blob: 6177e3bcd66bdc7b8fc74cb583d47bb89c4940ef [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
Christoph Lameterf1fd1062006-01-18 17:42:30 -080063 /* Can pages be swapped as part of reclaim? */
64 int may_swap;
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 /* This context's SWAP_CLUSTER_MAX. If freeing memory for
67 * suspend, we effectively ignore SWAP_CLUSTER_MAX.
68 * In this context, it doesn't matter that we scan the
69 * whole list at once. */
70 int swap_cluster_max;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -070071
72 int swappiness;
Nick Piggin408d8542006-09-25 23:31:27 -070073
74 int all_unreclaimable;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -070075
76 int order;
Balbir Singh66e17072008-02-07 00:13:56 -080077
78 /* Which cgroup do we reclaim from */
79 struct mem_cgroup *mem_cgroup;
80
81 /* Pluggable isolate pages callback */
82 unsigned long (*isolate_pages)(unsigned long nr, struct list_head *dst,
83 unsigned long *scanned, int order, int mode,
84 struct zone *z, struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -070085 int active, int file);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086};
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
89
90#ifdef ARCH_HAS_PREFETCH
91#define prefetch_prev_lru_page(_page, _base, _field) \
92 do { \
93 if ((_page)->lru.prev != _base) { \
94 struct page *prev; \
95 \
96 prev = lru_to_page(&(_page->lru)); \
97 prefetch(&prev->_field); \
98 } \
99 } while (0)
100#else
101#define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
102#endif
103
104#ifdef ARCH_HAS_PREFETCHW
105#define prefetchw_prev_lru_page(_page, _base, _field) \
106 do { \
107 if ((_page)->lru.prev != _base) { \
108 struct page *prev; \
109 \
110 prev = lru_to_page(&(_page->lru)); \
111 prefetchw(&prev->_field); \
112 } \
113 } while (0)
114#else
115#define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
116#endif
117
118/*
119 * From 0 .. 100. Higher means more swappy.
120 */
121int vm_swappiness = 60;
Andrew Mortonbd1e22b2006-06-23 02:03:47 -0700122long vm_total_pages; /* The total number of pages which the VM controls */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124static LIST_HEAD(shrinker_list);
125static DECLARE_RWSEM(shrinker_rwsem);
126
Balbir Singh00f0b822008-03-04 14:28:39 -0800127#ifdef CONFIG_CGROUP_MEM_RES_CTLR
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800128#define scanning_global_lru(sc) (!(sc)->mem_cgroup)
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -0800129#else
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800130#define scanning_global_lru(sc) (1)
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -0800131#endif
132
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800133static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone,
134 struct scan_control *sc)
135{
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800136 if (!scanning_global_lru(sc))
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -0800137 return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone);
138
KOSAKI Motohiro6e901572009-01-07 18:08:15 -0800139 return &zone->reclaim_stat;
140}
141
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -0800142static unsigned long zone_nr_pages(struct zone *zone, struct scan_control *sc,
143 enum lru_list lru)
144{
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -0800145 if (!scanning_global_lru(sc))
KOSAKI Motohiroa3d8e052009-01-07 18:08:19 -0800146 return mem_cgroup_zone_nr_pages(sc->mem_cgroup, zone, lru);
147
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -0800148 return zone_page_state(zone, NR_LRU_BASE + lru);
149}
150
151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152/*
153 * Add a shrinker callback to be called from the vm
154 */
Rusty Russell8e1f9362007-07-17 04:03:17 -0700155void register_shrinker(struct shrinker *shrinker)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156{
Rusty Russell8e1f9362007-07-17 04:03:17 -0700157 shrinker->nr = 0;
158 down_write(&shrinker_rwsem);
159 list_add_tail(&shrinker->list, &shrinker_list);
160 up_write(&shrinker_rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161}
Rusty Russell8e1f9362007-07-17 04:03:17 -0700162EXPORT_SYMBOL(register_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164/*
165 * Remove one
166 */
Rusty Russell8e1f9362007-07-17 04:03:17 -0700167void unregister_shrinker(struct shrinker *shrinker)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
169 down_write(&shrinker_rwsem);
170 list_del(&shrinker->list);
171 up_write(&shrinker_rwsem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172}
Rusty Russell8e1f9362007-07-17 04:03:17 -0700173EXPORT_SYMBOL(unregister_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175#define SHRINK_BATCH 128
176/*
177 * Call the shrink functions to age shrinkable caches
178 *
179 * Here we assume it costs one seek to replace a lru page and that it also
180 * takes a seek to recreate a cache object. With this in mind we age equal
181 * percentages of the lru and ageable caches. This should balance the seeks
182 * generated by these structures.
183 *
Simon Arlott183ff222007-10-20 01:27:18 +0200184 * If the vm encountered mapped pages on the LRU it increase the pressure on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 * slab to avoid swapping.
186 *
187 * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
188 *
189 * `lru_pages' represents the number of on-LRU pages in all the zones which
190 * are eligible for the caller's allocation attempt. It is used for balancing
191 * slab reclaim versus page reclaim.
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700192 *
193 * Returns the number of slab objects which we shrunk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 */
Andrew Morton69e05942006-03-22 00:08:19 -0800195unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask,
196 unsigned long lru_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 struct shrinker *shrinker;
Andrew Morton69e05942006-03-22 00:08:19 -0800199 unsigned long ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
201 if (scanned == 0)
202 scanned = SWAP_CLUSTER_MAX;
203
204 if (!down_read_trylock(&shrinker_rwsem))
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700205 return 1; /* Assume we'll be able to shrink next time */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207 list_for_each_entry(shrinker, &shrinker_list, list) {
208 unsigned long long delta;
209 unsigned long total_scan;
Rusty Russell8e1f9362007-07-17 04:03:17 -0700210 unsigned long max_pass = (*shrinker->shrink)(0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212 delta = (4 * scanned) / shrinker->seeks;
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800213 delta *= max_pass;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 do_div(delta, lru_pages + 1);
215 shrinker->nr += delta;
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800216 if (shrinker->nr < 0) {
217 printk(KERN_ERR "%s: nr=%ld\n",
Harvey Harrisond40cee22008-04-30 00:55:07 -0700218 __func__, shrinker->nr);
Andrea Arcangeliea164d72005-11-28 13:44:15 -0800219 shrinker->nr = max_pass;
220 }
221
222 /*
223 * Avoid risking looping forever due to too large nr value:
224 * never try to free more than twice the estimate number of
225 * freeable entries.
226 */
227 if (shrinker->nr > max_pass * 2)
228 shrinker->nr = max_pass * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230 total_scan = shrinker->nr;
231 shrinker->nr = 0;
232
233 while (total_scan >= SHRINK_BATCH) {
234 long this_scan = SHRINK_BATCH;
235 int shrink_ret;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700236 int nr_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Rusty Russell8e1f9362007-07-17 04:03:17 -0700238 nr_before = (*shrinker->shrink)(0, gfp_mask);
239 shrink_ret = (*shrinker->shrink)(this_scan, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 if (shrink_ret == -1)
241 break;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700242 if (shrink_ret < nr_before)
243 ret += nr_before - shrink_ret;
Christoph Lameterf8891e52006-06-30 01:55:45 -0700244 count_vm_events(SLABS_SCANNED, this_scan);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 total_scan -= this_scan;
246
247 cond_resched();
248 }
249
250 shrinker->nr += total_scan;
251 }
252 up_read(&shrinker_rwsem);
akpm@osdl.orgb15e0902005-06-21 17:14:35 -0700253 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254}
255
256/* Called without lock on whether page is mapped, so answer is unstable */
257static inline int page_mapping_inuse(struct page *page)
258{
259 struct address_space *mapping;
260
261 /* Page is in somebody's page tables. */
262 if (page_mapped(page))
263 return 1;
264
265 /* Be more reluctant to reclaim swapcache than pagecache */
266 if (PageSwapCache(page))
267 return 1;
268
269 mapping = page_mapping(page);
270 if (!mapping)
271 return 0;
272
273 /* File is mmap'd by somebody? */
274 return mapping_mapped(mapping);
275}
276
277static inline int is_page_cache_freeable(struct page *page)
278{
279 return page_count(page) - !!PagePrivate(page) == 2;
280}
281
282static int may_write_to_queue(struct backing_dev_info *bdi)
283{
Christoph Lameter930d9152006-01-08 01:00:47 -0800284 if (current->flags & PF_SWAPWRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return 1;
286 if (!bdi_write_congested(bdi))
287 return 1;
288 if (bdi == current->backing_dev_info)
289 return 1;
290 return 0;
291}
292
293/*
294 * We detected a synchronous write error writing a page out. Probably
295 * -ENOSPC. We need to propagate that into the address_space for a subsequent
296 * fsync(), msync() or close().
297 *
298 * The tricky part is that after writepage we cannot touch the mapping: nothing
299 * prevents it from being freed up. But we have a ref on the page and once
300 * that page is locked, the mapping is pinned.
301 *
302 * We're allowed to run sleeping lock_page() here because we know the caller has
303 * __GFP_FS.
304 */
305static void handle_write_error(struct address_space *mapping,
306 struct page *page, int error)
307{
308 lock_page(page);
Guillaume Chazarain3e9f45b2007-05-08 00:23:25 -0700309 if (page_mapping(page) == mapping)
310 mapping_set_error(mapping, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 unlock_page(page);
312}
313
Andy Whitcroftc661b072007-08-22 14:01:26 -0700314/* Request for sync pageout. */
315enum pageout_io {
316 PAGEOUT_IO_ASYNC,
317 PAGEOUT_IO_SYNC,
318};
319
Christoph Lameter04e62a22006-06-23 02:03:38 -0700320/* possible outcome of pageout() */
321typedef enum {
322 /* failed to write page out, page is locked */
323 PAGE_KEEP,
324 /* move page to the active list, page is locked */
325 PAGE_ACTIVATE,
326 /* page has been sent to the disk successfully, page is unlocked */
327 PAGE_SUCCESS,
328 /* page is clean and locked */
329 PAGE_CLEAN,
330} pageout_t;
331
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332/*
Andrew Morton1742f192006-03-22 00:08:21 -0800333 * pageout is called by shrink_page_list() for each dirty page.
334 * Calls ->writepage().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
Andy Whitcroftc661b072007-08-22 14:01:26 -0700336static pageout_t pageout(struct page *page, struct address_space *mapping,
337 enum pageout_io sync_writeback)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
339 /*
340 * If the page is dirty, only perform writeback if that write
341 * will be non-blocking. To prevent this allocation from being
342 * stalled by pagecache activity. But note that there may be
343 * stalls if we need to run get_block(). We could test
344 * PagePrivate for that.
345 *
346 * If this process is currently in generic_file_write() against
347 * this page's queue, we can perform writeback even if that
348 * will block.
349 *
350 * If the page is swapcache, write it back even if that would
351 * block, for some throttling. This happens by accident, because
352 * swap_backing_dev_info is bust: it doesn't reflect the
353 * congestion state of the swapdevs. Easy to fix, if needed.
354 * See swapfile.c:page_queue_congested().
355 */
356 if (!is_page_cache_freeable(page))
357 return PAGE_KEEP;
358 if (!mapping) {
359 /*
360 * Some data journaling orphaned pages can have
361 * page->mapping == NULL while being dirty with clean buffers.
362 */
akpm@osdl.org323aca62005-04-16 15:24:06 -0700363 if (PagePrivate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 if (try_to_free_buffers(page)) {
365 ClearPageDirty(page);
Harvey Harrisond40cee22008-04-30 00:55:07 -0700366 printk("%s: orphaned page\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 return PAGE_CLEAN;
368 }
369 }
370 return PAGE_KEEP;
371 }
372 if (mapping->a_ops->writepage == NULL)
373 return PAGE_ACTIVATE;
374 if (!may_write_to_queue(mapping->backing_dev_info))
375 return PAGE_KEEP;
376
377 if (clear_page_dirty_for_io(page)) {
378 int res;
379 struct writeback_control wbc = {
380 .sync_mode = WB_SYNC_NONE,
381 .nr_to_write = SWAP_CLUSTER_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700382 .range_start = 0,
383 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 .nonblocking = 1,
385 .for_reclaim = 1,
386 };
387
388 SetPageReclaim(page);
389 res = mapping->a_ops->writepage(page, &wbc);
390 if (res < 0)
391 handle_write_error(mapping, page, res);
Zach Brown994fc28c2005-12-15 14:28:17 -0800392 if (res == AOP_WRITEPAGE_ACTIVATE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 ClearPageReclaim(page);
394 return PAGE_ACTIVATE;
395 }
Andy Whitcroftc661b072007-08-22 14:01:26 -0700396
397 /*
398 * Wait on writeback if requested to. This happens when
399 * direct reclaiming a large contiguous area and the
400 * first attempt to free a range of pages fails.
401 */
402 if (PageWriteback(page) && sync_writeback == PAGEOUT_IO_SYNC)
403 wait_on_page_writeback(page);
404
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 if (!PageWriteback(page)) {
406 /* synchronous write or broken a_ops? */
407 ClearPageReclaim(page);
408 }
Andrew Mortone129b5c2006-09-27 01:50:00 -0700409 inc_zone_page_state(page, NR_VMSCAN_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 return PAGE_SUCCESS;
411 }
412
413 return PAGE_CLEAN;
414}
415
Andrew Mortona649fd92006-10-17 00:09:36 -0700416/*
Nick Piggine2867812008-07-25 19:45:30 -0700417 * Same as remove_mapping, but if the page is removed from the mapping, it
418 * gets returned with a refcount of 0.
Andrew Mortona649fd92006-10-17 00:09:36 -0700419 */
Nick Piggine2867812008-07-25 19:45:30 -0700420static int __remove_mapping(struct address_space *mapping, struct page *page)
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800421{
Nick Piggin28e4d962006-09-25 23:31:23 -0700422 BUG_ON(!PageLocked(page));
423 BUG_ON(mapping != page_mapping(page));
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800424
Nick Piggin19fd6232008-07-25 19:45:32 -0700425 spin_lock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800426 /*
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700427 * The non racy check for a busy page.
428 *
429 * Must be careful with the order of the tests. When someone has
430 * a ref to the page, it may be possible that they dirty it then
431 * drop the reference. So if PageDirty is tested before page_count
432 * here, then the following race may occur:
433 *
434 * get_user_pages(&page);
435 * [user mapping goes away]
436 * write_to(page);
437 * !PageDirty(page) [good]
438 * SetPageDirty(page);
439 * put_page(page);
440 * !page_count(page) [good, discard it]
441 *
442 * [oops, our write_to data is lost]
443 *
444 * Reversing the order of the tests ensures such a situation cannot
445 * escape unnoticed. The smp_rmb is needed to ensure the page->flags
446 * load is not satisfied before that of page->_count.
447 *
448 * Note that if SetPageDirty is always performed via set_page_dirty,
449 * and thus under tree_lock, then this ordering is not required.
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800450 */
Nick Piggine2867812008-07-25 19:45:30 -0700451 if (!page_freeze_refs(page, 2))
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800452 goto cannot_free;
Nick Piggine2867812008-07-25 19:45:30 -0700453 /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
454 if (unlikely(PageDirty(page))) {
455 page_unfreeze_refs(page, 2);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800456 goto cannot_free;
Nick Piggine2867812008-07-25 19:45:30 -0700457 }
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800458
459 if (PageSwapCache(page)) {
460 swp_entry_t swap = { .val = page_private(page) };
461 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700462 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800463 swap_free(swap);
Nick Piggine2867812008-07-25 19:45:30 -0700464 } else {
465 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700466 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800467 }
468
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800469 return 1;
470
471cannot_free:
Nick Piggin19fd6232008-07-25 19:45:32 -0700472 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800473 return 0;
474}
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476/*
Nick Piggine2867812008-07-25 19:45:30 -0700477 * Attempt to detach a locked page from its ->mapping. If it is dirty or if
478 * someone else has a ref on the page, abort and return 0. If it was
479 * successfully detached, return 1. Assumes the caller has a single ref on
480 * this page.
481 */
482int remove_mapping(struct address_space *mapping, struct page *page)
483{
484 if (__remove_mapping(mapping, page)) {
485 /*
486 * Unfreezing the refcount with 1 rather than 2 effectively
487 * drops the pagecache ref for us without requiring another
488 * atomic operation.
489 */
490 page_unfreeze_refs(page, 1);
491 return 1;
492 }
493 return 0;
494}
495
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700496/**
497 * putback_lru_page - put previously isolated page onto appropriate LRU list
498 * @page: page to be put back to appropriate lru list
499 *
500 * Add previously isolated @page to appropriate LRU list.
501 * Page may still be unevictable for other reasons.
502 *
503 * lru_lock must not be held, interrupts must be enabled.
504 */
505#ifdef CONFIG_UNEVICTABLE_LRU
506void putback_lru_page(struct page *page)
507{
508 int lru;
509 int active = !!TestClearPageActive(page);
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700510 int was_unevictable = PageUnevictable(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700511
512 VM_BUG_ON(PageLRU(page));
513
514redo:
515 ClearPageUnevictable(page);
516
517 if (page_evictable(page, NULL)) {
518 /*
519 * For evictable pages, we can use the cache.
520 * In event of a race, worst case is we end up with an
521 * unevictable page on [in]active list.
522 * We know how to handle that.
523 */
524 lru = active + page_is_file_cache(page);
525 lru_cache_add_lru(page, lru);
526 } else {
527 /*
528 * Put unevictable pages directly on zone's unevictable
529 * list.
530 */
531 lru = LRU_UNEVICTABLE;
532 add_page_to_unevictable_list(page);
533 }
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700534
535 /*
536 * page's status can change while we move it among lru. If an evictable
537 * page is on unevictable list, it never be freed. To avoid that,
538 * check after we added it to the list, again.
539 */
540 if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) {
541 if (!isolate_lru_page(page)) {
542 put_page(page);
543 goto redo;
544 }
545 /* This means someone else dropped this page from LRU
546 * So, it will be freed or putback to LRU again. There is
547 * nothing to do here.
548 */
549 }
550
Lee Schermerhornbbfd28e2008-10-18 20:26:40 -0700551 if (was_unevictable && lru != LRU_UNEVICTABLE)
552 count_vm_event(UNEVICTABLE_PGRESCUED);
553 else if (!was_unevictable && lru == LRU_UNEVICTABLE)
554 count_vm_event(UNEVICTABLE_PGCULLED);
555
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700556 put_page(page); /* drop ref from isolate */
557}
558
559#else /* CONFIG_UNEVICTABLE_LRU */
560
561void putback_lru_page(struct page *page)
562{
563 int lru;
564 VM_BUG_ON(PageLRU(page));
565
566 lru = !!TestClearPageActive(page) + page_is_file_cache(page);
567 lru_cache_add_lru(page, lru);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700568 put_page(page);
569}
570#endif /* CONFIG_UNEVICTABLE_LRU */
571
572
Nick Piggine2867812008-07-25 19:45:30 -0700573/*
Andrew Morton1742f192006-03-22 00:08:21 -0800574 * shrink_page_list() returns the number of reclaimed pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 */
Andrew Morton1742f192006-03-22 00:08:21 -0800576static unsigned long shrink_page_list(struct list_head *page_list,
Andy Whitcroftc661b072007-08-22 14:01:26 -0700577 struct scan_control *sc,
578 enum pageout_io sync_writeback)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
580 LIST_HEAD(ret_pages);
581 struct pagevec freed_pvec;
582 int pgactivate = 0;
Andrew Morton05ff5132006-03-22 00:08:20 -0800583 unsigned long nr_reclaimed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 cond_resched();
586
587 pagevec_init(&freed_pvec, 1);
588 while (!list_empty(page_list)) {
589 struct address_space *mapping;
590 struct page *page;
591 int may_enter_fs;
592 int referenced;
593
594 cond_resched();
595
596 page = lru_to_page(page_list);
597 list_del(&page->lru);
598
Nick Piggin529ae9a2008-08-02 12:01:03 +0200599 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 goto keep;
601
Nick Piggin725d7042006-09-25 23:30:55 -0700602 VM_BUG_ON(PageActive(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604 sc->nr_scanned++;
Christoph Lameter80e43422006-02-11 17:55:53 -0800605
Nick Pigginb291f002008-10-18 20:26:44 -0700606 if (unlikely(!page_evictable(page, NULL)))
607 goto cull_mlocked;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700608
Christoph Lameter80e43422006-02-11 17:55:53 -0800609 if (!sc->may_swap && page_mapped(page))
610 goto keep_locked;
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 /* Double the slab pressure for mapped and swapcache pages */
613 if (page_mapped(page) || PageSwapCache(page))
614 sc->nr_scanned++;
615
Andy Whitcroftc661b072007-08-22 14:01:26 -0700616 may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
617 (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
618
619 if (PageWriteback(page)) {
620 /*
621 * Synchronous reclaim is performed in two passes,
622 * first an asynchronous pass over the list to
623 * start parallel writeback, and a second synchronous
624 * pass to wait for the IO to complete. Wait here
625 * for any page for which writeback has already
626 * started.
627 */
628 if (sync_writeback == PAGEOUT_IO_SYNC && may_enter_fs)
629 wait_on_page_writeback(page);
Andrew Morton4dd4b922008-03-24 12:29:52 -0700630 else
Andy Whitcroftc661b072007-08-22 14:01:26 -0700631 goto keep_locked;
632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Balbir Singhbed71612008-02-07 00:14:01 -0800634 referenced = page_referenced(page, 1, sc->mem_cgroup);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 /* In active use or really unfreeable? Activate it. */
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700636 if (sc->order <= PAGE_ALLOC_COSTLY_ORDER &&
637 referenced && page_mapping_inuse(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 goto activate_locked;
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 /*
641 * Anonymous process memory has backing store?
642 * Try to allocate it some swap space here.
643 */
Nick Pigginb291f002008-10-18 20:26:44 -0700644 if (PageAnon(page) && !PageSwapCache(page)) {
Hugh Dickins63eb6b92008-11-19 15:36:37 -0800645 if (!(sc->gfp_mask & __GFP_IO))
646 goto keep_locked;
Hugh Dickinsac47b002009-01-06 14:39:39 -0800647 if (!add_to_swap(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 goto activate_locked;
Hugh Dickins63eb6b92008-11-19 15:36:37 -0800649 may_enter_fs = 1;
Nick Pigginb291f002008-10-18 20:26:44 -0700650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
652 mapping = page_mapping(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 /*
655 * The page is mapped into the page tables of one or more
656 * processes. Try to unmap it here.
657 */
658 if (page_mapped(page) && mapping) {
Christoph Lametera48d07a2006-02-01 03:05:38 -0800659 switch (try_to_unmap(page, 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 case SWAP_FAIL:
661 goto activate_locked;
662 case SWAP_AGAIN:
663 goto keep_locked;
Nick Pigginb291f002008-10-18 20:26:44 -0700664 case SWAP_MLOCK:
665 goto cull_mlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 case SWAP_SUCCESS:
667 ; /* try to free the page below */
668 }
669 }
670
671 if (PageDirty(page)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700672 if (sc->order <= PAGE_ALLOC_COSTLY_ORDER && referenced)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 goto keep_locked;
Andrew Morton4dd4b922008-03-24 12:29:52 -0700674 if (!may_enter_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 goto keep_locked;
Christoph Lameter52a83632006-02-01 03:05:28 -0800676 if (!sc->may_writepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 goto keep_locked;
678
679 /* Page is dirty, try to write it out here */
Andy Whitcroftc661b072007-08-22 14:01:26 -0700680 switch (pageout(page, mapping, sync_writeback)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 case PAGE_KEEP:
682 goto keep_locked;
683 case PAGE_ACTIVATE:
684 goto activate_locked;
685 case PAGE_SUCCESS:
Andrew Morton4dd4b922008-03-24 12:29:52 -0700686 if (PageWriteback(page) || PageDirty(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 goto keep;
688 /*
689 * A synchronous write - probably a ramdisk. Go
690 * ahead and try to reclaim the page.
691 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200692 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 goto keep;
694 if (PageDirty(page) || PageWriteback(page))
695 goto keep_locked;
696 mapping = page_mapping(page);
697 case PAGE_CLEAN:
698 ; /* try to free the page below */
699 }
700 }
701
702 /*
703 * If the page has buffers, try to free the buffer mappings
704 * associated with this page. If we succeed we try to free
705 * the page as well.
706 *
707 * We do this even if the page is PageDirty().
708 * try_to_release_page() does not perform I/O, but it is
709 * possible for a page to have PageDirty set, but it is actually
710 * clean (all its buffers are clean). This happens if the
711 * buffers were written out directly, with submit_bh(). ext3
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700712 * will do this, as well as the blockdev mapping.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 * try_to_release_page() will discover that cleanness and will
714 * drop the buffers and mark the page clean - it can be freed.
715 *
716 * Rarely, pages can have buffers and no ->mapping. These are
717 * the pages which were not successfully invalidated in
718 * truncate_complete_page(). We try to drop those buffers here
719 * and if that worked, and the page is no longer mapped into
720 * process address space (page_count == 1) it can be freed.
721 * Otherwise, leave the page on the LRU so it is swappable.
722 */
723 if (PagePrivate(page)) {
724 if (!try_to_release_page(page, sc->gfp_mask))
725 goto activate_locked;
Nick Piggine2867812008-07-25 19:45:30 -0700726 if (!mapping && page_count(page) == 1) {
727 unlock_page(page);
728 if (put_page_testzero(page))
729 goto free_it;
730 else {
731 /*
732 * rare race with speculative reference.
733 * the speculative reference will free
734 * this page shortly, so we may
735 * increment nr_reclaimed here (and
736 * leave it off the LRU).
737 */
738 nr_reclaimed++;
739 continue;
740 }
741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
743
Nick Piggine2867812008-07-25 19:45:30 -0700744 if (!mapping || !__remove_mapping(mapping, page))
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800745 goto keep_locked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Nick Piggina978d6f2008-10-18 20:26:58 -0700747 /*
748 * At this point, we have no other references and there is
749 * no way to pick any more up (removed from LRU, removed
750 * from pagecache). Can use non-atomic bitops now (and
751 * we obviously don't have to worry about waking up a process
752 * waiting on the page lock, because there are no references.
753 */
754 __clear_page_locked(page);
Nick Piggine2867812008-07-25 19:45:30 -0700755free_it:
Andrew Morton05ff5132006-03-22 00:08:20 -0800756 nr_reclaimed++;
Nick Piggine2867812008-07-25 19:45:30 -0700757 if (!pagevec_add(&freed_pvec, page)) {
758 __pagevec_free(&freed_pvec);
759 pagevec_reinit(&freed_pvec);
760 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 continue;
762
Nick Pigginb291f002008-10-18 20:26:44 -0700763cull_mlocked:
Hugh Dickins63d6c5a2009-01-06 14:39:38 -0800764 if (PageSwapCache(page))
765 try_to_free_swap(page);
Nick Pigginb291f002008-10-18 20:26:44 -0700766 unlock_page(page);
767 putback_lru_page(page);
768 continue;
769
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770activate_locked:
Rik van Riel68a223942008-10-18 20:26:23 -0700771 /* Not a candidate for swapping, so reclaim swap space. */
772 if (PageSwapCache(page) && vm_swap_full())
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800773 try_to_free_swap(page);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700774 VM_BUG_ON(PageActive(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 SetPageActive(page);
776 pgactivate++;
777keep_locked:
778 unlock_page(page);
779keep:
780 list_add(&page->lru, &ret_pages);
Nick Pigginb291f002008-10-18 20:26:44 -0700781 VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
783 list_splice(&ret_pages, page_list);
784 if (pagevec_count(&freed_pvec))
Nick Piggine2867812008-07-25 19:45:30 -0700785 __pagevec_free(&freed_pvec);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700786 count_vm_events(PGACTIVATE, pgactivate);
Andrew Morton05ff5132006-03-22 00:08:20 -0800787 return nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700790/* LRU Isolation modes. */
791#define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */
792#define ISOLATE_ACTIVE 1 /* Isolate active pages. */
793#define ISOLATE_BOTH 2 /* Isolate both active and inactive pages. */
794
795/*
796 * Attempt to remove the specified page from its LRU. Only take this page
797 * if it is of the appropriate PageActive status. Pages which are being
798 * freed elsewhere are also ignored.
799 *
800 * page: page to consider
801 * mode: one of the LRU isolation modes defined above
802 *
803 * returns 0 on success, -ve errno on failure.
804 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700805int __isolate_lru_page(struct page *page, int mode, int file)
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700806{
807 int ret = -EINVAL;
808
809 /* Only take pages on the LRU. */
810 if (!PageLRU(page))
811 return ret;
812
813 /*
814 * When checking the active state, we need to be sure we are
815 * dealing with comparible boolean values. Take the logical not
816 * of each.
817 */
818 if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
819 return ret;
820
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700821 if (mode != ISOLATE_BOTH && (!page_is_file_cache(page) != !file))
822 return ret;
823
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700824 /*
825 * When this function is being called for lumpy reclaim, we
826 * initially look into all LRU pages, active, inactive and
827 * unevictable; only give shrink_page_list evictable pages.
828 */
829 if (PageUnevictable(page))
830 return ret;
831
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700832 ret = -EBUSY;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800833
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700834 if (likely(get_page_unless_zero(page))) {
835 /*
836 * Be careful not to clear PageLRU until after we're
837 * sure the page is not being freed elsewhere -- the
838 * page release code relies on it.
839 */
840 ClearPageLRU(page);
841 ret = 0;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -0800842 mem_cgroup_del_lru(page);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700843 }
844
845 return ret;
846}
847
Christoph Lameter49d2e9c2006-01-08 01:00:48 -0800848/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 * zone->lru_lock is heavily contended. Some of the functions that
850 * shrink the lists perform better by taking out a batch of pages
851 * and working on them outside the LRU lock.
852 *
853 * For pagecache intensive workloads, this function is the hottest
854 * spot in the kernel (apart from copy_*_user functions).
855 *
856 * Appropriate locks must be held before calling this function.
857 *
858 * @nr_to_scan: The number of pages to look through on the list.
859 * @src: The LRU list to pull pages off.
860 * @dst: The temp list to put pages on to.
861 * @scanned: The number of pages that were scanned.
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700862 * @order: The caller's attempted allocation order
863 * @mode: One of the LRU isolation modes
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700864 * @file: True [1] if isolating file [!anon] pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 *
866 * returns how many pages were moved onto *@dst.
867 */
Andrew Morton69e05942006-03-22 00:08:19 -0800868static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
869 struct list_head *src, struct list_head *dst,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700870 unsigned long *scanned, int order, int mode, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
Andrew Morton69e05942006-03-22 00:08:19 -0800872 unsigned long nr_taken = 0;
Wu Fengguangc9b02d92006-03-22 00:08:23 -0800873 unsigned long scan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Wu Fengguangc9b02d92006-03-22 00:08:23 -0800875 for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700876 struct page *page;
877 unsigned long pfn;
878 unsigned long end_pfn;
879 unsigned long page_pfn;
880 int zone_id;
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 page = lru_to_page(src);
883 prefetchw_prev_lru_page(page, src, flags);
884
Nick Piggin725d7042006-09-25 23:30:55 -0700885 VM_BUG_ON(!PageLRU(page));
Nick Piggin8d438f92006-03-22 00:07:59 -0800886
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700887 switch (__isolate_lru_page(page, mode, file)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700888 case 0:
889 list_move(&page->lru, dst);
Nick Piggin7c8ee9a2006-03-22 00:08:03 -0800890 nr_taken++;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700891 break;
Nick Piggin46453a62006-03-22 00:07:58 -0800892
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700893 case -EBUSY:
894 /* else it is being freed elsewhere */
895 list_move(&page->lru, src);
896 continue;
897
898 default:
899 BUG();
900 }
901
902 if (!order)
903 continue;
904
905 /*
906 * Attempt to take all pages in the order aligned region
907 * surrounding the tag page. Only take those pages of
908 * the same active state as that tag page. We may safely
909 * round the target page pfn down to the requested order
910 * as the mem_map is guarenteed valid out to MAX_ORDER,
911 * where that page is in a different zone we will detect
912 * it from its zone id and abort this block scan.
913 */
914 zone_id = page_zone_id(page);
915 page_pfn = page_to_pfn(page);
916 pfn = page_pfn & ~((1 << order) - 1);
917 end_pfn = pfn + (1 << order);
918 for (; pfn < end_pfn; pfn++) {
919 struct page *cursor_page;
920
921 /* The target page is in the block, ignore it. */
922 if (unlikely(pfn == page_pfn))
923 continue;
924
925 /* Avoid holes within the zone. */
926 if (unlikely(!pfn_valid_within(pfn)))
927 break;
928
929 cursor_page = pfn_to_page(pfn);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700930
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700931 /* Check that we have not crossed a zone boundary. */
932 if (unlikely(page_zone_id(cursor_page) != zone_id))
933 continue;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700934 switch (__isolate_lru_page(cursor_page, mode, file)) {
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700935 case 0:
936 list_move(&cursor_page->lru, dst);
937 nr_taken++;
938 scan++;
939 break;
940
941 case -EBUSY:
942 /* else it is being freed elsewhere */
943 list_move(&cursor_page->lru, src);
944 default:
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700945 break; /* ! on LRU or wrong list */
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700946 }
947 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 }
949
950 *scanned = scan;
951 return nr_taken;
952}
953
Balbir Singh66e17072008-02-07 00:13:56 -0800954static unsigned long isolate_pages_global(unsigned long nr,
955 struct list_head *dst,
956 unsigned long *scanned, int order,
957 int mode, struct zone *z,
958 struct mem_cgroup *mem_cont,
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700959 int active, int file)
Balbir Singh66e17072008-02-07 00:13:56 -0800960{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700961 int lru = LRU_BASE;
Balbir Singh66e17072008-02-07 00:13:56 -0800962 if (active)
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700963 lru += LRU_ACTIVE;
964 if (file)
965 lru += LRU_FILE;
966 return isolate_lru_pages(nr, &z->lru[lru].list, dst, scanned, order,
967 mode, !!file);
Balbir Singh66e17072008-02-07 00:13:56 -0800968}
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970/*
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700971 * clear_active_flags() is a helper for shrink_active_list(), clearing
972 * any active bits from the pages in the list.
973 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700974static unsigned long clear_active_flags(struct list_head *page_list,
975 unsigned int *count)
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700976{
977 int nr_active = 0;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700978 int lru;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700979 struct page *page;
980
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700981 list_for_each_entry(page, page_list, lru) {
982 lru = page_is_file_cache(page);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700983 if (PageActive(page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700984 lru += LRU_ACTIVE;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700985 ClearPageActive(page);
986 nr_active++;
987 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700988 count[lru]++;
989 }
Andy Whitcroft5ad333e2007-07-17 04:03:16 -0700990
991 return nr_active;
992}
993
Nick Piggin62695a82008-10-18 20:26:09 -0700994/**
995 * isolate_lru_page - tries to isolate a page from its LRU list
996 * @page: page to isolate from its LRU list
997 *
998 * Isolates a @page from an LRU list, clears PageLRU and adjusts the
999 * vmstat statistic corresponding to whatever LRU list the page was on.
1000 *
1001 * Returns 0 if the page was removed from an LRU list.
1002 * Returns -EBUSY if the page was not on an LRU list.
1003 *
1004 * The returned page will have PageLRU() cleared. If it was found on
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001005 * the active list, it will have PageActive set. If it was found on
1006 * the unevictable list, it will have the PageUnevictable bit set. That flag
1007 * may need to be cleared by the caller before letting the page go.
Nick Piggin62695a82008-10-18 20:26:09 -07001008 *
1009 * The vmstat statistic corresponding to the list on which the page was
1010 * found will be decremented.
1011 *
1012 * Restrictions:
1013 * (1) Must be called with an elevated refcount on the page. This is a
1014 * fundamentnal difference from isolate_lru_pages (which is called
1015 * without a stable reference).
1016 * (2) the lru_lock must not be held.
1017 * (3) interrupts must be enabled.
1018 */
1019int isolate_lru_page(struct page *page)
1020{
1021 int ret = -EBUSY;
1022
1023 if (PageLRU(page)) {
1024 struct zone *zone = page_zone(page);
1025
1026 spin_lock_irq(&zone->lru_lock);
1027 if (PageLRU(page) && get_page_unless_zero(page)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001028 int lru = page_lru(page);
Nick Piggin62695a82008-10-18 20:26:09 -07001029 ret = 0;
1030 ClearPageLRU(page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001031
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001032 del_page_from_lru_list(zone, page, lru);
Nick Piggin62695a82008-10-18 20:26:09 -07001033 }
1034 spin_unlock_irq(&zone->lru_lock);
1035 }
1036 return ret;
1037}
1038
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001039/*
Andrew Morton1742f192006-03-22 00:08:21 -08001040 * shrink_inactive_list() is a helper for shrink_zone(). It returns the number
1041 * of reclaimed pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 */
Andrew Morton1742f192006-03-22 00:08:21 -08001043static unsigned long shrink_inactive_list(unsigned long max_scan,
Rik van Riel33c120e2008-10-18 20:26:36 -07001044 struct zone *zone, struct scan_control *sc,
1045 int priority, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
1047 LIST_HEAD(page_list);
1048 struct pagevec pvec;
Andrew Morton69e05942006-03-22 00:08:19 -08001049 unsigned long nr_scanned = 0;
Andrew Morton05ff5132006-03-22 00:08:20 -08001050 unsigned long nr_reclaimed = 0;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001051 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 pagevec_init(&pvec, 1);
1054
1055 lru_add_drain();
1056 spin_lock_irq(&zone->lru_lock);
Andrew Morton69e05942006-03-22 00:08:19 -08001057 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 struct page *page;
Andrew Morton69e05942006-03-22 00:08:19 -08001059 unsigned long nr_taken;
1060 unsigned long nr_scan;
1061 unsigned long nr_freed;
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001062 unsigned long nr_active;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001063 unsigned int count[NR_LRU_LISTS] = { 0, };
Rik van Riel33c120e2008-10-18 20:26:36 -07001064 int mode = ISOLATE_INACTIVE;
1065
1066 /*
1067 * If we need a large contiguous chunk of memory, or have
1068 * trouble getting a small set of contiguous pages, we
1069 * will reclaim both active and inactive pages.
1070 *
1071 * We use the same threshold as pageout congestion_wait below.
1072 */
1073 if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
1074 mode = ISOLATE_BOTH;
1075 else if (sc->order && priority < DEF_PRIORITY - 2)
1076 mode = ISOLATE_BOTH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
Balbir Singh66e17072008-02-07 00:13:56 -08001078 nr_taken = sc->isolate_pages(sc->swap_cluster_max,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001079 &page_list, &nr_scan, sc->order, mode,
1080 zone, sc->mem_cgroup, 0, file);
1081 nr_active = clear_active_flags(&page_list, count);
Andy Whitcrofte9187bd2007-08-22 14:01:25 -07001082 __count_vm_events(PGDEACTIVATE, nr_active);
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001083
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001084 __mod_zone_page_state(zone, NR_ACTIVE_FILE,
1085 -count[LRU_ACTIVE_FILE]);
1086 __mod_zone_page_state(zone, NR_INACTIVE_FILE,
1087 -count[LRU_INACTIVE_FILE]);
1088 __mod_zone_page_state(zone, NR_ACTIVE_ANON,
1089 -count[LRU_ACTIVE_ANON]);
1090 __mod_zone_page_state(zone, NR_INACTIVE_ANON,
1091 -count[LRU_INACTIVE_ANON]);
1092
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001093 if (scanning_global_lru(sc))
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001094 zone->pages_scanned += nr_scan;
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001095
1096 reclaim_stat->recent_scanned[0] += count[LRU_INACTIVE_ANON];
1097 reclaim_stat->recent_scanned[0] += count[LRU_ACTIVE_ANON];
1098 reclaim_stat->recent_scanned[1] += count[LRU_INACTIVE_FILE];
1099 reclaim_stat->recent_scanned[1] += count[LRU_ACTIVE_FILE];
1100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 spin_unlock_irq(&zone->lru_lock);
1102
Andrew Morton69e05942006-03-22 00:08:19 -08001103 nr_scanned += nr_scan;
Andy Whitcroftc661b072007-08-22 14:01:26 -07001104 nr_freed = shrink_page_list(&page_list, sc, PAGEOUT_IO_ASYNC);
1105
1106 /*
1107 * If we are direct reclaiming for contiguous pages and we do
1108 * not reclaim everything in the list, try again and wait
1109 * for IO to complete. This will stall high-order allocations
1110 * but that should be acceptable to the caller
1111 */
1112 if (nr_freed < nr_taken && !current_is_kswapd() &&
1113 sc->order > PAGE_ALLOC_COSTLY_ORDER) {
1114 congestion_wait(WRITE, HZ/10);
1115
1116 /*
1117 * The attempt at page out may have made some
1118 * of the pages active, mark them inactive again.
1119 */
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001120 nr_active = clear_active_flags(&page_list, count);
Andy Whitcroftc661b072007-08-22 14:01:26 -07001121 count_vm_events(PGDEACTIVATE, nr_active);
1122
1123 nr_freed += shrink_page_list(&page_list, sc,
1124 PAGEOUT_IO_SYNC);
1125 }
1126
Andrew Morton05ff5132006-03-22 00:08:20 -08001127 nr_reclaimed += nr_freed;
Nick Piggina74609f2006-01-06 00:11:20 -08001128 local_irq_disable();
1129 if (current_is_kswapd()) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001130 __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scan);
1131 __count_vm_events(KSWAPD_STEAL, nr_freed);
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001132 } else if (scanning_global_lru(sc))
Christoph Lameterf8891e52006-06-30 01:55:45 -07001133 __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001134
Shantanu Goel918d3f92006-12-29 16:48:59 -08001135 __count_zone_vm_events(PGSTEAL, zone, nr_freed);
Nick Piggina74609f2006-01-06 00:11:20 -08001136
Wu Fengguangfb8d14e2006-03-22 00:08:28 -08001137 if (nr_taken == 0)
1138 goto done;
1139
Nick Piggina74609f2006-01-06 00:11:20 -08001140 spin_lock(&zone->lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 /*
1142 * Put back any unfreeable pages.
1143 */
1144 while (!list_empty(&page_list)) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001145 int lru;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 page = lru_to_page(&page_list);
Nick Piggin725d7042006-09-25 23:30:55 -07001147 VM_BUG_ON(PageLRU(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 list_del(&page->lru);
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001149 if (unlikely(!page_evictable(page, NULL))) {
1150 spin_unlock_irq(&zone->lru_lock);
1151 putback_lru_page(page);
1152 spin_lock_irq(&zone->lru_lock);
1153 continue;
1154 }
1155 SetPageLRU(page);
1156 lru = page_lru(page);
1157 add_page_to_lru_list(zone, page, lru);
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001158 if (PageActive(page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001159 int file = !!page_is_file_cache(page);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001160 reclaim_stat->recent_rotated[file]++;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001161 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 if (!pagevec_add(&pvec, page)) {
1163 spin_unlock_irq(&zone->lru_lock);
1164 __pagevec_release(&pvec);
1165 spin_lock_irq(&zone->lru_lock);
1166 }
1167 }
Andrew Morton69e05942006-03-22 00:08:19 -08001168 } while (nr_scanned < max_scan);
Wu Fengguangfb8d14e2006-03-22 00:08:28 -08001169 spin_unlock(&zone->lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170done:
Wu Fengguangfb8d14e2006-03-22 00:08:28 -08001171 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 pagevec_release(&pvec);
Andrew Morton05ff5132006-03-22 00:08:20 -08001173 return nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001176/*
1177 * We are about to scan this zone at a certain priority level. If that priority
1178 * level is smaller (ie: more urgent) than the previous priority, then note
1179 * that priority level within the zone. This is done so that when the next
1180 * process comes in to scan this zone, it will immediately start out at this
1181 * priority level rather than having to build up its own scanning priority.
1182 * Here, this priority affects only the reclaim-mapped threshold.
1183 */
1184static inline void note_zone_scanning_priority(struct zone *zone, int priority)
1185{
1186 if (priority < zone->prev_priority)
1187 zone->prev_priority = priority;
1188}
1189
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001190/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 * This moves pages from the active list to the inactive list.
1192 *
1193 * We move them the other way if the page is referenced by one or more
1194 * processes, from rmap.
1195 *
1196 * If the pages are mostly unmapped, the processing is fast and it is
1197 * appropriate to hold zone->lru_lock across the whole operation. But if
1198 * the pages are mapped, the processing is slow (page_referenced()) so we
1199 * should drop zone->lru_lock around each page. It's impossible to balance
1200 * this, so instead we remove the pages from the LRU while processing them.
1201 * It is safe to rely on PG_active against the non-LRU pages in here because
1202 * nobody will play with that bit on a non-LRU page.
1203 *
1204 * The downside is that we have to touch page->_count against each page.
1205 * But we had to alter page->flags anyway.
1206 */
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001207
1208
Andrew Morton1742f192006-03-22 00:08:21 -08001209static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001210 struct scan_control *sc, int priority, int file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
Andrew Morton69e05942006-03-22 00:08:19 -08001212 unsigned long pgmoved;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 int pgdeactivate = 0;
Andrew Morton69e05942006-03-22 00:08:19 -08001214 unsigned long pgscanned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 LIST_HEAD(l_hold); /* The pages which were snipped off */
Christoph Lameterb69408e2008-10-18 20:26:14 -07001216 LIST_HEAD(l_inactive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 struct page *page;
1218 struct pagevec pvec;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001219 enum lru_list lru;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001220 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
1222 lru_add_drain();
1223 spin_lock_irq(&zone->lru_lock);
Balbir Singh66e17072008-02-07 00:13:56 -08001224 pgmoved = sc->isolate_pages(nr_pages, &l_hold, &pgscanned, sc->order,
1225 ISOLATE_ACTIVE, zone,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001226 sc->mem_cgroup, 1, file);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001227 /*
1228 * zone->pages_scanned is used for detect zone's oom
1229 * mem_cgroup remembers nr_scan by itself.
1230 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001231 if (scanning_global_lru(sc)) {
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001232 zone->pages_scanned += pgscanned;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001233 }
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001234 reclaim_stat->recent_scanned[!!file] += pgmoved;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001235
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001236 if (file)
1237 __mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved);
1238 else
1239 __mod_zone_page_state(zone, NR_ACTIVE_ANON, -pgmoved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 spin_unlock_irq(&zone->lru_lock);
1241
Rik van Riel556adec2008-10-18 20:26:34 -07001242 pgmoved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 while (!list_empty(&l_hold)) {
1244 cond_resched();
1245 page = lru_to_page(&l_hold);
1246 list_del(&page->lru);
Rik van Riel7e9cd482008-10-18 20:26:35 -07001247
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001248 if (unlikely(!page_evictable(page, NULL))) {
1249 putback_lru_page(page);
1250 continue;
1251 }
1252
Rik van Riel7e9cd482008-10-18 20:26:35 -07001253 /* page_referenced clears PageReferenced */
1254 if (page_mapping_inuse(page) &&
1255 page_referenced(page, 0, sc->mem_cgroup))
1256 pgmoved++;
1257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 list_add(&page->lru, &l_inactive);
1259 }
1260
Andrew Mortonb5557492009-01-06 14:40:13 -08001261 /*
1262 * Move the pages to the [file or anon] inactive list.
1263 */
1264 pagevec_init(&pvec, 1);
1265 pgmoved = 0;
1266 lru = LRU_BASE + file * LRU_FILE;
1267
Johannes Weiner2a1dc502008-12-01 03:00:35 +01001268 spin_lock_irq(&zone->lru_lock);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001269 /*
Rik van Riel7e9cd482008-10-18 20:26:35 -07001270 * Count referenced pages from currently used mappings as
1271 * rotated, even though they are moved to the inactive list.
1272 * This helps balance scan pressure between file and anonymous
1273 * pages in get_scan_ratio.
1274 */
KOSAKI Motohiro3e2f41f2009-01-07 18:08:20 -08001275 reclaim_stat->recent_rotated[!!file] += pgmoved;
Rik van Riel556adec2008-10-18 20:26:34 -07001276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 while (!list_empty(&l_inactive)) {
1278 page = lru_to_page(&l_inactive);
1279 prefetchw_prev_lru_page(page, &l_inactive, flags);
Nick Piggin725d7042006-09-25 23:30:55 -07001280 VM_BUG_ON(PageLRU(page));
Nick Piggin8d438f92006-03-22 00:07:59 -08001281 SetPageLRU(page);
Nick Piggin725d7042006-09-25 23:30:55 -07001282 VM_BUG_ON(!PageActive(page));
Nick Piggin4c84cac2006-03-22 00:08:00 -08001283 ClearPageActive(page);
1284
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001285 list_move(&page->lru, &zone->lru[lru].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001286 mem_cgroup_add_lru_list(page, lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 pgmoved++;
1288 if (!pagevec_add(&pvec, page)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001289 __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 spin_unlock_irq(&zone->lru_lock);
1291 pgdeactivate += pgmoved;
1292 pgmoved = 0;
1293 if (buffer_heads_over_limit)
1294 pagevec_strip(&pvec);
1295 __pagevec_release(&pvec);
1296 spin_lock_irq(&zone->lru_lock);
1297 }
1298 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001299 __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 pgdeactivate += pgmoved;
1301 if (buffer_heads_over_limit) {
1302 spin_unlock_irq(&zone->lru_lock);
1303 pagevec_strip(&pvec);
1304 spin_lock_irq(&zone->lru_lock);
1305 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001306 __count_zone_vm_events(PGREFILL, zone, pgscanned);
1307 __count_vm_events(PGDEACTIVATE, pgdeactivate);
1308 spin_unlock_irq(&zone->lru_lock);
Rik van Riel68a223942008-10-18 20:26:23 -07001309 if (vm_swap_full())
1310 pagevec_swap_free(&pvec);
Nick Piggina74609f2006-01-06 00:11:20 -08001311
1312 pagevec_release(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313}
1314
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001315static int inactive_anon_is_low_global(struct zone *zone)
KOSAKI Motohirof89eb902009-01-07 18:08:14 -08001316{
1317 unsigned long active, inactive;
1318
1319 active = zone_page_state(zone, NR_ACTIVE_ANON);
1320 inactive = zone_page_state(zone, NR_INACTIVE_ANON);
1321
1322 if (inactive * zone->inactive_ratio < active)
1323 return 1;
1324
1325 return 0;
1326}
1327
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001328/**
1329 * inactive_anon_is_low - check if anonymous pages need to be deactivated
1330 * @zone: zone to check
1331 * @sc: scan control of this context
1332 *
1333 * Returns true if the zone does not have enough inactive anon pages,
1334 * meaning some active anon pages need to be deactivated.
1335 */
1336static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc)
1337{
1338 int low;
1339
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001340 if (scanning_global_lru(sc))
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001341 low = inactive_anon_is_low_global(zone);
1342 else
KOSAKI Motohiroc772be92009-01-07 18:08:25 -08001343 low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001344 return low;
1345}
1346
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001347static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
Christoph Lameterb69408e2008-10-18 20:26:14 -07001348 struct zone *zone, struct scan_control *sc, int priority)
1349{
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001350 int file = is_file_lru(lru);
1351
Rik van Riel556adec2008-10-18 20:26:34 -07001352 if (lru == LRU_ACTIVE_FILE) {
1353 shrink_active_list(nr_to_scan, zone, sc, priority, file);
1354 return 0;
1355 }
1356
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001357 if (lru == LRU_ACTIVE_ANON && inactive_anon_is_low(zone, sc)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001358 shrink_active_list(nr_to_scan, zone, sc, priority, file);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001359 return 0;
1360 }
Rik van Riel33c120e2008-10-18 20:26:36 -07001361 return shrink_inactive_list(nr_to_scan, zone, sc, priority, file);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001362}
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364/*
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001365 * Determine how aggressively the anon and file LRU lists should be
1366 * scanned. The relative value of each set of LRU lists is determined
1367 * by looking at the fraction of the pages scanned we did rotate back
1368 * onto the active list instead of evict.
1369 *
1370 * percent[0] specifies how much pressure to put on ram/swap backed
1371 * memory, while percent[1] determines pressure on the file LRUs.
1372 */
1373static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
1374 unsigned long *percent)
1375{
1376 unsigned long anon, file, free;
1377 unsigned long anon_prio, file_prio;
1378 unsigned long ap, fp;
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001379 struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001380
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001381 /* If we have no swap space, do not bother scanning anon pages. */
1382 if (nr_swap_pages <= 0) {
1383 percent[0] = 0;
1384 percent[1] = 100;
1385 return;
1386 }
1387
KOSAKI Motohiroc9f299d2009-01-07 18:08:16 -08001388 anon = zone_nr_pages(zone, sc, LRU_ACTIVE_ANON) +
1389 zone_nr_pages(zone, sc, LRU_INACTIVE_ANON);
1390 file = zone_nr_pages(zone, sc, LRU_ACTIVE_FILE) +
1391 zone_nr_pages(zone, sc, LRU_INACTIVE_FILE);
Hugh Dickinsb9627162009-01-06 14:39:41 -08001392
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001393 if (scanning_global_lru(sc)) {
KOSAKI Motohiroeeee9a82009-01-07 18:08:17 -08001394 free = zone_page_state(zone, NR_FREE_PAGES);
1395 /* If we have very few page cache pages,
1396 force-scan anon pages. */
1397 if (unlikely(file + free <= zone->pages_high)) {
1398 percent[0] = 100;
1399 percent[1] = 0;
1400 return;
1401 }
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001402 }
1403
1404 /*
1405 * OK, so we have swap space and a fair amount of page cache
1406 * pages. We use the recently rotated / recently scanned
1407 * ratios to determine how valuable each cache is.
1408 *
1409 * Because workloads change over time (and to avoid overflow)
1410 * we keep these statistics as a floating average, which ends
1411 * up weighing recent references more than old ones.
1412 *
1413 * anon in [0], file in [1]
1414 */
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001415 if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001416 spin_lock_irq(&zone->lru_lock);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001417 reclaim_stat->recent_scanned[0] /= 2;
1418 reclaim_stat->recent_rotated[0] /= 2;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001419 spin_unlock_irq(&zone->lru_lock);
1420 }
1421
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001422 if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001423 spin_lock_irq(&zone->lru_lock);
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001424 reclaim_stat->recent_scanned[1] /= 2;
1425 reclaim_stat->recent_rotated[1] /= 2;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001426 spin_unlock_irq(&zone->lru_lock);
1427 }
1428
1429 /*
1430 * With swappiness at 100, anonymous and file have the same priority.
1431 * This scanning priority is essentially the inverse of IO cost.
1432 */
1433 anon_prio = sc->swappiness;
1434 file_prio = 200 - sc->swappiness;
1435
1436 /*
Rik van Riel00d80892008-11-19 15:36:44 -08001437 * The amount of pressure on anon vs file pages is inversely
1438 * proportional to the fraction of recently scanned pages on
1439 * each list that were recently referenced and in active use.
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001440 */
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001441 ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1);
1442 ap /= reclaim_stat->recent_rotated[0] + 1;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001443
KOSAKI Motohiro6e901572009-01-07 18:08:15 -08001444 fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1);
1445 fp /= reclaim_stat->recent_rotated[1] + 1;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001446
1447 /* Normalize to percentages */
1448 percent[0] = 100 * ap / (ap + fp + 1);
1449 percent[1] = 100 - percent[0];
1450}
1451
1452
1453/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 * This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
1455 */
Rik van Riela79311c2009-01-06 14:40:01 -08001456static void shrink_zone(int priority, struct zone *zone,
Andrew Morton05ff5132006-03-22 00:08:20 -08001457 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
Christoph Lameterb69408e2008-10-18 20:26:14 -07001459 unsigned long nr[NR_LRU_LISTS];
Christoph Lameter86959492006-03-22 00:08:18 -08001460 unsigned long nr_to_scan;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001461 unsigned long percent[2]; /* anon @ 0; file @ 1 */
Christoph Lameterb69408e2008-10-18 20:26:14 -07001462 enum lru_list l;
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001463 unsigned long nr_reclaimed = sc->nr_reclaimed;
1464 unsigned long swap_cluster_max = sc->swap_cluster_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001466 get_scan_ratio(zone, sc, percent);
1467
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001468 for_each_evictable_lru(l) {
KOSAKI Motohiro9439c1c2009-01-07 18:08:21 -08001469 int file = is_file_lru(l);
1470 int scan;
Johannes Weinere0f79b82008-10-18 20:26:55 -07001471
KOSAKI Motohiro9439c1c2009-01-07 18:08:21 -08001472 scan = zone_page_state(zone, NR_LRU_BASE + l);
1473 if (priority) {
1474 scan >>= priority;
1475 scan = (scan * percent[file]) / 100;
1476 }
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001477 if (scanning_global_lru(sc)) {
Johannes Weinere0f79b82008-10-18 20:26:55 -07001478 zone->lru[l].nr_scan += scan;
Christoph Lameterb69408e2008-10-18 20:26:14 -07001479 nr[l] = zone->lru[l].nr_scan;
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001480 if (nr[l] >= swap_cluster_max)
Christoph Lameterb69408e2008-10-18 20:26:14 -07001481 zone->lru[l].nr_scan = 0;
1482 else
1483 nr[l] = 0;
KOSAKI Motohiro9439c1c2009-01-07 18:08:21 -08001484 } else
1485 nr[l] = scan;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001486 }
1487
Rik van Riel556adec2008-10-18 20:26:34 -07001488 while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
1489 nr[LRU_INACTIVE_FILE]) {
Lee Schermerhorn894bc312008-10-18 20:26:39 -07001490 for_each_evictable_lru(l) {
Christoph Lameterb69408e2008-10-18 20:26:14 -07001491 if (nr[l]) {
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001492 nr_to_scan = min(nr[l], swap_cluster_max);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001493 nr[l] -= nr_to_scan;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001495 nr_reclaimed += shrink_list(l, nr_to_scan,
1496 zone, sc, priority);
Christoph Lameterb69408e2008-10-18 20:26:14 -07001497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 }
Rik van Riela79311c2009-01-06 14:40:01 -08001499 /*
1500 * On large memory systems, scan >> priority can become
1501 * really large. This is fine for the starting priority;
1502 * we want to put equal scanning pressure on each zone.
1503 * However, if the VM has a harder time of freeing pages,
1504 * with multiple processes reclaiming pages, the total
1505 * freeing target can get unreasonably large.
1506 */
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001507 if (nr_reclaimed > swap_cluster_max &&
Rik van Riela79311c2009-01-06 14:40:01 -08001508 priority < DEF_PRIORITY && !current_is_kswapd())
1509 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 }
1511
KOSAKI Motohiro01dbe5c2009-01-06 14:40:02 -08001512 sc->nr_reclaimed = nr_reclaimed;
1513
Rik van Riel556adec2008-10-18 20:26:34 -07001514 /*
1515 * Even if we did not try to evict anon pages at all, we want to
1516 * rebalance the anon lru active/inactive ratio.
1517 */
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001518 if (inactive_anon_is_low(zone, sc))
Rik van Riel556adec2008-10-18 20:26:34 -07001519 shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1520
Andrew Morton232ea4d2007-02-28 20:13:21 -08001521 throttle_vm_writeout(sc->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522}
1523
1524/*
1525 * This is the direct reclaim path, for page-allocating processes. We only
1526 * try to reclaim pages from zones which will satisfy the caller's allocation
1527 * request.
1528 *
1529 * We reclaim from a zone even if that zone is over pages_high. Because:
1530 * a) The caller may be trying to free *extra* pages to satisfy a higher-order
1531 * allocation or
1532 * b) The zones may be over pages_high but they must go *over* pages_high to
1533 * satisfy the `incremental min' zone defense algorithm.
1534 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 * If a zone is deemed to be full of pinned pages then just give it a light
1536 * scan then give up on it.
1537 */
Rik van Riela79311c2009-01-06 14:40:01 -08001538static void shrink_zones(int priority, struct zonelist *zonelist,
Andrew Morton05ff5132006-03-22 00:08:20 -08001539 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540{
Mel Gorman54a6eb52008-04-28 02:12:16 -07001541 enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
Mel Gormandd1a2392008-04-28 02:12:17 -07001542 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001543 struct zone *zone;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001544
Nick Piggin408d8542006-09-25 23:31:27 -07001545 sc->all_unreclaimable = 1;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001546 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08001547 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 continue;
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001549 /*
1550 * Take care memory controller reclaiming has small influence
1551 * to global LRU.
1552 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001553 if (scanning_global_lru(sc)) {
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001554 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1555 continue;
1556 note_zone_scanning_priority(zone, priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001558 if (zone_is_all_unreclaimable(zone) &&
1559 priority != DEF_PRIORITY)
1560 continue; /* Let kswapd poll it */
1561 sc->all_unreclaimable = 0;
1562 } else {
1563 /*
1564 * Ignore cpuset limitation here. We just want to reduce
1565 * # of used pages by us regardless of memory shortage.
1566 */
1567 sc->all_unreclaimable = 0;
1568 mem_cgroup_note_reclaim_priority(sc->mem_cgroup,
1569 priority);
1570 }
Nick Piggin408d8542006-09-25 23:31:27 -07001571
Rik van Riela79311c2009-01-06 14:40:01 -08001572 shrink_zone(priority, zone, sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 }
1574}
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001575
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576/*
1577 * This is the main entry point to direct page reclaim.
1578 *
1579 * If a full scan of the inactive list fails to free enough memory then we
1580 * are "out of memory" and something needs to be killed.
1581 *
1582 * If the caller is !__GFP_FS then the probability of a failure is reasonably
1583 * high - the zone may be full of dirty or under-writeback pages, which this
1584 * caller can't do much about. We kick pdflush and take explicit naps in the
1585 * hope that some of these pages can be written. But if the allocating task
1586 * holds filesystem locks which prevent writeout this might not work, and the
1587 * allocation attempt will fail.
Nishanth Aravamudana41f24e2008-04-29 00:58:25 -07001588 *
1589 * returns: 0, if no pages reclaimed
1590 * else, the number of pages reclaimed
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 */
Mel Gormandac1d272008-04-28 02:12:12 -07001592static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
Mel Gormandd1a2392008-04-28 02:12:17 -07001593 struct scan_control *sc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594{
1595 int priority;
kosaki.motohiro@jp.fujitsu.comc700be32008-06-12 15:21:27 -07001596 unsigned long ret = 0;
Andrew Morton69e05942006-03-22 00:08:19 -08001597 unsigned long total_scanned = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 struct reclaim_state *reclaim_state = current->reclaim_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 unsigned long lru_pages = 0;
Mel Gormandd1a2392008-04-28 02:12:17 -07001600 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001601 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -07001602 enum zone_type high_zoneidx = gfp_zone(sc->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
Keika Kobayashi873b4772008-07-25 01:48:52 -07001604 delayacct_freepages_start();
1605
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001606 if (scanning_global_lru(sc))
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001607 count_vm_event(ALLOCSTALL);
1608 /*
1609 * mem_cgroup will not do shrink_slab.
1610 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001611 if (scanning_global_lru(sc)) {
Mel Gorman54a6eb52008-04-28 02:12:16 -07001612 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001614 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1615 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001617 lru_pages += zone_lru_pages(zone);
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001618 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 }
1620
1621 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
Balbir Singh66e17072008-02-07 00:13:56 -08001622 sc->nr_scanned = 0;
Rik van Rielf7b7fd82005-11-28 13:44:07 -08001623 if (!priority)
1624 disable_swap_token();
Rik van Riela79311c2009-01-06 14:40:01 -08001625 shrink_zones(priority, zonelist, sc);
Balbir Singh66e17072008-02-07 00:13:56 -08001626 /*
1627 * Don't shrink slabs when reclaiming memory from
1628 * over limit cgroups
1629 */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001630 if (scanning_global_lru(sc)) {
Mel Gormandd1a2392008-04-28 02:12:17 -07001631 shrink_slab(sc->nr_scanned, sc->gfp_mask, lru_pages);
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -08001632 if (reclaim_state) {
Rik van Riela79311c2009-01-06 14:40:01 -08001633 sc->nr_reclaimed += reclaim_state->reclaimed_slab;
KAMEZAWA Hiroyuki91a45472008-02-07 00:14:29 -08001634 reclaim_state->reclaimed_slab = 0;
1635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 }
Balbir Singh66e17072008-02-07 00:13:56 -08001637 total_scanned += sc->nr_scanned;
Rik van Riela79311c2009-01-06 14:40:01 -08001638 if (sc->nr_reclaimed >= sc->swap_cluster_max) {
1639 ret = sc->nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 goto out;
1641 }
1642
1643 /*
1644 * Try to write back as many pages as we just scanned. This
1645 * tends to cause slow streaming writers to write data to the
1646 * disk smoothly, at the dirtying rate, which is nice. But
1647 * that's undesirable in laptop mode, where we *want* lumpy
1648 * writeout. So in laptop mode, write out the whole world.
1649 */
Balbir Singh66e17072008-02-07 00:13:56 -08001650 if (total_scanned > sc->swap_cluster_max +
1651 sc->swap_cluster_max / 2) {
Pekka J Enberg687a21c2005-06-28 20:44:55 -07001652 wakeup_pdflush(laptop_mode ? 0 : total_scanned);
Balbir Singh66e17072008-02-07 00:13:56 -08001653 sc->may_writepage = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 }
1655
1656 /* Take a nap, wait for some writeback to complete */
Andrew Morton4dd4b922008-03-24 12:29:52 -07001657 if (sc->nr_scanned && priority < DEF_PRIORITY - 2)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001658 congestion_wait(WRITE, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 }
Fernando Luis Vazquez Cao87547ee2008-07-29 22:33:42 -07001660 /* top priority shrink_zones still had more to do? don't OOM, then */
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001661 if (!sc->all_unreclaimable && scanning_global_lru(sc))
Rik van Riela79311c2009-01-06 14:40:01 -08001662 ret = sc->nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663out:
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001664 /*
1665 * Now that we've scanned all the zones at this priority level, note
1666 * that level within the zone so that the next thread which performs
1667 * scanning of this zone will immediately start out at this priority
1668 * level. This affects only the decision whether or not to bring
1669 * mapped pages onto the inactive list.
1670 */
1671 if (priority < 0)
1672 priority = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
KAMEZAWA Hiroyukie72e2bd2009-01-07 18:08:23 -08001674 if (scanning_global_lru(sc)) {
Mel Gorman54a6eb52008-04-28 02:12:16 -07001675 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
KAMEZAWA Hiroyuki1cfb4192008-02-07 00:14:37 -08001677 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
1678 continue;
1679
1680 zone->prev_priority = priority;
1681 }
1682 } else
1683 mem_cgroup_record_reclaim_priority(sc->mem_cgroup, priority);
1684
Keika Kobayashi873b4772008-07-25 01:48:52 -07001685 delayacct_freepages_end();
1686
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 return ret;
1688}
1689
Mel Gormandac1d272008-04-28 02:12:12 -07001690unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
1691 gfp_t gfp_mask)
Balbir Singh66e17072008-02-07 00:13:56 -08001692{
1693 struct scan_control sc = {
1694 .gfp_mask = gfp_mask,
1695 .may_writepage = !laptop_mode,
1696 .swap_cluster_max = SWAP_CLUSTER_MAX,
1697 .may_swap = 1,
1698 .swappiness = vm_swappiness,
1699 .order = order,
1700 .mem_cgroup = NULL,
1701 .isolate_pages = isolate_pages_global,
1702 };
1703
Mel Gormandd1a2392008-04-28 02:12:17 -07001704 return do_try_to_free_pages(zonelist, &sc);
Balbir Singh66e17072008-02-07 00:13:56 -08001705}
1706
Balbir Singh00f0b822008-03-04 14:28:39 -08001707#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Balbir Singh66e17072008-02-07 00:13:56 -08001708
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001709unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001710 gfp_t gfp_mask,
1711 bool noswap,
1712 unsigned int swappiness)
Balbir Singh66e17072008-02-07 00:13:56 -08001713{
1714 struct scan_control sc = {
Balbir Singh66e17072008-02-07 00:13:56 -08001715 .may_writepage = !laptop_mode,
1716 .may_swap = 1,
1717 .swap_cluster_max = SWAP_CLUSTER_MAX,
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08001718 .swappiness = swappiness,
Balbir Singh66e17072008-02-07 00:13:56 -08001719 .order = 0,
1720 .mem_cgroup = mem_cont,
1721 .isolate_pages = mem_cgroup_isolate_pages,
1722 };
Mel Gormandac1d272008-04-28 02:12:12 -07001723 struct zonelist *zonelist;
Balbir Singh66e17072008-02-07 00:13:56 -08001724
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08001725 if (noswap)
1726 sc.may_swap = 0;
1727
Mel Gormandd1a2392008-04-28 02:12:17 -07001728 sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
1729 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
1730 zonelist = NODE_DATA(numa_node_id())->node_zonelists;
1731 return do_try_to_free_pages(zonelist, &sc);
Balbir Singh66e17072008-02-07 00:13:56 -08001732}
1733#endif
1734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735/*
1736 * For kswapd, balance_pgdat() will work across all this node's zones until
1737 * they are all at pages_high.
1738 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 * Returns the number of pages which were actually freed.
1740 *
1741 * There is special handling here for zones which are full of pinned pages.
1742 * This can happen if the pages are all mlocked, or if they are all used by
1743 * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
1744 * What we do is to detect the case where all pages in the zone have been
1745 * scanned twice and there has been zero successful reclaim. Mark the zone as
1746 * dead and from now on, only perform a short scan. Basically we're polling
1747 * the zone for when the problem goes away.
1748 *
1749 * kswapd scans the zones in the highmem->normal->dma direction. It skips
1750 * zones which have free_pages > pages_high, but once a zone is found to have
1751 * free_pages <= pages_high, we scan that zone and the lower zones regardless
1752 * of the number of free pages in the lower zones. This interoperates with
1753 * the page allocator fallback scheme to ensure that aging of pages is balanced
1754 * across the zones.
1755 */
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001756static unsigned long balance_pgdat(pg_data_t *pgdat, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 int all_zones_ok;
1759 int priority;
1760 int i;
Andrew Morton69e05942006-03-22 00:08:19 -08001761 unsigned long total_scanned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 struct reclaim_state *reclaim_state = current->reclaim_state;
Andrew Morton179e9632006-03-22 00:08:18 -08001763 struct scan_control sc = {
1764 .gfp_mask = GFP_KERNEL,
1765 .may_swap = 1,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001766 .swap_cluster_max = SWAP_CLUSTER_MAX,
1767 .swappiness = vm_swappiness,
Andy Whitcroft5ad333e2007-07-17 04:03:16 -07001768 .order = order,
Balbir Singh66e17072008-02-07 00:13:56 -08001769 .mem_cgroup = NULL,
1770 .isolate_pages = isolate_pages_global,
Andrew Morton179e9632006-03-22 00:08:18 -08001771 };
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001772 /*
1773 * temp_priority is used to remember the scanning priority at which
1774 * this zone was successfully refilled to free_pages == pages_high.
1775 */
1776 int temp_priority[MAX_NR_ZONES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
1778loop_again:
1779 total_scanned = 0;
Rik van Riela79311c2009-01-06 14:40:01 -08001780 sc.nr_reclaimed = 0;
Christoph Lameterc0bbbc72006-06-11 15:22:26 -07001781 sc.may_writepage = !laptop_mode;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001782 count_vm_event(PAGEOUTRUN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001784 for (i = 0; i < pgdat->nr_zones; i++)
1785 temp_priority[i] = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787 for (priority = DEF_PRIORITY; priority >= 0; priority--) {
1788 int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */
1789 unsigned long lru_pages = 0;
1790
Rik van Rielf7b7fd82005-11-28 13:44:07 -08001791 /* The swap token gets in the way of swapout... */
1792 if (!priority)
1793 disable_swap_token();
1794
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 all_zones_ok = 1;
1796
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001797 /*
1798 * Scan in the highmem->dma direction for the highest
1799 * zone which needs scanning
1800 */
1801 for (i = pgdat->nr_zones - 1; i >= 0; i--) {
1802 struct zone *zone = pgdat->node_zones + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001804 if (!populated_zone(zone))
1805 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
David Rientjese815af92007-10-16 23:25:54 -07001807 if (zone_is_all_unreclaimable(zone) &&
1808 priority != DEF_PRIORITY)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001809 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810
Rik van Riel556adec2008-10-18 20:26:34 -07001811 /*
1812 * Do some background aging of the anon list, to give
1813 * pages a chance to be referenced before reclaiming.
1814 */
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08001815 if (inactive_anon_is_low(zone, &sc))
Rik van Riel556adec2008-10-18 20:26:34 -07001816 shrink_active_list(SWAP_CLUSTER_MAX, zone,
1817 &sc, priority, 0);
1818
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001819 if (!zone_watermark_ok(zone, order, zone->pages_high,
1820 0, 0)) {
1821 end_zone = i;
Andrew Mortone1dbeda2006-12-06 20:32:01 -08001822 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 }
Andrew Mortone1dbeda2006-12-06 20:32:01 -08001825 if (i < 0)
1826 goto out;
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 for (i = 0; i <= end_zone; i++) {
1829 struct zone *zone = pgdat->node_zones + i;
1830
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001831 lru_pages += zone_lru_pages(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 }
1833
1834 /*
1835 * Now scan the zone in the dma->highmem direction, stopping
1836 * at the last zone which needs scanning.
1837 *
1838 * We do this because the page allocator works in the opposite
1839 * direction. This prevents the page allocator from allocating
1840 * pages behind kswapd's direction of progress, which would
1841 * cause too much scanning of the lower zones.
1842 */
1843 for (i = 0; i <= end_zone; i++) {
1844 struct zone *zone = pgdat->node_zones + i;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07001845 int nr_slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846
Con Kolivasf3fe6512006-01-06 00:11:15 -08001847 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 continue;
1849
David Rientjese815af92007-10-16 23:25:54 -07001850 if (zone_is_all_unreclaimable(zone) &&
1851 priority != DEF_PRIORITY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 continue;
1853
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07001854 if (!zone_watermark_ok(zone, order, zone->pages_high,
1855 end_zone, 0))
1856 all_zones_ok = 0;
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001857 temp_priority[i] = priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 sc.nr_scanned = 0;
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001859 note_zone_scanning_priority(zone, priority);
Rik van Riel32a43302007-10-16 01:24:50 -07001860 /*
1861 * We put equal pressure on every zone, unless one
1862 * zone has way too many pages free already.
1863 */
1864 if (!zone_watermark_ok(zone, order, 8*zone->pages_high,
1865 end_zone, 0))
Rik van Riela79311c2009-01-06 14:40:01 -08001866 shrink_zone(priority, zone, &sc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 reclaim_state->reclaimed_slab = 0;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07001868 nr_slab = shrink_slab(sc.nr_scanned, GFP_KERNEL,
1869 lru_pages);
Rik van Riela79311c2009-01-06 14:40:01 -08001870 sc.nr_reclaimed += reclaim_state->reclaimed_slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 total_scanned += sc.nr_scanned;
David Rientjese815af92007-10-16 23:25:54 -07001872 if (zone_is_all_unreclaimable(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 continue;
akpm@osdl.orgb15e0902005-06-21 17:14:35 -07001874 if (nr_slab == 0 && zone->pages_scanned >=
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001875 (zone_lru_pages(zone) * 6))
David Rientjese815af92007-10-16 23:25:54 -07001876 zone_set_flag(zone,
1877 ZONE_ALL_UNRECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 /*
1879 * If we've done a decent amount of scanning and
1880 * the reclaim ratio is low, start doing writepage
1881 * even in laptop mode
1882 */
1883 if (total_scanned > SWAP_CLUSTER_MAX * 2 &&
Rik van Riela79311c2009-01-06 14:40:01 -08001884 total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 sc.may_writepage = 1;
1886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 if (all_zones_ok)
1888 break; /* kswapd: all done */
1889 /*
1890 * OK, kswapd is getting into trouble. Take a nap, then take
1891 * another pass across the zones.
1892 */
Andrew Morton4dd4b922008-03-24 12:29:52 -07001893 if (total_scanned && priority < DEF_PRIORITY - 2)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001894 congestion_wait(WRITE, HZ/10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896 /*
1897 * We do this so kswapd doesn't build up large priorities for
1898 * example when it is freeing in parallel with allocators. It
1899 * matches the direct reclaim path behaviour in terms of impact
1900 * on zone->*_priority.
1901 */
Rik van Riela79311c2009-01-06 14:40:01 -08001902 if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 break;
1904 }
1905out:
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001906 /*
1907 * Note within each zone the priority level at which this zone was
1908 * brought into a happy state. So that the next thread which scans this
1909 * zone will start out at that priority level.
1910 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 for (i = 0; i < pgdat->nr_zones; i++) {
1912 struct zone *zone = pgdat->node_zones + i;
1913
Martin Bligh3bb1a8522006-10-28 10:38:24 -07001914 zone->prev_priority = temp_priority[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 }
1916 if (!all_zones_ok) {
1917 cond_resched();
Rafael J. Wysocki83573762006-12-06 20:34:18 -08001918
1919 try_to_freeze();
1920
KOSAKI Motohiro73ce02e2009-01-06 14:40:33 -08001921 /*
1922 * Fragmentation may mean that the system cannot be
1923 * rebalanced for high-order allocations in all zones.
1924 * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX,
1925 * it means the zones have been fully scanned and are still
1926 * not balanced. For high-order allocations, there is
1927 * little point trying all over again as kswapd may
1928 * infinite loop.
1929 *
1930 * Instead, recheck all watermarks at order-0 as they
1931 * are the most important. If watermarks are ok, kswapd will go
1932 * back to sleep. High-order users can still perform direct
1933 * reclaim if they wish.
1934 */
1935 if (sc.nr_reclaimed < SWAP_CLUSTER_MAX)
1936 order = sc.order = 0;
1937
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 goto loop_again;
1939 }
1940
Rik van Riela79311c2009-01-06 14:40:01 -08001941 return sc.nr_reclaimed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942}
1943
1944/*
1945 * The background pageout daemon, started as a kernel thread
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001946 * from the init process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 *
1948 * This basically trickles out pages so that we have _some_
1949 * free memory available even if there is no other activity
1950 * that frees anything up. This is needed for things like routing
1951 * etc, where we otherwise might have all activity going on in
1952 * asynchronous contexts that cannot page things out.
1953 *
1954 * If there are applications that are active memory-allocators
1955 * (most normal use), this basically shouldn't matter.
1956 */
1957static int kswapd(void *p)
1958{
1959 unsigned long order;
1960 pg_data_t *pgdat = (pg_data_t*)p;
1961 struct task_struct *tsk = current;
1962 DEFINE_WAIT(wait);
1963 struct reclaim_state reclaim_state = {
1964 .reclaimed_slab = 0,
1965 };
Mike Travisc5f59f02008-04-04 18:11:10 -07001966 node_to_cpumask_ptr(cpumask, pgdat->node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Rusty Russell174596a2009-01-01 10:12:29 +10301968 if (!cpumask_empty(cpumask))
Mike Travisc5f59f02008-04-04 18:11:10 -07001969 set_cpus_allowed_ptr(tsk, cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 current->reclaim_state = &reclaim_state;
1971
1972 /*
1973 * Tell the memory management that we're a "memory allocator",
1974 * and that if we need more memory we should get access to it
1975 * regardless (see "__alloc_pages()"). "kswapd" should
1976 * never get caught in the normal page freeing logic.
1977 *
1978 * (Kswapd normally doesn't need memory anyway, but sometimes
1979 * you need a small amount of memory in order to be able to
1980 * page out something else, and this flag essentially protects
1981 * us from recursively trying to free more memory as we're
1982 * trying to free the first piece of memory in the first place).
1983 */
Christoph Lameter930d9152006-01-08 01:00:47 -08001984 tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
Rafael J. Wysocki83144182007-07-17 04:03:35 -07001985 set_freezable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
1987 order = 0;
1988 for ( ; ; ) {
1989 unsigned long new_order;
Christoph Lameter3e1d1d22005-06-24 23:13:50 -07001990
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
1992 new_order = pgdat->kswapd_max_order;
1993 pgdat->kswapd_max_order = 0;
1994 if (order < new_order) {
1995 /*
1996 * Don't sleep if someone wants a larger 'order'
1997 * allocation
1998 */
1999 order = new_order;
2000 } else {
Rafael J. Wysockib1296cc2007-05-06 14:50:48 -07002001 if (!freezing(current))
2002 schedule();
2003
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 order = pgdat->kswapd_max_order;
2005 }
2006 finish_wait(&pgdat->kswapd_wait, &wait);
2007
Rafael J. Wysockib1296cc2007-05-06 14:50:48 -07002008 if (!try_to_freeze()) {
2009 /* We can speed up thawing tasks if we don't call
2010 * balance_pgdat after returning from the refrigerator
2011 */
2012 balance_pgdat(pgdat, order);
2013 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 }
2015 return 0;
2016}
2017
2018/*
2019 * A zone is low on free memory, so wake its kswapd task to service it.
2020 */
2021void wakeup_kswapd(struct zone *zone, int order)
2022{
2023 pg_data_t *pgdat;
2024
Con Kolivasf3fe6512006-01-06 00:11:15 -08002025 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 return;
2027
2028 pgdat = zone->zone_pgdat;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08002029 if (zone_watermark_ok(zone, order, zone->pages_low, 0, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 return;
2031 if (pgdat->kswapd_max_order < order)
2032 pgdat->kswapd_max_order = order;
Paul Jackson02a0e532006-12-13 00:34:25 -08002033 if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 return;
Con Kolivas8d0986e2005-09-13 01:25:07 -07002035 if (!waitqueue_active(&pgdat->kswapd_wait))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 return;
Con Kolivas8d0986e2005-09-13 01:25:07 -07002037 wake_up_interruptible(&pgdat->kswapd_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038}
2039
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002040unsigned long global_lru_pages(void)
2041{
2042 return global_page_state(NR_ACTIVE_ANON)
2043 + global_page_state(NR_ACTIVE_FILE)
2044 + global_page_state(NR_INACTIVE_ANON)
2045 + global_page_state(NR_INACTIVE_FILE);
2046}
2047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048#ifdef CONFIG_PM
2049/*
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002050 * Helper function for shrink_all_memory(). Tries to reclaim 'nr_pages' pages
2051 * from LRU lists system-wide, for given pass and priority, and returns the
2052 * number of reclaimed pages
2053 *
2054 * For pass > 3 we also try to shrink the LRU lists that contain a few pages
2055 */
Nigel Cunninghame07aa052006-12-22 01:07:21 -08002056static unsigned long shrink_all_zones(unsigned long nr_pages, int prio,
2057 int pass, struct scan_control *sc)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002058{
2059 struct zone *zone;
Johannes Weiner0cb57252009-02-14 02:04:10 +01002060 unsigned long ret = 0;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002061
2062 for_each_zone(zone) {
Johannes Weiner0cb57252009-02-14 02:04:10 +01002063 enum lru_list l;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002064
2065 if (!populated_zone(zone))
2066 continue;
David Rientjese815af92007-10-16 23:25:54 -07002067 if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002068 continue;
2069
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002070 for_each_evictable_lru(l) {
Johannes Weiner0cb57252009-02-14 02:04:10 +01002071 enum zone_stat_item ls = NR_LRU_BASE + l;
2072 unsigned long lru_pages = zone_page_state(zone, ls);
2073
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002074 /* For pass = 0, we don't shrink the active list */
Johannes Weiner0cb57252009-02-14 02:04:10 +01002075 if (pass == 0 && (l == LRU_ACTIVE_ANON ||
2076 l == LRU_ACTIVE_FILE))
Christoph Lameterb69408e2008-10-18 20:26:14 -07002077 continue;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002078
Johannes Weiner0cb57252009-02-14 02:04:10 +01002079 zone->lru[l].nr_scan += (lru_pages >> prio) + 1;
Christoph Lameterb69408e2008-10-18 20:26:14 -07002080 if (zone->lru[l].nr_scan >= nr_pages || pass > 3) {
Johannes Weiner0cb57252009-02-14 02:04:10 +01002081 unsigned long nr_to_scan;
2082
Christoph Lameterb69408e2008-10-18 20:26:14 -07002083 zone->lru[l].nr_scan = 0;
Johannes Weiner0cb57252009-02-14 02:04:10 +01002084 nr_to_scan = min(nr_pages, lru_pages);
Christoph Lameterb69408e2008-10-18 20:26:14 -07002085 ret += shrink_list(l, nr_to_scan, zone,
2086 sc, prio);
2087 if (ret >= nr_pages)
2088 return ret;
2089 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002090 }
2091 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002092 return ret;
2093}
2094
2095/*
2096 * Try to free `nr_pages' of memory, system-wide, and return the number of
2097 * freed pages.
2098 *
2099 * Rather than trying to age LRUs the aim is to preserve the overall
2100 * LRU order by reclaiming preferentially
2101 * inactive > active > active referenced > active mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 */
Andrew Morton69e05942006-03-22 00:08:19 -08002103unsigned long shrink_all_memory(unsigned long nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002105 unsigned long lru_pages, nr_slab;
Andrew Morton69e05942006-03-22 00:08:19 -08002106 unsigned long ret = 0;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002107 int pass;
2108 struct reclaim_state reclaim_state;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002109 struct scan_control sc = {
2110 .gfp_mask = GFP_KERNEL,
2111 .may_swap = 0,
2112 .swap_cluster_max = nr_pages,
2113 .may_writepage = 1,
Balbir Singh66e17072008-02-07 00:13:56 -08002114 .isolate_pages = isolate_pages_global,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 };
2116
2117 current->reclaim_state = &reclaim_state;
Andrew Morton69e05942006-03-22 00:08:19 -08002118
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002119 lru_pages = global_lru_pages();
Christoph Lameter972d1a72006-09-25 23:31:51 -07002120 nr_slab = global_page_state(NR_SLAB_RECLAIMABLE);
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002121 /* If slab caches are huge, it's better to hit them first */
2122 while (nr_slab >= lru_pages) {
2123 reclaim_state.reclaimed_slab = 0;
2124 shrink_slab(nr_pages, sc.gfp_mask, lru_pages);
2125 if (!reclaim_state.reclaimed_slab)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 break;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002127
2128 ret += reclaim_state.reclaimed_slab;
2129 if (ret >= nr_pages)
2130 goto out;
2131
2132 nr_slab -= reclaim_state.reclaimed_slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002134
2135 /*
2136 * We try to shrink LRUs in 5 passes:
2137 * 0 = Reclaim from inactive_list only
2138 * 1 = Reclaim from active list but don't reclaim mapped
2139 * 2 = 2nd pass of type 1
2140 * 3 = Reclaim mapped (normal reclaim)
2141 * 4 = 2nd pass of type 3
2142 */
2143 for (pass = 0; pass < 5; pass++) {
2144 int prio;
2145
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002146 /* Force reclaiming mapped pages in the passes #3 and #4 */
Johannes Weiner30491032009-02-14 02:03:08 +01002147 if (pass > 2)
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002148 sc.may_swap = 1;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002149
2150 for (prio = DEF_PRIORITY; prio >= 0; prio--) {
2151 unsigned long nr_to_scan = nr_pages - ret;
2152
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002153 sc.nr_scanned = 0;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002154 ret += shrink_all_zones(nr_to_scan, prio, pass, &sc);
2155 if (ret >= nr_pages)
2156 goto out;
2157
2158 reclaim_state.reclaimed_slab = 0;
Andrew Morton76395d32007-01-05 16:37:05 -08002159 shrink_slab(sc.nr_scanned, sc.gfp_mask,
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002160 global_lru_pages());
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002161 ret += reclaim_state.reclaimed_slab;
2162 if (ret >= nr_pages)
2163 goto out;
2164
2165 if (sc.nr_scanned && prio < DEF_PRIORITY - 2)
Andrew Morton3fcfab12006-10-19 23:28:16 -07002166 congestion_wait(WRITE, HZ / 10);
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002167 }
Rafael J. Wysocki248a0302006-03-22 00:09:04 -08002168 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002169
2170 /*
2171 * If ret = 0, we could not shrink LRUs, but there may be something
2172 * in slab caches
2173 */
Andrew Morton76395d32007-01-05 16:37:05 -08002174 if (!ret) {
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002175 do {
2176 reclaim_state.reclaimed_slab = 0;
Rik van Riel4f98a2f2008-10-18 20:26:32 -07002177 shrink_slab(nr_pages, sc.gfp_mask, global_lru_pages());
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002178 ret += reclaim_state.reclaimed_slab;
2179 } while (ret < nr_pages && reclaim_state.reclaimed_slab > 0);
Andrew Morton76395d32007-01-05 16:37:05 -08002180 }
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002181
2182out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 current->reclaim_state = NULL;
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002184
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 return ret;
2186}
2187#endif
2188
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189/* It's optimal to keep kswapds on the same CPUs as their memory, but
2190 not required for correctness. So if the last cpu in a node goes
2191 away, we get changed to run anywhere: as the first one comes back,
2192 restore their cpu bindings. */
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002193static int __devinit cpu_callback(struct notifier_block *nfb,
Andrew Morton69e05942006-03-22 00:08:19 -08002194 unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195{
Yasunori Goto58c0a4a2007-10-16 01:25:40 -07002196 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07002198 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
Yasunori Goto58c0a4a2007-10-16 01:25:40 -07002199 for_each_node_state(nid, N_HIGH_MEMORY) {
Mike Travisc5f59f02008-04-04 18:11:10 -07002200 pg_data_t *pgdat = NODE_DATA(nid);
2201 node_to_cpumask_ptr(mask, pgdat->node_id);
2202
Rusty Russell3e597942009-01-01 10:12:24 +10302203 if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 /* One of our CPUs online: restore mask */
Mike Travisc5f59f02008-04-04 18:11:10 -07002205 set_cpus_allowed_ptr(pgdat->kswapd, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 }
2207 }
2208 return NOTIFY_OK;
2209}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
Yasunori Goto3218ae12006-06-27 02:53:33 -07002211/*
2212 * This kswapd start function will be called by init and node-hot-add.
2213 * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
2214 */
2215int kswapd_run(int nid)
2216{
2217 pg_data_t *pgdat = NODE_DATA(nid);
2218 int ret = 0;
2219
2220 if (pgdat->kswapd)
2221 return 0;
2222
2223 pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
2224 if (IS_ERR(pgdat->kswapd)) {
2225 /* failure at boot is fatal */
2226 BUG_ON(system_state == SYSTEM_BOOTING);
2227 printk("Failed to start kswapd on node %d\n",nid);
2228 ret = -1;
2229 }
2230 return ret;
2231}
2232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233static int __init kswapd_init(void)
2234{
Yasunori Goto3218ae12006-06-27 02:53:33 -07002235 int nid;
Andrew Morton69e05942006-03-22 00:08:19 -08002236
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 swap_setup();
Christoph Lameter9422ffb2007-10-16 01:25:31 -07002238 for_each_node_state(nid, N_HIGH_MEMORY)
Yasunori Goto3218ae12006-06-27 02:53:33 -07002239 kswapd_run(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 hotcpu_notifier(cpu_callback, 0);
2241 return 0;
2242}
2243
2244module_init(kswapd_init)
Christoph Lameter9eeff232006-01-18 17:42:31 -08002245
2246#ifdef CONFIG_NUMA
2247/*
2248 * Zone reclaim mode
2249 *
2250 * If non-zero call zone_reclaim when the number of free pages falls below
2251 * the watermarks.
Christoph Lameter9eeff232006-01-18 17:42:31 -08002252 */
2253int zone_reclaim_mode __read_mostly;
2254
Christoph Lameter1b2ffb72006-02-01 03:05:34 -08002255#define RECLAIM_OFF 0
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -07002256#define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
Christoph Lameter1b2ffb72006-02-01 03:05:34 -08002257#define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
2258#define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
2259
Christoph Lameter9eeff232006-01-18 17:42:31 -08002260/*
Christoph Lametera92f7122006-02-01 03:05:32 -08002261 * Priority for ZONE_RECLAIM. This determines the fraction of pages
2262 * of a node considered for each zone_reclaim. 4 scans 1/16th of
2263 * a zone.
2264 */
2265#define ZONE_RECLAIM_PRIORITY 4
2266
Christoph Lameter9eeff232006-01-18 17:42:31 -08002267/*
Christoph Lameter96146342006-07-03 00:24:13 -07002268 * Percentage of pages in a zone that must be unmapped for zone_reclaim to
2269 * occur.
2270 */
2271int sysctl_min_unmapped_ratio = 1;
2272
2273/*
Christoph Lameter0ff38492006-09-25 23:31:52 -07002274 * If the number of slab pages in a zone grows beyond this percentage then
2275 * slab reclaim needs to occur.
2276 */
2277int sysctl_min_slab_ratio = 5;
2278
2279/*
Christoph Lameter9eeff232006-01-18 17:42:31 -08002280 * Try to free up some pages from this zone through reclaim.
2281 */
Andrew Morton179e9632006-03-22 00:08:18 -08002282static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
Christoph Lameter9eeff232006-01-18 17:42:31 -08002283{
Christoph Lameter7fb2d462006-03-22 00:08:22 -08002284 /* Minimum pages needed in order to stay on node */
Andrew Morton69e05942006-03-22 00:08:19 -08002285 const unsigned long nr_pages = 1 << order;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002286 struct task_struct *p = current;
2287 struct reclaim_state reclaim_state;
Christoph Lameter86959492006-03-22 00:08:18 -08002288 int priority;
Andrew Morton179e9632006-03-22 00:08:18 -08002289 struct scan_control sc = {
2290 .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
2291 .may_swap = !!(zone_reclaim_mode & RECLAIM_SWAP),
Andrew Morton69e05942006-03-22 00:08:19 -08002292 .swap_cluster_max = max_t(unsigned long, nr_pages,
2293 SWAP_CLUSTER_MAX),
Andrew Morton179e9632006-03-22 00:08:18 -08002294 .gfp_mask = gfp_mask,
Rafael J. Wysockid6277db2006-06-23 02:03:18 -07002295 .swappiness = vm_swappiness,
Balbir Singh66e17072008-02-07 00:13:56 -08002296 .isolate_pages = isolate_pages_global,
Andrew Morton179e9632006-03-22 00:08:18 -08002297 };
Christoph Lameter83e33a42006-09-25 23:31:53 -07002298 unsigned long slab_reclaimable;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002299
2300 disable_swap_token();
Christoph Lameter9eeff232006-01-18 17:42:31 -08002301 cond_resched();
Christoph Lameterd4f77962006-02-24 13:04:22 -08002302 /*
2303 * We need to be able to allocate from the reserves for RECLAIM_SWAP
2304 * and we also need to be able to write out pages for RECLAIM_WRITE
2305 * and RECLAIM_SWAP.
2306 */
2307 p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002308 reclaim_state.reclaimed_slab = 0;
2309 p->reclaim_state = &reclaim_state;
Christoph Lameterc84db232006-02-01 03:05:29 -08002310
Christoph Lameter0ff38492006-09-25 23:31:52 -07002311 if (zone_page_state(zone, NR_FILE_PAGES) -
2312 zone_page_state(zone, NR_FILE_MAPPED) >
2313 zone->min_unmapped_pages) {
2314 /*
2315 * Free memory by calling shrink zone with increasing
2316 * priorities until we have enough memory freed.
2317 */
2318 priority = ZONE_RECLAIM_PRIORITY;
2319 do {
Martin Bligh3bb1a8522006-10-28 10:38:24 -07002320 note_zone_scanning_priority(zone, priority);
Rik van Riela79311c2009-01-06 14:40:01 -08002321 shrink_zone(priority, zone, &sc);
Christoph Lameter0ff38492006-09-25 23:31:52 -07002322 priority--;
Rik van Riela79311c2009-01-06 14:40:01 -08002323 } while (priority >= 0 && sc.nr_reclaimed < nr_pages);
Christoph Lameter0ff38492006-09-25 23:31:52 -07002324 }
Christoph Lameterc84db232006-02-01 03:05:29 -08002325
Christoph Lameter83e33a42006-09-25 23:31:53 -07002326 slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
2327 if (slab_reclaimable > zone->min_slab_pages) {
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002328 /*
Christoph Lameter7fb2d462006-03-22 00:08:22 -08002329 * shrink_slab() does not currently allow us to determine how
Christoph Lameter0ff38492006-09-25 23:31:52 -07002330 * many pages were freed in this zone. So we take the current
2331 * number of slab pages and shake the slab until it is reduced
2332 * by the same nr_pages that we used for reclaiming unmapped
2333 * pages.
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002334 *
Christoph Lameter0ff38492006-09-25 23:31:52 -07002335 * Note that shrink_slab will free memory on all zones and may
2336 * take a long time.
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002337 */
Christoph Lameter0ff38492006-09-25 23:31:52 -07002338 while (shrink_slab(sc.nr_scanned, gfp_mask, order) &&
Christoph Lameter83e33a42006-09-25 23:31:53 -07002339 zone_page_state(zone, NR_SLAB_RECLAIMABLE) >
2340 slab_reclaimable - nr_pages)
Christoph Lameter0ff38492006-09-25 23:31:52 -07002341 ;
Christoph Lameter83e33a42006-09-25 23:31:53 -07002342
2343 /*
2344 * Update nr_reclaimed by the number of slab pages we
2345 * reclaimed from this zone.
2346 */
Rik van Riela79311c2009-01-06 14:40:01 -08002347 sc.nr_reclaimed += slab_reclaimable -
Christoph Lameter83e33a42006-09-25 23:31:53 -07002348 zone_page_state(zone, NR_SLAB_RECLAIMABLE);
Christoph Lameter2a16e3f2006-02-01 03:05:35 -08002349 }
2350
Christoph Lameter9eeff232006-01-18 17:42:31 -08002351 p->reclaim_state = NULL;
Christoph Lameterd4f77962006-02-24 13:04:22 -08002352 current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
Rik van Riela79311c2009-01-06 14:40:01 -08002353 return sc.nr_reclaimed >= nr_pages;
Christoph Lameter9eeff232006-01-18 17:42:31 -08002354}
Andrew Morton179e9632006-03-22 00:08:18 -08002355
2356int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
2357{
Andrew Morton179e9632006-03-22 00:08:18 -08002358 int node_id;
David Rientjesd773ed62007-10-16 23:26:01 -07002359 int ret;
Andrew Morton179e9632006-03-22 00:08:18 -08002360
2361 /*
Christoph Lameter0ff38492006-09-25 23:31:52 -07002362 * Zone reclaim reclaims unmapped file backed pages and
2363 * slab pages if we are over the defined limits.
Christoph Lameter34aa1332006-06-30 01:55:37 -07002364 *
Christoph Lameter96146342006-07-03 00:24:13 -07002365 * A small portion of unmapped file backed pages is needed for
2366 * file I/O otherwise pages read by file I/O will be immediately
2367 * thrown out if the zone is overallocated. So we do not reclaim
2368 * if less than a specified percentage of the zone is used by
2369 * unmapped file backed pages.
Andrew Morton179e9632006-03-22 00:08:18 -08002370 */
Christoph Lameter34aa1332006-06-30 01:55:37 -07002371 if (zone_page_state(zone, NR_FILE_PAGES) -
Christoph Lameter0ff38492006-09-25 23:31:52 -07002372 zone_page_state(zone, NR_FILE_MAPPED) <= zone->min_unmapped_pages
2373 && zone_page_state(zone, NR_SLAB_RECLAIMABLE)
2374 <= zone->min_slab_pages)
Christoph Lameter96146342006-07-03 00:24:13 -07002375 return 0;
Andrew Morton179e9632006-03-22 00:08:18 -08002376
David Rientjesd773ed62007-10-16 23:26:01 -07002377 if (zone_is_all_unreclaimable(zone))
2378 return 0;
2379
Andrew Morton179e9632006-03-22 00:08:18 -08002380 /*
David Rientjesd773ed62007-10-16 23:26:01 -07002381 * Do not scan if the allocation should not be delayed.
Andrew Morton179e9632006-03-22 00:08:18 -08002382 */
David Rientjesd773ed62007-10-16 23:26:01 -07002383 if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
Andrew Morton179e9632006-03-22 00:08:18 -08002384 return 0;
2385
2386 /*
2387 * Only run zone reclaim on the local zone or on zones that do not
2388 * have associated processors. This will favor the local processor
2389 * over remote processors and spread off node memory allocations
2390 * as wide as possible.
2391 */
Christoph Lameter89fa3022006-09-25 23:31:55 -07002392 node_id = zone_to_nid(zone);
Christoph Lameter37c07082007-10-16 01:25:36 -07002393 if (node_state(node_id, N_CPU) && node_id != numa_node_id())
Andrew Morton179e9632006-03-22 00:08:18 -08002394 return 0;
David Rientjesd773ed62007-10-16 23:26:01 -07002395
2396 if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
2397 return 0;
2398 ret = __zone_reclaim(zone, gfp_mask, order);
2399 zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
2400
2401 return ret;
Andrew Morton179e9632006-03-22 00:08:18 -08002402}
Christoph Lameter9eeff232006-01-18 17:42:31 -08002403#endif
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002404
2405#ifdef CONFIG_UNEVICTABLE_LRU
2406/*
2407 * page_evictable - test whether a page is evictable
2408 * @page: the page to test
2409 * @vma: the VMA in which the page is or will be mapped, may be NULL
2410 *
2411 * Test whether page is evictable--i.e., should be placed on active/inactive
Nick Pigginb291f002008-10-18 20:26:44 -07002412 * lists vs unevictable list. The vma argument is !NULL when called from the
2413 * fault path to determine how to instantate a new page.
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002414 *
2415 * Reasons page might not be evictable:
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002416 * (1) page's mapping marked unevictable
Nick Pigginb291f002008-10-18 20:26:44 -07002417 * (2) page is part of an mlocked VMA
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002418 *
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002419 */
2420int page_evictable(struct page *page, struct vm_area_struct *vma)
2421{
2422
Lee Schermerhornba9ddf42008-10-18 20:26:42 -07002423 if (mapping_unevictable(page_mapping(page)))
2424 return 0;
2425
Nick Pigginb291f002008-10-18 20:26:44 -07002426 if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page)))
2427 return 0;
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002428
2429 return 1;
2430}
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002431
2432/**
2433 * check_move_unevictable_page - check page for evictability and move to appropriate zone lru list
2434 * @page: page to check evictability and move to appropriate lru list
2435 * @zone: zone page is in
2436 *
2437 * Checks a page for evictability and moves the page to the appropriate
2438 * zone lru list.
2439 *
2440 * Restrictions: zone->lru_lock must be held, page must be on LRU and must
2441 * have PageUnevictable set.
2442 */
2443static void check_move_unevictable_page(struct page *page, struct zone *zone)
2444{
2445 VM_BUG_ON(PageActive(page));
2446
2447retry:
2448 ClearPageUnevictable(page);
2449 if (page_evictable(page, NULL)) {
2450 enum lru_list l = LRU_INACTIVE_ANON + page_is_file_cache(page);
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002451
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002452 __dec_zone_state(zone, NR_UNEVICTABLE);
2453 list_move(&page->lru, &zone->lru[l].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002454 mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l);
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002455 __inc_zone_state(zone, NR_INACTIVE_ANON + l);
2456 __count_vm_event(UNEVICTABLE_PGRESCUED);
2457 } else {
2458 /*
2459 * rotate unevictable list
2460 */
2461 SetPageUnevictable(page);
2462 list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08002463 mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE);
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002464 if (page_evictable(page, NULL))
2465 goto retry;
2466 }
2467}
2468
2469/**
2470 * scan_mapping_unevictable_pages - scan an address space for evictable pages
2471 * @mapping: struct address_space to scan for evictable pages
2472 *
2473 * Scan all pages in mapping. Check unevictable pages for
2474 * evictability and move them to the appropriate zone lru list.
2475 */
2476void scan_mapping_unevictable_pages(struct address_space *mapping)
2477{
2478 pgoff_t next = 0;
2479 pgoff_t end = (i_size_read(mapping->host) + PAGE_CACHE_SIZE - 1) >>
2480 PAGE_CACHE_SHIFT;
2481 struct zone *zone;
2482 struct pagevec pvec;
2483
2484 if (mapping->nrpages == 0)
2485 return;
2486
2487 pagevec_init(&pvec, 0);
2488 while (next < end &&
2489 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
2490 int i;
2491 int pg_scanned = 0;
2492
2493 zone = NULL;
2494
2495 for (i = 0; i < pagevec_count(&pvec); i++) {
2496 struct page *page = pvec.pages[i];
2497 pgoff_t page_index = page->index;
2498 struct zone *pagezone = page_zone(page);
2499
2500 pg_scanned++;
2501 if (page_index > next)
2502 next = page_index;
2503 next++;
2504
2505 if (pagezone != zone) {
2506 if (zone)
2507 spin_unlock_irq(&zone->lru_lock);
2508 zone = pagezone;
2509 spin_lock_irq(&zone->lru_lock);
2510 }
2511
2512 if (PageLRU(page) && PageUnevictable(page))
2513 check_move_unevictable_page(page, zone);
2514 }
2515 if (zone)
2516 spin_unlock_irq(&zone->lru_lock);
2517 pagevec_release(&pvec);
2518
2519 count_vm_events(UNEVICTABLE_PGSCANNED, pg_scanned);
2520 }
2521
2522}
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002523
2524/**
2525 * scan_zone_unevictable_pages - check unevictable list for evictable pages
2526 * @zone - zone of which to scan the unevictable list
2527 *
2528 * Scan @zone's unevictable LRU lists to check for pages that have become
2529 * evictable. Move those that have to @zone's inactive list where they
2530 * become candidates for reclaim, unless shrink_inactive_zone() decides
2531 * to reactivate them. Pages that are still unevictable are rotated
2532 * back onto @zone's unevictable list.
2533 */
2534#define SCAN_UNEVICTABLE_BATCH_SIZE 16UL /* arbitrary lock hold batch size */
KOSAKI Motohiro14b90b22009-01-06 14:39:45 -08002535static void scan_zone_unevictable_pages(struct zone *zone)
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002536{
2537 struct list_head *l_unevictable = &zone->lru[LRU_UNEVICTABLE].list;
2538 unsigned long scan;
2539 unsigned long nr_to_scan = zone_page_state(zone, NR_UNEVICTABLE);
2540
2541 while (nr_to_scan > 0) {
2542 unsigned long batch_size = min(nr_to_scan,
2543 SCAN_UNEVICTABLE_BATCH_SIZE);
2544
2545 spin_lock_irq(&zone->lru_lock);
2546 for (scan = 0; scan < batch_size; scan++) {
2547 struct page *page = lru_to_page(l_unevictable);
2548
2549 if (!trylock_page(page))
2550 continue;
2551
2552 prefetchw_prev_lru_page(page, l_unevictable, flags);
2553
2554 if (likely(PageLRU(page) && PageUnevictable(page)))
2555 check_move_unevictable_page(page, zone);
2556
2557 unlock_page(page);
2558 }
2559 spin_unlock_irq(&zone->lru_lock);
2560
2561 nr_to_scan -= batch_size;
2562 }
2563}
2564
2565
2566/**
2567 * scan_all_zones_unevictable_pages - scan all unevictable lists for evictable pages
2568 *
2569 * A really big hammer: scan all zones' unevictable LRU lists to check for
2570 * pages that have become evictable. Move those back to the zones'
2571 * inactive list where they become candidates for reclaim.
2572 * This occurs when, e.g., we have unswappable pages on the unevictable lists,
2573 * and we add swap to the system. As such, it runs in the context of a task
2574 * that has possibly/probably made some previously unevictable pages
2575 * evictable.
2576 */
KOSAKI Motohiroff301532009-01-06 14:39:44 -08002577static void scan_all_zones_unevictable_pages(void)
Lee Schermerhornaf936a12008-10-18 20:26:53 -07002578{
2579 struct zone *zone;
2580
2581 for_each_zone(zone) {
2582 scan_zone_unevictable_pages(zone);
2583 }
2584}
2585
2586/*
2587 * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of
2588 * all nodes' unevictable lists for evictable pages
2589 */
2590unsigned long scan_unevictable_pages;
2591
2592int scan_unevictable_handler(struct ctl_table *table, int write,
2593 struct file *file, void __user *buffer,
2594 size_t *length, loff_t *ppos)
2595{
2596 proc_doulongvec_minmax(table, write, file, buffer, length, ppos);
2597
2598 if (write && *(unsigned long *)table->data)
2599 scan_all_zones_unevictable_pages();
2600
2601 scan_unevictable_pages = 0;
2602 return 0;
2603}
2604
2605/*
2606 * per node 'scan_unevictable_pages' attribute. On demand re-scan of
2607 * a specified node's per zone unevictable lists for evictable pages.
2608 */
2609
2610static ssize_t read_scan_unevictable_node(struct sys_device *dev,
2611 struct sysdev_attribute *attr,
2612 char *buf)
2613{
2614 return sprintf(buf, "0\n"); /* always zero; should fit... */
2615}
2616
2617static ssize_t write_scan_unevictable_node(struct sys_device *dev,
2618 struct sysdev_attribute *attr,
2619 const char *buf, size_t count)
2620{
2621 struct zone *node_zones = NODE_DATA(dev->id)->node_zones;
2622 struct zone *zone;
2623 unsigned long res;
2624 unsigned long req = strict_strtoul(buf, 10, &res);
2625
2626 if (!req)
2627 return 1; /* zero is no-op */
2628
2629 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2630 if (!populated_zone(zone))
2631 continue;
2632 scan_zone_unevictable_pages(zone);
2633 }
2634 return 1;
2635}
2636
2637
2638static SYSDEV_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
2639 read_scan_unevictable_node,
2640 write_scan_unevictable_node);
2641
2642int scan_unevictable_register_node(struct node *node)
2643{
2644 return sysdev_create_file(&node->sysdev, &attr_scan_unevictable_pages);
2645}
2646
2647void scan_unevictable_unregister_node(struct node *node)
2648{
2649 sysdev_remove_file(&node->sysdev, &attr_scan_unevictable_pages);
2650}
2651
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002652#endif