
827 * dealing with comparible boolean values. Take the logical not
828 * of each.
829 */
830 if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
831 return ret;
832
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700833 if (mode != ISOLATE_BOTH && (!page_is_file_cache(page) != !file))
834 return ret;
835
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700836 /*
837 * When this function is being called for lumpy reclaim, we
838 * initially look into all LRU pages, active, inactive and
839 * unevictable; only give shrink_page_list evictable pages.
840 */
841 if (PageUnevictable(page))
842 return ret;
843
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700844 ret = -EBUSY;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800845
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700846 if (likely(get_page_unless_zero(page))) {
847 /*
848 * Be careful not to clear PageLRU until after we're
849 * sure the page is not being freed elsewhere -- the
850 * page release code relies on it.
851 */
852 ClearPageLRU(page);
853 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800854 mem_cgroup_del_lru(page);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700855 }
856
857 return ret;
858}
859
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800860/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 * zone->lru_lock is heavily contended. Some of the functions that
862 * shrink the lists perform better by taking out a batch of pages
863 * and working on them outside the LRU lock.
864 *
865 * For pagecache intensive workloads, this function is the hottest
866 * spot in the kernel (apart from copy_*_user functions).
867 *
868 * Appropriate locks must be held before calling this function.
869 *
870 * @nr_to_scan: The number of pages to look through on the list.
871 * @src: The LRU list to pull pages off.
872 * @dst: The temp list to put pages on to.
873 * @scanned: The number of pages that were scanned.
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700874 * @order: The caller's attempted allocation order
875 * @mode: One of the LRU isolation modes
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700876 * @file: True [1] if isolating file [!anon] pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 *
878 * returns how many pages were moved onto *@dst.
879 */
Andrew Morton69e05942006-03-22 00:08:19 -0800880static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
881 struct list_head *src, struct list_head *dst,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700882 unsigned long *scanned, int order, int mode, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883{
Andrew Morton69e05942006-03-22 00:08:19 -0800884 unsigned long nr_taken = 0;
Wu Fengguangc9b02d92006-03-22 00:08:23 -0800885 unsigned long scan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Wu Fengguangc9b02d92006-03-22 00:08:23 -0800887 for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700888 struct page *page;
889 unsigned long pfn;
890 unsigned long end_pfn;
891 unsigned long page_pfn;
892 int zone_id;
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 page = lru_to_page(src);
895 prefetchw_prev_lru_page(page, src, flags);
896
Nick Piggin725d7042006-09-25 23:30:55 -0700897 VM_BUG_ON(!PageLRU(page));
Nick Piggin8d438f92006-03-22 00:07:59 -0800898
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700899 switch (__isolate_lru_page(page, mode, file)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700900 case 0:
901 list_move(&page->lru, dst);
Nick Piggin7c8ee9a2006-03-22 00:08:03 -0800902 nr_taken++;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700903 break;
Nick Piggin46453a62006-03-22 00:07:58 -0800904
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700905 case -EBUSY:
906 /* else it is being freed elsewhere */
907 list_move(&page->lru, src);
908 continue;
909
910 default:
911 BUG();
912 }
913
914 if (!order)
915 continue;
916
917 /*
918 * Attempt to take all pages in the order aligned region
919 * surrounding the tag page. Only take those pages of
920 * the same active state as that tag page. We may safely
921 * round the target page pfn down to the requested order
922 * as the mem_map is guarenteed valid out to MAX_ORDER,
923 * where that page is in a different zone we will detect
924 * it from its zone id and abort this block scan.
925 */
926 zone_id = page_zone_id(page);
927 page_pfn = page_to_pfn(page);
928 pfn = page_pfn & ~((1 << order) - 1);
929 end_pfn = pfn + (1 << order);
930 for (; pfn < end_pfn; pfn++) {
931 struct page *cursor_page;
932
933 /* The target page is in the block, ignore it. */
934 if (unlikely(pfn == page_pfn))
935 continue;
936
937 /* Avoid holes within the zone. */
938 if (unlikely(!pfn_valid_within(pfn)))
939 break;
940
941 cursor_page = pfn_to_page(pfn);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700942
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700943 /* Check that we have not crossed a zone boundary. */
944 if (unlikely(page_zone_id(cursor_page) != zone_id))
945 continue;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700946 switch (__isolate_lru_page(cursor_page, mode, file)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700947 case 0:
948 list_move(&cursor_page->lru, dst);
949 nr_taken++;
950 scan++;
951 break;
952
953 case -EBUSY:
954 /* else it is being freed elsewhere */
955 list_move(&cursor_page->lru, src);
956 default:
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700957 break; /* ! on LRU or wrong list */
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700958 }
959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
961
962 *scanned = scan;
963 return nr_taken;
964}
965
Balbir Singh66e17072008-02-07 00:13:56 -0800966static unsigned long isolate_pages_global(unsigned long nr,
967 struct list_head *dst,
968 unsigned long *scanned, int order,
969 int mode, struct zone *z,
970 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700971 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -0800972{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700973 int lru = LRU_BASE;
Balbir Singh66e17072008-02-07 00:13:56 -0800974 if (active)
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700975 lru += LRU_ACTIVE;
976 if (file)
977 lru += LRU_FILE;
978 return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order,
979 mode, !!file);
Balbir Singh66e17072008-02-07 00:13:56 -0800980}
981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982/*
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700983 * clear_active_flags() is a helper for shrink_active_list(), clearing
984 * any active bits from the pages in the list.
985 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700986static unsigned long clear_active_flags(struct list_head *page_list,
987 unsigned int *count)
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700988{
989 int nr_active = 0;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700990 int lru;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700991 struct page *page;
992
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700993 list_for_each_entry(page, page_list, lru) {
994 lru = page_is_file_cache(page);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700995 if (PageActive(page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700996 lru += LRU_ACTIVE;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700997 ClearPageActive(page);
998 nr_active++;
999 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001000 count[lru]++;
1001 }
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001002
1003 return nr_active;
1004}
1005
Nick Piggin62695a82008-10-18 20:26:09 -07001006/**
1007 * isolate_lru_page - tries to isolate a page from its LRU list
1008 * @page: page to isolate from its LRU list
1009 *
1010 * Isolates a @page from an LRU list, clears PageLRU and adjusts the
1011 * vmstat statistic corresponding to whatever LRU list the page was on.
1012 *
1013 * Returns 0 if the page was removed from an LRU list.
1014 * Returns -EBUSY if the page was not on an LRU list.
1015 *
1016 * The returned page will have PageLRU() cleared. If it was found on
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001017 * the active list, it will have PageActive set. If it was found on
1018 * the unevictable list, it will have the PageUnevictable bit set. That flag
1019 * may need to be cleared by the caller before letting the page go.
Nick Piggin62695a82008-10-18 20:26:09 -07001020 *
1021 * The vmstat statistic corresponding to the list on which the page was
1022 * found will be decremented.
1023 *
1024 * Restrictions:
1025 * (1) Must be called with an elevated refcount on the page. This is a
1026 * fundamentnal difference from isolate_lru_pages (which is called
1027 * without a stable reference).
1028 * (2) the lru_lock must not be held.
1029 * (3) interrupts must be enabled.
1030 */
1031int isolate_lru_page(struct page *page)
1032{
1033 int ret = -EBUSY;
1034
1035 if (PageLRU(page)) {
1036 struct zone *zone = page_zone(page);
1037
1038 spin_lock_irq(&zone->lru_lock);
1039 if (PageLRU(page) && get_page_unless_zero(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001040 int lru = page_lru(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001041 ret = 0;
1042 ClearPageLRU(page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001043
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001044 del_page_from_lru_list(zone, page, lru);
Nick Piggin62695a82008-10-18 20:26:09 -07001045 }
1046 spin_unlock_irq(&zone->lru_lock);
1047 }
1048 return ret;
1049}
1050
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001051/*
Andrew Morton1742f192006-03-22 00:08:21 -08001052 * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
1053 * of reclaimed pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 */
Andrew Morton1742f192006-03-22 00:08:21 -08001055static unsigned long shrink_inactive_list(unsigned long max_scan,
Rik van Riel33c120e2008-10-18 20:26:36 -07001056 struct zone *zone, struct scan_control *sc,
1057 int priority, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
1059 LIST_HEAD(page_list);
1060 struct pagevec pvec;
Andrew Morton69e05942006-03-22 00:08:19 -08001061 unsigned long nr_scanned = 0;
Andrew Morton05ff5132006-03-22 00:08:20 -08001062 unsigned long nr_reclaimed = 0;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001063 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065 pagevec_init(&pvec, 1);
1066
1067 lru_add_drain();
1068 spin_lock_irq(&zone->lru_lock);
Andrew Morton69e05942006-03-22 00:08:19 -08001069 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 struct page *page;
Andrew Morton69e05942006-03-22 00:08:19 -08001071 unsigned long nr_taken;
1072 unsigned long nr_scan;
1073 unsigned long nr_freed;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001074 unsigned long nr_active;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001075 unsigned int count[NR_LRU_LISTS] = { 0, };
Rik van Riel33c120e2008-10-18 20:26:36 -07001076 int mode = ISOLATE_INACTIVE;
1077
1078 /*
1079 * If we need a large contiguous chunk of memory, or have
1080 * trouble getting a small set of contiguous pages, we
1081 * will reclaim both active and inactive pages.
1082 *
1083 * We use the same threshold as pageout congestion_wait below.
1084 */
1085 if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
1086 mode = ISOLATE_BOTH;
1087 else if (sc->order && priority < DEF_PRIORITY - 2)
1088 mode = ISOLATE_BOTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Balbir Singh66e17072008-02-07 00:13:56 -08001090 nr_taken = sc->isolate_pages(sc->swap_cluster_max,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001091 &page_list, &nr_scan, sc->order, mode,
1092 zone, sc->mem_cgroup, 0, file);
1093 nr_active = clear_active_flags(&page_list, count);
Andy Whitcrofte9187bd2007-08-22 14:01:25 -07001094 __count_vm_events(PGDEACTIVATE, nr_active);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001095
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001096 __mod_zone_page_state(zone, NR_ACTIVE_FILE,
1097 -count[LRU_ACTIVE_FILE]);
1098 __mod_zone_page_state(zone, NR_INACTIVE_FILE,
1099 -count[LRU_INACTIVE_FILE]);
1100 __mod_zone_page_state(zone, NR_ACTIVE_ANON,
1101 -count[LRU_ACTIVE_ANON]);
1102 __mod_zone_page_state(zone, NR_INACTIVE_ANON,
1103 -count[LRU_INACTIVE_ANON]);
1104
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001105 if (scanning_global_lru(sc))
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001106 zone->pages_scanned += nr_scan;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001107
1108 reclaim_stat->recent_scanned[0] += count[LRU_INACTIVE_ANON];
1109 reclaim_stat->recent_scanned[0] += count[LRU_ACTIVE_ANON];
1110 reclaim_stat->recent_scanned[1] += count[LRU_INACTIVE_FILE];
1111 reclaim_stat->recent_scanned[1] += count[LRU_ACTIVE_FILE];
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 spin_unlock_irq(&zone->lru_lock);
1114
Andrew Morton69e05942006-03-22 00:08:19 -08001115 nr_scanned += nr_scan;
Andy Whitcroftc661b072007-08-22 14:01:26 -07001116 nr_freed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC);
1117
1118 /*
1119 * If we are direct reclaiming for contiguous pages and we do
1120 * not reclaim everything in the list, try again and wait
1121 * for IO to complete. This will stall high-order allocations
1122 * but that should be acceptable to the caller
1123 */
1124 if (nr_freed < nr_taken && !current_is_kswapd() &&
1125 sc->order > PAGE_ALLOC_COSTLY_ORDER) {
1126 congestion_wait(WRITE, HZ/10);
1127
1128 /*
1129 * The attempt at page out may have made some
1130 * of the pages active, mark them inactive again.
1131 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001132 nr_active = clear_active_flags(&page_list, count);
Andy Whitcroftc661b072007-08-22 14:01:26 -07001133 count_vm_events(PGDEACTIVATE, nr_active);
1134
1135 nr_freed += shrink_page_list(&page_list, sc,
1136 PAGEOUT_IO_SYNC);
1137 }
1138
Andrew Morton05ff5132006-03-22 00:08:20 -08001139 nr_reclaimed += nr_freed;
Nick Piggina74609f2006-01-06 00:11:20 -08001140 local_irq_disable();
1141 if (current_is_kswapd()) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001142 __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scan);
1143 __count_vm_events(KSWAPD_STEAL, nr_freed);
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001144 } else if (scanning_global_lru(sc))
Christoph Lameterf8891e52006-06-30 01:55:45 -07001145 __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001146
Shantanu Goel918d3f92006-12-29 16:48:59 -08001147 __count_zone_vm_events(PGSTEAL, zone, nr_freed);
Nick Piggina74609f2006-01-06 00:11:20 -08001148
Wu Fengguangfb8d14e2006-03-22 00:08:28 -08001149 if (nr_taken == 0)
1150 goto done;
1151
Nick Piggina74609f2006-01-06 00:11:20 -08001152 spin_lock(&zone->lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 /*
1154 * Put back any unfreeable pages.
1155 */
1156 while (!list_empty(&page_list)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001157 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 page = lru_to_page(&page_list);
Nick Piggin725d7042006-09-25 23:30:55 -07001159 VM_BUG_ON(PageLRU(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 list_del(&page->lru);
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001161 if (unlikely(!page_evictable(page, NULL))) {
1162 spin_unlock_irq(&zone->lru_lock);
1163 putback_lru_page(page);
1164 spin_lock_irq(&zone->lru_lock);
1165 continue;
1166 }
1167 SetPageLRU(page);
1168 lru = page_lru(page);
1169 add_page_to_lru_list(zone, page, lru);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001170 if (PageActive(page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001171 int file = !!page_is_file_cache(page);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001172 reclaim_stat->recent_rotated[file]++;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001173 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 if (!pagevec_add(&pvec, page)) {
1175 spin_unlock_irq(&zone->lru_lock);
1176 __pagevec_release(&pvec);
1177 spin_lock_irq(&zone->lru_lock);
1178 }
1179 }
Andrew Morton69e05942006-03-22 00:08:19 -08001180 } while (nr_scanned < max_scan);
Wu Fengguangfb8d14e2006-03-22 00:08:28 -08001181 spin_unlock(&zone->lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182done:
Wu Fengguangfb8d14e2006-03-22 00:08:28 -08001183 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 pagevec_release(&pvec);
Andrew Morton05ff5132006-03-22 00:08:20 -08001185 return nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001188/*
1189 * We are about to scan this zone at a certain priority level. If that priority
1190 * level is smaller (ie: more urgent) than the previous priority, then note
1191 * that priority level within the zone. This is done so that when the next
1192 * process comes in to scan this zone, it will immediately start out at this
1193 * priority level rather than having to build up its own scanning priority.
1194 * Here, this priority affects only the reclaim-mapped threshold.
1195 */
1196static inline void note_zone_scanning_priority(struct zone *zone, int priority)
1197{
1198 if (priority < zone->prev_priority)
1199 zone->prev_priority = priority;
1200}
1201
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001202/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 * This moves pages from the active list to the inactive list.
1204 *
1205 * We move them the other way if the page is referenced by one or more
1206 * processes, from rmap.
1207 *
1208 * If the pages are mostly unmapped, the processing is fast and it is
1209 * appropriate to hold zone->lru_lock across the whole operation. But if
1210 * the pages are mapped, the processing is slow (page_referenced()) so we
1211 * should drop zone->lru_lock around each page. It's impossible to balance
1212 * this, so instead we remove the pages from the LRU while processing them.
1213 * It is safe to rely on PG_active against the non-LRU pages in here because
1214 * nobody will play with that bit on a non-LRU page.
1215 *
1216 * The downside is that we have to touch page->_count against each page.
1217 * But we had to alter page->flags anyway.
1218 */
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001219
1220
Andrew Morton1742f192006-03-22 00:08:21 -08001221static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001222 struct scan_control *sc, int priority, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223{
Andrew Morton69e05942006-03-22 00:08:19 -08001224 unsigned long pgmoved;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 int pgdeactivate = 0;
Andrew Morton69e05942006-03-22 00:08:19 -08001226 unsigned long pgscanned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 LIST_HEAD(l_hold); /* The pages which were snipped off */
Christoph Lameterb69408e2008-10-18 20:26:14 -07001228 LIST_HEAD(l_inactive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 struct page *page;
1230 struct pagevec pvec;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001231 enum lru_list lru;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001232 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 lru_add_drain();
1235 spin_lock_irq(&zone->lru_lock);
Balbir Singh66e17072008-02-07 00:13:56 -08001236 pgmoved = sc->isolate_pages(nr_pages, &l_hold, &pgscanned, sc->order,
1237 ISOLATE_ACTIVE, zone,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001238 sc->mem_cgroup, 1, file);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001239 /*
1240 * zone->pages_scanned is used for detect zone's oom
1241 * mem_cgroup remembers nr_scan by itself.
1242 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001243 if (scanning_global_lru(sc)) {
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001244 zone->pages_scanned += pgscanned;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001245 }
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001246 reclaim_stat->recent_scanned[!!file] += pgmoved;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001247
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001248 if (file)
1249 __mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved);
1250 else
1251 __mod_zone_page_state(zone, NR_ACTIVE_ANON, -pgmoved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 spin_unlock_irq(&zone->lru_lock);
1253
Rik van Riel556adec2008-10-18 20:26:34 -07001254 pgmoved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 while (!list_empty(&l_hold)) {
1256 cond_resched();
1257 page = lru_to_page(&l_hold);
1258 list_del(&page->lru);
Rik van Riel7e9cd482008-10-18 20:26:35 -07001259
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001260 if (unlikely(!page_evictable(page, NULL))) {
1261 putback_lru_page(page);
1262 continue;
1263 }
1264
Rik van Riel7e9cd482008-10-18 20:26:35 -07001265 /* page_referenced clears PageReferenced */
1266 if (page_mapping_inuse(page) &&
1267 page_referenced(page, 0, sc->mem_cgroup))
1268 pgmoved++;
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 list_add(&page->lru, &l_inactive);
1271 }
1272
Andrew Mortonb5557492009-01-06 14:40:13 -08001273 /*
1274 * Move the pages to the [file or anon] inactive list.
1275 */
1276 pagevec_init(&pvec, 1);
Andrew Mortonb5557492009-01-06 14:40:13 -08001277 lru = LRU_BASE + file * LRU_FILE;
1278
Johannes Weiner2a1dc502008-12-01 03:00:35 +01001279 spin_lock_irq(&zone->lru_lock);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001280 /*
Rik van Riel7e9cd482008-10-18 20:26:35 -07001281 * Count referenced pages from currently used mappings as
1282 * rotated, even though they are moved to the inactive list.
1283 * This helps balance scan pressure between file and anonymous
1284 * pages in get_scan_ratio.
1285 */
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001286 reclaim_stat->recent_rotated[!!file] += pgmoved;
Rik van Riel556adec2008-10-18 20:26:34 -07001287
Daisuke Nishimura1d885522009-03-13 13:52:00 -07001288 pgmoved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 while (!list_empty(&l_inactive)) {
1290 page = lru_to_page(&l_inactive);
1291 prefetchw_prev_lru_page(page, &l_inactive, flags);
Nick Piggin725d7042006-09-25 23:30:55 -07001292 VM_BUG_ON(PageLRU(page));
Nick Piggin8d438f92006-03-22 00:07:59 -08001293 SetPageLRU(page);
Nick Piggin725d7042006-09-25 23:30:55 -07001294 VM_BUG_ON(!PageActive(page));
Nick Piggin4c84cac2006-03-22 00:08:00 -08001295 ClearPageActive(page);
1296
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001297 list_move(&page->lru, &zone->lru[lru].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001298 mem_cgroup_add_lru_list(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 pgmoved++;
1300 if (!pagevec_add(&pvec, page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001301 __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 spin_unlock_irq(&zone->lru_lock);
1303 pgdeactivate += pgmoved;
1304 pgmoved = 0;
1305 if (buffer_heads_over_limit)
1306 pagevec_strip(&pvec);
1307 __pagevec_release(&pvec);
1308 spin_lock_irq(&zone->lru_lock);
1309 }
1310 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001311 __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 pgdeactivate += pgmoved;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001313 __count_zone_vm_events(PGREFILL, zone, pgscanned);
1314 __count_vm_events(PGDEACTIVATE, pgdeactivate);
1315 spin_unlock_irq(&zone->lru_lock);
Johannes Weiner24434622009-03-31 15:23:12 -07001316 if (buffer_heads_over_limit)
1317 pagevec_strip(&pvec);
Nick Piggina74609f2006-01-06 00:11:20 -08001318 pagevec_release(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
1320
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001321static int inactive_anon_is_low_global(struct zone *zone)
KOSAKI Motohirof89eb902009-01-07 18:08:14 -08001322{
1323 unsigned long active, inactive;
1324
1325 active = zone_page_state(zone, NR_ACTIVE_ANON);
1326 inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1327
1328 if (inactive * zone->inactive_ratio < active)
1329 return 1;
1330
1331 return 0;
1332}
1333
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001334/**
1335 * inactive_anon_is_low - check if anonymous pages need to be deactivated
1336 * @zone: zone to check
1337 * @sc: scan control of this context
1338 *
1339 * Returns true if the zone does not have enough inactive anon pages,
1340 * meaning some active anon pages need to be deactivated.
1341 */
1342static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc)
1343{
1344 int low;
1345
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001346 if (scanning_global_lru(sc))
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001347 low = inactive_anon_is_low_global(zone);
1348 else
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001349 low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001350 return low;
1351}
1352
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001353static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
Christoph Lameterb69408e2008-10-18 20:26:14 -07001354 struct zone *zone, struct scan_control *sc, int priority)
1355{
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001356 int file = is_file_lru(lru);
1357
Rik van Riel556adec2008-10-18 20:26:34 -07001358 if (lru == LRU_ACTIVE_FILE) {
1359 shrink_active_list(nr_to_scan, zone, sc, priority, file);
1360 return 0;
1361 }
1362
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001363 if (lru == LRU_ACTIVE_ANON && inactive_anon_is_low(zone, sc)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001364 shrink_active_list(nr_to_scan, zone, sc, priority, file);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001365 return 0;
1366 }
Rik van Riel33c120e2008-10-18 20:26:36 -07001367 return shrink_inactive_list(nr_to_scan, zone, sc, priority, file);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001368}
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370/*
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001371 * Determine how aggressively the anon and file LRU lists should be
1372 * scanned. The relative value of each set of LRU lists is determined
1373 * by looking at the fraction of the pages scanned we did rotate back
1374 * onto the active list instead of evict.
1375 *
1376 * percent[0] specifies how much pressure to put on ram/swap backed
1377 * memory, while percent[1] determines pressure on the file LRUs.
1378 */
1379static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
1380 unsigned long *percent)
1381{
1382 unsigned long anon, file, free;
1383 unsigned long anon_prio, file_prio;
1384 unsigned long ap, fp;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001385 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001386
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001387 /* If we have no swap space, do not bother scanning anon pages. */
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07001388 if (!sc->may_swap || (nr_swap_pages <= 0)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001389 percent[0] = 0;
1390 percent[1] = 100;
1391 return;
1392 }
1393
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -08001394 anon = zone_nr_pages(zone, sc, LRU_ACTIVE_ANON) +
1395 zone_nr_pages(zone, sc, LRU_INACTIVE_ANON);
1396 file = zone_nr_pages(zone, sc, LRU_ACTIVE_FILE) +
1397 zone_nr_pages(zone, sc, LRU_INACTIVE_FILE);
Hugh Dickinsb9627162009-01-06 14:39:41 -08001398
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001399 if (scanning_global_lru(sc)) {
KOSAKI Motohiroeeee9a82009-01-07 18:08:17 -08001400 free = zone_page_state(zone, NR_FREE_PAGES);
1401 /* If we have very few page cache pages,
1402 force-scan anon pages. */
1403 if (unlikely(file + free <= zone->pages_high)) {
1404 percent[0] = 100;
1405 percent[1] = 0;
1406 return;
1407 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001408 }
1409
1410 /*
1411 * OK, so we have swap space and a fair amount of page cache
1412 * pages. We use the recently rotated / recently scanned
1413 * ratios to determine how valuable each cache is.
1414 *
1415 * Because workloads change over time (and to avoid overflow)
1416 * we keep these statistics as a floating average, which ends
1417 * up weighing recent references more than old ones.
1418 *
1419 * anon in [0], file in [1]
1420 */
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001421 if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001422 spin_lock_irq(&zone->lru_lock);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001423 reclaim_stat->recent_scanned[0] /= 2;
1424 reclaim_stat->recent_rotated[0] /= 2;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001425 spin_unlock_irq(&zone->lru_lock);
1426 }
1427
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001428 if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001429 spin_lock_irq(&zone->lru_lock);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001430 reclaim_stat->recent_scanned[1] /= 2;
1431 reclaim_stat->recent_rotated[1] /= 2;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001432 spin_unlock_irq(&zone->lru_lock);
1433 }
1434
1435 /*
1436 * With swappiness at 100, anonymous and file have the same priority.
1437 * This scanning priority is essentially the inverse of IO cost.
1438 */
1439 anon_prio = sc->swappiness;
1440 file_prio = 200 - sc->swappiness;
1441
1442 /*
Rik van Riel00d80892008-11-19 15:36:44 -08001443 * The amount of pressure on anon vs file pages is inversely
1444 * proportional to the fraction of recently scanned pages on
1445 * each list that were recently referenced and in active use.
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001446 */
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001447 ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1);
1448 ap /= reclaim_stat->recent_rotated[0] + 1;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001449
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001450 fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1);
1451 fp /= reclaim_stat->recent_rotated[1] + 1;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001452
1453 /* Normalize to percentages */
1454 percent[0] = 100 * ap / (ap + fp + 1);
1455 percent[1] = 100 - percent[0];
1456}
1457
1458
1459/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
1461 */
Rik van Riela79311c2009-01-06 14:40:01 -08001462static void shrink_zone(int priority, struct zone *zone,
Andrew Morton05ff5132006-03-22 00:08:20 -08001463 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464{
Christoph Lameterb69408e2008-10-18 20:26:14 -07001465 unsigned long nr[NR_LRU_LISTS];
Christoph Lameter86959492006-03-22 00:08:18 -08001466 unsigned long nr_to_scan;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001467 unsigned long percent[2]; /* anon @ 0; file @ 1 */
Christoph Lameterb69408e2008-10-18 20:26:14 -07001468 enum lru_list l;
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001469 unsigned long nr_reclaimed = sc->nr_reclaimed;
1470 unsigned long swap_cluster_max = sc->swap_cluster_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001472 get_scan_ratio(zone, sc, percent);
1473
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001474 for_each_evictable_lru(l) {
KOSAKI Motohiro9439c1c2009-01-07 18:08:21 -08001475 int file = is_file_lru(l);
Andrew Morton8713e012009-04-30 15:08:55 -07001476 unsigned long scan;
Johannes Weinere0f79b82008-10-18 20:26:55 -07001477
KOSAKI Motohirof272b7b2009-03-12 14:31:36 -07001478 scan = zone_nr_pages(zone, sc, l);
KOSAKI Motohiro9439c1c2009-01-07 18:08:21 -08001479 if (priority) {
1480 scan >>= priority;
1481 scan = (scan * percent[file]) / 100;
1482 }
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001483 if (scanning_global_lru(sc)) {
Johannes Weinere0f79b82008-10-18 20:26:55 -07001484 zone->lru[l].nr_scan += scan;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001485 nr[l] = zone->lru[l].nr_scan;
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001486 if (nr[l] >= swap_cluster_max)
Christoph Lameterb69408e2008-10-18 20:26:14 -07001487 zone->lru[l].nr_scan = 0;
1488 else
1489 nr[l] = 0;
KOSAKI Motohiro9439c1c2009-01-07 18:08:21 -08001490 } else
1491 nr[l] = scan;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001492 }
1493
Rik van Riel556adec2008-10-18 20:26:34 -07001494 while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
1495 nr[LRU_INACTIVE_FILE]) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001496 for_each_evictable_lru(l) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07001497 if (nr[l]) {
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001498 nr_to_scan = min(nr[l], swap_cluster_max);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001499 nr[l] -= nr_to_scan;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001501 nr_reclaimed += shrink_list(l, nr_to_scan,
1502 zone, sc, priority);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 }
Rik van Riela79311c2009-01-06 14:40:01 -08001505 /*
1506 * On large memory systems, scan >> priority can become
1507 * really large. This is fine for the starting priority;
1508 * we want to put equal scanning pressure on each zone.
1509 * However, if the VM has a harder time of freeing pages,
1510 * with multiple processes reclaiming pages, the total
1511 * freeing target can get unreasonably large.
1512 */
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001513 if (nr_reclaimed > swap_cluster_max &&
Rik van Riela79311c2009-01-06 14:40:01 -08001514 priority < DEF_PRIORITY && !current_is_kswapd())
1515 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 }
1517
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001518 sc->nr_reclaimed = nr_reclaimed;
1519
Rik van Riel556adec2008-10-18 20:26:34 -07001520 /*
1521 * Even if we did not try to evict anon pages at all, we want to
1522 * rebalance the anon lru active/inactive ratio.
1523 */
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001524 if (inactive_anon_is_low(zone, sc))
Rik van Riel556adec2008-10-18 20:26:34 -07001525 shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1526
Andrew Morton232ea4d2007-02-28 20:13:21 -08001527 throttle_vm_writeout(sc->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528}
1529
1530/*
1531 * This is the direct reclaim path, for page-allocating processes. We only
1532 * try to reclaim pages from zones which will satisfy the caller's allocation
1533 * request.
1534 *
1535 * We reclaim from a zone even if that zone is over pages_high. Because:
1536 * a) The caller may be trying to free *extra* pages to satisfy a higher-order
1537 * allocation or
1538 * b) The zones may be over pages_high but they must go *over* pages_high to
1539 * satisfy the `incremental min' zone defense algorithm.
1540 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 * If a zone is deemed to be full of pinned pages then just give it a light
1542 * scan then give up on it.
1543 */
Rik van Riela79311c2009-01-06 14:40:01 -08001544static void shrink_zones(int priority, struct zonelist *zonelist,
Andrew Morton05ff5132006-03-22 00:08:20 -08001545 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
Mel Gorman54a6eb52008-04-28 02:12:16 -07001547 enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
Mel Gormandd1a2392008-04-28 02:12:17 -07001548 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001549 struct zone *zone;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001550
Nick Piggin408d8542006-09-25 23:31:27 -07001551 sc->all_unreclaimable = 1;
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -07001552 for_each_zone_zonelist_nodemask(zone, z, zonelist, high_zoneidx,
1553 sc->nodemask) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08001554 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 continue;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001556 /*
1557 * Take care memory controller reclaiming has small influence
1558 * to global LRU.
1559 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001560 if (scanning_global_lru(sc)) {
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001561 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1562 continue;
1563 note_zone_scanning_priority(zone, priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001565 if (zone_is_all_unreclaimable(zone) &&
1566 priority != DEF_PRIORITY)
1567 continue; /* Let kswapd poll it */
1568 sc->all_unreclaimable = 0;
1569 } else {
1570 /*
1571 * Ignore cpuset limitation here. We just want to reduce
1572 * # of used pages by us regardless of memory shortage.
1573 */
1574 sc->all_unreclaimable = 0;
1575 mem_cgroup_note_reclaim_priority(sc->mem_cgroup,
1576 priority);
1577 }
Nick Piggin408d8542006-09-25 23:31:27 -07001578
Rik van Riela79311c2009-01-06 14:40:01 -08001579 shrink_zone(priority, zone, sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581}
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583/*
1584 * This is the main entry point to direct page reclaim.
1585 *
1586 * If a full scan of the inactive list fails to free enough memory then we
1587 * are "out of memory" and something needs to be killed.
1588 *
1589 * If the caller is !__GFP_FS then the probability of a failure is reasonably
1590 * high - the zone may be full of dirty or under-writeback pages, which this
1591 * caller can't do much about. We kick pdflush and take explicit naps in the
1592 * hope that some of these pages can be written. But if the allocating task
1593 * holds filesystem locks which prevent writeout this might not work, and the
1594 * allocation attempt will fail.
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001595 *
1596 * returns: 0, if no pages reclaimed
1597 * else, the number of pages reclaimed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 */
Mel Gormandac1d272008-04-28 02:12:12 -07001599static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
Mel Gormandd1a2392008-04-28 02:12:17 -07001600 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
1602 int priority;
kosaki.motohiro@jp.fujitsu.comc700be32008-06-12 15:21:27 -07001603 unsigned long ret = 0;
Andrew Morton69e05942006-03-22 00:08:19 -08001604 unsigned long total_scanned = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 struct reclaim_state *reclaim_state = current->reclaim_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 unsigned long lru_pages = 0;
Mel Gormandd1a2392008-04-28 02:12:17 -07001607 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001608 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -07001609 enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Keika Kobayashi873b4772008-07-25 01:48:52 -07001611 delayacct_freepages_start();
1612
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001613 if (scanning_global_lru(sc))
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001614 count_vm_event(ALLOCSTALL);
1615 /*
1616 * mem_cgroup will not do shrink_slab.
1617 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001618 if (scanning_global_lru(sc)) {
Mel Gorman54a6eb52008-04-28 02:12:16 -07001619 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001621 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1622 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001624 lru_pages += zone_lru_pages(zone);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 }
1627
1628 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
Balbir Singh66e17072008-02-07 00:13:56 -08001629 sc->nr_scanned = 0;
Rik van Rielf7b7fd82005-11-28 13:44:07 -08001630 if (!priority)
1631 disable_swap_token();
Rik van Riela79311c2009-01-06 14:40:01 -08001632 shrink_zones(priority, zonelist, sc);
Balbir Singh66e17072008-02-07 00:13:56 -08001633 /*
1634 * Don't shrink slabs when reclaiming memory from
1635 * over limit cgroups
1636 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001637 if (scanning_global_lru(sc)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001638 shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages);
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -08001639 if (reclaim_state) {
Rik van Riela79311c2009-01-06 14:40:01 -08001640 sc->nr_reclaimed += reclaim_state->reclaimed_slab;
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -08001641 reclaim_state->reclaimed_slab = 0;
1642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 }
Balbir Singh66e17072008-02-07 00:13:56 -08001644 total_scanned += sc->nr_scanned;
Rik van Riela79311c2009-01-06 14:40:01 -08001645 if (sc->nr_reclaimed >= sc->swap_cluster_max) {
1646 ret = sc->nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 goto out;
1648 }
1649
1650 /*
1651 * Try to write back as many pages as we just scanned. This
1652 * tends to cause slow streaming writers to write data to the
1653 * disk smoothly, at the dirtying rate, which is nice. But
1654 * that's undesirable in laptop mode, where we *want* lumpy
1655 * writeout. So in laptop mode, write out the whole world.
1656 */
Balbir Singh66e17072008-02-07 00:13:56 -08001657 if (total_scanned > sc->swap_cluster_max +
1658 sc->swap_cluster_max / 2) {
Pekka J Enberg687a21c2005-06-28 20:44:55 -07001659 wakeup_pdflush(laptop_mode ? 0 : total_scanned);
Balbir Singh66e17072008-02-07 00:13:56 -08001660 sc->may_writepage = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 }
1662
1663 /* Take a nap, wait for some writeback to complete */
Andrew Morton4dd4b922008-03-24 12:29:52 -07001664 if (sc->nr_scanned && priority < DEF_PRIORITY - 2)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001665 congestion_wait(WRITE, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 }
Fernando Luis Vazquez Cao87547ee2008-07-29 22:33:42 -07001667 /* top priority shrink_zones still had more to do? don't OOM, then */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001668 if (!sc->all_unreclaimable && scanning_global_lru(sc))
Rik van Riela79311c2009-01-06 14:40:01 -08001669 ret = sc->nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670out:
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001671 /*
1672 * Now that we've scanned all the zones at this priority level, note
1673 * that level within the zone so that the next thread which performs
1674 * scanning of this zone will immediately start out at this priority
1675 * level. This affects only the decision whether or not to bring
1676 * mapped pages onto the inactive list.
1677 */
1678 if (priority < 0)
1679 priority = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001681 if (scanning_global_lru(sc)) {
Mel Gorman54a6eb52008-04-28 02:12:16 -07001682 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001684 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1685 continue;
1686
1687 zone->prev_priority = priority;
1688 }
1689 } else
1690 mem_cgroup_record_reclaim_priority(sc->mem_cgroup, priority);
1691
Keika Kobayashi873b4772008-07-25 01:48:52 -07001692 delayacct_freepages_end();
1693
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 return ret;
1695}
1696
Mel Gormandac1d272008-04-28 02:12:12 -07001697unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -07001698 gfp_t gfp_mask, nodemask_t *nodemask)
Balbir Singh66e17072008-02-07 00:13:56 -08001699{
1700 struct scan_control sc = {
1701 .gfp_mask = gfp_mask,
1702 .may_writepage = !laptop_mode,
1703 .swap_cluster_max = SWAP_CLUSTER_MAX,
Johannes Weinera6dc60f82009-03-31 15:19:30 -07001704 .may_unmap = 1,
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07001705 .may_swap = 1,
Balbir Singh66e17072008-02-07 00:13:56 -08001706 .swappiness = vm_swappiness,
1707 .order = order,
1708 .mem_cgroup = NULL,
1709 .isolate_pages = isolate_pages_global,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -07001710 .nodemask = nodemask,
Balbir Singh66e17072008-02-07 00:13:56 -08001711 };
1712
Mel Gormandd1a2392008-04-28 02:12:17 -07001713 return do_try_to_free_pages(zonelist, &sc);
Balbir Singh66e17072008-02-07 00:13:56 -08001714}
1715
Balbir Singh00f0b822008-03-04 14:28:39 -08001716#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Balbir Singh66e17072008-02-07 00:13:56 -08001717
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001718unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001719 gfp_t gfp_mask,
1720 bool noswap,
1721 unsigned int swappiness)
Balbir Singh66e17072008-02-07 00:13:56 -08001722{
1723 struct scan_control sc = {
Balbir Singh66e17072008-02-07 00:13:56 -08001724 .may_writepage = !laptop_mode,
Johannes Weinera6dc60f82009-03-31 15:19:30 -07001725 .may_unmap = 1,
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07001726 .may_swap = !noswap,
Balbir Singh66e17072008-02-07 00:13:56 -08001727 .swap_cluster_max = SWAP_CLUSTER_MAX,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001728 .swappiness = swappiness,
Balbir Singh66e17072008-02-07 00:13:56 -08001729 .order = 0,
1730 .mem_cgroup = mem_cont,
1731 .isolate_pages = mem_cgroup_isolate_pages,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -07001732 .nodemask = NULL, /* we don't care the placement */
Balbir Singh66e17072008-02-07 00:13:56 -08001733 };
Mel Gormandac1d272008-04-28 02:12:12 -07001734 struct zonelist *zonelist;
Balbir Singh66e17072008-02-07 00:13:56 -08001735
Mel Gormandd1a2392008-04-28 02:12:17 -07001736 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
1737 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
1738 zonelist = NODE_DATA(numa_node_id())->node_zonelists;
1739 return do_try_to_free_pages(zonelist, &sc);
Balbir Singh66e17072008-02-07 00:13:56 -08001740}
1741#endif
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743/*
1744 * For kswapd, balance_pgdat() will work across all this node's zones until
1745 * they are all at pages_high.
1746 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 * Returns the number of pages which were actually freed.
1748 *
1749 * There is special handling here for zones which are full of pinned pages.
1750 * This can happen if the pages are all mlocked, or if they are all used by
1751 * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
1752 * What we do is to detect the case where all pages in the zone have been
1753 * scanned twice and there has been zero successful reclaim. Mark the zone as
1754 * dead and from now on, only perform a short scan. Basically we're polling
1755 * the zone for when the problem goes away.
1756 *
1757 * kswapd scans the zones in the highmem->normal->dma direction. It skips
1758 * zones which have free_pages > pages_high, but once a zone is found to have
1759 * free_pages <= pages_high, we scan that zone and the lower zones regardless
1760 * of the number of free pages in the lower zones. This interoperates with
1761 * the page allocator fallback scheme to ensure that aging of pages is balanced
1762 * across the zones.
1763 */
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001764static unsigned long balance_pgdat(pg_data_t *pgdat, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 int all_zones_ok;
1767 int priority;
1768 int i;
Andrew Morton69e05942006-03-22 00:08:19 -08001769 unsigned long total_scanned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 struct reclaim_state *reclaim_state = current->reclaim_state;
Andrew Morton179e9632006-03-22 00:08:18 -08001771 struct scan_control sc = {
1772 .gfp_mask = GFP_KERNEL,
Johannes Weinera6dc60f82009-03-31 15:19:30 -07001773 .may_unmap = 1,
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07001774 .may_swap = 1,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001775 .swap_cluster_max = SWAP_CLUSTER_MAX,
1776 .swappiness = vm_swappiness,
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001777 .order = order,
Balbir Singh66e17072008-02-07 00:13:56 -08001778 .mem_cgroup = NULL,
1779 .isolate_pages = isolate_pages_global,
Andrew Morton179e9632006-03-22 00:08:18 -08001780 };
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001781 /*
1782 * temp_priority is used to remember the scanning priority at which
1783 * this zone was successfully refilled to free_pages == pages_high.
1784 */
1785 int temp_priority[MAX_NR_ZONES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787loop_again:
1788 total_scanned = 0;
Rik van Riela79311c2009-01-06 14:40:01 -08001789 sc.nr_reclaimed = 0;
Christoph Lameterc0bbbc72006-06-11 15:22:26 -07001790 sc.may_writepage = !laptop_mode;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001791 count_vm_event(PAGEOUTRUN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001793 for (i = 0; i < pgdat->nr_zones; i++)
1794 temp_priority[i] = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
1796 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
1797 int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */
1798 unsigned long lru_pages = 0;
1799
Rik van Rielf7b7fd82005-11-28 13:44:07 -08001800 /* The swap token gets in the way of swapout... */
1801 if (!priority)
1802 disable_swap_token();
1803
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 all_zones_ok = 1;
1805
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001806 /*
1807 * Scan in the highmem->dma direction for the highest
1808 * zone which needs scanning
1809 */
1810 for (i = pgdat->nr_zones - 1; i >= 0; i--) {
1811 struct zone *zone = pgdat->node_zones + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001813 if (!populated_zone(zone))
1814 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
David Rientjese815af92007-10-16 23:25:54 -07001816 if (zone_is_all_unreclaimable(zone) &&
1817 priority != DEF_PRIORITY)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001818 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819
Rik van Riel556adec2008-10-18 20:26:34 -07001820 /*
1821 * Do some background aging of the anon list, to give
1822 * pages a chance to be referenced before reclaiming.
1823 */
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001824 if (inactive_anon_is_low(zone, &sc))
Rik van Riel556adec2008-10-18 20:26:34 -07001825 shrink_active_list(SWAP_CLUSTER_MAX, zone,
1826 &sc, priority, 0);
1827
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001828 if (!zone_watermark_ok(zone, order, zone->pages_high,
1829 0, 0)) {
1830 end_zone = i;
Andrew Mortone1dbeda2006-12-06 20:32:01 -08001831 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 }
Andrew Mortone1dbeda2006-12-06 20:32:01 -08001834 if (i < 0)
1835 goto out;
1836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 for (i = 0; i <= end_zone; i++) {
1838 struct zone *zone = pgdat->node_zones + i;
1839
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001840 lru_pages += zone_lru_pages(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 }
1842
1843 /*
1844 * Now scan the zone in the dma->highmem direction, stopping
1845 * at the last zone which needs scanning.
1846 *
1847 * We do this because the page allocator works in the opposite
1848 * direction. This prevents the page allocator from allocating
1849 * pages behind kswapd's direction of progress, which would
1850 * cause too much scanning of the lower zones.
1851 */
1852 for (i = 0; i <= end_zone; i++) {
1853 struct zone *zone = pgdat->node_zones + i;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07001854 int nr_slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Con Kolivasf3fe6512006-01-06 00:11:15 -08001856 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 continue;
1858
David Rientjese815af92007-10-16 23:25:54 -07001859 if (zone_is_all_unreclaimable(zone) &&
1860 priority != DEF_PRIORITY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 continue;
1862
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001863 if (!zone_watermark_ok(zone, order, zone->pages_high,
1864 end_zone, 0))
1865 all_zones_ok = 0;
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001866 temp_priority[i] = priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 sc.nr_scanned = 0;
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001868 note_zone_scanning_priority(zone, priority);
Rik van Riel32a43302007-10-16 01:24:50 -07001869 /*
1870 * We put equal pressure on every zone, unless one
1871 * zone has way too many pages free already.
1872 */
1873 if (!zone_watermark_ok(zone, order, 8*zone->pages_high,
1874 end_zone, 0))
Rik van Riela79311c2009-01-06 14:40:01 -08001875 shrink_zone(priority, zone, &sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 reclaim_state->reclaimed_slab = 0;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07001877 nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL,
1878 lru_pages);
Rik van Riela79311c2009-01-06 14:40:01 -08001879 sc.nr_reclaimed += reclaim_state->reclaimed_slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 total_scanned += sc.nr_scanned;
David Rientjese815af92007-10-16 23:25:54 -07001881 if (zone_is_all_unreclaimable(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 continue;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07001883 if (nr_slab == 0 && zone->pages_scanned >=
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001884 (zone_lru_pages(zone) * 6))
David Rientjese815af92007-10-16 23:25:54 -07001885 zone_set_flag(zone,
1886 ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 /*
1888 * If we've done a decent amount of scanning and
1889 * the reclaim ratio is low, start doing writepage
1890 * even in laptop mode
1891 */
1892 if (total_scanned > SWAP_CLUSTER_MAX * 2 &&
Rik van Riela79311c2009-01-06 14:40:01 -08001893 total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 sc.may_writepage = 1;
1895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 if (all_zones_ok)
1897 break; /* kswapd: all done */
1898 /*
1899 * OK, kswapd is getting into trouble. Take a nap, then take
1900 * another pass across the zones.
1901 */
Andrew Morton4dd4b922008-03-24 12:29:52 -07001902 if (total_scanned && priority < DEF_PRIORITY - 2)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001903 congestion_wait(WRITE, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
1905 /*
1906 * We do this so kswapd doesn't build up large priorities for
1907 * example when it is freeing in parallel with allocators. It
1908 * matches the direct reclaim path behaviour in terms of impact
1909 * on zone->*_priority.
1910 */
Rik van Riela79311c2009-01-06 14:40:01 -08001911 if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 break;
1913 }
1914out:
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001915 /*
1916 * Note within each zone the priority level at which this zone was
1917 * brought into a happy state. So that the next thread which scans this
1918 * zone will start out at that priority level.
1919 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 for (i = 0; i < pgdat->nr_zones; i++) {
1921 struct zone *zone = pgdat->node_zones + i;
1922
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001923 zone->prev_priority = temp_priority[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 }
1925 if (!all_zones_ok) {
1926 cond_resched();
Rafael J. Wysocki83573762006-12-06 20:34:18 -08001927
1928 try_to_freeze();
1929
KOSAKI Motohiro73ce02e2009-01-06 14:40:33 -08001930 /*
1931 * Fragmentation may mean that the system cannot be
1932 * rebalanced for high-order allocations in all zones.
1933 * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX,
1934 * it means the zones have been fully scanned and are still
1935 * not balanced. For high-order allocations, there is
1936 * little point trying all over again as kswapd may
1937 * infinite loop.
1938 *
1939 * Instead, recheck all watermarks at order-0 as they
1940 * are the most important. If watermarks are ok, kswapd will go
1941 * back to sleep. High-order users can still perform direct
1942 * reclaim if they wish.
1943 */
1944 if (sc.nr_reclaimed < SWAP_CLUSTER_MAX)
1945 order = sc.order = 0;
1946
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 goto loop_again;
1948 }
1949
Rik van Riela79311c2009-01-06 14:40:01 -08001950 return sc.nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951}
1952
1953/*
1954 * The background pageout daemon, started as a kernel thread
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001955 * from the init process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 *
1957 * This basically trickles out pages so that we have _some_
1958 * free memory available even if there is no other activity
1959 * that frees anything up. This is needed for things like routing
1960 * etc, where we otherwise might have all activity going on in
1961 * asynchronous contexts that cannot page things out.
1962 *
1963 * If there are applications that are active memory-allocators
1964 * (most normal use), this basically shouldn't matter.
1965 */
1966static int kswapd(void *p)
1967{
1968 unsigned long order;
1969 pg_data_t *pgdat = (pg_data_t*)p;
1970 struct task_struct *tsk = current;
1971 DEFINE_WAIT(wait);
1972 struct reclaim_state reclaim_state = {
1973 .reclaimed_slab = 0,
1974 };
Rusty Russella70f7302009-03-13 14:49:46 +10301975 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
Nick Piggincf40bd12009-01-21 08:12:39 +01001977 lockdep_set_current_reclaim_state(GFP_KERNEL);
1978
Rusty Russell174596a2009-01-01 10:12:29 +10301979 if (!cpumask_empty(cpumask))
Mike Travisc5f59f02008-04-04 18:11:10 -07001980 set_cpus_allowed_ptr(tsk, cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 current->reclaim_state = &reclaim_state;
1982
1983 /*
1984 * Tell the memory management that we're a "memory allocator",
1985 * and that if we need more memory we should get access to it
1986 * regardless (see "__alloc_pages()"). "kswapd" should
1987 * never get caught in the normal page freeing logic.
1988 *
1989 * (Kswapd normally doesn't need memory anyway, but sometimes
1990 * you need a small amount of memory in order to be able to
1991 * page out something else, and this flag essentially protects
1992 * us from recursively trying to free more memory as we're
1993 * trying to free the first piece of memory in the first place).
1994 */
Christoph Lameter930d9152006-01-08 01:00:47 -08001995 tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
Rafael J. Wysocki83144182007-07-17 04:03:35 -07001996 set_freezable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
1998 order = 0;
1999 for ( ; ; ) {
2000 unsigned long new_order;
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07002001
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
2003 new_order = pgdat->kswapd_max_order;
2004 pgdat->kswapd_max_order = 0;
2005 if (order < new_order) {
2006 /*
2007 * Don't sleep if someone wants a larger 'order'
2008 * allocation
2009 */
2010 order = new_order;
2011 } else {
Rafael J. Wysockib1296cc2007-05-06 14:50:48 -07002012 if (!freezing(current))
2013 schedule();
2014
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 order = pgdat->kswapd_max_order;
2016 }
2017 finish_wait(&pgdat->kswapd_wait, &wait);
2018
Rafael J. Wysockib1296cc2007-05-06 14:50:48 -07002019 if (!try_to_freeze()) {
2020 /* We can speed up thawing tasks if we don't call
2021 * balance_pgdat after returning from the refrigerator
2022 */
2023 balance_pgdat(pgdat, order);
2024 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 }
2026 return 0;
2027}
2028
2029/*
2030 * A zone is low on free memory, so wake its kswapd task to service it.
2031 */
2032void wakeup_kswapd(struct zone *zone, int order)
2033{
2034 pg_data_t *pgdat;
2035
Con Kolivasf3fe6512006-01-06 00:11:15 -08002036 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 return;
2038
2039 pgdat = zone->zone_pgdat;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002040 if (zone_watermark_ok(zone, order, zone->pages_low, 0, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 return;
2042 if (pgdat->kswapd_max_order < order)
2043 pgdat->kswapd_max_order = order;
Paul Jackson02a0e532006-12-13 00:34:25 -08002044 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 return;
Con Kolivas8d0986e2005-09-13 01:25:07 -07002046 if (!waitqueue_active(&pgdat->kswapd_wait))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return;
Con Kolivas8d0986e2005-09-13 01:25:07 -07002048 wake_up_interruptible(&pgdat->kswapd_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049}
2050
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002051unsigned long global_lru_pages(void)
2052{
2053 return global_page_state(NR_ACTIVE_ANON)
2054 + global_page_state(NR_ACTIVE_FILE)
2055 + global_page_state(NR_INACTIVE_ANON)
2056 + global_page_state(NR_INACTIVE_FILE);
2057}
2058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059#ifdef CONFIG_PM
2060/*
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002061 * Helper function for shrink_all_memory(). Tries to reclaim 'nr_pages' pages
MinChan Kimd9796772009-03-31 15:19:34 -07002062 * from LRU lists system-wide, for given pass and priority.
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002063 *
2064 * For pass > 3 we also try to shrink the LRU lists that contain a few pages
2065 */
MinChan Kimd9796772009-03-31 15:19:34 -07002066static void shrink_all_zones(unsigned long nr_pages, int prio,
Nigel Cunninghame07aa052006-12-22 01:07:21 -08002067 int pass, struct scan_control *sc)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002068{
2069 struct zone *zone;
MinChan Kimd9796772009-03-31 15:19:34 -07002070 unsigned long nr_reclaimed = 0;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002071
KOSAKI Motohiroee99c712009-03-31 15:19:31 -07002072 for_each_populated_zone(zone) {
Johannes Weiner0cb57252009-02-14 02:04:10 +01002073 enum lru_list l;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002074
David Rientjese815af92007-10-16 23:25:54 -07002075 if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002076 continue;
2077
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002078 for_each_evictable_lru(l) {
Johannes Weiner0cb57252009-02-14 02:04:10 +01002079 enum zone_stat_item ls = NR_LRU_BASE + l;
2080 unsigned long lru_pages = zone_page_state(zone, ls);
2081
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002082 /* For pass = 0, we don't shrink the active list */
Johannes Weiner0cb57252009-02-14 02:04:10 +01002083 if (pass == 0 && (l == LRU_ACTIVE_ANON ||
2084 l == LRU_ACTIVE_FILE))
Christoph Lameterb69408e2008-10-18 20:26:14 -07002085 continue;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002086
Johannes Weiner0cb57252009-02-14 02:04:10 +01002087 zone->lru[l].nr_scan += (lru_pages >> prio) + 1;
Christoph Lameterb69408e2008-10-18 20:26:14 -07002088 if (zone->lru[l].nr_scan >= nr_pages || pass > 3) {
Johannes Weiner0cb57252009-02-14 02:04:10 +01002089 unsigned long nr_to_scan;
2090
Christoph Lameterb69408e2008-10-18 20:26:14 -07002091 zone->lru[l].nr_scan = 0;
Johannes Weiner0cb57252009-02-14 02:04:10 +01002092 nr_to_scan = min(nr_pages, lru_pages);
MinChan Kimd9796772009-03-31 15:19:34 -07002093 nr_reclaimed += shrink_list(l, nr_to_scan, zone,
Christoph Lameterb69408e2008-10-18 20:26:14 -07002094 sc, prio);
MinChan Kimd9796772009-03-31 15:19:34 -07002095 if (nr_reclaimed >= nr_pages) {
Rafael J. Wysockia21e2552009-04-18 17:23:41 +02002096 sc->nr_reclaimed += nr_reclaimed;
MinChan Kimd9796772009-03-31 15:19:34 -07002097 return;
2098 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07002099 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002100 }
2101 }
Rafael J. Wysockia21e2552009-04-18 17:23:41 +02002102 sc->nr_reclaimed += nr_reclaimed;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002103}
2104
2105/*
2106 * Try to free `nr_pages' of memory, system-wide, and return the number of
2107 * freed pages.
2108 *
2109 * Rather than trying to age LRUs the aim is to preserve the overall
2110 * LRU order by reclaiming preferentially
2111 * inactive > active > active referenced > active mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 */
Andrew Morton69e05942006-03-22 00:08:19 -08002113unsigned long shrink_all_memory(unsigned long nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114{
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002115 unsigned long lru_pages, nr_slab;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002116 int pass;
2117 struct reclaim_state reclaim_state;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002118 struct scan_control sc = {
2119 .gfp_mask = GFP_KERNEL,
Johannes Weinera6dc60f82009-03-31 15:19:30 -07002120 .may_unmap = 0,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002121 .may_writepage = 1,
Balbir Singh66e17072008-02-07 00:13:56 -08002122 .isolate_pages = isolate_pages_global,
Rafael J. Wysockia21e2552009-04-18 17:23:41 +02002123 .nr_reclaimed = 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 };
2125
2126 current->reclaim_state = &reclaim_state;
Andrew Morton69e05942006-03-22 00:08:19 -08002127
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002128 lru_pages = global_lru_pages();
Christoph Lameter972d1a72006-09-25 23:31:51 -07002129 nr_slab = global_page_state(NR_SLAB_RECLAIMABLE);
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002130 /* If slab caches are huge, it's better to hit them first */
2131 while (nr_slab >= lru_pages) {
2132 reclaim_state.reclaimed_slab = 0;
2133 shrink_slab(nr_pages, sc.gfp_mask, lru_pages);
2134 if (!reclaim_state.reclaimed_slab)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 break;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002136
MinChan Kimd9796772009-03-31 15:19:34 -07002137 sc.nr_reclaimed += reclaim_state.reclaimed_slab;
2138 if (sc.nr_reclaimed >= nr_pages)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002139 goto out;
2140
2141 nr_slab -= reclaim_state.reclaimed_slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002143
2144 /*
2145 * We try to shrink LRUs in 5 passes:
2146 * 0 = Reclaim from inactive_list only
2147 * 1 = Reclaim from active list but don't reclaim mapped
2148 * 2 = 2nd pass of type 1
2149 * 3 = Reclaim mapped (normal reclaim)
2150 * 4 = 2nd pass of type 3
2151 */
2152 for (pass = 0; pass < 5; pass++) {
2153 int prio;
2154
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002155 /* Force reclaiming mapped pages in the passes #3 and #4 */
Johannes Weiner30491032009-02-14 02:03:08 +01002156 if (pass > 2)
Johannes Weinera6dc60f82009-03-31 15:19:30 -07002157 sc.may_unmap = 1;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002158
2159 for (prio = DEF_PRIORITY; prio >= 0; prio--) {
MinChan Kimd9796772009-03-31 15:19:34 -07002160 unsigned long nr_to_scan = nr_pages - sc.nr_reclaimed;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002161
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002162 sc.nr_scanned = 0;
Johannes Weiner9786bf82009-03-31 15:19:35 -07002163 sc.swap_cluster_max = nr_to_scan;
MinChan Kimd9796772009-03-31 15:19:34 -07002164 shrink_all_zones(nr_to_scan, prio, pass, &sc);
2165 if (sc.nr_reclaimed >= nr_pages)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002166 goto out;
2167
2168 reclaim_state.reclaimed_slab = 0;
Andrew Morton76395d32007-01-05 16:37:05 -08002169 shrink_slab(sc.nr_scanned, sc.gfp_mask,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002170 global_lru_pages());
MinChan Kimd9796772009-03-31 15:19:34 -07002171 sc.nr_reclaimed += reclaim_state.reclaimed_slab;
2172 if (sc.nr_reclaimed >= nr_pages)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002173 goto out;
2174
2175 if (sc.nr_scanned && prio < DEF_PRIORITY - 2)
Andrew Morton3fcfab12006-10-19 23:28:16 -07002176 congestion_wait(WRITE, HZ / 10);
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002177 }
Rafael J. Wysocki248a0302006-03-22 00:09:04 -08002178 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002179
2180 /*
MinChan Kimd9796772009-03-31 15:19:34 -07002181 * If sc.nr_reclaimed = 0, we could not shrink LRUs, but there may be
2182 * something in slab caches
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002183 */
MinChan Kimd9796772009-03-31 15:19:34 -07002184 if (!sc.nr_reclaimed) {
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002185 do {
2186 reclaim_state.reclaimed_slab = 0;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002187 shrink_slab(nr_pages, sc.gfp_mask, global_lru_pages());
MinChan Kimd9796772009-03-31 15:19:34 -07002188 sc.nr_reclaimed += reclaim_state.reclaimed_slab;
2189 } while (sc.nr_reclaimed < nr_pages &&
2190 reclaim_state.reclaimed_slab > 0);
Andrew Morton76395d32007-01-05 16:37:05 -08002191 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002192
MinChan Kimd9796772009-03-31 15:19:34 -07002193
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002194out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 current->reclaim_state = NULL;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002196
MinChan Kimd9796772009-03-31 15:19:34 -07002197 return sc.nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198}
2199#endif
2200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201/* It's optimal to keep kswapds on the same CPUs as their memory, but
2202 not required for correctness. So if the last cpu in a node goes
2203 away, we get changed to run anywhere: as the first one comes back,
2204 restore their cpu bindings. */
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002205static int __devinit cpu_callback(struct notifier_block *nfb,
Andrew Morton69e05942006-03-22 00:08:19 -08002206 unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
Yasunori Goto58c0a4a2007-10-16 01:25:40 -07002208 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002210 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
Yasunori Goto58c0a4a2007-10-16 01:25:40 -07002211 for_each_node_state(nid, N_HIGH_MEMORY) {
Mike Travisc5f59f02008-04-04 18:11:10 -07002212 pg_data_t *pgdat = NODE_DATA(nid);
Rusty Russella70f7302009-03-13 14:49:46 +10302213 const struct cpumask *mask;
2214
2215 mask = cpumask_of_node(pgdat->node_id);
Mike Travisc5f59f02008-04-04 18:11:10 -07002216
Rusty Russell3e597942009-01-01 10:12:24 +10302217 if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 /* One of our CPUs online: restore mask */
Mike Travisc5f59f02008-04-04 18:11:10 -07002219 set_cpus_allowed_ptr(pgdat->kswapd, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 }
2221 }
2222 return NOTIFY_OK;
2223}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
Yasunori Goto3218ae12006-06-27 02:53:33 -07002225/*
2226 * This kswapd start function will be called by init and node-hot-add.
2227 * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
2228 */
2229int kswapd_run(int nid)
2230{
2231 pg_data_t *pgdat = NODE_DATA(nid);
2232 int ret = 0;
2233
2234 if (pgdat->kswapd)
2235 return 0;
2236
2237 pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
2238 if (IS_ERR(pgdat->kswapd)) {
2239 /* failure at boot is fatal */
2240 BUG_ON(system_state == SYSTEM_BOOTING);
2241 printk("Failed to start kswapd on node %d\n",nid);
2242 ret = -1;
2243 }
2244 return ret;
2245}
2246
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247static int __init kswapd_init(void)
2248{
Yasunori Goto3218ae12006-06-27 02:53:33 -07002249 int nid;
Andrew Morton69e05942006-03-22 00:08:19 -08002250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 swap_setup();
Christoph Lameter9422ffb2007-10-16 01:25:31 -07002252 for_each_node_state(nid, N_HIGH_MEMORY)
Yasunori Goto3218ae12006-06-27 02:53:33 -07002253 kswapd_run(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 hotcpu_notifier(cpu_callback, 0);
2255 return 0;
2256}
2257
2258module_init(kswapd_init)
Christoph Lameter9eeff232006-01-18 17:42:31 -08002259
2260#ifdef CONFIG_NUMA
2261/*
2262 * Zone reclaim mode
2263 *
2264 * If non-zero call zone_reclaim when the number of free pages falls below
2265 * the watermarks.
Christoph Lameter9eeff232006-01-18 17:42:31 -08002266 */
2267int zone_reclaim_mode __read_mostly;
2268
Christoph Lameter1b2ffb72006-02-01 03:05:34 -08002269#define RECLAIM_OFF 0
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07002270#define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
Christoph Lameter1b2ffb72006-02-01 03:05:34 -08002271#define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
2272#define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
2273
Christoph Lameter9eeff232006-01-18 17:42:31 -08002274/*
Christoph Lametera92f7122006-02-01 03:05:32 -08002275 * Priority for ZONE_RECLAIM. This determines the fraction of pages
2276 * of a node considered for each zone_reclaim. 4 scans 1/16th of
2277 * a zone.
2278 */
2279#define ZONE_RECLAIM_PRIORITY 4
2280
Christoph Lameter9eeff232006-01-18 17:42:31 -08002281/*
Christoph Lameter96146342006-07-03 00:24:13 -07002282 * Percentage of pages in a zone that must be unmapped for zone_reclaim to
2283 * occur.
2284 */
2285int sysctl_min_unmapped_ratio = 1;
2286
2287/*
Christoph Lameter0ff38492006-09-25 23:31:52 -07002288 * If the number of slab pages in a zone grows beyond this percentage then
2289 * slab reclaim needs to occur.
2290 */
2291int sysctl_min_slab_ratio = 5;
2292
2293/*
Christoph Lameter9eeff232006-01-18 17:42:31 -08002294 * Try to free up some pages from this zone through reclaim.
2295 */
Andrew Morton179e9632006-03-22 00:08:18 -08002296static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
Christoph Lameter9eeff232006-01-18 17:42:31 -08002297{
Christoph Lameter7fb2d462006-03-22 00:08:22 -08002298 /* Minimum pages needed in order to stay on node */
Andrew Morton69e05942006-03-22 00:08:19 -08002299 const unsigned long nr_pages = 1 << order;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002300 struct task_struct *p = current;
2301 struct reclaim_state reclaim_state;
Christoph Lameter86959492006-03-22 00:08:18 -08002302 int priority;
Andrew Morton179e9632006-03-22 00:08:18 -08002303 struct scan_control sc = {
2304 .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
Johannes Weinera6dc60f82009-03-31 15:19:30 -07002305 .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP),
KOSAKI Motohiro2e2e4252009-04-21 12:24:57 -07002306 .may_swap = 1,
Andrew Morton69e05942006-03-22 00:08:19 -08002307 .swap_cluster_max = max_t(unsigned long, nr_pages,
2308 SWAP_CLUSTER_MAX),
Andrew Morton179e9632006-03-22 00:08:18 -08002309 .gfp_mask = gfp_mask,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002310 .swappiness = vm_swappiness,
Johannes Weinerbd2f6192009-03-31 15:19:38 -07002311 .order = order,
Balbir Singh66e17072008-02-07 00:13:56 -08002312 .isolate_pages = isolate_pages_global,
Andrew Morton179e9632006-03-22 00:08:18 -08002313 };
Christoph Lameter83e33a42006-09-25 23:31:53 -07002314 unsigned long slab_reclaimable;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002315
2316 disable_swap_token();
Christoph Lameter9eeff232006-01-18 17:42:31 -08002317 cond_resched();
Christoph Lameterd4f77962006-02-24 13:04:22 -08002318 /*
2319 * We need to be able to allocate from the reserves for RECLAIM_SWAP
2320 * and we also need to be able to write out pages for RECLAIM_WRITE
2321 * and RECLAIM_SWAP.
2322 */
2323 p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002324 reclaim_state.reclaimed_slab = 0;
2325 p->reclaim_state = &reclaim_state;
Christoph Lameterc84db232006-02-01 03:05:29 -08002326
Christoph Lameter0ff38492006-09-25 23:31:52 -07002327 if (zone_page_state(zone, NR_FILE_PAGES) -
2328 zone_page_state(zone, NR_FILE_MAPPED) >
2329 zone->min_unmapped_pages) {
2330 /*
2331 * Free memory by calling shrink zone with increasing
2332 * priorities until we have enough memory freed.
2333 */
2334 priority = ZONE_RECLAIM_PRIORITY;
2335 do {
Martin Bligh3bb1a8522006-10-28 10:38:24 -07002336 note_zone_scanning_priority(zone, priority);
Rik van Riela79311c2009-01-06 14:40:01 -08002337 shrink_zone(priority, zone, &sc);
Christoph Lameter0ff38492006-09-25 23:31:52 -07002338 priority--;
Rik van Riela79311c2009-01-06 14:40:01 -08002339 } while (priority >= 0 && sc.nr_reclaimed < nr_pages);
Christoph Lameter0ff38492006-09-25 23:31:52 -07002340 }
Christoph Lameterc84db232006-02-01 03:05:29 -08002341
Christoph Lameter83e33a42006-09-25 23:31:53 -07002342 slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2343 if (slab_reclaimable > zone->min_slab_pages) {
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002344 /*
Christoph Lameter7fb2d462006-03-22 00:08:22 -08002345 * shrink_slab() does not currently allow us to determine how
Christoph Lameter0ff38492006-09-25 23:31:52 -07002346 * many pages were freed in this zone. So we take the current
2347 * number of slab pages and shake the slab until it is reduced
2348 * by the same nr_pages that we used for reclaiming unmapped
2349 * pages.
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002350 *
Christoph Lameter0ff38492006-09-25 23:31:52 -07002351 * Note that shrink_slab will free memory on all zones and may
2352 * take a long time.
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002353 */
Christoph Lameter0ff38492006-09-25 23:31:52 -07002354 while (shrink_slab(sc.nr_scanned, gfp_mask, order) &&
Christoph Lameter83e33a42006-09-25 23:31:53 -07002355 zone_page_state(zone, NR_SLAB_RECLAIMABLE) >
2356 slab_reclaimable - nr_pages)
Christoph Lameter0ff38492006-09-25 23:31:52 -07002357 ;
Christoph Lameter83e33a42006-09-25 23:31:53 -07002358
2359 /*
2360 * Update nr_reclaimed by the number of slab pages we
2361 * reclaimed from this zone.
2362 */
Rik van Riela79311c2009-01-06 14:40:01 -08002363 sc.nr_reclaimed += slab_reclaimable -
Christoph Lameter83e33a42006-09-25 23:31:53 -07002364 zone_page_state(zone, NR_SLAB_RECLAIMABLE);
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002365 }
2366
Christoph Lameter9eeff232006-01-18 17:42:31 -08002367 p->reclaim_state = NULL;
Christoph Lameterd4f77962006-02-24 13:04:22 -08002368 current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
Rik van Riela79311c2009-01-06 14:40:01 -08002369 return sc.nr_reclaimed >= nr_pages;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002370}
Andrew Morton179e9632006-03-22 00:08:18 -08002371
2372int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
2373{
Andrew Morton179e9632006-03-22 00:08:18 -08002374 int node_id;
David Rientjesd773ed62007-10-16 23:26:01 -07002375 int ret;
Andrew Morton179e9632006-03-22 00:08:18 -08002376
2377 /*
Christoph Lameter0ff38492006-09-25 23:31:52 -07002378 * Zone reclaim reclaims unmapped file backed pages and
2379 * slab pages if we are over the defined limits.
Christoph Lameter34aa1332006-06-30 01:55:37 -07002380 *
Christoph Lameter96146342006-07-03 00:24:13 -07002381 * A small portion of unmapped file backed pages is needed for
2382 * file I/O otherwise pages read by file I/O will be immediately
2383 * thrown out if the zone is overallocated. So we do not reclaim
2384 * if less than a specified percentage of the zone is used by
2385 * unmapped file backed pages.
Andrew Morton179e9632006-03-22 00:08:18 -08002386 */
Christoph Lameter34aa1332006-06-30 01:55:37 -07002387 if (zone_page_state(zone, NR_FILE_PAGES) -
Christoph Lameter0ff38492006-09-25 23:31:52 -07002388 zone_page_state(zone, NR_FILE_MAPPED) <= zone->min_unmapped_pages
2389 && zone_page_state(zone, NR_SLAB_RECLAIMABLE)
2390 <= zone->min_slab_pages)
Christoph Lameter96146342006-07-03 00:24:13 -07002391 return 0;
Andrew Morton179e9632006-03-22 00:08:18 -08002392
David Rientjesd773ed62007-10-16 23:26:01 -07002393 if (zone_is_all_unreclaimable(zone))
2394 return 0;
2395
Andrew Morton179e9632006-03-22 00:08:18 -08002396 /*
David Rientjesd773ed62007-10-16 23:26:01 -07002397 * Do not scan if the allocation should not be delayed.
Andrew Morton179e9632006-03-22 00:08:18 -08002398 */
David Rientjesd773ed62007-10-16 23:26:01 -07002399 if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
Andrew Morton179e9632006-03-22 00:08:18 -08002400 return 0;
2401
2402 /*
2403 * Only run zone reclaim on the local zone or on zones that do not
2404 * have associated processors. This will favor the local processor
2405 * over remote processors and spread off node memory allocations
2406 * as wide as possible.
2407 */
Christoph Lameter89fa3022006-09-25 23:31:55 -07002408 node_id = zone_to_nid(zone);
Christoph Lameter37c07082007-10-16 01:25:36 -07002409 if (node_state(node_id, N_CPU) && node_id != numa_node_id())
Andrew Morton179e9632006-03-22 00:08:18 -08002410 return 0;
David Rientjesd773ed62007-10-16 23:26:01 -07002411
2412 if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
2413 return 0;
2414 ret = __zone_reclaim(zone, gfp_mask, order);
2415 zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
2416
2417 return ret;
Andrew Morton179e9632006-03-22 00:08:18 -08002418}
Christoph Lameter9eeff232006-01-18 17:42:31 -08002419#endif
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002420
2421#ifdef CONFIG_UNEVICTABLE_LRU
2422/*
2423 * page_evictable - test whether a page is evictable
2424 * @page: the page to test
2425 * @vma: the VMA in which the page is or will be mapped, may be NULL
2426 *
2427 * Test whether page is evictable--i.e., should be placed on active/inactive
Nick Pigginb291f002008-10-18 20:26:44 -07002428 * lists vs unevictable list. The vma argument is !NULL when called from the
2429 * fault path to determine how to instantate a new page.
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002430 *
2431 * Reasons page might not be evictable:
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002432 * (1) page's mapping marked unevictable
Nick Pigginb291f002008-10-18 20:26:44 -07002433 * (2) page is part of an mlocked VMA
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002434 *
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002435 */
2436int page_evictable(struct page *page, struct vm_area_struct *vma)
2437{
2438
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002439 if (mapping_unevictable(page_mapping(page)))
2440 return 0;
2441
Nick Pigginb291f002008-10-18 20:26:44 -07002442 if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page)))
2443 return 0;
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002444
2445 return 1;
2446}
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002447
2448/**
2449 * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list
2450 * @page: page to check evictability and move to appropriate lru list
2451 * @zone: zone page is in
2452 *
2453 * Checks a page for evictability and moves the page to the appropriate
2454 * zone lru list.
2455 *
2456 * Restrictions: zone->lru_lock must be held, page must be on LRU and must
2457 * have PageUnevictable set.
2458 */
2459static void check_move_unevictable_page(struct page *page, struct zone *zone)
2460{
2461 VM_BUG_ON(PageActive(page));
2462
2463retry:
2464 ClearPageUnevictable(page);
2465 if (page_evictable(page, NULL)) {
2466 enum lru_list l = LRU_INACTIVE_ANON + page_is_file_cache(page);
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002467
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002468 __dec_zone_state(zone, NR_UNEVICTABLE);
2469 list_move(&page->lru, &zone->lru[l].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002470 mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l);
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002471 __inc_zone_state(zone, NR_INACTIVE_ANON + l);
2472 __count_vm_event(UNEVICTABLE_PGRESCUED);
2473 } else {
2474 /*
2475 * rotate unevictable list
2476 */
2477 SetPageUnevictable(page);
2478 list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002479 mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE);
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002480 if (page_evictable(page, NULL))
2481 goto retry;
2482 }
2483}
2484
2485/**
2486 * scan_mapping_unevictable_pages - scan an address space for evictable pages
2487 * @mapping: struct address_space to scan for evictable pages
2488 *
2489 * Scan all pages in mapping. Check unevictable pages for
2490 * evictability and move them to the appropriate zone lru list.
2491 */
2492void scan_mapping_unevictable_pages(struct address_space *mapping)
2493{
2494 pgoff_t next = 0;
2495 pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >>
2496 PAGE_CACHE_SHIFT;
2497 struct zone *zone;
2498 struct pagevec pvec;
2499
2500 if (mapping->nrpages == 0)
2501 return;
2502
2503 pagevec_init(&pvec, 0);
2504 while (next < end &&
2505 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
2506 int i;
2507 int pg_scanned = 0;
2508
2509 zone = NULL;
2510
2511 for (i = 0; i < pagevec_count(&pvec); i++) {
2512 struct page *page = pvec.pages[i];
2513 pgoff_t page_index = page->index;
2514 struct zone *pagezone = page_zone(page);
2515
2516 pg_scanned++;
2517 if (page_index > next)
2518 next = page_index;
2519 next++;
2520
2521 if (pagezone != zone) {
2522 if (zone)
2523 spin_unlock_irq(&zone->lru_lock);
2524 zone = pagezone;
2525 spin_lock_irq(&zone->lru_lock);
2526 }
2527
2528 if (PageLRU(page) && PageUnevictable(page))
2529 check_move_unevictable_page(page, zone);
2530 }
2531 if (zone)
2532 spin_unlock_irq(&zone->lru_lock);
2533 pagevec_release(&pvec);
2534
2535 count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned);
2536 }
2537
2538}
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002539
2540/**
2541 * scan_zone_unevictable_pages - check unevictable list for evictable pages
2542 * @zone - zone of which to scan the unevictable list
2543 *
2544 * Scan @zone's unevictable LRU lists to check for pages that have become
2545 * evictable. Move those that have to @zone's inactive list where they
2546 * become candidates for reclaim, unless shrink_inactive_zone() decides
2547 * to reactivate them. Pages that are still unevictable are rotated
2548 * back onto @zone's unevictable list.
2549 */
2550#define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */
KOSAKI Motohiro14b90b22009-01-06 14:39:45 -08002551static void scan_zone_unevictable_pages(struct zone *zone)
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002552{
2553 struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list;
2554 unsigned long scan;
2555 unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE);
2556
2557 while (nr_to_scan > 0) {
2558 unsigned long batch_size = min(nr_to_scan,
2559 SCAN_UNEVICTABLE_BATCH_SIZE);
2560
2561 spin_lock_irq(&zone->lru_lock);
2562 for (scan = 0; scan < batch_size; scan++) {
2563 struct page *page = lru_to_page(l_unevictable);
2564
2565 if (!trylock_page(page))
2566 continue;
2567
2568 prefetchw_prev_lru_page(page, l_unevictable, flags);
2569
2570 if (likely(PageLRU(page) && PageUnevictable(page)))
2571 check_move_unevictable_page(page, zone);
2572
2573 unlock_page(page);
2574 }
2575 spin_unlock_irq(&zone->lru_lock);
2576
2577 nr_to_scan -= batch_size;
2578 }
2579}
2580
2581
2582/**
2583 * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages
2584 *
2585 * A really big hammer: scan all zones' unevictable LRU lists to check for
2586 * pages that have become evictable. Move those back to the zones'
2587 * inactive list where they become candidates for reclaim.
2588 * This occurs when, e.g., we have unswappable pages on the unevictable lists,
2589 * and we add swap to the system. As such, it runs in the context of a task
2590 * that has possibly/probably made some previously unevictable pages
2591 * evictable.
2592 */
KOSAKI Motohiroff301532009-01-06 14:39:44 -08002593static void scan_all_zones_unevictable_pages(void)
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002594{
2595 struct zone *zone;
2596
2597 for_each_zone(zone) {
2598 scan_zone_unevictable_pages(zone);
2599 }
2600}
2601
2602/*
2603 * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of
2604 * all nodes' unevictable lists for evictable pages
2605 */
2606unsigned long scan_unevictable_pages;
2607
2608int scan_unevictable_handler(struct ctl_table *table, int write,
2609 struct file *file, void __user *buffer,
2610 size_t *length, loff_t *ppos)
2611{
2612 proc_doulongvec_minmax(table, write, file, buffer, length, ppos);
2613
2614 if (write && *(unsigned long *)table->data)
2615 scan_all_zones_unevictable_pages();
2616
2617 scan_unevictable_pages = 0;
2618 return 0;
2619}
2620
2621/*
2622 * per node 'scan_unevictable_pages' attribute. On demand re-scan of
2623 * a specified node's per zone unevictable lists for evictable pages.
2624 */
2625
2626static ssize_t read_scan_unevictable_node(struct sys_device *dev,
2627 struct sysdev_attribute *attr,
2628 char *buf)
2629{
2630 return sprintf(buf, "0\n"); /* always zero; should fit... */
2631}
2632
2633static ssize_t write_scan_unevictable_node(struct sys_device *dev,
2634 struct sysdev_attribute *attr,
2635 const char *buf, size_t count)
2636{
2637 struct zone *node_zones = NODE_DATA(dev->id)->node_zones;
2638 struct zone *zone;
2639 unsigned long res;
2640 unsigned long req = strict_strtoul(buf, 10, &res);
2641
2642 if (!req)
2643 return 1; /* zero is no-op */
2644
2645 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2646 if (!populated_zone(zone))
2647 continue;
2648 scan_zone_unevictable_pages(zone);
2649 }
2650 return 1;
2651}
2652
2653
2654static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
2655 read_scan_unevictable_node,
2656 write_scan_unevictable_node);
2657
2658int scan_unevictable_register_node(struct node *node)
2659{
2660 return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages);
2661}
2662
2663void scan_unevictable_unregister_node(struct node *node)
2664{
2665 sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages);
2666}
2667
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002668#endif